diff options
Diffstat (limited to 'erts/emulator/sys')
34 files changed, 21443 insertions, 0 deletions
diff --git a/erts/emulator/sys/common/erl_check_io.c b/erts/emulator/sys/common/erl_check_io.c new file mode 100644 index 0000000000..218bd79584 --- /dev/null +++ b/erts/emulator/sys/common/erl_check_io.c @@ -0,0 +1,1912 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2006-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +/* + * Description: Check I/O + * + * Author: Rickard Green + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#define ERL_CHECK_IO_C__ +#define ERTS_WANT_BREAK_HANDLING +#ifndef WANT_NONBLOCKING +# define WANT_NONBLOCKING +#endif +#include "sys.h" +#include "global.h" +#include "erl_check_io.h" + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS +# define ERTS_DRV_EV_STATE_EXTRA_SIZE 128 +#else +# include "safe_hash.h" +# define DRV_EV_STATE_HTAB_SIZE 1024 +#endif + +typedef char EventStateType; +#define ERTS_EV_TYPE_NONE ((EventStateType) 0) +#define ERTS_EV_TYPE_DRV_SEL ((EventStateType) 1) /* driver_select */ +#define ERTS_EV_TYPE_DRV_EV ((EventStateType) 2) /* driver_event */ +#define ERTS_EV_TYPE_STOP_USE ((EventStateType) 3) /* pending stop_select */ + +typedef char EventStateFlags; +#define ERTS_EV_FLAG_USED ((EventStateFlags) 1) /* ERL_DRV_USE has been turned on */ + + +#if defined(ERTS_KERNEL_POLL_VERSION) +# define ERTS_CIO_EXPORT(FUNC) FUNC ## _kp +#elif defined(ERTS_NO_KERNEL_POLL_VERSION) +# define ERTS_CIO_EXPORT(FUNC) FUNC ## _nkp +#else +# define ERTS_CIO_EXPORT(FUNC) FUNC +#endif + +#define ERTS_CIO_HAVE_DRV_EVENT \ + (ERTS_POLL_USE_POLL && !ERTS_POLL_USE_KERNEL_POLL) + +#define ERTS_CIO_POLL_CTL ERTS_POLL_EXPORT(erts_poll_control) +#define ERTS_CIO_POLL_WAIT ERTS_POLL_EXPORT(erts_poll_wait) +#define ERTS_CIO_POLL_INTR ERTS_POLL_EXPORT(erts_poll_interrupt) +#define ERTS_CIO_POLL_INTR_TMD ERTS_POLL_EXPORT(erts_poll_interrupt_timed) +#define ERTS_CIO_NEW_POLLSET ERTS_POLL_EXPORT(erts_poll_create_pollset) +#define ERTS_CIO_FREE_POLLSET ERTS_POLL_EXPORT(erts_poll_destroy_pollset) +#define ERTS_CIO_POLL_MAX_FDS ERTS_POLL_EXPORT(erts_poll_max_fds) +#define ERTS_CIO_POLL_INIT ERTS_POLL_EXPORT(erts_poll_init) +#define ERTS_CIO_POLL_INFO ERTS_POLL_EXPORT(erts_poll_info) + +static struct pollset_info +{ + ErtsPollSet ps; + erts_smp_atomic_t in_poll_wait; /* set while doing poll */ +#ifdef ERTS_SMP + struct removed_fd* removed_list; /* list of deselected fd's*/ + erts_smp_spinlock_t removed_list_lock; +#endif +}pollset; +#define NUM_OF_POLLSETS 1 + +typedef struct { +#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS + SafeHashBucket hb; +#endif + ErtsSysFdType fd; + union { + ErtsDrvEventDataState *event; /* ERTS_EV_TYPE_DRV_EV */ + ErtsDrvSelectDataState *select; /* ERTS_EV_TYPE_DRV_SEL */ + erts_driver_t* drv_ptr; /* ERTS_EV_TYPE_STOP_USE */ + } driver; + ErtsPollEvents events; + unsigned short remove_cnt; /* number of removed_fd's referring to this fd */ + EventStateType type; + EventStateFlags flags; +} ErtsDrvEventState; + +#ifdef ERTS_SMP +struct removed_fd { + struct removed_fd *next; +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + ErtsSysFdType fd; +#else + ErtsDrvEventState* state; + #ifdef DEBUG + ErtsSysFdType fd; + #endif +#endif + +}; +#endif + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS +static int max_fds = -1; +#endif +#define DRV_EV_STATE_LOCK_CNT 16 +static union { + erts_smp_mtx_t lck; + byte _cache_line_alignment[64]; +}drv_ev_state_locks[DRV_EV_STATE_LOCK_CNT]; + +#ifdef ERTS_SMP +static ERTS_INLINE erts_smp_mtx_t* fd_mtx(ErtsSysFdType fd) +{ + int hash = (int)fd; +# ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS + hash ^= (hash >> 9); +# endif + return &drv_ev_state_locks[hash % DRV_EV_STATE_LOCK_CNT].lck; +} +#else +# define fd_mtx(fd) NULL +#endif + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + +static erts_smp_atomic_t drv_ev_state_len; +static ErtsDrvEventState *drv_ev_state; +static erts_smp_mtx_t drv_ev_state_grow_lock; /* prevent lock-hogging of racing growers */ + +#else +static SafeHash drv_ev_state_tab; +static int num_state_prealloc; +static ErtsDrvEventState *state_prealloc_first; +erts_smp_spinlock_t state_prealloc_lock; + +static ERTS_INLINE ErtsDrvEventState *hash_get_drv_ev_state(ErtsSysFdType fd) +{ + ErtsDrvEventState tmpl; + tmpl.fd = fd; + return (ErtsDrvEventState *) safe_hash_get(&drv_ev_state_tab, (void *) &tmpl); +} + +static ERTS_INLINE ErtsDrvEventState* hash_new_drv_ev_state(ErtsSysFdType fd) +{ + ErtsDrvEventState tmpl; + tmpl.fd = fd; + tmpl.driver.select = NULL; + tmpl.events = 0; + tmpl.remove_cnt = 0; + tmpl.type = ERTS_EV_TYPE_NONE; + tmpl.flags = 0; + return (ErtsDrvEventState *) safe_hash_put(&drv_ev_state_tab, (void *) &tmpl); +} + +static ERTS_INLINE void hash_erase_drv_ev_state(ErtsDrvEventState *state) +{ + ASSERT(state->remove_cnt == 0); + safe_hash_erase(&drv_ev_state_tab, (void *) state); +} + +#endif /* !ERTS_SYS_CONTINOUS_FD_NUMBERS */ + +static void stale_drv_select(Eterm id, ErtsDrvEventState *state, int mode); +static void select_steal(ErlDrvPort ix, ErtsDrvEventState *state, + int mode, int on); +static void print_select_op(erts_dsprintf_buf_t *dsbufp, + ErlDrvPort ix, ErtsSysFdType fd, int mode, int on); +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS +static void select_large_fd_error(ErlDrvPort, ErtsSysFdType, int, int); +#endif +#if ERTS_CIO_HAVE_DRV_EVENT +static void event_steal(ErlDrvPort ix, ErtsDrvEventState *state, + ErlDrvEventData event_data); +static void print_event_op(erts_dsprintf_buf_t *dsbufp, + ErlDrvPort, ErtsSysFdType, ErlDrvEventData); +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS +static void event_large_fd_error(ErlDrvPort, ErtsSysFdType, ErlDrvEventData); +#endif +#endif +static void steal_pending_stop_select(erts_dsprintf_buf_t*, ErlDrvPort, + ErtsDrvEventState*, int mode, int on); +static ERTS_INLINE Eterm +drvport2id(ErlDrvPort dp) +{ + Port *pp = erts_drvport2port(dp); + if (pp) + return pp->id; + else { + ASSERT(0); + return am_undefined; + } +} + +#ifdef ERTS_SMP +ERTS_SCHED_PREF_QUICK_ALLOC_IMPL(removed_fd, struct removed_fd, 64, ERTS_ALC_T_FD_LIST) +#endif + +static ERTS_INLINE void +remember_removed(ErtsDrvEventState *state, struct pollset_info* psi) +{ +#ifdef ERTS_SMP + struct removed_fd *fdlp; + ERTS_SMP_LC_ASSERT(erts_smp_lc_mtx_is_locked(fd_mtx(state->fd))); + if (erts_smp_atomic_read(&psi->in_poll_wait)) { + state->remove_cnt++; + ASSERT(state->remove_cnt > 0); + fdlp = removed_fd_alloc(); + #if defined(ERTS_SYS_CONTINOUS_FD_NUMBERS) || defined(DEBUG) + fdlp->fd = state->fd; + #endif + #ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS + fdlp->state = state; + #endif + erts_smp_spin_lock(&psi->removed_list_lock); + fdlp->next = psi->removed_list; + psi->removed_list = fdlp; + erts_smp_spin_unlock(&psi->removed_list_lock); + } +#endif +} + + +static ERTS_INLINE int +is_removed(ErtsDrvEventState *state) +{ +#ifdef ERTS_SMP + /* Note that there is a possible race here, where an fd is removed + (increasing remove_cnt) and then added again just before erts_poll_wait + is called by erts_check_io. Any polled event on the re-added fd will then + be falsely ignored. But that does not matter, as the event will trigger + again next time erl_check_io is called. */ + return state->remove_cnt > 0; +#else + return 0; +#endif +} + +static void +forget_removed(struct pollset_info* psi) +{ +#ifdef ERTS_SMP + struct removed_fd* fdlp; + struct removed_fd* tofree; + + /* Fast track: if (atomic_ptr(removed_list)==NULL) return; */ + + erts_smp_spin_lock(&psi->removed_list_lock); + fdlp = psi->removed_list; + psi->removed_list = NULL; + erts_smp_spin_unlock(&psi->removed_list_lock); + + while (fdlp) { + erts_driver_t* drv_ptr = NULL; + erts_smp_mtx_t* mtx; + ErtsSysFdType fd; + ErtsDrvEventState *state; + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + fd = fdlp->fd; + mtx = fd_mtx(fd); + erts_smp_mtx_lock(mtx); + state = &drv_ev_state[(int) fd]; +#else + state = fdlp->state; + fd = state->fd; + ASSERT(fd == fdlp->fd); + mtx = fd_mtx(fd); + erts_smp_mtx_lock(mtx); +#endif + ASSERT(state->remove_cnt > 0); + if (--state->remove_cnt == 0) { + switch (state->type) { + case ERTS_EV_TYPE_STOP_USE: + /* Now we can call stop_select */ + drv_ptr = state->driver.drv_ptr; + ASSERT(drv_ptr); + state->type = ERTS_EV_TYPE_NONE; + state->flags = 0; + state->driver.drv_ptr = NULL; + /* Fall through */ + case ERTS_EV_TYPE_NONE: +#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS + hash_erase_drv_ev_state(state); +#endif + break; + case ERTS_EV_TYPE_DRV_SEL: + case ERTS_EV_TYPE_DRV_EV: + break; + default: + ASSERT(0); + } + } + erts_smp_mtx_unlock(mtx); + if (drv_ptr) { + int was_unmasked = erts_block_fpe(); + (*drv_ptr->stop_select) (fd, NULL); + erts_unblock_fpe(was_unmasked); + if (drv_ptr->handle) { + erts_ddll_dereference_driver(drv_ptr->handle); + } + } + tofree = fdlp; + fdlp = fdlp->next; + removed_fd_free(tofree); + } +#endif /* ERTS_SMP */ +} + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS +static void +grow_drv_ev_state(int min_ix) +{ + int i; + int new_len = min_ix + 1 + ERTS_DRV_EV_STATE_EXTRA_SIZE; + if (new_len > max_fds) + new_len = max_fds; + + erts_smp_mtx_lock(&drv_ev_state_grow_lock); + if (erts_smp_atomic_read(&drv_ev_state_len) <= min_ix) { + for (i=0; i<DRV_EV_STATE_LOCK_CNT; i++) { /* lock all fd's */ + erts_smp_mtx_lock(&drv_ev_state_locks[i].lck); + } + drv_ev_state = (drv_ev_state + ? erts_realloc(ERTS_ALC_T_DRV_EV_STATE, + drv_ev_state, + sizeof(ErtsDrvEventState)*new_len) + : erts_alloc(ERTS_ALC_T_DRV_EV_STATE, + sizeof(ErtsDrvEventState)*new_len)); + for (i = erts_smp_atomic_read(&drv_ev_state_len); i < new_len; i++) { + drv_ev_state[i].fd = (ErtsSysFdType) i; + drv_ev_state[i].driver.select = NULL; + drv_ev_state[i].events = 0; + drv_ev_state[i].remove_cnt = 0; + drv_ev_state[i].type = ERTS_EV_TYPE_NONE; + drv_ev_state[i].flags = 0; + } + erts_smp_atomic_set(&drv_ev_state_len, new_len); + for (i=0; i<DRV_EV_STATE_LOCK_CNT; i++) { + erts_smp_mtx_unlock(&drv_ev_state_locks[i].lck); + } + } + /*else already grown by racing thread */ + + erts_smp_mtx_unlock(&drv_ev_state_grow_lock); +} +#endif /* ERTS_SYS_CONTINOUS_FD_NUMBERS */ + + +static ERTS_INLINE void +abort_task(Eterm id, ErtsPortTaskHandle *pthp, EventStateType type) +{ + if (is_nil(id)) { + ASSERT(type == ERTS_EV_TYPE_NONE + || !erts_port_task_is_scheduled(pthp)); + } + else if (erts_port_task_is_scheduled(pthp)) { + erts_port_task_abort(id, pthp); + ASSERT(erts_is_port_alive(id)); + } +} + +static ERTS_INLINE void +abort_tasks(ErtsDrvEventState *state, int mode) +{ + switch (mode) { + case 0: check_type: + switch (state->type) { +#if ERTS_CIO_HAVE_DRV_EVENT + case ERTS_EV_TYPE_DRV_EV: + abort_task(state->driver.event->port, + &state->driver.event->task, + ERTS_EV_TYPE_DRV_EV); + return; +#endif + case ERTS_EV_TYPE_NONE: + return; + default: + ASSERT(state->type == ERTS_EV_TYPE_DRV_SEL); + /* Fall through */ + } + case ERL_DRV_READ|ERL_DRV_WRITE: + case ERL_DRV_WRITE: + ASSERT(state->type == ERTS_EV_TYPE_DRV_SEL); + abort_task(state->driver.select->outport, + &state->driver.select->outtask, + state->type); + if (mode == ERL_DRV_WRITE) + break; + case ERL_DRV_READ: + ASSERT(state->type == ERTS_EV_TYPE_DRV_SEL); + abort_task(state->driver.select->inport, + &state->driver.select->intask, + state->type); + break; + default: + goto check_type; + } +} + +static void +deselect(ErtsDrvEventState *state, int mode) +{ + int do_wake = 0; + ErtsPollEvents rm_events; + ERTS_SMP_LC_ASSERT(erts_smp_lc_mtx_is_locked(fd_mtx(state->fd))); + ASSERT(state->events); + + abort_tasks(state, mode); + + if (!mode) + rm_events = state->events; + else { + rm_events = 0; + ASSERT(state->type == ERTS_EV_TYPE_DRV_SEL); + if (mode & ERL_DRV_READ) { + state->driver.select->inport = NIL; + rm_events |= ERTS_POLL_EV_IN; + } + if (mode & ERL_DRV_WRITE) { + state->driver.select->outport = NIL; + rm_events |= ERTS_POLL_EV_OUT; + } + } + + state->events = ERTS_CIO_POLL_CTL(pollset.ps, state->fd, rm_events, 0, &do_wake); + + if (!(state->events)) { + switch (state->type) { + case ERTS_EV_TYPE_DRV_SEL: + ASSERT(!erts_port_task_is_scheduled(&state->driver.select->intask)); + ASSERT(!erts_port_task_is_scheduled(&state->driver.select->outtask)); + erts_free(ERTS_ALC_T_DRV_SEL_D_STATE, + state->driver.select); + break; +#if ERTS_CIO_HAVE_DRV_EVENT + case ERTS_EV_TYPE_DRV_EV: + ASSERT(!erts_port_task_is_scheduled(&state->driver.event->task)); + erts_free(ERTS_ALC_T_DRV_EV_D_STATE, + state->driver.event); + break; +#endif + case ERTS_EV_TYPE_NONE: + break; + default: + ASSERT(0); + break; + } + + state->driver.select = NULL; + state->type = ERTS_EV_TYPE_NONE; + state->flags = 0; + remember_removed(state, &pollset); + } +} + + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS +# define IS_FD_UNKNOWN(state) ((state)->type == ERTS_EV_TYPE_NONE && (state)->remove_cnt == 0) +#else +# define IS_FD_UNKNOWN(state) ((state) == NULL) +#endif + + +int +ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix, + ErlDrvEvent e, + int mode, + int on) +{ + void (*stop_select_fn)(ErlDrvEvent, void*) = NULL; + Eterm id = drvport2id(ix); + ErtsSysFdType fd = (ErtsSysFdType) e; + ErtsPollEvents ctl_events = (ErtsPollEvents) 0; + ErtsPollEvents new_events, old_events; + ErtsDrvEventState *state; + int wake_poller; + int ret; + + ERTS_SMP_LC_ASSERT(erts_drvport2port(ix) + && erts_lc_is_port_locked(erts_drvport2port(ix))); + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + if ((unsigned)fd >= (unsigned)erts_smp_atomic_read(&drv_ev_state_len)) { + if (fd < 0) { + return -1; + } + if (fd >= max_fds) { + select_large_fd_error(ix, fd, mode, on); + return -1; + } + grow_drv_ev_state(fd); + } +#endif + + erts_smp_mtx_lock(fd_mtx(fd)); + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + state = &drv_ev_state[(int) fd]; +#else + state = hash_get_drv_ev_state(fd); /* may be NULL! */ +#endif + + if (!on && (mode&ERL_DRV_USE_NO_CALLBACK) == ERL_DRV_USE) { + if (IS_FD_UNKNOWN(state)) { + /* fast track to stop_select callback */ + stop_select_fn = erts_drvport2port(ix)->drv_ptr->stop_select; + ret = 0; + goto done_unknown; + } + mode |= (ERL_DRV_READ | ERL_DRV_WRITE); + wake_poller = 1; /* to eject fd from pollset (if needed) */ + } + else wake_poller = 0; + +#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS + if (state == NULL) { + state = hash_new_drv_ev_state(fd); + } +#endif + +#if ERTS_CIO_HAVE_DRV_EVENT + if (state->type == ERTS_EV_TYPE_DRV_EV) + select_steal(ix, state, mode, on); +#endif + if (state->type == ERTS_EV_TYPE_STOP_USE) { + erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); + print_select_op(dsbufp, ix, state->fd, mode, on); + steal_pending_stop_select(dsbufp, ix, state, mode, on); + if (state->type == ERTS_EV_TYPE_STOP_USE) { + ret = 0; + goto done; /* stop_select still pending */ + } + ASSERT(state->type == ERTS_EV_TYPE_NONE); + } + + if (mode & ERL_DRV_READ) { + if (state->type == ERTS_EV_TYPE_DRV_SEL) { + Eterm owner = state->driver.select->inport; + if (owner != id && is_not_nil(owner)) + select_steal(ix, state, mode, on); + } + ctl_events |= ERTS_POLL_EV_IN; + } + if (mode & ERL_DRV_WRITE) { + if (state->type == ERTS_EV_TYPE_DRV_SEL) { + Eterm owner = state->driver.select->outport; + if (owner != id && is_not_nil(owner)) + select_steal(ix, state, mode, on); + } + ctl_events |= ERTS_POLL_EV_OUT; + } + + ASSERT((state->type == ERTS_EV_TYPE_DRV_SEL) || + (state->type == ERTS_EV_TYPE_NONE && !state->events)); + + if (!on && !(state->flags & ERTS_EV_FLAG_USED) + && state->events && !(state->events & ~ctl_events)) { + /* Old driver removing all events. At least wake poller. + It will not make close() 100% safe but it will prevent + actions delayed by poll timeout. */ + wake_poller = 1; + } + + new_events = ERTS_CIO_POLL_CTL(pollset.ps, state->fd, ctl_events, on, &wake_poller); + + if (new_events & (ERTS_POLL_EV_ERR|ERTS_POLL_EV_NVAL)) { + if (state->type == ERTS_EV_TYPE_DRV_SEL && !state->events) { + state->type = ERTS_EV_TYPE_NONE; + state->flags = 0; + erts_free(ERTS_ALC_T_DRV_SEL_D_STATE, state->driver.select); + state->driver.select = NULL; + } + ret = -1; + goto done; + } + + old_events = state->events; + + ASSERT(on + ? (new_events == (state->events | ctl_events)) + : (new_events == (state->events & ~ctl_events))); + + ASSERT(state->type == ERTS_EV_TYPE_DRV_SEL + || state->type == ERTS_EV_TYPE_NONE); + + state->events = new_events; + if (ctl_events) { + if (on) { + if (state->type == ERTS_EV_TYPE_NONE) { + ErtsDrvSelectDataState *dsdsp + = erts_alloc(ERTS_ALC_T_DRV_SEL_D_STATE, + sizeof(ErtsDrvSelectDataState)); + dsdsp->inport = NIL; + dsdsp->outport = NIL; + erts_port_task_handle_init(&dsdsp->intask); + erts_port_task_handle_init(&dsdsp->outtask); + ASSERT(state->driver.select == NULL); + state->driver.select = dsdsp; + state->type = ERTS_EV_TYPE_DRV_SEL; + } + ASSERT(state->type == ERTS_EV_TYPE_DRV_SEL); + if (ctl_events & ERTS_POLL_EV_IN) + state->driver.select->inport = id; + if (ctl_events & ERTS_POLL_EV_OUT) + state->driver.select->outport = id; + if (mode & ERL_DRV_USE) { + state->flags |= ERTS_EV_FLAG_USED; + } + } + else { /* off */ + if (state->type == ERTS_EV_TYPE_DRV_SEL) { + if (ctl_events & ERTS_POLL_EV_IN) { + abort_tasks(state, ERL_DRV_READ); + state->driver.select->inport = NIL; + } + if (ctl_events & ERTS_POLL_EV_OUT) { + abort_tasks(state, ERL_DRV_WRITE); + state->driver.select->outport = NIL; + } + if (new_events == 0) { + ASSERT(!erts_port_task_is_scheduled(&state->driver.select->intask)); + ASSERT(!erts_port_task_is_scheduled(&state->driver.select->outtask)); + if (old_events != 0) { + remember_removed(state, &pollset); + } + if ((mode & ERL_DRV_USE) || !(state->flags & ERTS_EV_FLAG_USED)) { + state->type = ERTS_EV_TYPE_NONE; + state->flags = 0; + erts_free(ERTS_ALC_T_DRV_SEL_D_STATE, + state->driver.select); + state->driver.select = NULL; + } + /*else keep it, as fd will probably be selected upon again */ + } + } + if ((mode & ERL_DRV_USE_NO_CALLBACK) == ERL_DRV_USE) { + erts_driver_t* drv_ptr = erts_drvport2port(ix)->drv_ptr; + ASSERT(new_events==0); + if (state->remove_cnt == 0 || !wake_poller) { + /* Safe to close fd now as it is not in pollset + or there was no need to eject fd (kernel poll) */ + stop_select_fn = drv_ptr->stop_select; + } + else { + /* Not safe to close fd, postpone stop_select callback. */ + state->type = ERTS_EV_TYPE_STOP_USE; + state->driver.drv_ptr = drv_ptr; + if (drv_ptr->handle) { + erts_ddll_reference_referenced_driver(drv_ptr->handle); + } + } + } + } + } + + ret = 0; + +done:; +#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS + if (state->type == ERTS_EV_TYPE_NONE && state->remove_cnt == 0) { + hash_erase_drv_ev_state(state); + } +#endif +done_unknown: + erts_smp_mtx_unlock(fd_mtx(fd)); + if (stop_select_fn) { + int was_unmasked = erts_block_fpe(); + (*stop_select_fn)(e, NULL); + erts_unblock_fpe(was_unmasked); + } + return ret; +} + +int +ERTS_CIO_EXPORT(driver_event)(ErlDrvPort ix, + ErlDrvEvent e, + ErlDrvEventData event_data) +{ +#if !ERTS_CIO_HAVE_DRV_EVENT + return -1; +#else + ErtsSysFdType fd = (ErtsSysFdType) e; + ErtsPollEvents events; + ErtsPollEvents add_events; + ErtsPollEvents remove_events; + Eterm id = drvport2id(ix); + ErtsDrvEventState *state; + int do_wake = 0; + int ret; + + ERTS_SMP_LC_ASSERT(erts_drvport2port(ix) + && erts_lc_is_port_locked(erts_drvport2port(ix))); + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + if ((unsigned)fd >= (unsigned)erts_smp_atomic_read(&drv_ev_state_len)) { + if (fd < 0) + return -1; + if (fd >= max_fds) { + event_large_fd_error(ix, fd, event_data); + return -1; + } + grow_drv_ev_state(fd); + } +#endif + + erts_smp_mtx_lock(fd_mtx(fd)); + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + state = &drv_ev_state[(int) fd]; +#else + /* Could use hash_new directly, but want to keep the normal case fast */ + state = hash_get_drv_ev_state(fd); + if (state == NULL) { + state = hash_new_drv_ev_state(fd); + } +#endif + + switch (state->type) { + case ERTS_EV_TYPE_DRV_EV: + if (state->driver.event->port == id) break; + /*fall through*/ + case ERTS_EV_TYPE_DRV_SEL: + event_steal(ix, state, event_data); + break; + case ERTS_EV_TYPE_STOP_USE: { + erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); + print_event_op(dsbufp, ix, fd, event_data); + steal_pending_stop_select(dsbufp, ix, state, 0, 1); + break; + } + } + + ASSERT(state->type == ERTS_EV_TYPE_DRV_EV + || state->type == ERTS_EV_TYPE_NONE); + + events = state->events; + + if (!event_data) { + remove_events = events; + add_events = 0; + } + else { + remove_events = ~event_data->events & events; + add_events = ~events & event_data->events; + } + + if (add_events) { + events = ERTS_CIO_POLL_CTL(pollset.ps, state->fd, add_events, 1, &do_wake); + if (events & (ERTS_POLL_EV_ERR|ERTS_POLL_EV_NVAL)) { + ret = -1; + goto done; + } + } + if (remove_events) { + events = ERTS_CIO_POLL_CTL(pollset.ps, state->fd, remove_events, 0, &do_wake); + if (events & (ERTS_POLL_EV_ERR|ERTS_POLL_EV_NVAL)) { + ret = -1; + goto done; + } + } + if (event_data && event_data->events != 0) { + if (state->type == ERTS_EV_TYPE_DRV_EV) { + state->driver.event->removed_events &= ~add_events; + state->driver.event->removed_events |= remove_events; + } + else { + state->driver.event + = erts_alloc(ERTS_ALC_T_DRV_EV_D_STATE, + sizeof(ErtsDrvEventDataState)); + erts_port_task_handle_init(&state->driver.event->task); + state->driver.event->port = id; + state->driver.event->removed_events = (ErtsPollEvents) 0; + state->type = ERTS_EV_TYPE_DRV_EV; + } + state->driver.event->data = event_data; + } + else { + if (state->type == ERTS_EV_TYPE_DRV_EV) { + abort_tasks(state, 0); + erts_free(ERTS_ALC_T_DRV_EV_D_STATE, + state->driver.event); + } + state->driver.select = NULL; + state->type = ERTS_EV_TYPE_NONE; + remember_removed(state, &pollset); + } + state->events = events; + ASSERT(event_data ? events == event_data->events : events == 0); + + ret = 0; + +done: +#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS + if (state->type == ERTS_EV_TYPE_NONE && state->remove_cnt == 0) { + hash_erase_drv_ev_state(state); + } +#endif + erts_smp_mtx_unlock(fd_mtx(fd)); + return ret; +#endif +} + +static ERTS_INLINE int +chk_stale(Eterm id, ErtsDrvEventState *state, int mode) +{ + if (is_nil(id)) + return 0; + if (erts_is_port_alive(id)) + return 1; /* Steal */ + stale_drv_select(id, state, mode); + return 0; +} + +static int +need2steal(ErtsDrvEventState *state, int mode) +{ + int do_steal = 0; + switch (state->type) { + case ERTS_EV_TYPE_DRV_SEL: + if (mode & ERL_DRV_READ) + do_steal |= chk_stale(state->driver.select->inport, + state, + ERL_DRV_READ); + if (mode & ERL_DRV_WRITE) + do_steal |= chk_stale(state->driver.select->outport, + state, + ERL_DRV_WRITE); + break; +#if ERTS_CIO_HAVE_DRV_EVENT + case ERTS_EV_TYPE_DRV_EV: + do_steal |= chk_stale(state->driver.event->port, state, 0); + break; +#endif + case ERTS_EV_TYPE_STOP_USE: + ASSERT(0); + break; + default: + break; + } + return do_steal; +} + +static void +print_driver_name(erts_dsprintf_buf_t *dsbufp, Eterm id) +{ + ErtsPortNames *pnp = erts_get_port_names(id); + if (!pnp->name && !pnp->driver_name) + erts_dsprintf(dsbufp, "%s ", "<unknown>"); + else { + if (pnp->name) { + if (!pnp->driver_name || strcmp(pnp->driver_name, pnp->name) == 0) + erts_dsprintf(dsbufp, "%s ", pnp->name); + else + erts_dsprintf(dsbufp, "%s (%s) ", pnp->driver_name, pnp->name); + } + else if (pnp->driver_name) { + erts_dsprintf(dsbufp, "%s ", pnp->driver_name); + } + } + erts_free_port_names(pnp); +} + +static void +steal(erts_dsprintf_buf_t *dsbufp, ErtsDrvEventState *state, int mode) +{ + erts_dsprintf(dsbufp, "stealing control of fd=%d from ", (int) state->fd); + switch (state->type) { + case ERTS_EV_TYPE_DRV_SEL: { + int deselect_mode = 0; + Eterm iid = state->driver.select->inport; + Eterm oid = state->driver.select->outport; + if ((mode & ERL_DRV_READ) && (is_not_nil(iid))) { + erts_dsprintf(dsbufp, "input driver "); + print_driver_name(dsbufp, iid); + erts_dsprintf(dsbufp, "%T ", iid); + deselect_mode |= ERL_DRV_READ; + } + if ((mode & ERL_DRV_WRITE) && is_not_nil(oid)) { + if (deselect_mode) { + erts_dsprintf(dsbufp, "and "); + } + erts_dsprintf(dsbufp, "output driver "); + print_driver_name(dsbufp, oid); + erts_dsprintf(dsbufp, "%T ", oid); + deselect_mode |= ERL_DRV_WRITE; + } + if (deselect_mode) + deselect(state, deselect_mode); + else { + erts_dsprintf(dsbufp, "no one", (int) state->fd); + ASSERT(0); + } + erts_dsprintf(dsbufp, "\n"); + break; + } +#if ERTS_CIO_HAVE_DRV_EVENT + case ERTS_EV_TYPE_DRV_EV: { + Eterm eid = state->driver.event->port; + if (is_nil(eid)) { + erts_dsprintf(dsbufp, "no one", (int) state->fd); + ASSERT(0); + } + else { + erts_dsprintf(dsbufp, "event driver "); + print_driver_name(dsbufp, eid); + erts_dsprintf(dsbufp, "%T ", eid); + } + erts_dsprintf(dsbufp, "\n"); + deselect(state, 0); + break; + } +#endif + case ERTS_EV_TYPE_STOP_USE: { + ASSERT(0); + break; + } + default: + erts_dsprintf(dsbufp, "no one\n", (int) state->fd); + ASSERT(0); + } +} + +static void +print_select_op(erts_dsprintf_buf_t *dsbufp, + ErlDrvPort ix, ErtsSysFdType fd, int mode, int on) +{ + Port *pp = erts_drvport2port(ix); + erts_dsprintf(dsbufp, + "driver_select(%p, %d,%s%s%s%s, %d) " + "by ", + ix, + (int) fd, + mode & ERL_DRV_READ ? " ERL_DRV_READ" : "", + mode & ERL_DRV_WRITE ? " ERL_DRV_WRITE" : "", + mode & ERL_DRV_USE ? " ERL_DRV_USE" : "", + mode & (ERL_DRV_USE_NO_CALLBACK & ~ERL_DRV_USE) ? "_NO_CALLBACK" : "", + on); + print_driver_name(dsbufp, pp->id); + erts_dsprintf(dsbufp, "driver %T ", pp ? pp->id : NIL); +} + +static void +select_steal(ErlDrvPort ix, ErtsDrvEventState *state, int mode, int on) +{ + if (need2steal(state, mode)) { + erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); + print_select_op(dsbufp, ix, state->fd, mode, on); + steal(dsbufp, state, mode); + erts_send_error_to_logger_nogl(dsbufp); + } +} + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS +static void +large_fd_error_common(erts_dsprintf_buf_t *dsbufp, ErtsSysFdType fd) +{ + erts_dsprintf(dsbufp, + "fd=%d is larger than the largest allowed fd=%d\n", + (int) fd, max_fds - 1); +} + +static void +select_large_fd_error(ErlDrvPort ix, ErtsSysFdType fd, int mode, int on) +{ + erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); + print_select_op(dsbufp, ix, fd, mode, on); + erts_dsprintf(dsbufp, "failed: "); + large_fd_error_common(dsbufp, fd); + erts_send_error_to_logger_nogl(dsbufp); +} +#endif /* ERTS_SYS_CONTINOUS_FD_NUMBERS */ + + + +static void +steal_pending_stop_select(erts_dsprintf_buf_t *dsbufp, ErlDrvPort ix, + ErtsDrvEventState *state, int mode, int on) +{ + ASSERT(state->type == ERTS_EV_TYPE_STOP_USE); + erts_dsprintf(dsbufp, "failed: fd=%d (re)selected before stop_select " + "was called for driver %s\n", + (int) state->fd, state->driver.drv_ptr->name); + erts_send_error_to_logger_nogl(dsbufp); + + if (on) { + /* Either fd-owner changed its mind about closing + * or closed fd before stop_select callback and fd is now reused. + * In either case stop_select should not be called. + */ + state->type = ERTS_EV_TYPE_NONE; + state->flags = 0; + if (state->driver.drv_ptr->handle) { + erts_ddll_dereference_driver(state->driver.drv_ptr->handle); + } + state->driver.drv_ptr = NULL; + } + else if ((mode & ERL_DRV_USE_NO_CALLBACK) == ERL_DRV_USE) { + erts_driver_t* drv_ptr = erts_drvport2port(ix)->drv_ptr; + if (drv_ptr != state->driver.drv_ptr) { + /* Some other driver wants the stop_select callback */ + if (state->driver.drv_ptr->handle) { + erts_ddll_dereference_driver(state->driver.drv_ptr->handle); + } + if (drv_ptr->handle) { + erts_ddll_reference_referenced_driver(drv_ptr->handle); + } + state->driver.drv_ptr = drv_ptr; + } + } + +} + + +#if ERTS_CIO_HAVE_DRV_EVENT + +static void +print_event_op(erts_dsprintf_buf_t *dsbufp, + ErlDrvPort ix, ErtsSysFdType fd, ErlDrvEventData event_data) +{ + Port *pp = erts_drvport2port(ix); + erts_dsprintf(dsbufp, "driver_event(%p, %d, ", ix, (int) fd); + if (!event_data) + erts_dsprintf(dsbufp, "NULL"); + else + erts_dsprintf(dsbufp, "{0x%x, 0x%x}", + (unsigned int) event_data->events, + (unsigned int) event_data->revents); + erts_dsprintf(dsbufp, ") by "); + print_driver_name(dsbufp, pp->id); + erts_dsprintf(dsbufp, "driver %T ", pp ? pp->id : NIL); +} + +static void +event_steal(ErlDrvPort ix, ErtsDrvEventState *state, ErlDrvEventData event_data) +{ + if (need2steal(state, ERL_DRV_READ|ERL_DRV_WRITE)) { + erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); + print_event_op(dsbufp, ix, state->fd, event_data); + steal(dsbufp, state, ERL_DRV_READ|ERL_DRV_WRITE); + erts_send_error_to_logger_nogl(dsbufp); + } + else if (state->type == ERTS_EV_TYPE_DRV_SEL) { + ASSERT(state->flags & ERTS_EV_FLAG_USED); + deselect(state, 0); + } +} + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS +static void +event_large_fd_error(ErlDrvPort ix, ErtsSysFdType fd, ErlDrvEventData event_data) +{ + erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); + print_event_op(dsbufp, ix, fd, event_data); + erts_dsprintf(dsbufp, "failed: "); + large_fd_error_common(dsbufp, fd); + erts_send_error_to_logger_nogl(dsbufp); +} +#endif +#endif + +static ERTS_INLINE void +iready(Eterm id, ErtsDrvEventState *state) +{ + if (erts_port_task_schedule(id, + &state->driver.select->intask, + ERTS_PORT_TASK_INPUT, + (ErlDrvEvent) state->fd, + NULL) != 0) { + stale_drv_select(id, state, ERL_DRV_READ); + } +} + +static ERTS_INLINE void +oready(Eterm id, ErtsDrvEventState *state) +{ + if (erts_port_task_schedule(id, + &state->driver.select->outtask, + ERTS_PORT_TASK_OUTPUT, + (ErlDrvEvent) state->fd, + NULL) != 0) { + stale_drv_select(id, state, ERL_DRV_WRITE); + } +} + +#if ERTS_CIO_HAVE_DRV_EVENT +static ERTS_INLINE void +eready(Eterm id, ErtsDrvEventState *state, ErlDrvEventData event_data) +{ + if (erts_port_task_schedule(id, + &state->driver.event->task, + ERTS_PORT_TASK_EVENT, + (ErlDrvEvent) state->fd, + event_data) != 0) { + stale_drv_select(id, state, 0); + } +} +#endif + +static void bad_fd_in_pollset( ErtsDrvEventState *, Eterm, Eterm, ErtsPollEvents); + +void +ERTS_CIO_EXPORT(erts_check_io_interrupt)(int set) +{ + ERTS_CIO_POLL_INTR(pollset.ps, set); +} + +void +ERTS_CIO_EXPORT(erts_check_io_interrupt_timed)(int set, long msec) +{ + ERTS_CIO_POLL_INTR_TMD(pollset.ps, set, msec); +} + +void +ERTS_CIO_EXPORT(erts_check_io)(int do_wait) +{ + ErtsPollResFd pollres[256]; + int pollres_len; + SysTimeval wait_time; + int poll_ret, i; + + restart: + + /* Figure out timeout value */ + if (do_wait) { + erts_time_remaining(&wait_time); + } else { /* poll only */ + wait_time.tv_sec = 0; + wait_time.tv_usec = 0; + } + +#ifdef ERTS_ENABLE_LOCK_CHECK + erts_lc_check_exact(NULL, 0); /* No locks should be locked */ +#endif + erts_smp_activity_begin(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL); + pollres_len = sizeof(pollres)/sizeof(ErtsPollResFd); + + erts_smp_atomic_set(&pollset.in_poll_wait, 1); + + poll_ret = ERTS_CIO_POLL_WAIT(pollset.ps, pollres, &pollres_len, &wait_time); + +#ifdef ERTS_ENABLE_LOCK_CHECK + erts_lc_check_exact(NULL, 0); /* No locks should be locked */ +#endif + erts_smp_activity_end(ERTS_ACTIVITY_WAIT, NULL, NULL, NULL); + + erts_deliver_time(); /* sync the machine's idea of time */ + +#ifdef ERTS_BREAK_REQUESTED + if (ERTS_BREAK_REQUESTED) + erts_do_break_handling(); +#endif + + if (poll_ret != 0) { + erts_smp_atomic_set(&pollset.in_poll_wait, 0); + forget_removed(&pollset); + if (poll_ret == EAGAIN) { + goto restart; + } + + if (poll_ret != ETIMEDOUT + && poll_ret != EINTR +#ifdef ERRNO_BLOCK + && poll_ret != ERRNO_BLOCK +#endif + ) { + erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); + erts_dsprintf(dsbufp, "erts_poll_wait() failed: %s (%d)\n", + erl_errno_id(poll_ret), poll_ret); + erts_send_error_to_logger_nogl(dsbufp); + } + return; + } + + for (i = 0; i < pollres_len; i++) { + + ErtsSysFdType fd = (ErtsSysFdType) pollres[i].fd; + ErtsDrvEventState *state; + + erts_smp_mtx_lock(fd_mtx(fd)); + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + state = &drv_ev_state[ (int) fd]; +#else + state = hash_get_drv_ev_state(fd); + if (!state) { + goto next_pollres; + } +#endif + + /* Skip this fd if it was removed from pollset */ + if (is_removed(state)) { + goto next_pollres; + } + + switch (state->type) { + case ERTS_EV_TYPE_DRV_SEL: { /* Requested via driver_select()... */ + ErtsPollEvents revents; + ErtsPollEvents revent_mask; + + revent_mask = ~(ERTS_POLL_EV_IN|ERTS_POLL_EV_OUT); + revent_mask |= state->events; + revents = pollres[i].events & revent_mask; + + if (revents & ERTS_POLL_EV_ERR) { + /* + * Let the driver handle the error condition. Only input, + * only output, or nothing might have been selected. + * We *do not* want to call a callback that corresponds + * to an event not selected. revents might give us a clue + * on which one to call. + */ + if ((revents & ERTS_POLL_EV_IN) + || (!(revents & ERTS_POLL_EV_OUT) + && state->events & ERTS_POLL_EV_IN)) { + iready(state->driver.select->inport, state); + } + else if (state->events & ERTS_POLL_EV_OUT) { + oready(state->driver.select->outport, state); + } + } + else if (revents & (ERTS_POLL_EV_IN|ERTS_POLL_EV_OUT)) { + if (revents & ERTS_POLL_EV_OUT) { + oready(state->driver.select->outport, state); + } + /* Someone might have deselected input since revents + was read (true also on the non-smp emulator since + oready() may have been called); therefore, update + revents... */ + revents &= ~(~state->events & ERTS_POLL_EV_IN); + if (revents & ERTS_POLL_EV_IN) { + iready(state->driver.select->inport, state); + } + } + else if (revents & ERTS_POLL_EV_NVAL) { + bad_fd_in_pollset(state, + state->driver.select->inport, + state->driver.select->outport, + state->events); + } + break; + } + +#if ERTS_CIO_HAVE_DRV_EVENT + case ERTS_EV_TYPE_DRV_EV: { /* Requested via driver_event()... */ + ErlDrvEventData event_data; + ErtsPollEvents revents; + ASSERT(state->driver.event); + ASSERT(state->driver.event->data); + event_data = state->driver.event->data; + revents = pollres[i].events; + revents &= ~state->driver.event->removed_events; + + if (revents) { + event_data->events = state->events; + event_data->revents = revents; + + eready(state->driver.event->port, state, event_data); + } + break; + } +#endif + + case ERTS_EV_TYPE_NONE: /* Deselected ... */ + break; + + default: { /* Error */ + erts_dsprintf_buf_t *dsbufp; + dsbufp = erts_create_logger_dsbuf(); + erts_dsprintf(dsbufp, + "Invalid event request type for fd in erts_poll()! " + "fd=%d, event request type=%sd\n", (int) state->fd, + (int) state->type); + ASSERT(0); + deselect(state, 0); + break; + } + } + + next_pollres:; +#ifdef ERTS_SMP + erts_smp_mtx_unlock(fd_mtx(fd)); +#endif + } + + erts_smp_atomic_set(&pollset.in_poll_wait, 0); + forget_removed(&pollset); +} + +static void +bad_fd_in_pollset(ErtsDrvEventState *state, Eterm inport, + Eterm outport, ErtsPollEvents events) +{ + erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf(); + + if (events & (ERTS_POLL_EV_IN|ERTS_POLL_EV_OUT)) { + char *io_str; + Eterm port = NIL; + if ((events & ERTS_POLL_EV_IN) && (events & ERTS_POLL_EV_OUT)) { + io_str = "input/output"; + if (inport == outport) + port = inport; + } + else { + if (events & ERTS_POLL_EV_IN) { + io_str = "input"; + port = inport; + } + else { + io_str = "output"; + port = outport; + } + } + erts_dsprintf(dsbufp, + "Bad %s fd in erts_poll()! fd=%d, ", + io_str, (int) state->fd); + if (is_nil(port)) { + ErtsPortNames *ipnp = erts_get_port_names(inport); + ErtsPortNames *opnp = erts_get_port_names(outport); + erts_dsprintf(dsbufp, "ports=%T/%T, drivers=%s/%s, names=%s/%s\n", + is_nil(inport) ? am_undefined : inport, + is_nil(outport) ? am_undefined : outport, + ipnp->driver_name ? ipnp->driver_name : "<unknown>", + opnp->driver_name ? opnp->driver_name : "<unknown>", + ipnp->name ? ipnp->name : "<unknown>", + opnp->name ? opnp->name : "<unknown>"); + erts_free_port_names(ipnp); + erts_free_port_names(opnp); + } + else { + ErtsPortNames *pnp = erts_get_port_names(port); + erts_dsprintf(dsbufp, "port=%T, driver=%s, name=%s\n", + is_nil(port) ? am_undefined : port, + pnp->driver_name ? pnp->driver_name : "<unknown>", + pnp->name ? pnp->name : "<unknown>"); + erts_free_port_names(pnp); + } + } + else { + erts_dsprintf(dsbufp, "Bad fd in erts_poll()! fd=%d\n", (int) state->fd); + } + erts_send_error_to_logger_nogl(dsbufp); + + /* unmap entry */ + deselect(state, 0); +} + +static void +stale_drv_select(Eterm id, ErtsDrvEventState *state, int mode) +{ + erts_stale_drv_select(id, (ErlDrvEvent) state->fd, mode, 0); + deselect(state, mode); +} + +#ifndef ERTS_SYS_CONTINOUS_FD_NUMBERS +static SafeHashValue drv_ev_state_hash(void *des) +{ + SafeHashValue val = (SafeHashValue) ((ErtsDrvEventState *) des)->fd; + return val ^ (val >> 8); /* Good enough for aligned pointer values? */ +} + +static int drv_ev_state_cmp(void *des1, void *des2) +{ + return ( ((ErtsDrvEventState *) des1)->fd == ((ErtsDrvEventState *) des2)->fd + ? 0 : 1); +} + +static void *drv_ev_state_alloc(void *des_tmpl) +{ + ErtsDrvEventState *evstate; + erts_smp_spin_lock(&state_prealloc_lock); + if (state_prealloc_first == NULL) { + erts_smp_spin_unlock(&state_prealloc_lock); + evstate = (ErtsDrvEventState *) + erts_alloc(ERTS_ALC_T_DRV_EV_STATE, sizeof(ErtsDrvEventState)); + } else { + evstate = state_prealloc_first; + state_prealloc_first = (ErtsDrvEventState *) evstate->hb.next; + --num_state_prealloc; + erts_smp_spin_unlock(&state_prealloc_lock); + } + /* XXX: Already valid data if prealloced, could ignore template! */ + *evstate = *((ErtsDrvEventState *) des_tmpl); + + return (void *) evstate; +} + +static void drv_ev_state_free(void *des) +{ + erts_smp_spin_lock(&state_prealloc_lock); + ((ErtsDrvEventState *) des)->hb.next = &state_prealloc_first->hb; + state_prealloc_first = (ErtsDrvEventState *) des; + ++num_state_prealloc; + erts_smp_spin_unlock(&state_prealloc_lock); +} +#endif + +void +ERTS_CIO_EXPORT(erts_init_check_io)(void) +{ + erts_smp_atomic_init(&pollset.in_poll_wait, 0); + ERTS_CIO_POLL_INIT(); + pollset.ps = ERTS_CIO_NEW_POLLSET(); + +#ifdef ERTS_SMP + init_removed_fd_alloc(); + pollset.removed_list = NULL; + erts_smp_spinlock_init(&pollset.removed_list_lock, + "pollset_rm_list"); + { + int i; + for (i=0; i<DRV_EV_STATE_LOCK_CNT; i++) { +#ifdef ERTS_ENABLE_LOCK_COUNT + erts_smp_mtx_init_x(&drv_ev_state_locks[i].lck, "drv_ev_state", make_small(i)); +#else + erts_smp_mtx_init(&drv_ev_state_locks[i].lck, "drv_ev_state"); +#endif + } + } +#endif +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + max_fds = ERTS_CIO_POLL_MAX_FDS(); + erts_smp_atomic_init(&drv_ev_state_len, 0); + drv_ev_state = NULL; + erts_smp_mtx_init(&drv_ev_state_grow_lock, "drv_ev_state_grow"); +#else + { + SafeHashFunctions hf; + hf.hash = &drv_ev_state_hash; + hf.cmp = &drv_ev_state_cmp; + hf.alloc = &drv_ev_state_alloc; + hf.free = &drv_ev_state_free; + num_state_prealloc = 0; + state_prealloc_first = NULL; + erts_smp_spinlock_init(&state_prealloc_lock,"state_prealloc"); + + safe_hash_init(ERTS_ALC_T_DRV_EV_STATE, &drv_ev_state_tab, "drv_ev_state_tab", + DRV_EV_STATE_HTAB_SIZE, hf); + } +#endif +} + +int +ERTS_CIO_EXPORT(erts_check_io_max_files)(void) +{ +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + return max_fds; +#else + return ERTS_POLL_EXPORT(erts_poll_max_fds)(); +#endif +} + +Uint +ERTS_CIO_EXPORT(erts_check_io_size)(void) +{ + Uint res; + ErtsPollInfo pi; + ERTS_CIO_POLL_INFO(pollset.ps, &pi); + res = pi.memory_size; +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + res += sizeof(ErtsDrvEventState) * erts_smp_atomic_read(&drv_ev_state_len); +#else + res += safe_hash_table_sz(&drv_ev_state_tab); + { + SafeHashInfo hi; + safe_hash_get_info(&hi, &drv_ev_state_tab); + res += hi.objs * sizeof(ErtsDrvEventState); + } + erts_smp_spin_lock(&state_prealloc_lock); + res += num_state_prealloc * sizeof(ErtsDrvEventState); + erts_smp_spin_unlock(&state_prealloc_lock); +#endif + return res; +} + +Eterm +ERTS_CIO_EXPORT(erts_check_io_info)(void *proc) +{ + Process *p = (Process *) proc; + Eterm tags[15], values[15], res; + Uint sz, *szp, *hp, **hpp, memory_size; + Sint i; + ErtsPollInfo pi; + + ERTS_CIO_POLL_INFO(pollset.ps, &pi); + memory_size = pi.memory_size; +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + memory_size += sizeof(ErtsDrvEventState) * erts_smp_atomic_read(&drv_ev_state_len); +#else + memory_size += safe_hash_table_sz(&drv_ev_state_tab); + { + SafeHashInfo hi; + safe_hash_get_info(&hi, &drv_ev_state_tab); + memory_size += hi.objs * sizeof(ErtsDrvEventState); + } + erts_smp_spin_lock(&state_prealloc_lock); + memory_size += num_state_prealloc * sizeof(ErtsDrvEventState); + erts_smp_spin_unlock(&state_prealloc_lock); +#endif + + hpp = NULL; + szp = &sz; + sz = 0; + + bld_it: + i = 0; + + tags[i] = erts_bld_atom(hpp, szp, "name"); + values[i++] = erts_bld_atom(hpp, szp, "erts_poll"); + + tags[i] = erts_bld_atom(hpp, szp, "primary"); + values[i++] = erts_bld_atom(hpp, szp, pi.primary); + + tags[i] = erts_bld_atom(hpp, szp, "fallback"); + values[i++] = erts_bld_atom(hpp, szp, pi.fallback ? pi.fallback : "false"); + + tags[i] = erts_bld_atom(hpp, szp, "kernel_poll"); + values[i++] = erts_bld_atom(hpp, szp, + pi.kernel_poll ? pi.kernel_poll : "false"); + + tags[i] = erts_bld_atom(hpp, szp, "memory_size"); + values[i++] = erts_bld_uint(hpp, szp, memory_size); + + tags[i] = erts_bld_atom(hpp, szp, "total_poll_set_size"); + values[i++] = erts_bld_uint(hpp, szp, (Uint) pi.poll_set_size); + + if (pi.fallback) { + tags[i] = erts_bld_atom(hpp, szp, "fallback_poll_set_size"); + values[i++] = erts_bld_uint(hpp, szp, (Uint) pi.fallback_poll_set_size); + } + + tags[i] = erts_bld_atom(hpp, szp, "lazy_updates"); + values[i++] = pi.lazy_updates ? am_true : am_false; + + if (pi.lazy_updates) { + tags[i] = erts_bld_atom(hpp, szp, "pending_updates"); + values[i++] = erts_bld_uint(hpp, szp, (Uint) pi.pending_updates); + } + + tags[i] = erts_bld_atom(hpp, szp, "batch_updates"); + values[i++] = pi.batch_updates ? am_true : am_false; + + tags[i] = erts_bld_atom(hpp, szp, "concurrent_updates"); + values[i++] = pi.concurrent_updates ? am_true : am_false; + + tags[i] = erts_bld_atom(hpp, szp, "max_fds"); + values[i++] = erts_bld_uint(hpp, szp, (Uint) pi.max_fds); + +#ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS + tags[i] = erts_bld_atom(hpp, szp, "no_avoided_wakeups"); + values[i++] = erts_bld_uint(hpp, szp, (Uint) pi.no_avoided_wakeups); + + tags[i] = erts_bld_atom(hpp, szp, "no_avoided_interrupts"); + values[i++] = erts_bld_uint(hpp, szp, (Uint) pi.no_avoided_interrupts); + + tags[i] = erts_bld_atom(hpp, szp, "no_interrupt_timed"); + values[i++] = erts_bld_uint(hpp, szp, (Uint) pi.no_interrupt_timed); +#endif + + res = erts_bld_2tup_list(hpp, szp, i, tags, values); + + if (!hpp) { + hp = HAlloc(p, sz); + hpp = &hp; + szp = NULL; + goto bld_it; + } + + return res; +} + +static ERTS_INLINE ErtsPollEvents +print_events(ErtsPollEvents ev) +{ + int first = 1; + if(ev & ERTS_POLL_EV_IN) { + ev &= ~ERTS_POLL_EV_IN; + erts_printf("%s%s", first ? "" : "|", "IN"); + first = 0; + } + if(ev & ERTS_POLL_EV_OUT) { + ev &= ~ERTS_POLL_EV_OUT; + erts_printf("%s%s", first ? "" : "|", "OUT"); + first = 0; + } + /* The following should not appear... */ + if(ev & ERTS_POLL_EV_NVAL) { + erts_printf("%s%s", first ? "" : "|", "NVAL"); + first = 0; + } + if(ev & ERTS_POLL_EV_ERR) { + erts_printf("%s%s", first ? "" : "|", "ERR"); + first = 0; + } + if (ev) + erts_printf("%s0x%b32x", first ? "" : "|", (Uint32) ev); + return ev; +} + +typedef struct { + int used_fds; + int num_errors; +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + int internal_fds; + ErtsPollEvents *epep; +#endif +} IterDebugCounters; + +static void doit_erts_check_io_debug(void *vstate, void *vcounters) +{ + ErtsDrvEventState *state = (ErtsDrvEventState *) vstate; + IterDebugCounters *counters = (IterDebugCounters *) vcounters; + ErtsPollEvents cio_events = state->events; + ErtsSysFdType fd = state->fd; +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + int internal = 0; + ErtsPollEvents ep_events = counters->epep[(int) fd]; +#endif + int err = 0; + +#if defined(HAVE_FSTAT) && !defined(NO_FSTAT_ON_SYS_FD_TYPE) + struct stat stat_buf; +#endif + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + if (state->events || ep_events) { + if (ep_events & ERTS_POLL_EV_NVAL) { + ep_events &= ~ERTS_POLL_EV_NVAL; + internal = 1; + counters->internal_fds++; + } + else + counters->used_fds++; +#else + if (state->events) { + counters->used_fds++; +#endif + + erts_printf("fd=%d ", (int) fd); + +#if defined(HAVE_FSTAT) && !defined(NO_FSTAT_ON_SYS_FD_TYPE) + if (fstat((int) fd, &stat_buf) < 0) + erts_printf("type=unknown "); + else { + erts_printf("type="); +#ifdef S_ISSOCK + if (S_ISSOCK(stat_buf.st_mode)) + erts_printf("sock "); + else +#endif +#ifdef S_ISFIFO + if (S_ISFIFO(stat_buf.st_mode)) + erts_printf("fifo "); + else +#endif +#ifdef S_ISCHR + if (S_ISCHR(stat_buf.st_mode)) + erts_printf("chr "); + else +#endif +#ifdef S_ISDIR + if (S_ISDIR(stat_buf.st_mode)) + erts_printf("dir "); + else +#endif +#ifdef S_ISBLK + if (S_ISBLK(stat_buf.st_mode)) + erts_printf("blk "); + else +#endif +#ifdef S_ISREG + if (S_ISREG(stat_buf.st_mode)) + erts_printf("reg "); + else +#endif +#ifdef S_ISLNK + if (S_ISLNK(stat_buf.st_mode)) + erts_printf("lnk "); + else +#endif +#ifdef S_ISDOOR + if (S_ISDOOR(stat_buf.st_mode)) + erts_printf("door "); + else +#endif +#ifdef S_ISWHT + if (S_ISWHT(stat_buf.st_mode)) + erts_printf("wht "); + else +#endif +#ifdef S_ISXATTR + if (S_ISXATTR(stat_buf.st_mode)) + erts_printf("xattr "); + else +#endif + erts_printf("unknown "); + } +#else + erts_printf("type=unknown "); +#endif + + if (state->type == ERTS_EV_TYPE_DRV_SEL) { + erts_printf("driver_select "); + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + if (internal) { + erts_printf("internal "); + err = 1; + } + + if (cio_events == ep_events) { + erts_printf("ev="); + if (print_events(cio_events) != 0) + err = 1; + } + else { + err = 1; + erts_printf("cio_ev="); + print_events(cio_events); + erts_printf(" ep_ev="); + print_events(ep_events); + } +#else + if (print_events(cio_events) != 0) + err = 1; +#endif + erts_printf(" "); + if (cio_events & ERTS_POLL_EV_IN) { + Eterm id = state->driver.select->inport; + if (is_nil(id)) { + erts_printf("inport=none inname=none indrv=none "); + err = 1; + } + else { + ErtsPortNames *pnp = erts_get_port_names(id); + erts_printf(" inport=%T inname=%s indrv=%s ", + id, + pnp->name ? pnp->name : "unknown", + (pnp->driver_name + ? pnp->driver_name + : "unknown")); + erts_free_port_names(pnp); + } + } + if (cio_events & ERTS_POLL_EV_OUT) { + Eterm id = state->driver.select->outport; + if (is_nil(id)) { + erts_printf("outport=none outname=none outdrv=none "); + err = 1; + } + else { + ErtsPortNames *pnp = erts_get_port_names(id); + erts_printf(" outport=%T outname=%s outdrv=%s ", + id, + pnp->name ? pnp->name : "unknown", + (pnp->driver_name + ? pnp->driver_name + : "unknown")); + erts_free_port_names(pnp); + } + } + } + else if (state->type == ERTS_EV_TYPE_DRV_EV) { + Eterm id; + erts_printf("driver_event "); +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + if (internal) { + erts_printf("internal "); + err = 1; + } + if (cio_events == ep_events) { + erts_printf("ev=0x%b32x", (Uint32) cio_events); + } + else { + err = 1; + erts_printf("cio_ev=0x%b32x", (Uint32) cio_events); + erts_printf(" ep_ev=0x%b32x", (Uint32) ep_events); + } +#else + erts_printf("ev=0x%b32x", (Uint32) cio_events); +#endif + id = state->driver.event->port; + if (is_nil(id)) { + erts_printf(" port=none name=none drv=none "); + err = 1; + } + else { + ErtsPortNames *pnp = erts_get_port_names(id); + erts_printf(" port=%T name=%s drv=%s ", + id, + pnp->name ? pnp->name : "unknown", + (pnp->driver_name + ? pnp->driver_name + : "unknown")); + erts_free_port_names(pnp); + } + } +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + else if (internal) { + erts_printf("internal "); + if (cio_events) { + err = 1; + erts_printf("cio_ev="); + print_events(cio_events); + } + if (ep_events) { + erts_printf("ep_ev="); + print_events(ep_events); + } + } +#endif + else { + err = 1; + erts_printf("control_type=%d ", (int)state->type); +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + if (cio_events == ep_events) { + erts_printf("ev=0x%b32x", (Uint32) cio_events); + } + else { + erts_printf("cio_ev=0x%b32x", (Uint32) cio_events); + erts_printf(" ep_ev=0x%b32x", (Uint32) ep_events); + } +#else + erts_printf("ev=0x%b32x", (Uint32) cio_events); +#endif + } + + if (err) { + counters->num_errors++; + erts_printf(" ERROR"); + } + erts_printf("\n"); + } +} + +int +ERTS_CIO_EXPORT(erts_check_io_debug)(void) +{ +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + int fd, len; +#endif + IterDebugCounters counters; + ErtsDrvEventState null_des; + + null_des.driver.select = NULL; + null_des.events = 0; + null_des.remove_cnt = 0; + null_des.type = ERTS_EV_TYPE_NONE; + + erts_printf("--- fds in pollset --------------------------------------\n"); + +#ifdef ERTS_SMP +# ifdef ERTS_ENABLE_LOCK_CHECK + erts_lc_check_exact(NULL, 0); /* No locks should be locked */ +# endif + erts_block_system(0); /* stop the world to avoid messy locking */ +#endif + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + counters.epep = erts_alloc(ERTS_ALC_T_TMP, sizeof(ErtsPollEvents)*max_fds); + ERTS_POLL_EXPORT(erts_poll_get_selected_events)(pollset.ps, counters.epep, max_fds); + counters.internal_fds = 0; +#endif + counters.used_fds = 0; + counters.num_errors = 0; + +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + len = erts_smp_atomic_read(&drv_ev_state_len); + for (fd = 0; fd < len; fd++) { + doit_erts_check_io_debug((void *) &drv_ev_state[fd], (void *) &counters); + } + for ( ; fd < max_fds; fd++) { + null_des.fd = fd; + doit_erts_check_io_debug((void *) &null_des, (void *) &counters); + } +#else + safe_hash_for_each(&drv_ev_state_tab, &doit_erts_check_io_debug, (void *) &counters); +#endif + +#ifdef ERTS_SMP + erts_release_system(); +#endif + + erts_printf("\n"); + erts_printf("used fds=%d\n", counters.used_fds); +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + erts_printf("internal fds=%d\n", counters.internal_fds); +#endif + erts_printf("---------------------------------------------------------\n"); + fflush(stdout); +#ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS + erts_free(ERTS_ALC_T_TMP, (void *) counters.epep); +#endif + return counters.num_errors; +} + diff --git a/erts/emulator/sys/common/erl_check_io.h b/erts/emulator/sys/common/erl_check_io.h new file mode 100644 index 0000000000..9b45a63913 --- /dev/null +++ b/erts/emulator/sys/common/erl_check_io.h @@ -0,0 +1,96 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2006-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +/* + * Description: Check I/O + * + * Author: Rickard Green + */ + +#ifndef ERL_CHECK_IO_H__ +#define ERL_CHECK_IO_H__ + +#include "erl_sys_driver.h" + +#ifdef ERTS_ENABLE_KERNEL_POLL + +int driver_select_kp(ErlDrvPort, ErlDrvEvent, int, int); +int driver_select_nkp(ErlDrvPort, ErlDrvEvent, int, int); +int driver_event_kp(ErlDrvPort, ErlDrvEvent, ErlDrvEventData); +int driver_event_nkp(ErlDrvPort, ErlDrvEvent, ErlDrvEventData); +Uint erts_check_io_size_kp(void); +Uint erts_check_io_size_nkp(void); +Eterm erts_check_io_info_kp(void *); +Eterm erts_check_io_info_nkp(void *); +int erts_check_io_max_files_kp(void); +int erts_check_io_max_files_nkp(void); +void erts_check_io_interrupt_kp(int); +void erts_check_io_interrupt_nkp(int); +void erts_check_io_interrupt_timed_kp(int, long); +void erts_check_io_interrupt_timed_nkp(int, long); +void erts_check_io_kp(int); +void erts_check_io_nkp(int); +void erts_init_check_io_kp(void); +void erts_init_check_io_nkp(void); +int erts_check_io_debug_kp(void); +int erts_check_io_debug_nkp(void); + +#else /* !ERTS_ENABLE_KERNEL_POLL */ + +Uint erts_check_io_size(void); +Eterm erts_check_io_info(void *); +int erts_check_io_max_files(void); +void erts_check_io_interrupt(int); +void erts_check_io_interrupt_timed(int, long); +void erts_check_io(int); +void erts_init_check_io(void); + +#endif + +#endif /* ERL_CHECK_IO_H__ */ + +#if !defined(ERL_CHECK_IO_C__) && !defined(ERTS_ALLOC_C__) +#define ERL_CHECK_IO_INTERNAL__ +#endif + +#ifndef ERL_CHECK_IO_INTERNAL__ +#define ERL_CHECK_IO_INTERNAL__ +#include "erl_poll.h" +#include "erl_port_task.h" + +/* + * ErtsDrvEventDataState is used by driver_event() which is almost never + * used. We allocate ErtsDrvEventDataState separate since we dont wan't + * the size of ErtsDrvEventState to increase due to driver_event() + * information. + */ +typedef struct { + Eterm port; + ErlDrvEventData data; + ErtsPollEvents removed_events; + ErtsPortTaskHandle task; +} ErtsDrvEventDataState; + +typedef struct { + Eterm inport; + Eterm outport; + ErtsPortTaskHandle intask; + ErtsPortTaskHandle outtask; +} ErtsDrvSelectDataState; +#endif /* #ifndef ERL_CHECK_IO_INTERNAL__ */ diff --git a/erts/emulator/sys/common/erl_mseg.c b/erts/emulator/sys/common/erl_mseg.c new file mode 100644 index 0000000000..f4e21bc05f --- /dev/null +++ b/erts/emulator/sys/common/erl_mseg.c @@ -0,0 +1,1452 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2002-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +/* + * Description: A memory segment allocator. Segments that are deallocated + * are kept for a while in a segment "cache" before they are + * destroyed. When segments are allocated, cached segments + * are used if possible instead of creating new segments. + * + * Author: Rickard Green + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include "sys.h" +#include "erl_mseg.h" +#include "global.h" +#include "erl_threads.h" +#include "erl_mtrace.h" +#include "big.h" + +#if HAVE_ERTS_MSEG + +#if defined(USE_THREADS) && !defined(ERTS_SMP) +# define ERTS_THREADS_NO_SMP +#endif + +#define SEGTYPE ERTS_MTRACE_SEGMENT_ID + +#ifndef HAVE_GETPAGESIZE +#define HAVE_GETPAGESIZE 0 +#endif + +#ifdef _SC_PAGESIZE +# define GET_PAGE_SIZE sysconf(_SC_PAGESIZE) +#elif HAVE_GETPAGESIZE +# define GET_PAGE_SIZE getpagesize() +#else +# error "Page size unknown" + /* Implement some other way to get the real page size if needed! */ +#endif + +#define MAX_CACHE_SIZE 30 + +#undef MIN +#define MIN(X, Y) ((X) < (Y) ? (X) : (Y)) +#undef MAX +#define MAX(X, Y) ((X) > (Y) ? (X) : (Y)) + +#undef PAGE_MASK +#define INV_PAGE_MASK ((Uint) (page_size - 1)) +#define PAGE_MASK (~INV_PAGE_MASK) +#define PAGE_FLOOR(X) ((X) & PAGE_MASK) +#define PAGE_CEILING(X) PAGE_FLOOR((X) + INV_PAGE_MASK) +#define PAGES(X) ((X) >> page_shift) + +static int atoms_initialized; + +static Uint cache_check_interval; + +static void check_cache(void *unused); +static void mseg_clear_cache(void); +static int is_cache_check_scheduled; +#ifdef ERTS_THREADS_NO_SMP +static int is_cache_check_requested; +#endif + +#if HAVE_MMAP +/* Mmap ... */ + +#define MMAP_PROT (PROT_READ|PROT_WRITE) +#ifdef MAP_ANON +# define MMAP_FLAGS (MAP_ANON|MAP_PRIVATE) +# define MMAP_FD (-1) +#else +# define MMAP_FLAGS (MAP_PRIVATE) +# define MMAP_FD mmap_fd +static int mmap_fd; +#endif + +#if HAVE_MREMAP +# define HAVE_MSEG_RECREATE 1 +#else +# define HAVE_MSEG_RECREATE 0 +#endif + +#define CAN_PARTLY_DESTROY 1 +#else /* #if HAVE_MMAP */ +#define CAN_PARTLY_DESTROY 0 +#error "Not supported" +#endif /* #if HAVE_MMAP */ + + +#if defined(ERTS_MSEG_FAKE_SEGMENTS) +#undef CAN_PARTLY_DESTROY +#define CAN_PARTLY_DESTROY 0 +#endif + +static const ErtsMsegOpt_t default_opt = ERTS_MSEG_DEFAULT_OPT_INITIALIZER; + +typedef struct cache_desc_t_ { + void *seg; + Uint size; + struct cache_desc_t_ *next; + struct cache_desc_t_ *prev; +} cache_desc_t; + +typedef struct { + Uint32 giga_no; + Uint32 no; +} CallCounter; + +static int is_init_done; +static Uint page_size; +static Uint page_shift; + +static struct { + CallCounter alloc; + CallCounter dealloc; + CallCounter realloc; + CallCounter create; + CallCounter destroy; +#if HAVE_MSEG_RECREATE + CallCounter recreate; +#endif + CallCounter clear_cache; + CallCounter check_cache; +} calls; + +static cache_desc_t cache_descs[MAX_CACHE_SIZE]; +static cache_desc_t *free_cache_descs; +static cache_desc_t *cache; +static cache_desc_t *cache_end; +static Uint cache_hits; +static Uint cache_size; +static Uint min_cached_seg_size; +static Uint max_cached_seg_size; + +static Uint max_cache_size; +static Uint abs_max_cache_bad_fit; +static Uint rel_max_cache_bad_fit; + +#if CAN_PARTLY_DESTROY +static Uint min_seg_size; +#endif + +struct { + struct { + Uint watermark; + Uint no; + Uint sz; + } current; + struct { + Uint no; + Uint sz; + } max; + struct { + Uint no; + Uint sz; + } max_ever; +} segments; + +#define ERTS_MSEG_ALLOC_STAT(SZ) \ +do { \ + segments.current.no++; \ + if (segments.max.no < segments.current.no) \ + segments.max.no = segments.current.no; \ + if (segments.current.watermark < segments.current.no) \ + segments.current.watermark = segments.current.no; \ + segments.current.sz += (SZ); \ + if (segments.max.sz < segments.current.sz) \ + segments.max.sz = segments.current.sz; \ +} while (0) + +#define ERTS_MSEG_DEALLOC_STAT(SZ) \ +do { \ + ASSERT(segments.current.no > 0); \ + segments.current.no--; \ + ASSERT(segments.current.sz >= (SZ)); \ + segments.current.sz -= (SZ); \ +} while (0) + +#define ERTS_MSEG_REALLOC_STAT(OSZ, NSZ) \ +do { \ + ASSERT(segments.current.sz >= (OSZ)); \ + segments.current.sz -= (OSZ); \ + segments.current.sz += (NSZ); \ +} while (0) + +#define ONE_GIGA (1000000000) + +#define ZERO_CC(CC) (calls.CC.no = 0, calls.CC.giga_no = 0) + +#define INC_CC(CC) (calls.CC.no == ONE_GIGA - 1 \ + ? (calls.CC.giga_no++, calls.CC.no = 0) \ + : calls.CC.no++) + +#define DEC_CC(CC) (calls.CC.no == 0 \ + ? (calls.CC.giga_no--, \ + calls.CC.no = ONE_GIGA - 1) \ + : calls.CC.no--) + + +static erts_mtx_t mseg_mutex; /* Also needed when !USE_THREADS */ +static erts_mtx_t init_atoms_mutex; /* Also needed when !USE_THREADS */ + +#ifdef USE_THREADS +#ifdef ERTS_THREADS_NO_SMP +static erts_tid_t main_tid; +static int async_handle = -1; +#endif + +static void thread_safe_init(void) +{ + erts_mtx_init(&init_atoms_mutex, "mseg_init_atoms"); + erts_mtx_init(&mseg_mutex, "mseg"); +#ifdef ERTS_THREADS_NO_SMP + main_tid = erts_thr_self(); +#endif +} + +#endif + +static ErlTimer cache_check_timer; + +static ERTS_INLINE void +schedule_cache_check(void) +{ + if (!is_cache_check_scheduled && is_init_done) { +#ifdef ERTS_THREADS_NO_SMP + if (!erts_equal_tids(erts_thr_self(), main_tid)) { + if (!is_cache_check_requested) { + is_cache_check_requested = 1; + sys_async_ready(async_handle); + } + } + else +#endif + { + cache_check_timer.active = 0; + erl_set_timer(&cache_check_timer, + check_cache, + NULL, + NULL, + cache_check_interval); + is_cache_check_scheduled = 1; +#ifdef ERTS_THREADS_NO_SMP + is_cache_check_requested = 0; +#endif + } + } +} + +#ifdef ERTS_THREADS_NO_SMP + +static void +check_schedule_cache_check(void) +{ + erts_mtx_lock(&mseg_mutex); + if (is_cache_check_requested + && !is_cache_check_scheduled) { + schedule_cache_check(); + } + erts_mtx_unlock(&mseg_mutex); +} + +#endif + +static void +mseg_shutdown(void) +{ +#ifdef ERTS_SMP + erts_mtx_lock(&mseg_mutex); +#endif + mseg_clear_cache(); +#ifdef ERTS_SMP + erts_mtx_unlock(&mseg_mutex); +#endif +} + +static ERTS_INLINE void * +mseg_create(Uint size) +{ + void *seg; + + ASSERT(size % page_size == 0); + +#if defined(ERTS_MSEG_FAKE_SEGMENTS) + seg = erts_sys_alloc(ERTS_ALC_N_INVALID, NULL, size); +#elif HAVE_MMAP + seg = (void *) mmap((void *) 0, (size_t) size, + MMAP_PROT, MMAP_FLAGS, MMAP_FD, 0); + if (seg == (void *) MAP_FAILED) + seg = NULL; +#else +#error "Missing mseg_create() implementation" +#endif + + INC_CC(create); + + return seg; +} + +static ERTS_INLINE void +mseg_destroy(void *seg, Uint size) +{ +#if defined(ERTS_MSEG_FAKE_SEGMENTS) + erts_sys_free(ERTS_ALC_N_INVALID, NULL, seg); +#elif HAVE_MMAP + +#ifdef DEBUG + int res = +#endif + + munmap((void *) seg, size); + + ASSERT(size % page_size == 0); + ASSERT(res == 0); +#else +#error "Missing mseg_destroy() implementation" +#endif + + INC_CC(destroy); + +} + +#if HAVE_MSEG_RECREATE + +static ERTS_INLINE void * +mseg_recreate(void *old_seg, Uint old_size, Uint new_size) +{ + void *new_seg; + + ASSERT(old_size % page_size == 0); + ASSERT(new_size % page_size == 0); + +#if defined(ERTS_MSEG_FAKE_SEGMENTS) + new_seg = erts_sys_realloc(ERTS_ALC_N_INVALID, NULL, old_seg, new_size); +#elif HAVE_MREMAP + new_seg = (void *) mremap((void *) old_seg, + (size_t) old_size, + (size_t) new_size, + MREMAP_MAYMOVE); + if (new_seg == (void *) MAP_FAILED) + new_seg = NULL; +#else +#error "Missing mseg_recreate() implementation" +#endif + + INC_CC(recreate); + + return new_seg; +} + +#endif /* #if HAVE_MSEG_RECREATE */ + + +static ERTS_INLINE cache_desc_t * +alloc_cd(void) +{ + cache_desc_t *cd = free_cache_descs; + if (cd) + free_cache_descs = cd->next; + return cd; +} + +static ERTS_INLINE void +free_cd(cache_desc_t *cd) +{ + cd->next = free_cache_descs; + free_cache_descs = cd; +} + + +static ERTS_INLINE void +link_cd(cache_desc_t *cd) +{ + if (cache) + cache->prev = cd; + cd->next = cache; + cd->prev = NULL; + cache = cd; + + if (!cache_end) { + ASSERT(!cd->next); + cache_end = cd; + } + + cache_size++; +} + +static ERTS_INLINE void +end_link_cd(cache_desc_t *cd) +{ + if (cache_end) + cache_end->next = cd; + cd->next = NULL; + cd->prev = cache_end; + cache_end = cd; + + if (!cache) { + ASSERT(!cd->prev); + cache = cd; + } + + cache_size++; +} + +static ERTS_INLINE void +unlink_cd(cache_desc_t *cd) +{ + + if (cd->next) + cd->next->prev = cd->prev; + else + cache_end = cd->prev; + + if (cd->prev) + cd->prev->next = cd->next; + else + cache = cd->next; + ASSERT(cache_size > 0); + cache_size--; +} + +static ERTS_INLINE void +check_cache_limits(void) +{ + cache_desc_t *cd; + max_cached_seg_size = 0; + min_cached_seg_size = ~((Uint) 0); + for (cd = cache; cd; cd = cd->next) { + if (cd->size < min_cached_seg_size) + min_cached_seg_size = cd->size; + if (cd->size > max_cached_seg_size) + max_cached_seg_size = cd->size; + } + +} + +static ERTS_INLINE void +adjust_cache_size(int force_check_limits) +{ + cache_desc_t *cd; + int check_limits = force_check_limits; + Sint max_cached = ((Sint) segments.current.watermark + - (Sint) segments.current.no); + + while (((Sint) cache_size) > max_cached && ((Sint) cache_size) > 0) { + ASSERT(cache_end); + cd = cache_end; + if (!check_limits && + !(min_cached_seg_size < cd->size + && cd->size < max_cached_seg_size)) { + check_limits = 1; + } + if (erts_mtrace_enabled) + erts_mtrace_crr_free(SEGTYPE, SEGTYPE, cd->seg); + mseg_destroy(cd->seg, cd->size); + unlink_cd(cd); + free_cd(cd); + } + + if (check_limits) + check_cache_limits(); + +} + +static void +check_cache(void *unused) +{ +#ifdef ERTS_SMP + erts_mtx_lock(&mseg_mutex); +#endif + + is_cache_check_scheduled = 0; + + if (segments.current.watermark > segments.current.no) + segments.current.watermark--; + adjust_cache_size(0); + + if (cache_size) + schedule_cache_check(); + + INC_CC(check_cache); + +#ifdef ERTS_SMP + erts_mtx_unlock(&mseg_mutex); +#endif + +} + +static void +mseg_clear_cache(void) +{ + segments.current.watermark = 0; + + adjust_cache_size(1); + + ASSERT(!cache); + ASSERT(!cache_end); + ASSERT(!cache_size); + + segments.current.watermark = segments.current.no; + + INC_CC(clear_cache); +} + +static void * +mseg_alloc(ErtsAlcType_t atype, Uint *size_p, const ErtsMsegOpt_t *opt) +{ + + Uint max, min, diff_size, size; + cache_desc_t *cd, *cand_cd; + void *seg; + + INC_CC(alloc); + + size = PAGE_CEILING(*size_p); + +#if CAN_PARTLY_DESTROY + if (size < min_seg_size) + min_seg_size = size; +#endif + + if (!opt->cache) { + create_seg: + adjust_cache_size(0); + seg = mseg_create(size); + if (!seg) { + mseg_clear_cache(); + seg = mseg_create(size); + if (!seg) + size = 0; + } + + *size_p = size; + if (seg) { + if (erts_mtrace_enabled) + erts_mtrace_crr_alloc(seg, atype, ERTS_MTRACE_SEGMENT_ID, size); + ERTS_MSEG_ALLOC_STAT(size); + } + return seg; + } + + if (size > max_cached_seg_size) + goto create_seg; + + if (size < min_cached_seg_size) { + + diff_size = min_cached_seg_size - size; + + if (diff_size > abs_max_cache_bad_fit) + goto create_seg; + + if (100*PAGES(diff_size) > rel_max_cache_bad_fit*PAGES(size)) + goto create_seg; + + } + + max = 0; + min = ~((Uint) 0); + cand_cd = NULL; + + for (cd = cache; cd; cd = cd->next) { + if (cd->size >= size) { + if (!cand_cd) { + cand_cd = cd; + continue; + } + else if (cd->size < cand_cd->size) { + if (max < cand_cd->size) + max = cand_cd->size; + if (min > cand_cd->size) + min = cand_cd->size; + cand_cd = cd; + continue; + } + } + if (max < cd->size) + max = cd->size; + if (min > cd->size) + min = cd->size; + } + + min_cached_seg_size = min; + max_cached_seg_size = max; + + if (!cand_cd) + goto create_seg; + + diff_size = cand_cd->size - size; + + if (diff_size > abs_max_cache_bad_fit + || 100*PAGES(diff_size) > rel_max_cache_bad_fit*PAGES(size)) { + if (max_cached_seg_size < cand_cd->size) + max_cached_seg_size = cand_cd->size; + if (min_cached_seg_size > cand_cd->size) + min_cached_seg_size = cand_cd->size; + goto create_seg; + } + + cache_hits++; + + size = cand_cd->size; + seg = cand_cd->seg; + + unlink_cd(cand_cd); + free_cd(cand_cd); + + *size_p = size; + + if (erts_mtrace_enabled) { + erts_mtrace_crr_free(SEGTYPE, SEGTYPE, seg); + erts_mtrace_crr_alloc(seg, atype, SEGTYPE, size); + } + + if (seg) + ERTS_MSEG_ALLOC_STAT(size); + return seg; +} + + +static void +mseg_dealloc(ErtsAlcType_t atype, void *seg, Uint size, + const ErtsMsegOpt_t *opt) +{ + cache_desc_t *cd; + + ERTS_MSEG_DEALLOC_STAT(size); + + if (!opt->cache || max_cache_size == 0) { + if (erts_mtrace_enabled) + erts_mtrace_crr_free(atype, SEGTYPE, seg); + mseg_destroy(seg, size); + } + else { + int check_limits = 0; + + if (size < min_cached_seg_size) + min_cached_seg_size = size; + if (size > max_cached_seg_size) + max_cached_seg_size = size; + + if (!free_cache_descs) { + cd = cache_end; + if (!(min_cached_seg_size < cd->size + && cd->size < max_cached_seg_size)) { + check_limits = 1; + } + if (erts_mtrace_enabled) + erts_mtrace_crr_free(SEGTYPE, SEGTYPE, cd->seg); + mseg_destroy(cd->seg, cd->size); + unlink_cd(cd); + free_cd(cd); + } + + cd = alloc_cd(); + ASSERT(cd); + cd->seg = seg; + cd->size = size; + link_cd(cd); + + if (erts_mtrace_enabled) { + erts_mtrace_crr_free(atype, SEGTYPE, seg); + erts_mtrace_crr_alloc(seg, SEGTYPE, SEGTYPE, size); + } + + /* ASSERT(segments.current.watermark >= segments.current.no + cache_size); */ + + if (check_limits) + check_cache_limits(); + + schedule_cache_check(); + + } + + INC_CC(dealloc); +} + +static void * +mseg_realloc(ErtsAlcType_t atype, void *seg, Uint old_size, Uint *new_size_p, + const ErtsMsegOpt_t *opt) +{ + void *new_seg; + Uint new_size; + + if (!seg || !old_size) { + new_seg = mseg_alloc(atype, new_size_p, opt); + DEC_CC(alloc); + return new_seg; + } + + if (!(*new_size_p)) { + mseg_dealloc(atype, seg, old_size, opt); + DEC_CC(dealloc); + return NULL; + } + + new_seg = seg; + new_size = PAGE_CEILING(*new_size_p); + + if (new_size == old_size) + ; + else if (new_size < old_size) { + Uint shrink_sz = old_size - new_size; + +#if CAN_PARTLY_DESTROY + if (new_size < min_seg_size) + min_seg_size = new_size; +#endif + + if (shrink_sz < opt->abs_shrink_th + && 100*PAGES(shrink_sz) < opt->rel_shrink_th*PAGES(old_size)) { + new_size = old_size; + } + else { + +#if CAN_PARTLY_DESTROY + + if (shrink_sz > min_seg_size + && free_cache_descs + && opt->cache) { + cache_desc_t *cd; + + cd = alloc_cd(); + ASSERT(cd); + cd->seg = ((char *) seg) + new_size; + cd->size = shrink_sz; + end_link_cd(cd); + + if (erts_mtrace_enabled) { + erts_mtrace_crr_realloc(new_seg, + atype, + SEGTYPE, + seg, + new_size); + erts_mtrace_crr_alloc(cd->seg, SEGTYPE, SEGTYPE, cd->size); + } + schedule_cache_check(); + } + else { + if (erts_mtrace_enabled) + erts_mtrace_crr_realloc(new_seg, + atype, + SEGTYPE, + seg, + new_size); + mseg_destroy(((char *) seg) + new_size, shrink_sz); + } + +#elif HAVE_MSEG_RECREATE + + goto do_recreate; + +#else + + new_seg = mseg_alloc(atype, &new_size, opt); + if (!new_seg) + new_size = old_size; + else { + sys_memcpy(((char *) new_seg), + ((char *) seg), + MIN(new_size, old_size)); + mseg_dealloc(atype, seg, old_size, opt); + } + +#endif + + } + } + else { + + if (!opt->preserv) { + mseg_dealloc(atype, seg, old_size, opt); + new_seg = mseg_alloc(atype, &new_size, opt); + } + else { +#if HAVE_MSEG_RECREATE +#if !CAN_PARTLY_DESTROY + do_recreate: +#endif + new_seg = mseg_recreate((void *) seg, old_size, new_size); + if (erts_mtrace_enabled) + erts_mtrace_crr_realloc(new_seg, atype, SEGTYPE, seg, new_size); + if (!new_seg) + new_size = old_size; +#else + new_seg = mseg_alloc(atype, &new_size, opt); + if (!new_seg) + new_size = old_size; + else { + sys_memcpy(((char *) new_seg), + ((char *) seg), + MIN(new_size, old_size)); + mseg_dealloc(atype, seg, old_size, opt); + } +#endif + } + } + + INC_CC(realloc); + + *new_size_p = new_size; + + ERTS_MSEG_REALLOC_STAT(old_size, new_size); + + return new_seg; +} + +/* --- Info stuff ---------------------------------------------------------- */ + +static struct { + Eterm version; + + Eterm options; + Eterm amcbf; + Eterm rmcbf; + Eterm mcs; + Eterm cci; + + Eterm status; + Eterm cached_segments; + Eterm cache_hits; + Eterm segments; + Eterm segments_size; + Eterm segments_watermark; + + + Eterm calls; + Eterm mseg_alloc; + Eterm mseg_dealloc; + Eterm mseg_realloc; + Eterm mseg_create; + Eterm mseg_destroy; +#if HAVE_MSEG_RECREATE + Eterm mseg_recreate; +#endif + Eterm mseg_clear_cache; + Eterm mseg_check_cache; + +#ifdef DEBUG + Eterm end_of_atoms; +#endif +} am; + +static void ERTS_INLINE atom_init(Eterm *atom, char *name) +{ + *atom = am_atom_put(name, strlen(name)); +} +#define AM_INIT(AM) atom_init(&am.AM, #AM) + +static void +init_atoms(void) +{ +#ifdef DEBUG + Eterm *atom; +#endif + + erts_mtx_unlock(&mseg_mutex); + erts_mtx_lock(&init_atoms_mutex); + + if (!atoms_initialized) { +#ifdef DEBUG + for (atom = (Eterm *) &am; atom <= &am.end_of_atoms; atom++) { + *atom = THE_NON_VALUE; + } +#endif + + AM_INIT(version); + + AM_INIT(options); + AM_INIT(amcbf); + AM_INIT(rmcbf); + AM_INIT(mcs); + AM_INIT(cci); + + AM_INIT(status); + AM_INIT(cached_segments); + AM_INIT(cache_hits); + AM_INIT(segments); + AM_INIT(segments_size); + AM_INIT(segments_watermark); + + AM_INIT(calls); + AM_INIT(mseg_alloc); + AM_INIT(mseg_dealloc); + AM_INIT(mseg_realloc); + AM_INIT(mseg_create); + AM_INIT(mseg_destroy); +#if HAVE_MSEG_RECREATE + AM_INIT(mseg_recreate); +#endif + AM_INIT(mseg_clear_cache); + AM_INIT(mseg_check_cache); + +#ifdef DEBUG + for (atom = (Eterm *) &am; atom < &am.end_of_atoms; atom++) { + ASSERT(*atom != THE_NON_VALUE); + } +#endif + } + + erts_mtx_lock(&mseg_mutex); + atoms_initialized = 1; + erts_mtx_unlock(&init_atoms_mutex); +} + + +#define bld_uint erts_bld_uint +#define bld_cons erts_bld_cons +#define bld_tuple erts_bld_tuple +#define bld_string erts_bld_string +#define bld_2tup_list erts_bld_2tup_list + + +/* + * bld_unstable_uint() (instead of bld_uint()) is used when values may + * change between size check and actual build. This because a value + * that would fit a small when size check is done may need to be built + * as a big when the actual build is performed. Caller is required to + * HRelease after build. + */ +static ERTS_INLINE Eterm +bld_unstable_uint(Uint **hpp, Uint *szp, Uint ui) +{ + Eterm res = THE_NON_VALUE; + if (szp) + *szp += BIG_UINT_HEAP_SIZE; + if (hpp) { + if (IS_USMALL(0, ui)) + res = make_small(ui); + else { + res = uint_to_big(ui, *hpp); + *hpp += BIG_UINT_HEAP_SIZE; + } + } + return res; +} + +static ERTS_INLINE void +add_2tup(Uint **hpp, Uint *szp, Eterm *lp, Eterm el1, Eterm el2) +{ + *lp = bld_cons(hpp, szp, bld_tuple(hpp, szp, 2, el1, el2), *lp); +} + +static ERTS_INLINE void +add_3tup(Uint **hpp, Uint *szp, Eterm *lp, Eterm el1, Eterm el2, Eterm el3) +{ + *lp = bld_cons(hpp, szp, bld_tuple(hpp, szp, 3, el1, el2, el3), *lp); +} + +static ERTS_INLINE void +add_4tup(Uint **hpp, Uint *szp, Eterm *lp, + Eterm el1, Eterm el2, Eterm el3, Eterm el4) +{ + *lp = bld_cons(hpp, szp, bld_tuple(hpp, szp, 4, el1, el2, el3, el4), *lp); +} + +static Eterm +info_options(char *prefix, + int *print_to_p, + void *print_to_arg, + Uint **hpp, + Uint *szp) +{ + Eterm res = THE_NON_VALUE; + + if (print_to_p) { + int to = *print_to_p; + void *arg = print_to_arg; + erts_print(to, arg, "%samcbf: %bpu\n", prefix, abs_max_cache_bad_fit); + erts_print(to, arg, "%srmcbf: %bpu\n", prefix, rel_max_cache_bad_fit); + erts_print(to, arg, "%smcs: %bpu\n", prefix, max_cache_size); + erts_print(to, arg, "%scci: %bpu\n", prefix, cache_check_interval); + } + + if (hpp || szp) { + + if (!atoms_initialized) + init_atoms(); + + res = NIL; + add_2tup(hpp, szp, &res, + am.cci, + bld_uint(hpp, szp, cache_check_interval)); + add_2tup(hpp, szp, &res, + am.mcs, + bld_uint(hpp, szp, max_cache_size)); + add_2tup(hpp, szp, &res, + am.rmcbf, + bld_uint(hpp, szp, rel_max_cache_bad_fit)); + add_2tup(hpp, szp, &res, + am.amcbf, + bld_uint(hpp, szp, abs_max_cache_bad_fit)); + + } + + return res; +} + +static Eterm +info_calls(int *print_to_p, void *print_to_arg, Uint **hpp, Uint *szp) +{ + Eterm res = THE_NON_VALUE; + + if (print_to_p) { + +#define PRINT_CC(TO, TOA, CC) \ + if (calls.CC.giga_no == 0) \ + erts_print(TO, TOA, "mseg_%s calls: %bpu\n", #CC, calls.CC.no); \ + else \ + erts_print(TO, TOA, "mseg_%s calls: %bpu%09bpu\n", #CC, \ + calls.CC.giga_no, calls.CC.no) + + int to = *print_to_p; + void *arg = print_to_arg; + + PRINT_CC(to, arg, alloc); + PRINT_CC(to, arg, dealloc); + PRINT_CC(to, arg, realloc); + PRINT_CC(to, arg, create); + PRINT_CC(to, arg, destroy); +#if HAVE_MSEG_RECREATE + PRINT_CC(to, arg, recreate); +#endif + PRINT_CC(to, arg, clear_cache); + PRINT_CC(to, arg, check_cache); + +#undef PRINT_CC + + } + + if (hpp || szp) { + + res = NIL; + + add_3tup(hpp, szp, &res, + am.mseg_check_cache, + bld_unstable_uint(hpp, szp, calls.check_cache.giga_no), + bld_unstable_uint(hpp, szp, calls.check_cache.no)); + add_3tup(hpp, szp, &res, + am.mseg_clear_cache, + bld_unstable_uint(hpp, szp, calls.clear_cache.giga_no), + bld_unstable_uint(hpp, szp, calls.clear_cache.no)); + +#if HAVE_MSEG_RECREATE + add_3tup(hpp, szp, &res, + am.mseg_recreate, + bld_unstable_uint(hpp, szp, calls.recreate.giga_no), + bld_unstable_uint(hpp, szp, calls.recreate.no)); +#endif + add_3tup(hpp, szp, &res, + am.mseg_destroy, + bld_unstable_uint(hpp, szp, calls.destroy.giga_no), + bld_unstable_uint(hpp, szp, calls.destroy.no)); + add_3tup(hpp, szp, &res, + am.mseg_create, + bld_unstable_uint(hpp, szp, calls.create.giga_no), + bld_unstable_uint(hpp, szp, calls.create.no)); + + + add_3tup(hpp, szp, &res, + am.mseg_realloc, + bld_unstable_uint(hpp, szp, calls.realloc.giga_no), + bld_unstable_uint(hpp, szp, calls.realloc.no)); + add_3tup(hpp, szp, &res, + am.mseg_dealloc, + bld_unstable_uint(hpp, szp, calls.dealloc.giga_no), + bld_unstable_uint(hpp, szp, calls.dealloc.no)); + add_3tup(hpp, szp, &res, + am.mseg_alloc, + bld_unstable_uint(hpp, szp, calls.alloc.giga_no), + bld_unstable_uint(hpp, szp, calls.alloc.no)); + } + + return res; +} + +static Eterm +info_status(int *print_to_p, + void *print_to_arg, + int begin_new_max_period, + Uint **hpp, + Uint *szp) +{ + Eterm res = THE_NON_VALUE; + + if (segments.max_ever.no < segments.max.no) + segments.max_ever.no = segments.max.no; + if (segments.max_ever.sz < segments.max.sz) + segments.max_ever.sz = segments.max.sz; + + if (print_to_p) { + int to = *print_to_p; + void *arg = print_to_arg; + + erts_print(to, arg, "cached_segments: %bpu\n", cache_size); + erts_print(to, arg, "cache_hits: %bpu\n", cache_hits); + erts_print(to, arg, "segments: %bpu %bpu %bpu\n", + segments.current.no, segments.max.no, segments.max_ever.no); + erts_print(to, arg, "segments_size: %bpu %bpu %bpu\n", + segments.current.sz, segments.max.sz, segments.max_ever.sz); + erts_print(to, arg, "segments_watermark: %bpu\n", + segments.current.watermark); + } + + if (hpp || szp) { + res = NIL; + add_2tup(hpp, szp, &res, + am.segments_watermark, + bld_unstable_uint(hpp, szp, segments.current.watermark)); + add_4tup(hpp, szp, &res, + am.segments_size, + bld_unstable_uint(hpp, szp, segments.current.sz), + bld_unstable_uint(hpp, szp, segments.max.sz), + bld_unstable_uint(hpp, szp, segments.max_ever.sz)); + add_4tup(hpp, szp, &res, + am.segments, + bld_unstable_uint(hpp, szp, segments.current.no), + bld_unstable_uint(hpp, szp, segments.max.no), + bld_unstable_uint(hpp, szp, segments.max_ever.no)); + add_2tup(hpp, szp, &res, + am.cache_hits, + bld_unstable_uint(hpp, szp, cache_hits)); + add_2tup(hpp, szp, &res, + am.cached_segments, + bld_unstable_uint(hpp, szp, cache_size)); + + } + + if (begin_new_max_period) { + segments.max.no = segments.current.no; + segments.max.sz = segments.current.sz; + } + + return res; +} + +static Eterm +info_version(int *print_to_p, void *print_to_arg, Uint **hpp, Uint *szp) +{ + Eterm res = THE_NON_VALUE; + + if (print_to_p) { + erts_print(*print_to_p, print_to_arg, "version: %s\n", + ERTS_MSEG_VSN_STR); + } + + if (hpp || szp) { + res = bld_string(hpp, szp, ERTS_MSEG_VSN_STR); + } + + return res; +} + +/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *\ + * Exported functions * +\* */ + +Eterm +erts_mseg_info_options(int *print_to_p, void *print_to_arg, + Uint **hpp, Uint *szp) +{ + Eterm res; + + erts_mtx_lock(&mseg_mutex); + + res = info_options("option ", print_to_p, print_to_arg, hpp, szp); + + erts_mtx_unlock(&mseg_mutex); + + return res; +} + +Eterm +erts_mseg_info(int *print_to_p, + void *print_to_arg, + int begin_max_per, + Uint **hpp, + Uint *szp) +{ + Eterm res = THE_NON_VALUE; + Eterm atoms[4]; + Eterm values[4]; + + erts_mtx_lock(&mseg_mutex); + + if (hpp || szp) { + + if (!atoms_initialized) + init_atoms(); + + atoms[0] = am.version; + atoms[1] = am.options; + atoms[2] = am.status; + atoms[3] = am.calls; + } + + values[0] = info_version(print_to_p, print_to_arg, hpp, szp); + values[1] = info_options("option ", print_to_p, print_to_arg, hpp, szp); + values[2] = info_status(print_to_p, print_to_arg, begin_max_per, hpp, szp); + values[3] = info_calls(print_to_p, print_to_arg, hpp, szp); + + if (hpp || szp) + res = bld_2tup_list(hpp, szp, 4, atoms, values); + + erts_mtx_unlock(&mseg_mutex); + + return res; +} + +void * +erts_mseg_alloc_opt(ErtsAlcType_t atype, Uint *size_p, const ErtsMsegOpt_t *opt) +{ + void *seg; + erts_mtx_lock(&mseg_mutex); + seg = mseg_alloc(atype, size_p, opt); + erts_mtx_unlock(&mseg_mutex); + return seg; +} + +void * +erts_mseg_alloc(ErtsAlcType_t atype, Uint *size_p) +{ + return erts_mseg_alloc_opt(atype, size_p, &default_opt); +} + +void +erts_mseg_dealloc_opt(ErtsAlcType_t atype, void *seg, Uint size, + const ErtsMsegOpt_t *opt) +{ + erts_mtx_lock(&mseg_mutex); + mseg_dealloc(atype, seg, size, opt); + erts_mtx_unlock(&mseg_mutex); +} + +void +erts_mseg_dealloc(ErtsAlcType_t atype, void *seg, Uint size) +{ + erts_mseg_dealloc_opt(atype, seg, size, &default_opt); +} + +void * +erts_mseg_realloc_opt(ErtsAlcType_t atype, void *seg, Uint old_size, + Uint *new_size_p, const ErtsMsegOpt_t *opt) +{ + void *new_seg; + erts_mtx_lock(&mseg_mutex); + new_seg = mseg_realloc(atype, seg, old_size, new_size_p, opt); + erts_mtx_unlock(&mseg_mutex); + return new_seg; +} + +void * +erts_mseg_realloc(ErtsAlcType_t atype, void *seg, Uint old_size, + Uint *new_size_p) +{ + return erts_mseg_realloc_opt(atype, seg, old_size, new_size_p, &default_opt); +} + +void +erts_mseg_clear_cache(void) +{ + erts_mtx_lock(&mseg_mutex); + mseg_clear_cache(); + erts_mtx_unlock(&mseg_mutex); +} + +Uint +erts_mseg_no(void) +{ + Uint n; + erts_mtx_lock(&mseg_mutex); + n = segments.current.no; + erts_mtx_unlock(&mseg_mutex); + return n; +} + +Uint +erts_mseg_unit_size(void) +{ + return page_size; +} + +void +erts_mseg_init(ErtsMsegInit_t *init) +{ + unsigned i; + + atoms_initialized = 0; + is_init_done = 0; + + /* Options ... */ + + abs_max_cache_bad_fit = init->amcbf; + rel_max_cache_bad_fit = init->rmcbf; + max_cache_size = init->mcs; + cache_check_interval = init->cci; + + /* */ + +#ifdef USE_THREADS + thread_safe_init(); +#endif + +#if HAVE_MMAP && !defined(MAP_ANON) + mmap_fd = open("/dev/zero", O_RDWR); + if (mmap_fd < 0) + erl_exit(ERTS_ABORT_EXIT, "erts_mseg: unable to open /dev/zero\n"); +#endif + + page_size = GET_PAGE_SIZE; + + page_shift = 1; + while ((page_size >> page_shift) != 1) { + if ((page_size & (1 << (page_shift - 1))) != 0) + erl_exit(ERTS_ABORT_EXIT, + "erts_mseg: Unexpected page_size %bpu\n", page_size); + page_shift++; + } + + sys_memzero((void *) &calls, sizeof(calls)); + +#if CAN_PARTLY_DESTROY + min_seg_size = ~((Uint) 0); +#endif + + cache = NULL; + cache_end = NULL; + cache_hits = 0; + max_cached_seg_size = 0; + min_cached_seg_size = ~((Uint) 0); + cache_size = 0; + + is_cache_check_scheduled = 0; +#ifdef ERTS_THREADS_NO_SMP + is_cache_check_requested = 0; +#endif + + if (max_cache_size > MAX_CACHE_SIZE) + max_cache_size = MAX_CACHE_SIZE; + + if (max_cache_size > 0) { + for (i = 0; i < max_cache_size - 1; i++) + cache_descs[i].next = &cache_descs[i + 1]; + cache_descs[max_cache_size - 1].next = NULL; + free_cache_descs = &cache_descs[0]; + } + else + free_cache_descs = NULL; + + segments.current.watermark = 0; + segments.current.no = 0; + segments.current.sz = 0; + segments.max.no = 0; + segments.max.sz = 0; + segments.max_ever.no = 0; + segments.max_ever.sz = 0; +} + + +/* + * erts_mseg_late_init() have to be called after all allocators, + * threads and timers have been initialized. + */ +void +erts_mseg_late_init(void) +{ +#ifdef ERTS_THREADS_NO_SMP + int handle = + erts_register_async_ready_callback( + check_schedule_cache_check); +#endif + erts_mtx_lock(&mseg_mutex); + is_init_done = 1; +#ifdef ERTS_THREADS_NO_SMP + async_handle = handle; +#endif + if (cache_size) + schedule_cache_check(); + erts_mtx_unlock(&mseg_mutex); +} + +void +erts_mseg_exit(void) +{ + mseg_shutdown(); +} + +#endif /* #if HAVE_ERTS_MSEG */ + +unsigned long +erts_mseg_test(unsigned long op, + unsigned long a1, + unsigned long a2, + unsigned long a3) +{ + switch (op) { +#if HAVE_ERTS_MSEG + case 0x400: /* Have erts_mseg */ + return (unsigned long) 1; + case 0x401: + return (unsigned long) erts_mseg_alloc(ERTS_ALC_A_INVALID, (Uint *) a1); + case 0x402: + erts_mseg_dealloc(ERTS_ALC_A_INVALID, (void *) a1, (Uint) a2); + return (unsigned long) 0; + case 0x403: + return (unsigned long) erts_mseg_realloc(ERTS_ALC_A_INVALID, + (void *) a1, + (Uint) a2, + (Uint *) a3); + case 0x404: + erts_mseg_clear_cache(); + return (unsigned long) 0; + case 0x405: + return (unsigned long) erts_mseg_no(); + case 0x406: { + unsigned long res; + erts_mtx_lock(&mseg_mutex); + res = (unsigned long) cache_size; + erts_mtx_unlock(&mseg_mutex); + return res; + } +#else /* #if HAVE_ERTS_MSEG */ + case 0x400: /* Have erts_mseg */ + return (unsigned long) 0; +#endif /* #if HAVE_ERTS_MSEG */ + default: ASSERT(0); return ~((unsigned long) 0); + } + +} + + diff --git a/erts/emulator/sys/common/erl_mseg.h b/erts/emulator/sys/common/erl_mseg.h new file mode 100644 index 0000000000..1c5aa63e90 --- /dev/null +++ b/erts/emulator/sys/common/erl_mseg.h @@ -0,0 +1,97 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2002-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +#ifndef ERL_MSEG_H_ +#define ERL_MSEG_H_ + +#include "sys.h" +#include "erl_alloc_types.h" + +#ifndef HAVE_MMAP +# define HAVE_MMAP 0 +#endif +#ifndef HAVE_MREMAP +# define HAVE_MREMAP 0 +#endif + +#if HAVE_MMAP +# define HAVE_ERTS_MSEG 1 +#else +# define HAVE_ERTS_MSEG 0 +#endif + +#if HAVE_ERTS_MSEG + +#define ERTS_MSEG_VSN_STR "0.9" + +typedef struct { + Uint amcbf; + Uint rmcbf; + Uint mcs; + Uint cci; +} ErtsMsegInit_t; + +#define ERTS_MSEG_INIT_DEFAULT_INITIALIZER \ +{ \ + 4*1024*1024, /* amcbf: Absolute max cache bad fit */ \ + 20, /* rmcbf: Relative max cache bad fit */ \ + 5, /* mcs: Max cache size */ \ + 1000 /* cci: Cache check interval */ \ +} + +typedef struct { + int cache; + int preserv; + Uint abs_shrink_th; + Uint rel_shrink_th; +} ErtsMsegOpt_t; + +#define ERTS_MSEG_DEFAULT_OPT_INITIALIZER \ +{ \ + 1, /* Use cache */ \ + 1, /* Preserv data */ \ + 0, /* Absolute shrink threshold */ \ + 0 /* Relative shrink threshold */ \ +} + +void *erts_mseg_alloc(ErtsAlcType_t, Uint *); +void *erts_mseg_alloc_opt(ErtsAlcType_t, Uint *, const ErtsMsegOpt_t *); +void erts_mseg_dealloc(ErtsAlcType_t, void *, Uint); +void erts_mseg_dealloc_opt(ErtsAlcType_t, void *, Uint, const ErtsMsegOpt_t *); +void *erts_mseg_realloc(ErtsAlcType_t, void *, Uint, Uint *); +void *erts_mseg_realloc_opt(ErtsAlcType_t, void *, Uint, Uint *, + const ErtsMsegOpt_t *); +void erts_mseg_clear_cache(void); +Uint erts_mseg_no(void); +Uint erts_mseg_unit_size(void); +void erts_mseg_init(ErtsMsegInit_t *init); +void erts_mseg_late_init(void); /* Have to be called after all allocators, + threads and timers have been initialized. */ +void erts_mseg_exit(void); +Eterm erts_mseg_info_options(int *, void*, Uint **, Uint *); +Eterm erts_mseg_info(int *, void*, int, Uint **, Uint *); + +#endif /* #if HAVE_ERTS_MSEG */ + +unsigned long erts_mseg_test(unsigned long, + unsigned long, + unsigned long, + unsigned long); + +#endif /* #ifndef ERL_MSEG_H_ */ diff --git a/erts/emulator/sys/common/erl_mtrace_sys_wrap.c b/erts/emulator/sys/common/erl_mtrace_sys_wrap.c new file mode 100644 index 0000000000..408aa7e016 --- /dev/null +++ b/erts/emulator/sys/common/erl_mtrace_sys_wrap.c @@ -0,0 +1,245 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2004-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif +#include "sys.h" +#include "erl_mtrace.h" + +#ifdef ERTS_CAN_TRACK_MALLOC +#if defined(HAVE_END_SYMBOL) +extern char end; +#elif defined(HAVE__END_SYMBOL) +extern char _end; +#endif + +static int inited = 0; +static int init(void); + +static volatile char *heap_start = NULL; +static volatile char *heap_end = NULL; + +#if defined(ERTS___AFTER_MORECORE_HOOK_CAN_TRACK_MALLOC) /* ----------------- */ + +#ifdef HAVE_MALLOC_H +# include <malloc.h> +#endif + +#undef SBRK_0 +#define SBRK_0 sbrk(0) + +static void +init_hook(void) +{ + __after_morecore_hook = erts_mtrace_update_heap_size; + if (inited) + return; + heap_end = NULL; +#if defined(HAVE_END_SYMBOL) + heap_start = &end; +#elif defined(HAVE__END_SYMBOL) + heap_start = &_end; +#else + heap_start = SBRK_0; + if (heap_start == (SBRK_RET_TYPE) -1) { + heap_start = NULL; + return; + } +#endif + inited = 1; +} + +static int +init(void) +{ + init_hook(); + return inited; +} + +void (*__malloc_initialize_hook)(void) = init_hook; + +#elif defined(ERTS_BRK_WRAPPERS_CAN_TRACK_MALLOC) /* ------------------------ */ +#ifdef HAVE_DLFCN_H +# include <dlfcn.h> +#endif + +#undef SBRK_0 +#define SBRK_0 (*real_sbrk)(0) + +#ifndef HAVE_SBRK +# error no sbrk() +#endif +#if !defined(HAVE_END_SYMBOL) && !defined(HAVE__END_SYMBOL) +# error no 'end' nor '_end' +#endif + +static void update_heap_size(char *new_end); + +#define SBRK_IMPL(RET_TYPE, FUNC, ARG_TYPE) \ +RET_TYPE FUNC (ARG_TYPE); \ +static RET_TYPE (*real_ ## FUNC)(ARG_TYPE) = NULL; \ +RET_TYPE FUNC (ARG_TYPE arg) \ +{ \ + RET_TYPE res; \ + if (!inited && !init()) \ + return (RET_TYPE) -1; \ + res = (*real_ ## FUNC)(arg); \ + if (erts_mtrace_enabled && res != ((RET_TYPE) -1)) \ + update_heap_size((char *) (*real_ ## FUNC)(0)); \ + return res; \ +} + +#define BRK_IMPL(RET_TYPE, FUNC, ARG_TYPE) \ +RET_TYPE FUNC (ARG_TYPE); \ +static RET_TYPE (*real_ ## FUNC)(ARG_TYPE) = NULL; \ +RET_TYPE FUNC (ARG_TYPE arg) \ +{ \ + RET_TYPE res; \ + if (!inited && !init()) \ + return (RET_TYPE) -1; \ + res = (*real_ ## FUNC)(arg); \ + if (erts_mtrace_enabled && res != ((RET_TYPE) -1)) \ + update_heap_size((char *) arg); \ + return res; \ +} + +SBRK_IMPL(SBRK_RET_TYPE, sbrk, SBRK_ARG_TYPE) +#ifdef HAVE_BRK + BRK_IMPL(BRK_RET_TYPE, brk, BRK_ARG_TYPE) +#endif + +#ifdef HAVE__SBRK + SBRK_IMPL(SBRK_RET_TYPE, _sbrk, SBRK_ARG_TYPE) +#endif +#ifdef HAVE__BRK + BRK_IMPL(BRK_RET_TYPE, _brk, BRK_ARG_TYPE) +#endif + +#ifdef HAVE___SBRK + SBRK_IMPL(SBRK_RET_TYPE, __sbrk, SBRK_ARG_TYPE) +#endif +#ifdef HAVE___BRK + BRK_IMPL(BRK_RET_TYPE, __brk, BRK_ARG_TYPE) +#endif + +static int +init(void) +{ + if (inited) + return 1; + +#define INIT_XBRK_SYM(SYM) \ +do { \ + if (!real_ ## SYM) { \ + real_ ## SYM = dlsym(RTLD_NEXT, #SYM); \ + if (!real_ ## SYM) { \ + errno = ENOMEM; \ + return 0; \ + } \ + } \ +} while (0) + + heap_end = NULL; +#if defined(HAVE_END_SYMBOL) + heap_start = &end; +#elif defined(HAVE__END_SYMBOL) + heap_start = &_end; +#endif + + INIT_XBRK_SYM(sbrk); +#ifdef HAVE_BRK + INIT_XBRK_SYM(brk); +#endif +#ifdef HAVE__SBRK + INIT_XBRK_SYM(_sbrk); +#endif +#ifdef HAVE__BRK + INIT_XBRK_SYM(_brk); +#endif +#ifdef HAVE___SBRK + INIT_XBRK_SYM(__sbrk); +#endif +#ifdef HAVE___BRK + INIT_XBRK_SYM(__brk); +#endif + + return inited = 1; +#undef INIT_XBRK_SYM +} + +#endif /* #elif defined(ERTS_BRK_WRAPPERS_CAN_TRACK_MALLOC) */ /* ----------- */ + +static void +update_heap_size(char *new_end) +{ + volatile char *new_start, *old_start, *old_end; + Uint size; + + if (new_end == ((char *) -1)) + return; + + new_start = (old_start = heap_start); + old_end = heap_end; + heap_end = new_end; + if (new_end < old_start || !old_start) + heap_start = (new_start = new_end); + + size = (Uint) (new_end - new_start); + + if (!old_end) { + if (size) + erts_mtrace_crr_alloc((void *) new_start, + ERTS_ALC_A_SYSTEM, + ERTS_MTRACE_SEGMENT_ID, + size); + else + heap_end = NULL; + } + else { + if (old_end != new_end || old_start != new_start) { + + if (size) + erts_mtrace_crr_realloc((void *) new_start, + ERTS_ALC_A_SYSTEM, + ERTS_MTRACE_SEGMENT_ID, + (void *) old_start, + size); + else { + if (old_start) + erts_mtrace_crr_free(ERTS_ALC_A_SYSTEM, + ERTS_MTRACE_SEGMENT_ID, + (void *) old_start); + heap_end = NULL; + } + } + } +} + +#endif /* #ifdef ERTS_CAN_TRACK_MALLOC */ + +void +erts_mtrace_update_heap_size(void) +{ +#ifdef ERTS_CAN_TRACK_MALLOC + if (erts_mtrace_enabled && (inited || init())) + update_heap_size((char *) SBRK_0); +#endif +} + diff --git a/erts/emulator/sys/common/erl_poll.c b/erts/emulator/sys/common/erl_poll.c new file mode 100644 index 0000000000..169d4579a2 --- /dev/null +++ b/erts/emulator/sys/common/erl_poll.c @@ -0,0 +1,2693 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2006-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +/* + * Description: Poll interface suitable for ERTS with or without + * SMP support. + * + * The interface is currently implemented using: + * - select + * - poll + * - /dev/poll + * - epoll with poll or select as fallback + * - kqueue with poll or select as fallback + * + * Some time in the future it will also be + * implemented using Solaris ports. + * + * + * + * Author: Rickard Green + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#ifndef WANT_NONBLOCKING +# define WANT_NONBLOCKING +#endif +#define ERTS_WANT_GOT_SIGUSR1 + +#include "erl_poll.h" +#if ERTS_POLL_USE_KQUEUE +# include <sys/types.h> +# include <sys/event.h> +# include <sys/time.h> +#endif +#if ERTS_POLL_USE_SELECT +# ifdef SYS_SELECT_H +# include <sys/select.h> +# endif +# ifdef VXWORKS +# include <selectLib.h> +# endif +#endif +#ifndef VXWORKS +# ifdef NO_SYSCONF +# if ERTS_POLL_USE_SELECT +# include <sys/param.h> +# else +# include <limits.h> +# endif +# endif +#endif +#include "erl_driver.h" +#include "erl_alloc.h" + +#if !defined(ERTS_POLL_USE_EPOLL) \ + && !defined(ERTS_POLL_USE_DEVPOLL) \ + && !defined(ERTS_POLL_USE_POLL) \ + && !defined(ERTS_POLL_USE_SELECT) +#error "Missing implementation of erts_poll()" +#endif + +#if defined(ERTS_KERNEL_POLL_VERSION) && !ERTS_POLL_USE_KERNEL_POLL +#error "Missing kernel poll implementation of erts_poll()" +#endif + +#if defined(ERTS_NO_KERNEL_POLL_VERSION) && ERTS_POLL_USE_KERNEL_POLL +#error "Kernel poll used when it shouldn't be used" +#endif + +#if 0 +#define ERTS_POLL_DEBUG_PRINT +#endif + +#if defined(DEBUG) && 0 +#define HARD_DEBUG +#endif + +#define ERTS_POLL_USE_BATCH_UPDATE_POLLSET (ERTS_POLL_USE_DEVPOLL \ + || ERTS_POLL_USE_KQUEUE) +#define ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE \ + (defined(ERTS_SMP) || ERTS_POLL_USE_KERNEL_POLL || ERTS_POLL_USE_POLL) + +#define ERTS_POLL_USE_CONCURRENT_UPDATE \ + (defined(ERTS_SMP) && ERTS_POLL_USE_EPOLL) + +#define ERTS_POLL_COALESCE_KP_RES (ERTS_POLL_USE_KQUEUE || ERTS_POLL_USE_EPOLL) + +#define FDS_STATUS_EXTRA_FREE_SIZE 128 +#define POLL_FDS_EXTRA_FREE_SIZE 128 + +#ifdef ERTS_POLL_NEED_ASYNC_INTERRUPT_SUPPORT +# define ERTS_POLL_ASYNC_INTERRUPT_SUPPORT 1 +#else +# define ERTS_POLL_ASYNC_INTERRUPT_SUPPORT 0 +#endif + +#define ERTS_POLL_USE_WAKEUP_PIPE \ + (ERTS_POLL_ASYNC_INTERRUPT_SUPPORT || defined(ERTS_SMP)) + +#ifdef ERTS_SMP + +#define ERTS_POLLSET_LOCK(PS) \ + erts_smp_mtx_lock(&(PS)->mtx) +#define ERTS_POLLSET_UNLOCK(PS) \ + erts_smp_mtx_unlock(&(PS)->mtx) + +#define ERTS_POLLSET_SET_POLLED_CHK(PS) \ + ((int) erts_smp_atomic_xchg(&(PS)->polled, (long) 1)) +#define ERTS_POLLSET_UNSET_POLLED(PS) \ + erts_smp_atomic_set(&(PS)->polled, (long) 0) +#define ERTS_POLLSET_IS_POLLED(PS) \ + ((int) erts_smp_atomic_read(&(PS)->polled)) + +#define ERTS_POLLSET_SET_POLLER_WOKEN_CHK(PS) \ + ((int) erts_smp_atomic_xchg(&(PS)->woken, (long) 1)) +#define ERTS_POLLSET_SET_POLLER_WOKEN(PS) \ + erts_smp_atomic_set(&(PS)->woken, (long) 1) +#define ERTS_POLLSET_UNSET_POLLER_WOKEN(PS) \ + erts_smp_atomic_set(&(PS)->woken, (long) 0) +#define ERTS_POLLSET_IS_POLLER_WOKEN(PS) \ + ((int) erts_smp_atomic_read(&(PS)->woken)) + +#else + +#define ERTS_POLLSET_LOCK(PS) +#define ERTS_POLLSET_UNLOCK(PS) +#define ERTS_POLLSET_SET_POLLED_CHK(PS) 0 +#define ERTS_POLLSET_UNSET_POLLED(PS) +#define ERTS_POLLSET_IS_POLLED(PS) 0 + +#if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT + +/* + * Ideally, the ERTS_POLLSET_SET_POLLER_WOKEN_CHK(PS) operation would + * be atomic. This operation isn't, but we will do okay anyway. The + * "woken check" is only an optimization. The only requirement we have: + * If (PS)->woken is set to a value != 0 when interrupting, we have to + * write on the the wakeup pipe at least once. Multiple writes are okay. + */ +#define ERTS_POLLSET_SET_POLLER_WOKEN_CHK(PS) ((PS)->woken++) +#define ERTS_POLLSET_SET_POLLER_WOKEN(PS) ((PS)->woken = 1, (void) 0) +#define ERTS_POLLSET_UNSET_POLLER_WOKEN(PS) ((PS)->woken = 0, (void) 0) +#define ERTS_POLLSET_IS_POLLER_WOKEN(PS) ((PS)->woken) + +#else + +#define ERTS_POLLSET_SET_POLLER_WOKEN_CHK(PS) 1 +#define ERTS_POLLSET_SET_POLLER_WOKEN(PS) +#define ERTS_POLLSET_UNSET_POLLER_WOKEN(PS) +#define ERTS_POLLSET_IS_POLLER_WOKEN(PS) 1 + +#endif + +#endif + +#if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE +#define ERTS_POLLSET_SET_HAVE_UPDATE_REQUESTS(PS) \ + erts_smp_atomic_set(&(PS)->have_update_requests, (long) 1) +#define ERTS_POLLSET_UNSET_HAVE_UPDATE_REQUESTS(PS) \ + erts_smp_atomic_set(&(PS)->have_update_requests, (long) 0) +#define ERTS_POLLSET_HAVE_UPDATE_REQUESTS(PS) \ + ((int) erts_smp_atomic_read(&(PS)->have_update_requests)) +#else +#define ERTS_POLLSET_SET_HAVE_UPDATE_REQUESTS(PS) +#define ERTS_POLLSET_UNSET_HAVE_UPDATE_REQUESTS(PS) +#define ERTS_POLLSET_HAVE_UPDATE_REQUESTS(PS) 0 +#endif + +#if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT && !defined(ERTS_SMP) + +#define ERTS_POLLSET_UNSET_INTERRUPTED_CHK(PS) unset_interrupted_chk((PS)) +#define ERTS_POLLSET_UNSET_INTERRUPTED(PS) ((PS)->interrupt = 0, (void) 0) +#define ERTS_POLLSET_SET_INTERRUPTED(PS) ((PS)->interrupt = 1, (void) 0) +#define ERTS_POLLSET_IS_INTERRUPTED(PS) ((PS)->interrupt) + +#else + +#define ERTS_POLLSET_UNSET_INTERRUPTED_CHK(PS) \ + ((int) erts_smp_atomic_xchg(&(PS)->interrupt, (long) 0)) +#define ERTS_POLLSET_UNSET_INTERRUPTED(PS) \ + erts_smp_atomic_set(&(PS)->interrupt, (long) 0) +#define ERTS_POLLSET_SET_INTERRUPTED(PS) \ + erts_smp_atomic_set(&(PS)->interrupt, (long) 1) +#define ERTS_POLLSET_IS_INTERRUPTED(PS) \ + ((int) erts_smp_atomic_read(&(PS)->interrupt)) + +#endif + +#if ERTS_POLL_USE_FALLBACK +# if ERTS_POLL_USE_POLL +# define ERTS_POLL_NEED_FALLBACK(PS) ((PS)->no_poll_fds > 1) +# elif ERTS_POLL_USE_SELECT +# define ERTS_POLL_NEED_FALLBACK(PS) ((PS)->no_select_fds > 1) +# endif +#endif +/* + * --- Data types ------------------------------------------------------------ + */ + +#if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE +#define ERTS_POLLSET_UPDATE_REQ_BLOCK_SIZE 128 + +typedef struct ErtsPollSetUpdateRequestsBlock_ ErtsPollSetUpdateRequestsBlock; +struct ErtsPollSetUpdateRequestsBlock_ { + ErtsPollSetUpdateRequestsBlock *next; + int len; + int fds[ERTS_POLLSET_UPDATE_REQ_BLOCK_SIZE]; +}; + +#endif + + +#if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE +# define ERTS_POLL_FD_FLG_INURQ (((unsigned short) 1) << 0) +#endif +#if ERTS_POLL_USE_FALLBACK +# define ERTS_POLL_FD_FLG_INFLBCK (((unsigned short) 1) << 1) +# define ERTS_POLL_FD_FLG_USEFLBCK (((unsigned short) 1) << 2) +#endif +#if ERTS_POLL_USE_KERNEL_POLL || defined(ERTS_SMP) +# define ERTS_POLL_FD_FLG_RST (((unsigned short) 1) << 3) +#endif +typedef struct { +#if ERTS_POLL_USE_POLL + int pix; +#endif + ErtsPollEvents used_events; + ErtsPollEvents events; +#if ERTS_POLL_COALESCE_KP_RES + unsigned short res_ev_ix; +#endif +#if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE || ERTS_POLL_USE_FALLBACK + unsigned short flags; +#endif + +} ErtsFdStatus; + + +#if ERTS_POLL_COALESCE_KP_RES +/* res_ev_ix max value */ +#define ERTS_POLL_MAX_RES ((1 << sizeof(unsigned short)*8) - 1) +#endif + +#if ERTS_POLL_USE_KQUEUE + +#define ERTS_POLL_KQ_OP_HANDLED 1 +#define ERTS_POLL_KQ_OP_DEL_R 2 +#define ERTS_POLL_KQ_OP_DEL_W 3 +#define ERTS_POLL_KQ_OP_ADD_R 4 +#define ERTS_POLL_KQ_OP_ADD_W 5 +#define ERTS_POLL_KQ_OP_ADD2_R 6 +#define ERTS_POLL_KQ_OP_ADD2_W 7 + +#endif + +struct ErtsPollSet_ { + ErtsPollSet next; + int internal_fd_limit; + ErtsFdStatus *fds_status; + int no_of_user_fds; + int fds_status_len; +#if ERTS_POLL_USE_KERNEL_POLL + int kp_fd; + int res_events_len; +#if ERTS_POLL_USE_EPOLL + struct epoll_event *res_events; +#elif ERTS_POLL_USE_KQUEUE + struct kevent *res_events; +#elif ERTS_POLL_USE_DEVPOLL + struct pollfd *res_events; +#endif +#endif /* ERTS_POLL_USE_KERNEL_POLL */ +#if ERTS_POLL_USE_POLL + int next_poll_fds_ix; + int no_poll_fds; + int poll_fds_len; + struct pollfd*poll_fds; +#elif ERTS_POLL_USE_SELECT + int next_sel_fd; + int max_fd; +#if ERTS_POLL_USE_FALLBACK + int no_select_fds; +#endif + fd_set input_fds; + fd_set res_input_fds; + fd_set output_fds; + fd_set res_output_fds; +#endif +#if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE + ErtsPollSetUpdateRequestsBlock update_requests; + ErtsPollSetUpdateRequestsBlock *curr_upd_req_block; + erts_smp_atomic_t have_update_requests; +#endif +#ifdef ERTS_SMP + erts_smp_atomic_t polled; + erts_smp_atomic_t woken; + erts_smp_mtx_t mtx; +#elif ERTS_POLL_ASYNC_INTERRUPT_SUPPORT + volatile int woken; +#endif +#if ERTS_POLL_USE_WAKEUP_PIPE + int wake_fds[2]; +#endif +#if ERTS_POLL_USE_FALLBACK + int fallback_used; +#endif +#if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT && !defined(ERTS_SMP) + volatile int interrupt; +#else + erts_smp_atomic_t interrupt; +#endif + erts_smp_atomic_t timeout; +#ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS + erts_smp_atomic_t no_avoided_wakeups; + erts_smp_atomic_t no_avoided_interrupts; + erts_smp_atomic_t no_interrupt_timed; +#endif +}; + +#if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT && !defined(ERTS_SMP) + +static ERTS_INLINE int +unset_interrupted_chk(ErtsPollSet ps) +{ + /* This operation isn't atomic, but we have no need at all for an + atomic operation here... */ + int res = ps->interrupt; + ps->interrupt = 0; + return res; +} + +#endif + +static void fatal_error(char *format, ...); +static void fatal_error_async_signal_safe(char *error_str); + +static int max_fds = -1; +static ErtsPollSet pollsets; +static erts_smp_spinlock_t pollsets_lock; + +#if ERTS_POLL_USE_POLL + +static ERTS_INLINE short +ev2pollev(ErtsPollEvents ev) +{ +#if !ERTS_POLL_USE_FALLBACK || ERTS_POLL_USE_KQUEUE + return ERTS_POLL_EV_E2N(ev); +#else /* Note, we only map events we are interested in */ + short res_ev = (short) 0; + if (ev & ERTS_POLL_EV_IN) + res_ev |= ERTS_POLL_EV_NKP_IN; + if (ev & ERTS_POLL_EV_OUT) + res_ev |= ERTS_POLL_EV_NKP_OUT; + return res_ev; +#endif +} + +static ERTS_INLINE ErtsPollEvents +pollev2ev(short ev) +{ +#if !ERTS_POLL_USE_FALLBACK || ERTS_POLL_USE_KQUEUE + return ERTS_POLL_EV_N2E(ev); +#else /* Note, we only map events we are interested in */ + ErtsPollEvents res_ev = (ErtsPollEvents) 0; + if (ev & ERTS_POLL_EV_NKP_IN) + res_ev |= ERTS_POLL_EV_IN; + if (ev & ERTS_POLL_EV_NKP_OUT) + res_ev |= ERTS_POLL_EV_OUT; + if (ev & ERTS_POLL_EV_NKP_ERR) + res_ev |= ERTS_POLL_EV_ERR; + if (ev & ERTS_POLL_EV_NKP_NVAL) + res_ev |= ERTS_POLL_EV_NVAL; + return res_ev; +#endif +} + +#endif + +#ifdef HARD_DEBUG +static void check_poll_result(ErtsPollResFd pr[], int len); +#if ERTS_POLL_USE_DEVPOLL +static void check_poll_status(ErtsPollSet ps); +#endif /* ERTS_POLL_USE_DEVPOLL */ +#endif /* HARD_DEBUG */ +#ifdef ERTS_POLL_DEBUG_PRINT +static void print_misc_debug_info(void); +#endif + +/* + * --- Wakeup pipe ----------------------------------------------------------- + */ + +#if ERTS_POLL_USE_WAKEUP_PIPE + +static ERTS_INLINE void +wake_poller(ErtsPollSet ps) +{ + /* + * NOTE: This function might be called from signal handlers in the + * non-smp case; therefore, it has to be async-signal safe in + * the non-smp case. + */ + if (!ERTS_POLLSET_SET_POLLER_WOKEN_CHK(ps)) { + ssize_t res; + if (ps->wake_fds[1] < 0) + return; /* Not initialized yet */ + do { + /* write() is async-signal safe (according to posix) */ + res = write(ps->wake_fds[1], "!", 1); + } while (res < 0 && errno == EINTR); + if (res <= 0 && errno != ERRNO_BLOCK) { + fatal_error_async_signal_safe(__FILE__ + ":XXX:wake_poller(): " + "Failed to write on wakeup pipe\n"); + } + } +} + +static ERTS_INLINE void +cleanup_wakeup_pipe(ErtsPollSet ps) +{ + int fd = ps->wake_fds[0]; + int res; + do { + char buf[32]; + res = read(fd, buf, sizeof(buf)); + } while (res > 0 || (res < 0 && errno == EINTR)); + if (res < 0 && errno != ERRNO_BLOCK) { + fatal_error("%s:%d:cleanup_wakeup_pipe(): " + "Failed to read on wakeup pipe fd=%d: " + "%s (%d)\n", + __FILE__, __LINE__, + fd, + erl_errno_id(errno), errno); + } +} + +static void +create_wakeup_pipe(ErtsPollSet ps) +{ + int do_wake = 0; + int wake_fds[2]; + ps->wake_fds[0] = -1; + ps->wake_fds[1] = -1; + if (pipe(wake_fds) < 0) { + fatal_error("%s:%d:create_wakeup_pipe(): " + "Failed to create pipe: %s (%d)\n", + __FILE__, + __LINE__, + erl_errno_id(errno), + errno); + } + SET_NONBLOCKING(wake_fds[0]); + SET_NONBLOCKING(wake_fds[1]); + +#ifdef ERTS_POLL_DEBUG_PRINT + erts_printf("wakeup fds = {%d, %d}\n", wake_fds[0], wake_fds[1]); +#endif + + ERTS_POLL_EXPORT(erts_poll_control)(ps, + wake_fds[0], + ERTS_POLL_EV_IN, + 1, &do_wake); +#if ERTS_POLL_USE_FALLBACK + /* We depend on the wakeup pipe being handled by kernel poll */ + if (ps->fds_status[wake_fds[0]].flags & ERTS_POLL_FD_FLG_INFLBCK) + fatal_error("%s:%d:create_wakeup_pipe(): Internal error\n", + __FILE__, __LINE__); +#endif + if (ps->internal_fd_limit <= wake_fds[1]) + ps->internal_fd_limit = wake_fds[1] + 1; + if (ps->internal_fd_limit <= wake_fds[0]) + ps->internal_fd_limit = wake_fds[0] + 1; + ps->wake_fds[0] = wake_fds[0]; + ps->wake_fds[1] = wake_fds[1]; +} + +#endif /* ERTS_POLL_USE_WAKEUP_PIPE */ + +/* + * --- Poll set update requests ---------------------------------------------- + */ +#if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE + +static ERTS_INLINE void +enqueue_update_request(ErtsPollSet ps, int fd) +{ + ErtsPollSetUpdateRequestsBlock *urqbp; + + ASSERT(fd < ps->fds_status_len); + + if (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INURQ) + return; + + if (ps->update_requests.len == 0) + ERTS_POLLSET_SET_HAVE_UPDATE_REQUESTS(ps); + + urqbp = ps->curr_upd_req_block; + + if (urqbp->len == ERTS_POLLSET_UPDATE_REQ_BLOCK_SIZE) { + ASSERT(!urqbp->next); + urqbp = erts_alloc(ERTS_ALC_T_POLLSET_UPDREQ, + sizeof(ErtsPollSetUpdateRequestsBlock)); + ps->curr_upd_req_block->next = urqbp; + ps->curr_upd_req_block = urqbp; + urqbp->next = NULL; + urqbp->len = 0; + } + + ps->fds_status[fd].flags |= ERTS_POLL_FD_FLG_INURQ; + urqbp->fds[urqbp->len++] = fd; +} + +static ERTS_INLINE void +free_update_requests_block(ErtsPollSet ps, + ErtsPollSetUpdateRequestsBlock *urqbp) +{ + if (urqbp != &ps->update_requests) + erts_free(ERTS_ALC_T_POLLSET_UPDREQ, (void *) urqbp); + else { + urqbp->next = NULL; + urqbp->len = 0; + } +} + +#endif /* ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE */ + +/* + * --- Growing poll set structures ------------------------------------------- + */ + +#if ERTS_POLL_USE_KERNEL_POLL +static void +grow_res_events(ErtsPollSet ps, int new_len) +{ + size_t new_size = sizeof( +#if ERTS_POLL_USE_EPOLL + struct epoll_event +#elif ERTS_POLL_USE_DEVPOLL + struct pollfd +#elif ERTS_POLL_USE_KQUEUE + struct kevent +#endif + )*new_len; + /* We do not need to save previously stored data */ + if (ps->res_events) + erts_free(ERTS_ALC_T_POLL_RES_EVS, ps->res_events); + ps->res_events = erts_alloc(ERTS_ALC_T_POLL_RES_EVS, new_size); + ps->res_events_len = new_len; +} +#endif /* ERTS_POLL_USE_KERNEL_POLL */ + +#if ERTS_POLL_USE_POLL +static void +grow_poll_fds(ErtsPollSet ps, int min_ix) +{ + int i; + int new_len = min_ix + 1 + POLL_FDS_EXTRA_FREE_SIZE; + if (new_len > max_fds) + new_len = max_fds; + ps->poll_fds = (ps->poll_fds_len + ? erts_realloc(ERTS_ALC_T_POLL_FDS, + ps->poll_fds, + sizeof(struct pollfd)*new_len) + : erts_alloc(ERTS_ALC_T_POLL_FDS, + sizeof(struct pollfd)*new_len)); + for (i = ps->poll_fds_len; i < new_len; i++) { + ps->poll_fds[i].fd = -1; + ps->poll_fds[i].events = (short) 0; + ps->poll_fds[i].revents = (short) 0; + } + ps->poll_fds_len = new_len; +} +#endif + +static void +grow_fds_status(ErtsPollSet ps, int min_fd) +{ + int i; + int new_len = min_fd + 1 + FDS_STATUS_EXTRA_FREE_SIZE; + ASSERT(min_fd < max_fds); + if (new_len > max_fds) + new_len = max_fds; + ps->fds_status = (ps->fds_status_len + ? erts_realloc(ERTS_ALC_T_FD_STATUS, + ps->fds_status, + sizeof(ErtsFdStatus)*new_len) + : erts_alloc(ERTS_ALC_T_FD_STATUS, + sizeof(ErtsFdStatus)*new_len)); + for (i = ps->fds_status_len; i < new_len; i++) { +#if ERTS_POLL_USE_POLL + ps->fds_status[i].pix = -1; +#endif + ps->fds_status[i].used_events = (ErtsPollEvents) 0; + ps->fds_status[i].events = (ErtsPollEvents) 0; +#if ERTS_POLL_COALESCE_KP_RES + ps->fds_status[i].res_ev_ix = (unsigned short) ERTS_POLL_MAX_RES; +#endif +#if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE || ERTS_POLL_USE_FALLBACK + ps->fds_status[i].flags = (unsigned short) 0; +#endif + } + ps->fds_status_len = new_len; +} + +/* + * --- Selecting fd to poll on ----------------------------------------------- + */ + +#if ERTS_POLL_USE_FALLBACK +static int update_fallback_pollset(ErtsPollSet ps, int fd); +#endif + +static ERTS_INLINE int +need_update(ErtsPollSet ps, int fd) +{ +#if ERTS_POLL_USE_KERNEL_POLL + int reset; +#endif + + ASSERT(fd < ps->fds_status_len); + +#if ERTS_POLL_USE_KERNEL_POLL + reset = (int) (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_RST); + if (reset && !ps->fds_status[fd].used_events) { + ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_RST; + reset = 0; + } +#elif defined(ERTS_SMP) + ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_RST; +#endif + + if (ps->fds_status[fd].used_events != ps->fds_status[fd].events) + return 1; + +#if ERTS_POLL_USE_KERNEL_POLL + return reset; +#else + return 0; +#endif +} + +#if ERTS_POLL_USE_BATCH_UPDATE_POLLSET + +#if ERTS_POLL_USE_KQUEUE +#define ERTS_POLL_MIN_BATCH_BUF_SIZE 128 +#else +#define ERTS_POLL_MIN_BATCH_BUF_SIZE 64 +#endif + +typedef struct { + int len; + int size; +#if ERTS_POLL_USE_DEVPOLL + struct pollfd *buf; +#elif ERTS_POLL_USE_KQUEUE + struct kevent *buf; + struct kevent *ebuf; +#endif +} ErtsPollBatchBuf; + + +static ERTS_INLINE void +setup_batch_buf(ErtsPollSet ps, ErtsPollBatchBuf *bbp) +{ + bbp->len = 0; +#if ERTS_POLL_USE_DEVPOLL + bbp->size = ps->res_events_len; + bbp->buf = ps->res_events; +#elif ERTS_POLL_USE_KQUEUE + bbp->size = ps->res_events_len/2; + bbp->buf = ps->res_events; + bbp->ebuf = bbp->buf + bbp->size; +#endif +} + + +#if ERTS_POLL_USE_DEVPOLL + +static void +write_batch_buf(ErtsPollSet ps, ErtsPollBatchBuf *bbp) +{ + ssize_t wres; + char *buf = (char *) bbp->buf; + size_t buf_size = sizeof(struct pollfd)*bbp->len; + + while (1) { + wres = write(ps->kp_fd, (void *) buf, buf_size); + if (wres < 0) { + if (errno == EINTR) + continue; + fatal_error("%s:%d:write_batch_buf(): " + "Failed to write to /dev/poll: " + "%s (%d)\n", + __FILE__, __LINE__, + erl_errno_id(errno), errno); + } + buf_size -= wres; + if (buf_size <= 0) + break; + buf += wres; + } + + if (buf_size < 0) { + fatal_error("%s:%d:write_devpoll_buf(): Internal error\n", + __FILE__, __LINE__); + } + bbp->len = 0; +} + +#elif ERTS_POLL_USE_KQUEUE + +static void +write_batch_buf(ErtsPollSet ps, ErtsPollBatchBuf *bbp) +{ + int res; + int len = bbp->len; + struct kevent *buf = bbp->buf; + struct timespec ts = {0, 0}; + + do { + res = kevent(ps->kp_fd, buf, len, NULL, 0, &ts); + } while (res < 0 && errno == EINTR); + if (res < 0) { + int i; + struct kevent *ebuf = bbp->ebuf; + do { + res = kevent(ps->kp_fd, buf, len, ebuf, len, &ts); + } while (res < 0 && errno == EINTR); + if (res < 0) { + fatal_error("%s:%d: kevent() failed: %s (%d)\n", + __FILE__, __LINE__, erl_errno_id(errno), errno); + } + for (i = 0; i < res; i++) { + if (ebuf[i].flags & EV_ERROR) { + short filter; + int fd = (int) ebuf[i].ident; + + switch ((int) ebuf[i].udata) { + + /* + * Since we use a lazy update approach EV_DELETE will + * frequently fail. This since kqueue automatically + * removes a file descriptor that is closed from the + * poll set. + */ + case ERTS_POLL_KQ_OP_DEL_R: + case ERTS_POLL_KQ_OP_DEL_W: + case ERTS_POLL_KQ_OP_HANDLED: + break; + + /* + * According to the kqueue man page EVFILT_READ support + * does not imply EVFILT_WRITE support; therefore, + * if an EV_ADD fail, we may have to remove other + * events on this fd in the kqueue pollset before + * adding fd to the fallback pollset. + */ + case ERTS_POLL_KQ_OP_ADD_W: + if (ps->fds_status[fd].used_events & ERTS_POLL_EV_IN) { + filter = EVFILT_READ; + goto rm_add_fb; + } + goto add_fb; + case ERTS_POLL_KQ_OP_ADD_R: + if (ps->fds_status[fd].used_events & ERTS_POLL_EV_OUT) { + filter = EVFILT_WRITE; + goto rm_add_fb; + } + goto add_fb; + case ERTS_POLL_KQ_OP_ADD2_W: + case ERTS_POLL_KQ_OP_ADD2_R: { + int j; + for (j = i+1; j < res; j++) { + if (fd == (int) ebuf[j].ident) { + ebuf[j].udata = (void *) ERTS_POLL_KQ_OP_HANDLED; + if (!(ebuf[j].flags & EV_ERROR)) { + switch ((int) ebuf[j].udata) { + case ERTS_POLL_KQ_OP_ADD2_W: + filter = EVFILT_WRITE; + goto rm_add_fb; + case ERTS_POLL_KQ_OP_ADD2_R: + filter = EVFILT_READ; + goto rm_add_fb; + default: + fatal_error("%s:%d:write_batch_buf(): " + "Internal error", + __FILE__, __LINE__); + break; + } + } + goto add_fb; + } + } + /* The other add succeded... */ + filter = (((int) ebuf[i].udata == ERTS_POLL_KQ_OP_ADD2_W) + ? EVFILT_READ + : EVFILT_WRITE); + rm_add_fb: + { + struct kevent kev; + struct timespec ts = {0, 0}; + EV_SET(&kev, fd, filter, EV_DELETE, 0, 0, 0); + (void) kevent(ps->kp_fd, &kev, 1, NULL, 0, &ts); + } + + add_fb: + ps->fds_status[fd].flags |= ERTS_POLL_FD_FLG_USEFLBCK; + ASSERT(ps->fds_status[fd].used_events); + ps->fds_status[fd].used_events = 0; + ps->no_of_user_fds--; + update_fallback_pollset(ps, fd); + ASSERT(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK); + break; + } + default: + fatal_error("%s:%d:write_batch_buf(): Internal error", + __FILE__, __LINE__); + break; + } + } + } + } + bbp->len = 0; +} + +#endif /* ERTS_POLL_USE_KQUEUE */ + +static ERTS_INLINE void +batch_update_pollset(ErtsPollSet ps, int fd, ErtsPollBatchBuf *bbp) +{ + int buf_len; +#if ERTS_POLL_USE_DEVPOLL + short events; + struct pollfd *buf; +#elif ERTS_POLL_USE_KQUEUE + struct kevent *buf; +#endif + +#ifdef ERTS_POLL_DEBUG_PRINT + erts_printf("Doing lazy update on fd=%d\n", fd); +#endif + + if (!need_update(ps, fd)) + return; + + /* Make sure we have room for at least maximum no of entries + per fd */ + if (bbp->size - bbp->len < 2) + write_batch_buf(ps, bbp); + + buf_len = bbp->len; + buf = bbp->buf; + + ASSERT(fd < ps->fds_status_len); + +#if ERTS_POLL_USE_DEVPOLL + events = ERTS_POLL_EV_E2N(ps->fds_status[fd].events); + if (!events) { + buf[buf_len].events = POLLREMOVE; + ps->no_of_user_fds--; + } + else if (!ps->fds_status[fd].used_events) { + buf[buf_len].events = events; + ps->no_of_user_fds++; + } + else { + if ((ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_RST) + || (ps->fds_status[fd].used_events & ~events)) { + /* Reset or removed events... */ + buf[buf_len].fd = fd; + buf[buf_len].events = POLLREMOVE; + buf[buf_len++].revents = 0; + } + buf[buf_len].events = events; + } + buf[buf_len].fd = fd; + buf[buf_len++].revents = 0; + +#elif ERTS_POLL_USE_KQUEUE + + if (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK) { + if (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_USEFLBCK) + update_fallback_pollset(ps, fd); + else { /* Remove from fallback and try kqueue */ + ErtsPollEvents events = ps->fds_status[fd].events; + ps->fds_status[fd].events = (ErtsPollEvents) 0; + update_fallback_pollset(ps, fd); + ASSERT(!(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK)); + if (events) { + ps->fds_status[fd].events = events; + goto try_kqueue; + } + } + } + else { + ErtsPollEvents events, used_events; + int mod_w, mod_r; + try_kqueue: + events = ERTS_POLL_EV_E2N(ps->fds_status[fd].events); + used_events = ERTS_POLL_EV_E2N(ps->fds_status[fd].used_events); + if (!(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_RST)) { + if (!used_events && + (events & ERTS_POLL_EV_IN) && (events & ERTS_POLL_EV_OUT)) + goto do_add_rw; + mod_r = ((events & ERTS_POLL_EV_IN) + != (used_events & ERTS_POLL_EV_IN)); + mod_w = ((events & ERTS_POLL_EV_OUT) + != (used_events & ERTS_POLL_EV_OUT)); + goto do_mod; + } + else { /* Reset */ + if ((events & ERTS_POLL_EV_IN) && (events & ERTS_POLL_EV_OUT)) { + do_add_rw: + EV_SET(&buf[buf_len], fd, EVFILT_READ, EV_ADD, + 0, 0, (void *) ERTS_POLL_KQ_OP_ADD2_R); + buf_len++; + EV_SET(&buf[buf_len], fd, EVFILT_WRITE, EV_ADD, + 0, 0, (void *) ERTS_POLL_KQ_OP_ADD2_W); + buf_len++; + + } + else { + mod_r = 1; + mod_w = 1; + do_mod: + if (mod_r) { + if (events & ERTS_POLL_EV_IN) { + EV_SET(&buf[buf_len], fd, EVFILT_READ, EV_ADD, + 0, 0, (void *) ERTS_POLL_KQ_OP_ADD_R); + buf_len++; + } + else if (used_events & ERTS_POLL_EV_IN) { + EV_SET(&buf[buf_len], fd, EVFILT_READ, EV_DELETE, + 0, 0, (void *) ERTS_POLL_KQ_OP_DEL_R); + buf_len++; + } + } + if (mod_w) { + if (events & ERTS_POLL_EV_OUT) { + EV_SET(&buf[buf_len], fd, EVFILT_WRITE, EV_ADD, + 0, 0, (void *) ERTS_POLL_KQ_OP_ADD_W); + buf_len++; + } + else if (used_events & ERTS_POLL_EV_OUT) { + EV_SET(&buf[buf_len], fd, EVFILT_WRITE, EV_DELETE, + 0, 0, (void *) ERTS_POLL_KQ_OP_DEL_W); + buf_len++; + } + } + } + } + if (used_events) { + if (!events) { + ps->no_of_user_fds--; + } + } + else { + if (events) + ps->no_of_user_fds++; + } + ASSERT((events & ~(ERTS_POLL_EV_IN|ERTS_POLL_EV_OUT)) == 0); + ASSERT((used_events & ~(ERTS_POLL_EV_IN|ERTS_POLL_EV_OUT)) == 0); + } + +#endif + + ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_RST; + ps->fds_status[fd].used_events = ps->fds_status[fd].events; + + bbp->len = buf_len; +} + +#else /* !ERTS_POLL_USE_BATCH_UPDATE_POLLSET */ + +#if ERTS_POLL_USE_EPOLL +static int +#if ERTS_POLL_USE_CONCURRENT_UPDATE +conc_update_pollset(ErtsPollSet ps, int fd, int *update_fallback) +#else +update_pollset(ErtsPollSet ps, int fd) +#endif +{ + int res; + int op; + struct epoll_event epe_templ; + struct epoll_event epe; + + ASSERT(fd < ps->fds_status_len); + + if (!need_update(ps, fd)) + return 0; + +#ifdef ERTS_POLL_DEBUG_PRINT + erts_printf("Doing update on fd=%d\n", fd); +#endif + if (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK) { +#if ERTS_POLL_USE_CONCURRENT_UPDATE + if (!*update_fallback) { + *update_fallback = 1; + return 0; + } +#endif + if (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_USEFLBCK) { + return update_fallback_pollset(ps, fd); + } + else { /* Remove from fallback and try epoll */ + ErtsPollEvents events = ps->fds_status[fd].events; + ps->fds_status[fd].events = (ErtsPollEvents) 0; + res = update_fallback_pollset(ps, fd); + ASSERT(!(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK)); + if (!events) + return res; + ps->fds_status[fd].events = events; + } + } + + epe_templ.events = ERTS_POLL_EV_E2N(ps->fds_status[fd].events); + epe_templ.data.fd = fd; + +#ifdef VALGRIND + /* Silence invalid valgrind warning ... */ + memset((void *) &epe.data, 0, sizeof(epoll_data_t)); +#endif + + if (epe_templ.events && ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_RST) { + do { + /* We init 'epe' every time since epoll_ctl() may modify it + (not declared const and not documented as const). */ + epe.events = epe_templ.events; + epe.data.fd = epe_templ.data.fd; + res = epoll_ctl(ps->kp_fd, EPOLL_CTL_DEL, fd, &epe); + } while (res != 0 && errno == EINTR); + ps->no_of_user_fds--; + ps->fds_status[fd].used_events = 0; + } + + if (!epe_templ.events) { + /* A note on EPOLL_CTL_DEL: linux kernel versions before 2.6.9 + need a non-NULL event pointer even though it is ignored... */ + op = EPOLL_CTL_DEL; + ps->no_of_user_fds--; + } + else if (!ps->fds_status[fd].used_events) { + op = EPOLL_CTL_ADD; + ps->no_of_user_fds++; + } + else { + op = EPOLL_CTL_MOD; + } + + do { + /* We init 'epe' every time since epoll_ctl() may modify it + (not declared const and not documented as const). */ + epe.events = epe_templ.events; + epe.data.fd = epe_templ.data.fd; + res = epoll_ctl(ps->kp_fd, op, fd, &epe); + } while (res != 0 && errno == EINTR); + +#if defined(ERTS_POLL_DEBUG_PRINT) && 1 + { + int saved_errno = errno; + erts_printf("%s = epoll_ctl(%d, %s, %d, {Ox%x, %d})\n", + res == 0 ? "0" : erl_errno_id(errno), + ps->kp_fd, + (op == EPOLL_CTL_ADD + ? "EPOLL_CTL_ADD" + : (op == EPOLL_CTL_MOD + ? "EPOLL_CTL_MOD" + : (op == EPOLL_CTL_DEL + ? "EPOLL_CTL_DEL" + : "UNKNOWN"))), + fd, + epe_templ.events, + fd); + errno = saved_errno; + } +#endif + if (res == 0) + ps->fds_status[fd].used_events = ps->fds_status[fd].events; + else { + switch (op) { + case EPOLL_CTL_MOD: + epe.events = 0; + do { + /* We init 'epe' every time since epoll_ctl() may modify it + (not declared const and not documented as const). */ + epe.events = 0; + epe.data.fd = fd; + res = epoll_ctl(ps->kp_fd, EPOLL_CTL_DEL, fd, &epe); + } while (res != 0 && errno == EINTR); + ps->fds_status[fd].used_events = 0; + /* Fall through ... */ + case EPOLL_CTL_ADD: { + ps->fds_status[fd].flags |= ERTS_POLL_FD_FLG_USEFLBCK; + ps->no_of_user_fds--; +#if ERTS_POLL_USE_CONCURRENT_UPDATE + if (!*update_fallback) { + *update_fallback = 1; + return 0; + } +#endif + ASSERT(!(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK)); + res = update_fallback_pollset(ps, fd); + ASSERT(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK); + break; + } + case EPOLL_CTL_DEL: { + /* + * Since we use a lazy update approach EPOLL_CTL_DEL will + * frequently fail. This since epoll automatically removes + * a filedescriptor that is closed from the poll set. + */ + ps->fds_status[fd].used_events = 0; + res = 0; + break; + } + default: + fatal_error("%s:%d:update_pollset(): Internal error\n", + __FILE__, __LINE__); + break; + } + } + ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_RST; + return res; +} + +#if ERTS_POLL_USE_CONCURRENT_UPDATE +static int +update_pollset(ErtsPollSet ps, int fd) +{ + int update_fallback = 1; + return conc_update_pollset(ps, fd, &update_fallback); +} +#endif + +#endif /* ERTS_POLL_USE_EPOLL */ + +#endif /* ERTS_POLL_USE_BATCH_UPDATE_POLLSET */ + +#if ERTS_POLL_USE_POLL || ERTS_POLL_USE_SELECT || ERTS_POLL_USE_FALLBACK + +#if ERTS_POLL_USE_FALLBACK +static int update_fallback_pollset(ErtsPollSet ps, int fd) +#else +static int update_pollset(ErtsPollSet ps, int fd) +#endif +{ +#ifdef ERTS_POLL_DEBUG_PRINT +#if ERTS_POLL_USE_FALLBACK + erts_printf("Doing fallback update on fd=%d\n", fd); +#else + erts_printf("Doing update on fd=%d\n", fd); +#endif +#endif + + ASSERT(fd < ps->fds_status_len); +#if ERTS_POLL_USE_FALLBACK + ASSERT(ps->fds_status[fd].used_events + ? (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK) + : (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_USEFLBCK)); +#endif + + if (!need_update(ps, fd)) + return 0; + +#if ERTS_POLL_USE_FALLBACK + ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_RST; +#endif + +#if ERTS_POLL_USE_POLL /* --- poll -------------------------------- */ + if (!ps->fds_status[fd].events) { + int pix = ps->fds_status[fd].pix; + int last_pix; + if (pix < 0) { +#if ERTS_POLL_USE_FALLBACK + ASSERT(!(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK)); +#endif + return -1; + } +#if ERTS_POLL_USE_FALLBACK + ASSERT(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK); +#endif + ps->no_of_user_fds--; + last_pix = --ps->no_poll_fds; + if (pix != last_pix) { + /* Move last pix to this pix */ + ps->poll_fds[pix].fd = ps->poll_fds[last_pix].fd; + ps->poll_fds[pix].events = ps->poll_fds[last_pix].events; + ps->poll_fds[pix].revents = ps->poll_fds[last_pix].revents; + ps->fds_status[ps->poll_fds[pix].fd].pix = pix; + } + /* Clear last pix */ + ps->poll_fds[last_pix].fd = -1; + ps->poll_fds[last_pix].events = (short) 0; + ps->poll_fds[last_pix].revents = (short) 0; + /* Clear this fd status */ + ps->fds_status[fd].pix = -1; + ps->fds_status[fd].used_events = (ErtsPollEvents) 0; +#if ERTS_POLL_USE_FALLBACK + ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_INFLBCK; +#endif + } + else { + int pix = ps->fds_status[fd].pix; + if (pix < 0) { +#if ERTS_POLL_USE_FALLBACK + ASSERT(!(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK) + || fd == ps->kp_fd); +#endif + ps->no_of_user_fds++; + ps->fds_status[fd].pix = pix = ps->no_poll_fds++; + if (pix >= ps->poll_fds_len) + grow_poll_fds(ps, pix); + ps->poll_fds[pix].fd = fd; + ps->fds_status[fd].pix = pix; +#if ERTS_POLL_USE_FALLBACK + ps->fds_status[fd].flags |= ERTS_POLL_FD_FLG_INFLBCK; +#endif + } + +#if ERTS_POLL_USE_FALLBACK + ASSERT(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK); +#endif + + /* Events to be used in next poll */ + ps->poll_fds[pix].events = ev2pollev(ps->fds_status[fd].events); + if (ps->poll_fds[pix].revents) { + /* Remove result events that we should not poll for anymore */ + ps->poll_fds[pix].revents + &= ev2pollev(~(~ps->fds_status[fd].used_events + & ps->fds_status[fd].events)); + } + /* Save events to be used in next poll */ + ps->fds_status[fd].used_events = ps->fds_status[fd].events; + } + return 0; +#elif ERTS_POLL_USE_SELECT /* --- select ------------------------------ */ + { + ErtsPollEvents events = ps->fds_status[fd].events; + if ((ERTS_POLL_EV_IN & events) + != (ERTS_POLL_EV_IN & ps->fds_status[fd].used_events)) { + if (ERTS_POLL_EV_IN & events) { + FD_SET(fd, &ps->input_fds); + } + else { + FD_CLR(fd, &ps->input_fds); + } + } + if ((ERTS_POLL_EV_OUT & events) + != (ERTS_POLL_EV_OUT & ps->fds_status[fd].used_events)) { + if (ERTS_POLL_EV_OUT & events) { + FD_SET(fd, &ps->output_fds); + } + else { + FD_CLR(fd, &ps->output_fds); + } + } + + if (!ps->fds_status[fd].used_events) { + ASSERT(events); + ps->no_of_user_fds++; +#if ERTS_POLL_USE_FALLBACK + ps->no_select_fds++; + ps->fds_status[fd].flags |= ERTS_POLL_FD_FLG_INFLBCK; +#endif + } + else if (!events) { + ASSERT(ps->fds_status[fd].used_events); + ps->no_of_user_fds--; + ps->fds_status[fd].events = events; +#if ERTS_POLL_USE_FALLBACK + ps->no_select_fds--; + ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_INFLBCK; +#endif + } + + ps->fds_status[fd].used_events = events; + + if (events && fd > ps->max_fd) + ps->max_fd = fd; + else if (!events && fd == ps->max_fd) { + int max = ps->max_fd; + for (max = ps->max_fd; max >= 0; max--) + if (ps->fds_status[max].used_events) + break; + ps->max_fd = max; + } + } + return 0; +#endif +} + +#endif /* ERTS_POLL_USE_POLL || ERTS_POLL_USE_SELECT || ERTS_POLL_USE_FALLBACK */ + +#if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE + +static void +handle_update_requests(ErtsPollSet ps) +{ + ErtsPollSetUpdateRequestsBlock *urqbp = &ps->update_requests; +#if ERTS_POLL_USE_BATCH_UPDATE_POLLSET + ErtsPollBatchBuf bb; + setup_batch_buf(ps, &bb); +#endif + + while (urqbp) { + ErtsPollSetUpdateRequestsBlock *free_urqbp = urqbp; + int i; + int len = urqbp->len; + for (i = 0; i < len; i++) { + int fd = urqbp->fds[i]; + ASSERT(fd < ps->fds_status_len); + ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_INURQ; +#if ERTS_POLL_USE_BATCH_UPDATE_POLLSET + batch_update_pollset(ps, fd, &bb); +#else + update_pollset(ps, fd); +#endif + } + + free_urqbp = urqbp; + urqbp = urqbp->next; + + free_update_requests_block(ps, free_urqbp); + + } + +#if ERTS_POLL_USE_BATCH_UPDATE_POLLSET + if (bb.len) + write_batch_buf(ps, &bb); +#endif + + ps->curr_upd_req_block = &ps->update_requests; + +#if ERTS_POLL_USE_DEVPOLL && defined(HARD_DEBUG) + check_poll_status(ps); +#endif + + ERTS_POLLSET_UNSET_HAVE_UPDATE_REQUESTS(ps); +} + +#endif /* ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE */ + +static ERTS_INLINE ErtsPollEvents +poll_control(ErtsPollSet ps, int fd, ErtsPollEvents events, int on, + int *have_set_have_update_requests, + int *do_wake) +{ + ErtsPollEvents new_events; + + if (fd < ps->internal_fd_limit || fd >= max_fds) { + if (fd < 0) { + new_events = ERTS_POLL_EV_ERR; + goto done; + } +#if ERTS_POLL_USE_KERNEL_POLL + if (fd == ps->kp_fd) { + new_events = ERTS_POLL_EV_NVAL; + goto done; + } +#endif +#if ERTS_POLL_USE_WAKEUP_PIPE + if (fd == ps->wake_fds[0] || fd == ps->wake_fds[1]) { + new_events = ERTS_POLL_EV_NVAL; + goto done; + } +#endif + } + + if (fd >= ps->fds_status_len) + grow_fds_status(ps, fd); + + ASSERT(fd < ps->fds_status_len); + + new_events = ps->fds_status[fd].events; + + if (events == 0) { + *do_wake = 0; + goto done; + } + + if (on) + new_events |= events; + else + new_events &= ~events; + + if (new_events == (ErtsPollEvents) 0) { +#if ERTS_POLL_USE_KERNEL_POLL || defined(ERTS_SMP) + ps->fds_status[fd].flags |= ERTS_POLL_FD_FLG_RST; +#endif +#if ERTS_POLL_USE_FALLBACK + ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_USEFLBCK; +#endif + } + + ps->fds_status[fd].events = new_events; + + if (new_events == ps->fds_status[fd].used_events +#if ERTS_POLL_USE_KERNEL_POLL || defined(ERTS_SMP) + && !(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_RST) +#endif + ) { + *do_wake = 0; + goto done; + } + +#if !ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE + if (update_pollset(ps, fd) != 0) + new_events = ERTS_POLL_EV_ERR; +#else /* ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE */ + +#if ERTS_POLL_USE_CONCURRENT_UPDATE + if (ERTS_POLLSET_IS_POLLED(ps)) { + int update_fallback = 0; + conc_update_pollset(ps, fd, &update_fallback); + if (!update_fallback) { + *do_wake = 0; /* no need to wake kernel poller */ + goto done; + } + } +#endif + + enqueue_update_request(ps, fd); + +#ifdef ERTS_SMP + /* + * If new events have been added, we need to wake up the + * polling thread, but if events have been removed we don't. + */ + if ((new_events && (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_RST)) + || (~ps->fds_status[fd].used_events & new_events)) + *do_wake = 1; +#endif /* ERTS_SMP */ + +#endif /* ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE */ + + done: +#ifdef ERTS_POLL_DEBUG_PRINT + erts_printf("0x%x = poll_control(ps, %d, 0x%x, %s) do_wake=%d\n", + (int) new_events, fd, (int) events, (on ? "on" : "off"), *do_wake); +#endif + return new_events; +} + +void +ERTS_POLL_EXPORT(erts_poll_controlv)(ErtsPollSet ps, + ErtsPollControlEntry pcev[], + int len) +{ + int i; + int hshur = 0; + int do_wake; + int final_do_wake = 0; + + ERTS_POLLSET_LOCK(ps); + + for (i = 0; i < len; i++) { + do_wake = 0; + pcev[i].events = poll_control(ps, + pcev[i].fd, + pcev[i].events, + pcev[i].on, + &hshur, + &do_wake); + final_do_wake |= do_wake; + } + +#ifdef ERTS_SMP + if (final_do_wake) + wake_poller(ps); +#endif /* ERTS_SMP */ + + ERTS_POLLSET_UNLOCK(ps); +} + +ErtsPollEvents +ERTS_POLL_EXPORT(erts_poll_control)(ErtsPollSet ps, + ErtsSysFdType fd, + ErtsPollEvents events, + int on, + int* do_wake) /* In: Wake up polling thread */ + /* Out: Poller is woken */ +{ + int hshur = 0; + ErtsPollEvents res; + + ERTS_POLLSET_LOCK(ps); + + res = poll_control(ps, fd, events, on, &hshur, do_wake); + +#ifdef ERTS_SMP + if (*do_wake) { + wake_poller(ps); + } +#endif /* ERTS_SMP */ + + ERTS_POLLSET_UNLOCK(ps); + return res; +} + +/* + * --- Wait on poll set ------------------------------------------------------ + */ + +#if ERTS_POLL_USE_KERNEL_POLL + +static ERTS_INLINE int +save_kp_result(ErtsPollSet ps, ErtsPollResFd pr[], int max_res, int chk_fds_res) +{ + int res = 0; + int i; + int n = chk_fds_res < max_res ? chk_fds_res : max_res; +#if ERTS_POLL_USE_WAKEUP_PIPE + int wake_fd = ps->wake_fds[0]; +#endif + + for (i = 0; i < n; i++) { + +#if ERTS_POLL_USE_EPOLL /* --- epoll ------------------------------- */ + + if (ps->res_events[i].events) { + int fd = ps->res_events[i].data.fd; + int ix; + ErtsPollEvents revents; +#if ERTS_POLL_USE_WAKEUP_PIPE + if (fd == wake_fd) { + cleanup_wakeup_pipe(ps); + continue; + } +#endif + ASSERT(!(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK)); + /* epoll_wait() can repeat the same fd in result array... */ + ix = (int) ps->fds_status[fd].res_ev_ix; + ASSERT(ix >= 0); + if (ix >= res || pr[ix].fd != fd) { + ix = res; + pr[ix].fd = fd; + pr[ix].events = (ErtsPollEvents) 0; + } + + revents = ERTS_POLL_EV_N2E(ps->res_events[i].events); + pr[ix].events |= revents; + if (revents) { + if (res == ix) { + ps->fds_status[fd].res_ev_ix = (unsigned short) ix; + res++; + } + } + } + +#elif ERTS_POLL_USE_KQUEUE /* --- kqueue ------------------------------ */ + + struct kevent *ev; + int fd; + int ix; + + ev = &ps->res_events[i]; + fd = (int) ev->ident; + ASSERT(fd < ps->fds_status_len); + ASSERT(!(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INFLBCK)); + ix = (int) ps->fds_status[fd].res_ev_ix; + + ASSERT(ix >= 0); + if (ix >= res || pr[ix].fd != fd) { + ix = res; + pr[ix].fd = (int) ev->ident; + pr[ix].events = (ErtsPollEvents) 0; + } + + if (ev->filter == EVFILT_READ) { +#if ERTS_POLL_USE_WAKEUP_PIPE + if (fd == wake_fd) { + cleanup_wakeup_pipe(ps); + continue; + } +#endif + pr[ix].events |= ERTS_POLL_EV_IN; + } + else if (ev->filter == EVFILT_WRITE) + pr[ix].events |= ERTS_POLL_EV_OUT; + if (ev->flags & (EV_ERROR|EV_EOF)) { + if ((ev->flags & EV_ERROR) && (((int) ev->data) == EBADF)) + pr[ix].events |= ERTS_POLL_EV_NVAL; + else + pr[ix].events |= ERTS_POLL_EV_ERR; + } + if (pr[ix].events) { + if (res == ix) { + ps->fds_status[fd].res_ev_ix = (unsigned short) ix; + res++; + } + } + +#elif ERTS_POLL_USE_DEVPOLL /* --- devpoll ----------------------------- */ + + if (ps->res_events[i].revents) { + int fd = ps->res_events[i].fd; + ErtsPollEvents revents; +#if ERTS_POLL_USE_WAKEUP_PIPE + if (fd == wake_fd) { + cleanup_wakeup_pipe(ps); + continue; + } +#endif + revents = ERTS_POLL_EV_N2E(ps->res_events[i].events); + pr[res].fd = fd; + pr[res].events = revents; + res++; + } + +#endif + + } + + return res; +} + +#endif /* ERTS_POLL_USE_KERNEL_POLL */ + +#if ERTS_POLL_USE_FALLBACK + +static int +get_kp_results(ErtsPollSet ps, ErtsPollResFd pr[], int max_res) +{ + int res; +#if ERTS_POLL_USE_KQUEUE + struct timespec ts = {0, 0}; +#endif + + if (max_res > ps->res_events_len) + grow_res_events(ps, max_res); + + do { +#if ERTS_POLL_USE_EPOLL + res = epoll_wait(ps->kp_fd, ps->res_events, max_res, 0); +#elif ERTS_POLL_USE_KQUEUE + res = kevent(ps->kp_fd, NULL, 0, ps->res_events, max_res, &ts); +#endif + } while (res < 0 && errno == EINTR); + + if (res < 0) { + fatal_error("%s:%d: %s() failed: %s (%d)\n", + __FILE__, __LINE__, +#if ERTS_POLL_USE_EPOLL + "epoll_wait", +#elif ERTS_POLL_USE_KQUEUE + "kevent", +#endif + erl_errno_id(errno), errno); + } + + return save_kp_result(ps, pr, max_res, res); +} + +#endif /* ERTS_POLL_USE_FALLBACK */ + + + +static ERTS_INLINE int +save_poll_result(ErtsPollSet ps, ErtsPollResFd pr[], int max_res, + int chk_fds_res, int ebadf) +{ +#if ERTS_POLL_USE_DEVPOLL + return save_kp_result(ps, pr, max_res, chk_fds_res); +#elif ERTS_POLL_USE_FALLBACK + if (!ps->fallback_used) + return save_kp_result(ps, pr, max_res, chk_fds_res); + else +#endif /* ERTS_POLL_USE_FALLBACK */ + { + +#if ERTS_POLL_USE_POLL /* --- poll -------------------------------- */ + int res = 0; +#if ERTS_POLL_USE_WAKEUP_PIPE && !ERTS_POLL_USE_FALLBACK + int wake_fd = ps->wake_fds[0]; +#endif + int i, first_ix, end_ix; + + /* + * In order to be somewhat fair, we continue on the poll_fds + * index where we stopped last time. + */ + first_ix = i = ((ps->next_poll_fds_ix < ps->no_poll_fds) + ? ps->next_poll_fds_ix + : 0); + end_ix = ps->no_poll_fds; + + while (1) { + while (i < end_ix && res < max_res) { + if (ps->poll_fds[i].revents != (short) 0) { + int fd = ps->poll_fds[i].fd; + ErtsPollEvents revents; +#if ERTS_POLL_USE_FALLBACK + if (fd == ps->kp_fd) { + res += get_kp_results(ps, &pr[res], max_res-res); + i++; + continue; + } +#elif ERTS_POLL_USE_WAKEUP_PIPE + if (fd == wake_fd) { + cleanup_wakeup_pipe(ps); + i++; + continue; + } +#endif + revents = pollev2ev(ps->poll_fds[i].revents); + pr[res].fd = fd; + pr[res].events = revents; + res++; + } + i++; + } + if (res == max_res || i == first_ix) + break; + ASSERT(i == ps->no_poll_fds); + i = 0; + end_ix = first_ix; + } + + ps->next_poll_fds_ix = i; + return res; + +#elif ERTS_POLL_USE_SELECT /* --- select ------------------------------ */ + int res = 0; +#if ERTS_POLL_USE_WAKEUP_PIPE && !ERTS_POLL_USE_FALLBACK + int wake_fd = ps->wake_fds[0]; +#endif + int fd, first_fd, end_fd; + + /* + * In order to be fair, we continue on the fd where we stopped + * last time. + */ + first_fd = fd = ps->next_sel_fd <= ps->max_fd ? ps->next_sel_fd : 0; + end_fd = ps->max_fd + 1; + + if (!ebadf) { + while (1) { + while (fd < end_fd && res < max_res) { + + pr[res].events = (ErtsPollEvents) 0; + if (FD_ISSET(fd, &ps->res_input_fds)) { +#if ERTS_POLL_USE_FALLBACK + if (fd == ps->kp_fd) { + res += get_kp_results(ps, &pr[res], max_res-res); + fd++; + continue; + } +#elif ERTS_POLL_USE_WAKEUP_PIPE + if (fd == wake_fd) { + cleanup_wakeup_pipe(ps); + fd++; + continue; + } +#endif + pr[res].events |= ERTS_POLL_EV_IN; + } + if (FD_ISSET(fd, &ps->res_output_fds)) + pr[res].events |= ERTS_POLL_EV_OUT; + if (pr[res].events) { + pr[res].fd = fd; + res++; + } + fd++; + } + if (res == max_res || fd == first_fd) + break; + ASSERT(fd == ps->max_fd + 1); + fd = 0; + end_fd = first_fd; + } + } + else { + /* + * Bad file descriptors in poll set. + * + * This only happens when running poorly written + * drivers. This code could be optimized, but we + * don't bother since it should never happen... + */ + while (1) { + while (fd < end_fd && res < max_res) { + if (ps->fds_status[fd].events) { + int sres; + fd_set *iset = NULL; + fd_set *oset = NULL; + if (ps->fds_status[fd].events & ERTS_POLL_EV_IN) { + iset = &ps->res_input_fds; + FD_ZERO(iset); + FD_SET(fd, iset); + } + if (ps->fds_status[fd].events & ERTS_POLL_EV_OUT) { + oset = &ps->res_output_fds; + FD_ZERO(oset); + FD_SET(fd, oset); + + } + do { + /* Initiate 'tv' each time; + select() may modify it */ + SysTimeval tv = {0, 0}; + sres = select(ps->max_fd+1, iset, oset, NULL, &tv); + } while (sres < 0 && errno == EINTR); + if (sres < 0) { +#if ERTS_POLL_USE_FALLBACK + if (fd == ps->kp_fd) { + res += get_kp_results(ps, + &pr[res], + max_res-res); + fd++; + continue; + } +#elif ERTS_POLL_USE_WAKEUP_PIPE + if (fd == wake_fd) { + cleanup_wakeup_pipe(ps); + fd++; + continue; + } +#endif + pr[res].fd = fd; + pr[res].events = ERTS_POLL_EV_NVAL; + res++; + } + else if (sres > 0) { + pr[res].fd = fd; + if (iset && FD_ISSET(fd, iset)) { +#if ERTS_POLL_USE_FALLBACK + if (fd == ps->kp_fd) { + res += get_kp_results(ps, + &pr[res], + max_res-res); + fd++; + continue; + } +#elif ERTS_POLL_USE_WAKEUP_PIPE + if (fd == wake_fd) { + cleanup_wakeup_pipe(ps); + fd++; + continue; + } +#endif + pr[res].events |= ERTS_POLL_EV_IN; + } + if (oset && FD_ISSET(fd, oset)) { + pr[res].events |= ERTS_POLL_EV_OUT; + } + ASSERT(pr[res].events); + res++; + } + } + fd++; + } + if (res == max_res || fd == first_fd) + break; + ASSERT(fd == ps->max_fd + 1); + fd = 0; + end_fd = first_fd; + } + } + ps->next_sel_fd = fd; + return res; +#endif + } +} + +static ERTS_INLINE int +check_fd_events(ErtsPollSet ps, SysTimeval *tv, int max_res, int *ps_locked) +{ + ASSERT(!*ps_locked); + if (ps->no_of_user_fds == 0 && tv->tv_usec == 0 && tv->tv_sec == 0) { + /* Nothing to poll and zero timeout; done... */ + return 0; + } + else { + long timeout = tv->tv_sec*1000 + tv->tv_usec/1000; + ASSERT(timeout >= 0); + erts_smp_atomic_set(&ps->timeout, timeout); +#if ERTS_POLL_USE_FALLBACK + if (!(ps->fallback_used = ERTS_POLL_NEED_FALLBACK(ps))) { + +#if ERTS_POLL_USE_EPOLL /* --- epoll ------------------------------- */ + if (timeout > INT_MAX) + timeout = INT_MAX; + if (max_res > ps->res_events_len) + grow_res_events(ps, max_res); + return epoll_wait(ps->kp_fd, ps->res_events, max_res, (int)timeout); +#elif ERTS_POLL_USE_KQUEUE /* --- kqueue ------------------------------ */ + struct timespec ts; + ts.tv_sec = tv->tv_sec; + ts.tv_nsec = tv->tv_usec*1000; + if (max_res > ps->res_events_len) + grow_res_events(ps, max_res); + return kevent(ps->kp_fd, NULL, 0, ps->res_events, max_res, &ts); +#endif /* ----------------------------------------- */ + + } + else /* use fallback (i.e. poll() or select()) */ +#endif /* ERTS_POLL_USE_FALLBACK */ + { + +#if ERTS_POLL_USE_DEVPOLL /* --- devpoll ----------------------------- */ + /* + * The ioctl() will fail with EINVAL on Solaris 10 if dp_nfds + * is set too high. dp_nfds should not be set greater than + * the maximum number of file descriptors in the poll set. + */ + struct dvpoll poll_res; + int nfds = ps->no_of_user_fds; +#ifdef ERTS_SMP + nfds++; /* Wakeup pipe */ +#endif + if (timeout > INT_MAX) + timeout = INT_MAX; + poll_res.dp_nfds = nfds < max_res ? nfds : max_res; + if (poll_res.dp_nfds > ps->res_events_len) + grow_res_events(ps, poll_res.dp_nfds); + poll_res.dp_fds = ps->res_events; + poll_res.dp_timeout = (int) timeout; + return ioctl(ps->kp_fd, DP_POLL, &poll_res); +#elif ERTS_POLL_USE_POLL /* --- poll -------------------------------- */ + if (timeout > INT_MAX) + timeout = INT_MAX; + return poll(ps->poll_fds, ps->no_poll_fds, (int) timeout); +#elif ERTS_POLL_USE_SELECT /* --- select ------------------------------ */ + int res; + ps->res_input_fds = ps->input_fds; + ps->res_output_fds = ps->output_fds; + res = select(ps->max_fd + 1, + &ps->res_input_fds, + &ps->res_output_fds, + NULL, + tv); +#ifdef ERTS_SMP + if (res < 0 + && errno == EBADF + && ERTS_POLLSET_HAVE_UPDATE_REQUESTS(ps)) { + /* + * This may have happened because another thread deselected + * a fd in our poll set and then closed it, i.e. the driver + * behaved correctly. We wan't to avoid looking for a bad + * fd, that may even not exist anymore. Therefore, handle + * update requests and try again. + * + * We don't know how much of the timeout is left; therfore, + * we use a zero timeout. If no error occur and no events + * have triggered, we fake an EAGAIN error and let the caller + * restart us. + */ + SysTimeval zero_tv = {0, 0}; + *ps_locked = 1; + ERTS_POLLSET_LOCK(ps); + handle_update_requests(ps); + res = select(ps->max_fd + 1, + &ps->res_input_fds, + &ps->res_output_fds, + NULL, + &zero_tv); + if (res == 0) { + errno = EAGAIN; + res = -1; + } + } +#endif /* ERTS_SMP */ + return res; +#endif /* ----------------------------------------- */ + } + } +} + +int +ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet ps, + ErtsPollResFd pr[], + int *len, + SysTimeval *utvp) +{ + int res, no_fds; + int ebadf = 0; + int ps_locked; + SysTimeval *tvp; + SysTimeval itv; + + no_fds = *len; +#ifdef ERTS_POLL_MAX_RES + if (no_fds >= ERTS_POLL_MAX_RES) + no_fds = ERTS_POLL_MAX_RES; +#endif + + *len = 0; + + ASSERT(utvp); + + tvp = utvp; + +#ifdef ERTS_POLL_DEBUG_PRINT + erts_printf("Entering erts_poll_wait(), timeout=%d\n", + (int) tv->tv_sec*1000 + tv->tv_usec/1000); +#endif + + ERTS_POLLSET_UNSET_POLLER_WOKEN(ps); + if (ERTS_POLLSET_SET_POLLED_CHK(ps)) { + res = EINVAL; /* Another thread is in erts_poll_wait() + on this pollset... */ + goto done; + } + + if (ERTS_POLLSET_IS_INTERRUPTED(ps)) { + /* Interrupt use zero timeout */ + itv.tv_sec = 0; + itv.tv_usec = 0; + tvp = &itv; + } + +#if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE + if (ERTS_POLLSET_HAVE_UPDATE_REQUESTS(ps)) { + ERTS_POLLSET_LOCK(ps); + handle_update_requests(ps); + ERTS_POLLSET_UNLOCK(ps); + } +#endif + + ps_locked = 0; + res = check_fd_events(ps, tvp, no_fds, &ps_locked); + + ERTS_POLLSET_SET_POLLER_WOKEN(ps); + + if (res == 0) { + res = ETIMEDOUT; + } + else if (res < 0) { +#if ERTS_POLL_USE_SELECT + if (errno == EBADF) { + ebadf = 1; + goto save_results; + } +#endif + res = errno; + } + else { +#if ERTS_POLL_USE_SELECT + save_results: +#endif + +#ifdef ERTS_SMP + if (!ps_locked) { + ps_locked = 1; + ERTS_POLLSET_LOCK(ps); + } +#endif + + no_fds = save_poll_result(ps, pr, no_fds, res, ebadf); + +#ifdef HARD_DEBUG + check_poll_result(pr, no_fds); +#endif + + res = (no_fds == 0 + ? (ERTS_POLLSET_UNSET_INTERRUPTED_CHK(ps) ? EINTR : EAGAIN) + : 0); + *len = no_fds; + } + +#ifdef ERTS_SMP + if (ps_locked) + ERTS_POLLSET_UNLOCK(ps); + ERTS_POLLSET_UNSET_POLLED(ps); +#endif + + done: + erts_smp_atomic_set(&ps->timeout, LONG_MAX); +#ifdef ERTS_POLL_DEBUG_PRINT + erts_printf("Leaving %s = erts_poll_wait()\n", + res == 0 ? "0" : erl_errno_id(res)); +#endif + + return res; +} + +/* + * --- Interrupt a thread doing erts_poll_wait() ----------------------------- + */ + +void +ERTS_POLL_EXPORT(erts_poll_interrupt)(ErtsPollSet ps, int set) +{ + /* + * NOTE: This function might be called from signal handlers in the + * non-smp case; therefore, it has to be async-signal safe in + * the non-smp case. + */ + if (set) { + ERTS_POLLSET_SET_INTERRUPTED(ps); +#if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT || defined(ERTS_SMP) + wake_poller(ps); +#endif + } + else { + ERTS_POLLSET_UNSET_INTERRUPTED(ps); + } +} + +/* + * erts_poll_interrupt_timed(): + * If 'set' != 0, interrupt thread blocked in erts_poll_wait() if it + * is not guaranteed that it will timeout before 'msec' milli seconds. + */ +void +ERTS_POLL_EXPORT(erts_poll_interrupt_timed)(ErtsPollSet ps, int set, long msec) +{ + if (set) { + if (erts_smp_atomic_read(&ps->timeout) > msec) { + ERTS_POLLSET_SET_INTERRUPTED(ps); +#if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT || defined(ERTS_SMP) + wake_poller(ps); +#endif + } +#ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS + else { + if (ERTS_POLLSET_IS_POLLED(ps)) + erts_smp_atomic_inc(&ps->no_avoided_wakeups); + erts_smp_atomic_inc(&ps->no_avoided_interrupts); + } + erts_smp_atomic_inc(&ps->no_interrupt_timed); +#endif + } + else { + ERTS_POLLSET_UNSET_INTERRUPTED(ps); + } +} + +int +ERTS_POLL_EXPORT(erts_poll_max_fds)(void) +{ + return max_fds; +} +/* + * --- Initialization -------------------------------------------------------- + */ + +#ifdef VXWORKS +extern int erts_vxworks_max_files; +#endif + +void +ERTS_POLL_EXPORT(erts_poll_init)(void) +{ + erts_smp_spinlock_init(&pollsets_lock, "pollsets_lock"); + pollsets = NULL; + + errno = 0; + +#if defined(VXWORKS) + max_fds = erts_vxworks_max_files; +#elif !defined(NO_SYSCONF) + max_fds = sysconf(_SC_OPEN_MAX); +#elif ERTS_POLL_USE_SELECT + max_fds = NOFILE; +#else + max_fds = OPEN_MAX; +#endif + +#if ERTS_POLL_USE_SELECT && defined(FD_SETSIZE) + if (max_fds > FD_SETSIZE) + max_fds = FD_SETSIZE; +#endif + + if (max_fds < 0) + fatal_error("erts_poll_init(): Failed to get max number of files: %s\n", + erl_errno_id(errno)); + +#ifdef ERTS_POLL_DEBUG_PRINT + print_misc_debug_info(); +#endif +} + +ErtsPollSet +ERTS_POLL_EXPORT(erts_poll_create_pollset)(void) +{ +#if ERTS_POLL_USE_KERNEL_POLL + int kp_fd; +#endif + ErtsPollSet ps = erts_alloc(ERTS_ALC_T_POLLSET, + sizeof(struct ErtsPollSet_)); + ps->internal_fd_limit = 0; + ps->fds_status = NULL; + ps->fds_status_len = 0; + ps->no_of_user_fds = 0; +#if ERTS_POLL_USE_KERNEL_POLL + ps->kp_fd = -1; +#if ERTS_POLL_USE_EPOLL + kp_fd = epoll_create(256); + ps->res_events_len = 0; + ps->res_events = NULL; +#elif ERTS_POLL_USE_DEVPOLL + kp_fd = open("/dev/poll", O_RDWR); + ps->res_events_len = 0; + ps->res_events = NULL; +#elif ERTS_POLL_USE_KQUEUE + kp_fd = kqueue(); + ps->res_events_len = 0; + ps->res_events = NULL; +#endif + if (kp_fd < 0) + fatal_error("erts_poll_create_pollset(): Failed to " +#if ERTS_POLL_USE_EPOLL + "create epoll set" +#elif ERTS_POLL_USE_DEVPOLL + "to open /dev/poll" +#elif ERTS_POLL_USE_KQUEUE + "create kqueue" +#endif + ": %s (%d)\n", + erl_errno_id(errno), errno); +#endif /* ERTS_POLL_USE_KERNEL_POLL */ +#if ERTS_POLL_USE_BATCH_UPDATE_POLLSET + /* res_events is also used as write buffer */ + grow_res_events(ps, ERTS_POLL_MIN_BATCH_BUF_SIZE); +#endif +#if ERTS_POLL_USE_POLL + ps->next_poll_fds_ix = 0; + ps->no_poll_fds = 0; + ps->poll_fds_len = 0; + ps->poll_fds = NULL; +#elif ERTS_POLL_USE_SELECT + ps->next_sel_fd = 0; + ps->max_fd = -1; +#if ERTS_POLL_USE_FALLBACK + ps->no_select_fds = 0; +#endif + FD_ZERO(&ps->input_fds); + FD_ZERO(&ps->res_input_fds); + FD_ZERO(&ps->output_fds); + FD_ZERO(&ps->res_output_fds); +#endif +#if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE + ps->update_requests.next = NULL; + ps->update_requests.len = 0; + ps->curr_upd_req_block = &ps->update_requests; + erts_smp_atomic_init(&ps->have_update_requests, 0); +#endif +#ifdef ERTS_SMP + erts_smp_atomic_init(&ps->polled, 0); + erts_smp_atomic_init(&ps->woken, 0); + erts_smp_mtx_init(&ps->mtx, "pollset"); +#elif ERTS_POLL_ASYNC_INTERRUPT_SUPPORT + ps->woken = 0; +#endif +#if ERTS_POLL_USE_WAKEUP_PIPE + create_wakeup_pipe(ps); +#endif +#if ERTS_POLL_USE_FALLBACK + if (kp_fd >= ps->fds_status_len) + grow_fds_status(ps, kp_fd); + /* Force kernel poll fd into fallback (poll/select) set */ + ps->fds_status[kp_fd].flags + |= ERTS_POLL_FD_FLG_INFLBCK|ERTS_POLL_FD_FLG_USEFLBCK; + { + int do_wake = 0; + ERTS_POLL_EXPORT(erts_poll_control)(ps, kp_fd, ERTS_POLL_EV_IN, 1, + &do_wake); + } +#endif +#if ERTS_POLL_USE_KERNEL_POLL + if (ps->internal_fd_limit <= kp_fd) + ps->internal_fd_limit = kp_fd + 1; + ps->kp_fd = kp_fd; +#endif +#if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT && !defined(ERTS_SMP) + ps->interrupt = 0; +#else + erts_smp_atomic_init(&ps->interrupt, 0); +#endif + erts_smp_atomic_init(&ps->timeout, LONG_MAX); +#ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS + erts_smp_atomic_init(&ps->no_avoided_wakeups, 0); + erts_smp_atomic_init(&ps->no_avoided_interrupts, 0); + erts_smp_atomic_init(&ps->no_interrupt_timed, 0); +#endif +#if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE + handle_update_requests(ps); +#endif +#if ERTS_POLL_USE_FALLBACK + ps->fallback_used = 0; +#endif + ps->no_of_user_fds = 0; /* Don't count wakeup pipe and fallback fd */ + + erts_smp_spin_lock(&pollsets_lock); + ps->next = pollsets; + pollsets = ps; + erts_smp_spin_unlock(&pollsets_lock); + + return ps; +} + +void +ERTS_POLL_EXPORT(erts_poll_destroy_pollset)(ErtsPollSet ps) +{ + + if (ps->fds_status) + erts_free(ERTS_ALC_T_FD_STATUS, (void *) ps->fds_status); + +#if ERTS_POLL_USE_EPOLL + if (ps->kp_fd >= 0) + close(ps->kp_fd); + if (ps->res_events) + erts_free(ERTS_ALC_T_POLL_RES_EVS, (void *) ps->res_events); +#elif ERTS_POLL_USE_DEVPOLL + if (ps->kp_fd >= 0) + close(ps->kp_fd); + if (ps->res_events) + erts_free(ERTS_ALC_T_POLL_RES_EVS, (void *) ps->res_events); +#elif ERTS_POLL_USE_POLL + if (ps->poll_fds) + erts_free(ERTS_ALC_T_POLL_FDS, (void *) ps->poll_fds); +#elif ERTS_POLL_USE_SELECT +#endif +#if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE + { + ErtsPollSetUpdateRequestsBlock *urqbp = ps->update_requests.next; + while (urqbp) { + ErtsPollSetUpdateRequestsBlock *free_urqbp = urqbp; + urqbp = urqbp->next; + free_update_requests_block(ps, free_urqbp); + } + } +#endif +#ifdef ERTS_SMP + erts_smp_mtx_destroy(&ps->mtx); +#endif +#if ERTS_POLL_USE_WAKEUP_PIPE + if (ps->wake_fds[0] >= 0) + close(ps->wake_fds[0]); + if (ps->wake_fds[1] >= 0) + close(ps->wake_fds[1]); +#endif + + erts_smp_spin_lock(&pollsets_lock); + if (ps == pollsets) + pollsets = pollsets->next; + else { + ErtsPollSet prev_ps; + for (prev_ps = pollsets; ps != prev_ps->next; prev_ps = prev_ps->next); + ASSERT(ps == prev_ps->next); + prev_ps->next = ps->next; + } + erts_smp_spin_unlock(&pollsets_lock); + + erts_free(ERTS_ALC_T_POLLSET, (void *) ps); +} + +/* + * --- Info ------------------------------------------------------------------ + */ + +void +ERTS_POLL_EXPORT(erts_poll_info)(ErtsPollSet ps, ErtsPollInfo *pip) +{ +#if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE + int pending_updates; +#endif + Uint size = 0; + + ERTS_POLLSET_LOCK(ps); + + size += sizeof(struct ErtsPollSet_); + size += ps->fds_status_len*sizeof(ErtsFdStatus); + +#if ERTS_POLL_USE_EPOLL + size += ps->res_events_len*sizeof(struct epoll_event); +#elif ERTS_POLL_USE_DEVPOLL + size += ps->res_events_len*sizeof(struct pollfd); +#elif ERTS_POLL_USE_KQUEUE + size += ps->res_events_len*sizeof(struct kevent); +#endif + +#if ERTS_POLL_USE_POLL + size += ps->poll_fds_len*sizeof(struct pollfd); +#elif ERTS_POLL_USE_SELECT +#endif + +#if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE + { + ErtsPollSetUpdateRequestsBlock *urqbp = ps->update_requests.next; + pending_updates = ps->update_requests.len; + while (urqbp) { + size += sizeof(ErtsPollSetUpdateRequestsBlock); + pending_updates += urqbp->len; + } + } +#endif + + pip->primary = +#if ERTS_POLL_USE_KQUEUE + "kqueue" +#elif ERTS_POLL_USE_EPOLL + "epoll" +#elif ERTS_POLL_USE_DEVPOLL + "/dev/poll" +#elif ERTS_POLL_USE_POLL + "poll" +#elif ERTS_POLL_USE_SELECT + "select" +#endif + ; + + pip->fallback = +#if !ERTS_POLL_USE_FALLBACK + NULL +#elif ERTS_POLL_USE_POLL + "poll" +#elif ERTS_POLL_USE_SELECT + "select" +#endif + ; + + pip->kernel_poll = +#if !ERTS_POLL_USE_KERNEL_POLL + NULL +#elif ERTS_POLL_USE_KQUEUE + "kqueue" +#elif ERTS_POLL_USE_EPOLL + "epoll" +#elif ERTS_POLL_USE_DEVPOLL + "/dev/poll" +#endif + ; + + pip->memory_size = size; + + pip->poll_set_size = ps->no_of_user_fds; +#ifdef ERTS_SMP + pip->poll_set_size++; /* Wakeup pipe */ +#endif + + pip->fallback_poll_set_size = +#if !ERTS_POLL_USE_FALLBACK + 0 +#elif ERTS_POLL_USE_POLL + ps->no_poll_fds +#elif ERTS_POLL_USE_SELECT + ps->no_select_fds +#endif + ; + +#if ERTS_POLL_USE_FALLBACK + /* If only kp_fd is in fallback poll set we don't use fallback... */ + if (pip->fallback_poll_set_size == 1) + pip->fallback_poll_set_size = 0; + else + pip->poll_set_size++; /* kp_fd */ +#endif + + pip->lazy_updates = +#if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE + 1 +#else + 0 +#endif + ; + + pip->pending_updates = +#if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE + pending_updates +#else + 0 +#endif + ; + + pip->batch_updates = +#if ERTS_POLL_USE_BATCH_UPDATE_POLLSET + 1 +#else + 0 +#endif + ; + + pip->concurrent_updates = +#if ERTS_POLL_USE_CONCURRENT_UPDATE + 1 +#else + 0 +#endif + ; + + pip->max_fds = max_fds; + +#ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS + pip->no_avoided_wakeups = erts_smp_atomic_read(&ps->no_avoided_wakeups); + pip->no_avoided_interrupts = erts_smp_atomic_read(&ps->no_avoided_interrupts); + pip->no_interrupt_timed = erts_smp_atomic_read(&ps->no_interrupt_timed); +#endif + + ERTS_POLLSET_UNLOCK(ps); + +} + +/* + * Fatal error... + */ + +#ifndef ERTS_GOT_SIGUSR1 +# define ERTS_GOT_SIGUSR1 0 +#endif + +static void +fatal_error(char *format, ...) +{ + va_list ap; + + if (ERTS_IS_CRASH_DUMPING || ERTS_GOT_SIGUSR1) { + /* + * Crash dump writing and reception of sigusr1 (which will + * result in a crash dump) closes all file descriptors. This + * typically results in a fatal error for erts_poll() (wakeup + * pipes and kernel poll fds are closed). + * + * We ignore the error and let the crash dump writing continue... + */ + return; + } + va_start(ap, format); + erts_vfprintf(stderr, format, ap); + va_end(ap); + abort(); +} + +static void +fatal_error_async_signal_safe(char *error_str) +{ + if (ERTS_IS_CRASH_DUMPING || ERTS_GOT_SIGUSR1) { + /* See comment above in fatal_error() */ + return; + } + if (error_str) { + int len = 0; + while (error_str[len]) + len++; + if (len) + (void) write(2, error_str, len); /* async signal safe */ + } + abort(); +} + +/* + * --- Debug ----------------------------------------------------------------- + */ + +void +ERTS_POLL_EXPORT(erts_poll_get_selected_events)(ErtsPollSet ps, + ErtsPollEvents ev[], + int len) +{ + int fd; + ERTS_POLLSET_LOCK(ps); + for (fd = 0; fd < len; fd++) { + if (fd >= ps->fds_status_len) + ev[fd] = 0; + else { + ev[fd] = ps->fds_status[fd].events; +#if ERTS_POLL_USE_WAKEUP_PIPE + if (fd == ps->wake_fds[0] || fd == ps->wake_fds[1]) + ev[fd] |= ERTS_POLL_EV_NVAL; +#endif +#if ERTS_POLL_USE_KERNEL_POLL + if (fd == ps->kp_fd) + ev[fd] |= ERTS_POLL_EV_NVAL; +#endif + } + } + ERTS_POLLSET_UNLOCK(ps); + +} + +#ifdef HARD_DEBUG + +static void +check_poll_result(ErtsPollResFd pr[], int len) +{ + int i, j; + + for (i = 0; i < len; i++) { + ASSERT(pr[i].fd >= 0); + ASSERT(pr[i].fd < max_fds); + for (j = 0; j < len; j++) { + ASSERT(i == j || pr[i].fd != pr[j].fd); + } + } +} + + +#if ERTS_POLL_USE_DEVPOLL + +static void +check_poll_status(ErtsPollSet ps) +{ + int i; + for (i = 0; i < ps->fds_status_len; i++) { + int ires; + struct pollfd dp_fd; + short events = ERTS_POLL_EV_E2N(ps->fds_status[i].events); + + dp_fd.fd = i; + dp_fd.events = (short) 0; + dp_fd.revents = (short) 0; + + ires = ioctl(ps->kp_fd, DP_ISPOLLED, &dp_fd); + + if (ires == 0) { + ASSERT(!events); + } + else if (ires == 1) { + ASSERT(events); + ASSERT(events == dp_fd.revents); + } + else { + ASSERT(0); + } + ASSERT(dp_fd.fd == i); + ASSERT(ps->fds_status[i].events == ps->fds_status[i].used_events); + } +} + +#endif /* ERTS_POLL_USE_DEVPOLL */ +#endif /* HARD_DEBUG */ + +#ifdef ERTS_POLL_DEBUG_PRINT +static void +print_misc_debug_info(void) +{ + erts_printf("erts_poll using: %s lazy_updates:%s batch_updates:%s\n", +#if ERTS_POLL_USE_KQUEUE + "kqueue" +#elif ERTS_POLL_USE_EPOLL + "epoll" +#elif ERTS_POLL_USE_DEVPOLL + "/dev/poll" +#endif +#if ERTS_POLL_USE_FALLBACK + "-" +#endif +#if ERTS_POLL_USE_POLL + "poll" +#elif ERTS_POLL_USE_SELECT + "select" +#endif + , +#if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE + "true" +#else + "false" +#endif + , +#if ERTS_POLL_USE_BATCH_UPDATE_POLLSET + "true" +#else + "false" +#endif + ); + + erts_printf("ERTS_POLL_EV_IN=0x%x\n" + "ERTS_POLL_EV_OUT=0x%x\n" + "ERTS_POLL_EV_NVAL=0x%x\n" + "ERTS_POLL_EV_ERR=0x%x\n", + ERTS_POLL_EV_IN, + ERTS_POLL_EV_OUT, + ERTS_POLL_EV_NVAL, + ERTS_POLL_EV_ERR); + +#ifdef FD_SETSIZE + erts_printf("FD_SETSIZE=%d\n", FD_SETSIZE); +#endif +} + +#endif diff --git a/erts/emulator/sys/common/erl_poll.h b/erts/emulator/sys/common/erl_poll.h new file mode 100644 index 0000000000..725a77a152 --- /dev/null +++ b/erts/emulator/sys/common/erl_poll.h @@ -0,0 +1,246 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2006-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +/* + * Description: Poll interface suitable for ERTS with or without + * SMP support. + * + * Author: Rickard Green + */ + +#ifndef ERL_POLL_H__ +#define ERL_POLL_H__ + +#include "sys.h" + +#if 0 +#define ERTS_POLL_COUNT_AVOIDED_WAKEUPS +#endif + +#ifdef ERTS_ENABLE_KERNEL_POLL +# if defined(ERTS_KERNEL_POLL_VERSION) +# define ERTS_POLL_EXPORT(FUNC) FUNC ## _kp +# else +# define ERTS_POLL_EXPORT(FUNC) FUNC ## _nkp +# undef ERTS_POLL_DISABLE_KERNEL_POLL +# define ERTS_POLL_DISABLE_KERNEL_POLL +# endif +#else +# define ERTS_POLL_EXPORT(FUNC) FUNC +# undef ERTS_POLL_DISABLE_KERNEL_POLL +# define ERTS_POLL_DISABLE_KERNEL_POLL +#endif + +#ifdef ERTS_POLL_DISABLE_KERNEL_POLL +# undef HAVE_SYS_EPOLL_H +# undef HAVE_SYS_EVENT_H +# undef HAVE_SYS_DEVPOLL_H +#endif + +#undef ERTS_POLL_USE_KERNEL_POLL +#define ERTS_POLL_USE_KERNEL_POLL 0 + +#undef ERTS_POLL_USE_KQUEUE +#define ERTS_POLL_USE_KQUEUE 0 +#undef ERTS_POLL_USE_EPOLL +#define ERTS_POLL_USE_EPOLL 0 +#undef ERTS_POLL_USE_DEVPOLL +#define ERTS_POLL_USE_DEVPOLL 0 +#undef ERTS_POLL_USE_POLL +#define ERTS_POLL_USE_POLL 0 +#undef ERTS_POLL_USE_SELECT +#define ERTS_POLL_USE_SELECT 0 + +#if defined(HAVE_SYS_EVENT_H) +# undef ERTS_POLL_USE_KQUEUE +# define ERTS_POLL_USE_KQUEUE 1 +# undef ERTS_POLL_USE_KERNEL_POLL +# define ERTS_POLL_USE_KERNEL_POLL 1 +#elif defined(HAVE_SYS_EPOLL_H) +# undef ERTS_POLL_USE_EPOLL +# define ERTS_POLL_USE_EPOLL 1 +# undef ERTS_POLL_USE_KERNEL_POLL +# define ERTS_POLL_USE_KERNEL_POLL 1 +#elif defined(HAVE_SYS_DEVPOLL_H) +# undef ERTS_POLL_USE_DEVPOLL +# define ERTS_POLL_USE_DEVPOLL 1 +# undef ERTS_POLL_USE_KERNEL_POLL +# define ERTS_POLL_USE_KERNEL_POLL 1 +#endif + +#define ERTS_POLL_USE_FALLBACK (ERTS_POLL_USE_KQUEUE || ERTS_POLL_USE_EPOLL) + +#if !ERTS_POLL_USE_KERNEL_POLL || ERTS_POLL_USE_FALLBACK +# if defined(ERTS_USE_POLL) +# undef ERTS_POLL_USE_POLL +# define ERTS_POLL_USE_POLL 1 +# elif !defined(__WIN32__) +# undef ERTS_POLL_USE_SELECT +# define ERTS_POLL_USE_SELECT 1 +# endif +#endif + +typedef Uint32 ErtsPollEvents; +#undef ERTS_POLL_EV_E2N + +#if defined(__WIN32__) /* --- win32 ------------------------------- */ + +#define ERTS_POLL_EV_IN 1 +#define ERTS_POLL_EV_OUT 2 +#define ERTS_POLL_EV_ERR 4 +#define ERTS_POLL_EV_NVAL 8 + +#elif ERTS_POLL_USE_EPOLL /* --- epoll ------------------------------- */ + +#include <sys/epoll.h> + +#define ERTS_POLL_EV_E2N(EV) \ + ((__uint32_t) (EV)) +#define ERTS_POLL_EV_N2E(EV) \ + ((ErtsPollEvents) (EV)) + +#define ERTS_POLL_EV_IN ERTS_POLL_EV_N2E(EPOLLIN) +#define ERTS_POLL_EV_OUT ERTS_POLL_EV_N2E(EPOLLOUT) +#define ERTS_POLL_EV_NVAL ERTS_POLL_EV_N2E(EPOLLET) +#define ERTS_POLL_EV_ERR ERTS_POLL_EV_N2E(EPOLLERR|EPOLLHUP) + +#elif ERTS_POLL_USE_DEVPOLL /* --- devpoll ----------------------------- */ + +#include <sys/devpoll.h> + +#define ERTS_POLL_EV_E2N(EV) \ + ((short) ((EV) & ~((~((ErtsPollEvents) 0)) << 8*SIZEOF_SHORT))) +#define ERTS_POLL_EV_N2E(EV) \ + ((ErtsPollEvents) ((unsigned short) (EV))) + +#define ERTS_POLL_EV_IN ERTS_POLL_EV_N2E(POLLIN) +#define ERTS_POLL_EV_OUT ERTS_POLL_EV_N2E(POLLOUT) +#define ERTS_POLL_EV_NVAL ERTS_POLL_EV_N2E(POLLNVAL) +#define ERTS_POLL_EV_ERR ERTS_POLL_EV_N2E(POLLERR|POLLHUP) + +#elif ERTS_POLL_USE_KQUEUE /* --- kqueue ------------------------------ */ +/* Kqueue use fallback defines (poll() or select()) */ +#endif + +#if ERTS_POLL_USE_POLL /* --- poll -------------------------------- */ + +#include <poll.h> + +#define ERTS_POLL_EV_NKP_E2N(EV) \ + ((short) ((EV) & ~((~((ErtsPollEvents) 0)) << 8*SIZEOF_SHORT))) +#define ERTS_POLL_EV_NKP_N2E(EV) \ + ((ErtsPollEvents) ((unsigned short) (EV))) + +/* At least on FreeBSD, we need POLLRDNORM for normal files, not POLLIN. */ +/* Whether this is a bug in FreeBSD, I don't know. */ +#ifdef POLLRDNORM +#define ERTS_POLL_EV_NKP_IN ERTS_POLL_EV_N2E(POLLIN|POLLRDNORM) +#else +#define ERTS_POLL_EV_NKP_IN ERTS_POLL_EV_N2E(POLLIN) +#endif +#define ERTS_POLL_EV_NKP_OUT ERTS_POLL_EV_N2E(POLLOUT) +#define ERTS_POLL_EV_NKP_NVAL ERTS_POLL_EV_N2E(POLLNVAL) +#define ERTS_POLL_EV_NKP_ERR ERTS_POLL_EV_N2E(POLLERR|POLLHUP) + +#elif ERTS_POLL_USE_SELECT /* --- select ------------------------------ */ + +#define ERTS_POLL_EV_NKP_E2N(EV) (EV) +#define ERTS_POLL_EV_NKP_N2E(EV) (EV) + +#define ERTS_POLL_EV_NKP_IN (((ErtsPollEvents) 1) << 0) +#define ERTS_POLL_EV_NKP_OUT (((ErtsPollEvents) 1) << 1) +#define ERTS_POLL_EV_NKP_NVAL (((ErtsPollEvents) 1) << 2) +#define ERTS_POLL_EV_NKP_ERR (((ErtsPollEvents) 1) << 3) + +#endif /* ----------------------------------------- */ + + +#if !defined(ERTS_POLL_EV_E2N) && defined(ERTS_POLL_EV_NKP_E2N) +/* poll(), select(), and kqueue() */ + +#define ERTS_POLL_EV_E2N(EV) ERTS_POLL_EV_NKP_E2N((EV)) +#define ERTS_POLL_EV_N2E(EV) ERTS_POLL_EV_NKP_N2E((EV)) + +#define ERTS_POLL_EV_IN ERTS_POLL_EV_NKP_IN +#define ERTS_POLL_EV_OUT ERTS_POLL_EV_NKP_OUT +#define ERTS_POLL_EV_NVAL ERTS_POLL_EV_NKP_NVAL +#define ERTS_POLL_EV_ERR ERTS_POLL_EV_NKP_ERR + +#endif + +typedef struct ErtsPollSet_ *ErtsPollSet; + +typedef struct { + ErtsSysFdType fd; + ErtsPollEvents events; + int on; +} ErtsPollControlEntry; + +typedef struct { + ErtsSysFdType fd; + ErtsPollEvents events; +} ErtsPollResFd; + +typedef struct { + char *primary; + char *fallback; + char *kernel_poll; + Uint memory_size; + int poll_set_size; + int fallback_poll_set_size; + int lazy_updates; + int pending_updates; + int batch_updates; + int concurrent_updates; + int max_fds; +#ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS + long no_avoided_wakeups; + long no_avoided_interrupts; + long no_interrupt_timed; +#endif +} ErtsPollInfo; + +void ERTS_POLL_EXPORT(erts_poll_interrupt)(ErtsPollSet, + int); +void ERTS_POLL_EXPORT(erts_poll_interrupt_timed)(ErtsPollSet, + int, + long); +ErtsPollEvents ERTS_POLL_EXPORT(erts_poll_control)(ErtsPollSet, + ErtsSysFdType, + ErtsPollEvents, + int on, + int* wake_poller); +void ERTS_POLL_EXPORT(erts_poll_controlv)(ErtsPollSet, + ErtsPollControlEntry [], + int on); +int ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet, + ErtsPollResFd [], + int *, + SysTimeval *); +int ERTS_POLL_EXPORT(erts_poll_max_fds)(void); +void ERTS_POLL_EXPORT(erts_poll_info)(ErtsPollSet, + ErtsPollInfo *); +ErtsPollSet ERTS_POLL_EXPORT(erts_poll_create_pollset)(void); +void ERTS_POLL_EXPORT(erts_poll_destroy_pollset)(ErtsPollSet); +void ERTS_POLL_EXPORT(erts_poll_init)(void); +void ERTS_POLL_EXPORT(erts_poll_get_selected_events)(ErtsPollSet, + ErtsPollEvents [], + int); + +#endif /* #ifndef ERL_POLL_H__ */ diff --git a/erts/emulator/sys/unix/driver_int.h b/erts/emulator/sys/unix/driver_int.h new file mode 100644 index 0000000000..a7ee8087ab --- /dev/null +++ b/erts/emulator/sys/unix/driver_int.h @@ -0,0 +1,41 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 1997-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ +/* + * System dependant driver declarations + */ + +#ifndef __DRIVER_INT_H__ +#define __DRIVER_INT_H__ + +#ifdef HAVE_SYS_UIO_H +#include <sys/types.h> +#include <sys/uio.h> + +typedef struct iovec SysIOVec; + +#else + +typedef struct { + char* iov_base; + int iov_len; +} SysIOVec; + +#endif + +#endif diff --git a/erts/emulator/sys/unix/erl9_start.c b/erts/emulator/sys/unix/erl9_start.c new file mode 100644 index 0000000000..578062d7e2 --- /dev/null +++ b/erts/emulator/sys/unix/erl9_start.c @@ -0,0 +1,130 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2002-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif +#include "sys.h" +#include "erl_vm.h" +#include "global.h" +#include <stdio.h> +#include <string.h> +#include <stdlib.h> + +/* + * XXX This is a temporary dummy to make sys.c happy until we'll rewrite it. + */ +unsigned preloaded_size_ring0 = 1; +unsigned char preloaded_ring0[1] = {0}; + +Preload pre_loaded[] = { + {"ring0", 1, preloaded_ring0}, + {0, 0, 0} +}; + +int +main(int argc, char** argv) +{ + char sbuf[1024]; + struct { + void* p; + int sz; + } bins[2]; + int bin_num = 0; + FILE* fp; + char* progname = argv[0]; + char* eq; + + argv++, argc--; + + if (argc > 0 && argv[0][0] == '-') { + argv++, argc--; + } + if (argc < 1) { + abort(); + } + if ((fp = fopen(argv[0], "r")) == NULL) { + abort(); + } + + /* Needs to be called before any memory allocation */ + erts_short_init(); + + while (fgets(sbuf, sizeof sbuf, fp)) { + if (sbuf[0] == '#') { + continue; /* Comment */ + } else if (sbuf[0] == 'e' && strncmp("exec", sbuf, 4) == 0) { + continue; /* Comment ;-) */ + } else if ((eq = strchr(sbuf, '=')) != NULL) { + char* val; + char* p = strchr(sbuf, '\n'); + if (p) { + *p = '\0'; + } + *eq = '\0'; + val = erts_read_env(sbuf); + if (val == NULL) { + *eq = '='; + erts_sys_putenv(sbuf, eq - &sbuf[0]); + } + erts_free_read_env(val); + } else if (sbuf[0] == ':' && '0' <= sbuf[1] && sbuf[1] <= '9') { + int load_size = atoi(sbuf+1); + void* bin; + + bin = malloc(load_size); + if (fread(bin, 1, load_size, fp) != load_size) { + abort(); + } + bins[bin_num].p = bin; + bins[bin_num].sz = load_size; + bin_num++; + } else if (strcmp(sbuf, "--end--\n") == 0) { + int rval; + Eterm mod = NIL; + char *val; + + fclose(fp); + + if (bin_num != 2) { + abort(); + } + + val = erts_read_env("ERLBREAKHANDLER"); + if (val) { + init_break_handler(); + } + erts_free_read_env(val); + + if ((rval = erts_load_module(NULL, 0, NIL, &mod, bins[0].p, bins[0].sz)) < 0) { + fprintf(stderr, "%s: Load of initial module failed: %d\n", + progname, rval); + abort(); + } + erts_first_process(mod, bins[1].p, bins[1].sz, argc, argv); + free(bins[0].p); + free(bins[1].p); + process_main(); + abort(); + } else { + fprintf(stderr, "%s: bad line: %s\n", progname, sbuf); + abort(); + } + } + abort(); +} diff --git a/erts/emulator/sys/unix/erl_child_setup.c b/erts/emulator/sys/unix/erl_child_setup.c new file mode 100644 index 0000000000..7c6e4a2f37 --- /dev/null +++ b/erts/emulator/sys/unix/erl_child_setup.c @@ -0,0 +1,122 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2002-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +/* + * After a vfork() (or fork()) the child exec()s to this program which + * sets up the child and exec()s to the user program (see spawn_start() + * in sys.c and ticket OTP-4389). + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#define NEED_CHILD_SETUP_DEFINES +#include "sys.h" +#include "erl_misc_utils.h" + +#ifdef SIG_SIGSET /* Old SysV */ +void sys_sigrelease(int sig) +{ + sigrelse(sig); +} +#else /* !SIG_SIGSET */ +#ifdef SIG_SIGNAL /* Old BSD */ +sys_sigrelease(int sig) +{ + sigsetmask(sigblock(0) & ~sigmask(sig)); +} +#else /* !SIG_SIGNAL */ /* The True Way - POSIX!:-) */ +void sys_sigrelease(int sig) +{ + sigset_t mask; + + sigemptyset(&mask); + sigaddset(&mask, sig); + sigprocmask(SIG_UNBLOCK, &mask, (sigset_t *)NULL); +} +#endif /* !SIG_SIGNAL */ +#endif /* !SIG_SIGSET */ + +int +main(int argc, char *argv[]) +{ + int i, from, to; + int erts_spawn_executable = 0; + + /* OBSERVE! + * Keep child setup after fork() (implemented in sys.c) up to date + * if changes are made here. + */ + + if (argc != CS_ARGV_NO_OF_ARGS) { + if (argc < CS_ARGV_NO_OF_ARGS) { + return 1; + } else { + erts_spawn_executable = 1; + } + } + + if (strcmp("false", argv[CS_ARGV_UNBIND_IX]) != 0) + if (erts_unbind_from_cpu_str(argv[CS_ARGV_UNBIND_IX]) != 0) + return 1; + + for (i = 0; i < CS_ARGV_NO_OF_DUP2_OPS; i++) { + if (argv[CS_ARGV_DUP2_OP_IX(i)][0] == '-' + && argv[CS_ARGV_DUP2_OP_IX(i)][1] == '\0') + break; + if (sscanf(argv[CS_ARGV_DUP2_OP_IX(i)], "%d:%d", &from, &to) != 2) + return 1; + if (dup2(from, to) < 0) + return 1; + } + + if (sscanf(argv[CS_ARGV_FD_CR_IX], "%d:%d", &from, &to) != 2) + return 1; + for (i = from; i <= to; i++) + (void) close(i); + + if (!(argv[CS_ARGV_WD_IX][0] == '.' && argv[CS_ARGV_WD_IX][1] == '\0') + && chdir(argv[CS_ARGV_WD_IX]) < 0) + return 1; + +#if defined(USE_SETPGRP_NOARGS) /* SysV */ + (void) setpgrp(); +#elif defined(USE_SETPGRP) /* BSD */ + (void) setpgrp(0, getpid()); +#else /* POSIX */ + (void) setsid(); +#endif + + sys_sigrelease(SIGCHLD); + sys_sigrelease(SIGINT); + sys_sigrelease(SIGUSR1); + + if (erts_spawn_executable) { + if (argv[CS_ARGV_NO_OF_ARGS + 1] == NULL) { + execl(argv[CS_ARGV_NO_OF_ARGS],argv[CS_ARGV_NO_OF_ARGS], + (char *) NULL); + } else { + execv(argv[CS_ARGV_NO_OF_ARGS],&(argv[CS_ARGV_NO_OF_ARGS + 1])); + } + } else { + execl("/bin/sh", "sh", "-c", argv[CS_ARGV_CMD_IX], (char *) NULL); + } + return 1; +} diff --git a/erts/emulator/sys/unix/erl_main.c b/erts/emulator/sys/unix/erl_main.c new file mode 100644 index 0000000000..b26f93f77e --- /dev/null +++ b/erts/emulator/sys/unix/erl_main.c @@ -0,0 +1,31 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2000-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif +#include "sys.h" +#include "erl_vm.h" +#include "global.h" + +int +main(int argc, char **argv) +{ + erl_start(argc, argv); + return 0; +} diff --git a/erts/emulator/sys/unix/erl_unix_sys.h b/erts/emulator/sys/unix/erl_unix_sys.h new file mode 100644 index 0000000000..2d5ef882f6 --- /dev/null +++ b/erts/emulator/sys/unix/erl_unix_sys.h @@ -0,0 +1,339 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 1997-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + * + * This file handles differences between different Unix systems. + * This should be the only place with conditional compilation + * depending on the type of OS. + */ + +#ifndef _ERL_UNIX_SYS_H +#define _ERL_UNIX_SYS_H + +#include <stdio.h> +#include <math.h> +#include <limits.h> +#include <stdlib.h> +#include <string.h> +#ifndef QNX +#include <memory.h> +#endif + +#if defined(__sun__) && defined(__SVR4) && !defined(__EXTENSIONS__) +# define __EXTENSIONS__ +# include <sys/types.h> +# undef __EXTENSIONS__ +#else +# include <sys/types.h> +#endif +#include <sys/stat.h> +#include <sys/param.h> +#include <fcntl.h> +#include "erl_errno.h" +#include <signal.h> + + +#if HAVE_SYS_SOCKETIO_H +# include <sys/socketio.h> +#endif +#if HAVE_SYS_SOCKIO_H +# include <sys/sockio.h> +#endif + +#ifdef HAVE_NET_ERRNO_H +#include <net/errno.h> +#endif + +#ifdef HAVE_DIRENT_H +# include <dirent.h> +#endif + +#ifdef HAVE_UNISTD_H +# include <unistd.h> +#endif + +#ifndef HAVE_MMAP +# define HAVE_MMAP 0 +#endif + +#if HAVE_MMAP +# include <sys/mman.h> +#endif + +#if TIME_WITH_SYS_TIME +# include <sys/time.h> +# include <time.h> +#else +# if HAVE_SYS_TIME_H +# include <sys/time.h> +# else +# include <time.h> +# endif +#endif + +#include <sys/times.h> + +#ifdef HAVE_IEEEFP_H +#include <ieeefp.h> +#endif + +#ifdef QNX +#include <process.h> +#include <sys/qnx_glob.h> +#endif + +#include <pwd.h> + +#ifndef HZ +#define HZ 60 +#endif + +#ifdef NETDB_H_NEEDS_IN_H +#include <netinet/in.h> +#endif +#include <netdb.h> + +/* + * Make sure that MAXPATHLEN is defined. + */ +#ifdef GETHRTIME_WITH_CLOCK_GETTIME +#undef HAVE_GETHRTIME +#define HAVE_GETHRTIME 1 +#endif + +#ifndef MAXPATHLEN +# ifdef PATH_MAX +# define MAXPATHLEN PATH_MAX +# else +# define MAXPATHLEN 2048 +# endif +#endif + +/* File descriptors are numbers anc consecutively allocated on Unix */ +#define ERTS_SYS_CONTINOUS_FD_NUMBERS + +#define HAVE_ERTS_CHECK_IO_DEBUG +int erts_check_io_debug(void); + + +#ifndef ENABLE_CHILD_WAITER_THREAD +# undef ERTS_POLL_NEED_ASYNC_INTERRUPT_SUPPORT +# define ERTS_POLL_NEED_ASYNC_INTERRUPT_SUPPORT +# ifdef ERTS_SMP +# define ERTS_SMP_SCHEDULERS_NEED_TO_CHECK_CHILDREN +void erts_check_children(void); +# endif +#endif + +typedef void *GETENV_STATE; + +/* +** For the erl_timer_sup module. +*/ + +typedef struct timeval SysTimeval; + +#define sys_gettimeofday(Arg) ((void) gettimeofday((Arg), NULL)) + +typedef struct tms SysTimes; + +extern int erts_ticks_per_sec; + +#define SYS_CLK_TCK (erts_ticks_per_sec) + +#define sys_times(Arg) times(Arg) + +#define ERTS_WRAP_SYS_TIMES 1 +extern int erts_ticks_per_sec_wrap; +#define SYS_CLK_TCK_WRAP (erts_ticks_per_sec_wrap) +extern clock_t sys_times_wrap(void); + +#ifdef HAVE_GETHRTIME +#ifdef GETHRTIME_WITH_CLOCK_GETTIME +typedef long long SysHrTime; + +extern SysHrTime sys_gethrtime(void); +#define sys_init_hrtime() /* Nothing */ + +#else /* Real gethrtime (Solaris) */ + +typedef hrtime_t SysHrTime; + +#define sys_gethrtime() gethrtime() +#define sys_init_hrtime() /* Nothing */ + +#endif /* GETHRTIME_WITH_CLOCK_GETTIME */ +#endif /* HAVE_GETHRTIME */ + +#if (defined(HAVE_GETHRVTIME) || defined(HAVE_CLOCK_GETTIME)) +typedef long long SysCpuTime; +typedef struct timespec SysTimespec; + +#if defined(HAVE_GETHRVTIME) +#define sys_gethrvtime() gethrvtime() +#define sys_get_proc_cputime(t,tp) (t) = sys_gethrvtime(), \ + (tp).tv_sec = (time_t)((t)/1000000000LL), \ + (tp).tv_nsec = (long)((t)%1000000000LL) +int sys_start_hrvtime(void); +int sys_stop_hrvtime(void); + +#elif defined(HAVE_CLOCK_GETTIME) +#define sys_clock_gettime(cid,tp) clock_gettime((cid),&(tp)) +#define sys_get_proc_cputime(t,tp) sys_clock_gettime(CLOCK_PROCESS_CPUTIME_ID,(tp)) + +#endif +#endif + +/* No use in having other resolutions than 1 Ms. */ +#define SYS_CLOCK_RESOLUTION 1 + +/* These are defined in sys.c */ +#if defined(SIG_SIGSET) /* Old SysV */ +RETSIGTYPE (*sys_sigset())(); +#elif defined(SIG_SIGNAL) /* Old BSD */ +RETSIGTYPE (*sys_sigset())(); +#else +RETSIGTYPE (*sys_sigset(int, RETSIGTYPE (*func)(int)))(int); +#endif +extern void sys_sigrelease(int); +extern void sys_sigblock(int); +extern void sys_stop_cat(void); + +/* + * Handling of floating point exceptions. + */ + +#ifdef USE_ISINF_ISNAN /* simulate finite() */ +# define finite(f) (!isinf(f) && !isnan(f)) +# define HAVE_FINITE +#endif + +#ifdef NO_FPE_SIGNALS + +#define erts_get_current_fp_exception() NULL +#ifdef ERTS_SMP +#define erts_thread_init_fp_exception() do{}while(0) +#endif +# define __ERTS_FP_CHECK_INIT(fpexnp) do {} while (0) +# define __ERTS_FP_ERROR(fpexnp, f, Action) if (!finite(f)) { Action; } else {} +# define __ERTS_FP_ERROR_THOROUGH(fpexnp, f, Action) __ERTS_FP_ERROR(fpexnp, f, Action) +# define __ERTS_SAVE_FP_EXCEPTION(fpexnp) +# define __ERTS_RESTORE_FP_EXCEPTION(fpexnp) + +#define erts_sys_block_fpe() 0 +#define erts_sys_unblock_fpe(x) do{}while(0) + +#else /* !NO_FPE_SIGNALS */ + +extern volatile unsigned long *erts_get_current_fp_exception(void); +#ifdef ERTS_SMP +extern void erts_thread_init_fp_exception(void); +#endif +# if (defined(__i386__) || defined(__x86_64__)) && defined(__GNUC__) +# define erts_fwait(fpexnp,f) \ + __asm__ __volatile__("fwait" : "=m"(*(fpexnp)) : "m"(f)) +# elif (defined(__powerpc__) || defined(__ppc__)) && defined(__GNUC__) +# define erts_fwait(fpexnp,f) \ + __asm__ __volatile__("" : "=m"(*(fpexnp)) : "fm"(f)) +# elif defined(__sparc__) && defined(__linux__) && defined(__GNUC__) +# define erts_fwait(fpexnp,f) \ + __asm__ __volatile__("" : "=m"(*(fpexnp)) : "em"(f)) +# else +# define erts_fwait(fpexnp,f) \ + __asm__ __volatile__("" : "=m"(*(fpexnp)) : "g"(f)) +# endif +# if (defined(__i386__) || defined(__x86_64__)) && defined(__GNUC__) + extern void erts_restore_fpu(void); +# else +# define erts_restore_fpu() /*empty*/ +# endif +# if (!defined(__GNUC__) || \ + (__GNUC__ < 2) || \ + (__GNUC__ == 2 && __GNUC_MINOR < 96)) && \ + !defined(__builtin_expect) +# define __builtin_expect(x, expected_value) (x) +# endif +static __inline__ int erts_check_fpe(volatile unsigned long *fp_exception, double f) +{ + erts_fwait(fp_exception, f); + if (__builtin_expect(*fp_exception == 0, 1)) + return 0; + *fp_exception = 0; + erts_restore_fpu(); + return 1; +} +# undef erts_fwait +# undef erts_restore_fpu +extern void erts_fp_check_init_error(volatile unsigned long *fp_exception); +static __inline__ void __ERTS_FP_CHECK_INIT(volatile unsigned long *fp_exception) +{ + if (__builtin_expect(*fp_exception == 0, 1)) + return; + erts_fp_check_init_error(fp_exception); +} +# define __ERTS_FP_ERROR(fpexnp, f, Action) do { if (erts_check_fpe((fpexnp),(f))) { Action; } } while (0) +# define __ERTS_SAVE_FP_EXCEPTION(fpexnp) unsigned long old_erl_fp_exception = *(fpexnp) +# define __ERTS_RESTORE_FP_EXCEPTION(fpexnp) \ + do { *(fpexnp) = old_erl_fp_exception; } while (0) + /* This is for library calls where we don't trust the external + code to always throw floating-point exceptions on errors. */ +static __inline__ int erts_check_fpe_thorough(volatile unsigned long *fp_exception, double f) +{ + return erts_check_fpe(fp_exception, f) || !finite(f); +} +# define __ERTS_FP_ERROR_THOROUGH(fpexnp, f, Action) \ + do { if (erts_check_fpe_thorough((fpexnp),(f))) { Action; } } while (0) + +int erts_sys_block_fpe(void); +void erts_sys_unblock_fpe(int); + +#endif /* !NO_FPE_SIGNALS */ + +#define ERTS_FP_CHECK_INIT(p) __ERTS_FP_CHECK_INIT(&(p)->fp_exception) +#define ERTS_FP_ERROR(p, f, A) __ERTS_FP_ERROR(&(p)->fp_exception, f, A) +#define ERTS_FP_ERROR_THOROUGH(p, f, A) __ERTS_FP_ERROR_THOROUGH(&(p)->fp_exception, f, A) + + +#ifdef NEED_CHILD_SETUP_DEFINES +/* The child setup argv[] */ +#define CS_ARGV_PROGNAME_IX 0 /* Program name */ +#define CS_ARGV_UNBIND_IX 1 /* Unbind from cpu */ +#define CS_ARGV_WD_IX 2 /* Working directory */ +#define CS_ARGV_CMD_IX 3 /* Command */ +#define CS_ARGV_FD_CR_IX 4 /* Fd close range */ +#define CS_ARGV_DUP2_OP_IX(N) ((N) + 5) /* dup2 operations */ + +#define CS_ARGV_NO_OF_DUP2_OPS 3 /* Number of dup2 ops */ +#define CS_ARGV_NO_OF_ARGS 8 /* Number of arguments */ +#endif /* #ifdef NEED_CHILD_SETUP_DEFINES */ + +/* Threads */ +#ifdef USE_THREADS +extern int init_async(int); +extern int exit_async(void); +#endif + +#define ERTS_EXIT_AFTER_DUMP _exit + +#ifdef ERTS_TIMER_THREAD +struct erts_iwait; /* opaque for clients */ +extern struct erts_iwait *erts_iwait_init(void); +extern void erts_iwait_wait(struct erts_iwait *iwait, struct timeval *delay); +extern void erts_iwait_interrupt(struct erts_iwait *iwait); +#endif /* ERTS_TIMER_THREAD */ + +#endif /* #ifndef _ERL_UNIX_SYS_H */ diff --git a/erts/emulator/sys/unix/erl_unix_sys_ddll.c b/erts/emulator/sys/unix/erl_unix_sys_ddll.c new file mode 100644 index 0000000000..336d9586c4 --- /dev/null +++ b/erts/emulator/sys/unix/erl_unix_sys_ddll.c @@ -0,0 +1,280 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2006-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +/* + * Interface functions to the dynamic linker using dl* functions. + * (As far as I know it works on SunOS 4, 5, Linux and FreeBSD. /Seb) + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include "sys.h" +#include "erl_vm.h" +#include "global.h" +#ifdef HAVE_DLFCN_H +#include <dlfcn.h> +#endif + + +/* some systems do not have RTLD_NOW defined, and require the "mode" + * argument to dload() always be 1. + */ +#ifndef RTLD_NOW +# define RTLD_NOW 1 +#endif + +#define MAX_NAME_LEN 255 /* XXX should we get the system path size? */ +#define EXT_LEN 3 +#define FILE_EXT ".so" /* extension appended to the filename */ + +static char **errcodes = NULL; +static int num_errcodes = 0; +static int num_errcodes_allocated = 0; + +#define my_strdup(WHAT) my_strdup_in(ERTS_ALC_T_DDLL_ERRCODES, WHAT); + +static char *my_strdup_in(ErtsAlcType_t type, char *what) +{ + char *res = erts_alloc(type, strlen(what) + 1); + strcpy(res, what); + return res; +} + + +static int find_errcode(char *string, ErtsSysDdllError* err) +{ + int i; + + if (err != NULL) { + erts_sys_ddll_free_error(err); /* in case we ignored an earlier error */ + err->str = my_strdup_in(ERTS_ALC_T_DDLL_TMP_BUF, string); + return 0; + } + for(i=0;i<num_errcodes;++i) { + if (!strcmp(string, errcodes[i])) { + return i; + } + } + if (num_errcodes_allocated == num_errcodes) { + errcodes = (num_errcodes_allocated == 0) + ? erts_alloc(ERTS_ALC_T_DDLL_ERRCODES, + (num_errcodes_allocated = 10) * sizeof(char *)) + : erts_realloc(ERTS_ALC_T_DDLL_ERRCODES, errcodes, + (num_errcodes_allocated += 10) * sizeof(char *)); + } + errcodes[num_errcodes++] = my_strdup(string); + return (num_errcodes - 1); +} + +void erl_sys_ddll_init(void) { +#if defined(HAVE_DLOPEN) && defined(ERTS_NEED_DLOPEN_BEFORE_DLERROR) + /* + * dlopen() needs to be called before we make the first call to + * dlerror(); otherwise, dlerror() might dump core. At least + * some versions of linuxthread suffer from this bug. + */ + void *handle = dlopen("/nonexistinglib", RTLD_NOW); + if (handle) + dlclose(handle); +#endif + return; +} + +/* + * Open a shared object + */ +int erts_sys_ddll_open2(char *full_name, void **handle, ErtsSysDdllError* err) +{ +#if defined(HAVE_DLOPEN) + char* dlname; + int len = sys_strlen(full_name); + int ret; + + dlname = erts_alloc(ERTS_ALC_T_TMP, len + EXT_LEN + 1); + sys_strcpy(dlname, full_name); + sys_strcpy(dlname+len, FILE_EXT); + + ret = erts_sys_ddll_open_noext(dlname, handle, err); + + erts_free(ERTS_ALC_T_TMP, (void *) dlname); + return ret; +#else + return ERL_DE_ERROR_NO_DDLL_FUNCTIONALITY; +#endif +} + +int erts_sys_ddll_open_noext(char *dlname, void **handle, ErtsSysDdllError* err) +{ + int ret = ERL_DE_NO_ERROR; + char *str; + dlerror(); + if ((*handle = dlopen(dlname, RTLD_NOW)) == NULL) { + str = dlerror(); + + if (err == NULL) { + /* + * Remove prefix filename to avoid exploading number of + * error codes on extreme usage. + */ + if (strstr(str,dlname) == str) { + char *save_str = str; + str += strlen(dlname); + while (*str == ':' || *str == ' ') { + ++str; + } + if (*str == '\0') { /* Better with filename than nothing... */ + str = save_str; + } + } + } + ret = ERL_DE_DYNAMIC_ERROR_OFFSET - find_errcode(str, err); + } + return ret; +} + +/* + * Find a symbol in the shared object + */ +int erts_sys_ddll_sym2(void *handle, char *func_name, void **function, + ErtsSysDdllError* err) +{ +#if defined(HAVE_DLOPEN) + void *sym; + char *e; + int ret; + dlerror(); + sym = dlsym(handle, func_name); + if ((e = dlerror()) != NULL) { + ret = ERL_DE_DYNAMIC_ERROR_OFFSET - find_errcode(e, err); + } else { + *function = sym; + ret = ERL_DE_NO_ERROR; + } + return ret; +#else + return ERL_DE_ERROR_NO_DDLL_FUNCTIONALITY; +#endif +} + +/* XXX:PaN These two will be changed with new driver interface! */ + +/* + * Load the driver init function, might appear under different names depending on object arch... + */ + +int erts_sys_ddll_load_driver_init(void *handle, void **function) +{ + void *fn; + int res; + if ((res = erts_sys_ddll_sym2(handle, "driver_init", &fn, NULL)) != ERL_DE_NO_ERROR) { + res = erts_sys_ddll_sym2(handle, "_driver_init", &fn, NULL); + } + if (res == ERL_DE_NO_ERROR) { + *function = fn; + } + return res; +} + +int erts_sys_ddll_load_nif_init(void *handle, void **function, ErtsSysDdllError* err) +{ + void *fn; + int res; + if ((res = erts_sys_ddll_sym2(handle, "nif_init", &fn, err)) != ERL_DE_NO_ERROR) { + res = erts_sys_ddll_sym2(handle, "_nif_init", &fn, err); + } + if (res == ERL_DE_NO_ERROR) { + *function = fn; + } + return res; +} + +/* + * Call the driver_init function, whatever it's really called, simple on unix... +*/ +void *erts_sys_ddll_call_init(void *function) { + void *(*initfn)(void) = function; + return (*initfn)(); +} +void *erts_sys_ddll_call_nif_init(void *function) { + return erts_sys_ddll_call_init(function); +} + + + +/* + * Close a chared object + */ +int erts_sys_ddll_close2(void *handle, ErtsSysDdllError* err) +{ +#if defined(HAVE_DLOPEN) + int ret; + char *s; + dlerror(); + if (dlclose(handle) == 0) { + ret = ERL_DE_NO_ERROR; + } else { + if ((s = dlerror()) == NULL) { + find_errcode("unspecified error", err); + ret = ERL_DE_ERROR_UNSPECIFIED; + } else { + ret = ERL_DE_DYNAMIC_ERROR_OFFSET - find_errcode(s, err); + } + } + return ret; +#else + return ERL_DE_ERROR_NO_DDLL_FUNCTIONALITY; +#endif +} + + +/* + * Return string that describes the (current) error + */ +char *erts_sys_ddll_error(int code) +{ + int actual_code; + + if (code > ERL_DE_DYNAMIC_ERROR_OFFSET) { + return "Unspecified error"; + } + actual_code = -1*(code - ERL_DE_DYNAMIC_ERROR_OFFSET); +#if defined(HAVE_DLOPEN) + { + char *msg; + + if (actual_code >= num_errcodes) { + msg = "Unknown dlload error"; + } else { + msg = errcodes[actual_code]; + } + return msg; + } +#endif + return "no error"; +} + +void erts_sys_ddll_free_error(ErtsSysDdllError* err) +{ + if (err->str != NULL) { + erts_free(ERTS_ALC_T_DDLL_TMP_BUF, err->str); + } +} + diff --git a/erts/emulator/sys/unix/sys.c b/erts/emulator/sys/unix/sys.c new file mode 100644 index 0000000000..183525b222 --- /dev/null +++ b/erts/emulator/sys/unix/sys.c @@ -0,0 +1,3346 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 1996-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#ifdef ISC32 +#define _POSIX_SOURCE +#define _XOPEN_SOURCE +#endif + +#include <sys/times.h> /* ! */ +#include <time.h> +#include <signal.h> +#include <sys/wait.h> +#include <sys/uio.h> +#include <termios.h> +#include <ctype.h> +#include <sys/utsname.h> + +#ifdef ISC32 +#include <sys/bsdtypes.h> +#endif + +#include <termios.h> +#ifdef HAVE_FCNTL_H +#include <fcntl.h> +#endif +#ifdef HAVE_SYS_IOCTL_H +#include <sys/ioctl.h> +#endif + +#define NEED_CHILD_SETUP_DEFINES +#define ERTS_WANT_BREAK_HANDLING +#define ERTS_WANT_GOT_SIGUSR1 +#define WANT_NONBLOCKING /* must define this to pull in defs from sys.h */ +#include "sys.h" + +#ifdef USE_THREADS +#include "erl_threads.h" +#endif + +#include "erl_mseg.h" + +extern char **environ; +static erts_smp_rwmtx_t environ_rwmtx; + +#define MAX_VSIZE 16 /* Max number of entries allowed in an I/O + * vector sock_sendv(). + */ + +/* + * Don't need global.h, but bif_table.h (included by bif.h), + * won't compile otherwise + */ +#include "global.h" +#include "bif.h" + +#include "erl_sys_driver.h" +#include "erl_check_io.h" + +#ifndef DISABLE_VFORK +#define DISABLE_VFORK 0 +#endif + +#ifdef USE_THREADS +# ifdef ENABLE_CHILD_WAITER_THREAD +# define CHLDWTHR ENABLE_CHILD_WAITER_THREAD +# else +# define CHLDWTHR 0 +# endif +#else +# define CHLDWTHR 0 +#endif +/* + * [OTP-3906] + * Solaris signal management gets confused when threads are used and a + * lot of child processes dies. The confusion results in that SIGCHLD + * signals aren't delivered to the emulator which in turn results in + * a lot of defunct processes in the system. + * + * The problem seems to appear when a signal is frequently + * blocked/unblocked at the same time as the signal is frequently + * propagated. The child waiter thread is a workaround for this problem. + * The SIGCHLD signal is always blocked (in all threads), and the child + * waiter thread fetches the signal by a call to sigwait(). See + * child_waiter(). + */ + +typedef struct ErtsSysReportExit_ ErtsSysReportExit; +struct ErtsSysReportExit_ { + ErtsSysReportExit *next; + Eterm port; + int pid; + int ifd; + int ofd; +#if CHLDWTHR && !defined(ERTS_SMP) + int status; +#endif +}; + +static ErtsSysReportExit *report_exit_list; +#if CHLDWTHR && !defined(ERTS_SMP) +static ErtsSysReportExit *report_exit_transit_list; +#endif + +extern int check_async_ready(void); +extern int driver_interrupt(int, int); +/*EXTERN_FUNCTION(void, increment_time, (int));*/ +/*EXTERN_FUNCTION(int, next_time, (_VOID_));*/ +extern void do_break(void); + +extern void erl_sys_args(int*, char**); + +/* The following two defs should probably be moved somewhere else */ + +extern void erts_sys_init_float(void); + +extern void erl_crash_dump(char* file, int line, char* fmt, ...); + +#define DIR_SEPARATOR_CHAR '/' + +#if defined(DEBUG) +#define ERL_BUILD_TYPE_MARKER ".debug" +#elif defined(PURIFY) +#define ERL_BUILD_TYPE_MARKER ".purify" +#elif defined(QUANTIFY) +#define ERL_BUILD_TYPE_MARKER ".quantify" +#elif defined(PURECOV) +#define ERL_BUILD_TYPE_MARKER ".purecov" +#elif defined(VALGRIND) +#define ERL_BUILD_TYPE_MARKER ".valgrind" +#else /* opt */ +#define ERL_BUILD_TYPE_MARKER +#endif + +#define CHILD_SETUP_PROG_NAME "child_setup" ERL_BUILD_TYPE_MARKER +#if !DISABLE_VFORK +static char *child_setup_prog; +#endif + +#ifdef DEBUG +static int debug_log = 0; +#endif + +#ifdef ERTS_SMP +erts_smp_atomic_t erts_got_sigusr1; +#define ERTS_SET_GOT_SIGUSR1 \ + erts_smp_atomic_set(&erts_got_sigusr1, 1) +#define ERTS_UNSET_GOT_SIGUSR1 \ + erts_smp_atomic_set(&erts_got_sigusr1, 0) +static erts_smp_atomic_t have_prepared_crash_dump; +#define ERTS_PREPARED_CRASH_DUMP \ + ((int) erts_smp_atomic_xchg(&have_prepared_crash_dump, 1)) +#else +volatile int erts_got_sigusr1; +#define ERTS_SET_GOT_SIGUSR1 (erts_got_sigusr1 = 1) +#define ERTS_UNSET_GOT_SIGUSR1 (erts_got_sigusr1 = 0) +static volatile int have_prepared_crash_dump; +#define ERTS_PREPARED_CRASH_DUMP \ + (have_prepared_crash_dump++) +#endif + +static erts_smp_atomic_t sys_misc_mem_sz; + +#if defined(ERTS_SMP) +static void smp_sig_notify(char c); +static int sig_notify_fds[2] = {-1, -1}; +#elif defined(USE_THREADS) +static int async_fd[2]; +#endif + +#if CHLDWTHR || defined(ERTS_SMP) +erts_mtx_t chld_stat_mtx; +#endif +#if CHLDWTHR +static erts_tid_t child_waiter_tid; +/* chld_stat_mtx is used to protect against concurrent accesses + of the driver_data fields pid, alive, and status. */ +erts_cnd_t chld_stat_cnd; +static long children_alive; +#define CHLD_STAT_LOCK erts_mtx_lock(&chld_stat_mtx) +#define CHLD_STAT_UNLOCK erts_mtx_unlock(&chld_stat_mtx) +#define CHLD_STAT_WAIT erts_cnd_wait(&chld_stat_cnd, &chld_stat_mtx) +#define CHLD_STAT_SIGNAL erts_cnd_signal(&chld_stat_cnd) +#elif defined(ERTS_SMP) /* ------------------------------------------------- */ +#define CHLD_STAT_LOCK erts_mtx_lock(&chld_stat_mtx) +#define CHLD_STAT_UNLOCK erts_mtx_unlock(&chld_stat_mtx) + +#else /* ------------------------------------------------------------------- */ +#define CHLD_STAT_LOCK +#define CHLD_STAT_UNLOCK +static volatile int children_died; +#endif + + +static struct fd_data { + char pbuf[4]; /* hold partial packet bytes */ + int psz; /* size of pbuf */ + char *buf; + char *cpos; + int sz; + int remain; /* for input on fd */ +} *fd_data; /* indexed by fd */ + +/* static FUNCTION(int, write_fill, (int, char*, int)); unused? */ +static FUNCTION(void, note_child_death, (int, int)); + +#if CHLDWTHR +static FUNCTION(void *, child_waiter, (void *)); +#endif + +/********************* General functions ****************************/ + +/* This is used by both the drivers and general I/O, must be set early */ +static int max_files = -1; + +/* + * a few variables used by the break handler + */ +#ifdef ERTS_SMP +erts_smp_atomic_t erts_break_requested; +#define ERTS_SET_BREAK_REQUESTED \ + erts_smp_atomic_set(&erts_break_requested, (long) 1) +#define ERTS_UNSET_BREAK_REQUESTED \ + erts_smp_atomic_set(&erts_break_requested, (long) 0) +#else +volatile int erts_break_requested = 0; +#define ERTS_SET_BREAK_REQUESTED (erts_break_requested = 1) +#define ERTS_UNSET_BREAK_REQUESTED (erts_break_requested = 0) +#endif +/* set early so the break handler has access to initial mode */ +static struct termios initial_tty_mode; +static int replace_intr = 0; +/* assume yes initially, ttsl_init will clear it */ +int using_oldshell = 1; + +#ifdef ERTS_ENABLE_KERNEL_POLL + +int erts_use_kernel_poll = 0; + +struct { + int (*select)(ErlDrvPort, ErlDrvEvent, int, int); + int (*event)(ErlDrvPort, ErlDrvEvent, ErlDrvEventData); + void (*check_io_interrupt)(int); + void (*check_io_interrupt_tmd)(int, long); + void (*check_io)(int); + Uint (*size)(void); + Eterm (*info)(void *); + int (*check_io_debug)(void); +} io_func = {0}; + + +int +driver_select(ErlDrvPort port, ErlDrvEvent event, int mode, int on) +{ + return (*io_func.select)(port, event, mode, on); +} + +int +driver_event(ErlDrvPort port, ErlDrvEvent event, ErlDrvEventData event_data) +{ + return (*io_func.event)(port, event, event_data); +} + +Eterm erts_check_io_info(void *p) +{ + return (*io_func.info)(p); +} + +int +erts_check_io_debug(void) +{ + return (*io_func.check_io_debug)(); +} + + +static void +init_check_io(void) +{ + if (erts_use_kernel_poll) { + io_func.select = driver_select_kp; + io_func.event = driver_event_kp; + io_func.check_io_interrupt = erts_check_io_interrupt_kp; + io_func.check_io_interrupt_tmd = erts_check_io_interrupt_timed_kp; + io_func.check_io = erts_check_io_kp; + io_func.size = erts_check_io_size_kp; + io_func.info = erts_check_io_info_kp; + io_func.check_io_debug = erts_check_io_debug_kp; + erts_init_check_io_kp(); + max_files = erts_check_io_max_files_kp(); + } + else { + io_func.select = driver_select_nkp; + io_func.event = driver_event_nkp; + io_func.check_io_interrupt = erts_check_io_interrupt_nkp; + io_func.check_io_interrupt_tmd = erts_check_io_interrupt_timed_nkp; + io_func.check_io = erts_check_io_nkp; + io_func.size = erts_check_io_size_nkp; + io_func.info = erts_check_io_info_nkp; + io_func.check_io_debug = erts_check_io_debug_nkp; + erts_init_check_io_nkp(); + max_files = erts_check_io_max_files_nkp(); + } +} + +#define ERTS_CHK_IO_INTR (*io_func.check_io_interrupt) +#define ERTS_CHK_IO_INTR_TMD (*io_func.check_io_interrupt_tmd) +#define ERTS_CHK_IO (*io_func.check_io) +#define ERTS_CHK_IO_SZ (*io_func.size) + +#else /* !ERTS_ENABLE_KERNEL_POLL */ + +static void +init_check_io(void) +{ + erts_init_check_io(); + max_files = erts_check_io_max_files(); +} + +#define ERTS_CHK_IO_INTR erts_check_io_interrupt +#define ERTS_CHK_IO_INTR_TMD erts_check_io_interrupt_timed +#define ERTS_CHK_IO erts_check_io +#define ERTS_CHK_IO_SZ erts_check_io_size + +#endif + +#ifdef ERTS_SMP +void +erts_sys_schedule_interrupt(int set) +{ + ERTS_CHK_IO_INTR(set); +} + +void +erts_sys_schedule_interrupt_timed(int set, long msec) +{ + ERTS_CHK_IO_INTR_TMD(set, msec); +} +#endif + +Uint +erts_sys_misc_mem_sz(void) +{ + Uint res = ERTS_CHK_IO_SZ(); + res += erts_smp_atomic_read(&sys_misc_mem_sz); + return res; +} + +/* + * reset the terminal to the original settings on exit + */ +void sys_tty_reset(void) +{ + if (using_oldshell && !replace_intr) { + SET_BLOCKING(0); + } + else if (isatty(0)) { + tcsetattr(0,TCSANOW,&initial_tty_mode); + } +} + +#ifdef __tile__ +/* Direct malloc to spread memory around the caches of multiple tiles. */ +#include <malloc.h> +MALLOC_USE_HASH(1); +#endif + +#ifdef USE_THREADS +static void *ethr_internal_alloc(size_t size) +{ + return erts_alloc_fnf(ERTS_ALC_T_ETHR_INTERNAL, (Uint) size); +} +static void *ethr_internal_realloc(void *ptr, size_t size) +{ + return erts_realloc_fnf(ERTS_ALC_T_ETHR_INTERNAL, ptr, (Uint) size); +} +static void ethr_internal_free(void *ptr) +{ + erts_free(ERTS_ALC_T_ETHR_INTERNAL, ptr); +} + +#ifdef ERTS_THR_HAVE_SIG_FUNCS +/* + * Child thread inherits parents signal mask at creation. In order to + * guarantee that the main thread will receive all SIGINT, SIGCHLD, and + * SIGUSR1 signals sent to the process, we block these signals in the + * parent thread when creating a new thread. + */ + +static sigset_t thr_create_sigmask; + +#endif /* #ifdef ERTS_THR_HAVE_SIG_FUNCS */ + +typedef struct { +#ifdef ERTS_THR_HAVE_SIG_FUNCS + sigset_t saved_sigmask; +#endif + int unbind_child; +} erts_thr_create_data_t; + +/* + * thr_create_prepare() is called in parent thread before thread creation. + * Returned value is passed as argument to thr_create_cleanup(). + */ +static void * +thr_create_prepare(void) +{ + erts_thr_create_data_t *tcdp; + ErtsSchedulerData *esdp; + + tcdp = erts_alloc(ERTS_ALC_T_TMP, sizeof(erts_thr_create_data_t)); + +#ifdef ERTS_THR_HAVE_SIG_FUNCS + erts_thr_sigmask(SIG_BLOCK, &thr_create_sigmask, &tcdp->saved_sigmask); +#endif + esdp = erts_get_scheduler_data(); + tcdp->unbind_child = esdp && erts_is_scheduler_bound(esdp); + + return (void *) tcdp; +} + + +/* thr_create_cleanup() is called in parent thread after thread creation. */ +static void +thr_create_cleanup(void *vtcdp) +{ + erts_thr_create_data_t *tcdp = (erts_thr_create_data_t *) vtcdp; + +#ifdef ERTS_THR_HAVE_SIG_FUNCS + /* Restore signalmask... */ + erts_thr_sigmask(SIG_SETMASK, &tcdp->saved_sigmask, NULL); +#endif + + erts_free(ERTS_ALC_T_TMP, tcdp); +} + +static void +thr_create_prepare_child(void *vtcdp) +{ + erts_thr_create_data_t *tcdp = (erts_thr_create_data_t *) vtcdp; + +#ifndef NO_FPE_SIGNALS + /* + * We do not want fp exeptions in other threads than the + * scheduler threads. We enable fpe explicitly in the scheduler + * threads after this. + */ + erts_thread_disable_fpe(); +#endif + + if (tcdp->unbind_child) { + erts_smp_rwmtx_rlock(&erts_cpu_bind_rwmtx); + erts_unbind_from_cpu(erts_cpuinfo); + erts_smp_rwmtx_runlock(&erts_cpu_bind_rwmtx); + } + +} + +#endif /* #ifdef USE_THREADS */ + +void +erts_sys_pre_init(void) +{ + erts_printf_add_cr_to_stdout = 1; + erts_printf_add_cr_to_stderr = 1; +#ifdef USE_THREADS + { + erts_thr_init_data_t eid = ERTS_THR_INIT_DATA_DEF_INITER; + eid.alloc = ethr_internal_alloc; + eid.realloc = ethr_internal_realloc; + eid.free = ethr_internal_free; + + eid.thread_create_child_func = thr_create_prepare_child; + /* Before creation in parent */ + eid.thread_create_prepare_func = thr_create_prepare; + /* After creation in parent */ + eid.thread_create_parent_func = thr_create_cleanup, + +#ifdef ERTS_THR_HAVE_SIG_FUNCS + sigemptyset(&thr_create_sigmask); + sigaddset(&thr_create_sigmask, SIGINT); /* block interrupt */ + sigaddset(&thr_create_sigmask, SIGCHLD); /* block child signals */ + sigaddset(&thr_create_sigmask, SIGUSR1); /* block user defined signal */ +#endif + + erts_thr_init(&eid); + + report_exit_list = NULL; + +#ifdef ERTS_ENABLE_LOCK_COUNT + erts_lcnt_init(); +#endif + +#if CHLDWTHR || defined(ERTS_SMP) + erts_mtx_init(&chld_stat_mtx, "child_status"); +#endif +#if CHLDWTHR +#ifndef ERTS_SMP + report_exit_transit_list = NULL; +#endif + erts_cnd_init(&chld_stat_cnd); + children_alive = 0; +#endif + } +#ifdef ERTS_SMP + erts_smp_atomic_init(&erts_break_requested, 0); + erts_smp_atomic_init(&erts_got_sigusr1, 0); + erts_smp_atomic_init(&have_prepared_crash_dump, 0); +#else + erts_break_requested = 0; + erts_got_sigusr1 = 0; + have_prepared_crash_dump = 0; +#endif +#if !CHLDWTHR && !defined(ERTS_SMP) + children_died = 0; +#endif +#endif /* USE_THREADS */ + erts_smp_atomic_init(&sys_misc_mem_sz, 0); + erts_smp_rwmtx_init(&environ_rwmtx, "environ"); +} + +void +erl_sys_init(void) +{ +#if !DISABLE_VFORK + int res; + char bindir[MAXPATHLEN]; + size_t bindirsz = sizeof(bindir); + Uint csp_path_sz; + + res = erts_sys_getenv("BINDIR", bindir, &bindirsz); + if (res != 0) { + if (res < 0) + erl_exit(-1, + "Environment variable BINDIR is not set\n"); + if (res > 0) + erl_exit(-1, + "Value of environment variable BINDIR is too large\n"); + } + if (bindir[0] != DIR_SEPARATOR_CHAR) + erl_exit(-1, + "Environment variable BINDIR does not contain an" + " absolute path\n"); + csp_path_sz = (strlen(bindir) + + 1 /* DIR_SEPARATOR_CHAR */ + + sizeof(CHILD_SETUP_PROG_NAME) + + 1); + child_setup_prog = erts_alloc(ERTS_ALC_T_CS_PROG_PATH, csp_path_sz); + erts_smp_atomic_add(&sys_misc_mem_sz, csp_path_sz); + sprintf(child_setup_prog, + "%s%c%s", + bindir, + DIR_SEPARATOR_CHAR, + CHILD_SETUP_PROG_NAME); +#endif + +#ifdef USE_SETLINEBUF + setlinebuf(stdout); +#else + setvbuf(stdout, (char *)NULL, _IOLBF, BUFSIZ); +#endif + + erts_sys_init_float(); + + /* we save this so the break handler can set and reset it properly */ + /* also so that we can reset on exit (break handler or not) */ + if (isatty(0)) { + tcgetattr(0,&initial_tty_mode); + } + tzset(); /* Required at least for NetBSD with localtime_r() */ +} + +/* signal handling */ + +#ifdef SIG_SIGSET /* Old SysV */ +RETSIGTYPE (*sys_sigset(sig, func))() +int sig; +RETSIGTYPE (*func)(); +{ + return(sigset(sig, func)); +} +void sys_sigblock(int sig) +{ + sighold(sig); +} +void sys_sigrelease(int sig) +{ + sigrelse(sig); +} +#else /* !SIG_SIGSET */ +#ifdef SIG_SIGNAL /* Old BSD */ +RETSIGTYPE (*sys_sigset(sig, func))(int, int) +int sig; +RETSIGTYPE (*func)(); +{ + return(signal(sig, func)); +} +sys_sigblock(int sig) +{ + sigblock(sig); +} +sys_sigrelease(int sig) +{ + sigsetmask(sigblock(0) & ~sigmask(sig)); +} +#else /* !SIG_SIGNAL */ /* The True Way - POSIX!:-) */ +RETSIGTYPE (*sys_sigset(int sig, RETSIGTYPE (*func)(int)))(int) +{ + struct sigaction act, oact; + + sigemptyset(&act.sa_mask); + act.sa_flags = 0; + act.sa_handler = func; + sigaction(sig, &act, &oact); + return(oact.sa_handler); +} + +#ifdef USE_THREADS +#undef sigprocmask +#define sigprocmask erts_thr_sigmask +#endif + +void sys_sigblock(int sig) +{ + sigset_t mask; + + sigemptyset(&mask); + sigaddset(&mask, sig); + sigprocmask(SIG_BLOCK, &mask, (sigset_t *)NULL); +} + +void sys_sigrelease(int sig) +{ + sigset_t mask; + + sigemptyset(&mask); + sigaddset(&mask, sig); + sigprocmask(SIG_UNBLOCK, &mask, (sigset_t *)NULL); +} +#endif /* !SIG_SIGNAL */ +#endif /* !SIG_SIGSET */ + +#if (0) /* not used? -- gordon */ +static void (*break_func)(); +static RETSIGTYPE break_handler(int sig) +{ +#ifdef QNX + /* Turn off SIGCHLD during break processing */ + sys_sigblock(SIGCHLD); +#endif + (*break_func)(); +#ifdef QNX + sys_sigrelease(SIGCHLD); +#endif +} +#endif /* 0 */ + +static ERTS_INLINE void +prepare_crash_dump(void) +{ + int i, max; + char env[21]; /* enough to hold any 64-bit integer */ + size_t envsz; + + if (ERTS_PREPARED_CRASH_DUMP) + return; /* We have already been called */ + + /* Make sure we unregister at epmd (unknown fd) and get at least + one free filedescriptor (for erl_crash.dump) */ + max = max_files; + if (max < 1024) + max = 1024; + for (i = 3; i < max; i++) { +#if defined(ERTS_SMP) + /* We don't want to close the signal notification pipe... */ + if (i == sig_notify_fds[0] || i == sig_notify_fds[1]) + continue; +#elif defined(USE_THREADS) + /* We don't want to close the async notification pipe... */ + if (i == async_fd[0] || i == async_fd[1]) + continue; +#endif + close(i); + } + + envsz = sizeof(env); + i = erts_sys_getenv("ERL_CRASH_DUMP_NICE", env, &envsz); + if (i >= 0) { + int nice_val; + nice_val = i != 0 ? 0 : atoi(env); + if (nice_val > 39) { + nice_val = 39; + } + nice(nice_val); + } + + envsz = sizeof(env); + i = erts_sys_getenv("ERL_CRASH_DUMP_SECONDS", env, &envsz); + if (i >= 0) { + unsigned sec; + sec = (unsigned) i != 0 ? 0 : atoi(env); + alarm(sec); + } + +} + +void +erts_sys_prepare_crash_dump(void) +{ + prepare_crash_dump(); +} + +static ERTS_INLINE void +break_requested(void) +{ + /* + * just set a flag - checked for and handled by + * scheduler threads erts_check_io() (not signal handler). + */ +#ifdef DEBUG + fprintf(stderr,"break!\n"); +#endif + if (ERTS_BREAK_REQUESTED) + erl_exit(ERTS_INTR_EXIT, ""); + + ERTS_SET_BREAK_REQUESTED; + ERTS_CHK_IO_INTR(1); /* Make sure we don't sleep in poll */ +} + +/* set up signal handlers for break and quit */ +#if (defined(SIG_SIGSET) || defined(SIG_SIGNAL)) +static RETSIGTYPE request_break(void) +#else +static RETSIGTYPE request_break(int signum) +#endif +{ +#ifdef ERTS_SMP + smp_sig_notify('I'); +#else + break_requested(); +#endif +} + +static ERTS_INLINE void +sigusr1_exit(void) +{ + /* We do this at interrupt level, since the main reason for + wanting to generate a crash dump in this way is that the emulator + is hung somewhere, so it won't be able to poll any flag we set here. + */ + ERTS_SET_GOT_SIGUSR1; + prepare_crash_dump(); + erl_exit(1, "Received SIGUSR1\n"); +} + +#ifdef ETHR_UNUSABLE_SIGUSRX +#warning "Unusable SIGUSR1 & SIGUSR2. Disabling use of these signals" +#endif + +#ifndef ETHR_UNUSABLE_SIGUSRX + +#if (defined(SIG_SIGSET) || defined(SIG_SIGNAL)) +static RETSIGTYPE user_signal1(void) +#else +static RETSIGTYPE user_signal1(int signum) +#endif +{ +#ifdef ERTS_SMP + smp_sig_notify('1'); +#else + sigusr1_exit(); +#endif +} + +#ifdef QUANTIFY +#if (defined(SIG_SIGSET) || defined(SIG_SIGNAL)) +static RETSIGTYPE user_signal2(void) +#else +static RETSIGTYPE user_signal2(int signum) +#endif +{ +#ifdef ERTS_SMP + smp_sig_notify('2'); +#else + quantify_save_data(); +#endif +} +#endif + +#endif /* #ifndef ETHR_UNUSABLE_SIGUSRX */ + +static void +quit_requested(void) +{ + erl_exit(ERTS_INTR_EXIT, ""); +} + +#if (defined(SIG_SIGSET) || defined(SIG_SIGNAL)) +static RETSIGTYPE do_quit(void) +#else +static RETSIGTYPE do_quit(int signum) +#endif +{ +#ifdef ERTS_SMP + smp_sig_notify('Q'); +#else + quit_requested(); +#endif +} + +/* Disable break */ +void erts_set_ignore_break(void) { + sys_sigset(SIGINT, SIG_IGN); + sys_sigset(SIGQUIT, SIG_IGN); + sys_sigset(SIGTSTP, SIG_IGN); +} + +/* Don't use ctrl-c for break handler but let it be + used by the shell instead (see user_drv.erl) */ +void erts_replace_intr(void) { + struct termios mode; + + if (isatty(0)) { + tcgetattr(0, &mode); + + /* here's an example of how to replace ctrl-c with ctrl-u */ + /* mode.c_cc[VKILL] = 0; + mode.c_cc[VINTR] = CKILL; */ + + mode.c_cc[VINTR] = 0; /* disable ctrl-c */ + tcsetattr(0, TCSANOW, &mode); + replace_intr = 1; + } +} + +void init_break_handler(void) +{ + sys_sigset(SIGINT, request_break); +#ifndef ETHR_UNUSABLE_SIGUSRX + sys_sigset(SIGUSR1, user_signal1); +#ifdef QUANTIFY + sys_sigset(SIGUSR2, user_signal2); +#endif +#endif /* #ifndef ETHR_UNUSABLE_SIGUSRX */ + sys_sigset(SIGQUIT, do_quit); +} + +int sys_max_files(void) +{ + return(max_files); +} + +static void block_signals(void) +{ +#if !CHLDWTHR + sys_sigblock(SIGCHLD); +#endif +#ifndef ERTS_SMP + sys_sigblock(SIGINT); +#ifndef ETHR_UNUSABLE_SIGUSRX + sys_sigblock(SIGUSR1); +#endif +#endif +} + +static void unblock_signals(void) +{ + /* Update erl_child_setup.c if changed */ +#if !CHLDWTHR + sys_sigrelease(SIGCHLD); +#endif +#ifndef ERTS_SMP + sys_sigrelease(SIGINT); +#ifndef ETHR_UNUSABLE_SIGUSRX + sys_sigrelease(SIGUSR1); +#endif /* #ifndef ETHR_UNUSABLE_SIGUSRX */ +#endif +} +/************************** Time stuff **************************/ +#ifdef HAVE_GETHRTIME +#ifdef GETHRTIME_WITH_CLOCK_GETTIME + +SysHrTime sys_gethrtime(void) +{ + struct timespec ts; + long long result; + if (clock_gettime(CLOCK_MONOTONIC,&ts) != 0) { + erl_exit(1,"Fatal, could not get clock_monotonic value!, " + "errno = %d\n", errno); + } + result = ((long long) ts.tv_sec) * 1000000000LL + + ((long long) ts.tv_nsec); + return (SysHrTime) result; +} +#endif +#endif + +/************************** OS info *******************************/ + +/* Used by erlang:info/1. */ +/* (This code was formerly in drv.XXX/XXX_os_drv.c) */ + +char os_type[] = "unix"; + +static int +get_number(char **str_ptr) +{ + char* s = *str_ptr; /* Pointer to beginning of string. */ + char* dot; /* Pointer to dot in string or NULL. */ + + if (!isdigit((int) *s)) + return 0; + if ((dot = strchr(s, '.')) == NULL) { + *str_ptr = s+strlen(s); + return atoi(s); + } else { + *dot = '\0'; + *str_ptr = dot+1; + return atoi(s); + } +} + +void +os_flavor(char* namebuf, /* Where to return the name. */ + unsigned size) /* Size of name buffer. */ +{ + static int called = 0; + static struct utsname uts; /* Information about the system. */ + + if (!called) { + char* s; + + (void) uname(&uts); + called = 1; + for (s = uts.sysname; *s; s++) { + if (isupper((int) *s)) { + *s = tolower((int) *s); + } + } + } + strcpy(namebuf, uts.sysname); +} + +void +os_version(pMajor, pMinor, pBuild) +int* pMajor; /* Pointer to major version. */ +int* pMinor; /* Pointer to minor version. */ +int* pBuild; /* Pointer to build number. */ +{ + struct utsname uts; /* Information about the system. */ + char* release; /* Pointer to the release string: + * X.Y or X.Y.Z. + */ + + (void) uname(&uts); + release = uts.release; + *pMajor = get_number(&release); + *pMinor = get_number(&release); + *pBuild = get_number(&release); +} + +void init_getenv_state(GETENV_STATE *state) +{ + erts_smp_rwmtx_rlock(&environ_rwmtx); + *state = NULL; +} + +char *getenv_string(GETENV_STATE *state0) +{ + char **state = (char **) *state0; + char *cp; + + ERTS_SMP_LC_ASSERT(erts_smp_lc_rwmtx_is_rlocked(&environ_rwmtx)); + + if (state == NULL) + state = environ; + + cp = *state++; + *state0 = (GETENV_STATE) state; + + return cp; +} + +void fini_getenv_state(GETENV_STATE *state) +{ + *state = NULL; + erts_smp_rwmtx_runlock(&environ_rwmtx); +} + + +/************************** Port I/O *******************************/ + + + +/* I. Common stuff */ + +/* + * Decreasing the size of it below 16384 is not allowed. + */ + +/* II. The spawn/fd/vanilla drivers */ + +#define ERTS_SYS_READ_BUF_SZ (64*1024) + +/* This data is shared by these drivers - initialized by spawn_init() */ +static struct driver_data { + int port_num, ofd, packet_bytes; + ErtsSysReportExit *report_exit; + int pid; + int alive; + int status; +} *driver_data; /* indexed by fd */ + +/* Driver interfaces */ +static ErlDrvData spawn_start(ErlDrvPort, char*, SysDriverOpts*); +static ErlDrvData fd_start(ErlDrvPort, char*, SysDriverOpts*); +static int fd_control(ErlDrvData, unsigned int, char *, int, char **, int); +static ErlDrvData vanilla_start(ErlDrvPort, char*, SysDriverOpts*); +static int spawn_init(void); +static void fd_stop(ErlDrvData); +static void stop(ErlDrvData); +static void ready_input(ErlDrvData, ErlDrvEvent); +static void ready_output(ErlDrvData, ErlDrvEvent); +static void output(ErlDrvData, char*, int); +static void outputv(ErlDrvData, ErlIOVec*); +static void stop_select(ErlDrvEvent, void*); + +struct erl_drv_entry spawn_driver_entry = { + spawn_init, + spawn_start, + stop, + output, + ready_input, + ready_output, + "spawn", + NULL, + NULL, + NULL, + NULL, + NULL, + NULL, + NULL, + NULL, + NULL, + ERL_DRV_EXTENDED_MARKER, + ERL_DRV_EXTENDED_MAJOR_VERSION, + ERL_DRV_EXTENDED_MINOR_VERSION, + ERL_DRV_FLAG_USE_PORT_LOCKING, + NULL, NULL, + stop_select +}; +struct erl_drv_entry fd_driver_entry = { + NULL, + fd_start, + fd_stop, + output, + ready_input, + ready_output, + "fd", + NULL, + NULL, + fd_control, + NULL, + outputv, + NULL, /* ready_async */ + NULL, /* flush */ + NULL, /* call */ + NULL, /* event */ + ERL_DRV_EXTENDED_MARKER, + ERL_DRV_EXTENDED_MAJOR_VERSION, + ERL_DRV_EXTENDED_MINOR_VERSION, + 0, /* ERL_DRV_FLAGs */ + NULL, /* handle2 */ + NULL, /* process_exit */ + stop_select +}; +struct erl_drv_entry vanilla_driver_entry = { + NULL, + vanilla_start, + stop, + output, + ready_input, + ready_output, + "vanilla", + NULL, + NULL, + NULL, + NULL, + NULL, + NULL, + NULL, /* flush */ + NULL, /* call */ + NULL, /* event */ + ERL_DRV_EXTENDED_MARKER, + ERL_DRV_EXTENDED_MAJOR_VERSION, + ERL_DRV_EXTENDED_MINOR_VERSION, + 0, /* ERL_DRV_FLAGs */ + NULL, /* handle2 */ + NULL, /* process_exit */ + stop_select +}; + +#if defined(USE_THREADS) && !defined(ERTS_SMP) +static int async_drv_init(void); +static ErlDrvData async_drv_start(ErlDrvPort, char*, SysDriverOpts*); +static void async_drv_stop(ErlDrvData); +static void async_drv_input(ErlDrvData, ErlDrvEvent); + +/* INTERNAL use only */ + +struct erl_drv_entry async_driver_entry = { + async_drv_init, + async_drv_start, + async_drv_stop, + NULL, + async_drv_input, + NULL, + "async", + NULL, + NULL, + NULL, + NULL, + NULL, + NULL +}; +#endif + +/* Handle SIGCHLD signals. */ +#if (defined(SIG_SIGSET) || defined(SIG_SIGNAL)) +static RETSIGTYPE onchld(void) +#else +static RETSIGTYPE onchld(int signum) +#endif +{ +#if CHLDWTHR + ASSERT(0); /* We should *never* catch a SIGCHLD signal */ +#elif defined(ERTS_SMP) + smp_sig_notify('C'); +#else + children_died = 1; + ERTS_CHK_IO_INTR(1); /* Make sure we don't sleep in poll */ +#endif +} + +static int set_driver_data(int port_num, + int ifd, + int ofd, + int packet_bytes, + int read_write, + int exit_status, + int pid) +{ + ErtsSysReportExit *report_exit; + + if (!exit_status) + report_exit = NULL; + else { + report_exit = erts_alloc(ERTS_ALC_T_PRT_REP_EXIT, + sizeof(ErtsSysReportExit)); + report_exit->next = report_exit_list; + report_exit->port = erts_port[port_num].id; + report_exit->pid = pid; + report_exit->ifd = read_write & DO_READ ? ifd : -1; + report_exit->ofd = read_write & DO_WRITE ? ofd : -1; +#if CHLDWTHR && !defined(ERTS_SMP) + report_exit->status = 0; +#endif + report_exit_list = report_exit; + } + + if (read_write & DO_READ) { + driver_data[ifd].packet_bytes = packet_bytes; + driver_data[ifd].port_num = port_num; + driver_data[ifd].report_exit = report_exit; + driver_data[ifd].pid = pid; + driver_data[ifd].alive = 1; + driver_data[ifd].status = 0; + if (read_write & DO_WRITE) { + driver_data[ifd].ofd = ofd; + if (ifd != ofd) + driver_data[ofd] = driver_data[ifd]; /* structure copy */ + } else { /* DO_READ only */ + driver_data[ifd].ofd = -1; + } + (void) driver_select(port_num, ifd, (ERL_DRV_READ|ERL_DRV_USE), 1); + return(ifd); + } else { /* DO_WRITE only */ + driver_data[ofd].packet_bytes = packet_bytes; + driver_data[ofd].port_num = port_num; + driver_data[ofd].report_exit = report_exit; + driver_data[ofd].ofd = ofd; + driver_data[ofd].pid = pid; + driver_data[ofd].alive = 1; + driver_data[ofd].status = 0; + return(ofd); + } +} + +static int spawn_init() +{ + int i; +#if CHLDWTHR + erts_thr_opts_t thr_opts = ERTS_THR_OPTS_DEFAULT_INITER; + thr_opts.detached = 0; + thr_opts.suggested_stack_size = 0; /* Smallest possible */ +#endif + + sys_sigset(SIGPIPE, SIG_IGN); /* Ignore - we'll handle the write failure */ + driver_data = (struct driver_data *) + erts_alloc(ERTS_ALC_T_DRV_TAB, max_files * sizeof(struct driver_data)); + erts_smp_atomic_add(&sys_misc_mem_sz, + max_files * sizeof(struct driver_data)); + + for (i = 0; i < max_files; i++) + driver_data[i].pid = -1; + +#if CHLDWTHR + sys_sigblock(SIGCHLD); +#endif + + sys_sigset(SIGCHLD, onchld); /* Reap children */ + +#if CHLDWTHR + erts_thr_create(&child_waiter_tid, child_waiter, NULL, &thr_opts); +#endif + + return 1; +} + +static void close_pipes(int ifd[2], int ofd[2], int read_write) +{ + if (read_write & DO_READ) { + (void) close(ifd[0]); + (void) close(ifd[1]); + } + if (read_write & DO_WRITE) { + (void) close(ofd[0]); + (void) close(ofd[1]); + } +} + +static void init_fd_data(int fd, int prt) +{ + fd_data[fd].buf = NULL; + fd_data[fd].cpos = NULL; + fd_data[fd].remain = 0; + fd_data[fd].sz = 0; + fd_data[fd].psz = 0; +} + +static char **build_unix_environment(char *block) +{ + int i; + int j; + int len; + char *cp; + char **cpp; + char** old_env; + + ERTS_SMP_LC_ASSERT(erts_smp_lc_rwmtx_is_rlocked(&environ_rwmtx)); + + cp = block; + len = 0; + while (*cp != '\0') { + cp += strlen(cp) + 1; + len++; + } + old_env = environ; + while (*old_env++ != NULL) { + len++; + } + + cpp = (char **) erts_alloc_fnf(ERTS_ALC_T_ENVIRONMENT, + sizeof(char *) * (len+1)); + if (cpp == NULL) { + return NULL; + } + + cp = block; + len = 0; + while (*cp != '\0') { + cpp[len] = cp; + cp += strlen(cp) + 1; + len++; + } + + i = len; + for (old_env = environ; *old_env; old_env++) { + char* old = *old_env; + + for (j = 0; j < len; j++) { + char *s, *t; + + s = cpp[j]; + t = old; + while (*s == *t && *s != '=') { + s++, t++; + } + if (*s == '=' && *t == '=') { + break; + } + } + + if (j == len) { /* New version not found */ + cpp[len++] = old; + } + } + + for (j = 0; j < i; j++) { + if (cpp[j][strlen(cpp[j])-1] == '=') { + cpp[j] = cpp[--len]; + } + } + + cpp[len] = NULL; + return cpp; +} + +/* + [arndt] In most Unix systems, including Solaris 2.5, 'fork' allocates memory + in swap space for the child of a 'fork', whereas 'vfork' does not do this. + The natural call to use here is therefore 'vfork'. Due to a bug in + 'vfork' in Solaris 2.5 (apparently fixed in 2.6), using 'vfork' + can be dangerous in what seems to be these circumstances: + If the child code under a vfork sets the signal action to SIG_DFL + (or SIG_IGN) + for any signal which was previously set to a signal handler, the + state of the parent is clobbered, so that the later arrival of + such a signal yields a sigsegv in the parent. If the signal was + not set to a signal handler, but ignored, all seems to work. + If you change the forking code below, beware of this. + */ + +static ErlDrvData spawn_start(ErlDrvPort port_num, char* name, SysDriverOpts* opts) +{ +#define CMD_LINE_PREFIX_STR "exec " +#define CMD_LINE_PREFIX_STR_SZ (sizeof(CMD_LINE_PREFIX_STR) - 1) + + int ifd[2], ofd[2], len, pid, i; + char **volatile new_environ; /* volatile since a vfork() then cannot + cause 'new_environ' to be clobbered + in the parent process. */ + int saved_errno; + long res; + char *cmd_line; +#ifndef QNX + int unbind; +#endif +#if !DISABLE_VFORK + int no_vfork; + size_t no_vfork_sz = sizeof(no_vfork); + + no_vfork = (erts_sys_getenv("ERL_NO_VFORK", + (char *) &no_vfork, + &no_vfork_sz) >= 0); +#endif + + switch (opts->read_write) { + case DO_READ: + if (pipe(ifd) < 0) + return ERL_DRV_ERROR_ERRNO; + if (ifd[0] >= max_files) { + close_pipes(ifd, ofd, opts->read_write); + errno = EMFILE; + return ERL_DRV_ERROR_ERRNO; + } + ofd[1] = -1; /* keep purify happy */ + break; + case DO_WRITE: + if (pipe(ofd) < 0) return ERL_DRV_ERROR_ERRNO; + if (ofd[1] >= max_files) { + close_pipes(ifd, ofd, opts->read_write); + errno = EMFILE; + return ERL_DRV_ERROR_ERRNO; + } + ifd[0] = -1; /* keep purify happy */ + break; + case DO_READ|DO_WRITE: + if (pipe(ifd) < 0) return ERL_DRV_ERROR_ERRNO; + errno = EMFILE; /* default for next two conditions */ + if (ifd[0] >= max_files || pipe(ofd) < 0) { + close_pipes(ifd, ofd, DO_READ); + return ERL_DRV_ERROR_ERRNO; + } + if (ofd[1] >= max_files) { + close_pipes(ifd, ofd, opts->read_write); + errno = EMFILE; + return ERL_DRV_ERROR_ERRNO; + } + break; + default: + ASSERT(0); + return ERL_DRV_ERROR_GENERAL; + } + + if (opts->spawn_type == ERTS_SPAWN_EXECUTABLE) { + /* started with spawn_executable, not with spawn */ + len = strlen(name); + cmd_line = (char *) erts_alloc_fnf(ERTS_ALC_T_TMP, len + 1); + if (!cmd_line) { + close_pipes(ifd, ofd, opts->read_write); + errno = ENOMEM; + return ERL_DRV_ERROR_ERRNO; + } + memcpy((void *) cmd_line,(void *) name, len); + cmd_line[len] = '\0'; + if (access(cmd_line,X_OK) != 0) { + int save_errno = errno; + erts_free(ERTS_ALC_T_TMP, cmd_line); + errno = save_errno; + return ERL_DRV_ERROR_ERRNO; + } + } else { + /* make the string suitable for giving to "sh" */ + len = strlen(name); + cmd_line = (char *) erts_alloc_fnf(ERTS_ALC_T_TMP, + CMD_LINE_PREFIX_STR_SZ + len + 1); + if (!cmd_line) { + close_pipes(ifd, ofd, opts->read_write); + errno = ENOMEM; + return ERL_DRV_ERROR_ERRNO; + } + memcpy((void *) cmd_line, + (void *) CMD_LINE_PREFIX_STR, + CMD_LINE_PREFIX_STR_SZ); + memcpy((void *) (cmd_line + CMD_LINE_PREFIX_STR_SZ), (void *) name, len); + cmd_line[CMD_LINE_PREFIX_STR_SZ + len] = '\0'; + } + + erts_smp_rwmtx_rlock(&environ_rwmtx); + + if (opts->envir == NULL) { + new_environ = environ; + } else if ((new_environ = build_unix_environment(opts->envir)) == NULL) { + erts_smp_rwmtx_runlock(&environ_rwmtx); + erts_free(ERTS_ALC_T_TMP, (void *) cmd_line); + errno = ENOMEM; + return ERL_DRV_ERROR_ERRNO; + } + +#ifndef QNX + /* Block child from SIGINT and SIGUSR1. Must be before fork() + to be safe. */ + block_signals(); + + CHLD_STAT_LOCK; + + unbind = erts_is_scheduler_bound(NULL); + if (unbind) + erts_smp_rwmtx_rlock(&erts_cpu_bind_rwmtx); + +#if !DISABLE_VFORK + /* See fork/vfork discussion before this function. */ + if (no_vfork) { +#endif + + DEBUGF(("Using fork\n")); + pid = fork(); + + if (pid == 0) { + /* The child! Setup child... */ + + if (unbind && erts_unbind_from_cpu(erts_cpuinfo) != 0) + goto child_error; + + /* OBSERVE! + * Keep child setup after vfork() (implemented below and in + * erl_child_setup.c) up to date if changes are made here. + */ + + if (opts->use_stdio) { + if (opts->read_write & DO_READ) { + /* stdout for process */ + if (dup2(ifd[1], 1) < 0) + goto child_error; + if(opts->redir_stderr) + /* stderr for process */ + if (dup2(ifd[1], 2) < 0) + goto child_error; + } + if (opts->read_write & DO_WRITE) + /* stdin for process */ + if (dup2(ofd[0], 0) < 0) + goto child_error; + } + else { /* XXX will fail if ofd[0] == 4 (unlikely..) */ + if (opts->read_write & DO_READ) + if (dup2(ifd[1], 4) < 0) + goto child_error; + if (opts->read_write & DO_WRITE) + if (dup2(ofd[0], 3) < 0) + goto child_error; + } + + for (i = opts->use_stdio ? 3 : 5; i < max_files; i++) + (void) close(i); + + if (opts->wd && chdir(opts->wd) < 0) + goto child_error; + +#if defined(USE_SETPGRP_NOARGS) /* SysV */ + (void) setpgrp(); +#elif defined(USE_SETPGRP) /* BSD */ + (void) setpgrp(0, getpid()); +#else /* POSIX */ + (void) setsid(); +#endif + + unblock_signals(); + + if (opts->spawn_type == ERTS_SPAWN_EXECUTABLE) { + if (opts->argv == NULL) { + execle(cmd_line,cmd_line,(char *) NULL, new_environ); + } else { + if (opts->argv[0] == erts_default_arg0) { + opts->argv[0] = cmd_line; + } + execve(cmd_line, opts->argv, new_environ); + if (opts->argv[0] == cmd_line) { + opts->argv[0] = erts_default_arg0; + } + } + } else { + execle("/bin/sh", "sh", "-c", cmd_line, (char *) NULL, new_environ); + } + child_error: + _exit(1); + } +#if !DISABLE_VFORK + } + else { /* Use vfork() */ + char **cs_argv= erts_alloc(ERTS_ALC_T_TMP,(CS_ARGV_NO_OF_ARGS + 1)* + sizeof(char *)); + char fd_close_range[44]; /* 44 bytes are enough to */ + char dup2_op[CS_ARGV_NO_OF_DUP2_OPS][44]; /* hold any "%d:%d" string */ + /* on a 64-bit machine. */ + + /* Setup argv[] for the child setup program (implemented in + erl_child_setup.c) */ + i = 0; + if (opts->use_stdio) { + if (opts->read_write & DO_READ){ + /* stdout for process */ + sprintf(&dup2_op[i++][0], "%d:%d", ifd[1], 1); + if(opts->redir_stderr) + /* stderr for process */ + sprintf(&dup2_op[i++][0], "%d:%d", ifd[1], 2); + } + if (opts->read_write & DO_WRITE) + /* stdin for process */ + sprintf(&dup2_op[i++][0], "%d:%d", ofd[0], 0); + } else { /* XXX will fail if ofd[0] == 4 (unlikely..) */ + if (opts->read_write & DO_READ) + sprintf(&dup2_op[i++][0], "%d:%d", ifd[1], 4); + if (opts->read_write & DO_WRITE) + sprintf(&dup2_op[i++][0], "%d:%d", ofd[0], 3); + } + for (; i < CS_ARGV_NO_OF_DUP2_OPS; i++) + strcpy(&dup2_op[i][0], "-"); + sprintf(fd_close_range, "%d:%d", opts->use_stdio ? 3 : 5, max_files-1); + + cs_argv[CS_ARGV_PROGNAME_IX] = child_setup_prog; + cs_argv[CS_ARGV_WD_IX] = opts->wd ? opts->wd : "."; + cs_argv[CS_ARGV_UNBIND_IX] + = (unbind ? erts_get_unbind_from_cpu_str(erts_cpuinfo) : "false"); + cs_argv[CS_ARGV_FD_CR_IX] = fd_close_range; + for (i = 0; i < CS_ARGV_NO_OF_DUP2_OPS; i++) + cs_argv[CS_ARGV_DUP2_OP_IX(i)] = &dup2_op[i][0]; + + if (opts->spawn_type == ERTS_SPAWN_EXECUTABLE) { + int num = 0; + int j = 0; + if (opts->argv != NULL) { + for(; opts->argv[num] != NULL; ++num) + ; + } + cs_argv = erts_realloc(ERTS_ALC_T_TMP,cs_argv, (CS_ARGV_NO_OF_ARGS + 1 + num + 1) * sizeof(char *)); + cs_argv[CS_ARGV_CMD_IX] = "-"; + cs_argv[CS_ARGV_NO_OF_ARGS] = cmd_line; + if (opts->argv != NULL) { + for (;opts->argv[j] != NULL; ++j) { + if (opts->argv[j] == erts_default_arg0) { + cs_argv[CS_ARGV_NO_OF_ARGS + 1 + j] = cmd_line; + } else { + cs_argv[CS_ARGV_NO_OF_ARGS + 1 + j] = opts->argv[j]; + } + } + } + cs_argv[CS_ARGV_NO_OF_ARGS + 1 + j] = NULL; + } else { + cs_argv[CS_ARGV_CMD_IX] = cmd_line; /* Command */ + cs_argv[CS_ARGV_NO_OF_ARGS] = NULL; + } + DEBUGF(("Using vfork\n")); + pid = vfork(); + + if (pid == 0) { + /* The child! */ + + /* Observe! + * OTP-4389: The child setup program (implemented in + * erl_child_setup.c) will perform the necessary setup of the + * child before it execs to the user program. This because + * vfork() only allow an *immediate* execve() or _exit() in the + * child. + */ + execve(child_setup_prog, cs_argv, new_environ); + _exit(1); + } + erts_free(ERTS_ALC_T_TMP,cs_argv); + } +#endif + + if (unbind) + erts_smp_rwmtx_runlock(&erts_cpu_bind_rwmtx); + + if (pid == -1) { + saved_errno = errno; + CHLD_STAT_UNLOCK; + erts_smp_rwmtx_runlock(&environ_rwmtx); + erts_free(ERTS_ALC_T_TMP, (void *) cmd_line); + unblock_signals(); + close_pipes(ifd, ofd, opts->read_write); + errno = saved_errno; + return ERL_DRV_ERROR_ERRNO; + } +#else /* QNX */ + if (opts->use_stdio) { + if (opts->read_write & DO_READ) + qnx_spawn_options.iov[1] = ifd[1]; /* stdout for process */ + if (opts->read_write & DO_WRITE) + qnx_spawn_options.iov[0] = ofd[0]; /* stdin for process */ + } + else { + if (opts->read_write & DO_READ) + qnx_spawn_options.iov[4] = ifd[1]; + if (opts->read_write & DO_WRITE) + qnx_spawn_options.iov[3] = ofd[0]; + } + /* Close fds on exec */ + for (i = 3; i < max_files; i++) + fcntl(i, F_SETFD, 1); + + qnx_spawn_options.flags = _SPAWN_SETSID; + if ((pid = spawnl(P_NOWAIT, "/bin/sh", "/bin/sh", "-c", cmd_line, + (char *) 0)) < 0) { + erts_free(ERTS_ALC_T_TMP, (void *) cmd_line); + reset_qnx_spawn(); + erts_smp_rwmtx_runlock(&environ_rwmtx); + close_pipes(ifd, ofd, opts->read_write); + return ERL_DRV_ERROR_GENERAL; + } + reset_qnx_spawn(); +#endif /* QNX */ + + erts_free(ERTS_ALC_T_TMP, (void *) cmd_line); + + if (new_environ != environ) + erts_free(ERTS_ALC_T_ENVIRONMENT, (void *) new_environ); + + if (opts->read_write & DO_READ) + (void) close(ifd[1]); + if (opts->read_write & DO_WRITE) + (void) close(ofd[0]); + + if (opts->read_write & DO_READ) { + SET_NONBLOCKING(ifd[0]); + init_fd_data(ifd[0], port_num); + } + if (opts->read_write & DO_WRITE) { + SET_NONBLOCKING(ofd[1]); + init_fd_data(ofd[1], port_num); + } + + res = set_driver_data(port_num, ifd[0], ofd[1], opts->packet_bytes, + opts->read_write, opts->exit_status, pid); + /* Don't unblock SIGCHLD until now, since the call above must + first complete putting away the info about our new subprocess. */ + unblock_signals(); + +#if CHLDWTHR + ASSERT(children_alive >= 0); + + if (!(children_alive++)) + CHLD_STAT_SIGNAL; /* Wake up child waiter thread if no children + was alive before we fork()ed ... */ +#endif + /* Don't unlock chld_stat_mtx until now of the same reason as above */ + CHLD_STAT_UNLOCK; + + erts_smp_rwmtx_runlock(&environ_rwmtx); + + return (ErlDrvData)res; +#undef CMD_LINE_PREFIX_STR +#undef CMD_LINE_PREFIX_STR_SZ +} + +#ifdef QNX +static reset_qnx_spawn() +{ + int i; + + /* Reset qnx_spawn_options */ + qnx_spawn_options.flags = 0; + qnx_spawn_options.iov[0] = 0xff; + qnx_spawn_options.iov[1] = 0xff; + qnx_spawn_options.iov[2] = 0xff; + qnx_spawn_options.iov[3] = 0xff; +} +#endif + +#define FD_DEF_HEIGHT 24 +#define FD_DEF_WIDTH 80 +/* Control op */ +#define FD_CTRL_OP_GET_WINSIZE 100 + +static int fd_get_window_size(int fd, Uint32 *width, Uint32 *height) +{ +#ifdef TIOCGWINSZ + struct winsize ws; + if (ioctl(fd,TIOCGWINSZ,&ws) == 0) { + *width = (Uint32) ws.ws_col; + *height = (Uint32) ws.ws_row; + return 0; + } +#endif + return -1; +} + +static int fd_control(ErlDrvData drv_data, + unsigned int command, + char *buf, int len, + char **rbuf, int rlen) +{ + int fd = (int)(long)drv_data; + char resbuff[2*sizeof(Uint32)]; + switch (command) { + case FD_CTRL_OP_GET_WINSIZE: + { + Uint32 w,h; + if (fd_get_window_size(fd,&w,&h)) + return 0; + memcpy(resbuff,&w,sizeof(Uint32)); + memcpy(resbuff+sizeof(Uint32),&h,sizeof(Uint32)); + } + break; + default: + return 0; + } + if (rlen < 2*sizeof(Uint32)) { + *rbuf = driver_alloc(2*sizeof(Uint32)); + } + memcpy(*rbuf,resbuff,2*sizeof(Uint32)); + return 2*sizeof(Uint32); +} + +static ErlDrvData fd_start(ErlDrvPort port_num, char* name, + SysDriverOpts* opts) +{ + ErlDrvData res; + + if (((opts->read_write & DO_READ) && opts->ifd >= max_files) || + ((opts->read_write & DO_WRITE) && opts->ofd >= max_files)) + return ERL_DRV_ERROR_GENERAL; + + /* + * Historical: + * + * "Note about nonblocking I/O. + * + * At least on Solaris, setting the write end of a TTY to nonblocking, + * will set the input end to nonblocking as well (and vice-versa). + * If erl is run in a pipeline like this: cat | erl + * the input end of the TTY will be the standard input of cat. + * And cat is not prepared to handle nonblocking I/O." + * + * Actually, the reason for this is not that the tty itself gets set + * in non-blocking mode, but that the "input end" (cat's stdin) and + * the "output end" (erlang's stdout) are typically the "same" file + * descriptor, dup()'ed from a single fd by one of this process' + * ancestors. + * + * The workaround for this problem used to be a rather bad kludge, + * interposing an extra process ("internal cat") between erlang's + * stdout and the original stdout, allowing erlang to set its stdout + * in non-blocking mode without affecting the stdin of the preceding + * process in the pipeline - and being a kludge, it caused all kinds + * of weird problems. + * + * So, this is the current logic: + * + * The only reason to set non-blocking mode on the output fd at all is + * if it's something that can cause a write() to block, of course, + * i.e. primarily if it points to a tty, socket, pipe, or fifo. + * + * If we don't set non-blocking mode when we "should" have, and output + * becomes blocked, the entire runtime system will be suspended - this + * is normally bad of course, and can happen fairly "easily" - e.g. user + * hits ^S on tty - but doesn't necessarily happen. + * + * If we do set non-blocking mode when we "shouldn't" have, the runtime + * system will end up seeing EOF on the input fd (due to the preceding + * process dying), which typically will cause the entire runtime system + * to terminate immediately (due to whatever erlang process is seeing + * the EOF taking it as a signal to halt the system). This is *very* bad. + * + * I.e. we should take a conservative approach, and only set non- + * blocking mode when we a) need to, and b) are reasonably certain + * that it won't be a problem. And as in the example above, the problem + * occurs when input fd and output fd point to different "things". + * + * However, determining that they are not just the same "type" of + * "thing", but actually the same instance of that type of thing, is + * unreasonably complex in many/most cases. + * + * Also, with pipes, sockets, and fifos it's far from obvious that the + * user *wants* non-blocking output: If you're running erlang inside + * some complex pipeline, you're probably not running a real-time system + * that must never stop, but rather *want* it to suspend if the output + * channel is "full". + * + * So, the bottom line: We will only set the output fd non-blocking if + * it points to a tty, and either a) the input fd also points to a tty, + * or b) we can make sure that setting the output fd non-blocking + * doesn't interfere with someone else's input, via a somewhat milder + * kludge than the above. + * + * Also keep in mind that while this code is almost exclusively run as + * a result of an erlang open_port({fd,0,1}, ...), that isn't the only + * case - it can be called with any old pre-existing file descriptors, + * the relations between which (if they're even two) we can only guess + * at - still, we try our best... + */ + + if (opts->read_write & DO_READ) { + init_fd_data(opts->ifd, port_num); + } + if (opts->read_write & DO_WRITE) { + init_fd_data(opts->ofd, port_num); + + /* If we don't have a read end, all bets are off - no non-blocking. */ + if (opts->read_write & DO_READ) { + + if (isatty(opts->ofd)) { /* output fd is a tty:-) */ + + if (isatty(opts->ifd)) { /* input fd is also a tty */ + + /* To really do this "right", we should also check that + input and output fd point to the *same* tty - but + this seems like overkill; ttyname() isn't for free, + and this is a very common case - and it's hard to + imagine a scenario where setting non-blocking mode + here would cause problems - go ahead and do it. */ + + SET_NONBLOCKING(opts->ofd); + + } else { /* output fd is a tty, input fd isn't */ + + /* This is a "problem case", but also common (see the + example above) - i.e. it makes sense to try a bit + harder before giving up on non-blocking mode: Try to + re-open the tty that the output fd points to, and if + successful replace the original one with the "new" fd + obtained this way, and set *that* one in non-blocking + mode. (Yes, this is a kludge.) + + However, re-opening the tty may fail in a couple of + (unusual) cases: + + 1) The name of the tty (or an equivalent one, i.e. + same major/minor number) can't be found, because + it actually lives somewhere other than /dev (or + wherever ttyname() looks for it), and isn't + equivalent to any of those that do live in the + "standard" place - this should be *very* unusual. + + 2) Permissions on the tty don't allow us to open it - + it's perfectly possible to have an fd open to an + object whose permissions wouldn't allow us to open + it. This is not as unusual as it sounds, one case + is if the user has su'ed to someone else (not + root) - we have a read/write fd open to the tty + (because it has been inherited all the way down + here), but we have neither read nor write + permission for the tty. + + In these cases, we finally give up, and don't set the + output fd in non-blocking mode. */ + + char *tty; + int nfd; + + if ((tty = ttyname(opts->ofd)) != NULL && + (nfd = open(tty, O_WRONLY)) != -1) { + dup2(nfd, opts->ofd); + close(nfd); + SET_NONBLOCKING(opts->ofd); + } + } + } + } + } + CHLD_STAT_LOCK; + res = (ErlDrvData)(long)set_driver_data(port_num, opts->ifd, opts->ofd, + opts->packet_bytes, + opts->read_write, 0, -1); + CHLD_STAT_UNLOCK; + return res; +} + +static void clear_fd_data(int fd) +{ + if (fd_data[fd].sz > 0) { + erts_free(ERTS_ALC_T_FD_ENTRY_BUF, (void *) fd_data[fd].buf); + ASSERT(erts_smp_atomic_read(&sys_misc_mem_sz) >= fd_data[fd].sz); + erts_smp_atomic_add(&sys_misc_mem_sz, -1*fd_data[fd].sz); + } + fd_data[fd].buf = NULL; + fd_data[fd].sz = 0; + fd_data[fd].remain = 0; + fd_data[fd].cpos = NULL; + fd_data[fd].psz = 0; +} + +static void nbio_stop_fd(int prt, int fd) +{ + driver_select(prt,fd,DO_READ|DO_WRITE,0); + clear_fd_data(fd); + SET_BLOCKING(fd); +} + +static void fd_stop(ErlDrvData fd) /* Does not close the fds */ +{ + int ofd; + + nbio_stop_fd(driver_data[(int)(long)fd].port_num, (int)(long)fd); + ofd = driver_data[(int)(long)fd].ofd; + if (ofd != (int)(long)fd && ofd != -1) + nbio_stop_fd(driver_data[(int)(long)fd].port_num, (int)(long)ofd); +} + +static ErlDrvData vanilla_start(ErlDrvPort port_num, char* name, + SysDriverOpts* opts) +{ + int flags, fd; + ErlDrvData res; + + flags = (opts->read_write == DO_READ ? O_RDONLY : + opts->read_write == DO_WRITE ? O_WRONLY|O_CREAT|O_TRUNC : + O_RDWR|O_CREAT); + if ((fd = open(name, flags, 0666)) < 0) + return ERL_DRV_ERROR_GENERAL; + if (fd >= max_files) { + close(fd); + return ERL_DRV_ERROR_GENERAL; + } + SET_NONBLOCKING(fd); + init_fd_data(fd, port_num); + + CHLD_STAT_LOCK; + res = (ErlDrvData)(long)set_driver_data(port_num, fd, fd, + opts->packet_bytes, + opts->read_write, 0, -1); + CHLD_STAT_UNLOCK; + return res; +} + +/* Note that driver_data[fd].ifd == fd if the port was opened for reading, */ +/* otherwise (i.e. write only) driver_data[fd].ofd = fd. */ + +static void stop(ErlDrvData fd) +{ + int prt, ofd; + + prt = driver_data[(int)(long)fd].port_num; + nbio_stop_fd(prt, (int)(long)fd); + + ofd = driver_data[(int)(long)fd].ofd; + if (ofd != (int)(long)fd && (int)(long)ofd != -1) + nbio_stop_fd(prt, ofd); + else + ofd = -1; + + CHLD_STAT_LOCK; + + /* Mark as unused. Maybe resetting the 'port_num' slot is better? */ + driver_data[(int)(long)fd].pid = -1; + + CHLD_STAT_UNLOCK; + + /* SMP note: Close has to be last thing done (open file descriptors work + as locks on driver_data[] entries) */ + driver_select(prt, (int)(long)fd, ERL_DRV_USE, 0); /* close(fd); */ + if (ofd >= 0) { + driver_select(prt, (int)(long)ofd, ERL_DRV_USE, 0); /* close(ofd); */ + } +} + +static void outputv(ErlDrvData e, ErlIOVec* ev) +{ + int fd = (int)(long)e; + int ix = driver_data[fd].port_num; + int pb = driver_data[fd].packet_bytes; + int ofd = driver_data[fd].ofd; + int n; + int sz; + char lb[4]; + char* lbp; + int len = ev->size; + + /* (len > ((unsigned long)-1 >> (4-pb)*8)) */ + if (((pb == 2) && (len > 0xffff)) || (pb == 1 && len > 0xff)) { + driver_failure_posix(ix, EINVAL); + return; /* -1; */ + } + put_int32(len, lb); + lbp = lb + (4-pb); + + ev->iov[0].iov_base = lbp; + ev->iov[0].iov_len = pb; + ev->size += pb; + if ((sz = driver_sizeq(ix)) > 0) { + driver_enqv(ix, ev, 0); + if (sz + ev->size >= (1 << 13)) + set_busy_port(ix, 1); + } + else { + int vsize = ev->vsize > MAX_VSIZE ? MAX_VSIZE : ev->vsize; + + n = writev(ofd, (const void *) (ev->iov), vsize); + if (n == ev->size) + return; /* 0;*/ + if (n < 0) { + if ((errno != EINTR) && (errno != ERRNO_BLOCK)) { + driver_failure_posix(ix, errno); + return; /* -1;*/ + } + n = 0; + } + driver_enqv(ix, ev, n); /* n is the skip value */ + driver_select(ix, ofd, ERL_DRV_WRITE|ERL_DRV_USE, 1); + } + /* return 0;*/ +} + + +static void output(ErlDrvData e, char* buf, int len) +{ + int fd = (int)(long)e; + int ix = driver_data[fd].port_num; + int pb = driver_data[fd].packet_bytes; + int ofd = driver_data[fd].ofd; + int n; + int sz; + char lb[4]; + char* lbp; + struct iovec iv[2]; + + /* (len > ((unsigned long)-1 >> (4-pb)*8)) */ + if (((pb == 2) && (len > 0xffff)) || (pb == 1 && len > 0xff)) { + driver_failure_posix(ix, EINVAL); + return; /* -1; */ + } + put_int32(len, lb); + lbp = lb + (4-pb); + + if ((sz = driver_sizeq(ix)) > 0) { + driver_enq(ix, lbp, pb); + driver_enq(ix, buf, len); + if (sz + len + pb >= (1 << 13)) + set_busy_port(ix, 1); + } + else { + iv[0].iov_base = lbp; + iv[0].iov_len = pb; /* should work for pb=0 */ + iv[1].iov_base = buf; + iv[1].iov_len = len; + n = writev(ofd, iv, 2); + if (n == pb+len) + return; /* 0; */ + if (n < 0) { + if ((errno != EINTR) && (errno != ERRNO_BLOCK)) { + driver_failure_posix(ix, errno); + return; /* -1; */ + } + n = 0; + } + if (n < pb) { + driver_enq(ix, lbp+n, pb-n); + driver_enq(ix, buf, len); + } + else { + n -= pb; + driver_enq(ix, buf+n, len-n); + } + driver_select(ix, ofd, ERL_DRV_WRITE|ERL_DRV_USE, 1); + } + return; /* 0; */ +} + +static int port_inp_failure(int port_num, int ready_fd, int res) + /* Result: 0 (eof) or -1 (error) */ +{ + int err = errno; + + ASSERT(res <= 0); + (void) driver_select(port_num, ready_fd, ERL_DRV_READ|ERL_DRV_WRITE, 0); + clear_fd_data(ready_fd); + if (res == 0) { + if (driver_data[ready_fd].report_exit) { + CHLD_STAT_LOCK; + + if (driver_data[ready_fd].alive) { + /* + * We have eof and want to report exit status, but the process + * hasn't exited yet. When it does report_exit_status() will + * driver_select() this fd which will make sure that we get + * back here with driver_data[ready_fd].alive == 0 and + * driver_data[ready_fd].status set. + */ + CHLD_STAT_UNLOCK; + return 0; + } + else { + int status = driver_data[ready_fd].status; + CHLD_STAT_UNLOCK; + + /* We need not be prepared for stopped/continued processes. */ + if (WIFSIGNALED(status)) + status = 128 + WTERMSIG(status); + else + status = WEXITSTATUS(status); + + driver_report_exit(driver_data[ready_fd].port_num, status); + } + } + driver_failure_eof(port_num); + } else { + driver_failure_posix(port_num, err); + } + return 0; +} + +/* fd is the drv_data that is returned from the */ +/* initial start routine */ +/* ready_fd is the descriptor that is ready to read */ + +static void ready_input(ErlDrvData e, ErlDrvEvent ready_fd) +{ + int fd = (int)(long)e; + int port_num; + int packet_bytes; + int res; + Uint h; + + port_num = driver_data[fd].port_num; + packet_bytes = driver_data[fd].packet_bytes; + + if (packet_bytes == 0) { + byte *read_buf = (byte *) erts_alloc(ERTS_ALC_T_SYS_READ_BUF, + ERTS_SYS_READ_BUF_SZ); + res = read(ready_fd, read_buf, ERTS_SYS_READ_BUF_SZ); + if (res < 0) { + if ((errno != EINTR) && (errno != ERRNO_BLOCK)) + port_inp_failure(port_num, ready_fd, res); + } + else if (res == 0) + port_inp_failure(port_num, ready_fd, res); + else + driver_output(port_num, (char*) read_buf, res); + erts_free(ERTS_ALC_T_SYS_READ_BUF, (void *) read_buf); + } + else if (fd_data[ready_fd].remain > 0) { /* We try to read the remainder */ + /* space is allocated in buf */ + res = read(ready_fd, fd_data[ready_fd].cpos, + fd_data[ready_fd].remain); + if (res < 0) { + if ((errno != EINTR) && (errno != ERRNO_BLOCK)) + port_inp_failure(port_num, ready_fd, res); + } + else if (res == 0) { + port_inp_failure(port_num, ready_fd, res); + } + else if (res == fd_data[ready_fd].remain) { /* we're done */ + driver_output(port_num, fd_data[ready_fd].buf, + fd_data[ready_fd].sz); + clear_fd_data(ready_fd); + } + else { /* if (res < fd_data[ready_fd].remain) */ + fd_data[ready_fd].cpos += res; + fd_data[ready_fd].remain -= res; + } + } + else if (fd_data[ready_fd].remain == 0) { /* clean fd */ + byte *read_buf = (byte *) erts_alloc(ERTS_ALC_T_SYS_READ_BUF, + ERTS_SYS_READ_BUF_SZ); + /* We make one read attempt and see what happens */ + res = read(ready_fd, read_buf, ERTS_SYS_READ_BUF_SZ); + if (res < 0) { + if ((errno != EINTR) && (errno != ERRNO_BLOCK)) + port_inp_failure(port_num, ready_fd, res); + } + else if (res == 0) { /* eof */ + port_inp_failure(port_num, ready_fd, res); + } + else if (res < packet_bytes - fd_data[ready_fd].psz) { + memcpy(fd_data[ready_fd].pbuf+fd_data[ready_fd].psz, + read_buf, res); + fd_data[ready_fd].psz += res; + } + else { /* if (res >= packet_bytes) */ + unsigned char* cpos = read_buf; + int bytes_left = res; + + while (1) { + int psz = fd_data[ready_fd].psz; + char* pbp = fd_data[ready_fd].pbuf + psz; + + while(bytes_left && (psz < packet_bytes)) { + *pbp++ = *cpos++; + bytes_left--; + psz++; + } + + if (psz < packet_bytes) { + fd_data[ready_fd].psz = psz; + break; + } + fd_data[ready_fd].psz = 0; + + switch (packet_bytes) { + case 1: h = get_int8(fd_data[ready_fd].pbuf); break; + case 2: h = get_int16(fd_data[ready_fd].pbuf); break; + case 4: h = get_int32(fd_data[ready_fd].pbuf); break; + default: ASSERT(0); return; /* -1; */ + } + + if (h <= (bytes_left)) { + driver_output(port_num, (char*) cpos, h); + cpos += h; + bytes_left -= h; + continue; + } + else { /* The last message we got was split */ + char *buf = erts_alloc_fnf(ERTS_ALC_T_FD_ENTRY_BUF, h); + if (!buf) { + errno = ENOMEM; + port_inp_failure(port_num, ready_fd, -1); + } + else { + erts_smp_atomic_add(&sys_misc_mem_sz, h); + sys_memcpy(buf, cpos, bytes_left); + fd_data[ready_fd].buf = buf; + fd_data[ready_fd].sz = h; + fd_data[ready_fd].remain = h - bytes_left; + fd_data[ready_fd].cpos = buf + bytes_left; + } + break; + } + } + } + erts_free(ERTS_ALC_T_SYS_READ_BUF, (void *) read_buf); + } +} + + +/* fd is the drv_data that is returned from the */ +/* initial start routine */ +/* ready_fd is the descriptor that is ready to read */ + +static void ready_output(ErlDrvData e, ErlDrvEvent ready_fd) +{ + int fd = (int)(long)e; + int ix = driver_data[fd].port_num; + int n; + struct iovec* iv; + int vsize; + + + if ((iv = (struct iovec*) driver_peekq(ix, &vsize)) == NULL) { + driver_select(ix, ready_fd, ERL_DRV_WRITE, 0); + return; /* 0; */ + } + vsize = vsize > MAX_VSIZE ? MAX_VSIZE : vsize; + if ((n = writev(ready_fd, iv, vsize)) > 0) { + if (driver_deq(ix, n) == 0) + set_busy_port(ix, 0); + } + else if (n < 0) { + if (errno == ERRNO_BLOCK || errno == EINTR) + return; /* 0; */ + else { + int res = errno; + driver_select(ix, ready_fd, ERL_DRV_WRITE, 0); + driver_failure_posix(ix, res); + return; /* -1; */ + } + } + return; /* 0; */ +} + +static void stop_select(ErlDrvEvent fd, void* _) +{ + close((int)fd); +} + +/* +** Async opertation support +*/ +#if defined(USE_THREADS) && !defined(ERTS_SMP) +static void +sys_async_ready_failed(int fd, int r, int err) +{ + char buf[120]; + sprintf(buf, "sys_async_ready(): Fatal error: fd=%d, r=%d, errno=%d\n", + fd, r, err); + (void) write(2, buf, strlen(buf)); + abort(); +} + +/* called from threads !! */ +void sys_async_ready(int fd) +{ + int r; + while (1) { + r = write(fd, "0", 1); /* signal main thread fd MUST be async_fd[1] */ + if (r == 1) { + DEBUGF(("sys_async_ready(): r = 1\r\n")); + break; + } + if (r < 0 && errno == EINTR) { + DEBUGF(("sys_async_ready(): r = %d\r\n", r)); + continue; + } + sys_async_ready_failed(fd, r, errno); + } +} + +static int async_drv_init(void) +{ + async_fd[0] = -1; + async_fd[1] = -1; + return 0; +} + +static ErlDrvData async_drv_start(ErlDrvPort port_num, + char* name, SysDriverOpts* opts) +{ + if (async_fd[0] != -1) + return ERL_DRV_ERROR_GENERAL; + if (pipe(async_fd) < 0) + return ERL_DRV_ERROR_GENERAL; + + DEBUGF(("async_drv_start: %d\r\n", port_num)); + + SET_NONBLOCKING(async_fd[0]); + driver_select(port_num, async_fd[0], ERL_DRV_READ, 1); + + if (init_async(async_fd[1]) < 0) + return ERL_DRV_ERROR_GENERAL; + return (ErlDrvData)port_num; +} + +static void async_drv_stop(ErlDrvData e) +{ + int port_num = (int)(long)e; + + DEBUGF(("async_drv_stop: %d\r\n", port_num)); + + exit_async(); + + driver_select(port_num, async_fd[0], ERL_DRV_READ, 0); + + close(async_fd[0]); + close(async_fd[1]); + async_fd[0] = async_fd[1] = -1; +} + + +static void async_drv_input(ErlDrvData e, ErlDrvEvent fd) +{ + char *buf[32]; + DEBUGF(("async_drv_input\r\n")); + while (read((int) fd, (void *) buf, 32) > 0); /* fd MUST be async_fd[0] */ + check_async_ready(); /* invoke all async_ready */ +} +#endif + +void erts_do_break_handling(void) +{ + struct termios temp_mode; + int saved = 0; + + /* + * Most functions that do_break() calls are intentionally not thread safe; + * therefore, make sure that all threads but this one are blocked before + * proceeding! + */ + erts_smp_block_system(0); + /* + * NOTE: since we allow gc we are not allowed to lock + * (any) process main locks while blocking system... + */ + + /* during break we revert to initial settings */ + /* this is done differently for oldshell */ + if (using_oldshell && !replace_intr) { + SET_BLOCKING(1); + } + else if (isatty(0)) { + tcgetattr(0,&temp_mode); + tcsetattr(0,TCSANOW,&initial_tty_mode); + saved = 1; + } + + /* call the break handling function, reset the flag */ + do_break(); + + ERTS_UNSET_BREAK_REQUESTED; + + fflush(stdout); + + /* after break we go back to saved settings */ + if (using_oldshell && !replace_intr) { + SET_NONBLOCKING(1); + } + else if (saved) { + tcsetattr(0,TCSANOW,&temp_mode); + } + + erts_smp_release_system(); +} + +/* Fills in the systems representation of the jam/beam process identifier. +** The Pid is put in STRING representation in the supplied buffer, +** no interpretatione of this should be done by the rest of the +** emulator. The buffer should be at least 21 bytes long. +*/ +void sys_get_pid(char *buffer){ + pid_t p = getpid(); + /* Assume the pid is scalar and can rest in an unsigned long... */ + sprintf(buffer,"%lu",(unsigned long) p); +} + +int +erts_sys_putenv(char *buffer, int sep_ix) +{ + int res; + char *env; +#ifdef HAVE_COPYING_PUTENV + env = buffer; +#else + Uint sz = strlen(buffer)+1; + env = erts_alloc(ERTS_ALC_T_PUTENV_STR, sz); + erts_smp_atomic_add(&sys_misc_mem_sz, sz); + strcpy(env,buffer); +#endif + erts_smp_rwmtx_rwlock(&environ_rwmtx); + res = putenv(env); + erts_smp_rwmtx_rwunlock(&environ_rwmtx); + return res; +} + +int +erts_sys_getenv(char *key, char *value, size_t *size) +{ + char *orig_value; + int res; + erts_smp_rwmtx_rlock(&environ_rwmtx); + orig_value = getenv(key); + if (!orig_value) + res = -1; + else { + size_t len = sys_strlen(orig_value); + if (len >= *size) { + *size = len + 1; + res = 1; + } + else { + *size = len; + sys_memcpy((void *) value, (void *) orig_value, len+1); + res = 0; + } + } + erts_smp_rwmtx_runlock(&environ_rwmtx); + return res; +} + +void +sys_init_io(void) +{ + fd_data = (struct fd_data *) + erts_alloc(ERTS_ALC_T_FD_TAB, max_files * sizeof(struct fd_data)); + erts_smp_atomic_add(&sys_misc_mem_sz, + max_files * sizeof(struct fd_data)); + +#ifdef USE_THREADS +#ifdef ERTS_SMP + if (init_async(-1) < 0) + erl_exit(1, "Failed to initialize async-threads\n"); +#else + { + /* This is speical stuff, starting a driver from the + * system routines, but is a nice way of handling stuff + * the erlang way + */ + SysDriverOpts dopts; + int ret; + + sys_memset((void*)&dopts, 0, sizeof(SysDriverOpts)); + add_driver_entry(&async_driver_entry); + ret = erts_open_driver(NULL, NIL, "async", &dopts, NULL); + DEBUGF(("open_driver = %d\n", ret)); + if (ret < 0) + erl_exit(1, "Failed to open async driver\n"); + erts_port[ret].status |= ERTS_PORT_SFLG_IMMORTAL; + } +#endif +#endif + +} + +#if (0) /* unused? */ +static int write_fill(fd, buf, len) +int fd, len; +char *buf; +{ + int i, done = 0; + + do { + if ((i = write(fd, buf+done, len-done)) < 0) { + if (errno != EINTR) + return (i); + i = 0; + } + done += i; + } while (done < len); + return (len); +} +#endif + +extern const char pre_loaded_code[]; +extern Preload pre_loaded[]; + +void erts_sys_alloc_init(void) +{ + elib_ensure_initialized(); +} + +void *erts_sys_alloc(ErtsAlcType_t t, void *x, Uint sz) +{ + void *res = malloc((size_t) sz); +#if HAVE_ERTS_MSEG + if (!res) { + erts_mseg_clear_cache(); + return malloc((size_t) sz); + } +#endif + return res; +} + +void *erts_sys_realloc(ErtsAlcType_t t, void *x, void *p, Uint sz) +{ + void *res = realloc(p, (size_t) sz); +#if HAVE_ERTS_MSEG + if (!res) { + erts_mseg_clear_cache(); + return realloc(p, (size_t) sz); + } +#endif + return res; +} + +void erts_sys_free(ErtsAlcType_t t, void *x, void *p) +{ + free(p); +} + +/* Return a pointer to a vector of names of preloaded modules */ + +Preload* +sys_preloaded(void) +{ + return pre_loaded; +} + +/* Return a pointer to preloaded code for module "module" */ +unsigned char* +sys_preload_begin(Preload* p) +{ + return p->code; +} + +/* Clean up if allocated */ +void sys_preload_end(Preload* p) +{ + /* Nothing */ +} + +/* Read a key from console (?) */ + +int sys_get_key(fd) +int fd; +{ + int c; + unsigned char rbuf[64]; + + fflush(stdout); /* Flush query ??? */ + + if ((c = read(fd,rbuf,64)) <= 0) { + return c; + } + + return rbuf[0]; +} + + +#ifdef DEBUG + +extern int erts_initialized; +void +erl_assert_error(char* expr, char* file, int line) +{ + fflush(stdout); + fprintf(stderr, "Assertion failed: %s in %s, line %d\n", + expr, file, line); + fflush(stderr); +#if !defined(ERTS_SMP) && 0 + /* Writing a crashdump from a failed assertion when smp support + * is enabled almost a guaranteed deadlocking, don't even bother. + * + * It could maybe be useful (but I'm not convinced) to write the + * crashdump if smp support is disabled... + */ + if (erts_initialized) + erl_crash_dump(file, line, "Assertion failed: %s\n", expr); +#endif + abort(); +} + +void +erl_debug(char* fmt, ...) +{ + char sbuf[1024]; /* Temporary buffer. */ + va_list va; + + if (debug_log) { + va_start(va, fmt); + vsprintf(sbuf, fmt, va); + va_end(va); + fprintf(stderr, "%s", sbuf); + } +} + +#endif /* DEBUG */ + +static ERTS_INLINE void +report_exit_status(ErtsSysReportExit *rep, int status) +{ + Port *pp; +#ifdef ERTS_SMP + CHLD_STAT_UNLOCK; +#endif + pp = erts_id2port_sflgs(rep->port, + NULL, + 0, + ERTS_PORT_SFLGS_INVALID_DRIVER_LOOKUP); +#ifdef ERTS_SMP + CHLD_STAT_LOCK; +#endif + if (pp) { + if (rep->ifd >= 0) { + driver_data[rep->ifd].alive = 0; + driver_data[rep->ifd].status = status; + (void) driver_select((ErlDrvPort) internal_port_index(pp->id), + rep->ifd, + (ERL_DRV_READ|ERL_DRV_USE), + 1); + } + if (rep->ofd >= 0) { + driver_data[rep->ofd].alive = 0; + driver_data[rep->ofd].status = status; + (void) driver_select((ErlDrvPort) internal_port_index(pp->id), + rep->ofd, + (ERL_DRV_WRITE|ERL_DRV_USE), + 1); + } + erts_port_release(pp); + } + erts_free(ERTS_ALC_T_PRT_REP_EXIT, rep); +} + +#if !CHLDWTHR /* ---------------------------------------------------------- */ + +#define ERTS_REPORT_EXIT_STATUS report_exit_status + +static int check_children(void) +{ + int res = 0; + int pid; + int status; + +#ifndef ERTS_SMP + if (children_died) +#endif + { + sys_sigblock(SIGCHLD); + CHLD_STAT_LOCK; + while ((pid = waitpid(-1, &status, WNOHANG)) > 0) + note_child_death(pid, status); +#ifndef ERTS_SMP + children_died = 0; +#endif + CHLD_STAT_UNLOCK; + sys_sigrelease(SIGCHLD); + res = 1; + } + return res; +} + +#ifdef ERTS_SMP + +void +erts_check_children(void) +{ + (void) check_children(); +} + +#endif + +#elif CHLDWTHR && defined(ERTS_SMP) /* ------------------------------------- */ + +#define ERTS_REPORT_EXIT_STATUS report_exit_status + +#define check_children() (0) + + +#else /* CHLDWTHR && !defined(ERTS_SMP) ------------------------------------ */ + +#define ERTS_REPORT_EXIT_STATUS initiate_report_exit_status + +static ERTS_INLINE void +initiate_report_exit_status(ErtsSysReportExit *rep, int status) +{ + rep->next = report_exit_transit_list; + rep->status = status; + report_exit_transit_list = rep; + /* + * We need the scheduler thread to call check_children(). + * If the scheduler thread is sleeping in a poll with a + * timeout, we need to wake the scheduler thread. We use the + * functionality of the async driver to do this, instead of + * implementing yet another driver doing the same thing. A + * little bit ugly, but it works... + */ + sys_async_ready(async_fd[1]); +} + +static int check_children(void) +{ + int res; + ErtsSysReportExit *rep; + CHLD_STAT_LOCK; + rep = report_exit_transit_list; + res = rep != NULL; + while (rep) { + ErtsSysReportExit *curr_rep = rep; + rep = rep->next; + report_exit_status(curr_rep, curr_rep->status); + } + report_exit_transit_list = NULL; + CHLD_STAT_UNLOCK; + return res; +} + +#endif /* ------------------------------------------------------------------ */ + +static void note_child_death(int pid, int status) +{ + ErtsSysReportExit **repp = &report_exit_list; + ErtsSysReportExit *rep = report_exit_list; + + while (rep) { + if (pid == rep->pid) { + *repp = rep->next; + ERTS_REPORT_EXIT_STATUS(rep, status); + break; + } + repp = &rep->next; + rep = rep->next; + } +} + +#if CHLDWTHR + +static void * +child_waiter(void *unused) +{ + int pid; + int status; + +#ifdef ERTS_ENABLE_LOCK_CHECK + erts_lc_set_thread_name("child waiter"); +#endif + + while(1) { +#ifdef DEBUG + int waitpid_errno; +#endif + pid = waitpid(-1, &status, 0); +#ifdef DEBUG + waitpid_errno = errno; +#endif + CHLD_STAT_LOCK; + if (pid < 0) { + ASSERT(waitpid_errno == ECHILD); + } + else { + children_alive--; + ASSERT(children_alive >= 0); + note_child_death(pid, status); + } + while (!children_alive) + CHLD_STAT_WAIT; /* Wait for children to wait on... :) */ + CHLD_STAT_UNLOCK; + } + + return NULL; +} + +#endif + +/* + * Called from schedule() when it runs out of runnable processes, + * or when Erlang code has performed INPUT_REDUCTIONS reduction + * steps. runnable == 0 iff there are no runnable Erlang processes. + */ +void +erl_sys_schedule(int runnable) +{ +#ifdef ERTS_SMP + ERTS_CHK_IO(!runnable); + ERTS_SMP_LC_ASSERT(!ERTS_LC_IS_BLOCKING); +#else + ERTS_CHK_IO_INTR(0); + if (runnable) { + ERTS_CHK_IO(0); /* Poll for I/O */ + check_async_ready(); /* Check async completions */ + } else { + int wait_for_io = !check_async_ready(); + if (wait_for_io) + wait_for_io = !check_children(); + ERTS_CHK_IO(wait_for_io); + } + (void) check_children(); +#endif +} + + +#ifdef ERTS_SMP + +static erts_smp_tid_t sig_dispatcher_tid; + +static void +smp_sig_notify(char c) +{ + int res; + do { + /* write() is async-signal safe (according to posix) */ + res = write(sig_notify_fds[1], &c, 1); + } while (res < 0 && errno == EINTR); + if (res != 1) { + char msg[] = + "smp_sig_notify(): Failed to notify signal-dispatcher thread " + "about received signal"; + (void) write(2, msg, sizeof(msg)); + abort(); + } +} + +static void * +signal_dispatcher_thread_func(void *unused) +{ + int initialized = 0; +#if !CHLDWTHR + int notify_check_children = 0; +#endif +#ifdef ERTS_ENABLE_LOCK_CHECK + erts_lc_set_thread_name("signal_dispatcher"); +#endif + while (1) { + char buf[32]; + int res, i; + /* Block on read() waiting for a signal notification to arrive... */ + res = read(sig_notify_fds[0], (void *) &buf[0], 32); + if (res < 0) { + if (errno == EINTR) + continue; + erl_exit(ERTS_ABORT_EXIT, + "signal-dispatcher thread got unexpected error: %s (%d)\n", + erl_errno_id(errno), + errno); + } + for (i = 0; i < res; i++) { + /* + * NOTE 1: The signal dispatcher thread should not do work + * that takes a substantial amount of time (except + * perhaps in test and debug builds). It needs to + * be responsive, i.e, it should only dispatch work + * to other threads. + * + * NOTE 2: The signal dispatcher thread is not a blockable + * thread (i.e., it hasn't called + * erts_register_blockable_thread()). This is + * intentional. We want to be able to interrupt + * writing of a crash dump by hitting C-c twice. + * Since it isn't a blockable thread it is important + * that it doesn't change the state of any data that + * a blocking thread expects to have exclusive access + * to (unless the signal dispatcher itself explicitly + * is blocking all blockable threads). + */ + switch (buf[i]) { + case 0: /* Emulator initialized */ + initialized = 1; +#if !CHLDWTHR + if (!notify_check_children) +#endif + break; +#if !CHLDWTHR + case 'C': /* SIGCHLD */ + if (initialized) + erts_smp_notify_check_children_needed(); + else + notify_check_children = 1; + break; +#endif + case 'I': /* SIGINT */ + break_requested(); + break; + case 'Q': /* SIGQUIT */ + quit_requested(); + break; + case '1': /* SIGUSR1 */ + sigusr1_exit(); + break; +#ifdef QUANTIFY + case '2': /* SIGUSR2 */ + quantify_save_data(); /* Might take a substantial amount of + time, but this is a test/debug + build */ + break; +#endif + default: + erl_exit(ERTS_ABORT_EXIT, + "signal-dispatcher thread received unknown " + "signal notification: '%c'\n", + buf[i]); + } + } + ERTS_SMP_LC_ASSERT(!ERTS_LC_IS_BLOCKING); + } + return NULL; +} + +static void +init_smp_sig_notify(void) +{ + erts_smp_thr_opts_t thr_opts = ERTS_SMP_THR_OPTS_DEFAULT_INITER; + thr_opts.detached = 1; + + if (pipe(sig_notify_fds) < 0) { + erl_exit(ERTS_ABORT_EXIT, + "Failed to create signal-dispatcher pipe: %s (%d)\n", + erl_errno_id(errno), + errno); + } + + /* Start signal handler thread */ + erts_smp_thr_create(&sig_dispatcher_tid, + signal_dispatcher_thread_func, + NULL, + &thr_opts); +} + +void +erts_sys_main_thread(void) +{ + erts_thread_disable_fpe(); + /* Become signal receiver thread... */ +#ifdef ERTS_ENABLE_LOCK_CHECK + erts_lc_set_thread_name("signal_receiver"); +#endif + + smp_sig_notify(0); /* Notify initialized */ + while (1) { + /* Wait for a signal to arrive... */ +#ifdef DEBUG + int res = +#else + (void) +#endif + select(0, NULL, NULL, NULL, NULL); + ASSERT(res < 0); + ASSERT(errno == EINTR); + } +} + +#endif /* ERTS_SMP */ + +#ifdef ERTS_ENABLE_KERNEL_POLL /* get_value() is currently only used when + kernel-poll is enabled */ + +/* Get arg marks argument as handled by + putting NULL in argv */ +static char * +get_value(char* rest, char** argv, int* ip) +{ + char *param = argv[*ip]+1; + argv[*ip] = NULL; + if (*rest == '\0') { + char *next = argv[*ip + 1]; + if (next[0] == '-' + && next[1] == '-' + && next[2] == '\0') { + erts_fprintf(stderr, "bad \"%s\" value: \n", param); + erts_usage(); + } + (*ip)++; + argv[*ip] = NULL; + return next; + } + return rest; +} + +#endif /* ERTS_ENABLE_KERNEL_POLL */ + +void +erl_sys_args(int* argc, char** argv) +{ + int i, j; + + i = 1; + + ASSERT(argc && argv); + + while (i < *argc) { + if(argv[i][0] == '-') { + switch (argv[i][1]) { +#ifdef ERTS_ENABLE_KERNEL_POLL + case 'K': { + char *arg = get_value(argv[i] + 2, argv, &i); + if (strcmp("true", arg) == 0) { + erts_use_kernel_poll = 1; + } + else if (strcmp("false", arg) == 0) { + erts_use_kernel_poll = 0; + } + else { + erts_fprintf(stderr, "bad \"K\" value: %s\n", arg); + erts_usage(); + } + break; + } +#endif + case '-': + goto done_parsing; + default: + break; + } + } + i++; + } + + done_parsing: + +#ifdef ERTS_ENABLE_KERNEL_POLL + if (erts_use_kernel_poll) { + char no_kp[10]; + size_t no_kp_sz = sizeof(no_kp); + int res = erts_sys_getenv("ERL_NO_KERNEL_POLL", no_kp, &no_kp_sz); + if (res > 0 + || (res == 0 + && sys_strcmp("false", no_kp) != 0 + && sys_strcmp("FALSE", no_kp) != 0)) { + erts_use_kernel_poll = 0; + } + } +#endif + + init_check_io(); + +#ifdef ERTS_SMP + init_smp_sig_notify(); +#endif + + /* Handled arguments have been marked with NULL. Slide arguments + not handled towards the beginning of argv. */ + for (i = 0, j = 0; i < *argc; i++) { + if (argv[i]) + argv[j++] = argv[i]; + } + *argc = j; +} + + +#ifdef ERTS_TIMER_THREAD + +/* + * Interruptible-wait facility: low-level synchronisation state + * and methods that are implementation dependent. + * + * Constraint: Every implementation must define 'struct erts_iwait' + * with a field 'erts_smp_atomic_t state;'. + */ + +/* values for struct erts_iwait's state field */ +#define IWAIT_WAITING 0 +#define IWAIT_AWAKE 1 +#define IWAIT_INTERRUPT 2 + +#if 0 /* XXX: needs feature test in erts/configure.in */ + +/* + * This is an implementation of the interruptible wait facility on + * top of Linux-specific futexes. + */ +#include <asm/unistd.h> +#define FUTEX_WAIT 0 +#define FUTEX_WAKE 1 +static int sys_futex(void *futex, int op, int val, const struct timespec *timeout) +{ + return syscall(__NR_futex, futex, op, val, timeout); +} + +struct erts_iwait { + erts_smp_atomic_t state; /* &state.counter is our futex */ +}; + +static void iwait_lowlevel_init(struct erts_iwait *iwait) { /* empty */ } + +static void iwait_lowlevel_wait(struct erts_iwait *iwait, struct timeval *delay) +{ + struct timespec timeout; + int res; + + timeout.tv_sec = delay->tv_sec; + timeout.tv_nsec = delay->tv_usec * 1000; + res = sys_futex((void*)&iwait->state.counter, FUTEX_WAIT, IWAIT_WAITING, &timeout); + if (res < 0 && errno != ETIMEDOUT && errno != EWOULDBLOCK && errno != EINTR) + perror("FUTEX_WAIT"); +} + +static void iwait_lowlevel_interrupt(struct erts_iwait *iwait) +{ + int res = sys_futex((void*)&iwait->state.counter, FUTEX_WAKE, 1, NULL); + if (res < 0) + perror("FUTEX_WAKE"); +} + +#else /* using poll() or select() */ + +/* + * This is an implementation of the interruptible wait facility on + * top of pipe(), poll() or select(), read(), and write(). + */ +struct erts_iwait { + erts_smp_atomic_t state; + int read_fd; /* wait polls and reads this fd */ + int write_fd; /* interrupt writes this fd */ +}; + +static void iwait_lowlevel_init(struct erts_iwait *iwait) +{ + int fds[2]; + + if (pipe(fds) < 0) { + perror("pipe()"); + exit(1); + } + iwait->read_fd = fds[0]; + iwait->write_fd = fds[1]; +} + +#if defined(ERTS_USE_POLL) + +#include <sys/poll.h> +#define PERROR_POLL "poll()" + +static int iwait_lowlevel_poll(int read_fd, struct timeval *delay) +{ + struct pollfd pollfd; + int timeout; + + pollfd.fd = read_fd; + pollfd.events = POLLIN; + pollfd.revents = 0; + timeout = delay->tv_sec * 1000 + delay->tv_usec / 1000; + return poll(&pollfd, 1, timeout); +} + +#else /* !ERTS_USE_POLL */ + +#include <sys/select.h> +#define PERROR_POLL "select()" + +static int iwait_lowlevel_poll(int read_fd, struct timeval *delay) +{ + fd_set readfds; + + FD_ZERO(&readfds); + FD_SET(read_fd, &readfds); + return select(read_fd + 1, &readfds, NULL, NULL, delay); +} + +#endif /* !ERTS_USE_POLL */ + +static void iwait_lowlevel_wait(struct erts_iwait *iwait, struct timeval *delay) +{ + int res; + char buf[64]; + + res = iwait_lowlevel_poll(iwait->read_fd, delay); + if (res > 0) + (void)read(iwait->read_fd, buf, sizeof buf); + else if (res < 0 && errno != EINTR) + perror(PERROR_POLL); +} + +static void iwait_lowlevel_interrupt(struct erts_iwait *iwait) +{ + int res = write(iwait->write_fd, "!", 1); + if (res < 0) + perror("write()"); +} + +#endif /* using poll() or select() */ + +#if 0 /* not using poll() or select() */ +/* + * This is an implementation of the interruptible wait facility on + * top of pthread_cond_timedwait(). This has two problems: + * 1. pthread_cond_timedwait() requires an absolute time point, + * so the relative delay must be converted to absolute time. + * Worse, this breaks if the machine's time is adjusted while + * we're preparing to wait. + * 2. Each cond operation requires additional mutex lock/unlock operations. + * + * Problem 2 is probably not too bad on Linux (they'll just become + * relatively cheap futex operations), but problem 1 is the real killer. + * Only use this implementation if no better alternatives are available! + */ +struct erts_iwait { + erts_smp_atomic_t state; + pthread_cond_t cond; + pthread_mutex_t mutex; +}; + +static void iwait_lowlevel_init(struct erts_iwait *iwait) +{ + iwait->cond = (pthread_cond_t) PTHREAD_COND_INITIALIZER; + iwait->mutex = (pthread_mutex_t) PTHREAD_MUTEX_INITIALIZER; +} + +static void iwait_lowlevel_wait(struct erts_iwait *iwait, struct timeval *delay) +{ + struct timeval tmp; + struct timespec timeout; + + /* Due to pthread_cond_timedwait()'s use of absolute + time, this must be the real gettimeofday(), _not_ + the "smoothed" one beam/erl_time_sup.c implements. */ + gettimeofday(&tmp, NULL); + + tmp.tv_sec += delay->tv_sec; + tmp.tv_usec += delay->tv_usec; + if (tmp.tv_usec >= 1000*1000) { + tmp.tv_usec -= 1000*1000; + tmp.tv_sec += 1; + } + timeout.tv_sec = tmp.tv_sec; + timeout.tv_nsec = tmp.tv_usec * 1000; + pthread_mutex_lock(&iwait->mutex); + pthread_cond_timedwait(&iwait->cond, &iwait->mutex, &timeout); + pthread_mutex_unlock(&iwait->mutex); +} + +static void iwait_lowlevel_interrupt(struct erts_iwait *iwait) +{ + pthread_mutex_lock(&iwait->mutex); + pthread_cond_signal(&iwait->cond); + pthread_mutex_unlock(&iwait->mutex); +} + +#endif /* not using POLL */ + +/* + * Interruptible-wait facility. This is just a wrapper around the + * low-level synchronisation code, where we maintain our logical + * state in order to suppress some state transitions. + */ + +struct erts_iwait *erts_iwait_init(void) +{ + struct erts_iwait *iwait = malloc(sizeof *iwait); + if (!iwait) { + perror("malloc"); + exit(1); + } + iwait_lowlevel_init(iwait); + erts_smp_atomic_init(&iwait->state, IWAIT_AWAKE); + return iwait; +} + +void erts_iwait_wait(struct erts_iwait *iwait, struct timeval *delay) +{ + if (erts_smp_atomic_xchg(&iwait->state, IWAIT_WAITING) != IWAIT_INTERRUPT) + iwait_lowlevel_wait(iwait, delay); + erts_smp_atomic_set(&iwait->state, IWAIT_AWAKE); +} + +void erts_iwait_interrupt(struct erts_iwait *iwait) +{ + if (erts_smp_atomic_xchg(&iwait->state, IWAIT_INTERRUPT) == IWAIT_WAITING) + iwait_lowlevel_interrupt(iwait); +} + +#endif /* ERTS_TIMER_THREAD */ diff --git a/erts/emulator/sys/unix/sys_float.c b/erts/emulator/sys/unix/sys_float.c new file mode 100644 index 0000000000..15da6ab45c --- /dev/null +++ b/erts/emulator/sys/unix/sys_float.c @@ -0,0 +1,815 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2001-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include "sys.h" +#include "global.h" +#include "erl_process.h" + + +#ifdef NO_FPE_SIGNALS + +void +erts_sys_init_float(void) +{ +# ifdef SIGFPE + sys_sigset(SIGFPE, SIG_IGN); /* Ignore so we can test for NaN and Inf */ +# endif +} + +static ERTS_INLINE void set_current_fp_exception(unsigned long pc) +{ + /* nothing to do */ +} + +#else /* !NO_FPE_SIGNALS */ + +#ifdef ERTS_SMP +static erts_tsd_key_t fpe_key; + +/* once-only initialisation early in the main thread (via erts_sys_init_float()) */ +static void erts_init_fp_exception(void) +{ + /* XXX: the wrappers prevent using a pthread destructor to + deallocate the key's value; so when/where do we do that? */ + erts_tsd_key_create(&fpe_key); +} + +void erts_thread_init_fp_exception(void) +{ + unsigned long *fpe = erts_alloc(ERTS_ALC_T_FP_EXCEPTION, sizeof(*fpe)); + *fpe = 0L; + erts_tsd_set(fpe_key, fpe); +} + +static ERTS_INLINE volatile unsigned long *erts_thread_get_fp_exception(void) +{ + return (volatile unsigned long*)erts_tsd_get(fpe_key); +} +#else /* !SMP */ +#define erts_init_fp_exception() /*empty*/ +static volatile unsigned long fp_exception; +#define erts_thread_get_fp_exception() (&fp_exception) +#endif /* SMP */ + +volatile unsigned long *erts_get_current_fp_exception(void) +{ + Process *c_p; + + c_p = erts_get_current_process(); + if (c_p) + return &c_p->fp_exception; + return erts_thread_get_fp_exception(); +} + +static void set_current_fp_exception(unsigned long pc) +{ + volatile unsigned long *fpexnp = erts_get_current_fp_exception(); + ASSERT(fpexnp != NULL); + *fpexnp = pc; +} + +void erts_fp_check_init_error(volatile unsigned long *fpexnp) +{ + char buf[64]; + snprintf(buf, sizeof buf, "ERTS_FP_CHECK_INIT at %p: detected unhandled FPE at %p\r\n", + __builtin_return_address(0), (void*)*fpexnp); + write(2, buf, strlen(buf)); + *fpexnp = 0; +#if defined(__i386__) || defined(__x86_64__) + erts_restore_fpu(); +#endif +} + +/* Is there no standard identifier for Darwin/MacOSX ? */ +#if defined(__APPLE__) && defined(__MACH__) && !defined(__DARWIN__) +#define __DARWIN__ 1 +#endif + +#if (defined(__i386__) || defined(__x86_64__)) && defined(__GNUC__) + +static void unmask_x87(void) +{ + unsigned short cw; + + __asm__ __volatile__("fstcw %0" : "=m"(cw)); + cw &= ~(0x01|0x04|0x08); /* unmask IM, ZM, OM */ + __asm__ __volatile__("fldcw %0" : : "m"(cw)); +} + +/* mask x87 FPE, return true if the previous state was unmasked */ +static int mask_x87(void) +{ + unsigned short cw; + int unmasked; + + __asm__ __volatile__("fstcw %0" : "=m"(cw)); + unmasked = (cw & (0x01|0x04|0x08)) == 0; + /* or just set cw = 0x37f */ + cw |= (0x01|0x04|0x08); /* mask IM, ZM, OM */ + __asm__ __volatile__("fldcw %0" : : "m"(cw)); + return unmasked; +} + +static void unmask_sse2(void) +{ + unsigned int mxcsr; + + __asm__ __volatile__("stmxcsr %0" : "=m"(mxcsr)); + mxcsr &= ~(0x003F|0x0680); /* clear exn flags, unmask OM, ZM, IM (not PM, UM, DM) */ + __asm__ __volatile__("ldmxcsr %0" : : "m"(mxcsr)); +} + +/* mask SSE2 FPE, return true if the previous state was unmasked */ +static int mask_sse2(void) +{ + unsigned int mxcsr; + int unmasked; + + __asm__ __volatile__("stmxcsr %0" : "=m"(mxcsr)); + unmasked = (mxcsr & 0x0680) == 0; + /* or just set mxcsr = 0x1f80 */ + mxcsr &= ~0x003F; /* clear exn flags */ + mxcsr |= 0x0680; /* mask OM, ZM, IM (not PM, UM, DM) */ + __asm__ __volatile__("ldmxcsr %0" : : "m"(mxcsr)); + return unmasked; +} + +#if defined(__x86_64__) + +static inline int cpu_has_sse2(void) { return 1; } + +#else /* !__x86_64__ */ + +/* + * Check if an x86-32 processor has SSE2. + */ +static unsigned int xor_eflags(unsigned int mask) +{ + unsigned int eax, edx; + + eax = mask; /* eax = mask */ + __asm__("pushfl\n\t" + "popl %0\n\t" /* edx = original EFLAGS */ + "xorl %0, %1\n\t" /* eax = mask ^ EFLAGS */ + "pushl %1\n\t" + "popfl\n\t" /* new EFLAGS = mask ^ original EFLAGS */ + "pushfl\n\t" + "popl %1\n\t" /* eax = new EFLAGS */ + "xorl %0, %1\n\t" /* eax = new EFLAGS ^ old EFLAGS */ + "pushl %0\n\t" + "popfl" /* restore original EFLAGS */ + : "=d"(edx), "=a"(eax) + : "1"(eax)); + return eax; +} + +static __inline__ unsigned int cpuid_eax(unsigned int op) +{ + unsigned int eax, save_ebx; + + /* In PIC mode i386 reserves EBX. So we must save + and restore it ourselves to not upset gcc. */ + __asm__( + "movl %%ebx, %1\n\t" + "cpuid\n\t" + "movl %1, %%ebx" + : "=a"(eax), "=m"(save_ebx) + : "0"(op) + : "cx", "dx"); + return eax; +} + +static __inline__ unsigned int cpuid_edx(unsigned int op) +{ + unsigned int eax, edx, save_ebx; + + /* In PIC mode i386 reserves EBX. So we must save + and restore it ourselves to not upset gcc. */ + __asm__( + "movl %%ebx, %2\n\t" + "cpuid\n\t" + "movl %2, %%ebx" + : "=a"(eax), "=d"(edx), "=m"(save_ebx) + : "0"(op) + : "cx"); + return edx; +} + +/* The AC bit, bit #18, is a new bit introduced in the EFLAGS + * register on the Intel486 processor to generate alignment + * faults. This bit cannot be set on the Intel386 processor. + */ +static __inline__ int is_386(void) +{ + return ((xor_eflags(1<<18) >> 18) & 1) == 0; +} + +/* Newer x86 processors have a CPUID instruction, as indicated by + * the ID bit (#21) in EFLAGS being modifiable. + */ +static __inline__ int has_CPUID(void) +{ + return (xor_eflags(1<<21) >> 21) & 1; +} + +static int cpu_has_sse2(void) +{ + unsigned int maxlev, features; + static int has_sse2 = -1; + + if (has_sse2 >= 0) + return has_sse2; + has_sse2 = 0; + + if (is_386()) + return 0; + if (!has_CPUID()) + return 0; + maxlev = cpuid_eax(0); + /* Intel A-step Pentium had a preliminary version of CPUID. + It also didn't have SSE2. */ + if ((maxlev & 0xFFFFFF00) == 0x0500) + return 0; + /* If max level is zero then CPUID cannot report any features. */ + if (maxlev == 0) + return 0; + features = cpuid_edx(1); + has_sse2 = (features & (1 << 26)) != 0; + + return has_sse2; +} +#endif /* !__x86_64__ */ + +static void unmask_fpe(void) +{ + __asm__ __volatile__("fnclex"); + unmask_x87(); + if (cpu_has_sse2()) + unmask_sse2(); +} + +static void unmask_fpe_conditional(int unmasked) +{ + if (unmasked) + unmask_fpe(); +} + +/* mask x86 FPE, return true if the previous state was unmasked */ +static int mask_fpe(void) +{ + int unmasked; + + unmasked = mask_x87(); + if (cpu_has_sse2()) + unmasked |= mask_sse2(); + return unmasked; +} + +void erts_restore_fpu(void) +{ + __asm__ __volatile__("fninit"); + unmask_x87(); + if (cpu_has_sse2()) + unmask_sse2(); +} + +#elif defined(__sparc__) && defined(__linux__) + +#if defined(__arch64__) +#define LDX "ldx" +#define STX "stx" +#else +#define LDX "ld" +#define STX "st" +#endif + +static void unmask_fpe(void) +{ + unsigned long fsr; + + __asm__(STX " %%fsr, %0" : "=m"(fsr)); + fsr &= ~(0x1FUL << 23); /* clear FSR[TEM] field */ + fsr |= (0x1AUL << 23); /* enable NV, OF, DZ exceptions */ + __asm__ __volatile__(LDX " %0, %%fsr" : : "m"(fsr)); +} + +static void unmask_fpe_conditional(int unmasked) +{ + if (unmasked) + unmask_fpe(); +} + +/* mask SPARC FPE, return true if the previous state was unmasked */ +static int mask_fpe(void) +{ + unsigned long fsr; + int unmasked; + + __asm__(STX " %%fsr, %0" : "=m"(fsr)); + unmasked = ((fsr >> 23) & 0x1A) == 0x1A; + fsr &= ~(0x1FUL << 23); /* clear FSR[TEM] field */ + __asm__ __volatile__(LDX " %0, %%fsr" : : "m"(fsr)); + return unmasked; +} + +#elif (defined(__powerpc__) && defined(__linux__)) || (defined(__ppc__) && defined(__DARWIN__)) + +#if defined(__linux__) +#include <sys/prctl.h> + +static void set_fpexc_precise(void) +{ + if (prctl(PR_SET_FPEXC, PR_FP_EXC_PRECISE) < 0) { + perror("PR_SET_FPEXC"); + exit(1); + } +} + +#elif defined(__DARWIN__) + +#include <mach/mach.h> +#include <pthread.h> + +/* + * FE0 FE1 MSR bits + * 0 0 floating-point exceptions disabled + * 0 1 floating-point imprecise nonrecoverable + * 1 0 floating-point imprecise recoverable + * 1 1 floating-point precise mode + * + * Apparently: + * - Darwin 5.5 (MacOS X <= 10.1) starts with FE0 == FE1 == 0, + * and resets FE0 and FE1 to 0 after each SIGFPE. + * - Darwin 6.0 (MacOS X 10.2) starts with FE0 == FE1 == 1, + * and does not reset FE0 or FE1 after a SIGFPE. + */ +#define FE0_MASK (1<<11) +#define FE1_MASK (1<<8) + +/* a thread cannot get or set its own MSR bits */ +static void *fpu_fpe_enable(void *arg) +{ + thread_t t = *(thread_t*)arg; + struct ppc_thread_state state; + unsigned int state_size = PPC_THREAD_STATE_COUNT; + + if (thread_get_state(t, PPC_THREAD_STATE, (natural_t*)&state, &state_size) != KERN_SUCCESS) { + perror("thread_get_state"); + exit(1); + } + if ((state.srr1 & (FE1_MASK|FE0_MASK)) != (FE1_MASK|FE0_MASK)) { +#if 1 + /* This would also have to be performed in the SIGFPE handler + to work around the MSR reset older Darwin releases do. */ + state.srr1 |= (FE1_MASK|FE0_MASK); + thread_set_state(t, PPC_THREAD_STATE, (natural_t*)&state, state_size); +#else + fprintf(stderr, "srr1 == 0x%08x, your Darwin is too old\n", state.srr1); + exit(1); +#endif + } + return NULL; /* Ok, we appear to be on Darwin 6.0 or later */ +} + +static void set_fpexc_precise(void) +{ + thread_t self = mach_thread_self(); + pthread_t enabler; + + if (pthread_create(&enabler, NULL, fpu_fpe_enable, &self)) { + perror("pthread_create"); + } else if (pthread_join(enabler, NULL)) { + perror("pthread_join"); + } +} + +#endif + +static void set_fpscr(unsigned int fpscr) +{ + union { + double d; + unsigned int fpscr[2]; + } u; + + u.fpscr[0] = 0xFFF80000; + u.fpscr[1] = fpscr; + __asm__ __volatile__("mtfsf 255,%0" : : "f"(u.d)); +} + +static unsigned int get_fpscr(void) +{ + union { + double d; + unsigned int fpscr[2]; + } u; + + __asm__("mffs %0" : "=f"(u.d)); + return u.fpscr[1]; +} + +static void unmask_fpe(void) +{ + set_fpexc_precise(); + set_fpscr(0x80|0x40|0x10); /* VE, OE, ZE; not UE or XE */ +} + +static void unmask_fpe_conditional(int unmasked) +{ + if (unmasked) + unmask_fpe(); +} + +/* mask PowerPC FPE, return true if the previous state was unmasked */ +static int mask_fpe(void) +{ + int unmasked; + + unmasked = (get_fpscr() & (0x80|0x40|0x10)) == (0x80|0x40|0x10); + set_fpscr(0x00); + return unmasked; +} + +#else + +static void unmask_fpe(void) +{ + fpsetmask(FP_X_INV | FP_X_OFL | FP_X_DZ); +} + +static void unmask_fpe_conditional(int unmasked) +{ + if (unmasked) + unmask_fpe(); +} + +/* mask IEEE FPE, return true if previous state was unmasked */ +static int mask_fpe(void) +{ + const fp_except unmasked_mask = FP_X_INV | FP_X_OFL | FP_X_DZ; + fp_except old_mask; + + old_mask = fpsetmask(0); + return (old_mask & unmasked_mask) == unmasked_mask; +} + +#endif + +#if (defined(__linux__) && (defined(__i386__) || defined(__x86_64__) || defined(__sparc__) || defined(__powerpc__))) || (defined(__DARWIN__) && (defined(__i386__) || defined(__x86_64__) || defined(__ppc__))) || (defined(__FreeBSD__) && (defined(__x86_64__) || defined(__i386__))) || (defined(__OpenBSD__) && defined(__x86_64__)) || (defined(__sun__) && defined(__x86_64__)) + +#if defined(__linux__) && defined(__i386__) +#if !defined(X86_FXSR_MAGIC) +#define X86_FXSR_MAGIC 0x0000 +#endif +#elif defined(__FreeBSD__) && defined(__x86_64__) +#include <sys/types.h> +#include <machine/fpu.h> +#elif defined(__FreeBSD__) && defined(__i386__) +#include <sys/types.h> +#include <machine/npx.h> +#elif defined(__DARWIN__) +#include <machine/signal.h> +#elif defined(__OpenBSD__) && defined(__x86_64__) +#include <sys/types.h> +#include <machine/fpu.h> +#endif +#if !(defined(__OpenBSD__) && defined(__x86_64__)) +#include <ucontext.h> +#endif +#include <string.h> + +#if defined(__linux__) && defined(__x86_64__) +#define mc_pc(mc) ((mc)->gregs[REG_RIP]) +#elif defined(__linux__) && defined(__i386__) +#define mc_pc(mc) ((mc)->gregs[REG_EIP]) +#elif defined(__DARWIN__) && defined(__i386__) +#ifdef DARWIN_MODERN_MCONTEXT +#define mc_pc(mc) ((mc)->__ss.__eip) +#else +#define mc_pc(mc) ((mc)->ss.eip) +#endif +#elif defined(__DARWIN__) && defined(__x86_64__) +#ifdef DARWIN_MODERN_MCONTEXT +#define mc_pc(mc) ((mc)->__ss.__rip) +#else +#define mc_pc(mc) ((mc)->ss.rip) +#endif +#elif defined(__FreeBSD__) && defined(__x86_64__) +#define mc_pc(mc) ((mc)->mc_rip) +#elif defined(__FreeBSD__) && defined(__i386__) +#define mc_pc(mc) ((mc)->mc_eip) +#elif defined(__OpenBSD__) && defined(__x86_64__) +#define mc_pc(mc) ((mc)->sc_rip) +#elif defined(__sun__) && defined(__x86_64__) +#define mc_pc(mc) ((mc)->gregs[REG_RIP]) +#endif + +static void fpe_sig_action(int sig, siginfo_t *si, void *puc) +{ + ucontext_t *uc = puc; + unsigned long pc; + +#if defined(__linux__) +#if defined(__x86_64__) + mcontext_t *mc = &uc->uc_mcontext; + fpregset_t fpstate = mc->fpregs; + pc = mc_pc(mc); + /* A failed SSE2 instruction will restart. To avoid + looping we mask SSE2 exceptions now and unmask them + again later in erts_check_fpe()/erts_restore_fpu(). + On RISCs we update PC to skip the failed instruction, + but the ever increasing complexity of the x86 instruction + set encoding makes that a poor solution here. */ + fpstate->mxcsr = 0x1F80; + fpstate->swd &= ~0xFF; +#elif defined(__i386__) + mcontext_t *mc = &uc->uc_mcontext; + fpregset_t fpstate = mc->fpregs; + pc = mc_pc(mc); + if ((fpstate->status >> 16) == X86_FXSR_MAGIC) + ((struct _fpstate*)fpstate)->mxcsr = 0x1F80; + fpstate->sw &= ~0xFF; +#elif defined(__sparc__) && defined(__arch64__) + /* on SPARC the 3rd parameter points to a sigcontext not a ucontext */ + struct sigcontext *sc = (struct sigcontext*)puc; + pc = sc->sigc_regs.tpc; + sc->sigc_regs.tpc = sc->sigc_regs.tnpc; + sc->sigc_regs.tnpc += 4; +#elif defined(__sparc__) + /* on SPARC the 3rd parameter points to a sigcontext not a ucontext */ + struct sigcontext *sc = (struct sigcontext*)puc; + pc = sc->si_regs.pc; + sc->si_regs.pc = sc->si_regs.npc; + sc->si_regs.npc = (unsigned long)sc->si_regs.npc + 4; +#elif defined(__powerpc__) +#if defined(__powerpc64__) + mcontext_t *mc = &uc->uc_mcontext; + unsigned long *regs = &mc->gp_regs[0]; +#else + mcontext_t *mc = uc->uc_mcontext.uc_regs; + unsigned long *regs = &mc->gregs[0]; +#endif + pc = regs[PT_NIP]; + regs[PT_NIP] += 4; + regs[PT_FPSCR] = 0x80|0x40|0x10; /* VE, OE, ZE; not UE or XE */ +#endif +#elif defined(__DARWIN__) && (defined(__i386__) || defined(__x86_64__)) +#ifdef DARWIN_MODERN_MCONTEXT + mcontext_t mc = uc->uc_mcontext; + pc = mc_pc(mc); + mc->__fs.__fpu_mxcsr = 0x1F80; + *(unsigned short *)&mc->__fs.__fpu_fsw &= ~0xFF; +#else + mcontext_t mc = uc->uc_mcontext; + pc = mc_pc(mc); + mc->fs.fpu_mxcsr = 0x1F80; + *(unsigned short *)&mc->fs.fpu_fsw &= ~0xFF; +#endif /* DARWIN_MODERN_MCONTEXT */ +#elif defined(__DARWIN__) && defined(__ppc__) + mcontext_t mc = uc->uc_mcontext; + pc = mc->ss.srr0; + mc->ss.srr0 += 4; + mc->fs.fpscr = 0x80|0x40|0x10; +#elif defined(__FreeBSD__) && defined(__x86_64__) + mcontext_t *mc = &uc->uc_mcontext; + struct savefpu *savefpu = (struct savefpu*)&mc->mc_fpstate; + struct envxmm *envxmm = &savefpu->sv_env; + pc = mc_pc(mc); + envxmm->en_mxcsr = 0x1F80; + envxmm->en_sw &= ~0xFF; +#elif defined(__FreeBSD__) && defined(__i386__) + mcontext_t *mc = &uc->uc_mcontext; + union savefpu *savefpu = (union savefpu*)&mc->mc_fpstate; + pc = mc_pc(mc); + if (mc->mc_fpformat == _MC_FPFMT_XMM) { + struct envxmm *envxmm = &savefpu->sv_xmm.sv_env; + envxmm->en_mxcsr = 0x1F80; + envxmm->en_sw &= ~0xFF; + } else { + struct env87 *env87 = &savefpu->sv_87.sv_env; + env87->en_sw &= ~0xFF; + } +#elif defined(__OpenBSD__) && defined(__x86_64__) + struct fxsave64 *fxsave = uc->sc_fpstate; + pc = mc_pc(uc); + fxsave->fx_mxcsr = 0x1F80; + fxsave->fx_fsw &= ~0xFF; +#elif defined(__sun__) && defined(__x86_64__) + mcontext_t *mc = &uc->uc_mcontext; + struct fpchip_state *fpstate = &mc->fpregs.fp_reg_set.fpchip_state; + pc = mc_pc(mc); + fpstate->mxcsr = 0x1F80; + fpstate->sw &= ~0xFF; +#endif +#if 0 + { + char buf[64]; + snprintf(buf, sizeof buf, "%s: FPE at %p\r\n", __FUNCTION__, (void*)pc); + write(2, buf, strlen(buf)); + } +#endif + set_current_fp_exception(pc); +} + +static void erts_thread_catch_fp_exceptions(void) +{ + struct sigaction act; + memset(&act, 0, sizeof act); + act.sa_sigaction = fpe_sig_action; + act.sa_flags = SA_SIGINFO; + sigaction(SIGFPE, &act, NULL); + unmask_fpe(); +} + +#else /* !((__linux__ && (__i386__ || __x86_64__ || __powerpc__)) || (__DARWIN__ && (__i386__ || __x86_64__ || __ppc__))) */ + +static void fpe_sig_handler(int sig) +{ + set_current_fp_exception(1); /* XXX: convert to sigaction so we can get the trap PC */ +} + +static void erts_thread_catch_fp_exceptions(void) +{ + sys_sigset(SIGFPE, fpe_sig_handler); + unmask_fpe(); +} + +#endif /* (__linux__ && (__i386__ || __x86_64__ || __powerpc__)) || (__DARWIN__ && (__i386__ || __x86_64__ || __ppc__))) */ + +/* once-only initialisation early in the main thread */ +void erts_sys_init_float(void) +{ + erts_init_fp_exception(); + erts_thread_catch_fp_exceptions(); + erts_printf_block_fpe = erts_sys_block_fpe; + erts_printf_unblock_fpe = erts_sys_unblock_fpe; +} + +#endif /* NO_FPE_SIGNALS */ + +void erts_thread_init_float(void) +{ +#ifdef ERTS_SMP + /* This allows Erlang schedulers to leave Erlang-process context + and still have working FP exceptions. XXX: is this needed? */ + erts_thread_init_fp_exception(); +#endif + +#ifndef NO_FPE_SIGNALS + /* NOTE: + * erts_thread_disable_fpe() is called in all threads at + * creation. We at least need to call unmask_fpe() + */ +#if defined(__DARWIN__) || defined(__FreeBSD__) + /* Darwin (7.9.0) does not appear to propagate FP exception settings + to a new thread from its parent. So if we want FP exceptions, we + must manually re-enable them in each new thread. + FreeBSD 6.1 appears to suffer from a similar issue. */ + erts_thread_catch_fp_exceptions(); +#else + unmask_fpe(); +#endif + +#endif +} + +void erts_thread_disable_fpe(void) +{ +#if !defined(NO_FPE_SIGNALS) + (void)mask_fpe(); +#endif +} + +#if !defined(NO_FPE_SIGNALS) +int erts_sys_block_fpe(void) +{ + return mask_fpe(); +} + +void erts_sys_unblock_fpe(int unmasked) +{ + unmask_fpe_conditional(unmasked); +} +#endif + +/* The following check is incorporated from the Vee machine */ + +#define ISDIGIT(d) ((d) >= '0' && (d) <= '9') + +/* + ** Convert a double to ascii format 0.dddde[+|-]ddd + ** return number of characters converted + ** + ** These two functions should maybe use localeconv() to pick up + ** the current radix character, but since it is uncertain how + ** expensive such a system call is, and since no-one has heard + ** of other radix characters than '.' and ',' an ad-hoc + ** low execution time solution is used instead. + */ + +int +sys_double_to_chars(double fp, char *buf) +{ + char *s = buf; + + (void) sprintf(buf, "%.20e", fp); + /* Search upto decimal point */ + if (*s == '+' || *s == '-') s++; + while (ISDIGIT(*s)) s++; + if (*s == ',') *s++ = '.'; /* Replace ',' with '.' */ + /* Scan to end of string */ + while (*s) s++; + return s-buf; /* i.e strlen(buf) */ +} + +/* Float conversion */ + +int +sys_chars_to_double(char* buf, double* fp) +{ +#ifndef NO_FPE_SIGNALS + volatile unsigned long *fpexnp = erts_get_current_fp_exception(); +#endif + char *s = buf, *t, *dp; + + /* Robert says that something like this is what he really wanted: + * (The [.,] radix test is NOT what Robert wanted - it was added later) + * + * 7 == sscanf(Tbuf, "%[+-]%[0-9][.,]%[0-9]%[eE]%[+-]%[0-9]%s", ....); + * if (*s2 == 0 || *s3 == 0 || *s4 == 0 || *s6 == 0 || *s7) + * break; + */ + + /* Scan string to check syntax. */ + if (*s == '+' || *s == '-') s++; + if (!ISDIGIT(*s)) /* Leading digits. */ + return -1; + while (ISDIGIT(*s)) s++; + if (*s != '.' && *s != ',') /* Decimal part. */ + return -1; + dp = s++; /* Remember decimal point pos just in case */ + if (!ISDIGIT(*s)) + return -1; + while (ISDIGIT(*s)) s++; + if (*s == 'e' || *s == 'E') { + /* There is an exponent. */ + s++; + if (*s == '+' || *s == '-') s++; + if (!ISDIGIT(*s)) + return -1; + while (ISDIGIT(*s)) s++; + } + if (*s) /* That should be it */ + return -1; + +#ifdef NO_FPE_SIGNALS + errno = 0; +#endif + __ERTS_FP_CHECK_INIT(fpexnp); + *fp = strtod(buf, &t); + __ERTS_FP_ERROR_THOROUGH(fpexnp, *fp, return -1); + if (t != s) { /* Whole string not scanned */ + /* Try again with other radix char */ + *dp = (*dp == '.') ? ',' : '.'; + errno = 0; + __ERTS_FP_CHECK_INIT(fpexnp); + *fp = strtod(buf, &t); + __ERTS_FP_ERROR_THOROUGH(fpexnp, *fp, return -1); + } + +#ifdef NO_FPE_SIGNALS + if (errno == ERANGE && (*fp == 0.0 || *fp == HUGE_VAL || *fp == -HUGE_VAL)) { + return -1; + } +#endif + return 0; +} + +int +matherr(struct exception *exc) +{ +#if !defined(NO_FPE_SIGNALS) + set_current_fp_exception((unsigned long)__builtin_return_address(0)); +#endif + return 1; +} diff --git a/erts/emulator/sys/unix/sys_time.c b/erts/emulator/sys/unix/sys_time.c new file mode 100644 index 0000000000..fcce54a2c4 --- /dev/null +++ b/erts/emulator/sys/unix/sys_time.c @@ -0,0 +1,134 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2005-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +/* These need to be undef:ed to not break activation of + * micro level process accounting on /proc/self + */ +#ifdef _LARGEFILE_SOURCE +# undef _LARGEFILE_SOURCE +#endif +#ifdef _FILE_OFFSET_BITS +# undef _FILE_OFFSET_BITS +#endif + +#include "sys.h" +#include "global.h" + +#ifdef NO_SYSCONF +# define TICKS_PER_SEC() HZ +#else +#define TICKS_PER_SEC() sysconf(_SC_CLK_TCK) +#endif + +#ifdef HAVE_GETHRVTIME_PROCFS_IOCTL +# include <unistd.h> +# include <sys/types.h> +# include <sys/stat.h> +# include <sys/signal.h> +# include <sys/fault.h> +# include <sys/syscall.h> +# include <sys/procfs.h> +# include <fcntl.h> +#endif + +/******************* Routines for time measurement *********************/ + +int erts_ticks_per_sec = 0; /* Will be SYS_CLK_TCK in erl_unix_sys.h */ +int erts_ticks_per_sec_wrap = 0; /* Will be SYS_CLK_TCK_WRAP */ +static int ticks_bsr = 0; /* Shift wrapped tick value this much to the right */ + +/* + * init timers, chose a tick length, and return it. + * Unix is priviliged when it comes to time, as erl_time_sup.c + * does almost everything. Other platforms have to + * emulate Unix in this sense. + */ +int sys_init_time(void) +{ + /* + * This (erts_ticks_per_sec) is only for times() (CLK_TCK), + * the resolution is always one millisecond.. + */ + if ((erts_ticks_per_sec = TICKS_PER_SEC()) < 0) + erl_exit(1, "Can't get clock ticks/sec\n"); + if (erts_ticks_per_sec >= 1000) { + /* Workaround for beta linux kernels, need to be done in runtime + to make erlang run on both 2.4 and 2.5 kernels. In the future, + the kernel ticks might as + well be used as a high res timer instead, but that's for when the + majority uses kernels with HZ == 1024 */ + ticks_bsr = 3; + } else { + ticks_bsr = 0; + } + erts_ticks_per_sec_wrap = (erts_ticks_per_sec >> ticks_bsr); + return SYS_CLOCK_RESOLUTION; +} + +clock_t sys_times_wrap(void) +{ + SysTimes dummy; + clock_t result = (sys_times(&dummy) >> ticks_bsr); + return result; +} + + + + +#ifdef HAVE_GETHRVTIME_PROCFS_IOCTL + +int sys_start_hrvtime(void) +{ + long msacct = PR_MSACCT; + int fd; + + if ( (fd = open("/proc/self", O_WRONLY)) == -1) { + return -1; + } + if (ioctl(fd, PIOCSET, &msacct) < 0) { + close(fd); + return -2; + } + close(fd); + return 0; +} + +int sys_stop_hrvtime(void) +{ + long msacct = PR_MSACCT; + int fd; + + if ( (fd = open("/proc/self", O_WRONLY)) == -1) { + return -1; + } + if (ioctl(fd, PIOCRESET, &msacct) < 0) { + close(fd); + return -2; + } + close(fd); + return 0; +} + +#endif /* HAVE_GETHRVTIME_PROCFS_IOCTL */ + + diff --git a/erts/emulator/sys/vxworks/driver_int.h b/erts/emulator/sys/vxworks/driver_int.h new file mode 100644 index 0000000000..f6bc71a799 --- /dev/null +++ b/erts/emulator/sys/vxworks/driver_int.h @@ -0,0 +1,30 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 1997-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ +/*---------------------------------------------------------------------- +** Purpose : System dependant driver declarations +**---------------------------------------------------------------------- */ + +#ifndef __DRIVER_INT_H__ +#define __DRIVER_INT_H__ + +#include <ioLib.h> + +typedef struct iovec SysIOVec; + +#endif diff --git a/erts/emulator/sys/vxworks/erl_main.c b/erts/emulator/sys/vxworks/erl_main.c new file mode 100644 index 0000000000..c9b44a635a --- /dev/null +++ b/erts/emulator/sys/vxworks/erl_main.c @@ -0,0 +1,45 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2000-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif +#include "sys.h" +#include "erl_vm.h" + +#if defined(__GNUC__) +/* + * The generated assembler does the usual trick (relative + * branch-and-link to next instruction) to get a copy of the + * instruction ptr. Instead of branching to an explicit zero offset, + * it branches to the symbol `__eabi' --- which is expected to be + * undefined and thus zero (if it is defined as non-zero, things will + * be interesting --- as in the Chinese curse). To shut up the VxWorks + * linker, we define `__eabi' as zero. + * + * This is just a work around. It's really Wind River's GCC's code + * generator that should be fixed. + */ +__asm__(".equ __eabi, 0"); +#endif + +void +erl_main(int argc, char **argv) +{ + erl_start(argc, argv); +} diff --git a/erts/emulator/sys/vxworks/erl_vxworks_sys.h b/erts/emulator/sys/vxworks/erl_vxworks_sys.h new file mode 100644 index 0000000000..ae46403600 --- /dev/null +++ b/erts/emulator/sys/vxworks/erl_vxworks_sys.h @@ -0,0 +1,183 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 1997-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ +#ifndef __ERL_VXWORKS_SYS_H__ +#define __ERL_VXWORKS_SYS_H__ + +/* stdarg.h don't work without this one... */ +#include <vxWorks.h> + +#include <stdio.h> +#include <math.h> +#include <limits.h> +#include <stdlib.h> +#define index StringIndexFunctionThatIDontWantDeclared +#include <string.h> +#undef index + + + +#include <sys/times.h> +#include <time.h>/* xxxP */ + +#include <dirent.h> +#include <sys/stat.h> + +/* xxxP from unix_sys.h begin */ + +/* + * Make sure that MAXPATHLEN is defined. + */ + +#ifndef MAXPATHLEN +# ifdef PATH_MAX +# define MAXPATHLEN PATH_MAX +# else +# define MAXPATHLEN 2048 +# endif +#endif + +/* xxxP end */ + + +/* Unimplemented math functions */ +#define NO_ASINH +#define NO_ACOSH +#define NO_ATANH +#define NO_ERF +#define NO_ERFC + +/* Stuff that is useful for port programs, drivers, etc */ +#ifndef VXWORKS +#define VXWORKS +#endif + +#define DONT_USE_MAIN +#define NO_FSYNC +#define NO_MKDIR_MODE +#define NO_UMASK +#define NO_SYMBOLIC_LINKS +#define NO_DEVICE_FILES +#define NO_UID +#define NO_ACCESS +#define NO_FCNTL +#define NO_SYSLOG +#define NO_SYSCONF +#define NO_PWD /* XXX Means what? */ +#define NO_DAEMON +/* This chooses ~250 reductions instead of 500 in config.h */ +#if (CPU == CPU32) +#define SLOW_PROCESSOR +#endif + +/* + * Even though we does not always have small memories on VxWorks + * we certainly does not have virtual memory. + */ +#if !defined(LARGE_MEMORY) +#define SMALL_MEMORY +#endif + +/*************** Floating point exception handling ***************/ + +/* There are no known ways to customize the handling of invalid floating + point operations, such as matherr() or ieee_handler(), in VxWorks 5.1. */ + +#if (CPU == MC68040 || CPU == CPU32 || CPU == PPC860 || CPU == PPC32 || \ + CPU == PPC603 || CPU == PPC604 || CPU == SIMSPARCSOLARIS) + +/* VxWorks 5.1 on Motorola 68040 never generates SIGFPE, but sets the + result of invalid floating point ops to Inf and NaN - unfortunately + the way to test for those values is undocumented and hidden in a + "private" include file... */ +/* Haven't found any better way, as of yet, for ppc860 xxxP*/ + +#include <private/mathP.h> +#define NO_FPE_SIGNALS +#define erts_get_current_fp_exception() NULL +#define __ERTS_FP_CHECK_INIT(fpexnp) do {} while (0) +#define __ERTS_FP_ERROR(fpexnp, f, Action) if (isInf(f) || isNan(f)) { Action; } else {} +#define __ERTS_FP_ERROR_THOROUGH(fpexnp, f, Action) __ERTS_FP_ERROR(fpexnp, f, Action) +#define __ERTS_SAVE_FP_EXCEPTION(fpexnp) +#define __ERTS_RESTORE_FP_EXCEPTION(fpexnp) + +#define ERTS_FP_CHECK_INIT(p) __ERTS_FP_CHECK_INIT(&(p)->fp_exception) +#define ERTS_FP_ERROR(p, f, A) __ERTS_FP_ERROR(&(p)->fp_exception, f, A) +#define ERTS_SAVE_FP_EXCEPTION(p) __ERTS_SAVE_FP_EXCEPTION(&(p)->fp_exception) +#define ERTS_RESTORE_FP_EXCEPTION(p) __ERTS_RESTORE_FP_EXCEPTION(&(p)->fp_exception) +#define ERTS_FP_ERROR_THOROUGH(p, f, A) __ERTS_FP_ERROR_THOROUGH(&(p)->fp_exception, f, A) + +#define erts_sys_block_fpe() 0 +#define erts_sys_unblock_fpe(x) do{}while(0) + +#if (CPU == PPC603) +/* Need fppLib to change the Floating point registers + (fix_registers in sys.c)*/ + +#include <fppLib.h> + +#endif /* PPC603 */ + +#else + +Unsupported CPU value ! + +#endif + +typedef void *GETENV_STATE; + +#define HAVE_GETHRTIME + +extern int erts_clock_rate; + +#define SYS_CLK_TCK (erts_clock_rate) + +#define SYS_CLOCK_RESOLUTION 1 + +typedef struct _vxworks_tms { + clock_t tms_utime; + clock_t tms_stime; + clock_t tms_cutime; + clock_t tms_cstime; +} SysTimes; + +typedef long long SysHrTime; + +typedef struct timeval SysTimeval; + +extern int sys_init_hrtime(void); +extern SysHrTime sys_gethrtime(void); +extern void sys_gettimeofday(SysTimeval *tvp); +extern clock_t sys_times(SysTimes *t); + +#define SIZEOF_SHORT 2 +#define SIZEOF_INT 4 +#define SIZEOF_LONG 4 +#define SIZEOF_VOID_P 4 +#define SIZEOF_SIZE_T 4 +#define SIZEOF_OFF_T 4 + +/* + * Temporary buffer *only* used in sys code. + */ +#define SYS_TMP_BUF_SIZE 65536 + +/* Need to be able to interrupt erts_poll_wait() from signal handler */ +#define ERTS_POLL_NEED_ASYNC_INTERRUPT_SUPPORT + +#endif /* __ERL_VXWORKS_SYS_H__ */ diff --git a/erts/emulator/sys/vxworks/erl_vxworks_sys_ddll.c b/erts/emulator/sys/vxworks/erl_vxworks_sys_ddll.c new file mode 100644 index 0000000000..c56c633b2f --- /dev/null +++ b/erts/emulator/sys/vxworks/erl_vxworks_sys_ddll.c @@ -0,0 +1,253 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2006-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +/* + * Interface functions to the dynamic linker using dl* functions. + * (As far as I know it works on SunOS 4, 5, Linux and FreeBSD. /Seb) + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif +#include <vxWorks.h> +#include <stdio.h> +#include <string.h> +#include <stdarg.h> +#include <a_out.h> +#include <symLib.h> +#include <loadLib.h> +#include <unldLib.h> +#include <moduleLib.h> +#include <sysSymTbl.h> +#include "sys.h" +#include "global.h" +#include "erl_alloc.h" +#include "erl_driver.h" + +#define EXT_LEN 4 +#define FILE_EXT ".eld" +#define ALT_FILE_EXT ".o" +/* ALT_FILE_EXT must not be longer than FILE_EXT */ +#define DRIVER_INIT_SUFFIX "_init" + +static MODULE_ID get_mid(char *); +static FUNCPTR lookup(char *); + +typedef enum { + NoError, + ModuleNotFound, + ModuleNotUnloadable, + UnknownError +} FakeSytemError; + +static char *errcode_tab[] = { + "No error", + "Module/file not found", + "Module cannot be unloaded", + "Unknown error" +}; + +void erl_sys_ddll_init(void) { + return; +} +/* + * Open a shared object + */ +int erts_sys_ddll_open2(char *full_name, void **handle, ErtsSysDdllError* err) +{ + int len; + + if (erts_sys_ddll_open_noext(full_name, handle, err) == ERL_DE_NO_ERROR) { + return ERL_DE_NO_ERROR; + } + if ((len = sys_strlen(full_name)) > PATH_MAX-EXT_LEN) { + return ERL_DE_LOAD_ERROR_NAME_TO_LONG; + } else { + static char dlname[PATH_MAX + 1]; + + sys_strcpy(dlname, full_name); + sys_strcpy(dlname+len, FILE_EXT); + if (erts_sys_ddll_open_noext(dlname, handle, err) == ERL_DE_NO_ERROR) { + return ERL_DE_NO_ERROR; + } + sys_strcpy(dlname+len, ALT_FILE_EXT); + return erts_sys_ddll_open_noext(dlname, handle, err); + } +} +int erts_sys_ddll_open_noext(char *dlname, void **handle, ErtsSysDdllError* err) +{ + MODULE_ID mid; + + if((mid = get_mid(dlname)) == NULL) { + return ERL_DE_DYNAMIC_ERROR_OFFSET - ((int) ModuleNotFound); + } + *handle = (void *) mid; + return ERL_DE_NO_ERROR; +} + +/* + * Find a symbol in the shared object + */ +#define PREALLOC_BUFFER_SIZE 256 +int erts_sys_ddll_sym2(void *handle, char *func_name, void **function, ErtsSysDdllError* err) +{ + FUNCPTR proc; + static char statbuf[PREALLOC_BUFFER_SIZE]; + char *buf = statbuf; + int need; + + if ((proc = lookup(func_name)) == NULL) { + if ((need = strlen(func_name)+2) > PREALLOC_BUFFER_SIZE) { + buf = erts_alloc(ERTS_ALC_T_DDLL_TMP_BUF,need); + } + buf[0] = '_'; + sys_strcpy(buf+1,func_name); + proc = lookup(buf); + if (buf != statbuf) { + erts_free(ERTS_ALC_T_DDLL_TMP_BUF, buf); + } + if (proc == NULL) { + return ERL_DE_LOOKUP_ERROR_NOT_FOUND; + } + } + *function = (void *) proc; + return ERL_DE_NO_ERROR; +} + +/* XXX:PaN These two will be changed with new driver interface! */ + +/* + * Load the driver init function, might appear under different names depending on object arch... + */ + +int erts_sys_ddll_load_driver_init(void *handle, void **function) +{ + MODULE_ID mid = (MODULE_ID) handle; + char *modname; + char *cp; + static char statbuf[PREALLOC_BUFFER_SIZE]; + char *fname = statbuf; + int len; + int res; + void *func; + int need; + + if((modname = moduleNameGet(mid)) == NULL) { + return ERL_DE_DYNAMIC_ERROR_OFFSET - ((int) ModuleNotFound); + } + + if((cp = strrchr(modname, '.')) == NULL) { + len = strlen(modname); + } else { + len = cp - modname; + } + + need = len + strlen(DRIVER_INIT_SUFFIX) + 1; + if (need > PREALLOC_BUFFER_SIZE) { + fname = erts_alloc(ERTS_ALC_T_DDLL_TMP_BUF, need); /* erts_alloc exits on failure */ + } + sys_strncpy(fname, modname, len); + fname[len] = '\0'; + sys_strcat(fname, DRIVER_INIT_SUFFIX); + res = erts_sys_ddll_sym(handle, fname, &func); + if (fname != statbuf) { + erts_free(ERTS_ALC_T_DDLL_TMP_BUF, fname); + } + if ( res != ERL_DE_NO_ERROR) { + return res; + } + *function = func; + return ERL_DE_NO_ERROR; +} + +int erts_sys_ddll_load_nif_init(void *handle, void **function, ErtsSysDdllError* err) +{ + /* NIFs not implemented for vxworks */ + return ERL_DE_ERROR_NO_DDLL_FUNCTIONALITY; +} + +/* + * Call the driver_init function, whatever it's really called, simple on unix... +*/ +void *erts_sys_ddll_call_init(void *function) { + void *(*initfn)(void) = function; + return (*initfn)(); +} +void *erts_sys_ddll_call_nif_init(void *function) { + return erts_sys_ddll_call_init(function); +} + + +/* + * Close a chared object + */ +int erts_sys_ddll_close2(void *handle, ErtsSysDdllError* err) +{ + MODULE_ID mid = (MODULE_ID) handle; + if (unld(mid, 0) < 0) { + return ERL_DE_DYNAMIC_ERROR_OFFSET - ((int) ModuleNotUnloadable); + } + return ERL_DE_NO_ERROR; +} + +/* + * Return string that describes the (current) error + */ +char *erts_sys_ddll_error(int code) +{ + int actual_code; + if (code > ERL_DE_DYNAMIC_ERROR_OFFSET) { + return "Unspecified error"; + } + actual_code = -1*(code - ERL_DE_DYNAMIC_ERROR_OFFSET); + if (actual_code > ((int) UnknownError)) { + actual_code = UnknownError; + } + return errcode_tab[actual_code]; +} + +static FUNCPTR lookup(char *sym) +{ + FUNCPTR entry; + SYM_TYPE type; + + if (symFindByNameAndType(sysSymTbl, sym, (char **)&entry, + &type, N_EXT | N_TEXT, N_EXT | N_TEXT) != OK) { + return NULL ; + } + return entry; +} + +static MODULE_ID get_mid(char* name) +{ + int fd; + MODULE_ID mid = NULL; + + if((fd = open(name, O_RDONLY, 0664)) >= 0) { + mid = loadModule(fd, GLOBAL_SYMBOLS); + close(fd); + } + return mid; +} + +void erts_sys_ddll_free_error(ErtsSysDdllError* err) +{ + /* NYI */ +} + diff --git a/erts/emulator/sys/vxworks/sys.c b/erts/emulator/sys/vxworks/sys.c new file mode 100644 index 0000000000..abddc7e107 --- /dev/null +++ b/erts/emulator/sys/vxworks/sys.c @@ -0,0 +1,2594 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 1997-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ +/* + * system-dependent functions + * + */ +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif +#include <vxWorks.h> +#include <version.h> +#include <string.h> +#include <types.h> +#include <sigLib.h> +#include <ioLib.h> +#include <iosLib.h> +#include <envLib.h> +#include <fioLib.h> +#include <stdlib.h> +#include <stdio.h> +#include <errno.h> +#include <symLib.h> +#include <sysLib.h> +#include <sysSymTbl.h> +#include <loadLib.h> +#include <taskLib.h> +#include <taskVarLib.h> +#include <taskHookLib.h> +#include <tickLib.h> +#include <time.h> +#include <rngLib.h> +#include <semLib.h> +#include <selectLib.h> +#include <sockLib.h> +#include <a_out.h> +#include <wdLib.h> +#include <timers.h> +#include <ctype.h> +#include <sys/stat.h> +#include <sys/socket.h> +#include <netinet/in.h> +#include <netinet/tcp.h> +#include <stdarg.h> + + +#ifndef WANT_NONBLOCKING +#define WANT_NONBLOCKING +#endif + +#include "sys.h" +#include "erl_alloc.h" + +/* don't need global.h, but bif_table.h (included by bif.h) won't compile otherwise */ +#include "global.h" +#include "bif.h" + +#include "erl_sys_driver.h" + +#include "elib_stat.h" + +#include "reclaim_private.h" /* Some more or less private reclaim facilities */ + +#ifndef RETSIGTYPE +#define RETSIGTYPE void +#endif + +EXTERN_FUNCTION(void, erl_start, (int, char**)); +EXTERN_FUNCTION(void, erl_exit, (int n, char*, _DOTS_)); +EXTERN_FUNCTION(void, erl_error, (char*, va_list)); +EXTERN_FUNCTION(int, driver_interrupt, (int, int)); +EXTERN_FUNCTION(void, increment_time, (int)); +EXTERN_FUNCTION(int, next_time, (_VOID_)); +EXTERN_FUNCTION(void, set_reclaim_free_function, (FreeFunction)); +EXTERN_FUNCTION(int, erl_mem_info_get, (MEM_PART_STATS *)); +EXTERN_FUNCTION(void, erl_crash_dump, (char* file, int line, char* fmt, ...)); + +#define ISREG(st) (((st).st_mode&S_IFMT) == S_IFREG) + +/* these are defined in usrLib.c */ +extern int spTaskPriority, spTaskOptions; + +/* forward declarations */ +static FUNCTION(FUNCPTR, lookup, (char*)); +static FUNCTION(int, read_fill, (int, char*, int)); +#if (CPU == SPARC) +static FUNCTION(RETSIGTYPE, fpe_sig_handler, (int)); /*where is this fun? */ +#elif (CPU == PPC603) +static FUNCTION(void, fix_registers, (void)); +#endif +static FUNCTION(void, close_pipes, (int*, int*, int)); +static FUNCTION(void, delete_hook, (void)); +static FUNCTION(void, initialize_allocation, (void)); + +FUNCTION(STATUS, uxPipeDrv, (void)); +FUNCTION(STATUS, pipe, (int*)); +FUNCTION(void, uxPipeShow, (int)); + +void erl_main(int argc, char **argv); +void argcall(char *args); + +/* Malloc-realted functions called from the VxWorks shell */ +EXTERN_FUNCTION(int, erl_set_memory_block, + (int, int, int, int, int, int, int, int, int, int)); +EXTERN_FUNCTION(int, erl_memory_show, + (int, int, int, int, int, int, int, int, int, int)); + +#define DEFAULT_PORT_STACK_SIZE 100000 +static int port_stack_size; + +static int erlang_id = 0; /* Inited at loading, set/reset at each run */ + +/* interval time reported to emulator */ +static int sys_itime; + +/* XXX - This is defined in .../config/all/configAll.h (NUM_FILES), + and not easily accessible at compile or run time - however, + in VxWorks 5.1 it is stored in the (undocumented?) maxFiles variable; + probably shouldn't depend on it, but we try to pick it up... */ +static int max_files = 50; /* default configAll.h */ + +int erts_vxworks_max_files; + +/* + * used by the break handler (set by signal handler on ctl-c) + */ +volatile int erts_break_requested; + +/********************* General functions ****************************/ + +Uint +erts_sys_misc_mem_sz(void) +{ + Uint res = erts_check_io_size(); + /* res += FIXME */ + return res; +} + +/* + * XXX This declaration should not be here. + */ +void erl_sys_schedule_loop(void); + +#ifdef SOFTDEBUG +static void do_trace(int line, char *file, char *format, ...) +{ + va_list va; + int tid = taskIdSelf(); + char buff[512]; + + va_start(va, format); + sprintf(buff,"Trace: Task: 0x%08x, %s:%d - ", + tid, file, line); + vsprintf(buff + strlen(buff), format, va); + va_end(va); + strcat(buff,"\r\n"); + write(2,buff,strlen(buff)); +} + +#define TRACE() do_trace(__LINE__, __FILE__,"") +#define TRACEF(Args...) do_trace(__LINE__,__FILE__, ## Args) +#endif + +void +erts_sys_pre_init(void) +{ + if (erlang_id != 0) { + /* NOTE: This particular case must *not* call erl_exit() */ + erts_fprintf(stderr, "Sorry, erlang is already running (as task %d)\n", + erlang_id); + exit(1); + } + + /* This must be done as early as possible... */ + if(!reclaim_init()) + fprintf(stderr, "Warning : reclaim facility should be initiated before " + "erlang is started!\n"); + erts_vxworks_max_files = max_files = reclaim_max_files(); + + /* Floating point exceptions */ +#if (CPU == SPARC) + sys_sigset(SIGFPE, fpe_sig_handler); +#elif (CPU == PPC603) + fix_registers(); +#endif + + /* register the private delete hook in reclaim */ + save_delete_hook((FUNCPTR)delete_hook, (caddr_t)0); + erlang_id = taskIdSelf(); +#ifdef DEBUG + printf("emulator task id = 0x%x\n", erlang_id); +#endif +} + +void erts_sys_alloc_init(void) +{ + initialize_allocation(); +} + +void +erl_sys_init(void) +{ + setvbuf(stdout, (char *)NULL, _IOLBF, BUFSIZ); + /* XXX Bug in VxWorks stdio loses fputch()'ed output after the + setvbuf() but before a *printf(), and possibly worse (malloc + errors, crash?) - so let's give it a *printf().... */ + fprintf(stdout, "%s",""); +} + +void +erl_sys_args(int* argc, char** argv) +{ + erts_init_check_io(); + max_files = erts_check_io_max_files(); + ASSERT(max_files <= erts_vxworks_max_files); +} + +/* + * Called from schedule() when it runs out of runnable processes, + * or when Erlang code has performed INPUT_REDUCTIONS reduction + * steps. runnable == 0 iff there are no runnable Erlang processes. + */ +void +erl_sys_schedule(int runnable) +{ + erts_check_io_interrupt(0); + erts_check_io(!runnable); +} + +void erts_do_break_handling(void) +{ + SET_BLOCKING(0); + /* call the break handling function, reset the flag */ + do_break(); + erts_break_requested = 0; + SET_NONBLOCKING(0); +} + +/* signal handling */ +RETSIGTYPE (*sys_sigset(sig, func))() + int sig; + RETSIGTYPE (*func)(); +{ + struct sigaction act, oact; + + sigemptyset(&act.sa_mask); + act.sa_flags = 0; + act.sa_handler = func; + sigaction(sig, &act, &oact); + return(oact.sa_handler); +} + +void sys_sigblock(int sig) +{ + sigset_t mask; + + sigemptyset(&mask); + sigaddset(&mask, sig); + sigprocmask(SIG_BLOCK, &mask, (sigset_t *)NULL); +} + +void sys_sigrelease(int sig) +{ + sigset_t mask; + + sigemptyset(&mask); + sigaddset(&mask, sig); + sigprocmask(SIG_UNBLOCK, &mask, (sigset_t *)NULL); +} + +void +erts_sys_prepare_crash_dump(void) +{ + +} + +/* register signal handlers XXX - they don't work, need to find out why... */ +/* set up signal handlers for break and quit */ +static void request_break(void) +{ + /* just set a flag - checked for and handled + * in main thread (not signal handler). + * see check_io() + */ +#ifdef DEBUG + fprintf(stderr,"break!\n"); +#endif + erts_break_requested = 1; + erts_check_io_interrupt(1); /* Make sure we don't sleep in erts_poll_wait */ +} + +static void do_quit(void) +{ + halt_0(0); +} + +void erts_set_ignore_break(void) { +} + +void init_break_handler(void) +{ + sys_sigset(SIGINT, request_break); + sys_sigset(SIGQUIT, do_quit); +} + +void erts_replace_intr(void) { +} + +int sys_max_files(void) +{ + return(max_files); +} + +/******************* Routines for time measurement *********************/ + +int sys_init_time(void) +{ + erts_clock_rate = sysClkRateGet(); + /* + ** One could imagine that it would be better returning + ** a resolution more near the clock rate, like in: + ** return 1000 / erts_clock_rate; + ** but tests show that such isn't the case (rounding errors?) + ** Well, we go for the Unix variant of returning 1 + ** as a constant virtual clock rate. + */ + return SYS_CLOCK_RESOLUTION; +} + +int erts_clock_rate; +static volatile int ticks_inuse; +static volatile unsigned long ticks_collected; /* will wrap */ +static WDOG_ID watchdog_id; +static ULONG user_time; +static int this_task_id, sys_itime; +static SysHrTime hrtime_wrap; +static unsigned long last_tick_count; + +static void tolerant_time_clockint(int count) +{ + if (watchdog_id != NULL) { + if (taskIsReady(this_task_id)) + user_time += 1; + ++count; + if (!ticks_inuse) { + ticks_collected += count; + count = 0; + } + wdStart(watchdog_id, 1, (FUNCPTR)tolerant_time_clockint, count); + } +} + +int sys_init_hrtime(void) +{ + this_task_id = taskIdSelf(); /* OK, this only works for one single task + in the system... */ + user_time = 0; + + ticks_inuse = 0; + ticks_collected = 0; + hrtime_wrap = 0; + last_tick_count = 0; + + sys_itime = 1000 / erts_clock_rate; + watchdog_id = wdCreate(); + wdStart(watchdog_id, 1, (FUNCPTR) tolerant_time_clockint, 0); + return 0; +} + +SysHrTime sys_gethrtime(void) +{ + SysHrTime ticks; + + ++ticks_inuse; + ticks = (SysHrTime) (ticks_collected & 0x7FFFFFFF); + ticks_inuse = 0; + if (ticks < (SysHrTime) last_tick_count) { + hrtime_wrap += 1UL << 31; + } + last_tick_count = ticks; + return (ticks + hrtime_wrap) * ((SysHrTime) (1000000000UL / + erts_clock_rate)); +} + +void sys_gettimeofday(SysTimeval *tvp) +{ + struct timespec now; + + clock_gettime(CLOCK_REALTIME, &now); + tvp->tv_sec = now.tv_sec; + tvp->tv_usec = now.tv_nsec / 1000; +} + +clock_t sys_times(SysTimes *t) +{ + t->tms_stime = t->tms_cutime = t->tms_cstime = 0; + ++ticks_inuse; + t->tms_utime = user_time; + ticks_inuse = 0; + return tickGet(); /* The best we can do... */ +} + +/* This is called when *this task* is deleted */ +static void delete_hook(void) +{ + if (watchdog_id != NULL) { + wdDelete(watchdog_id); + watchdog_id = NULL; + } + erlang_id = 0; + this_task_id = 0; +} + +/************************** OS info *******************************/ + +/* Used by erlang:info/1. */ +/* (This code was formerly in drv.XXX/XXX_os_drv.c) */ + +#define MAX_VER_STR 9 /* Number of characters to + consider in version string */ + +static FUNCTION(int, get_number, (char** str_ptr)); + +char os_type[] = "vxworks"; + +static int +get_number(char **str_ptr) +{ + char* s = *str_ptr; /* Pointer to beginning of string. */ + char* dot; /* Pointer to dot in string or NULL. */ + + if (!isdigit(*s)) + return 0; + if ((dot = strchr(s, '.')) == NULL) { + *str_ptr = s+strlen(s); + return atoi(s); + } else { + *dot = '\0'; + *str_ptr = dot+1; + return atoi(s); + } +} + +/* namebuf; Where to return the name. */ +/* size; Size of name buffer. */ +void +os_flavor(char *namebuf, unsigned size) +{ + strcpy(namebuf, "-"); +} + +/* int* pMajor; Pointer to major version. */ +/* int* pMinor; Pointer to minor version. */ +/* int* pBuild; Pointer to build number. */ +void +os_version(int *pMajor, int *pMinor, int *pBuild) +{ + char os_ver[MAX_VER_STR+2]; + char* release; /* Pointer to the release string: + * X.Y or X.Y.Z. + */ + strncpy(os_ver, vxWorksVersion, MAX_VER_STR); + release = os_ver; + *pMajor = get_number(&release); + *pMinor = get_number(&release); + *pBuild = get_number(&release); +} + +void init_getenv_state(GETENV_STATE *state) +{ + *state = NULL; +} + +char *getenv_string(GETENV_STATE *state0) +{ + return NULL; +} + +void fini_getenv_state(GETENV_STATE *state) +{ + *state = NULL; +} + +/************************** Port I/O *******************************/ + + +/* I. Common stuff */ + +#define TMP_BUF_MAX (tmp_buf_size - 1024) +static byte *tmp_buf; +static Uint tmp_buf_size; + +/* II. The spawn/fd/vanilla drivers */ + +/* This data is shared by these drivers - initialized by spawn_init() */ +static struct driver_data { + int port_num, ofd, packet_bytes, report_exit; + int exitcode, exit_reported; /* For returning of exit codes. */ +} *driver_data; /* indexed by fd */ + +/* + * Locking only for exitcodes and exit_reported, one global sem for all + * spawn ports as this is rare. + */ +static SEM_ID driver_data_sem = NULL; +/* + * Also locking when looking up entries in the load table + */ +static SEM_ID entry_data_sem = NULL; + +/* We maintain a linked fifo queue of these structs in order */ +/* to manage unfinnished reads/and writes on differenet fd's */ + +typedef struct pend { + char *cpos; + int fd; + int remain; + struct pend *next; + char buf[1]; /* this is a trick to be able to malloc one chunk */ +} Pend; + +static struct fd_data { + int inport, outport; + char *buf, *cpos; + int sz, remain; /* for input on fd */ + Pend* pending; /* pending outputs */ + +} *fd_data; /* indexed by fd */ + + +/* Driver interfaces */ +static ErlDrvData spawn_start(ErlDrvPort port_num, char *name, SysDriverOpts* opts); +static ErlDrvData fd_start(ErlDrvPort port_num, char *name, SysDriverOpts* opts); +static ErlDrvData vanilla_start(ErlDrvPort port_num, char *name, SysDriverOpts* opts); +static int spawn_init(void); +static void fd_stop(ErlDrvData); +static void stop(ErlDrvData); +static void ready_input(ErlDrvData fd, ErlDrvEvent ready_fd); +static void ready_output(ErlDrvData fd, ErlDrvEvent ready_fd); +static void output(ErlDrvData fd, char *buf, int len); +static void stop_select(ErlDrvEvent, void*); + +struct erl_drv_entry spawn_driver_entry = { + spawn_init, + spawn_start, + stop, + output, + ready_input, + ready_output, + "spawn", + NULL, /* finish */ + NULL, /* handle */ + NULL, /* control */ + NULL, /* timeout */ + NULL, /* outputv */ + NULL, /* ready_async */ + NULL, /* flush */ + NULL, /* call */ + NULL, /* event */ + ERL_DRV_EXTENDED_MARKER, + ERL_DRV_EXTENDED_MAJOR_VERSION, + ERL_DRV_EXTENDED_MINOR_VERSION, + 0, /* ERL_DRV_FLAGs */ + NULL, /* handle2 */ + NULL, /* process_exit */ + stop_select + +}; +struct erl_drv_entry fd_driver_entry = { + NULL, + fd_start, + fd_stop, + output, + ready_input, + ready_output, + "fd", + NULL, /* finish */ + NULL, /* handle */ + NULL, /* control */ + NULL, /* timeout */ + NULL, /* outputv */ + NULL, /* ready_async */ + NULL, /* flush */ + NULL, /* call */ + NULL, /* event */ + ERL_DRV_EXTENDED_MARKER, + ERL_DRV_EXTENDED_MAJOR_VERSION, + ERL_DRV_EXTENDED_MINOR_VERSION, + 0, /* ERL_DRV_FLAGs */ + NULL, /* handle2 */ + NULL, /* process_exit */ + stop_select +}; +struct erl_drv_entry vanilla_driver_entry = { + NULL, + vanilla_start, + stop, + output, + ready_input, + ready_output, + "vanilla", + NULL, /* finish */ + NULL, /* handle */ + NULL, /* control */ + NULL, /* timeout */ + NULL, /* outputv */ + NULL, /* ready_async */ + NULL, /* flush */ + NULL, /* call */ + NULL, /* event */ + ERL_DRV_EXTENDED_MARKER, + ERL_DRV_EXTENDED_MAJOR_VERSION, + ERL_DRV_EXTENDED_MINOR_VERSION, + 0, /* ERL_DRV_FLAGs */ + NULL, /* handle2 */ + NULL, /* process_exit */ + stop_select +}; + +/* +** Set up enough of the driver_data structure to be able to report exit status. +** Some things may be initiated again, but that is no real problem. +*/ +static int pre_set_driver_data(int ifd, int ofd, + int read_write, int report_exit) { + if (read_write & DO_READ) { + driver_data[ifd].report_exit = report_exit; + driver_data[ifd].exitcode = 0; + driver_data[ifd].exit_reported = 0; + if (read_write & DO_WRITE) { + driver_data[ifd].ofd = ofd; + if (ifd != ofd) { + driver_data[ofd] = driver_data[ifd]; + driver_data[ofd].report_exit = 0; + } + } else { /* DO_READ only */ + driver_data[ifd].ofd = -1; + } + return(ifd); + } else { /* DO_WRITE only */ + driver_data[ofd].report_exit = 0; + driver_data[ofd].exitcode = 0; + driver_data[ofd].exit_reported = 0; + driver_data[ofd].ofd = ofd; + return(ofd); + } +} + +/* +** Set up the driver_data structure, it may have been initiated +** partly by the function above, but we dont care. +*/ +static int set_driver_data(int port_num, int ifd, int ofd, + int packet_bytes, int read_write, + int report_exit) +{ + if (read_write & DO_READ) { + driver_data[ifd].packet_bytes = packet_bytes; + driver_data[ifd].port_num = port_num; + driver_data[ifd].report_exit = report_exit; + if (read_write & DO_WRITE) { + driver_data[ifd].ofd = ofd; + if (ifd != ofd) { + driver_data[ofd] = driver_data[ifd]; + driver_data[ofd].report_exit = 0; + } + } else { /* DO_READ only */ + driver_data[ifd].ofd = -1; + } + (void) driver_select(port_num, ifd, ERL_DRV_READ|ERL_DRV_USE, 1); + return(ifd); + } else { /* DO_WRITE only */ + driver_data[ofd].packet_bytes = packet_bytes; + driver_data[ofd].port_num = port_num; + driver_data[ofd].report_exit = 0; + driver_data[ofd].ofd = ofd; + return(ofd); + } +} + +static int need_new_sems = 1; + +static int spawn_init(void) +{ + char *stackenv; + int size; + driver_data = (struct driver_data *) + erts_alloc(ERTS_ALC_T_DRV_TAB, max_files * sizeof(struct driver_data)); + if (need_new_sems) { + driver_data_sem = semMCreate + (SEM_Q_PRIORITY | SEM_DELETE_SAFE | SEM_INVERSION_SAFE); + entry_data_sem = semMCreate + (SEM_Q_PRIORITY | SEM_DELETE_SAFE | SEM_INVERSION_SAFE); + } + if (driver_data_sem == NULL || entry_data_sem == NULL) { + erl_exit(1,"Could not allocate driver locking semaphore."); + } + need_new_sems = 0; + + (void)uxPipeDrv(); /* Install pipe driver */ + + if ((stackenv = getenv("ERLPORTSTACKSIZE")) != NULL && + (size = atoi(stackenv)) > 0) + port_stack_size = size; + else + port_stack_size = DEFAULT_PORT_STACK_SIZE; + return 0; +} + +/* Argv has to be built vith the save_xxx routines, not with whathever + sys_xxx2 has in mind... */ +#define argv_alloc save_malloc +#define argv_realloc save_realloc +#define argv_free save_free +/* Build argv, return argc or -1 on failure */ +static int build_argv(char *name, char ***argvp) +{ + int argvsize = 10, argc = 0; + char *args, *arglast = NULL, *argp; + char **argv; + +#ifdef DEBUG + fdprintf(2, "Building argv, %s =>\n", name); +#endif + if ((argv = (char **)argv_alloc(argvsize * sizeof(char *))) == NULL) + return(-1); + if ((args = argv_alloc(strlen(name) + 1)) == NULL) + return(-1); + strcpy(args, name); + argp = strtok_r(args, " \t", &arglast); + while (argp != NULL) { + if (argc + 1 >= argvsize) { + argvsize += 10; + argv = (char **)argv_realloc((char *)argv, argvsize*sizeof(char *)); + if (argv == NULL) { + argv_free(args); + return(-1); + } + } +#ifdef DEBUG + fdprintf(2, "%s\n", argp); +#endif + argv[argc++] = argp; + argp = strtok_r((char *)NULL, " \t", &arglast); + } + argv[argc] = NULL; + *argvp = argv; + return(argc); +} +#undef argv_alloc +#undef argv_realloc +#undef argv_free + + +/* Lookup and return global text symbol or NULL on failure + Symbol name is null-terminated and without the leading '_' */ +static FUNCPTR +lookup(char *sym) +{ + char buf[256]; + char *symname = buf; + int len; + FUNCPTR entry; + SYM_TYPE type; + + len = strlen(sym); + if (len > 254 && (symname = malloc(len+2)) == NULL) + return(NULL); +#if defined _ARCH_PPC || defined SIMSPARCSOLARIS + /* GCC for PPC and SIMSPARC doesn't add a leading _ to symbols */ + strcpy(symname, sym); +#else + sprintf(symname, "_%s", sym); +#endif + if (symFindByNameAndType(sysSymTbl, symname, (char **)&entry, + &type, N_EXT | N_TEXT, N_EXT | N_TEXT) != OK) + entry = NULL; + if (symname != buf) + free(symname); + return(entry); +} + +/* This function is spawned to build argc, argv, lookup the symbol to call, + connect and set up file descriptors, and make the actual call. + N.B. 'name' was allocated by the Erlang task (through plain_malloc) and + is freed by this port program task. + Note: 'name' may be a path containing '/'. */ + +static void call_proc(char *name, int ifd, int ofd, int read_write, + int redir_stderr, int driver_index, + int p6, int p7, int p8, int p9) +{ + int argc; + char **argv, *bname; + FUNCPTR entry; + int ret = -1; + + /* Must consume 'name' */ + argc = build_argv(name, &argv); + plain_free(name); + /* Find basename of path */ + if ((bname = strrchr(argv[0], '/')) != NULL) { + bname++; + } else { + bname = argv[0]; + } +#ifdef DEBUG + fdprintf(2, "Port program name: %s\n", bname); +#endif + semTake(entry_data_sem, WAIT_FOREVER); + + if (argc > 0) { + if ((entry = lookup(bname)) == NULL) { + int fd; + char *fn; + /* NOTE: We don't check the return value of loadModule, + since that was incompatibly changed from 5.0.2b to 5.1, + but rather do a repeated lookup(). */ + if ((fd = open(argv[0], O_RDONLY)) > 0) { + (void) loadModule(fd, GLOBAL_SYMBOLS); + close(fd); + entry = lookup(bname); + } + if (entry == NULL) { + /* filename == func failed, try func.o */ + if ((fn = malloc(strlen(argv[0]) + 3)) != NULL) { /* ".o\0" */ + strcpy(fn, argv[0]); + strcat(fn, ".o"); + if ((fd = open(fn, O_RDONLY)) > 0) { + (void) loadModule(fd, GLOBAL_SYMBOLS); + close(fd); + entry = lookup(bname); + } + free(fn); + } + } + } + } else { + entry = NULL; + } + semGive(entry_data_sem); + + if (read_write & DO_READ) { /* emulator read */ + save_fd(ofd); + ioTaskStdSet(0, 1, ofd); /* stdout for process */ + if(redir_stderr) + ioTaskStdSet(0, 2, ofd);/* stderr for process */ + } + if (read_write & DO_WRITE) { /* emulator write */ + save_fd(ifd); + ioTaskStdSet(0, 0, ifd); /* stdin for process */ + } + if (entry != NULL) { + ret = (*entry)(argc, argv, (char **)NULL); /* NULL for envp */ + } else { + fdprintf(2, "Could not exec \"%s\"\n", argv[0]); + ret = -1; + } + if (driver_data[driver_index].report_exit) { + semTake(driver_data_sem, WAIT_FOREVER); + driver_data[driver_index].exitcode = ret; + driver_data[driver_index].exit_reported = 1; + semGive(driver_data_sem); + } + /* We *don't* want to close the pipes here, but let the delete + hook take care of it - it might want to flush stdout and there'd + better be an open descriptor to flush to... */ + exit(ret); +} + +static void close_pipes(int ifd[2], int ofd[2], int read_write) +{ + if (read_write & DO_READ) { + (void) close(ifd[0]); + (void) close(ifd[1]); + } + if (read_write & DO_WRITE) { + (void) close(ofd[0]); + (void) close(ofd[1]); + } +} + +static void init_fd_data(int fd, int port_unused_argument) +{ + SET_NONBLOCKING(fd); + fd_data[fd].pending = NULL; + fd_data[fd].buf = fd_data[fd].cpos = NULL; + fd_data[fd].remain = fd_data[fd].sz = 0; +} + +static ErlDrvData spawn_start(ErlDrvPort port_num, char *name,SysDriverOpts* opts) +{ + int ifd[2], ofd[2], len, nl, id; + char taskname[11], *progname, *bname; + char *space_in_command; + int packet_bytes = opts->packet_bytes; + int read_write = opts->read_write; + int use_stdio = opts->use_stdio; + int redir_stderr = opts->redir_stderr; + int driver_index; + + if (!use_stdio){ + return (ErlDrvData) -3; + } + + /* Create pipes and set the Erlang task as owner of its + * read and write ends (through save_fd()). + */ + switch (read_write) { + case DO_READ: + if (pipe(ifd) < 0){ + return (ErlDrvData) -2; + } + if (ifd[0] >= max_files) { + close_pipes(ifd, ofd, read_write); + errno = ENFILE; + return (ErlDrvData) -2; + } + save_fd(ifd[0]); + break; + case DO_WRITE: + if (pipe(ofd) < 0) { + return (ErlDrvData) -2; + } + if (ofd[1] >= max_files) { + close_pipes(ifd, ofd, read_write); + errno = ENFILE; + return (ErlDrvData) -2; + } + save_fd(ofd[1]); + break; + case DO_READ|DO_WRITE: + if (pipe(ifd) < 0){ + return (ErlDrvData) -2; + } + if (ifd[0] >= max_files || pipe(ofd) < 0) { + close_pipes(ifd, ofd, DO_READ); + errno = ENFILE; + return (ErlDrvData) -2; + } + if (ofd[1] >= max_files) { + close_pipes(ifd, ofd, read_write); + errno = ENFILE; + return (ErlDrvData) -2; + } + save_fd(ifd[0]); + save_fd(ofd[1]); + break; + default: + return (ErlDrvData) -1; + } + + /* Allocate space for program name to be freed by the + * spawned task. We use plain_malloc so that the allocated + * space is not owned by the Erlang task. + */ + + if ((progname = plain_malloc(strlen(name) + 1)) == NULL) { + close_pipes(ifd, ofd, read_write); + errno = ENOMEM; + return (ErlDrvData) -2; + } + strcpy(progname, name); + + /* Check if name contains a space + * (e.g "port_test -o/home/gandalf/tornado/wind/target/erlang") + */ + if ((space_in_command = strrchr(progname, ' ')) != NULL) { + *space_in_command = '\0'; + } + + /* resulting in "port_test" */ + if ((bname = strrchr(progname, '/')) != NULL) + bname++; + else + bname = progname; + + /* resulting in "port_test" */ + len = strlen(bname); + nl = len > 10 ? 10 : len; + strncpy(taskname, bname, nl); + taskname[nl] = '\0'; + if (space_in_command != NULL) + *space_in_command = ' '; + driver_index = pre_set_driver_data(ifd[0], ofd[1], + read_write, opts->exit_status); + + /* resetting to "port_test -o/home/gandalf/tornado/wind/target/erlang" */ + if ((id = taskSpawn(taskname, spTaskPriority, spTaskOptions, + port_stack_size, (FUNCPTR)call_proc, (int)progname, + ofd[0], ifd[1], read_write, redir_stderr, driver_index, + 0,0,0,0)) + == ERROR) { + close_pipes(ifd, ofd, read_write); + plain_free(progname); /* only when spawn fails */ + errno = ENOMEM; + return (ErlDrvData) -2; + } +#ifdef DEBUG + fdprintf(2, "Spawned %s as %s[0x%x]\n", name, taskname, id); +#endif + if (read_write & DO_READ) + init_fd_data(ifd[0], port_num); + if (read_write & DO_WRITE) + init_fd_data(ofd[1], port_num); + return (ErlDrvData) (set_driver_data(port_num, ifd[0], ofd[1], + packet_bytes,read_write, + opts->exit_status)); +} + +static ErlDrvData fd_start(ErlDrvPort port_num, char *name, SysDriverOpts* opts) +{ + if (((opts->read_write & DO_READ) && opts->ifd >= max_files) || + ((opts->read_write & DO_WRITE) && opts->ofd >= max_files)) { + return (ErlDrvData) -1; + } + + if (opts->read_write & DO_READ) + init_fd_data(opts->ifd, port_num); + if (opts->read_write & DO_WRITE) + init_fd_data(opts->ofd, port_num); + return (ErlDrvData) (set_driver_data(port_num, opts->ifd, opts->ofd, + opts->packet_bytes, opts->read_write, 0)); +} + +static void clear_fd_data(int fd) +{ + + if (fd_data[fd].sz > 0) + erts_free(ERTS_ALC_T_FD_ENTRY_BUF, (void *) fd_data[fd].buf); + fd_data[fd].buf = NULL; + fd_data[fd].sz = 0; + fd_data[fd].remain = 0; + fd_data[fd].cpos = NULL; +} + +static void nbio_stop_fd(int port_num, int fd) +{ + Pend *p, *p1; + + driver_select(port_num, fd, ERL_DRV_READ|ERL_DRV_WRITE, 0); + clear_fd_data(fd); + p = fd_data[fd].pending; + SET_BLOCKING(fd); + while (p) { + p1 = p->next; + free(p); + p = p1; + } + fd_data[fd].pending = NULL; +} + +static void fd_stop(ErlDrvData drv_data) +{ + int ofd; + int fd = (int) drv_data; + + nbio_stop_fd(driver_data[fd].port_num, (int)fd); + ofd = driver_data[fd].ofd; + if (ofd != fd && ofd != -1) + nbio_stop_fd(driver_data[fd].port_num, (int)ofd); /* XXX fd = ofd? */ +} + +static ErlDrvData +vanilla_start(ErlDrvPort port_num, char *name, SysDriverOpts* opts) +{ + int flags, fd; + struct stat statbuf; + + DEBUGF(("vanilla_start, name: %s [r=%1i w=%1i]\n", name, + opts->read_write & DO_READ, + opts->read_write & DO_WRITE)); + + flags = (opts->read_write == DO_READ ? O_RDONLY : + opts->read_write == DO_WRITE ? O_WRONLY|O_CREAT|O_TRUNC : + O_RDWR|O_CREAT); + if ((fd = open(name, flags, 0666)) < 0){ + errno = ENFILE; + return (ErlDrvData) -2; + } + if (fd >= max_files) { + close(fd); + errno = ENFILE; + return (ErlDrvData) -2; + } + if (fstat(fd, &statbuf) < 0) { + close(fd); + errno = ENFILE; + return (ErlDrvData) -2; + } + + /* Return error for reading regular files (doesn't work) */ + if (ISREG(statbuf) && ((opts->read_write) & DO_READ)) { + close(fd); + return (ErlDrvData) -3; + } + init_fd_data(fd, port_num); + return (ErlDrvData) (set_driver_data(port_num, fd, fd, + opts->packet_bytes, opts->read_write, 0)); +} + +/* Note that driver_data[fd].ifd == fd if the port was opened for reading, */ +/* otherwise (i.e. write only) driver_data[fd].ofd = fd. */ + +static void stop(ErlDrvData drv_data) +{ + int port_num, ofd; + int fd = (int) drv_data; + + port_num = driver_data[fd].port_num; + nbio_stop_fd(port_num, fd); + driver_select(port_num, fd, ERL_DRV_USE, 0); /* close(fd) */ + + ofd = driver_data[fd].ofd; + if (ofd != fd && ofd != -1) { + nbio_stop_fd(port_num, ofd); + driver_select(port_num, ofd, ERL_DRV_USE, 0); /* close(fd) */ + } +} + +static int sched_write(int port_num,int fd, char *buf, int len, int pb) +{ + Pend *p, *p2, *p3; + int p_bytes = len; + + p = (Pend*) erts_alloc_fnf(ERTS_ALC_T_PEND_DATA, pb + len + sizeof(Pend)); + if (!p) { + driver_failure(port_num, -1); + return(-1); + } + + switch(pb) { + case 4: put_int32(len, p->buf); break; + case 2: put_int16(len, p->buf); break; + case 1: put_int8(len, p->buf); break; + case 0: break; /* Handles this case too */ + } + sys_memcpy(p->buf + pb, buf, len); + driver_select(port_num, fd, ERL_DRV_WRITE|ERL_DRV_USE, 1); + p->cpos = p->buf; + p->fd = fd; + p->next = NULL; + p->remain = len + pb; + p2 = fd_data[fd].pending; + if (p2 == NULL) + fd_data[fd].pending = p; + else { + p3 = p2->next; + while(p3) { + p_bytes += p2->remain; + p2 = p2->next; + p3 = p3->next; + } + p2->next = p; + } + if (p_bytes > (1 << 13)) /* More than 8 k pending */ + set_busy_port(port_num, 1); + return(0); +} + +/* Fd is the value returned as drv_data by the start func */ +static void output(ErlDrvData drv_data, char *buf, int len) +{ + int buf_done, port_num, wval, pb, ofd; + byte lb[4]; + struct iovec iv[2]; + int fd = (int) drv_data; + + pb = driver_data[fd].packet_bytes; + port_num = driver_data[fd].port_num; + + if ((ofd = driver_data[fd].ofd) == -1) { + return; + } + + if (fd_data[ofd].pending) { + sched_write(port_num, ofd, buf, len, pb); + return; + } + + if ((pb == 2 && len > 65535) || (pb == 1 && len > 255)) { + driver_failure_posix(port_num, EINVAL); + return; + } + if (pb == 0) { + wval = write(ofd, buf, len); + } else { + lb[0] = (len >> 24) & 255; /* MSB */ + lb[1] = (len >> 16) & 255; + lb[2] = (len >> 8) & 255; + lb[3] = len & 255; /* LSB */ + iv[0].iov_base = (char*) lb + (4 - pb); + iv[0].iov_len = pb; + iv[1].iov_base = buf; + iv[1].iov_len = len; + wval = writev(ofd, iv, 2); + } + if (wval == pb + len ) { + return; + } + if (wval < 0) { + if ((errno == EINTR) || (errno == ERRNO_BLOCK)) { + if (pb) { + sched_write(port_num, ofd, buf ,len, pb); + } else if (pb == 0) { + sched_write(port_num, ofd, buf ,len, 0); + } + return; + } + driver_failure_posix(driver_data[fd].port_num, EINVAL); + return; + } + if (wval < pb) { + sched_write(port_num, ofd, (lb +4 -pb) + wval, pb-wval, 0); + sched_write(port_num, ofd, buf ,len, 0); + return; + } + + /* we now know that wval < (pb + len) */ + buf_done = wval - pb; + sched_write(port_num, ofd, buf + buf_done, len - buf_done,0); +} + +static void stop_select(ErlDrvEvent fd, void* _) +{ + close((int)fd); +} + +static int ensure_header(int fd,char *buf,int packet_size, int sofar) +{ + int res = 0; + int remaining = packet_size - sofar; + + SET_BLOCKING(fd); + if (read_fill(fd, buf+sofar, remaining) != remaining) + return -1; + switch (packet_size) { + case 1: res = get_int8(buf); break; + case 2: res = get_int16(buf); break; + case 4: res = get_int32(buf); break; + } + SET_NONBLOCKING(fd); + return(res); +} + +static int port_inp_failure(int port_num, int ready_fd, int res) +{ + (void) driver_select(port_num, ready_fd, ERL_DRV_READ|ERL_DRV_WRITE, 0); + clear_fd_data(ready_fd); + if (res == 0) { + if (driver_data[ready_fd].report_exit) { + int tmpexit = 0; + int reported; + /* Lock the driver_data structure */ + semTake(driver_data_sem, WAIT_FOREVER); + if ((reported = driver_data[ready_fd].exit_reported)) + tmpexit = driver_data[ready_fd].exitcode; + semGive(driver_data_sem); + if (reported) { + erts_fprintf(stderr,"Exitcode %d reported\r\n", tmpexit); + driver_report_exit(port_num, tmpexit); + } + } + driver_failure_eof(port_num); + } else { + driver_failure(port_num, res); + } + return 0; +} + +/* fd is the drv_data that is returned from the */ +/* initial start routine */ +/* ready_fd is the descriptor that is ready to read */ + +static void ready_input(ErlDrvData drv_data, ErlDrvEvent drv_event) +{ + int port_num, packet_bytes, res; + Uint h = 0; + char *buf; + int fd = (int) drv_data; + int ready_fd = (int) drv_event; + + port_num = driver_data[fd].port_num; + packet_bytes = driver_data[fd].packet_bytes; + + if (packet_bytes == 0) { + if ((res = read(ready_fd, tmp_buf, tmp_buf_size)) > 0) { + driver_output(port_num, (char*)tmp_buf, res); + return; + } + port_inp_failure(port_num, ready_fd, res); + return; + } + + if (fd_data[ready_fd].remain > 0) { /* We try to read the remainder */ + /* space is allocated in buf */ + res = read(ready_fd, fd_data[ready_fd].cpos, + fd_data[ready_fd].remain); + if (res < 0) { + if ((errno == EINTR) || (errno == ERRNO_BLOCK)) { + ; + } else { + port_inp_failure(port_num, ready_fd, res); + } + } else if (res == 0) { + port_inp_failure(port_num, ready_fd, res); + } else if (res == fd_data[ready_fd].remain) { /* we're done */ + driver_output(port_num, fd_data[ready_fd].buf, + fd_data[ready_fd].sz); + clear_fd_data(ready_fd); + } else { /* if (res < fd_data[ready_fd].remain) */ + fd_data[ready_fd].cpos += res; + fd_data[ready_fd].remain -= res; + } + return; + } + + + if (fd_data[ready_fd].remain == 0) { /* clean fd */ + /* We make one read attempt and see what happens */ + res = read(ready_fd, tmp_buf, tmp_buf_size); + if (res < 0) { + if ((errno == EINTR) || (errno == ERRNO_BLOCK)) + return; + port_inp_failure(port_num, ready_fd, res); + return; + } + else if (res == 0) { /* eof */ + port_inp_failure(port_num, ready_fd, res); + return; + } + else if (res < packet_bytes) { /* Ugly case... get at least */ + if ((h = ensure_header(ready_fd, tmp_buf, packet_bytes, res))==-1) { + port_inp_failure(port_num, ready_fd, -1); + return; + } + buf = erts_alloc_fnf(ERTS_ALC_T_FD_ENTRY_BUF, h); + if (!buf) { + port_inp_failure(port_num, ready_fd, -1); + return; + } + fd_data[ready_fd].buf = buf; + fd_data[ready_fd].sz = h; + fd_data[ready_fd].remain = h; + fd_data[ready_fd].cpos = buf; + return; + } + else { /* if (res >= packet_bytes) */ + unsigned char* cpos = tmp_buf; + int bytes_left = res; + while (1) { /* driver_output as many as possible */ + if (bytes_left == 0) { + clear_fd_data(ready_fd); + return; + } + if (bytes_left < packet_bytes) { /* Yet an ugly case */ + if((h=ensure_header(ready_fd, cpos, + packet_bytes, bytes_left))==-1) { + port_inp_failure(port_num, ready_fd, -1); + return; + } + buf = erts_alloc_fnf(ERTS_ALC_T_FD_ENTRY_BUF, h); + if (!buf) + port_inp_failure(port_num, ready_fd, -1); + fd_data[ready_fd].buf = buf; + fd_data[ready_fd].sz = h; + fd_data[ready_fd].remain = h; + fd_data[ready_fd].cpos = buf; + return; + } + switch (packet_bytes) { + case 1: h = get_int8(cpos); cpos += 1; break; + case 2: h = get_int16(cpos); cpos += 2; break; + case 4: h = get_int32(cpos); cpos += 4; break; + } + bytes_left -= packet_bytes; + /* we've got the header, now check if we've got the data */ + if (h <= (bytes_left)) { + driver_output(port_num, (char*) cpos, h); + cpos += h; + bytes_left -= h; + continue; + } + else { /* The last message we got was split */ + buf = erts_alloc_fnf(ERTS_ALC_T_FD_ENTRY_BUF, h); + if (!buf) { + port_inp_failure(port_num, ready_fd, -1); + } + sys_memcpy(buf, cpos, bytes_left); + fd_data[ready_fd].buf = buf; + fd_data[ready_fd].sz = h; + fd_data[ready_fd].remain = h - bytes_left; + fd_data[ready_fd].cpos = buf + bytes_left; + return; + } + } + return; + } + } + fprintf(stderr, "remain %d \n", fd_data[ready_fd].remain); + port_inp_failure(port_num, ready_fd, -1); +} + + +/* fd is the drv_data that is returned from the */ +/* initial start routine */ +/* ready_fd is the descriptor that is ready to read */ + +static void ready_output(ErlDrvData drv_data, ErlDrvEvent drv_event) +{ + Pend *p; + int wval; + + int fd = (int) drv_data; + int ready_fd = (int) drv_event; + + while(1) { + if ((p = fd_data[ready_fd].pending) == NULL) { + driver_select(driver_data[fd].port_num, ready_fd, + ERL_DRV_WRITE, 0); + return; + } + wval = write(p->fd, p->cpos, p->remain); + if (wval == p->remain) { + fd_data[ready_fd].pending = p->next; + erts_free(ERTS_ALC_T_PEND_DATA, p); + if (fd_data[ready_fd].pending == NULL) { + driver_select(driver_data[fd].port_num, ready_fd, + ERL_DRV_WRITE, 0); + set_busy_port(driver_data[fd].port_num, 0); + return; + } + else + continue; + } + else if (wval < 0) { + if (errno == ERRNO_BLOCK || errno == EINTR) + return; + else { + driver_select(driver_data[fd].port_num, ready_fd, + ERL_DRV_WRITE, 0); + driver_failure(driver_data[fd].port_num, -1); + return; + } + } + else if (wval < p->remain) { + p->cpos += wval; + p->remain -= wval; + return; + } + } +} + +/* Fills in the systems representation of the jam/beam process identifier. +** The Pid is put in STRING representation in the supplied buffer, +** no interpretatione of this should be done by the rest of the +** emulator. The buffer should be at least 21 bytes long. +*/ +void sys_get_pid(char *buffer){ + int p = taskIdSelf(); /* Hmm, may be negative??? requires some GB of + memory to make the TCB address convert to a + negative value. */ + sprintf(buffer,"%d", p); +} + +int +erts_sys_putenv(char *buffer, int sep_ix) +{ + return putenv(buffer); +} + +int +erts_sys_getenv(char *key, char *value, size_t *size) +{ + char *orig_value; + int res; + orig_value = getenv(key); + if (!orig_value) + res = -1; + else { + size_t len = sys_strlen(orig_value); + if (len >= *size) { + *size = len + 1; + res = 1; + } + else { + *size = len; + sys_memcpy((void *) value, (void *) orig_value, len+1); + res = 0; + } + } + return res; +} + +void +sys_init_io(void) +{ + tmp_buf = (byte *) erts_alloc(ERTS_ALC_T_SYS_TMP_BUF, SYS_TMP_BUF_SIZE); + tmp_buf_size = SYS_TMP_BUF_SIZE; + fd_data = (struct fd_data *) + erts_alloc(ERTS_ALC_T_FD_TAB, max_files * sizeof(struct fd_data)); +} + + +/* Fill buffer, return buffer length, 0 for EOF, < 0 for error. */ + +static int read_fill(int fd, char *buf, int len) +{ + int i, got = 0; + do { + if ((i = read(fd, buf+got, len-got)) <= 0) { + return i; + } + got += i; + } while (got < len); + return (len); +} + + +/************************** Misc... *******************************/ + +extern const char pre_loaded_code[]; +extern char* const pre_loaded[]; + + +/* Float conversion */ + +int sys_chars_to_double(char *buf, double *fp) +{ + char *s = buf; + + /* The following check is incorporated from the Vee machine */ + +#define ISDIGIT(d) ((d) >= '0' && (d) <= '9') + + /* Robert says that something like this is what he really wanted: + * + * 7 == sscanf(Tbuf, "%[+-]%[0-9].%[0-9]%[eE]%[+-]%[0-9]%s", ....); + * if (*s2 == 0 || *s3 == 0 || *s4 == 0 || *s6 == 0 || *s7) + * break; + */ + + /* Scan string to check syntax. */ + if (*s == '+' || *s == '-') + s++; + + if (!ISDIGIT(*s)) /* Leading digits. */ + return -1; + while (ISDIGIT(*s)) s++; + if (*s++ != '.') /* Decimal part. */ + return -1; + if (!ISDIGIT(*s)) + return -1; + while (ISDIGIT(*s)) s++; + if (*s == 'e' || *s == 'E') { + /* There is an exponent. */ + s++; + if (*s == '+' || *s == '-') + s++; + if (!ISDIGIT(*s)) + return -1; + while (ISDIGIT(*s)) s++; + } + if (*s) /* That should be it */ + return -1; + + if (sscanf(buf, "%lf", fp) != 1) + return -1; + return 0; +} + +/* + ** Convert a double to ascii format 0.dddde[+|-]ddd + ** return number of characters converted + */ + +int sys_double_to_chars(double fp, char *buf) +{ + (void) sprintf(buf, "%.20e", fp); + return strlen(buf); +} + + +/* Floating point exceptions */ + +#if (CPU == SPARC) +jmp_buf fpe_jmp; + +RETSIGTYPE fpe_sig_handler(int sig) +{ + longjmp(fpe_jmp, 1); +} + +#elif (CPU == PPC603) +static void fix_registers(void){ + FP_CONTEXT fpcontext; + fppSave(&fpcontext); + fpcontext.fpcsr &= ~(_PPC_FPSCR_INIT); + fppRestore(&fpcontext); +} +#endif + + +/* Return a pointer to a vector of names of preloaded modules */ + +Preload* sys_preloaded(void) +{ + return (Preload *) pre_loaded; +} + +/* Return a pointer to preloaded code for module "module" */ +unsigned char* sys_preload_begin(Preload *pp) +{ + return pp->code; +} + +/* Clean up if allocated */ +void sys_preload_end(Preload *pp) +{ + /* Nothing */ +} + +/* Read a key from console (?) */ + +int sys_get_key(int fd) +{ + int c; + unsigned char rbuf[64]; + + fflush(stdout); /* Flush query ??? */ + + if ((c = read(fd,rbuf,64)) <= 0) + return c; + return rbuf[0]; +} + + +/* A real printf that does the equivalent of fprintf(stdout, ...) */ + +/* ARGSUSED */ +static STATUS +stdio_write(char *buf, int nchars, int fp) +{ + if (fwrite(buf, sizeof(char), nchars, (FILE *)fp) == 0) + return(ERROR); + return(OK); +} + +int real_printf(const char *fmt, ...) +{ + va_list ap; + int err; + + va_start(ap, fmt); + err = fioFormatV(fmt, ap, stdio_write, (int)stdout); + va_end(ap); + return(err); +} + + +/* + * Little function to do argc, argv calls from (e.g.) VxWorks shell + * The arguments should be in the form of a single ""-enclosed string + * NOTE: This isn't really part of the emulator, just included here + * so we can use the handy functions and memory reclamation. + */ +void argcall(char *args) +{ + int argc; + char **argv; + FUNCPTR entry; + + if (args != NULL) { + if ((argc = build_argv(args, &argv)) > 0) { + if ((entry = lookup(argv[0])) != NULL) + (*entry)(argc, argv, (char **)NULL); /* NULL for envp */ + else + fprintf(stderr, "Couldn't find %s\n", argv[0]); + } else + fprintf(stderr, "Failed to build argv!\n"); + } else + fprintf(stderr, "No argument list!\n"); +} + + +/* That concludes the Erlang stuff - now we just need to implement an OS... + - Just kidding, but resource reclamation isn't the strength of VxWorks */ +#undef calloc +#undef free +#undef cfree +#undef malloc +#undef realloc +#undef open +#undef creat +#undef socket +#undef accept +#undef close +#undef fopen +#undef fdopen +#undef freopen +#undef fclose + +/********************* Using elib_malloc ****************************/ +/* This gives us yet another level of malloc wrappers. The purpouse */ +/* is to be able to select between different varieties of memory */ +/* allocation without recompiling. */ +/* Maybe the performance is somewhat degraded by this, but */ +/* on the other hand, performance may be much better if the most */ +/* suiting malloc is used (not to mention the much lower */ +/* fragmentation). */ +/* /Patrik N */ +/********************************************************************/ + +/* + * I don't want to include the whole elib header, especially + * as it uses char * for generic pointers. Let's fool ANSI C instead. + */ +extern void *elib_malloc(size_t); +extern void *elib_realloc(void *, size_t); +extern void elib_free(void *); +extern void elib_init(void *, int); +extern void elib_force_init(void *, int); +extern size_t elib_sizeof(void *); + +/* Flags */ +#define USING_ELIB_MALLOC 1 /* We are using the elib_malloc */ +#define WARN_MALLOC_MIX 2 /* Warn if plain malloc or save_malloc + is mixed with sys_free2 or + sys_realloc2 */ +#define REALLOC_MOVES 4 /* Always move on realloc + (less fragmentation) */ +#define USER_POOL 8 /* The user supplied the memory + pool, it was not save_alloced. */ +#define RECLAIM_USER_POOL 16 /* Use the reclaim mechanism in the + user pool. */ +#define NEW_USER_POOL 32 /* The user pool is newly suppllied, + any old pool should be discarded */ + + +#define ELIB_LOCK \ +if(alloc_flags & USING_ELIB_MALLOC) \ + semTake(elib_malloc_sem, WAIT_FOREVER) + +#define ELIB_UNLOCK \ +if(alloc_flags & USING_ELIB_MALLOC) \ + semGive(elib_malloc_sem) + +#define USER_RECLAIM() ((alloc_flags & USING_ELIB_MALLOC) && \ + (alloc_flags & USER_POOL) && \ + (alloc_flags & RECLAIM_USER_POOL)) + +/* + * Global state + * The use of function pointers for the malloc/realloc/free functions + * is actually only useful in the malloc case, we must know what kind of + * realloc/free we are going to use, so we could call elib_xxx directly. + * However, as the overhead is small and this construction makes it + * fairly easy to add another malloc algorithm, the function pointers + * are used in realloc/free to. + */ +static MallocFunction actual_alloc = &save_malloc; +static ReallocFunction actual_realloc = &save_realloc; +static FreeFunction actual_free = &save_free; +static int alloc_flags = 0; +static int alloc_pool_size = 0; +static void *alloc_pool_ptr = NULL; +static SEM_ID elib_malloc_sem = NULL; + +/* + * Descide if we should use the save_free instead of elib_free or, + * in the case of the free used in a delete hook, if we should + * use plain free instead of elib_free. + */ +static int use_save_free(void *ptr){ + register int diff = ((char *) ptr) - ((char *) alloc_pool_ptr); + /* + * Hmmm... should it be save_free even if diff is exactly 0? + * The answer is Yes if the whole area is save_alloced and No if not, + * so reclaim_free_hook is NOT run in the case of one save_alloced area. + */ + return (!(alloc_flags & USING_ELIB_MALLOC) || + (diff < 0 || diff >= alloc_pool_size)); +} + +/* + * A free function used by the task deletion hook for the save_xxx functions. + * Set with the set_reclaim_free_function function. + */ +static void reclaim_free_hook(void *ptr){ + if(use_save_free(ptr)){ + free(ptr); + } else { + ELIB_LOCK; + (*actual_free)(ptr); + ELIB_UNLOCK; + } +} + + +/* + * Initialize, sets the values of pointers based on + * either nothing (the default) or what's set previously by the + * erl_set_memory_block function. + */ +static void initialize_allocation(void){ + set_reclaim_free_function(NULL); + if(alloc_pool_size == 0){ + actual_alloc = (void *(*)(size_t))&save_malloc; + actual_realloc = (void *(*)(void *, size_t))&save_realloc; + actual_free = &save_free; + alloc_flags &= ~(USING_ELIB_MALLOC | USER_POOL | RECLAIM_USER_POOL); + } else { + if(elib_malloc_sem == NULL) + elib_malloc_sem = semMCreate + (SEM_Q_PRIORITY | SEM_DELETE_SAFE | SEM_INVERSION_SAFE); + if(elib_malloc_sem == NULL) + erl_exit(1,"Could not create mutex semaphore for elib_malloc"); + if(!(alloc_flags & USER_POOL)){ + if((alloc_pool_ptr = save_malloc(alloc_pool_size)) == NULL) + erl_exit(1,"Erlang set to allocate a %d byte block initially;" + " not enough memory available.", alloc_pool_size); + elib_force_init(alloc_pool_ptr, alloc_pool_size); + } else if(alloc_flags & NEW_USER_POOL){ + elib_force_init(alloc_pool_ptr, alloc_pool_size); + } + actual_alloc=&elib_malloc; + actual_realloc=&elib_realloc; + actual_free=&elib_free; + alloc_flags |= USING_ELIB_MALLOC; + /* We MUST see to that the right free function is used + otherwise we'll get a very nasty crash! */ + if(USER_RECLAIM()) + set_reclaim_free_function(&reclaim_free_hook); + } + alloc_flags &= ~(NEW_USER_POOL); /* It's never new after initialization*/ +} + +/* This does not exist on other platforms, we just use it in sys.c + and the BSD resolver */ +void *sys_calloc2(Uint nelem, Uint elsize){ + void *ptr = erts_alloc_fnf(ERTS_ALC_T_UNDEF, nelem*elsize); + if(ptr != NULL) + memset(ptr,0,nelem*elsize); + return ptr; +} + +/* + * The malloc wrapper + */ +void * +erts_sys_alloc(ErtsAlcType_t type, void *extra, Uint size) +{ + register void *ret; + ELIB_LOCK; + if(USER_RECLAIM()) + ret = save_malloc2((size_t)size,actual_alloc); + else + ret = (*actual_alloc)((size_t)size); + ELIB_UNLOCK; + return ret; +} + +/* + * The realloc wrapper, may respond to the "realloc-always-moves" flag + * if the area is initially allocated with elib_malloc. + */ +void * +erts_sys_realloc(ErtsAlcType_t type, void *extra, void *ptr, Uint size) +{ + register void *ret; + if(use_save_free(ptr)){ + if((alloc_flags & WARN_MALLOC_MIX) && + (alloc_flags & USING_ELIB_MALLOC)) + erts_fprintf(stderr,"Warning, save_malloced data realloced " + "by sys_realloc2\n"); + return save_realloc(ptr, (size_t) size); + } else { + ELIB_LOCK; + if((alloc_flags & REALLOC_MOVES) && + (alloc_flags & USING_ELIB_MALLOC)){ + size_t osz = elib_sizeof(ptr); + if(USER_RECLAIM()) + ret = save_malloc2((size_t) size, actual_alloc); + else + ret = (*actual_alloc)((size_t) size); + if(ret != NULL){ + memcpy(ret,ptr,(((size_t)size) < osz) ? ((size_t)size) : osz); + if(USER_RECLAIM()) + save_free2(ptr,actual_free); + else + (*actual_free)(ptr); + } + } else { + if(USER_RECLAIM()) + ret = save_realloc2(ptr,(size_t)size,actual_realloc); + else + ret = (*actual_realloc)(ptr,(size_t)size); + } + ELIB_UNLOCK; + return ret; + } +} + +/* + * Wrapped free(). + */ +void +erts_sys_free(ErtsAlcType_t type, void *extra, void *ptr) +{ + if(use_save_free(ptr)){ + /* + * This might happen when linked in drivers use save_malloc etc + * directly. + */ + if((alloc_flags & WARN_MALLOC_MIX) && + (alloc_flags & USING_ELIB_MALLOC)) + erts_fprintf(stderr,"Warning, save_malloced data freed by " + "sys_free2\n"); + save_free(ptr); + } else { + ELIB_LOCK; + if(USER_RECLAIM()) + save_free2(ptr,actual_free); + else + (*actual_free)(ptr); + ELIB_UNLOCK; + } +} + +/* + * External interface to be called before erlang is started + * Parameters: + * isize: The size of the memory block where erlang should malloc(). + * iptr: (optional) A pointer to a user supplied memory block of + * size isize. + * warn_save: Instructs sys_free2 and sys_realloc2 to warn if + * memory allocation/reallocation/freeing is mixed between + * pure malloc/save_malloc/sys_alloc2 routines (only + * warns if elib is actually used in the sys_alloc2 routines). + * realloc_moves: Always allocate a fresh memory block on reallocation + * (less fragmentation). + * reclaim_in_supplied: Use memory reclaim mechanisms inside the user + * supplied area, this makes one area reusable between + * starts of erlang and might be nice for drivers etc. + */ + +int erl_set_memory_block(int isize, int iptr, int warn_save, + int realloc_moves, int reclaim_in_supplied, int p5, + int p6, int p7, int p8, int p9){ + if(erlang_id != 0){ + erts_fprintf(stderr,"Error, cannot set erlang memory block while an " + "erlang task is running!\n"); + return 1; + } + if(isize < 8 * 1024 *1024) + erts_fprintf(stderr, + "Warning, the memory pool of %dMb may be to small to " + "run erlang in!\n", isize / (1024 * 1024)); + alloc_pool_size = (size_t) isize; + alloc_pool_ptr = (void *) iptr; + alloc_flags = 0; + /* USING_ELIB_MALLOC gets set by the initialization routine */ + if((void *)iptr != NULL) + alloc_flags |= (USER_POOL | NEW_USER_POOL); + if(realloc_moves) + alloc_flags |= REALLOC_MOVES; + if(warn_save) + alloc_flags |= WARN_MALLOC_MIX; + if((void *)iptr != NULL && reclaim_in_supplied) + alloc_flags |= RECLAIM_USER_POOL; + return 0; +} + +/* External statistics interface */ +int erl_memory_show(int p0, int p1, int p2, int p3, int p4, int p5, + int p6, int p7, int p8, int p9){ + struct elib_stat statistics; + if(!(alloc_flags & USING_ELIB_MALLOC) && erlang_id != 0){ + erts_printf("Using plain save_alloc, use memShow instead.\n"); + return 1; + } + if(erlang_id == 0 && !((alloc_flags & USER_POOL) && + !(alloc_flags & NEW_USER_POOL))){ + erts_printf("Sorry, no allocation statistics until erlang " + "is started.\n"); + return 1; + } + erts_printf("Allocation settings:\n"); + erts_printf("Using elib_malloc with memory pool size of %lu bytes.\n", + (unsigned long) alloc_pool_size); + erts_printf("Realloc-always-moves is %s\n", + (alloc_flags & REALLOC_MOVES) ? "on" : "off"); + erts_printf("Warnings about mixed malloc/free's are %s\n", + (alloc_flags & WARN_MALLOC_MIX) ? "on" : "off"); + if(alloc_flags & USER_POOL){ + erts_printf("The memory block used by elib is user supplied " + "at 0x%08x.\n", (unsigned int) alloc_pool_ptr); + if(alloc_flags & RECLAIM_USER_POOL) + erts_printf("Allocated memory within the user supplied pool\n" + " will be automatically reclaimed at task exit.\n"); + } else { + erts_printf("The memory block used by elib is save_malloc'ed " + "at 0x%08x.\n", (unsigned int) alloc_pool_ptr); + } +#ifdef NO_FIX_ALLOC + erts_printf("Fix_alloc is disabled in this build\n"); +#endif + erts_printf("Statistics from elib_malloc:\n"); + ELIB_LOCK; + + elib_stat(&statistics); + ELIB_UNLOCK; + erts_printf("Type Size (bytes) Number of blocks\n"); + erts_printf("============= ============ ================\n"); + erts_printf("Total: %12lu %16lu\n", + (unsigned long) statistics.mem_total*4, + (unsigned long) statistics.mem_blocks); + erts_printf("Allocated: %12lu %16lu\n", + (unsigned long) statistics.mem_alloc*4, + (unsigned long) statistics.mem_blocks-statistics.free_blocks); + erts_printf("Free: %12lu %16lu\n", + (unsigned long) statistics.mem_free*4, + (unsigned long) statistics.free_blocks); + erts_printf("Largest free: %12lu -\n\n", + (unsigned long) statistics.max_free*4); + return 0; +} + + +/* +** More programmer friendly (as opposed to user friendly ;-) interface +** to the memory statistics. Resembles the VxWorks memPartInfoGet but +** does not take a partition id as parameter... +*/ +int erl_mem_info_get(MEM_PART_STATS *stats){ + struct elib_stat statistics; + if(!(alloc_flags & USING_ELIB_MALLOC)) + return -1; + ELIB_LOCK; + elib_stat(&statistics); + ELIB_UNLOCK; + stats->numBytesFree = statistics.mem_free*4; + stats->numBlocksFree = statistics.free_blocks; + stats->maxBlockSizeFree = statistics.max_free*4; + stats->numBytesAlloc = statistics.mem_alloc*4; + stats->numBlocksAlloc = statistics.mem_blocks-statistics.free_blocks; + return 0; +} + +/********************* Pipe driver **********************************/ +/* + * Purpose: Pipe driver with Unix (unnamed) pipe semantics. + * Author: Peter Hogfeldt ([email protected]) from an outline + * by Per Hedeland ([email protected]). + * + * Note: This driver must *not* use the reclaim facilities, hence it + * is placed here. (after the #undef's of open,malloc etc) + * + * This driver supports select() and non-blocking I/O via + * ioctl(fd, FIONBIO, val). + * + * 1997-03-21 Peter Hogfeldt + * Added non-blocking I/O. + * + */ + +/* + * SEMAPHORES + * + * Each end of a pipe has two semaphores: semExcl for serialising access to + * the pipe end, and semBlock for blocking I/O. + * + * reader->semBlock is available (full) if and only if the pipe is + * not empty, or the write end is closed. Otherwise + * it is unavailable (empty). It is initially + * unavailable. + * + * writer->semBlock is available (full) if and only if the pipe is + * not full, or if the reader end is closed. + * Otherwise it is unavailable. It is initially + * available. + */ + +#define UXPIPE_SIZE 4096 + +/* Forward declaration */ +typedef struct uxPipeDev UXPIPE_DEV; + +/* + * Pipe descriptor (one for each open pipe). + */ +typedef struct { + int drvNum; + UXPIPE_DEV *reader, *writer; + RING_ID ringId; +} UXPIPE; + +/* + * Device descriptor (one for each of the read and write + * ends of an open pipe). + */ +struct uxPipeDev { + UXPIPE *pipe; + int blocking; + SEL_WAKEUP_LIST wakeupList; + SEM_ID semExcl; + SEM_ID semBlock; +}; + +int uxPipeDrvNum = 0; /* driver number of pipe driver */ + +#define PIPE_NAME "/uxpipe" /* only used internally */ +#define PIPE_READ "/r" /* ditto */ +#define PIPE_WRITE "/w" /* ditto */ + +LOCAL char pipeRead[64], pipeWrite[64]; +LOCAL DEV_HDR devHdr; +LOCAL UXPIPE *newPipe; /* communicate btwn open()s in pipe() */ +LOCAL SEM_ID pipeSem; /* mutual exclusion in pipe() */ + +/* forward declarations */ +LOCAL int uxPipeOpen(DEV_HDR *pDv, char *name, int mode); +LOCAL int uxPipeClose(UXPIPE_DEV *pDev); +LOCAL int uxPipeRead(UXPIPE_DEV *pDev, char *buffer, int maxbytes); +LOCAL int uxPipeWrite(UXPIPE_DEV *pDev, char *buffer, int nbytes); +LOCAL STATUS uxPipeIoctl(FAST UXPIPE_DEV *pDev, FAST int function, int arg); + + +/*************************************************************************** + * + * uxPipeDrv - install Unix pipe driver + * + * This routine initializes the Unix pipe driver. It must be called + * before any other routine in this driver. + * + * RETURNS: + * OK, or ERROR if I/O system is unable to install driver. + */ + +STATUS +uxPipeDrv(void) +{ + if (uxPipeDrvNum > 0) + return (OK); /* driver already installed */ + if ((uxPipeDrvNum = iosDrvInstall((FUNCPTR) NULL, (FUNCPTR) NULL, + uxPipeOpen, uxPipeClose, uxPipeRead, + uxPipeWrite, uxPipeIoctl)) == ERROR) + return (ERROR); + if (iosDevAdd(&devHdr, PIPE_NAME, uxPipeDrvNum) == ERROR) + return (ERROR); + strcpy(pipeRead, PIPE_NAME); + strcat(pipeRead, PIPE_READ); + strcpy(pipeWrite, PIPE_NAME); + strcat(pipeWrite, PIPE_WRITE); + if ((pipeSem = semMCreate(SEM_Q_PRIORITY | SEM_DELETE_SAFE)) == NULL) + return (ERROR); + return (OK); +} + +/*************************************************************************** + * + * uxPipeOpen - open a pipe + * + * RETURNS: Pointer to device descriptor, or ERROR if memory cannot be + * allocated (errno = ENOMEM), or invalid argument (errno = EINVAL). + */ + +/* + * DEV_HDR *pDv; pointer to device header (dummy) + * char *name; name of pipe to open ("/r" or "/w") + * int mode; access mode (O_RDONLY or O_WRONLY) + */ +LOCAL int +uxPipeOpen(DEV_HDR *pDv, char *name, int mode) +{ + UXPIPE_DEV *reader, *writer; + + if (mode == O_RDONLY && strcmp(name, PIPE_READ) == 0) { + /* reader open */ + if ((newPipe = (UXPIPE *) malloc(sizeof(UXPIPE))) != NULL) { + if ((newPipe->ringId = rngCreate(UXPIPE_SIZE)) != NULL) { + if ((reader = (UXPIPE_DEV *) malloc(sizeof(UXPIPE_DEV))) != NULL) { + if ((reader->semExcl = semBCreate(SEM_Q_FIFO, SEM_FULL)) != NULL) { + if ((reader->semBlock = semBCreate(SEM_Q_FIFO, SEM_EMPTY)) != NULL) { + reader->pipe = newPipe; + reader->blocking = 1; + selWakeupListInit(&reader->wakeupList); + newPipe->reader = reader; + newPipe->writer = NULL; + newPipe->drvNum = uxPipeDrvNum; + return ((int) reader); + } + semDelete(reader->semExcl); + } + free(reader); + } + rngDelete(newPipe->ringId); + } + free(newPipe); + newPipe = NULL; + errno = ENOMEM; + } + } else if (mode == O_WRONLY && strcmp(name, PIPE_WRITE) == 0) { + /* writer open */ + if (newPipe != NULL && + (writer = (UXPIPE_DEV *) malloc(sizeof(UXPIPE_DEV))) != NULL) { + if ((writer->semExcl = semBCreate(SEM_Q_FIFO, SEM_FULL)) != NULL) { + if ((writer->semBlock = semBCreate(SEM_Q_FIFO, SEM_FULL)) != NULL) { + writer->blocking = 1; + writer->pipe = newPipe; + selWakeupListInit(&writer->wakeupList); + newPipe->writer = writer; + newPipe = NULL; + return ((int) writer); + } + semDelete(writer->semExcl); + } + free(writer); + } + if (newPipe != NULL) + free(newPipe); + newPipe = NULL; + errno = ENOMEM; + } else { + errno = EINVAL; + } + return (ERROR); +} + +/*************************************************************************** + * + * uxPipeClose - close read or write end of a pipe. + * + * RETURNS: + * OK, or ERROR if device descriptor does not refer to an open read or + write end of a pipe (errno = EBADF). + */ + +LOCAL int +uxPipeClose(UXPIPE_DEV *pDev) +{ + UXPIPE *pajp = pDev->pipe; + + taskLock(); + if (pDev == pajp->reader) { + /* Close this end */ + semDelete(pDev->semExcl); + semDelete(pDev->semBlock); + free(pDev); + pajp->reader = NULL; + /* Inform the other end */ + if (pajp->writer != NULL) { + selWakeupAll(&pajp->writer->wakeupList, SELWRITE); + semGive(pajp->writer->semBlock); + } + } else if (pDev == pajp->writer) { + /* Close this end */ + semDelete(pDev->semExcl); + semDelete(pDev->semBlock); + free(pDev); + pajp->writer = NULL; + /* Inform the other end */ + if (pajp->reader != NULL) { + selWakeupAll(&pajp->reader->wakeupList, SELREAD); + semGive(pajp->reader->semBlock); + } + } else { + errno = EBADF; + taskUnlock(); + return (ERROR); + } + if (pajp->reader == NULL && pajp->writer == NULL) { + rngDelete(pajp->ringId); + pajp->drvNum = 0; + free(pajp); + } + taskUnlock(); + return (OK); +} +/*************************************************************************** + * + * uxPipeRead - read from a pipe. + * + * Reads at most maxbytes bytes from the pipe. Blocks if blocking mode is + * set and the pipe is empty. + * + * RETURNS: + * number of bytes read, 0 on EOF, or ERROR if device descriptor does + * not refer to an open read end of a pipe (errno = EBADF), or if + * non-blocking mode is set and the pipe is empty (errno = EWOULDBLOCK). + */ + +LOCAL int +uxPipeRead(UXPIPE_DEV *pDev, char *buffer, int maxbytes) +{ + UXPIPE *pajp = pDev->pipe; + int nbytes = 0; + + if (pDev != pajp->reader) { + errno = EBADF; + return (ERROR); + } + if (maxbytes == 0) + return (0); + semTake(pDev->semExcl, WAIT_FOREVER); + /* + * Note that semBlock may be full, although there is nothing to read. + * This happens e.g. after the following sequence of operations: a + * reader task blocks, a writer task writes two times (the first + * write unblocks the reader task, the second write makes semBlock + * full). + */ + while (nbytes == 0) { + if (pDev->blocking) + semTake(pDev->semBlock, WAIT_FOREVER); + /* + * Reading and updating of the write end must not be interleaved + * with a write from another task - hence we lock this task. + */ + taskLock(); + nbytes = rngBufGet(pajp->ringId, buffer, maxbytes); + if (nbytes > 0) { + /* Give own semaphore if bytes remain or if write end is closed */ + if ((!rngIsEmpty(pajp->ringId) || pajp->writer == NULL) && + pDev->blocking) + semGive(pDev->semBlock); + /* Inform write end */ + if (pajp->writer != NULL) { + if (pajp->writer->blocking) + semGive(pajp->writer->semBlock); + selWakeupAll(&pajp->writer->wakeupList, SELWRITE); + } + } else if (pajp->writer == NULL) { + nbytes = 0; /* EOF */ + /* Give semaphore when write end is closed */ + if (pDev->blocking) + semGive(pDev->semBlock); + taskUnlock(); + semGive(pDev->semExcl); + return (nbytes); + } else if (!pDev->blocking) { + taskUnlock(); + semGive(pDev->semExcl); + errno = EWOULDBLOCK; + return (ERROR); + } + taskUnlock(); + } + semGive(pDev->semExcl); + return (nbytes); +} + +/*************************************************************************** + * + * uxPipeWrite - write to a pipe. + * + * Writes nbytes bytes to the pipe. Blocks if blocking mode is set, and if + * the pipe is full. + * + * RETURNS: + * number of bytes written, or ERROR if the device descriptor does not + * refer to an open write end of a pipe (errno = EBADF); or if the read end + * of the pipe is closed (errno = EPIPE); or if non-blocking mode is set + * and the pipe is full (errno = EWOULDBLOCK). + * + */ + +LOCAL int +uxPipeWrite(UXPIPE_DEV *pDev, char *buffer, int nbytes) +{ + + UXPIPE *pajp = pDev->pipe; + int sofar = 0, written; + + if (pDev != pajp->writer) { + errno = EBADF; + return (ERROR); + } + if (pajp->reader == NULL) { + errno = EPIPE; + return (ERROR); + } + if (nbytes == 0) + return (0); + semTake(pDev->semExcl, WAIT_FOREVER); + while (sofar < nbytes) { + if (pDev->blocking) + semTake(pDev->semBlock, WAIT_FOREVER); + if (pajp->reader == NULL) { + errno = EPIPE; + semGive(pDev->semBlock); + semGive(pDev->semExcl); + return (ERROR); + } + /* Writing and updating of the read end must not be interleaved + * with a read from another task - hence we lock this task. + */ + taskLock(); + written = rngBufPut(pajp->ringId, buffer + sofar, nbytes - sofar); + sofar += written; + /* Inform the read end if we really wrote something */ + if (written > 0 && pajp->reader != NULL) { + selWakeupAll(&pajp->reader->wakeupList, SELREAD); + if (pajp->reader->blocking) + semGive(pajp->reader->semBlock); + } + taskUnlock(); + if (!pDev->blocking) { + if (sofar == 0) { + errno = EWOULDBLOCK; + sofar = ERROR; + } + break; + } + } + /* Give own semaphore if space remains */ + if (!rngIsFull(pajp->ringId) && pDev->blocking) + semGive(pDev->semBlock); + semGive(pDev->semExcl); + return (sofar); +} + +/*************************************************************************** + * + * uxPipeIoctl - do device specific I/O control + * + * RETURNS: + * OK or ERROR. + */ + +LOCAL STATUS +uxPipeIoctl(FAST UXPIPE_DEV *pDev, FAST int function, int arg) + +{ + UXPIPE *pajp = pDev->pipe; + int status = OK; + + switch (function) { + case FIONBIO: + pDev->blocking = (*(int *)arg) ? 0 : 1; + break; + case FIOSELECT: + taskLock(); + selNodeAdd(&pDev->wakeupList, (SEL_WAKEUP_NODE *) arg); + if (selWakeupType((SEL_WAKEUP_NODE *) arg) == SELREAD && + pDev == pajp->reader && + (!rngIsEmpty(pajp->ringId) || pajp->writer == NULL)) + selWakeup((SEL_WAKEUP_NODE *) arg); + if (selWakeupType((SEL_WAKEUP_NODE *) arg) == SELWRITE && + pDev == pajp->writer && + (!rngIsFull(pajp->ringId) || pajp->reader == NULL)) + selWakeup((SEL_WAKEUP_NODE *) arg); + taskUnlock(); + break; + case FIOUNSELECT: + selNodeDelete(&pDev->wakeupList, (SEL_WAKEUP_NODE *) arg); + break; + default: + status = ERROR; + break; + } + return (status); +} + +/*************************************************************************** + * + * pipe - create an intertask channel + * + * Creates a pipe. fd[0] (fd[1]) is the read (write) file descriptor. + * + * RETURNS: + * OK or ERROR, if the pipe could not be created. + */ + +STATUS +pipe(int fd[2]) +{ + semTake(pipeSem, WAIT_FOREVER); + if ((fd[0] = open(pipeRead, O_RDONLY, 0)) != ERROR) { + if ((fd[1] = open(pipeWrite, O_WRONLY, 0)) != ERROR) { + semGive(pipeSem); + return (OK); + } + (void) close(fd[0]); + } + errno &= 0xFFFF; + if((errno & 0xFFFF) == EINTR) /* Why on earth EINTR??? */ + errno = ENFILE; /* It means we are out of file descriptors...*/ + semGive(pipeSem); + return (ERROR); +} + +/*************************************************************************** + * + * uxPipeShow - display pipe information + * + * RETURNS: + * N/A. + */ + +void +uxPipeShow(int fd) +{ + UXPIPE_DEV *pDev; + UXPIPE *pajp; + int drvValue; + + if ((drvValue = iosFdValue(fd)) == ERROR) { + erts_fprintf(stderr, "Error: file descriptor invalid\n"); + return; + } + pDev = (UXPIPE_DEV *)drvValue; + pajp = pDev->pipe; + if (pajp->drvNum != uxPipeDrvNum) { + erts_fprintf(stderr, "Error: Not a ux pipe device\n"); + return; + } + erts_fprintf(stderr, "Device : 0x%x\n", (int) pDev); + erts_fprintf(stderr, "Buffer size : %d\n", UXPIPE_SIZE); + erts_fprintf(stderr, "Bytes in buffer : %d\n\n", rngNBytes(pajp->ringId)); + erts_fprintf(stderr, "READ END\n\n"); + if (pajp->reader != NULL) { + erts_fprintf(stderr, "Mode : "); + erts_fprintf(stderr, "%s\n", + (pajp->reader->blocking) ? "blocking" : "non-blocking"); + } + erts_fprintf(stderr, "Status : "); + if (pajp->reader != NULL) { + erts_fprintf(stderr, "OPEN\n"); + erts_fprintf(stderr, "Wake-up list : %d\n\n", + selWakeupListLen(&pajp->reader->wakeupList)); + erts_fprintf(stderr, "Exclusion Semaphore\n"); + semShow(pajp->reader->semExcl, 1); + erts_fprintf(stderr, "Blocking Semaphore\n"); + semShow(pajp->reader->semBlock, 1); + } else + erts_fprintf(stderr, "CLOSED\n\n"); + erts_fprintf(stderr, "WRITE END\n\n"); + if (pajp->writer != NULL) { + erts_fprintf(stderr, "Mode : "); + erts_fprintf(stderr, "%s\n", + (pajp->writer->blocking) ? "blocking" : "non-blocking"); + } + erts_fprintf(stderr, "Status : "); + if (pajp->writer != NULL) { + erts_fprintf(stderr, "OPEN\n"); + erts_fprintf(stderr, "Wake-up list : %d\n\n", + selWakeupListLen(&pajp->writer->wakeupList)); + erts_fprintf(stderr, "Exclusion Semaphore\n"); + semShow(pajp->writer->semExcl, 1); + erts_fprintf(stderr, "Blocking Semaphore\n"); + semShow(pajp->writer->semBlock, 1); + } else + erts_fprintf(stderr, "CLOSED\n\n"); +} + +#ifdef DEBUG +void +erl_assert_error(char* expr, char* file, int line) +{ + fflush(stdout); + fprintf(stderr, "Assertion failed: %s in %s, line %d\n", + expr, file, line); + fflush(stderr); + erl_crash_dump(file, line, "Assertion failed: %s\n", expr); + abort(); +} +void +erl_debug(char* fmt, ...) +{ + char sbuf[1024]; /* Temporary buffer. */ + va_list va; + + va_start(va, fmt); + vsprintf(sbuf, fmt, va); + va_end(va); + fprintf(stderr, "%s\n", sbuf); +} +#endif diff --git a/erts/emulator/sys/win32/dosmap.c b/erts/emulator/sys/win32/dosmap.c new file mode 100644 index 0000000000..15416a66c5 --- /dev/null +++ b/erts/emulator/sys/win32/dosmap.c @@ -0,0 +1,282 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 1998-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ +/* + * _dosmaperr: maps Windows OS errors to Unix System V errno values + * + * Contributor: Michael Regen + */ + +/* Only use for win32 if linking to MSVCR??.DLL and not if statically linking + to LIBCMT.LIB */ +#if defined(WIN32) && defined(_MT) && defined(_DLL) + +#include <errno.h> +#include <winerror.h> +#include <stdlib.h> + +/* Position in table = Windows OS error -> Posix errno +** An exception for ERROR_NOT_ENOUGH_QUOTA - 1816 is in _dosmaperr +*/ +static const unsigned char errMapTable[] = { + EINVAL, /* ERROR_SUCCESS 0 */ + EINVAL, /* ERROR_INVALID_FUNCTION 1 */ + ENOENT, /* ERROR_FILE_NOT_FOUND 2 */ + ENOENT, /* ERROR_PATH_NOT_FOUND 3 */ + EMFILE, /* ERROR_TOO_MANY_OPEN_FILES 4 */ + EACCES, /* ERROR_ACCESS_DENIED 5 */ + EBADF, /* ERROR_INVALID_HANDLE 6 */ + ENOMEM, /* ERROR_ARENA_TRASHED 7 */ + ENOMEM, /* ERROR_NOT_ENOUGH_MEMORY 8 */ + ENOMEM, /* ERROR_INVALID_BLOCK 9 */ + E2BIG, /* ERROR_BAD_ENVIRONMENT 10 */ + ENOEXEC, /* ERROR_BAD_FORMAT 11 */ + EINVAL, /* ERROR_INVALID_ACCESS 12 */ + EINVAL, /* ERROR_INVALID_DATA 13 */ + EINVAL, /* ERROR_OUTOFMEMORY 14 */ + ENOENT, /* ERROR_INVALID_DRIVE 15 */ + EACCES, /* ERROR_CURRENT_DIRECTORY 16 */ + EXDEV, /* ERROR_NOT_SAME_DEVICE 17 */ + ENOENT, /* ERROR_NO_MORE_FILES 18 */ + EACCES, /* ERROR_WRITE_PROTECT 19 */ + EACCES, /* ERROR_BAD_UNIT 20 */ + EACCES, /* ERROR_NOT_READY 21 */ + EACCES, /* ERROR_BAD_COMMAND 22 */ + EACCES, /* ERROR_CRC 23 */ + EACCES, /* ERROR_BAD_LENGTH 24 */ + EACCES, /* ERROR_SEEK 25 */ + EACCES, /* ERROR_NOT_DOS_DISK 26 */ + EACCES, /* ERROR_SECTOR_NOT_FOUND 27 */ + EACCES, /* ERROR_OUT_OF_PAPER 28 */ + EACCES, /* ERROR_WRITE_FAULT 29 */ + EACCES, /* ERROR_READ_FAULT 30 */ + EACCES, /* ERROR_GEN_FAILURE 31 */ + EACCES, /* ERROR_SHARING_VIOLATION 32 */ + EACCES, /* ERROR_LOCK_VIOLATION 33 */ + EACCES, /* ERROR_WRONG_DISK 34 */ + EACCES, /* 35 */ + EACCES, /* ERROR_SHARING_BUFFER_EXCEEDED 36 */ + EINVAL, /* 37 */ + EINVAL, /* ERROR_HANDLE_EOF 38 */ + EINVAL, /* ERROR_HANDLE_DISK_FULL 39 */ + EINVAL, /* 40 */ + EINVAL, /* 41 */ + EINVAL, /* 42 */ + EINVAL, /* 43 */ + EINVAL, /* 44 */ + EINVAL, /* 45 */ + EINVAL, /* 46 */ + EINVAL, /* 47 */ + EINVAL, /* 48 */ + EINVAL, /* 49 */ + EINVAL, /* ERROR_NOT_SUPPORTED 50 */ + EINVAL, /* ERROR_REM_NOT_LIST 51 */ + EINVAL, /* ERROR_DUP_NAME 52 */ + ENOENT, /* ERROR_BAD_NETPATH 53 */ + EINVAL, /* ERROR_NETWORK_BUSY 54 */ + EINVAL, /* ERROR_DEV_NOT_EXIST 55 */ + EINVAL, /* ERROR_TOO_MANY_CMDS 56 */ + EINVAL, /* ERROR_ADAP_HDW_ERR 57 */ + EINVAL, /* ERROR_BAD_NET_RESP 58 */ + EINVAL, /* ERROR_UNEXP_NET_ERR 59 */ + EINVAL, /* ERROR_BAD_REM_ADAP 60 */ + EINVAL, /* ERROR_PRINTQ_FULL 61 */ + EINVAL, /* ERROR_NO_SPOOL_SPACE 62 */ + EINVAL, /* ERROR_PRINT_CANCELLED 63 */ + EINVAL, /* ERROR_NETNAME_DELETED 64 */ + EACCES, /* ERROR_NETWORK_ACCESS_DENIED 65 */ + EINVAL, /* ERROR_BAD_DEV_TYPE 66 */ + ENOENT, /* ERROR_BAD_NET_NAME 67 */ + EINVAL, /* ERROR_TOO_MANY_NAMES 68 */ + EINVAL, /* ERROR_TOO_MANY_SESS 69 */ + EINVAL, /* ERROR_SHARING_PAUSED 70 */ + EINVAL, /* ERROR_REQ_NOT_ACCEP 71 */ + EINVAL, /* ERROR_REDIR_PAUSED 72 */ + EINVAL, /* 73 */ + EINVAL, /* 74 */ + EINVAL, /* 75 */ + EINVAL, /* 76 */ + EINVAL, /* 77 */ + EINVAL, /* 78 */ + EINVAL, /* 79 */ + EEXIST, /* ERROR_FILE_EXISTS 80 */ + EINVAL, /* 81 */ + EACCES, /* ERROR_CANNOT_MAKE 82 */ + EACCES, /* ERROR_FAIL_I24 83 */ + EINVAL, /* ERROR_OUT_OF_STRUCTURES 84 */ + EINVAL, /* ERROR_ALREADY_ASSIGNED 85 */ + EINVAL, /* ERROR_INVALID_PASSWORD 86 */ + EINVAL, /* ERROR_INVALID_PARAMETER 87 */ + EINVAL, /* ERROR_NET_WRITE_FAULT 88 */ + EAGAIN, /* ERROR_NO_PROC_SLOTS 89 */ + EINVAL, /* 90 */ + EINVAL, /* 91 */ + EINVAL, /* 92 */ + EINVAL, /* 93 */ + EINVAL, /* 94 */ + EINVAL, /* 95 */ + EINVAL, /* 96 */ + EINVAL, /* 97 */ + EINVAL, /* 98 */ + EINVAL, /* 99 */ + EINVAL, /* ERROR_TOO_MANY_SEMAPHORES 100 */ + EINVAL, /* ERROR_EXCL_SEM_ALREADY_OWNED 101 */ + EINVAL, /* ERROR_SEM_IS_SET 102 */ + EINVAL, /* ERROR_TOO_MANY_SEM_REQUESTS 103 */ + EINVAL, /* ERROR_INVALID_AT_INTERRUPT_TIME 104 */ + EINVAL, /* ERROR_SEM_OWNER_DIED 105 */ + EINVAL, /* ERROR_SEM_USER_LIMIT 106 */ + EINVAL, /* ERROR_DISK_CHANGE 107 */ + EACCES, /* ERROR_DRIVE_LOCKED 108 */ + EPIPE, /* ERROR_BROKEN_PIPE 109 */ + EINVAL, /* ERROR_OPEN_FAILED 110 */ + EINVAL, /* ERROR_BUFFER_OVERFLOW 111 */ + ENOSPC, /* ERROR_DISK_FULL 112 */ + EINVAL, /* ERROR_NO_MORE_SEARCH_HANDLES 113 */ + EBADF, /* ERROR_INVALID_TARGET_HANDLE 114 */ + EINVAL, /* 115 */ + EINVAL, /* 116 */ + EINVAL, /* ERROR_INVALID_CATEGORY 117 */ + EINVAL, /* ERROR_INVALID_VERIFY_SWITCH 118 */ + EINVAL, /* ERROR_BAD_DRIVER_LEVEL 119 */ + EINVAL, /* ERROR_CALL_NOT_IMPLEMENTED 120 */ + EINVAL, /* ERROR_SEM_TIMEOUT 121 */ + EINVAL, /* ERROR_INSUFFICIENT_BUFFER 122 */ + EINVAL, /* ERROR_INVALID_NAME 123 */ + EINVAL, /* ERROR_INVALID_LEVEL 124 */ + EINVAL, /* ERROR_NO_VOLUME_LABEL 125 */ + EINVAL, /* ERROR_MOD_NOT_FOUND 126 */ + EINVAL, /* ERROR_PROC_NOT_FOUND 127 */ + ECHILD, /* ERROR_WAIT_NO_CHILDREN 128 */ + ECHILD, /* ERROR_CHILD_NOT_COMPLETE 129 */ + EBADF, /* ERROR_DIRECT_ACCESS_HANDLE 130 */ + EINVAL, /* ERROR_NEGATIVE_SEEK 131 */ + EACCES, /* ERROR_SEEK_ON_DEVICE 132 */ + EINVAL, /* ERROR_IS_JOIN_TARGET 133 */ + EINVAL, /* ERROR_IS_JOINED 134 */ + EINVAL, /* ERROR_IS_SUBSTED 135 */ + EINVAL, /* ERROR_NOT_JOINED 136 */ + EINVAL, /* ERROR_NOT_SUBSTED 137 */ + EINVAL, /* ERROR_JOIN_TO_JOIN 138 */ + EINVAL, /* ERROR_SUBST_TO_SUBST 139 */ + EINVAL, /* ERROR_JOIN_TO_SUBST 140 */ + EINVAL, /* ERROR_SUBST_TO_JOIN 141 */ + EINVAL, /* ERROR_BUSY_DRIVE 142 */ + EINVAL, /* ERROR_SAME_DRIVE 143 */ + EINVAL, /* ERROR_DIR_NOT_ROOT 144 */ + ENOTEMPTY, /* ERROR_DIR_NOT_EMPTY 145 */ + EINVAL, /* ERROR_IS_SUBST_PATH 146 */ + EINVAL, /* ERROR_IS_JOIN_PATH 147 */ + EINVAL, /* ERROR_PATH_BUSY 148 */ + EINVAL, /* ERROR_IS_SUBST_TARGET 149 */ + EINVAL, /* ERROR_SYSTEM_TRACE 150 */ + EINVAL, /* ERROR_INVALID_EVENT_COUNT 151 */ + EINVAL, /* ERROR_TOO_MANY_MUXWAITERS 152 */ + EINVAL, /* ERROR_INVALID_LIST_FORMAT 153 */ + EINVAL, /* ERROR_LABEL_TOO_LONG 154 */ + EINVAL, /* ERROR_TOO_MANY_TCBS 155 */ + EINVAL, /* ERROR_SIGNAL_REFUSED 156 */ + EINVAL, /* ERROR_DISCARDED 157 */ + EACCES, /* ERROR_NOT_LOCKED 158 */ + EINVAL, /* ERROR_BAD_THREADID_ADDR 159 */ + EINVAL, /* ERROR_BAD_ARGUMENTS 160 */ + ENOENT, /* ERROR_BAD_PATHNAME 161 */ + EINVAL, /* ERROR_SIGNAL_PENDING 162 */ + EINVAL, /* 163 */ + EAGAIN, /* ERROR_MAX_THRDS_REACHED 164 */ + EINVAL, /* 165 */ + EINVAL, /* 166 */ + EACCES, /* ERROR_LOCK_FAILED 167 */ + EINVAL, /* 168 */ + EINVAL, /* 169 */ + EINVAL, /* ERROR_BUSY 170 */ + EINVAL, /* 171 */ + EINVAL, /* 172 */ + EINVAL, /* ERROR_CANCEL_VIOLATION 173 */ + EINVAL, /* ERROR_ATOMIC_LOCKS_NOT_SUPPORTED 174 */ + EINVAL, /* 175 */ + EINVAL, /* 176 */ + EINVAL, /* 177 */ + EINVAL, /* 178 */ + EINVAL, /* 179 */ + EINVAL, /* ERROR_INVALID_SEGMENT_NUMBER 180 */ + EINVAL, /* 181 */ + EINVAL, /* ERROR_INVALID_ORDINAL 182 */ + EEXIST, /* ERROR_ALREADY_EXISTS 183 */ + EINVAL, /* 184 */ + EINVAL, /* 185 */ + EINVAL, /* ERROR_INVALID_FLAG_NUMBER 186 */ + EINVAL, /* ERROR_SEM_NOT_FOUND 187 */ + ENOEXEC, /* ERROR_INVALID_STARTING_CODESEG 188 */ + ENOEXEC, /* ERROR_INVALID_STACKSEG 189 */ + ENOEXEC, /* ERROR_INVALID_MODULETYPE 190 */ + ENOEXEC, /* ERROR_INVALID_EXE_SIGNATURE 191 */ + ENOEXEC, /* ERROR_EXE_MARKED_INVALID 192 */ + ENOEXEC, /* ERROR_BAD_EXE_FORMAT 193 */ + ENOEXEC, /* ERROR_ITERATED_DATA_EXCEEDS_64k 194 */ + ENOEXEC, /* ERROR_INVALID_MINALLOCSIZE 195 */ + ENOEXEC, /* ERROR_DYNLINK_FROM_INVALID_RING 196 */ + ENOEXEC, /* ERROR_IOPL_NOT_ENABLED 197 */ + ENOEXEC, /* ERROR_INVALID_SEGDPL 198 */ + ENOEXEC, /* ERROR_AUTODATASEG_EXCEEDS_64k 199 */ + ENOEXEC, /* ERROR_RING2SEG_MUST_BE_MOVABLE 200 */ + ENOEXEC, /* ERROR_RELOC_CHAIN_XEEDS_SEGLIM 201 */ + ENOEXEC, /* ERROR_INFLOOP_IN_RELOC_CHAIN 202 */ + EINVAL, /* ERROR_ENVVAR_NOT_FOUND 203 */ + EINVAL, /* 204 */ + EINVAL, /* ERROR_NO_SIGNAL_SENT 205 */ + ENOENT, /* ERROR_FILENAME_EXCED_RANGE 206 */ + EINVAL, /* ERROR_RING2_STACK_IN_USE 207 */ + EINVAL, /* ERROR_META_EXPANSION_TOO_LONG 208 */ + EINVAL, /* ERROR_INVALID_SIGNAL_NUMBER 209 */ + EINVAL, /* ERROR_THREAD_1_INACTIVE 210 */ + EINVAL, /* 211 */ + EINVAL, /* ERROR_LOCKED 212 */ + EINVAL, /* 213 */ + EINVAL, /* ERROR_TOO_MANY_MODULES 214 */ + EAGAIN /* ERROR_NESTING_NOT_ALLOWED 215 */ +}; + +/* size of the table */ +#define ERRMAPTABLESIZE (sizeof(errMapTable)/sizeof(errMapTable[0])) + +/* +** void __cdecl _dosmaperr(winerrno) +** +** Takes a Windows error number and tries to map it to a Unix System V errno. +** Sets: +** _doserrno = Windows error number +** errno = Unix System V errno. +*/ +void __cdecl _dosmaperr(unsigned long winerrno) +{ + _doserrno = winerrno; + + if (winerrno >= ERRMAPTABLESIZE) { + if (winerrno == ERROR_NOT_ENOUGH_QUOTA) { /* exception for 1816 */ + errno = ENOMEM; + } else { + errno = EINVAL; + } + } else { + errno = (unsigned int) errMapTable[winerrno]; + } +} + +#endif /* WIN32 && _MT && _DLL */ + diff --git a/erts/emulator/sys/win32/driver_int.h b/erts/emulator/sys/win32/driver_int.h new file mode 100644 index 0000000000..97e188816e --- /dev/null +++ b/erts/emulator/sys/win32/driver_int.h @@ -0,0 +1,39 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 1997-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ +/*---------------------------------------------------------------------- +** Purpose : System dependant driver declarations +**---------------------------------------------------------------------- */ + +#ifndef __DRIVER_INT_H__ +#define __DRIVER_INT_H__ + +#if !defined __WIN32__ +# define __WIN32__ +#endif + +/* + * This structure can be cast to a WSABUF structure. + */ + +typedef struct _SysIOVec { + unsigned long iov_len; + char* iov_base; +} SysIOVec; + +#endif diff --git a/erts/emulator/sys/win32/erl.def b/erts/emulator/sys/win32/erl.def new file mode 100644 index 0000000000..59e940847d --- /dev/null +++ b/erts/emulator/sys/win32/erl.def @@ -0,0 +1,4 @@ +EXPORTS + erl_start + sys_get_key + sys_primitive_init diff --git a/erts/emulator/sys/win32/erl_main.c b/erts/emulator/sys/win32/erl_main.c new file mode 100644 index 0000000000..5471bffb52 --- /dev/null +++ b/erts/emulator/sys/win32/erl_main.c @@ -0,0 +1,29 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2000-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif +#include "sys.h" +#include "global.h" + +void +main(int argc, char **argv) +{ + erl_start(argc, argv); +} diff --git a/erts/emulator/sys/win32/erl_poll.c b/erts/emulator/sys/win32/erl_poll.c new file mode 100644 index 0000000000..d816cc2c07 --- /dev/null +++ b/erts/emulator/sys/win32/erl_poll.c @@ -0,0 +1,1361 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2007-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#define WANT_NONBLOCKING + +#include "sys.h" +#include "erl_alloc.h" +#include "erl_poll.h" + +/* + * Some debug macros + */ + +/*#define HARDDEBUG */ +#ifdef HARDDEBUG +#ifdef HARDTRACE +#define HARDTRACEF(X) my_debug_printf##X +#else +#define HARDTRACEF(X) +#endif + +#define HARDDEBUGF(X) my_debug_printf##X +static void my_debug_printf(char *fmt, ...) +{ + char buffer[1024]; + va_list args; + + va_start(args, fmt); + erts_vsnprintf(buffer,1024,fmt,args); + va_end(args); + erts_fprintf(stderr,"%s\r\n",buffer); +} +#else +#define HARDTRACEF(X) +#define HARDDEBUGF(X) +#endif + +#ifdef DEBUG +#define NoMansLandFill 0xFD /* fill no-man's land with this */ +#define DeadLandFill 0xDD /* fill free objects with this */ +#define CleanLandFill 0xCD /* fill new objects with this */ + +static void consistency_check(struct _Waiter* w); +static void* debug_alloc(ErtsAlcType_t, Uint); +static void* debug_realloc(ErtsAlcType_t, void *, Uint, Uint); + +# define SEL_ALLOC debug_alloc +# define SEL_REALLOC debug_realloc +# define SEL_FREE erts_free + +static void *debug_alloc(ErtsAlcType_t type, Uint size) +{ + void* p = erts_alloc(type, size); + memset(p, CleanLandFill, size); + return p; +} + +static void *debug_realloc(ErtsAlcType_t type, void *ptr, Uint prev_size, + Uint size) +{ + void *p; + size_t fill_size; + void *fill_ptr; + + if (prev_size > size) { + size_t fill_size = (size_t) (prev_size - size); + void *fill_ptr = (void *) (((char *) ptr) + size); + memset(fill_ptr, NoMansLandFill, fill_size); + } + + p = erts_realloc(type, ptr, size); + + if (size > prev_size) { + size_t fill_size = (size_t) (size - prev_size); + void *fill_ptr = (void *) (((char *) p) + prev_size); + memset(fill_ptr, CleanLandFill, fill_size); + } + + return p; +} +#else +# define SEL_ALLOC erts_alloc +# define SEL_REALLOC realloc_wrap +# define SEL_FREE erts_free + +static ERTS_INLINE void * +realloc_wrap(ErtsAlcType_t t, void *p, Uint ps, Uint s) +{ + return erts_realloc(t, p, s); +} +#endif + + +#ifdef HARD_POLL_DEBUG +#define OP_SELECT 1 +#define OP_DESELECT 2 +#define OP_FIRED 3 +#define OP_READ_BEGIN 4 +#define OP_READ_DONE 5 +#define OP_WRITE_BEGIN 6 +#define OP_WRITE_DONE 7 +#define OP_REPORTED 8 +#define OP_DIED 9 +#define OP_ASYNC_INIT 10 +#define OP_ASYNC_IMMED 11 +#define OP_FD_MOVED 12 + +static struct { + int op; + ErtsSysFdType active; + int xdata; +} debug_save_ops[1024]; + +static int num_debug_save_ops = 0; + +static ErtsSysFdType active_debug_fd; +static int active_debug_fd_set = 0; + +static erts_mtx_t save_ops_mtx; + +static void poll_debug_init(void) +{ + erts_mtx_init(&save_ops_mtx, "save_ops_lock"); +} + +void poll_debug_set_active_fd(ErtsSysFdType fd) +{ + erts_mtx_lock(&save_ops_mtx); + active_debug_fd_set = 1; + active_debug_fd = fd; + erts_mtx_unlock(&save_ops_mtx); +} + +static void do_save_op(ErtsSysFdType fd, int op, int xdata) +{ + erts_mtx_lock(&save_ops_mtx); + if (fd == active_debug_fd && num_debug_save_ops < 1024) { + int x = num_debug_save_ops++; + debug_save_ops[x].op = op; + debug_save_ops[x].active = fd; + debug_save_ops[x].xdata = xdata; + } + erts_mtx_unlock(&save_ops_mtx); +} + +void poll_debug_moved(ErtsSysFdType fd, int s1, int s2) +{ + do_save_op(fd,OP_FD_MOVED,s1 | (s2 << 16)); +} + +void poll_debug_select(ErtsSysFdType fd, int mode) +{ + do_save_op(fd,OP_SELECT,mode); +} + +void poll_debug_deselect(ErtsSysFdType fd) +{ + do_save_op(fd,OP_DESELECT,0); +} + +void poll_debug_fired(ErtsSysFdType fd) +{ + do_save_op(fd,OP_FIRED,0); +} + +void poll_debug_read_begin(ErtsSysFdType fd) +{ + do_save_op(fd,OP_READ_BEGIN,0); +} + +void poll_debug_read_done(ErtsSysFdType fd, int bytes) +{ + do_save_op(fd,OP_READ_DONE,bytes); +} + +void poll_debug_async_initialized(ErtsSysFdType fd) +{ + do_save_op(fd,OP_ASYNC_INIT,0); +} + +void poll_debug_async_immediate(ErtsSysFdType fd, int bytes) +{ + do_save_op(fd,OP_ASYNC_IMMED,bytes); +} + +void poll_debug_write_begin(ErtsSysFdType fd) +{ + do_save_op(fd,OP_WRITE_BEGIN,0); +} + +void poll_debug_write_done(ErtsSysFdType fd, int bytes) +{ + do_save_op(fd,OP_WRITE_DONE,bytes); +} + +void poll_debug_reported(ErtsSysFdType fd, int mode) +{ + do_save_op(fd,OP_REPORTED,mode); +} + +void poll_debug_died(ErtsSysFdType fd) +{ + do_save_op(fd,OP_DIED,0); +} + +#endif /* DEBUG */ + +/* + * End of debug macros + */ + + + +/* + * Handles that we poll, but that are actually signalled from outside + * this module + */ + +extern HANDLE erts_service_event; +extern HANDLE erts_sys_break_event; + + +/* + * The structure we hold for each event (i.e. fd) + */ +typedef struct _EventData { + HANDLE event; /* For convenience. */ + ErtsPollEvents mode; /* The current select mode. */ + struct _EventData *next; /* Next in free or delete lists. */ +} EventData; + +/* + * The structure to represent a waiter thread + */ +typedef struct _Waiter { + HANDLE events[MAXIMUM_WAIT_OBJECTS]; /* The events. */ + EventData* evdata[MAXIMUM_WAIT_OBJECTS]; /* Pointers to associated data. */ + int active_events; /* Number of events to wait for */ + int total_events; /* Total number of events in the arrays. */ + int highwater; /* Events processed up to here */ + EventData evdata_heap[MAXIMUM_WAIT_OBJECTS]; /* Pre-allocated EventDatas */ + EventData* first_free_evdata; /* Index of first free EventData object. */ + HANDLE go_ahead; /* The waiter may continue. (Auto-reset) */ + void *xdata; /* used when thread parameter */ + erts_tid_t this; /* Thread "handle" of this waiter */ + erts_mtx_t mtx; /* Mutex for updating/reading pollset, but the + currently used set require thread stopping + to be updated */ +} Waiter; + +/* + * The structure for a pollset. There can currently be only one... + */ +struct ErtsPollSet_ { + Waiter** waiter; + int allocated_waiters; /* Size ow waiter array */ + int num_waiters; /* Number of waiter threads. */ + erts_atomic_t sys_io_ready; /* Tells us there is I/O ready (already). */ + int restore_events; /* Tells us to restore waiters events + next time around */ + HANDLE event_io_ready; /* To be used when waiting for io */ + /* These are used to wait for workers to enter standby */ + volatile int standby_wait_counter; /* Number of threads to wait for */ + CRITICAL_SECTION standby_crit; /* CS to guard the counter */ + HANDLE standby_wait_event; /* Event signalled when counte == 0 */ +#ifdef ERTS_SMP + erts_smp_atomic_t woken; + erts_smp_mtx_t mtx; + erts_smp_atomic_t interrupt; +#endif + erts_smp_atomic_t timeout; +}; + +#ifdef ERTS_SMP + +#define ERTS_POLLSET_LOCK(PS) \ + erts_smp_mtx_lock(&(PS)->mtx) +#define ERTS_POLLSET_UNLOCK(PS) \ + erts_smp_mtx_unlock(&(PS)->mtx) +#define ERTS_POLLSET_SET_POLLED_CHK(PS) \ + ((int) erts_smp_atomic_xchg(&(PS)->polled, (long) 1)) +#define ERTS_POLLSET_SET_POLLED(PS) \ + erts_smp_atomic_set(&(PS)->polled, (long) 1) +#define ERTS_POLLSET_UNSET_POLLED(PS) \ + erts_smp_atomic_set(&(PS)->polled, (long) 0) +#define ERTS_POLLSET_IS_POLLED(PS) \ + ((int) erts_smp_atomic_read(&(PS)->polled)) +#define ERTS_POLLSET_SET_POLLER_WOKEN_CHK(PS) \ + ((int) erts_smp_atomic_xchg(&(PS)->woken, (long) 1)) +#define ERTS_POLLSET_SET_POLLER_WOKEN(PS) \ + erts_smp_atomic_set(&(PS)->woken, (long) 1) +#define ERTS_POLLSET_UNSET_POLLER_WOKEN(PS) \ + erts_smp_atomic_set(&(PS)->woken, (long) 0) +#define ERTS_POLLSET_IS_POLLER_WOKEN(PS) \ + ((int) erts_smp_atomic_read(&(PS)->woken)) + +#define ERTS_POLLSET_UNSET_INTERRUPTED_CHK(PS) \ + ((int) erts_smp_atomic_xchg(&(PS)->interrupt, (long) 0)) +#define ERTS_POLLSET_UNSET_INTERRUPTED(PS) \ + erts_smp_atomic_set(&(PS)->interrupt, (long) 0) +#define ERTS_POLLSET_SET_INTERRUPTED(PS) \ + erts_smp_atomic_set(&(PS)->interrupt, (long) 1) +#define ERTS_POLLSET_IS_INTERRUPTED(PS) \ + ((int) erts_smp_atomic_read(&(PS)->interrupt)) + +#else + +#define ERTS_POLLSET_LOCK(PS) +#define ERTS_POLLSET_UNLOCK(PS) +#define ERTS_POLLSET_SET_POLLED_CHK(PS) 0 +#define ERTS_POLLSET_UNSET_POLLED(PS) +#define ERTS_POLLSET_IS_POLLED(PS) 0 +#define ERTS_POLLSET_SET_POLLER_WOKEN_CHK(PS) 1 +#define ERTS_POLLSET_SET_POLLER_WOKEN(PS) +#define ERTS_POLLSET_UNSET_POLLER_WOKEN(PS) +#define ERTS_POLLSET_IS_POLLER_WOKEN(PS) 1 + + +#endif + +/* + * While atomics are not yet implemented for windows in the common library... + * + * MSDN doc states that SMP machines and old compilers require + * InterLockedExchange to properly read and write interlocked + * variables, otherwise the processors might reschedule + * the access and order of atomics access is destroyed... + * While they only mention it in white-papers, the problem + * in VS2003 is due to the IA64 arch, so we can still count + * on the CPU not rescheduling the access to volatile in X86 arch using + * even the slightly older compiler... + * + * So here's (hopefully) a subset of the generally working atomic + * variable access... + */ + +#if defined(__GNUC__) +# if defined(__i386__) || defined(__x86_64__) +# define VOLATILE_IN_SEQUENCE 1 +# else +# define VOLATILE_IN_SEQUENCE 0 +# endif +#elif defined(_MSC_VER) +# if _MSC_VER < 1300 +# define VOLATILE_IN_SEQUENCE 0 /* Dont trust really old compilers */ +# else +# if defined(_M_IX86) +# define VOLATILE_IN_SEQUENCE 1 +# else /* I.e. IA64 */ +# if _MSC_VER >= 1400 +# define VOLATILE_IN_SEQUENCE 1 +# else +# define VOLATILE_IN_SEQUENCE 0 +# endif +# endif +# endif +#else +# define VOLATILE_IN_SEQUENCE 0 +#endif + + + +/* + * Communication with sys_interrupt + */ + +#ifdef ERTS_SMP +extern erts_smp_atomic_t erts_break_requested; +#define ERTS_SET_BREAK_REQUESTED \ + erts_smp_atomic_set(&erts_break_requested, (long) 1) +#define ERTS_UNSET_BREAK_REQUESTED \ + erts_smp_atomic_set(&erts_break_requested, (long) 0) +#else +extern volatile int erts_break_requested; +#define ERTS_SET_BREAK_REQUESTED (erts_break_requested = 1) +#define ERTS_UNSET_BREAK_REQUESTED (erts_break_requested = 0) +#endif + +static erts_mtx_t break_waiter_lock; +static HANDLE break_happened_event; +static erts_atomic_t break_waiter_state; +#define BREAK_WAITER_GOT_BREAK 1 +#define BREAK_WAITER_GOT_HALT 2 + + +/* + * Forward declarations + */ + +static void *threaded_waiter(void *param); +static void *break_waiter(void *param); + +/* + * Sychronization macros and functions + */ +#define START_WAITER(PS, w) \ + SetEvent((w)->go_ahead) + +#define STOP_WAITER(PS,w) \ +do { \ + setup_standby_wait((PS),1); \ + SetEvent((w)->events[0]); \ + wait_standby(PS); \ +} while(0) + +#define START_WAITERS(PS) \ +do { \ + int i; \ + for (i = 0; i < (PS)->num_waiters; i++) { \ + SetEvent((PS)->waiter[i]->go_ahead); \ + } \ + } while(0) + +#define STOP_WAITERS(PS) \ +do { \ + int i; \ + setup_standby_wait((PS),(PS)->num_waiters); \ + for (i = 0; i < (PS)->num_waiters; i++) { \ + SetEvent((PS)->waiter[i]->events[0]); \ + } \ + wait_standby(PS); \ + } while(0) + +#if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT && !defined(ERTS_SMP) + +static ERTS_INLINE int +unset_interrupted_chk(ErtsPollSet ps) +{ + /* This operation isn't atomic, but we have no need at all for an + atomic operation here... */ + int res = ps->interrupt; + ps->interrupt = 0; + return res; +} + +#endif + +#ifdef ERTS_SMP +static ERTS_INLINE void +wake_poller(ErtsPollSet ps) +{ + if (!ERTS_POLLSET_SET_POLLER_WOKEN_CHK(ps)) { + SetEvent(ps->event_io_ready); + } +} +#endif + +static void setup_standby_wait(ErtsPollSet ps, int num_threads) +{ + EnterCriticalSection(&(ps->standby_crit)); + ps->standby_wait_counter = num_threads; + ResetEvent(ps->standby_wait_event); + LeaveCriticalSection(&(ps->standby_crit)); +} + +static void signal_standby(ErtsPollSet ps) +{ + EnterCriticalSection(&(ps->standby_crit)); + --(ps->standby_wait_counter); + if (ps->standby_wait_counter < 0) { + LeaveCriticalSection(&(ps->standby_crit)); + erl_exit(1,"Standby signalled by more threads than expected"); + } + if (!(ps->standby_wait_counter)) { + SetEvent(ps->standby_wait_event); + } + LeaveCriticalSection(&(ps->standby_crit)); +} + +static void wait_standby(ErtsPollSet ps) +{ + WaitForSingleObject(ps->standby_wait_event,INFINITE); +} + +static void remove_event_from_set(Waiter *w, int j) +{ + w->evdata[j]->event = INVALID_HANDLE_VALUE; + w->evdata[j]->mode = 0; + w->evdata[j]->next = w->first_free_evdata; + w->first_free_evdata = w->evdata[j]; + + /* + * If the event is active, we will overwrite it + * with the last active event and make the hole + * the first non-active event. + */ + + if (j < w->active_events) { + w->active_events--; + w->highwater--; + w->total_events--; + w->events[j] = w->events[w->active_events]; + w->evdata[j] = w->evdata[w->active_events]; + w->events[w->active_events] = w->events[w->highwater]; + w->evdata[w->active_events] = w->evdata[w->highwater]; + w->events[w->highwater] = w->events[w->total_events]; + w->evdata[w->highwater] = w->evdata[w->total_events]; + } else if (j < w->highwater) { + w->highwater--; + w->total_events--; + w->events[j] = w->events[w->highwater]; + w->evdata[j] = w->evdata[w->highwater]; + w->events[w->highwater] = w->events[w->total_events]; + w->evdata[w->highwater] = w->evdata[w->total_events]; + } else { + w->total_events--; + w->events[j] = w->events[w->total_events]; + w->evdata[j] = w->evdata[w->total_events]; + } + +#ifdef DEBUG + w->events[w->total_events] = (HANDLE) CleanLandFill; + w->evdata[w->total_events] = (EventData *) CleanLandFill; + consistency_check(w); +#endif +} + +/* + * Thread handling + */ + +#ifdef DEBUG +static void consistency_check(Waiter* w) +{ + int i; + + ASSERT(w->active_events <= w->total_events); + ASSERT(w->evdata[0] == NULL); + + for (i = 1; i < w->total_events; i++) { + ASSERT(w->events[i] == w->evdata[i]->event); + ASSERT(w->evdata[i]->mode != 0); + } +} + +#endif + +static void new_waiter(ErtsPollSet ps) +{ + register Waiter* w; + DWORD tid; /* Id for thread. */ + erts_tid_t thread; + int i; + int tres; + + if (ps->num_waiters == ps->allocated_waiters) { + Uint old_size = sizeof(Waiter *)*ps->allocated_waiters; + ps->allocated_waiters += 64; + ps->waiter = SEL_REALLOC(ERTS_ALC_T_WAITER_OBJ, + (void *) ps->waiter, + old_size, + sizeof(Waiter *) * (ps->allocated_waiters)); + } + + w = (Waiter *) SEL_ALLOC(ERTS_ALC_T_WAITER_OBJ, sizeof(Waiter)); + ps->waiter[ps->num_waiters] = w; + + w->events[0] = CreateAutoEvent(FALSE); + w->evdata[0] = NULL; /* Should never be used. */ + w->active_events = 1; + w->highwater = 1; + w->total_events = 1; + erts_mtx_init(&w->mtx, "pollwaiter"); + + + /* + * Form the free list of EventData objects. + */ + + w->evdata_heap[0].next = 0; /* Last in free list. */ + for (i = 1; i < MAXIMUM_WAIT_OBJECTS; i++) { + w->evdata_heap[i].next = w->evdata_heap+i-1; + } + w->first_free_evdata = w->evdata_heap+MAXIMUM_WAIT_OBJECTS-1; + + /* + * Create the other events. + */ + + w->go_ahead = CreateAutoEvent(FALSE); + + /* + * Create the thread. + */ + w->xdata = ps; + erts_thr_create(&thread, &threaded_waiter, w, NULL); + w->this = thread; + + /* + * Finally, done. + */ + + (ps->num_waiters)++; +} + +static void *break_waiter(void *param) +{ + HANDLE harr[2]; + int i = 0; + harr[i++] = erts_sys_break_event; + if (erts_service_event != NULL) { + harr[i++] = erts_service_event; + } + + for(;;) { + switch (WaitForMultipleObjects(i,harr,FALSE,INFINITE)) { + case WAIT_OBJECT_0: + ResetEvent(harr[0]); + erts_mtx_lock(&break_waiter_lock); + erts_atomic_set(&break_waiter_state,BREAK_WAITER_GOT_BREAK); + SetEvent(break_happened_event); + erts_mtx_unlock(&break_waiter_lock); + break; + case (WAIT_OBJECT_0+1): + ResetEvent(harr[1]); + erts_mtx_lock(&break_waiter_lock); + erts_atomic_set(&break_waiter_state,BREAK_WAITER_GOT_HALT); + SetEvent(break_happened_event); + erts_mtx_unlock(&break_waiter_lock); + break; + default: + erl_exit(1,"Unexpected event in break_waiter"); + } + } +} + +static void *threaded_waiter(void *param) +{ + register Waiter* w = (Waiter *) param; + ErtsPollSet ps = (ErtsPollSet) w->xdata; +#ifdef HARD_POLL_DEBUG2 + HANDLE oold_fired[64]; + int num_oold_fired; + HANDLE old_fired[64]; + int num_old_fired = 0; + HANDLE fired[64]; + int num_fired = 0; + HANDLE errors[1024]; + int num_errors = 0; + HANDLE save_events[64]; + int save_active_events; + int save_total_events; + int save_highwater; +#endif + + again: + WaitForSingleObject(w->go_ahead, INFINITE); + /* Atomic enough when just checking, skip lock */ + if (w->total_events == 0) { + return NULL; + } + if (w->active_events == 0) { + goto again; + } + ASSERT(w->evdata[0] == NULL); +#ifdef HARD_POLL_DEBUG2 + num_oold_fired = num_old_fired; + memcpy(oold_fired,old_fired,num_old_fired*sizeof(HANDLE)); + num_old_fired = num_fired; + memcpy(old_fired,fired,num_fired*sizeof(HANDLE)); + num_fired = 0; +#endif + for (;;) { + int i; + int j; +#ifdef HARD_POLL_DEBUG2 + erts_mtx_lock(&w->mtx); + memcpy(save_events,w->events,w->active_events*sizeof(HANDLE)); + save_active_events = w->active_events; + save_total_events = w->total_events; + save_highwater = w->highwater; + erts_mtx_unlock(&w->mtx); +#endif + i = WaitForMultipleObjects(w->active_events, w->events, FALSE, INFINITE); + switch (i) { + case WAIT_FAILED: + DEBUGF(("Wait failed: %s\n", last_error())); + erts_mtx_lock(&w->mtx); + /* Dont wait for our signal event */ + for (j = 1; j < w->active_events; j++) { + int tmp; + if ((tmp = WaitForSingleObject(w->events[j], 0)) + == WAIT_FAILED) { + DEBUGF(("Invalid handle: i = %d, handle = 0x%0x\n", + j, w->events[j])); +#ifdef HARD_POLL_DEBUG2 + if (num_errors < 1024) + errors[num_errors++] = w->events[j]; +#endif +#ifdef HARD_POLL_DEBUG + poll_debug_died(w->events[j]); +#endif + remove_event_from_set(w,j); +#ifdef DEBUG + consistency_check(w); +#endif + } else if (tmp == WAIT_OBJECT_0) { + i = WAIT_OBJECT_0 + j; + goto event_happened; + } + } + erts_mtx_unlock(&w->mtx); + break; + case WAIT_OBJECT_0: + signal_standby(ps); + goto again; +#ifdef DEBUG + case WAIT_TIMEOUT: + ASSERT(0); +#endif + default: + erts_mtx_lock(&w->mtx); +#ifdef HARD_POLL_DEBUG2 + { + int x = memcmp(save_events,w->events,w->active_events*sizeof(HANDLE)); + ASSERT(x == 0 && save_active_events == w->active_events); + } +#endif +event_happened: +#ifdef DEBUG + consistency_check(w); +#endif + ASSERT(WAIT_OBJECT_0 < i && i < WAIT_OBJECT_0+w->active_events); + if (!erts_atomic_xchg(&ps->sys_io_ready,1)) { + HARDDEBUGF(("SET EventIoReady (%d)",erts_atomic_read(&ps->sys_io_ready))); + SetEvent(ps->event_io_ready); + } else { + HARDDEBUGF(("DONT SET EventIoReady")); + } + + /* + * The main thread wont start working on our arrays untill we're + * stopped, so we can work in peace although the main thread runs + */ + ASSERT(i >= WAIT_OBJECT_0+1); + i -= WAIT_OBJECT_0; + ASSERT(i >= 1); + w->active_events--; + HARDDEBUGF(("i = %d, a,h,t = %d,%d,%d",i, + w->active_events, w->highwater, w->total_events)); +#ifdef HARD_POLL_DEBUG2 + fired[num_fired++] = w->events[i]; +#endif +#ifdef HARD_POLL_DEBUG + poll_debug_fired(w->events[i]); +#endif + if (i < w->active_events) { + HANDLE te = w->events[i]; + EventData* tp = w->evdata[i]; + w->events[i] = w->events[w->active_events]; + w->evdata[i] = w->evdata[w->active_events]; + w->events[w->active_events] = te; + w->evdata[w->active_events] = tp; + } + HARDDEBUGF(("i = %d, a,h,t = %d,%d,%d",i, + w->active_events, w->highwater, w->total_events)); +#ifdef DEBUG + consistency_check(w); +#endif + erts_mtx_unlock(&w->mtx); + break; + } + } +} + +/* + * The actual adding and removing from pollset utilities + */ + +static int set_driver_select(ErtsPollSet ps, HANDLE event, ErtsPollEvents mode) +{ + int i; + int best_waiter = -1; /* The waiter with lowest number of events. */ + int lowest = MAXIMUM_WAIT_OBJECTS; /* Lowest number of events + * in any waiter. + */ + EventData* ev; + Waiter* w; + + /* + * Find the waiter which is least busy. + */ + +#ifdef HARD_POLL_DEBUG + poll_debug_select(event, mode); +#endif + + /* total_events can no longer be read without the lock, it's changed in the waiter */ + for (i = 0; i < ps->num_waiters; i++) { + erts_mtx_lock(&(ps->waiter[i]->mtx)); + if (ps->waiter[i]->total_events < lowest) { + lowest = ps->waiter[i]->total_events; + best_waiter = i; + } + erts_mtx_unlock(&(ps->waiter[i]->mtx)); + } + + /* + * Stop the selected waiter, or start a new waiter if all were busy. + */ + + if (best_waiter >= 0) { + w = ps->waiter[best_waiter]; + STOP_WAITER(ps,w); + erts_mtx_lock(&w->mtx); + } else { + new_waiter(ps); + w = ps->waiter[(ps->num_waiters)-1]; + erts_mtx_lock(&w->mtx); + } + +#ifdef DEBUG + consistency_check(w); +#endif + + /* + * Allocate and initialize an EventData structure. + */ + + ev = w->first_free_evdata; + w->first_free_evdata = ev->next; + ev->event = event; + ev->mode = mode; + ev->next = NULL; + + /* + * At this point, the selected waiter (newly-created or not) is + * standing by. Put the new event into the active part of the array. + */ + + if (w->active_events < w->total_events) { + /* + * Move the first event beyond the active part of the array to + * the very end to make place for the new event. + */ + +#ifdef HARD_POLL_DEBUG + poll_debug_moved(w->events[w->highwater],w->highwater,w->total_events); +#endif + w->events[w->total_events] = w->events[w->highwater]; + w->evdata[w->total_events] = w->evdata[w->highwater]; +#ifdef HARD_POLL_DEBUG + poll_debug_moved(w->events[w->active_events],w->active_events,w->highwater); +#endif + w->events[w->highwater] = w->events[w->active_events]; + w->evdata[w->highwater] = w->evdata[w->active_events]; + + } + w->events[w->active_events] = event; + w->evdata[w->active_events] = ev; + w->active_events++; + w->highwater++; + w->total_events++; + +#ifdef DEBUG + consistency_check(w); +#endif + erts_mtx_unlock(&w->mtx); + START_WAITER(ps,w); + HARDDEBUGF(("add select %d %d %d %d",best_waiter, + w->active_events,w->highwater,w->total_events)); + return mode; +} + + +static int cancel_driver_select(ErtsPollSet ps, HANDLE event) +{ + int i; + + ASSERT(event != INVALID_HANDLE_VALUE); + restart: + for (i = 0; i < ps->num_waiters; i++) { + Waiter* w = ps->waiter[i]; + int j; + + erts_mtx_lock(&w->mtx); +#ifdef DEBUG + consistency_check(w); +#endif + for (j = 0; j < w->total_events; j++) { + if (w->events[j] == event) { + int stopped = 0; + /* + * Free the event's EventData structure. + */ + + if (j < w->active_events) { + HARDDEBUGF(("Stopped in remove select")); + stopped = 1; + erts_mtx_unlock(&w->mtx); + STOP_WAITER(ps,w); + erts_mtx_lock(&w->mtx); + if ( j >= w->active_events || w->events[j] != event) { + /* things happened while unlocked */ + START_WAITER(ps,w); + erts_mtx_unlock(&w->mtx); + goto restart; + } + } +#ifdef HARD_POLL_DEBUG + poll_debug_deselect(w->events[j]); +#endif + remove_event_from_set(w, j); + if (stopped) { + START_WAITER(ps,w); + } + HARDDEBUGF(("removed select %d,%d %d %d %d",i,j, + w->active_events,w->highwater,w->total_events)); + break; + } + } + erts_mtx_unlock(&w->mtx); + } + return 0; +} + +/* + * Interface functions + */ + +void erts_poll_interrupt(ErtsPollSet ps, int set /* bool */) +{ + HARDTRACEF(("In erts_poll_interrupt(%d)",set)); +#ifdef ERTS_SMP + if (set) { + ERTS_POLLSET_SET_INTERRUPTED(ps); + wake_poller(ps); + } + else { + ERTS_POLLSET_UNSET_INTERRUPTED(ps); + } +#endif + HARDTRACEF(("Out erts_poll_interrupt(%d)",set)); +} + +void erts_poll_interrupt_timed(ErtsPollSet ps, + int set /* bool */, + long msec) +{ + HARDTRACEF(("In erts_poll_interrupt_timed(%d,%ld)",set,msec)); +#ifdef ERTS_SMP + if (set) { + if (erts_smp_atomic_read(&ps->timeout) > msec) { + ERTS_POLLSET_SET_INTERRUPTED(ps); + wake_poller(ps); + } + } + else { + ERTS_POLLSET_UNSET_INTERRUPTED(ps); + } +#endif + HARDTRACEF(("Out erts_poll_interrupt_timed")); +} + + +/* + * Windows is special, there is actually only one event type, and + * the only difference between ERTS_POLL_EV_IN and ERTS_POLL_EV_OUT + * is which driver callback will eventually be called. + */ +static ErtsPollEvents do_poll_control(ErtsPollSet ps, + ErtsSysFdType fd, + ErtsPollEvents pe, + int on /* bool */) +{ + HANDLE event = (HANDLE) fd; + ErtsPollEvents mode; + ErtsPollEvents result; + ASSERT(event != INVALID_HANDLE_VALUE); + + if (on) { + if (pe & ERTS_POLL_EV_IN || !(pe & ERTS_POLL_EV_OUT )) { + mode = ERTS_POLL_EV_IN; + } else { + mode = ERTS_POLL_EV_OUT; /* ready output only in this case */ + } + result = set_driver_select(ps, event, mode); + } else { + result = cancel_driver_select(ps, event); + } + return result; +} + +ErtsPollEvents erts_poll_control(ErtsPollSet ps, + ErtsSysFdType fd, + ErtsPollEvents pe, + int on, + int* do_wake) /* In: Wake up polling thread */ + /* Out: Poller is woken */ +{ + ErtsPollEvents result; + HARDTRACEF(("In erts_poll_control(0x%08X, %u, %d)",(unsigned long) fd, (unsigned) pe, on)); + ERTS_POLLSET_LOCK(ps); + result=do_poll_control(ps,fd,pe,on); + ERTS_POLLSET_UNLOCK(ps); + *do_wake = 0; /* Never any need to wake polling threads on windows */ + HARDTRACEF(("Out erts_poll_control -> %u",(unsigned) result)); + return result; +} + +void erts_poll_controlv(ErtsPollSet ps, + ErtsPollControlEntry pcev[], + int len) +{ + int i; + int hshur = 0; + int do_wake = 0; + + HARDTRACEF(("In erts_poll_controlv(%d)",len)); + ERTS_POLLSET_LOCK(ps); + + for (i = 0; i < len; i++) { + pcev[i].events = do_poll_control(ps, + pcev[i].fd, + pcev[i].events, + pcev[i].on); + } + ERTS_POLLSET_LOCK(ps); + HARDTRACEF(("Out erts_poll_controlv")); +} + +int erts_poll_wait(ErtsPollSet ps, + ErtsPollResFd pr[], + int *len, + SysTimeval *utvp) +{ + SysTimeval *tvp = utvp; + SysTimeval itv; + int no_fds; + DWORD timeout; + EventData* ev; + int res = 0; + int num = 0; + int n; + int i; + int break_state; + + HARDTRACEF(("In erts_poll_wait")); + ERTS_POLLSET_LOCK(ps); + + if (!erts_atomic_read(&ps->sys_io_ready) && ps->restore_events) { + HARDDEBUGF(("Restore events: %d",ps->num_waiters)); + ps->restore_events = 0; + for (i = 0; i < ps->num_waiters; ++i) { + Waiter* w = ps->waiter[i]; + erts_mtx_lock(&w->mtx); + HARDDEBUGF(("Maybe reset %d %d %d %d",i, + w->active_events,w->highwater,w->total_events)); + if (w->active_events < w->total_events) { + erts_mtx_unlock(&w->mtx); + STOP_WAITER(ps,w); + HARDDEBUGF(("Need reset %d %d %d %d",i, + w->active_events,w->highwater,w->total_events)); + erts_mtx_lock(&w->mtx); + /* Need reset, just check that it doesn't have got more to tell */ + if (w->highwater != w->active_events) { + HARDDEBUGF(("Oups!")); + /* Oups, got signalled before we took the lock, can't reset */ + if(erts_atomic_read(&ps->sys_io_ready) == 0) { + erl_exit(1,"Internal error: " + "Inconsistent io structures in erl_poll.\n"); + } + START_WAITER(ps,w); + erts_mtx_unlock(&w->mtx); + ps->restore_events = 1; + continue; + } + w->active_events = w->highwater = w->total_events; + START_WAITER(ps,w); + erts_mtx_unlock(&w->mtx); + } else { + erts_mtx_unlock(&w->mtx); + } + } + } + + no_fds = *len; + +#ifdef ERTS_POLL_MAX_RES + if (no_fds >= ERTS_POLL_MAX_RES) + no_fds = ERTS_POLL_MAX_RES; +#endif + + + ResetEvent(ps->event_io_ready); + ERTS_POLLSET_UNSET_POLLER_WOKEN(ps); + +#ifdef ERTS_SMP + if (ERTS_POLLSET_IS_INTERRUPTED(ps)) { + /* Interrupt use zero timeout */ + itv.tv_sec = 0; + itv.tv_usec = 0; + tvp = &itv; + } +#endif + + timeout = tvp->tv_sec * 1000 + tvp->tv_usec / 1000; + /*HARDDEBUGF(("timeout = %ld",(long) timeout));*/ + erts_smp_atomic_set(&ps->timeout, timeout); + + if (timeout > 0 && ! erts_atomic_read(&ps->sys_io_ready) && ! erts_atomic_read(&break_waiter_state)) { + HANDLE harr[2] = {ps->event_io_ready, break_happened_event}; + int num_h = 2; + + HARDDEBUGF(("Start waiting %d [%d]",num_h, (long) timeout)); + ERTS_POLLSET_UNLOCK(ps); + WaitForMultipleObjects(num_h, harr, FALSE, timeout); + ERTS_POLLSET_LOCK(ps); + HARDDEBUGF(("Stop waiting %d [%d]",num_h, (long) timeout)); + } + + ERTS_UNSET_BREAK_REQUESTED; + if(erts_atomic_read(&break_waiter_state)) { + erts_mtx_lock(&break_waiter_lock); + break_state = erts_atomic_read(&break_waiter_state); + erts_atomic_set(&break_waiter_state,0); + ResetEvent(break_happened_event); + erts_mtx_unlock(&break_waiter_lock); + switch (break_state) { + case BREAK_WAITER_GOT_BREAK: + ERTS_SET_BREAK_REQUESTED; + break; + case BREAK_WAITER_GOT_HALT: + erl_exit(0,""); + break; + default: + break; + } + } + + ERTS_POLLSET_SET_POLLER_WOKEN(ps); + + if (!erts_atomic_read(&ps->sys_io_ready)) { + res = EINTR; + HARDDEBUGF(("EINTR!")); + goto done; + } + + erts_atomic_set(&ps->sys_io_ready,0); + + n = ps->num_waiters; + + for (i = 0; i < n; i++) { + Waiter* w = ps->waiter[i]; + int j; + int first; + int last; + erts_mtx_lock(&w->mtx); +#ifdef DEBUG + consistency_check(w); +#endif + + first = w->active_events; + last = w->highwater; + w->highwater = w->active_events; + + for (j = last-1; j >= first; --j) { + if (num >= no_fds) { + w->highwater=j+1; + erts_mtx_unlock(&w->mtx); + /* This might mean we still have data to report, set + back the global flag! */ + erts_atomic_set(&ps->sys_io_ready,1); + HARDDEBUGF(("To many FD's to report!")); + goto done; + } + HARDDEBUGF(("SET! Restore events")); + ps->restore_events = 1; + HARDDEBUGF(("Report %d,%d",i,j)); + pr[num].fd = (ErtsSysFdType) w->events[j]; + pr[num].events = w->evdata[j]->mode; +#ifdef HARD_POLL_DEBUG + poll_debug_reported(w->events[j],w->highwater | (j << 16)); + poll_debug_reported(w->events[j],first | (last << 16)); +#endif + ++num; + } + +#ifdef DEBUG + consistency_check(w); +#endif + erts_mtx_unlock(&w->mtx); + } + done: + erts_smp_atomic_set(&ps->timeout, LONG_MAX); + *len = num; + ERTS_POLLSET_UNLOCK(ps); + HARDTRACEF(("Out erts_poll_wait")); + return res; + +} + +int erts_poll_max_fds(void) +{ + int res = sys_max_files(); + HARDTRACEF(("In/Out erts_poll_max_fds -> %d",res)); + return res; +} + +void erts_poll_info(ErtsPollSet ps, + ErtsPollInfo *pip) +{ + Uint size = 0; + Uint num_events = 0; + int i; + + HARDTRACEF(("In erts_poll_info")); + ERTS_POLLSET_LOCK(ps); + + size += sizeof(struct ErtsPollSet_); + size += sizeof(Waiter *) * ps->allocated_waiters; + for (i = 0; i < ps->num_waiters; ++i) { + Waiter *w = ps->waiter[i]; + if (w != NULL) { + size += sizeof(Waiter); + erts_mtx_lock(&w->mtx); + size += sizeof(EventData) * w->total_events; + num_events += (w->total_events - 1); /* First event is internal */ + erts_mtx_unlock(&w->mtx); + } + } + + pip->primary = "WaitForMultipleObjects"; + + pip->fallback = NULL; + + pip->kernel_poll = NULL; + + pip->memory_size = size; + + pip->poll_set_size = num_events; + + pip->fallback_poll_set_size = 0; + + pip->lazy_updates = 0; + + pip->pending_updates = 0; + + pip->batch_updates = 0; + + pip->concurrent_updates = 0; + ERTS_POLLSET_UNLOCK(ps); + + pip->max_fds = erts_poll_max_fds(); + HARDTRACEF(("Out erts_poll_info")); + +} + +ErtsPollSet erts_poll_create_pollset(void) +{ + ErtsPollSet ps = SEL_ALLOC(ERTS_ALC_T_POLLSET, + sizeof(struct ErtsPollSet_)); + HARDTRACEF(("In erts_poll_create_pollset")); + + ps->num_waiters = 0; + ps->allocated_waiters = 64; + ps->waiter = SEL_ALLOC(ERTS_ALC_T_WAITER_OBJ, + sizeof(Waiter *)*ps->allocated_waiters); + InitializeCriticalSection(&(ps->standby_crit)); + ps->standby_wait_counter = 0; + ps->event_io_ready = CreateManualEvent(FALSE); + ps->standby_wait_event = CreateManualEvent(FALSE); + erts_atomic_init(&ps->sys_io_ready,0); + ps->restore_events = 0; + +#ifdef ERTS_SMP + erts_smp_atomic_init(&ps->woken, 0); + erts_smp_mtx_init(&ps->mtx, "pollset"); + erts_smp_atomic_init(&ps->interrupt, 0); +#endif + erts_smp_atomic_init(&ps->timeout, LONG_MAX); + + HARDTRACEF(("Out erts_poll_create_pollset")); + return ps; +} + +void erts_poll_destroy_pollset(ErtsPollSet ps) +{ + int i; + HARDTRACEF(("In erts_poll_destroy_pollset")); + ERTS_POLLSET_LOCK(ps); + STOP_WAITERS(ps); + for (i=0;i<ps->num_waiters;++i) { + Waiter *w = ps->waiter[i]; + void *dummy; + erts_tid_t t = w->this; + /* Assume we're alone, no locking here... */ + w->active_events = w->total_events = w->highwater = 0; + START_WAITER(ps,w); + erts_thr_join(t,&dummy); + CloseHandle(w->go_ahead); + CloseHandle(w->events[0]); + erts_mtx_destroy(&w->mtx); + SEL_FREE(ERTS_ALC_T_WAITER_OBJ, (void *) w); + } + SEL_FREE(ERTS_ALC_T_WAITER_OBJ,ps->waiter); + CloseHandle(ps->event_io_ready); + CloseHandle(ps->standby_wait_event); + ERTS_POLLSET_UNLOCK(ps); +#ifdef ERTS_SMP + erts_smp_mtx_destroy(&ps->mtx); +#endif + SEL_FREE(ERTS_ALC_T_POLLSET, (void *) ps); + HARDTRACEF(("Out erts_poll_destroy_pollset")); +} + +/* + * Actually mostly initializes the friend module sys_interrupt... + */ +void erts_poll_init(void) +{ + erts_tid_t thread; + +#ifdef HARD_POLL_DEBUG + poll_debug_init(); +#endif + + HARDTRACEF(("In erts_poll_init")); + erts_sys_break_event = CreateManualEvent(FALSE); + + erts_mtx_init(&break_waiter_lock,"break_waiter_lock"); + break_happened_event = CreateManualEvent(FALSE); + erts_atomic_init(&break_waiter_state, 0); + + erts_thr_create(&thread, &break_waiter, NULL, NULL); + ERTS_UNSET_BREAK_REQUESTED; + HARDTRACEF(("Out erts_poll_init")); +} + +/* + * Non windows friendly interface, not used when fd's are not continous + */ +void erts_poll_get_selected_events(ErtsPollSet ps, + ErtsPollEvents ev[], + int len) +{ + int i; + HARDTRACEF(("In erts_poll_get_selected_events")); + for (i = 0; i < len; ++i) + ev[i] = 0; + HARDTRACEF(("Out erts_poll_get_selected_events")); +} diff --git a/erts/emulator/sys/win32/erl_win32_sys_ddll.c b/erts/emulator/sys/win32/erl_win32_sys_ddll.c new file mode 100644 index 0000000000..a19f49af10 --- /dev/null +++ b/erts/emulator/sys/win32/erl_win32_sys_ddll.c @@ -0,0 +1,206 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2006-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +/* + * Interface functions to the dynamic linker using dl* functions. + * (As far as I know it works on SunOS 4, 5, Linux and FreeBSD. /Seb) + */ + +#include <windows.h> + +#define GET_ERTS_ALC_TEST +#include "sys.h" +#include "global.h" +#include "erl_alloc.h" + +#include "erl_driver.h" +#include "erl_win_dyn_driver.h" + +#include "erl_nif.h" + +#define EXT_LEN 4 +#define FILE_EXT ".dll" + +static DWORD tls_index = 0; +static TWinDynDriverCallbacks wddc; +static TWinDynNifCallbacks nif_callbacks; + +void erl_sys_ddll_init(void) { + tls_index = TlsAlloc(); + ERL_INIT_CALLBACK_STRUCTURE(wddc); + +#define ERL_NIF_API_FUNC_DECL(RET,NAME,ARGS) nif_callbacks.NAME = NAME +#include "erl_nif_api_funcs.h" +#undef ERL_NIF_API_FUNC_DECL + + return; +} + +/* + * Open a shared object + */ +int erts_sys_ddll_open2(char *full_name, void **handle, ErtsSysDdllError* err) +{ + int len; + char dlname[MAXPATHLEN + 1]; + + if ((len = sys_strlen(full_name)) >= MAXPATHLEN - EXT_LEN) { + if (err != NULL) { + err->str = "Library name too long"; + } + return ERL_DE_LOAD_ERROR_NAME_TO_LONG; + } + sys_strcpy(dlname, full_name); + sys_strcpy(dlname+len, FILE_EXT); + return erts_sys_ddll_open_noext(dlname, handle, err); +} +int erts_sys_ddll_open_noext(char *dlname, void **handle, ErtsSysDdllError* err) +{ + HINSTANCE hinstance; + + if ((hinstance = LoadLibrary(dlname)) == NULL) { + int code = ERL_DE_DYNAMIC_ERROR_OFFSET - GetLastError(); + if (err != NULL) { + err->str = erts_sys_ddll_error(code); + } + return code; + } else { + *handle = (void *) hinstance; + return ERL_DE_NO_ERROR; + } +} + +/* + * Find a symbol in the shared object + */ +int erts_sys_ddll_sym2(void *handle, char *func_name, void **function, + ErtsSysDdllError* err) +{ + FARPROC proc; + if ((proc = GetProcAddress( (HINSTANCE) handle, func_name)) == NULL) { + int code = ERL_DE_DYNAMIC_ERROR_OFFSET - GetLastError(); + if (err != NULL) { + err->str = erts_sys_ddll_error(code); + } + return code; + } + *function = (void *) proc; + return ERL_DE_NO_ERROR; +} + +/* XXX:PaN These two will be changed with new driver interface! */ + +/* + * Load the driver init function, might appear under different names depending on object arch... + */ + +int erts_sys_ddll_load_driver_init(void *handle, void **function) +{ + void *fn; + int res; + if ((res = erts_sys_ddll_sym(handle, "driver_init", &fn)) != ERL_DE_NO_ERROR) { + return res; + } + *function = fn; + return res; +} + +int erts_sys_ddll_load_nif_init(void *handle, void **function, ErtsSysDdllError* err) +{ + void *fn; + int res; + if ((res = erts_sys_ddll_sym2(handle, "nif_init", &fn, err)) != ERL_DE_NO_ERROR) { + return res; + } + *function = fn; + return res; +} + + +/* + * Call the driver_init function, whatever it's really called, simple on unix... +*/ +void *erts_sys_ddll_call_init(void *function) { + void *(*initfn)(TWinDynDriverCallbacks *) = function; + return (*initfn)(&wddc); +} + +void *erts_sys_ddll_call_nif_init(void *function) { + void *(*initfn)(TWinDynNifCallbacks *) = function; + return (*initfn)(&nif_callbacks); +} + + +/* + * Close a chared object + */ +int erts_sys_ddll_close2(void *handle, ErtsSysDdllError* err) +{ + if (!FreeLibrary((HINSTANCE) handle)) { + int code = ERL_DE_DYNAMIC_ERROR_OFFSET - GetLastError(); + if (err != NULL) { + err->str = erts_sys_ddll_error(code); + } + return code; + } + return ERL_DE_NO_ERROR; +} + +/* + * Return string that describes the (current) error + */ +#define MAX_ERROR 255 +char *erts_sys_ddll_error(int code) +{ + int actual_code; + char *local_ptr; + if (code > ERL_DE_DYNAMIC_ERROR_OFFSET) { + return "Unspecified error"; + } + actual_code = -1*(code - ERL_DE_DYNAMIC_ERROR_OFFSET); + + local_ptr = TlsGetValue(tls_index); + if (local_ptr == NULL) { + local_ptr = erts_alloc(ERTS_ALC_T_DDLL_ERRCODES, MAX_ERROR); + TlsSetValue(tls_index,local_ptr); + } + if (!FormatMessage( + FORMAT_MESSAGE_FROM_SYSTEM, + NULL, + (DWORD) actual_code, + MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT), + local_ptr, + MAX_ERROR, NULL )) { + return "Unspecified error"; + } else { + char *ptr = local_ptr + strlen(local_ptr) - 1; + while (ptr >= local_ptr && (*ptr == '\r' || *ptr == '\n')) { + *ptr-- = '\0'; + } + } + return local_ptr; +} + +void erts_sys_ddll_free_error(ErtsSysDdllError* err) +{ + /* err->str may be either a static string or reused as thread local data, + * so wo don't bother free it. + */ +} + diff --git a/erts/emulator/sys/win32/erl_win_dyn_driver.h b/erts/emulator/sys/win32/erl_win_dyn_driver.h new file mode 100644 index 0000000000..4949998abc --- /dev/null +++ b/erts/emulator/sys/win32/erl_win_dyn_driver.h @@ -0,0 +1,489 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2003-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +/* + * Include file for erlang driver writers using dynamic drivers on windows. + */ + +/* Maybe this should be auto generated, but I'll leave that for now... */ + +#ifndef _ERL_WIN_DYN_DRIVER_H +#define _ERL_WIN_DYN_DRIVER_H + +#define WDD_FTYPE(FunctionName) TWinDynDriver##FunctionName + +#define WDD_TYPEDEF(RetType, FunctionName, Params) \ + typedef RetType WDD_FTYPE(FunctionName) Params + +WDD_TYPEDEF(int, null_func,(void)); +WDD_TYPEDEF(int, driver_failure_atom,(ErlDrvPort, char *)); +WDD_TYPEDEF(int, driver_failure_posix,(ErlDrvPort, int)); +WDD_TYPEDEF(int, driver_failure,(ErlDrvPort, int)); +WDD_TYPEDEF(int, driver_exit, (ErlDrvPort, int)); +WDD_TYPEDEF(int, driver_failure_eof, (ErlDrvPort)); +WDD_TYPEDEF(int, driver_select, (ErlDrvPort, ErlDrvEvent, int, int)); +WDD_TYPEDEF(int, driver_event, (ErlDrvPort, ErlDrvEvent,ErlDrvEventData)); +WDD_TYPEDEF(int, driver_output, (ErlDrvPort, char *, int)); +WDD_TYPEDEF(int, driver_output2, (ErlDrvPort, char *, int,char *, int)); +WDD_TYPEDEF(int, driver_output_binary, (ErlDrvPort, char *, int,ErlDrvBinary*, int, int)); +WDD_TYPEDEF(int, driver_outputv, (ErlDrvPort, char*, int, ErlIOVec *,int)); +WDD_TYPEDEF(int, driver_vec_to_buf, (ErlIOVec *, char *, int)); +WDD_TYPEDEF(int, driver_set_timer, (ErlDrvPort, unsigned long)); +WDD_TYPEDEF(int, driver_cancel_timer, (ErlDrvPort)); +WDD_TYPEDEF(int, driver_read_timer, (ErlDrvPort, unsigned long *)); +WDD_TYPEDEF(char *, erl_errno_id, (int)); +WDD_TYPEDEF(void, set_busy_port, (ErlDrvPort, int)); +WDD_TYPEDEF(void, set_port_control_flags, (ErlDrvPort, int)); +WDD_TYPEDEF(int, get_port_flags, (ErlDrvPort)); +WDD_TYPEDEF(ErlDrvBinary *, driver_alloc_binary, (int)); +WDD_TYPEDEF(ErlDrvBinary *, driver_realloc_binary, (ErlDrvBinary *, int)); +WDD_TYPEDEF(void, driver_free_binary, (ErlDrvBinary *)); +WDD_TYPEDEF(void *, driver_alloc, (size_t)); +WDD_TYPEDEF(void *, driver_realloc, (void *, size_t)); +WDD_TYPEDEF(void, driver_free, (void *)); +WDD_TYPEDEF(int, driver_enq, (ErlDrvPort, char*, int)); +WDD_TYPEDEF(int, driver_pushq, (ErlDrvPort, char*, int)); +WDD_TYPEDEF(int, driver_deq, (ErlDrvPort, int)); +WDD_TYPEDEF(int, driver_sizeq, (ErlDrvPort)); +WDD_TYPEDEF(int, driver_enq_bin, (ErlDrvPort, ErlDrvBinary *, int,int)); +WDD_TYPEDEF(int, driver_pushq_bin, (ErlDrvPort, ErlDrvBinary *, int,int)); +WDD_TYPEDEF(int, driver_peekqv, (ErlDrvPort, ErlIOVec *)); +WDD_TYPEDEF(SysIOVec *, driver_peekq, (ErlDrvPort, int *)); +WDD_TYPEDEF(int, driver_enqv, (ErlDrvPort, ErlIOVec *, int)); +WDD_TYPEDEF(int, driver_pushqv, (ErlDrvPort, ErlIOVec *, int)); +WDD_TYPEDEF(void, add_driver_entry, (ErlDrvEntry *)); +WDD_TYPEDEF(int, remove_driver_entry, (ErlDrvEntry *)); +WDD_TYPEDEF(ErlDrvTermData, driver_mk_atom, (char*)); +WDD_TYPEDEF(ErlDrvTermData, driver_mk_port,(ErlDrvPort)); +WDD_TYPEDEF(ErlDrvTermData, driver_connected,(ErlDrvPort)); +WDD_TYPEDEF(ErlDrvTermData, driver_caller,(ErlDrvPort)); +WDD_TYPEDEF(ErlDrvTermData, driver_mk_term_nil,(void)); +WDD_TYPEDEF(int, driver_output_term, (ErlDrvPort, ErlDrvTermData*, int)); +WDD_TYPEDEF(int, driver_send_term, (ErlDrvPort, ErlDrvTermData, ErlDrvTermData*, int)); +WDD_TYPEDEF(long, driver_async, (ErlDrvPort,unsigned int*,void (*)(void*),void*,void (*)(void*))); +WDD_TYPEDEF(int, driver_async_cancel, (unsigned int)); +WDD_TYPEDEF(int, driver_lock_driver, (ErlDrvPort)); +WDD_TYPEDEF(void *, driver_dl_open, (char *)); +WDD_TYPEDEF(void *, driver_dl_sym, (void *, char *)); +WDD_TYPEDEF(int, driver_dl_close, (void *)); +WDD_TYPEDEF(char *, driver_dl_error, (void)); +WDD_TYPEDEF(unsigned long, erts_alc_test, (unsigned long, + unsigned long, + unsigned long, + unsigned long)); +WDD_TYPEDEF(long, driver_binary_get_refc, (ErlDrvBinary *dbp)); +WDD_TYPEDEF(long, driver_binary_inc_refc, (ErlDrvBinary *dbp)); +WDD_TYPEDEF(long, driver_binary_dec_refc, (ErlDrvBinary *dbp)); +WDD_TYPEDEF(ErlDrvPDL, driver_pdl_create, (ErlDrvPort)); +WDD_TYPEDEF(void, driver_pdl_lock, (ErlDrvPDL)); +WDD_TYPEDEF(void, driver_pdl_unlock, (ErlDrvPDL)); +WDD_TYPEDEF(long, driver_pdl_get_refc, (ErlDrvPDL)); +WDD_TYPEDEF(long, driver_pdl_inc_refc, (ErlDrvPDL)); +WDD_TYPEDEF(long, driver_pdl_dec_refc, (ErlDrvPDL)); +WDD_TYPEDEF(void, driver_system_info, (ErlDrvSysInfo *, size_t)); +WDD_TYPEDEF(int, driver_get_now, (ErlDrvNowData *)); +WDD_TYPEDEF(int, driver_monitor_process, (ErlDrvPort port, + ErlDrvTermData process, + ErlDrvMonitor *monitor)); +WDD_TYPEDEF(int, driver_demonitor_process, (ErlDrvPort port, + const ErlDrvMonitor *monitor)); +WDD_TYPEDEF(ErlDrvTermData, driver_get_monitored_process, + (ErlDrvPort port, const ErlDrvMonitor *monitor)); +WDD_TYPEDEF(int, driver_compare_monitors, + (const ErlDrvMonitor *, const ErlDrvMonitor *)); +WDD_TYPEDEF(ErlDrvMutex *, erl_drv_mutex_create, (char *name)); +WDD_TYPEDEF(void, erl_drv_mutex_destroy, (ErlDrvMutex *mtx)); +WDD_TYPEDEF(int, erl_drv_mutex_trylock, (ErlDrvMutex *mtx)); +WDD_TYPEDEF(void, erl_drv_mutex_lock, (ErlDrvMutex *mtx)); +WDD_TYPEDEF(void, erl_drv_mutex_unlock, (ErlDrvMutex *mtx)); +WDD_TYPEDEF(ErlDrvCond *, erl_drv_cond_create, (char *name)); +WDD_TYPEDEF(void, erl_drv_cond_destroy, (ErlDrvCond *cnd)); +WDD_TYPEDEF(void, erl_drv_cond_signal, (ErlDrvCond *cnd)); +WDD_TYPEDEF(void, erl_drv_cond_broadcast, (ErlDrvCond *cnd)); +WDD_TYPEDEF(void, erl_drv_cond_wait, (ErlDrvCond *cnd, ErlDrvMutex *mtx)); +WDD_TYPEDEF(ErlDrvRWLock *, erl_drv_rwlock_create, (char *name)); +WDD_TYPEDEF(void, erl_drv_rwlock_destroy, (ErlDrvRWLock *rwlck)); +WDD_TYPEDEF(int, erl_drv_rwlock_tryrlock, (ErlDrvRWLock *rwlck)); +WDD_TYPEDEF(void, erl_drv_rwlock_rlock, (ErlDrvRWLock *rwlck)); +WDD_TYPEDEF(void, erl_drv_rwlock_runlock, (ErlDrvRWLock *rwlck)); +WDD_TYPEDEF(int, erl_drv_rwlock_tryrwlock, (ErlDrvRWLock *rwlck)); +WDD_TYPEDEF(void, erl_drv_rwlock_rwlock, (ErlDrvRWLock *rwlck)); +WDD_TYPEDEF(void, erl_drv_rwlock_rwunlock, (ErlDrvRWLock *rwlck)); +WDD_TYPEDEF(int, erl_drv_tsd_key_create, (char *name, ErlDrvTSDKey *key)); +WDD_TYPEDEF(void, erl_drv_tsd_key_destroy, (ErlDrvTSDKey key)); +WDD_TYPEDEF(void, erl_drv_tsd_set, (ErlDrvTSDKey key, void *data)); +WDD_TYPEDEF(void *, erl_drv_tsd_get, (ErlDrvTSDKey key)); +WDD_TYPEDEF(ErlDrvThreadOpts *, erl_drv_thread_opts_create, (char *name)); +WDD_TYPEDEF(void, erl_drv_thread_opts_destroy, (ErlDrvThreadOpts *opts)); +WDD_TYPEDEF(int, erl_drv_thread_create, (char *name, + ErlDrvTid *tid, + void * (*func)(void *), + void *args, + ErlDrvThreadOpts *opts)); +WDD_TYPEDEF(ErlDrvTid, erl_drv_thread_self, (void)); +WDD_TYPEDEF(int, erl_drv_equal_tids, (ErlDrvTid tid1, ErlDrvTid tid2)); +WDD_TYPEDEF(void, erl_drv_thread_exit, (void *resp)); +WDD_TYPEDEF(int, erl_drv_thread_join, (ErlDrvTid, void **respp)); +WDD_TYPEDEF(int, erl_drv_putenv, (char *key, char *value)); +WDD_TYPEDEF(int, erl_drv_getenv, (char *key, char *value, size_t *value_size)); + +typedef struct { + WDD_FTYPE(null_func) *null_func; + WDD_FTYPE(driver_failure_atom) *driver_failure_atom; + WDD_FTYPE(driver_failure_posix) *driver_failure_posix; + WDD_FTYPE(driver_failure) *driver_failure; + WDD_FTYPE(driver_exit) *driver_exit; + WDD_FTYPE(driver_failure_eof) *driver_failure_eof; + WDD_FTYPE(driver_select) *driver_select; + WDD_FTYPE(driver_event) *driver_event; + WDD_FTYPE(driver_output) *driver_output; + WDD_FTYPE(driver_output2) *driver_output2; + WDD_FTYPE(driver_output_binary) *driver_output_binary; + WDD_FTYPE(driver_outputv) *driver_outputv; + WDD_FTYPE(driver_vec_to_buf) *driver_vec_to_buf; + WDD_FTYPE(driver_set_timer) *driver_set_timer; + WDD_FTYPE(driver_cancel_timer) *driver_cancel_timer; + WDD_FTYPE(driver_read_timer) *driver_read_timer; + WDD_FTYPE(erl_errno_id) *erl_errno_id; + WDD_FTYPE(set_busy_port)* set_busy_port; + WDD_FTYPE(set_port_control_flags) *set_port_control_flags; + WDD_FTYPE(get_port_flags) *get_port_flags; + WDD_FTYPE(driver_alloc_binary) *driver_alloc_binary; + WDD_FTYPE(driver_realloc_binary) *driver_realloc_binary; + WDD_FTYPE(driver_free_binary) *driver_free_binary; + WDD_FTYPE(driver_alloc) *driver_alloc; + WDD_FTYPE(driver_realloc) *driver_realloc; + WDD_FTYPE(driver_free) *driver_free; + WDD_FTYPE(driver_enq) *driver_enq; + WDD_FTYPE(driver_pushq) *driver_pushq; + WDD_FTYPE(driver_deq) *driver_deq; + WDD_FTYPE(driver_sizeq) *driver_sizeq; + WDD_FTYPE(driver_enq_bin)* driver_enq_bin; + WDD_FTYPE(driver_pushq_bin) *driver_pushq_bin; + WDD_FTYPE(driver_peekqv) *driver_peekqv; + WDD_FTYPE(driver_peekq) *driver_peekq; + WDD_FTYPE(driver_enqv) *driver_enqv; + WDD_FTYPE(driver_pushqv) *driver_pushqv; + WDD_FTYPE(add_driver_entry) *add_driver_entry; + WDD_FTYPE(remove_driver_entry) *remove_driver_entry; + WDD_FTYPE(driver_mk_atom) *driver_mk_atom; + WDD_FTYPE(driver_mk_port) *driver_mk_port; + WDD_FTYPE(driver_connected) *driver_connected; + WDD_FTYPE(driver_caller) *driver_caller; + WDD_FTYPE(driver_mk_term_nil) *driver_mk_term_nil; + WDD_FTYPE(driver_output_term) *driver_output_term; + WDD_FTYPE(driver_send_term) *driver_send_term; + WDD_FTYPE(driver_async) *driver_async; + WDD_FTYPE(driver_async_cancel) *driver_async_cancel; + WDD_FTYPE(driver_lock_driver) *driver_lock_driver; + WDD_FTYPE(driver_dl_open) *driver_dl_open; + WDD_FTYPE(driver_dl_sym) *driver_dl_sym; + WDD_FTYPE(driver_dl_close) *driver_dl_close; + WDD_FTYPE(driver_dl_error) *driver_dl_error; + WDD_FTYPE(erts_alc_test) *erts_alc_test; + WDD_FTYPE(driver_binary_get_refc) *driver_binary_get_refc; + WDD_FTYPE(driver_binary_inc_refc) *driver_binary_inc_refc; + WDD_FTYPE(driver_binary_dec_refc) *driver_binary_dec_refc; + WDD_FTYPE(driver_pdl_create) *driver_pdl_create; + WDD_FTYPE(driver_pdl_lock) *driver_pdl_lock; + WDD_FTYPE(driver_pdl_unlock) *driver_pdl_unlock; + WDD_FTYPE(driver_pdl_get_refc) *driver_pdl_get_refc; + WDD_FTYPE(driver_pdl_inc_refc) *driver_pdl_inc_refc; + WDD_FTYPE(driver_pdl_dec_refc) *driver_pdl_dec_refc; + WDD_FTYPE(driver_system_info) *driver_system_info; + WDD_FTYPE(driver_get_now) *driver_get_now; + WDD_FTYPE(driver_monitor_process) *driver_monitor_process; + WDD_FTYPE(driver_demonitor_process) *driver_demonitor_process; + WDD_FTYPE(driver_get_monitored_process) *driver_get_monitored_process; + WDD_FTYPE(driver_compare_monitors) *driver_compare_monitors; + WDD_FTYPE(erl_drv_mutex_create) *erl_drv_mutex_create; + WDD_FTYPE(erl_drv_mutex_destroy) *erl_drv_mutex_destroy; + WDD_FTYPE(erl_drv_mutex_trylock) *erl_drv_mutex_trylock; + WDD_FTYPE(erl_drv_mutex_lock) *erl_drv_mutex_lock; + WDD_FTYPE(erl_drv_mutex_unlock) *erl_drv_mutex_unlock; + WDD_FTYPE(erl_drv_cond_create) *erl_drv_cond_create; + WDD_FTYPE(erl_drv_cond_destroy) *erl_drv_cond_destroy; + WDD_FTYPE(erl_drv_cond_signal) *erl_drv_cond_signal; + WDD_FTYPE(erl_drv_cond_broadcast) *erl_drv_cond_broadcast; + WDD_FTYPE(erl_drv_cond_wait) *erl_drv_cond_wait; + WDD_FTYPE(erl_drv_rwlock_create) *erl_drv_rwlock_create; + WDD_FTYPE(erl_drv_rwlock_destroy) *erl_drv_rwlock_destroy; + WDD_FTYPE(erl_drv_rwlock_tryrlock) *erl_drv_rwlock_tryrlock; + WDD_FTYPE(erl_drv_rwlock_rlock) *erl_drv_rwlock_rlock; + WDD_FTYPE(erl_drv_rwlock_runlock) *erl_drv_rwlock_runlock; + WDD_FTYPE(erl_drv_rwlock_tryrwlock) *erl_drv_rwlock_tryrwlock; + WDD_FTYPE(erl_drv_rwlock_rwlock) *erl_drv_rwlock_rwlock; + WDD_FTYPE(erl_drv_rwlock_rwunlock) *erl_drv_rwlock_rwunlock; + WDD_FTYPE(erl_drv_tsd_key_create) *erl_drv_tsd_key_create; + WDD_FTYPE(erl_drv_tsd_key_destroy) *erl_drv_tsd_key_destroy; + WDD_FTYPE(erl_drv_tsd_set) *erl_drv_tsd_set; + WDD_FTYPE(erl_drv_tsd_get) *erl_drv_tsd_get; + WDD_FTYPE(erl_drv_thread_opts_create) *erl_drv_thread_opts_create; + WDD_FTYPE(erl_drv_thread_opts_destroy) *erl_drv_thread_opts_destroy; + WDD_FTYPE(erl_drv_thread_create) *erl_drv_thread_create; + WDD_FTYPE(erl_drv_thread_self) *erl_drv_thread_self; + WDD_FTYPE(erl_drv_equal_tids) *erl_drv_equal_tids; + WDD_FTYPE(erl_drv_thread_exit) *erl_drv_thread_exit; + WDD_FTYPE(erl_drv_thread_join) *erl_drv_thread_join; + WDD_FTYPE(erl_drv_putenv) *erl_drv_putenv; + WDD_FTYPE(erl_drv_getenv) *erl_drv_getenv; + /* Add new calls here */ +} TWinDynDriverCallbacks; + +/* This header is included explicitly by the ddll static driver, it musn't define things then */ +#ifndef STATIC_ERLANG_DRIVER + +extern TWinDynDriverCallbacks WinDynDriverCallbacks; + +#define null_func (WinDynDriverCallbacks.null_func) +#define driver_failure_atom (WinDynDriverCallbacks.driver_failure_atom) +#define driver_failure_posix (WinDynDriverCallbacks.driver_failure_posix) +#define driver_failure (WinDynDriverCallbacks.driver_failure) +#define driver_exit (WinDynDriverCallbacks.driver_exit) +#define driver_failure_eof (WinDynDriverCallbacks.driver_failure_eof) +#define driver_select (WinDynDriverCallbacks.driver_select) +#define driver_event (WinDynDriverCallbacks.driver_event) +#define driver_output (WinDynDriverCallbacks.driver_output) +#define driver_output2 (WinDynDriverCallbacks.driver_output2) +#define driver_output_binary (WinDynDriverCallbacks.driver_output_binary) +#define driver_outputv (WinDynDriverCallbacks.driver_outputv) +#define driver_vec_to_buf (WinDynDriverCallbacks.driver_vec_to_buf) +#define driver_set_timer (WinDynDriverCallbacks.driver_set_timer) +#define driver_cancel_timer (WinDynDriverCallbacks.driver_cancel_timer) +#define driver_read_timer (WinDynDriverCallbacks.driver_read_timer) +#define erl_errno_id (WinDynDriverCallbacks.erl_errno_id) +#define set_busy_port (WinDynDriverCallbacks.set_busy_port) +#define set_port_control_flags (WinDynDriverCallbacks.set_port_control_flags) +#define get_port_flags (WinDynDriverCallbacks.get_port_flags) +#define driver_alloc_binary (WinDynDriverCallbacks.driver_alloc_binary) +#define driver_realloc_binary (WinDynDriverCallbacks.driver_realloc_binary) +#define driver_free_binary (WinDynDriverCallbacks.driver_free_binary) +#define driver_alloc (WinDynDriverCallbacks.driver_alloc) +#define driver_realloc (WinDynDriverCallbacks.driver_realloc) +#define driver_free (WinDynDriverCallbacks.driver_free) +#define driver_enq (WinDynDriverCallbacks.driver_enq) +#define driver_pushq (WinDynDriverCallbacks.driver_pushq) +#define driver_deq (WinDynDriverCallbacks.driver_deq) +#define driver_sizeq (WinDynDriverCallbacks.driver_sizeq) +#define driver_enq_bin (WinDynDriverCallbacks.driver_enq_bin) +#define driver_pushq_bin (WinDynDriverCallbacks.driver_pushq_bin) +#define driver_peekqv (WinDynDriverCallbacks.driver_peekqv) +#define driver_peekq (WinDynDriverCallbacks.driver_peekq) +#define driver_enqv (WinDynDriverCallbacks.driver_enqv) +#define driver_pushqv (WinDynDriverCallbacks.driver_pushqv) +#define add_driver_entry (WinDynDriverCallbacks.add_driver_entry) +#define remove_driver_entry (WinDynDriverCallbacks.remove_driver_entry) +#define driver_mk_atom (WinDynDriverCallbacks.driver_mk_atom) +#define driver_mk_port (WinDynDriverCallbacks.driver_mk_port) +#define driver_connected (WinDynDriverCallbacks.driver_connected) +#define driver_caller (WinDynDriverCallbacks.driver_caller) +#define driver_mk_term_nil (WinDynDriverCallbacks.driver_mk_term_nil) +#define driver_output_term (WinDynDriverCallbacks.driver_output_term) +#define driver_send_term (WinDynDriverCallbacks.driver_send_term) +#define driver_async (WinDynDriverCallbacks.driver_async) +#define driver_async_cancel (WinDynDriverCallbacks.driver_async_cancel) +#define driver_lock_driver (WinDynDriverCallbacks.driver_lock_driver) +#define driver_dl_open (WinDynDriverCallbacks.driver_dl_open) +#define driver_dl_sym (WinDynDriverCallbacks.driver_dl_sym) +#define driver_dl_close (WinDynDriverCallbacks.driver_dl_close) +#define driver_dl_error (WinDynDriverCallbacks.driver_dl_error) +#define erts_alc_test (WinDynDriverCallbacks.erts_alc_test) +#define driver_binary_get_refc (WinDynDriverCallbacks.driver_binary_get_refc) +#define driver_binary_inc_refc (WinDynDriverCallbacks.driver_binary_inc_refc) +#define driver_binary_dec_refc (WinDynDriverCallbacks.driver_binary_dec_refc) +#define driver_pdl_create (WinDynDriverCallbacks.driver_pdl_create) +#define driver_pdl_lock (WinDynDriverCallbacks.driver_pdl_lock) +#define driver_pdl_unlock (WinDynDriverCallbacks.driver_pdl_unlock) +#define driver_pdl_get_refc (WinDynDriverCallbacks.driver_pdl_get_refc) +#define driver_pdl_inc_refc (WinDynDriverCallbacks.driver_pdl_inc_refc) +#define driver_pdl_dec_refc (WinDynDriverCallbacks.driver_pdl_dec_refc) +#define driver_system_info (WinDynDriverCallbacks.driver_system_info) +#define driver_get_now (WinDynDriverCallbacks.driver_get_now) +#define driver_monitor_process \ +(WinDynDriverCallbacks.driver_monitor_process) +#define driver_demonitor_process \ +(WinDynDriverCallbacks.driver_demonitor_process) +#define driver_get_monitored_process \ +(WinDynDriverCallbacks.driver_get_monitored_process) +#define driver_compare_monitors \ +(WinDynDriverCallbacks.driver_compare_monitors) +#define erl_drv_mutex_create (WinDynDriverCallbacks.erl_drv_mutex_create) +#define erl_drv_mutex_destroy (WinDynDriverCallbacks.erl_drv_mutex_destroy) +#define erl_drv_mutex_trylock (WinDynDriverCallbacks.erl_drv_mutex_trylock) +#define erl_drv_mutex_lock (WinDynDriverCallbacks.erl_drv_mutex_lock) +#define erl_drv_mutex_unlock (WinDynDriverCallbacks.erl_drv_mutex_unlock) +#define erl_drv_cond_create (WinDynDriverCallbacks.erl_drv_cond_create) +#define erl_drv_cond_destroy (WinDynDriverCallbacks.erl_drv_cond_destroy) +#define erl_drv_cond_signal (WinDynDriverCallbacks.erl_drv_cond_signal) +#define erl_drv_cond_broadcast (WinDynDriverCallbacks.erl_drv_cond_broadcast) +#define erl_drv_cond_wait (WinDynDriverCallbacks.erl_drv_cond_wait) +#define erl_drv_rwlock_create (WinDynDriverCallbacks.erl_drv_rwlock_create) +#define erl_drv_rwlock_destroy (WinDynDriverCallbacks.erl_drv_rwlock_destroy) +#define erl_drv_rwlock_tryrlock (WinDynDriverCallbacks.erl_drv_rwlock_tryrlock) +#define erl_drv_rwlock_rlock (WinDynDriverCallbacks.erl_drv_rwlock_rlock) +#define erl_drv_rwlock_runlock (WinDynDriverCallbacks.erl_drv_rwlock_runlock) +#define erl_drv_rwlock_tryrwlock \ +(WinDynDriverCallbacks.erl_drv_rwlock_tryrwlock) +#define erl_drv_rwlock_rwlock (WinDynDriverCallbacks.erl_drv_rwlock_rwlock) +#define erl_drv_rwlock_rwunlock (WinDynDriverCallbacks.erl_drv_rwlock_rwunlock) +#define erl_drv_tsd_key_create (WinDynDriverCallbacks.erl_drv_tsd_key_create) +#define erl_drv_tsd_key_destroy (WinDynDriverCallbacks.erl_drv_tsd_key_destroy) +#define erl_drv_tsd_set (WinDynDriverCallbacks.erl_drv_tsd_set) +#define erl_drv_tsd_get (WinDynDriverCallbacks.erl_drv_tsd_get) +#define erl_drv_thread_opts_create \ +(WinDynDriverCallbacks.erl_drv_thread_opts_create) +#define erl_drv_thread_opts_destroy \ +(WinDynDriverCallbacks.erl_drv_thread_opts_destroy) +#define erl_drv_thread_create (WinDynDriverCallbacks.erl_drv_thread_create) +#define erl_drv_thread_self (WinDynDriverCallbacks.erl_drv_thread_self) +#define erl_drv_equal_tids (WinDynDriverCallbacks.erl_drv_equal_tids) +#define erl_drv_thread_exit (WinDynDriverCallbacks.erl_drv_thread_exit) +#define erl_drv_thread_join (WinDynDriverCallbacks.erl_drv_thread_join) +#define erl_drv_putenv (WinDynDriverCallbacks.erl_drv_putenv) +#define erl_drv_getenv (WinDynDriverCallbacks.erl_drv_getenv) + +/* The only variable in the interface... */ +#define driver_term_nil (driver_mk_term_nil()) + +#include <stdio.h> +#include <stdlib.h> + +#define DRIVER_INIT(DriverName) \ +ErlDrvEntry *erl_dyndriver_real_driver_init(void); \ +TWinDynDriverCallbacks WinDynDriverCallbacks; \ +__declspec(dllexport) ErlDrvEntry *driver_init(TWinDynDriverCallbacks *callbacks) \ +{ \ + memcpy(&WinDynDriverCallbacks,callbacks,sizeof(TWinDynDriverCallbacks)); \ + return erl_dyndriver_real_driver_init(); \ +} \ +ErlDrvEntry *erl_dyndriver_real_driver_init(void) + +/* This is to make erl_driver.h avoid changing what's done here */ +#define ERL_DRIVER_TYPES_ONLY + +#else /* defined(STATIC_ERLANG_DRIVER) */ +/* This is for the ddll driver */ + +#define ERL_INIT_CALLBACK_STRUCTURE(W) \ +do { \ +((W).null_func) = null_func; \ +((W).driver_failure_atom) = driver_failure_atom; \ +((W).driver_failure_posix) = driver_failure_posix; \ +((W).driver_failure) = driver_failure; \ +((W).driver_exit) = driver_exit; \ +((W).driver_failure_eof) = driver_failure_eof; \ +((W).driver_select) = driver_select; \ +((W).driver_event) = driver_event; \ +((W).driver_output) = driver_output; \ +((W).driver_output2) = driver_output2; \ +((W).driver_output_binary) = driver_output_binary; \ +((W).driver_outputv) = driver_outputv; \ +((W).driver_vec_to_buf) = driver_vec_to_buf; \ +((W).driver_set_timer) = driver_set_timer; \ +((W).driver_cancel_timer) = driver_cancel_timer; \ +((W).driver_read_timer) = driver_read_timer; \ +((W).erl_errno_id) = erl_errno_id; \ +((W).set_busy_port) = set_busy_port; \ +((W).set_port_control_flags) = set_port_control_flags; \ +((W).get_port_flags) = get_port_flags; \ +((W).driver_alloc_binary) = driver_alloc_binary; \ +((W).driver_realloc_binary) = driver_realloc_binary; \ +((W).driver_free_binary) = driver_free_binary; \ +((W).driver_alloc) = driver_alloc; \ +((W).driver_realloc) = driver_realloc; \ +((W).driver_free) = driver_free; \ +((W).driver_enq) = driver_enq; \ +((W).driver_pushq) = driver_pushq; \ +((W).driver_deq) = driver_deq; \ +((W).driver_sizeq) = driver_sizeq; \ +((W).driver_enq_bin) = driver_enq_bin; \ +((W).driver_pushq_bin) = driver_pushq_bin; \ +((W).driver_peekqv) = driver_peekqv; \ +((W).driver_peekq) = driver_peekq; \ +((W).driver_enqv) = driver_enqv; \ +((W).driver_pushqv) = driver_pushqv; \ +((W).add_driver_entry) = add_driver_entry; \ +((W).remove_driver_entry) = remove_driver_entry; \ +((W).driver_mk_atom) = driver_mk_atom; \ +((W).driver_mk_port) = driver_mk_port; \ +((W).driver_connected) = driver_connected; \ +((W).driver_caller) = driver_caller; \ +((W).driver_mk_term_nil) = driver_mk_term_nil; \ +((W).driver_output_term) = driver_output_term; \ +((W).driver_send_term) = driver_send_term; \ +((W).driver_async) = driver_async; \ +((W).driver_async_cancel) = driver_async_cancel; \ +((W).driver_lock_driver) = driver_lock_driver; \ +((W).driver_dl_open) = driver_dl_open; \ +((W).driver_dl_sym) = driver_dl_sym; \ +((W).driver_dl_close) = driver_dl_close; \ +((W).driver_dl_error) = driver_dl_error; \ +((W).erts_alc_test) = erts_alc_test; \ +((W).driver_binary_get_refc) = driver_binary_get_refc; \ +((W).driver_binary_inc_refc) = driver_binary_inc_refc; \ +((W).driver_binary_dec_refc) = driver_binary_dec_refc; \ +((W).driver_pdl_create) = driver_pdl_create; \ +((W).driver_pdl_lock) = driver_pdl_lock; \ +((W).driver_pdl_unlock) = driver_pdl_unlock; \ +((W).driver_pdl_get_refc) = driver_pdl_get_refc; \ +((W).driver_pdl_inc_refc) = driver_pdl_inc_refc; \ +((W).driver_pdl_dec_refc) = driver_pdl_dec_refc; \ +((W).driver_system_info) = driver_system_info; \ +((W).driver_get_now) = driver_get_now; \ +((W).driver_monitor_process) = driver_monitor_process; \ +((W).driver_demonitor_process) = driver_demonitor_process; \ +((W).driver_get_monitored_process) = driver_get_monitored_process; \ +((W).driver_compare_monitors) = driver_compare_monitors;\ +((W).erl_drv_mutex_create) = erl_drv_mutex_create; \ +((W).erl_drv_mutex_destroy) = erl_drv_mutex_destroy; \ +((W).erl_drv_mutex_trylock) = erl_drv_mutex_trylock; \ +((W).erl_drv_mutex_lock) = erl_drv_mutex_lock; \ +((W).erl_drv_mutex_unlock) = erl_drv_mutex_unlock; \ +((W).erl_drv_cond_create) = erl_drv_cond_create; \ +((W).erl_drv_cond_destroy) = erl_drv_cond_destroy; \ +((W).erl_drv_cond_signal) = erl_drv_cond_signal; \ +((W).erl_drv_cond_broadcast) = erl_drv_cond_broadcast; \ +((W).erl_drv_cond_wait) = erl_drv_cond_wait; \ +((W).erl_drv_rwlock_create) = erl_drv_rwlock_create; \ +((W).erl_drv_rwlock_destroy) = erl_drv_rwlock_destroy; \ +((W).erl_drv_rwlock_tryrlock) = erl_drv_rwlock_tryrlock;\ +((W).erl_drv_rwlock_rlock) = erl_drv_rwlock_rlock; \ +((W).erl_drv_rwlock_runlock) = erl_drv_rwlock_runlock; \ +((W).erl_drv_rwlock_tryrwlock) = erl_drv_rwlock_tryrwlock;\ +((W).erl_drv_rwlock_rwlock) = erl_drv_rwlock_rwlock; \ +((W).erl_drv_rwlock_rwunlock) = erl_drv_rwlock_rwunlock;\ +((W).erl_drv_tsd_key_create) = erl_drv_tsd_key_create; \ +((W).erl_drv_tsd_key_destroy) = erl_drv_tsd_key_destroy;\ +((W).erl_drv_tsd_set) = erl_drv_tsd_set; \ +((W).erl_drv_tsd_get) = erl_drv_tsd_get; \ +((W).erl_drv_thread_opts_create) = erl_drv_thread_opts_create;\ +((W).erl_drv_thread_opts_destroy) = erl_drv_thread_opts_destroy;\ +((W).erl_drv_thread_create) = erl_drv_thread_create; \ +((W).erl_drv_thread_self) = erl_drv_thread_self; \ +((W).erl_drv_equal_tids) = erl_drv_equal_tids; \ +((W).erl_drv_thread_exit) = erl_drv_thread_exit; \ +((W).erl_drv_thread_join) = erl_drv_thread_join; \ +((W).erl_drv_putenv) = erl_drv_putenv; \ +((W).erl_drv_getenv) = erl_drv_getenv; \ +} while (0) + + + +#endif /* STATIC_ERLANG_DRIVER */ +#endif /* _ERL_WIN_DYN_DRIVER_H */ diff --git a/erts/emulator/sys/win32/erl_win_sys.h b/erts/emulator/sys/win32/erl_win_sys.h new file mode 100644 index 0000000000..92d8577537 --- /dev/null +++ b/erts/emulator/sys/win32/erl_win_sys.h @@ -0,0 +1,212 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 1997-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ +/* + * This file handles differences between operating systems. + * This should be the only place with conditional compilation + * depending on the type of OS. + */ + +#ifndef _ERL_WIN_SYS_H +#define _ERL_WIN_SYS_H + +#define HAS_STDARG + +#ifdef __GNUC__ +#ifdef pid_t +/* Really... */ +#undef pid_t +#endif +#endif +#include <stdio.h> +#include <math.h> +#include <stdlib.h> +#include <string.h> +#include <stdarg.h> +#include <limits.h> +#include <process.h> +#include <malloc.h> +#ifndef __GNUC__ +#include <direct.h> +#endif +#include "erl_errno.h" +#include <io.h> +#include <sys/types.h> +#include <sys/stat.h> +#include <fcntl.h> +#include <time.h> +#include <sys/timeb.h> +#pragma comment(linker,"/manifestdependency:\"type='win32' "\ + "name='Microsoft.Windows.Common-Controls' "\ + "version='6.0.0.0' processorArchitecture='*' "\ + "publicKeyToken='6595b64144ccf1df' language='*'\"") + +#define WIN32_LEAN_AND_MEAN +#include <windows.h> +#undef WIN32_LEAN_AND_MEAN + +/* + * Define MAXPATHLEN in terms of MAXPATH if available. + */ + +#ifndef MAXPATH +#define MAXPATH MAX_PATH +#endif /* MAXPATH */ + +#ifndef MAXPATHLEN +#define MAXPATHLEN MAXPATH +#endif /* MAXPATHLEN */ + +/* + * Various configuration options, used to be in the Makefile. + */ + +#define NO_ASINH +#define NO_ACOSH +#define NO_ATANH +#define NO_ERF +#define NO_ERFC + +#define NO_SYSLOG +#define NO_SYSCONF +#define NO_DAEMON +#define NO_PWD +/*#define HAVE_MEMMOVE*/ + +#define strncasecmp _strnicmp + +/* + * Practial Windows specific macros. + */ + +#define CreateAutoEvent(state) CreateEvent(NULL, FALSE, state, NULL) +#define CreateManualEvent(state) CreateEvent(NULL, TRUE, state, NULL) + + +/* + * Our own type of "FD's" + */ +#define ERTS_SYS_FD_TYPE HANDLE +#define NO_FSTAT_ON_SYS_FD_TYPE 1 /* They are events, not files */ + +#define HAVE_ERTS_CHECK_IO_DEBUG +int erts_check_io_debug(void); + +/* + * For erl_time_sup + */ +#define HAVE_GETHRTIME + +#define sys_init_hrtime() /* Nothing */ + +#define SYS_CLK_TCK 1000 +#define SYS_CLOCK_RESOLUTION 1 + +typedef struct { + long tv_sec; + long tv_usec; +} SysTimeval; + +typedef struct { + clock_t tms_utime; + clock_t tms_stime; + clock_t tms_cutime; + clock_t tms_cstime; +} SysTimes; + +#define HAVE_INT64 1 +#if defined (__GNUC__) +typedef unsigned long long Uint64; +typedef long long Sint64; + +typedef long long SysHrTime; +#else +typedef ULONGLONG Uint64; +typedef LONGLONG Sint64; + +typedef LONGLONG SysHrTime; +#endif + +extern int sys_init_time(void); +extern void sys_gettimeofday(SysTimeval *tv); +extern SysHrTime sys_gethrtime(void); +extern clock_t sys_times(SysTimes *buffer); + +extern char *win_build_environment(char *); + +typedef struct { + char *environment_strings; + char *next_string; +} GETENV_STATE; + +void erts_sys_env_init(void); + +/* + ** These are to avoid irritating warnings + */ +#pragma warning(disable : 4244) +#pragma warning(disable : 4018) + +/* + * Floating point support. + */ + +extern volatile int erl_fp_exception; + +#include <float.h> +#if defined (__GNUC__) +int _finite(double x); +#endif +#endif + +/*#define NO_FPE_SIGNALS*/ +#define erts_get_current_fp_exception() NULL +#define __ERTS_FP_CHECK_INIT(fpexnp) do {} while (0) +#define __ERTS_FP_ERROR(fpexnp, f, Action) if (!_finite(f)) { Action; } else {} +#define __ERTS_FP_ERROR_THOROUGH(fpexnp, f, Action) __ERTS_FP_ERROR(fpexnp, f, Action) +#define __ERTS_SAVE_FP_EXCEPTION(fpexnp) +#define __ERTS_RESTORE_FP_EXCEPTION(fpexnp) + +#define ERTS_FP_CHECK_INIT(p) __ERTS_FP_CHECK_INIT(&(p)->fp_exception) +#define ERTS_FP_ERROR(p, f, A) __ERTS_FP_ERROR(&(p)->fp_exception, f, A) +#define ERTS_SAVE_FP_EXCEPTION(p) __ERTS_SAVE_FP_EXCEPTION(&(p)->fp_exception) +#define ERTS_RESTORE_FP_EXCEPTION(p) __ERTS_RESTORE_FP_EXCEPTION(&(p)->fp_exception) +#define ERTS_FP_ERROR_THOROUGH(p, f, A) __ERTS_FP_ERROR_THOROUGH(&(p)->fp_exception, f, A) + +#define erts_sys_block_fpe() 0 +#define erts_sys_unblock_fpe(x) do{}while(0) + +#define SIZEOF_SHORT 2 +#define SIZEOF_INT 4 +#define SIZEOF_LONG 4 +#define SIZEOF_VOID_P 4 +#define SIZEOF_SIZE_T 4 +#define SIZEOF_OFF_T 4 + +/* + * Seems to be missing. + */ +#ifndef __GNUC__ +typedef long ssize_t; +#endif + +/* Threads */ +#ifdef USE_THREADS +int init_async(int); +int exit_async(void); +#endif diff --git a/erts/emulator/sys/win32/sys.c b/erts/emulator/sys/win32/sys.c new file mode 100644 index 0000000000..3194493ac8 --- /dev/null +++ b/erts/emulator/sys/win32/sys.c @@ -0,0 +1,3093 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 1996-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ +/* + * system-dependent functions + * + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include "sys.h" +#include "erl_alloc.h" +#include "erl_sys_driver.h" +#include "global.h" +#include "erl_threads.h" +#include "../../drivers/win32/win_con.h" + + +void erts_sys_init_float(void); + +void erl_start(int, char**); +void erl_exit(int n, char*, _DOTS_); +void erl_error(char*, va_list); +void erl_crash_dump(char*, int, char*, ...); + +/* + * Microsoft-specific function to map a WIN32 error code to a Posix errno. + */ +extern void _dosmaperr(DWORD); + +#ifdef ERL_RUN_SHARED_LIB +#ifdef __argc +#undef __argc +#endif +#define __argc e_argc +#ifdef __argv +#undef __argv +#endif +#define __argv e_argv +#endif + +static void init_console(); +static int get_and_remove_option(int* argc, char** argv, const char* option); +static char *get_and_remove_option2(int *argc, char **argv, + const char *option); +static int init_async_io(struct async_io* aio, int use_threads); +static void release_async_io(struct async_io* aio, ErlDrvPort); +static void async_read_file(struct async_io* aio, LPVOID buf, DWORD numToRead); +static int async_write_file(struct async_io* aio, LPVOID buf, DWORD numToWrite); +static int get_overlapped_result(struct async_io* aio, + LPDWORD pBytesRead, BOOL wait); +static FUNCTION(BOOL, CreateChildProcess, (char *, HANDLE, HANDLE, + HANDLE, LPHANDLE, BOOL, + LPVOID, LPTSTR, unsigned, + char **, int *)); +static int create_pipe(LPHANDLE, LPHANDLE, BOOL, BOOL); +static int ApplicationType(const char* originalName, char fullPath[MAX_PATH], + BOOL search_in_path, BOOL handle_quotes, + int *error_return); + +HANDLE erts_service_event; + +#ifdef ERTS_SMP +static erts_smp_tsd_key_t win32_errstr_key; +#endif + +static erts_smp_atomic_t pipe_creation_counter; + +static erts_smp_mtx_t sys_driver_data_lock; + + +/* Results from ApplicationType is one of */ +#define APPL_NONE 0 +#define APPL_DOS 1 +#define APPL_WIN3X 2 +#define APPL_WIN32 3 + +static FUNCTION(int, driver_write, (long, HANDLE, byte*, int)); +static void common_stop(int); +static int create_file_thread(struct async_io* aio, int mode); +static DWORD WINAPI threaded_reader(LPVOID param); +static DWORD WINAPI threaded_writer(LPVOID param); +static DWORD WINAPI threaded_exiter(LPVOID param); + +#ifdef DEBUG +static void debug_console(void); +#endif + +BOOL WINAPI ctrl_handler(DWORD dwCtrlType); + +#define PORT_BUFSIZ 4096 + +#define PORT_FREE (-1) +#define PORT_EXITING (-2) + +#define DRV_BUF_ALLOC(SZ) \ + erts_alloc_fnf(ERTS_ALC_T_DRV_DATA_BUF, (SZ)) +#define DRV_BUF_REALLOC(P, SZ) \ + erts_realloc_fnf(ERTS_ALC_T_DRV_DATA_BUF, (P), (SZ)) +#define DRV_BUF_FREE(P) \ + erts_free(ERTS_ALC_T_DRV_DATA_BUF, (P)) + +/********************* General functions ****************************/ + +/* + * Whether create_pipe() should use a named pipe or an anonymous. + * (Named pipes are not supported on Windows 95.) + */ + +static int max_files = 1024; + +static BOOL use_named_pipes; +static BOOL win_console = FALSE; + + +static OSVERSIONINFO int_os_version; /* Version information for Win32. */ + + +/* This is the system's main function (which may or may not be called "main") + - do general system-dependent initialization + - call erl_start() to parse arguments and do other init +*/ + +static erts_smp_atomic_t sys_misc_mem_sz; + +HMODULE beam_module = NULL; + +void erl_sys_init(); + +void erl_sys_args(int* argc, char** argv); + +int nohup; +#ifndef __GNUC__ +void erts_sys_invalid_parameter_handler(const wchar_t * expression, + const wchar_t * function, + const wchar_t * file, + unsigned int line, + uintptr_t pReserved + ) +{ +#ifdef DEBUG + fprintf(stderr, + "Debug: Invalid parameter\"%ls\" " + "(detected in \"%ls\" [%ls:%d]) \n", + (expression) ? expression : L"(unknown)", + (function) ? function : L"(unknown)", + (file) ? file : L"(unknown)", + line); +#endif + return; +} +#endif + +void sys_primitive_init(HMODULE beam) +{ +#ifndef __GNUC__ + /* Initialize this module handle (the beam.dll module handle) and + take care of the standard library's aggressive invalid parameter + handling... */ + _set_invalid_parameter_handler(&erts_sys_invalid_parameter_handler); +#endif + beam_module = (HMODULE) beam; +} + +Uint +erts_sys_misc_mem_sz(void) +{ + Uint res = (Uint) erts_check_io_size(); + res += (Uint) erts_smp_atomic_read(&sys_misc_mem_sz); + return res; +} + +void erl_sys_args(int* argc, char** argv) +{ + char *event_name; + nohup = get_and_remove_option(argc, argv, "-nohup"); + +#ifdef DEBUG + /* + * Start a debug console if -console option given. + */ + + if (get_and_remove_option(argc, argv, "-console")) { + debug_console(); + } +#endif + + if (nohup && (event_name = get_and_remove_option2(argc, argv, + "-service_event"))) { + if ((erts_service_event = + OpenEvent(EVENT_ALL_ACCESS,FALSE,event_name)) == NULL) { + erts_fprintf(stderr, + "Warning: could not open service event: %s\r\n", + event_name); + } + } else { + erts_service_event = NULL; + } + +#ifdef DEBUG + /* + * Given the "-threads" option, always use threads instead of + * named pipes. + */ + + if (get_and_remove_option(argc, argv, "-threads")) { + use_named_pipes = FALSE; + } +#endif +} + +void +erts_sys_prepare_crash_dump(void) +{ + /* Windows - free file descriptors are hopefully available */ + return; +} + +static void +init_console() +{ + char* mode = erts_read_env("ERL_CONSOLE_MODE"); + + if (!mode || strcmp(mode, "window") == 0) { + win_console = TRUE; + ConInit(); + /*nohup = 0;*/ + } else if (strncmp(mode, "tty:", 4) == 0) { + if (mode[5] == 'c') { + setvbuf(stdout, NULL, _IONBF, 0); + } + if (mode[6] == 'c') { + setvbuf(stderr, NULL, _IONBF, 0); + } + } + + erts_free_read_env(mode); +} + +int sys_max_files() +{ + return max_files; +} + +/* + * Looks for the given option in the argv vector. If it is found, + * it will be removed from the argv vector. + * + * If the return value indicates that the option was found and removed, + * it is the responsibility of the caller to decrement the value of argc. + * + * Returns: 0 if the option wasn't found, 1 if it was found + */ + +static int +get_and_remove_option(argc, argv, option) + int* argc; /* Number of arguments. */ + char* argv[]; /* The argument vector. */ + const char* option; /* Option to search for and remove. */ +{ + int i; + + for (i = 1; i < *argc; i++) { + if (strcmp(argv[i], option) == 0) { + (*argc)--; + while (i < *argc) { + argv[i] = argv[i+1]; + i++; + } + argv[i] = NULL; + return 1; + } + } + return 0; +} + +static char *get_and_remove_option2(int *argc, char **argv, + const char *option) +{ + char *ret; + int i; + + for (i = 1; i < *argc; i++) { + if (strcmp(argv[i], option) == 0) { + if (i+1 < *argc) { + ret = argv[i+1]; + (*argc) -= 2; + while (i < *argc) { + argv[i] = argv[i+2]; + i++; + } + argv[i] = NULL; + return ret; + } + } + } + return NULL; +} + + +/************************** OS info *******************************/ + +/* Used by erlang:info/1. */ +/* (This code was formerly in drv.XXX/XXX_os_drv.c) */ + +char os_type[] = "win32"; + +void +os_flavor(namebuf, size) +char* namebuf; /* Where to return the name. */ +unsigned size; /* Size of name buffer. */ +{ + switch (int_os_version.dwPlatformId) { + case VER_PLATFORM_WIN32_WINDOWS: + strcpy(namebuf, "windows"); + break; + case VER_PLATFORM_WIN32_NT: + strcpy(namebuf, "nt"); + break; + default: /* Can't happen. */ + strcpy(namebuf, "unknown"); + break; + } +} + +void +os_version(pMajor, pMinor, pBuild) +int* pMajor; /* Pointer to major version. */ +int* pMinor; /* Pointer to minor version. */ +int* pBuild; /* Pointer to build number. */ +{ + *pMajor = int_os_version.dwMajorVersion; + *pMinor = int_os_version.dwMinorVersion; + *pBuild = int_os_version.dwBuildNumber; +} + +/************************** Port I/O *******************************/ + +/* I. Common stuff */ + +/* II. The spawn/fd/vanilla drivers */ + +/* + * Definitions for driver flags. + */ + +#define DF_OVR_READY 1 /* Overlapped result is ready. */ +#define DF_EXIT_THREAD 2 /* The thread should exit. */ +#define DF_XLAT_CR 4 /* The thread should translate CRs. */ +#define DF_DROP_IF_INVH 8 /* Drop packages instead of crash if + invalid handle (stderr) */ + +#define OV_BUFFER_PTR(dp) ((LPVOID) ((dp)->ov.Internal)) +#define OV_NUM_TO_READ(dp) ((dp)->ov.InternalHigh) + +/* + * This data is used to make overlapped I/O operations work on both + * Windows NT (using true overlapped I/O) and Windows 95 (using threads). + */ + +typedef struct async_io { + unsigned flags; /* Driver flags, definitions found above. */ + HANDLE thread; /* If -1, overlapped I/O is used (Windows NT). + * Otherwise, it is the handle of the thread used + * for simulating overlapped I/O (Windows 95 and + * the console for Windows NT). + */ + HANDLE fd; /* Handle for file or pipe. */ +#ifdef ERTS_SMP + int async_io_active; /* if true, a close of the file will signal the event in ov */ +#endif + OVERLAPPED ov; /* Control structure for overlapped reading. + * When overlapped reading is simulated with + * a thread, the fields are used as follows: + * ov.Internal - Read buffer. + * ov.InternalHigh - Number of bytes to read. + * See macros above. + */ + HANDLE ioAllowed; /* The thread will wait for this event + * before starting a new read or write. + */ + DWORD pendingError; /* Used to delay presentating an error to Erlang + * until the check_io function is entered. + */ + DWORD bytesTransferred; /* Bytes read or write in the last operation. + * Valid only when DF_OVR_READY is set. + */ +} AsyncIo; + + +/* + * Input thread for fd_driver (if fd_driver is running). + */ +static AsyncIo* fd_driver_input = NULL; +static BOOL (WINAPI *fpSetHandleInformation)(HANDLE,DWORD,DWORD); + +/* + * This data is used by the spawn and vanilla drivers. + * There will be one entry for each port, even if the input + * and output HANDLES are different. Since handles are not + * guaranteed to be small numbers in Win32, we cannot index + * with them. I.e. the index for each entry is not equal to + * none of the file handles. + */ + +typedef struct driver_data { + int totalNeeded; /* Total number of bytes needed to fill + * up the packet header or packet. */ + int bytesInBuffer; /* Number of bytes read so far in + * the input buffer. + */ + int inBufSize; /* Size of input buffer. */ + byte *inbuf; /* Buffer to use for overlapped read. */ + int outBufSize; /* Size of output buffer. */ + byte *outbuf; /* Buffer to use for overlapped write. */ + ErlDrvPort port_num; /* The port number. */ + int packet_bytes; /* 0: continous stream, 1, 2, or 4: the number + * of bytes in the packet header. + */ + HANDLE port_pid; /* PID of the port process. */ + AsyncIo in; /* Control block for overlapped reading. */ + AsyncIo out; /* Control block for overlapped writing. */ + int report_exit; /* Do report exit status for the port */ +} DriverData; + +static DriverData* driver_data; /* Pointer to array of driver data. */ + +/* Driver interfaces */ +static ErlDrvData spawn_start(ErlDrvPort, char*, SysDriverOpts*); +static ErlDrvData fd_start(ErlDrvPort, char*, SysDriverOpts*); +static ErlDrvData vanilla_start(ErlDrvPort, char*, SysDriverOpts*); +static int spawn_init(void); +static int fd_init(void); +static void fd_stop(ErlDrvData); +static void stop(ErlDrvData); +static void output(ErlDrvData, char*, int); +static void ready_input(ErlDrvData, ErlDrvEvent); +static void ready_output(ErlDrvData, ErlDrvEvent); +static void stop_select(ErlDrvEvent, void*); + +struct erl_drv_entry spawn_driver_entry = { + spawn_init, + spawn_start, + stop, + output, + ready_input, + ready_output, + "spawn", + NULL, /* finish */ + NULL, /* handle */ + NULL, /* control */ + NULL, /* timeout */ + NULL, /* outputv */ + NULL, /* ready_async */ + NULL, /* flush */ + NULL, /* call */ + NULL, /* event */ + ERL_DRV_EXTENDED_MARKER, + ERL_DRV_EXTENDED_MAJOR_VERSION, + ERL_DRV_EXTENDED_MINOR_VERSION, + 0, /* ERL_DRV_FLAGs */ + NULL, + NULL, /* process_exit */ + stop_select +}; + +#ifdef HARD_POLL_DEBUG +extern void poll_debug_set_active_fd(ErtsSysFdType fd); +extern void poll_debug_read_begin(ErtsSysFdType fd); +extern void poll_debug_read_done(ErtsSysFdType fd, int bytes); +extern void poll_debug_async_initialized(ErtsSysFdType fd); +extern void poll_debug_async_immediate(ErtsSysFdType fd, int bytes); +extern void poll_debug_write_begin(ErtsSysFdType fd); +extern void poll_debug_write_done(ErtsSysFdType fd, int bytes); +#endif + +extern int null_func(void); + +struct erl_drv_entry fd_driver_entry = { + fd_init, + fd_start, + fd_stop, + output, + ready_input, + ready_output, + "fd", + NULL, /* finish */ + NULL, /* handle */ + NULL, /* control */ + NULL, /* timeout */ + NULL, /* outputv */ + NULL, /* ready_async */ + NULL, /* flush */ + NULL, /* call */ + NULL, /* event */ + ERL_DRV_EXTENDED_MARKER, + ERL_DRV_EXTENDED_MAJOR_VERSION, + ERL_DRV_EXTENDED_MINOR_VERSION, + 0, /* ERL_DRV_FLAGs */ + NULL, + NULL, /* process_exit */ + stop_select +}; + +struct erl_drv_entry vanilla_driver_entry = { + null_func, + vanilla_start, + stop, + output, + ready_input, + ready_output, + "vanilla", + NULL, /* finish */ + NULL, /* handle */ + NULL, /* control */ + NULL, /* timeout */ + NULL, /* outputv */ + NULL, /* ready_async */ + NULL, /* flush */ + NULL, /* call */ + NULL, /* event */ + ERL_DRV_EXTENDED_MARKER, + ERL_DRV_EXTENDED_MAJOR_VERSION, + ERL_DRV_EXTENDED_MINOR_VERSION, + 0, /* ERL_DRV_FLAGs */ + NULL, + NULL, /* process_exit */ + stop_select +}; + +#if defined(USE_THREADS) && !defined(ERTS_SMP) + +static int async_drv_init(void); +static ErlDrvData async_drv_start(ErlDrvPort, char*, SysDriverOpts*); +static void async_drv_stop(ErlDrvData); +static void async_drv_input(ErlDrvData, ErlDrvEvent); + +/* INTERNAL use only */ + +void null_output(ErlDrvData drv_data, char* buf, int len) +{ +} + +void null_ready_output(ErlDrvData drv_data, ErlDrvEvent event) +{ +} + +struct erl_drv_entry async_driver_entry = { + async_drv_init, + async_drv_start, + async_drv_stop, + null_output, + async_drv_input, + null_ready_output, + "async", + NULL, /* finish */ + NULL, /* handle */ + NULL, /* control */ + NULL, /* timeout */ + NULL, /* outputv */ + NULL, /* ready_async */ + NULL, /* flush */ + NULL, /* call */ + NULL, /* event */ + ERL_DRV_EXTENDED_MARKER, + ERL_DRV_EXTENDED_MAJOR_VERSION, + ERL_DRV_EXTENDED_MINOR_VERSION, + 0, /* ERL_DRV_FLAGs */ + NULL, + NULL, /* process_exit */ + stop_select +}; + +#endif + +/* + * Initialises a DriverData structure. + * + * Results: Returns a pointer to a DriverData structure, or NULL + * if the initialsation failed. + */ + +static DriverData* +new_driver_data(port_num, packet_bytes, wait_objs_required, use_threads) + int port_num; /* The port number. */ + int packet_bytes; /* Number of bytes in header. */ + int wait_objs_required; /* The number objects this port is going + /* wait for (typically 1 or 2). */ + int use_threads; /* TRUE if threads are intended to be used. */ +{ + DriverData* dp; + + erts_smp_mtx_lock(&sys_driver_data_lock); + + DEBUGF(("new_driver_data(port_num %d, pb %d)\n", + port_num, packet_bytes)); + + /* + * We used to test first at all that there is enough room in the + * array used by WaitForMultipleObjects(), but that is not necessary + * any more, since driver_select() can't fail. + */ + + /* + * Search for a free slot. + */ + + for (dp = driver_data; dp < driver_data+max_files; dp++) { + if (dp->port_num == PORT_FREE) { + dp->bytesInBuffer = 0; + dp->totalNeeded = packet_bytes; + dp->inBufSize = PORT_BUFSIZ; + dp->inbuf = DRV_BUF_ALLOC(dp->inBufSize); + if (dp->inbuf == NULL) { + erts_smp_mtx_unlock(&sys_driver_data_lock); + return NULL; + } + erts_smp_atomic_add(&sys_misc_mem_sz, dp->inBufSize); + dp->outBufSize = 0; + dp->outbuf = NULL; + dp->port_num = port_num; + dp->packet_bytes = packet_bytes; + dp->port_pid = INVALID_HANDLE_VALUE; + if (init_async_io(&dp->in, use_threads) == -1) + break; + if (init_async_io(&dp->out, use_threads) == -1) + break; + erts_smp_mtx_unlock(&sys_driver_data_lock); + return dp; + } + } + + /* + * Error or no free driver data. + */ + + if (dp < driver_data+max_files) { + release_async_io(&dp->in, dp->port_num); + release_async_io(&dp->out, dp->port_num); + } + erts_smp_mtx_unlock(&sys_driver_data_lock); + return NULL; +} + +static void +release_driver_data(DriverData* dp) +{ + erts_smp_mtx_lock(&sys_driver_data_lock); + +#ifdef ERTS_SMP + /* This is a workaround for the fact that CancelIo cant cancel + requests issued by another thread and that we still cant use + CancelIoEx as that's only availabele in Vista etc. */ + if(dp->in.async_io_active && dp->in.fd != INVALID_HANDLE_VALUE) { + CloseHandle(dp->in.fd); + dp->in.fd = INVALID_HANDLE_VALUE; + DEBUGF(("Waiting for the in event thingie")); + WaitForSingleObject(dp->in.ov.hEvent,INFINITE); + DEBUGF(("...done\n")); + } + if(dp->out.async_io_active && dp->out.fd != INVALID_HANDLE_VALUE) { + CloseHandle(dp->out.fd); + dp->out.fd = INVALID_HANDLE_VALUE; + DEBUGF(("Waiting for the out event thingie")); + WaitForSingleObject(dp->out.ov.hEvent,INFINITE); + DEBUGF(("...done\n")); + } +#else + if (dp->out.thread == (HANDLE) -1 && dp->in.fd != INVALID_HANDLE_VALUE) { + CancelIo(dp->in.fd); + } + if (dp->out.thread == (HANDLE) -1 && dp->out.fd != INVALID_HANDLE_VALUE) { + CancelIo(dp->out.fd); + } +#endif + + if (dp->inbuf != NULL) { + ASSERT(erts_smp_atomic_read(&sys_misc_mem_sz) >= dp->inBufSize); + erts_smp_atomic_add(&sys_misc_mem_sz, -1*dp->inBufSize); + DRV_BUF_FREE(dp->inbuf); + dp->inBufSize = 0; + dp->inbuf = NULL; + } + ASSERT(dp->inBufSize == 0); + + if (dp->outbuf != NULL) { + ASSERT(erts_smp_atomic_read(&sys_misc_mem_sz) >= dp->outBufSize); + erts_smp_atomic_add(&sys_misc_mem_sz, -1*dp->outBufSize); + DRV_BUF_FREE(dp->outbuf); + dp->outBufSize = 0; + dp->outbuf = NULL; + } + ASSERT(dp->outBufSize == 0); + + if (dp->port_pid != INVALID_HANDLE_VALUE) { + CloseHandle(dp->port_pid); + dp->port_pid = INVALID_HANDLE_VALUE; + } + + release_async_io(&dp->in, dp->port_num); + release_async_io(&dp->out, dp->port_num); + + /* + * This must be last, because this function might be executed from + * the exit thread. + */ + + dp->port_num = PORT_FREE; + erts_smp_mtx_unlock(&sys_driver_data_lock); +} + +/* + * Stores input and output file descriptors in the DriverData structure, + * and calls driver_select(). + * + * This function fortunately can't fail! + */ + +static ErlDrvData +set_driver_data(dp, ifd, ofd, read_write, report_exit) + DriverData* dp; + HANDLE ifd; + HANDLE ofd; + int read_write; + int report_exit; +{ + int index = dp - driver_data; + int result; + + dp->in.fd = ifd; + dp->out.fd = ofd; + dp->report_exit = report_exit; + + if (read_write & DO_READ) { + result = driver_select(dp->port_num, (ErlDrvEvent)dp->in.ov.hEvent, + ERL_DRV_READ|ERL_DRV_USE, 1); + ASSERT(result != -1); + async_read_file(&dp->in, dp->inbuf, dp->inBufSize); + } + + if (read_write & DO_WRITE) { + result = driver_select(dp->port_num, (ErlDrvEvent)dp->out.ov.hEvent, + ERL_DRV_WRITE|ERL_DRV_USE, 1); + ASSERT(result != -1); + } + return (ErlDrvData)index; +} + +/* + * Initialises an AsyncIo structure. + */ + +static int +init_async_io(AsyncIo* aio, int use_threads) +{ + aio->flags = 0; + aio->thread = (HANDLE) -1; + aio->fd = INVALID_HANDLE_VALUE; + aio->ov.hEvent = NULL; + aio->ov.Offset = 0L; + aio->ov.OffsetHigh = 0L; + aio->ioAllowed = NULL; + aio->pendingError = 0; + aio->bytesTransferred = 0; +#ifdef ERTS_SMP + aio->async_io_active = 0; +#endif + aio->ov.hEvent = CreateManualEvent(FALSE); + if (aio->ov.hEvent == NULL) + return -1; + if (use_threads) { + aio->ioAllowed = CreateAutoEvent(FALSE); + if (aio->ioAllowed == NULL) + return -1; + } + return 0; +} + +/* + * Releases everything allocated in an AsyncIo structure. + */ + +static void +release_async_io(AsyncIo* aio, ErlDrvPort port_num) +{ + aio->flags = 0; + + if (aio->thread != (HANDLE) -1) + CloseHandle(aio->thread); + aio->thread = (HANDLE) -1; + + if (aio->fd != INVALID_HANDLE_VALUE) + CloseHandle(aio->fd); + aio->fd = INVALID_HANDLE_VALUE; + + if (aio->ov.hEvent != NULL) { + (void) driver_select(port_num, + (ErlDrvEvent)aio->ov.hEvent, + ERL_DRV_USE, 0); + /* was CloseHandle(aio->ov.hEvent); */ + } + + aio->ov.hEvent = NULL; + + if (aio->ioAllowed != NULL) + CloseHandle(aio->ioAllowed); + aio->ioAllowed = NULL; +} + +/* ---------------------------------------------------------------------- + * async_read_file -- + * Initiaties an asynchronous file read, or simulates that using + * the thread associated with this driver data. To get the results, + * call get_overlapped_result(). + * + * Results: + * None. + * ---------------------------------------------------------------------- + */ + +static void +async_read_file(aio, buf, numToRead) + AsyncIo* aio; /* Pointer to driver data. */ + LPVOID buf; /* Pointer to buffer to receive data. */ + DWORD numToRead; /* Number of bytes to read. */ +{ + aio->pendingError = NO_ERROR; +#ifdef HARD_POLL_DEBUG + poll_debug_async_initialized(aio->ov.hEvent); +#endif + if (aio->thread != (HANDLE) -1) { + DEBUGF(("async_read_file: signaling thread 0x%x, event 0x%x\n", + aio->thread, aio->ioAllowed)); + OV_BUFFER_PTR(aio) = buf; + OV_NUM_TO_READ(aio) = numToRead; + ResetEvent(aio->ov.hEvent); + SetEvent(aio->ioAllowed); + } else { +#ifdef ERTS_SMP + aio->async_io_active = 1; /* Will get 0 when the event actually happened */ +#endif + if (ReadFile(aio->fd, buf, numToRead, + &aio->bytesTransferred, &aio->ov)) { + DEBUGF(("async_read_file: ReadFile() suceeded: %d bytes\n", + aio->bytesTransferred)); +#ifdef HARD_POLL_DEBUG + poll_debug_async_immediate(aio->ov.hEvent, aio->bytesTransferred); +#endif + aio->flags |= DF_OVR_READY; + SetEvent(aio->ov.hEvent); + } else { + DWORD error = GetLastError(); + if (error != ERROR_IO_PENDING) { +#ifdef HARD_POLL_DEBUG + poll_debug_async_immediate(aio->ov.hEvent, 0); +#endif + aio->pendingError = error; + SetEvent(aio->ov.hEvent); + } + DEBUGF(("async_read_file: ReadFile() -> %s\n", win32_errorstr(error))); + } + } +} + +/* ---------------------------------------------------------------------- + * async_write_file -- + * Initiaties an asynchronous file write, or simulates that using + * the output thread associated with this driver data. + * To get the results, call get_overlapped_result(). + * + * Results: + * None. + * ---------------------------------------------------------------------- + */ +static int +async_write_file(aio, buf, numToWrite) + AsyncIo* aio; /* Pointer to async control block. */ + LPVOID buf; /* Pointer to buffer with data to write. */ + DWORD numToWrite; /* Number of bytes to write. */ +{ + aio->pendingError = NO_ERROR; + if (aio->thread != (HANDLE) -1) { + DEBUGF(("async_write_file: signaling thread 0x%x, event 0x%x\n", + aio->thread, aio->ioAllowed)); + OV_BUFFER_PTR(aio) = buf; + OV_NUM_TO_READ(aio) = numToWrite; + ResetEvent(aio->ov.hEvent); + SetEvent(aio->ioAllowed); + } else { +#ifdef ERTS_SMP + aio->async_io_active = 1; /* Will get 0 when the event actually happened */ +#endif + if (WriteFile(aio->fd, buf, numToWrite, + &aio->bytesTransferred, &aio->ov)) { + DEBUGF(("async_write_file: WriteFile() suceeded: %d bytes\n", + aio->bytesTransferred)); +#ifdef ERTS_SMP + aio->async_io_active = 0; /* The event will not be signalled */ +#endif + ResetEvent(aio->ov.hEvent); + return TRUE; + } else { + DWORD error = GetLastError(); + if (error != ERROR_IO_PENDING) { + aio->pendingError = error; + SetEvent(aio->ov.hEvent); + } + DEBUGF(("async_write_file: WriteFile() -> %s\n", win32_errorstr(error))); + } + } + return FALSE; +} + +/* ---------------------------------------------------------------------- + * get_overlapped_result -- + * + * Results: + * Returns the error code for the overlapped result, or NO_ERROR + * if no error. + * ---------------------------------------------------------------------- + */ +static int +get_overlapped_result(aio, pBytesRead, wait) + AsyncIo* aio; /* Pointer to async control block. */ + LPDWORD pBytesRead; /* Where to place the number of bytes + * transferred. + */ + BOOL wait; /* If true, wait until result is ready. */ +{ + DWORD error = NO_ERROR; /* Error status from last function. */ + + if (aio->thread != (HANDLE) -1) { + + /* + * Simulate overlapped io with a thread. + */ + DEBUGF(("get_overlapped_result: about to wait for event 0x%x\n", + aio->ov.hEvent)); + error = WaitForSingleObject(aio->ov.hEvent, wait ? INFINITE : 0); + switch (error) { + case WAIT_OBJECT_0: + error = aio->pendingError; + aio->pendingError = NO_ERROR; + *pBytesRead = aio->bytesTransferred; + ResetEvent(aio->ov.hEvent); + DEBUGF(("get_overlapped_result -> %s\n", + win32_errorstr(error))); + return error; + case WAIT_TIMEOUT: + DEBUGF(("get_overlapped_result -> %s\n", + ERROR_IO_INCOMPLETE)); + return ERROR_IO_INCOMPLETE; + case WAIT_FAILED: /* XXX: Shouldn't happen? */ + error = GetLastError(); + DEBUGF(("get_overlapped_result (WAIT_FAILED) -> %s\n", + win32_errorstr(error))); + return error; + } + } else if (aio->pendingError != NO_ERROR) { /* Pending error. */ + error = aio->pendingError; + aio->pendingError = NO_ERROR; + ResetEvent(aio->ov.hEvent); + DEBUGF(("get_overlapped_result: pending error: %s\n", + win32_errorstr(error))); + return error; + } else if (aio->flags & DF_OVR_READY) { /* Operation succeded. */ + aio->flags &= ~DF_OVR_READY; + *pBytesRead = aio->bytesTransferred; + ResetEvent(aio->ov.hEvent); + DEBUGF(("get_overlapped_result: delayed success: %d bytes\n", + aio->bytesTransferred)); + } else if (!GetOverlappedResult(aio->fd, &aio->ov, pBytesRead, wait)) { + error = GetLastError(); + ResetEvent(aio->ov.hEvent); + DEBUGF(("get_overlapped_result: error: %s\n", win32_errorstr(error))); + return error; + } else { /* Success. */ + DEBUGF(("get_overlapped_result: success\n")); + ResetEvent(aio->ov.hEvent); + } + return NO_ERROR; +} + +static int +fd_init(void) +{ + char kernel_dll_name[] = "kernel32"; + HMODULE module; + module = GetModuleHandle(kernel_dll_name); + fpSetHandleInformation = (module != NULL) ? + (BOOL (WINAPI *)(HANDLE,DWORD,DWORD)) + GetProcAddress(module,"SetHandleInformation") : + NULL; + + return 0; +} +static int +spawn_init() +{ + int i; + + driver_data = (struct driver_data *) + erts_alloc(ERTS_ALC_T_DRV_TAB, max_files * sizeof(struct driver_data)); + erts_smp_atomic_add(&sys_misc_mem_sz, max_files*sizeof(struct driver_data)); + for (i = 0; i < max_files; i++) + driver_data[i].port_num = PORT_FREE; + return 0; +} + +static ErlDrvData +spawn_start(ErlDrvPort port_num, char* name, SysDriverOpts* opts) +{ + HANDLE hToChild = INVALID_HANDLE_VALUE; /* Write handle to child. */ + HANDLE hFromChild = INVALID_HANDLE_VALUE; /* Read handle from child. */ + HANDLE hChildStdin = INVALID_HANDLE_VALUE; /* Child's stdin. */ + HANDLE hChildStdout = INVALID_HANDLE_VALUE; /* Child's stout. */ + HANDLE hChildStderr = INVALID_HANDLE_VALUE; /* Child's sterr. */ + int close_child_stderr = 0; + DriverData* dp; /* Pointer to driver data. */ + ErlDrvData retval = ERL_DRV_ERROR_GENERAL; /* Return value. */ + int ok; + int neededSelects = 0; + SECURITY_ATTRIBUTES sa = {sizeof(SECURITY_ATTRIBUTES), NULL, TRUE}; + char* envir = opts->envir; + int errno_return = -1; + + if (opts->read_write & DO_READ) + neededSelects++; + if (opts->read_write & DO_WRITE) + neededSelects++; + + if ((dp = new_driver_data(port_num, opts->packet_bytes, neededSelects, + !use_named_pipes)) == NULL) + return ERL_DRV_ERROR_GENERAL; + + /* + * Create two pipes to communicate with the port program. + */ + + if (opts->read_write & DO_READ) { + if (!create_pipe(&hFromChild, &hChildStdout, FALSE, + opts->overlapped_io)) + goto error; + } else { + hChildStdout = CreateFile("nul", GENERIC_WRITE, 0, + &sa, OPEN_EXISTING, + FILE_ATTRIBUTE_NORMAL, NULL); + DEBUGF(("Created nul file for hChildStdout = %d\n",hChildStdout)); + } + if (opts->read_write & DO_WRITE) { + if (!create_pipe(&hChildStdin, &hToChild, TRUE, opts->overlapped_io)) { + CloseHandle(hFromChild); + hFromChild = INVALID_HANDLE_VALUE; + CloseHandle(hChildStdout); + goto error; + } + } else { + hChildStdin = CreateFile("nul", GENERIC_READ, 0, + &sa, OPEN_EXISTING, + FILE_ATTRIBUTE_NORMAL, NULL); + DEBUGF(("Created nul file for hChildStdin = %d\n",hChildStdin)); + } + + /* + * Make sure that standard error is valid handle, because a Command Prompt + * window not work properly otherwise. We leave standard error alone if + * it is okay and no redirection was specified. + */ + hChildStderr = GetStdHandle(STD_ERROR_HANDLE); + if (opts->redir_stderr) { + hChildStderr = hChildStdout; + } else if (hChildStderr == INVALID_HANDLE_VALUE || hChildStderr == 0) { + hChildStderr = CreateFile("nul", GENERIC_WRITE, 0, &sa, OPEN_EXISTING, + FILE_ATTRIBUTE_NORMAL, NULL); + close_child_stderr = 1; + } + if (fpSetHandleInformation != NULL) { + (*fpSetHandleInformation)(hChildStderr, HANDLE_FLAG_INHERIT, 1); + } + /* + * Spawn the port program. + */ + + DEBUGF(("Spawning \"%s\"\n", name)); + envir = win_build_environment(envir); + ok = CreateChildProcess(name, + hChildStdin, + hChildStdout, + hChildStderr, + &dp->port_pid, + opts->hide_window, + (LPVOID) envir, + (LPTSTR) opts->wd, + opts->spawn_type, + opts->argv, + &errno_return); + CloseHandle(hChildStdin); + CloseHandle(hChildStdout); + if (close_child_stderr && hChildStderr != INVALID_HANDLE_VALUE && + hChildStderr != 0) { + CloseHandle(hChildStderr); + } + if (envir != NULL) { + erts_free(ERTS_ALC_T_ENVIRONMENT, envir); + } + + if (!ok) { + dp->port_pid = INVALID_HANDLE_VALUE; + if (errno_return >= 0) { + retval = ERL_DRV_ERROR_ERRNO; + } + } else { + if (!use_named_pipes) { + if ((opts->read_write & DO_READ) && + !create_file_thread(&dp->in, DO_READ)) + goto error; + if ((opts->read_write & DO_WRITE) && + !create_file_thread(&dp->out, DO_WRITE)) { + dp->in.flags = DF_EXIT_THREAD; + SetEvent(dp->in.ioAllowed); + WaitForSingleObject(dp->in.thread, INFINITE); + dp->in.thread = (HANDLE) -1; + goto error; + } + } +#ifdef HARD_POLL_DEBUG + if (strncmp(name,"inet_gethost",12) == 0) { + erts_printf("Debugging \"%s\"\n", name); + poll_debug_set_active_fd(dp->in.ov.hEvent); + } +#endif + retval = set_driver_data(dp, hFromChild, hToChild, opts->read_write, + opts->exit_status); + } + + if (retval != ERL_DRV_ERROR_GENERAL && retval != ERL_DRV_ERROR_ERRNO) + return retval; + + error: + if (hFromChild != INVALID_HANDLE_VALUE) + CloseHandle(hFromChild); + if (hToChild != INVALID_HANDLE_VALUE) + CloseHandle(hToChild); + release_driver_data(dp); + if (retval == ERL_DRV_ERROR_ERRNO) { + errno = errno_return; + } + return retval; +} + +static int +create_file_thread(AsyncIo* aio, int mode) +{ + DWORD tid; /* Id for thread. */ + + aio->thread = (HANDLE) + _beginthreadex(NULL, 0, + (mode & DO_WRITE) ? threaded_writer : threaded_reader, + aio, 0, &tid); + + return aio->thread != (HANDLE) -1; +} + +/* + * A helper function used by CreateChildProcess(). + * Parses a command line with arguments and returns the length of the + * first part containing the program name. + * Example: input = "\"Program Files\"\\erl arg1 arg2" + * gives 19 as result. + * The length returned is equivalent with length(argv[0]) if the + * comman line should have been prepared by _setargv for the main function +*/ +int parse_command(char* cmd){ +#define NORMAL 2 +#define STRING 1 +#define STOP 0 + int i =0; + int state = NORMAL; + while (cmd[i]) { + switch (cmd[i]) { + case '"': + if (state == NORMAL) + state = STRING; + else + state = NORMAL; + break; + case '\\': + if ((state == STRING) && (cmd[i+1]=='"')) + i++; + break; + case ' ': + if (state == NORMAL) + state = STOP; + break; + default: + break; + } + if (state == STOP) { + return i; + } + i++; + } + return i; +} + +BOOL need_quotes(char *str) +{ + int in_quote = 0; + int backslashed = 0; + int naked_space = 0; + while (*str != '\0') { + switch (*str) { + case '\\' : + backslashed = !backslashed; + break; + case '"': + if (backslashed) { + backslashed=0; + } else { + in_quote = !in_quote; + } + break; + case ' ': + backslashed = 0; + if (!(backslashed || in_quote)) { + naked_space++; + } + break; + default: + backslashed = 0; + } + ++str; + } + return (naked_space > 0); +} + + + +/* + *---------------------------------------------------------------------- + * + * CreateChildProcess -- + * + * Create a child process that has pipes as its + * standard input, output, and error. The child process runs + * synchronously under Win32s and asynchronously under Windows NT + * and Windows 95, and runs with the same environment variables + * as the creating process. + * + * The complete Windows search path is searched to find the specified + * executable. If an executable by the given name is not found, + * automatically tries appending ".com", ".exe", and ".bat" to the + * executable name. + * + * Results: + * The return value is FALSE if there was a problem creating the child process. + * Otherwise, the return value is 0 and *phPid is + * filled with the process id of the child process. + * + * Side effects: + * A process is created. + * + *---------------------------------------------------------------------- + */ + +static BOOL +CreateChildProcess +( + char *origcmd, /* Command line for child process (including + * name of executable). Or whole executable if st is + * ERTS_SPAWN_EXECUTABLE + */ + HANDLE hStdin, /* The standard input handle for child. */ + HANDLE hStdout, /* The standard output handle for child. */ + HANDLE hStderr, /* The standard error handle for child. */ + LPHANDLE phPid, /* Pointer to variable to received PID. */ + BOOL hide, /* Hide the window unconditionally. */ + LPVOID env, /* Environment for the child */ + LPTSTR wd, /* Working dir for the child */ + unsigned st, /* Flags for spawn, tells us how to interpret origcmd */ + char **argv, /* Argument vector if given. */ + int *errno_return /* Place to put an errno in in case of failure */ + ) +{ + PROCESS_INFORMATION piProcInfo = {0}; + STARTUPINFO siStartInfo = {0}; + BOOL ok = FALSE; + int applType; + /* Not to be changed for different types of executables */ + int staticCreateFlags = GetPriorityClass(GetCurrentProcess()); + int createFlags = DETACHED_PROCESS; + char *newcmdline = NULL; + char execPath[MAX_PATH]; + int cmdlength; + char* thecommand; + LPTSTR appname = NULL; + HANDLE hProcess = GetCurrentProcess(); + + *errno_return = -1; + + siStartInfo.cb = sizeof(STARTUPINFO); + siStartInfo.dwFlags = STARTF_USESTDHANDLES; + siStartInfo.hStdInput = hStdin; + siStartInfo.hStdOutput = hStdout; + siStartInfo.hStdError = hStderr; + + + if (st != ERTS_SPAWN_EXECUTABLE) { + /* + * Parse out the program name from the command line (it can be quoted and + * contain spaces). + */ + newcmdline = erts_alloc(ERTS_ALC_T_TMP, 2048); + cmdlength = parse_command(origcmd); + thecommand = (char *) erts_alloc(ERTS_ALC_T_TMP, cmdlength+1); + strncpy(thecommand, origcmd, cmdlength); + thecommand[cmdlength] = '\0'; + DEBUGF(("spawn command: %s\n", thecommand)); + + applType = ApplicationType(thecommand, execPath, TRUE, + TRUE, errno_return); + DEBUGF(("ApplicationType returned for (%s) is %d\n", thecommand, applType)); + erts_free(ERTS_ALC_T_TMP, (void *) thecommand); + if (applType == APPL_NONE) { + erts_free(ERTS_ALC_T_TMP,newcmdline); + return FALSE; + } + newcmdline[0] = '\0'; + + if (applType == APPL_DOS) { + /* + * Under NT, 16-bit DOS applications will not run unless they + * can be attached to a console. Run the 16-bit program as + * a normal process inside of a hidden console application, + * and then run that hidden console as a detached process. + */ + + siStartInfo.wShowWindow = SW_HIDE; + siStartInfo.dwFlags |= STARTF_USESHOWWINDOW; + createFlags = CREATE_NEW_CONSOLE; + strcat(newcmdline, "cmd.exe /c "); + } else if (hide) { + DEBUGF(("hiding window\n")); + siStartInfo.wShowWindow = SW_HIDE; + siStartInfo.dwFlags |= STARTF_USESHOWWINDOW; + createFlags = 0; + } + + strcat(newcmdline, execPath); + strcat(newcmdline, origcmd+cmdlength); + } else { /* ERTS_SPAWN_EXECUTABLE */ + int run_cmd = 0; + applType = ApplicationType(origcmd, execPath, FALSE, FALSE, + errno_return); + if (applType == APPL_NONE) { + return FALSE; + } + if (applType == APPL_DOS) { + /* + * See comment above + */ + + siStartInfo.wShowWindow = SW_HIDE; + siStartInfo.dwFlags |= STARTF_USESHOWWINDOW; + createFlags = CREATE_NEW_CONSOLE; + run_cmd = 1; + } else if (hide) { + DEBUGF(("hiding window\n")); + siStartInfo.wShowWindow = SW_HIDE; + siStartInfo.dwFlags |= STARTF_USESHOWWINDOW; + createFlags = 0; + } + if (run_cmd) { + char cmdPath[MAX_PATH]; + int cmdType; + cmdType = ApplicationType("cmd.exe", cmdPath, TRUE, FALSE, errno_return); + if (cmdType == APPL_NONE || cmdType == APPL_DOS) { + return FALSE; + } + appname = (char *) erts_alloc(ERTS_ALC_T_TMP, strlen(cmdPath)+1); + strcpy(appname,cmdPath); + } else { + appname = (char *) erts_alloc(ERTS_ALC_T_TMP, strlen(execPath)+1); + strcpy(appname,execPath); + } + if (argv == NULL) { + BOOL orig_need_q = need_quotes(execPath); + char *ptr; + int ocl = strlen(execPath); + if (run_cmd) { + newcmdline = (char *) erts_alloc(ERTS_ALC_T_TMP, + ocl + ((orig_need_q) ? 3 : 1) + + 11); + memcpy(newcmdline,"cmd.exe /c ",11); + ptr = newcmdline + 11; + } else { + newcmdline = (char *) erts_alloc(ERTS_ALC_T_TMP, + ocl + ((orig_need_q) ? 3 : 1)); + ptr = newcmdline; + } + if (orig_need_q) { + *ptr++ = '"'; + } + memcpy(ptr,execPath,ocl); + ptr += ocl; + if (orig_need_q) { + *ptr++ = '"'; + } + *ptr = '\0'; + } else { + int sum = 1; /* '\0' */ + char **ar = argv; + char *n; + char *save_arg0 = NULL; + if (argv[0] == erts_default_arg0 || run_cmd) { + save_arg0 = argv[0]; + argv[0] = execPath; + } + if (run_cmd) { + sum += 11; /* cmd.exe /c */ + } + while (*ar != NULL) { + sum += strlen(*ar); + if (need_quotes(*ar)) { + sum += 2; /* quotes */ + } + sum++; /* space */ + ++ar; + } + ar = argv; + newcmdline = erts_alloc(ERTS_ALC_T_TMP, sum); + n = newcmdline; + if (run_cmd) { + memcpy(n,"cmd.exe /c ",11); + n += 11; + } + while (*ar != NULL) { + int q = need_quotes(*ar); + sum = strlen(*ar); + if (q) { + *n++ = '"'; + } + memcpy(n,*ar,sum); + n += sum; + if (q) { + *n++ = '"'; + } + *n++ = ' '; + ++ar; + } + ASSERT(n > newcmdline); + *(n-1) = '\0'; + if (save_arg0 != NULL) { + argv[0] = save_arg0; + } + } + + } + DEBUGF(("Creating child process: %s, createFlags = %d\n", newcmdline, createFlags)); + ok = CreateProcess(appname, + newcmdline, + NULL, + NULL, + TRUE, + createFlags | staticCreateFlags, + env, + wd, + &siStartInfo, + &piProcInfo); + + if (newcmdline != NULL) { + erts_free(ERTS_ALC_T_TMP,newcmdline); + } + if (appname != NULL) { + erts_free(ERTS_ALC_T_TMP,appname); + } + if (!ok) { + DEBUGF(("CreateProcess failed: %s\n", last_error())); + if (*errno_return < 0) { + *errno_return = EACCES; + } + return FALSE; + } + CloseHandle(piProcInfo.hThread); /* Necessary to avoid resource leak. */ + *phPid = piProcInfo.hProcess; + + if (applType == APPL_DOS) { + WaitForSingleObject(hProcess, 50); + } + + /* + * When an application spawns a process repeatedly, a new thread + * instance will be created for each process but the previous + * instances may not be cleaned up. This results in a significant + * virtual memory loss each time the process is spawned. If there + * is a WaitForInputIdle() call between CreateProcess() and + * CloseHandle(), the problem does not occur. PSS ID Number: Q124121 + */ + + WaitForInputIdle(piProcInfo.hProcess, 5000); + + return ok; +} + +/* + * Note, inheritRead == FALSE means "inhetitWrite", i e one of the + * pipe ends is always expected to be inherited. The pipe end that should + * be inherited is opened without overlapped io flags, as the child program + * would expect stdout not to demand overlapped I/O. + */ +static int create_pipe(HANDLE *phRead, HANDLE *phWrite, BOOL inheritRead, BOOL overlapped_io) +{ + SECURITY_ATTRIBUTES sa = {sizeof(SECURITY_ATTRIBUTES), NULL, TRUE}; + char pipe_name[128]; /* Name of pipe. */ + Uint calls; + + /* + * If we should't use named pipes, create anonmous pipes. + */ + + if (!use_named_pipes) { + int success; + HANDLE non_inherited; /* Non-inherited copy of handle. */ + + if (!CreatePipe(phRead, phWrite, &sa, 0)) { + DEBUGF(("Error creating anonyomous pipe: %s\n", last_error())); + return FALSE; + } + + if (inheritRead) { + success = DuplicateHandle(GetCurrentProcess(), *phWrite, + GetCurrentProcess(), &non_inherited, 0, + FALSE, DUPLICATE_SAME_ACCESS); + CloseHandle(*phWrite); + *phWrite = non_inherited; + } else { + success = DuplicateHandle(GetCurrentProcess(), *phRead, + GetCurrentProcess(), &non_inherited, 0, + FALSE, DUPLICATE_SAME_ACCESS); + CloseHandle(*phRead); + *phRead = non_inherited; + } + return success; + } + + + /* + * Otherwise, create named pipes. + */ + + calls = (Uint) erts_smp_atomic_inctest(&pipe_creation_counter); + sprintf(pipe_name, "\\\\.\\pipe\\erlang44_%d_%d", + getpid(), calls); + + DEBUGF(("Creating pipe %s\n", pipe_name)); + sa.bInheritHandle = inheritRead; + if ((*phRead = CreateNamedPipe(pipe_name, + PIPE_ACCESS_INBOUND | + ((inheritRead && !overlapped_io) ? 0 : FILE_FLAG_OVERLAPPED), + PIPE_TYPE_BYTE | PIPE_READMODE_BYTE, + 1, + 0, + 0, + 2000, + &sa)) == NULL) { + DEBUGF(("Error creating pipe: %s\n", last_error())); + return FALSE; + } + + sa.bInheritHandle = !inheritRead; + if ((*phWrite = CreateFile(pipe_name, + GENERIC_WRITE, + 0, /* No sharing */ + &sa, + OPEN_EXISTING, + FILE_ATTRIBUTE_NORMAL | + ((inheritRead || overlapped_io) ? FILE_FLAG_OVERLAPPED : 0), + NULL)) == INVALID_HANDLE_VALUE) { + CloseHandle(*phRead); + DEBUGF(("Error opening other end of pipe: %s\n", last_error())); + return FALSE; + } + return TRUE; +} + + + + +static int ApplicationType +( + const char *originalName, /* Name of the application to find. */ + char fullPath[MAX_PATH], /* Filled with complete path to + * application. */ + BOOL search_in_path, /* If we should search the system wide path */ + BOOL handle_quotes, /* If we should handle quotes around executable */ + int *error_return /* A place to put an error code */ + ) +{ + int applType, i; + HANDLE hFile; + char *ext, *rest; + char buf[2]; + DWORD read; + IMAGE_DOS_HEADER header; + static char extensions[][5] = {"", ".com", ".exe", ".bat"}; + int is_quoted; + int len; + + /* Look for the program as an external program. First try the name + * as it is, then try adding .com, .exe, and .bat, in that order, to + * the name, looking for an executable. + * NOTE! that we does not support execution of .com programs on Windows NT + * + * + * Using the raw SearchPath() procedure doesn't do quite what is + * necessary. If the name of the executable already contains a '.' + * character, it will not try appending the specified extension when + * searching (in other words, SearchPath will not find the program + * "a.b.exe" if the arguments specified "a.b" and ".exe"). + * So, first look for the file as it is named. Then manually append + * the extensions, looking for a match. (') + */ + + len = strlen(originalName); + is_quoted = handle_quotes && len > 0 && originalName[0] == '"' && + originalName[len-1] == '"'; + + applType = APPL_NONE; + *error_return = ENOENT; + for (i = 0; i < (int) (sizeof(extensions) / sizeof(extensions[0])); i++) { + if(is_quoted) { + lstrcpyn(fullPath, originalName+1, MAX_PATH - 7); + len = strlen(fullPath); + if(len > 0) { + fullPath[len-1] = '\0'; + } + } else { + lstrcpyn(fullPath, originalName, MAX_PATH - 5); + } + lstrcat(fullPath, extensions[i]); + SearchPath((search_in_path) ? NULL : ".", fullPath, NULL, MAX_PATH, fullPath, &rest); + + /* + * Ignore matches on directories or data files, return if identified + * a known type. + */ + + if (GetFileAttributes(fullPath) & FILE_ATTRIBUTE_DIRECTORY) { + continue; + } + + ext = strrchr(fullPath, '.'); + if ((ext != NULL) && (strcmpi(ext, ".bat") == 0)) { + *error_return = EACCES; + applType = APPL_DOS; + break; + } + + hFile = CreateFile(fullPath, GENERIC_READ, FILE_SHARE_READ, NULL, + OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, NULL); + if (hFile == INVALID_HANDLE_VALUE) { + continue; + } + + *error_return = EACCES; /* If considered an error, + it's an access error */ + header.e_magic = 0; + ReadFile(hFile, (void *) &header, sizeof(header), &read, NULL); + if (header.e_magic != IMAGE_DOS_SIGNATURE) { + /* + * Doesn't have the magic number for relocatable executables. If + * filename ends with .com, assume it's a DOS application anyhow. + * Note that we didn't make this assumption at first, because some + * supposed .com files are really 32-bit executables with all the + * magic numbers and everything. + */ + + CloseHandle(hFile); + if ((ext != NULL) && (strcmpi(ext, ".com") == 0)) { + applType = APPL_DOS; + break; + } + continue; + } + if (header.e_lfarlc != sizeof(header)) { + /* + * All Windows 3.X and Win32 and some DOS programs have this value + * set here. If it doesn't, assume that since it already had the + * other magic number it was a DOS application. + */ + + CloseHandle(hFile); + applType = APPL_DOS; + break; + } + + /* + * The DWORD at header.e_lfanew points to yet another magic number. + */ + + buf[0] = '\0'; + SetFilePointer(hFile, header.e_lfanew, NULL, FILE_BEGIN); + ReadFile(hFile, (void *) buf, 2, &read, NULL); + CloseHandle(hFile); + + if ((buf[0] == 'L') && (buf[1] == 'E')) { + applType = APPL_DOS; + } else if ((buf[0] == 'N') && (buf[1] == 'E')) { + applType = APPL_WIN3X; + } else if ((buf[0] == 'P') && (buf[1] == 'E')) { + applType = APPL_WIN32; + } else { + continue; + } + break; + } + + if (applType == APPL_NONE) { + return APPL_NONE; + } + + if ((applType == APPL_DOS) || (applType == APPL_WIN3X)) { + /* + * Replace long path name of executable with short path name for + * 16-bit applications. Otherwise the application may not be able + * to correctly parse its own command line to separate off the + * application name from the arguments. + */ + + GetShortPathName(fullPath, fullPath, MAX_PATH); + } + if (is_quoted) { + /* restore quotes on quoted program name */ + len = strlen(fullPath); + memmove(fullPath+1,fullPath,len); + fullPath[0]='"'; + fullPath[len+1]='"'; + fullPath[len+2]='\0'; + } + return applType; +} + +/* + * Thread function used to emulate overlapped reading. + */ + +DWORD WINAPI +threaded_reader(LPVOID param) +{ + AsyncIo* aio = (AsyncIo *) param; + HANDLE thread = GetCurrentThread(); + char* buf; + DWORD numToRead; + + for (;;) { + WaitForSingleObject(aio->ioAllowed, INFINITE); + if (aio->flags & DF_EXIT_THREAD) + break; + buf = OV_BUFFER_PTR(aio); + numToRead = OV_NUM_TO_READ(aio); + aio->pendingError = 0; + if (!ReadFile(aio->fd, buf, numToRead, &aio->bytesTransferred, NULL)) + aio->pendingError = GetLastError(); + else if (aio->flags & DF_XLAT_CR) { + char *s; + int n; + + n = aio->bytesTransferred; + for (s = buf; s < buf+n; s++) { + if (*s == '\r') { + if (s < buf + n - 1 && s[1] == '\n') { + memmove(s, s+1, (buf+n - s - 1)); + --n; + } else { + *s = '\n'; + } + } + } + aio->bytesTransferred = n; + } + SetEvent(aio->ov.hEvent); + if ((aio->flags & DF_XLAT_CR) == 0 && aio->bytesTransferred == 0) { + break; + } + if (aio->pendingError != NO_ERROR) { + break; + } + if (aio->flags & DF_EXIT_THREAD) + break; + } + return 0; +} + +/* + * Thread function used to emulate overlapped writing + */ + +DWORD WINAPI +threaded_writer(LPVOID param) +{ + AsyncIo* aio = (AsyncIo *) param; + HANDLE thread = GetCurrentThread(); + char* buf; + DWORD numToWrite; + int ok; + + for (;;) { + WaitForSingleObject(aio->ioAllowed, INFINITE); + if (aio->flags & DF_EXIT_THREAD) + break; + buf = OV_BUFFER_PTR(aio); + numToWrite = OV_NUM_TO_READ(aio); + aio->pendingError = 0; + ok = WriteFile(aio->fd, buf, numToWrite, &aio->bytesTransferred, NULL); + if (!ok) { + aio->pendingError = GetLastError(); + if (aio->pendingError == ERROR_INVALID_HANDLE && + aio->flags & DF_DROP_IF_INVH) { + /* This is standard error and we'we got an + invalid standard error FD (non-inheritable) from parent. + Just drop the message and be happy. */ + aio->pendingError = 0; + aio->bytesTransferred = numToWrite; + } else if (aio->pendingError == ERROR_NOT_ENOUGH_MEMORY) { + /* This could be a console, which limits utput to 64kbytes, + which might translate to less on a unicode system. + Try 16k chunks and see if it works before giving up. */ + int done = 0; + DWORD transferred; + aio->pendingError = 0; + aio->bytesTransferred = 0; + ok = 1; + while (ok && (numToWrite - done) > 0x4000) { + ok = WriteFile(aio->fd, buf + done, 0x4000, &transferred, NULL); + aio->bytesTransferred += transferred; + done += 0x4000; + } + if (ok && (numToWrite - done) > 0) { + ok = WriteFile(aio->fd, buf + done, (numToWrite - done), + &transferred, NULL); + aio->bytesTransferred += transferred; + } + if (!ok) { + aio->pendingError = GetLastError(); + } + } + } + SetEvent(aio->ov.hEvent); + if (aio->pendingError != NO_ERROR || aio->bytesTransferred == 0) + break; + if (aio->flags & DF_EXIT_THREAD) + break; + } + CloseHandle(aio->fd); + aio->fd = INVALID_HANDLE_VALUE; + return 0; +} + +static HANDLE +translate_fd(int fd) +{ + DWORD access; + HANDLE handle; + + switch (fd) { + case 0: + access = GENERIC_READ; + handle = GetStdHandle(STD_INPUT_HANDLE); + break; + case 1: + access = GENERIC_WRITE; + handle = GetStdHandle(STD_OUTPUT_HANDLE); + break; + case 2: + access = GENERIC_WRITE; + handle = GetStdHandle(STD_ERROR_HANDLE); + break; + default: + return (HANDLE) fd; + } + DEBUGF(("translate_fd(%d) -> std(%d)\n", fd, handle)); + + if (handle == INVALID_HANDLE_VALUE || handle == 0) { + handle = CreateFile("nul", access, 0, + NULL, OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, NULL); + } + DEBUGF(("translate_fd(%d) -> %d\n", fd, handle)); + return handle; +} + +static ErlDrvData +fd_start(ErlDrvPort port_num, char* name, SysDriverOpts* opts) +{ + DriverData* dp; + int is_std_error = (opts->ofd == 2); + + opts->ifd = (int) translate_fd(opts->ifd); + opts->ofd = (int) translate_fd(opts->ofd); + if ((dp = new_driver_data(port_num, opts->packet_bytes, 2, TRUE)) == NULL) + return ERL_DRV_ERROR_GENERAL; + + if (!create_file_thread(&dp->in, DO_READ)) { + dp->port_num = PORT_FREE; + return ERL_DRV_ERROR_GENERAL; + } + + if (!create_file_thread(&dp->out, DO_WRITE)) { + dp->port_num = PORT_FREE; + return ERL_DRV_ERROR_GENERAL; + } + + fd_driver_input = &(dp->in); + dp->in.flags = DF_XLAT_CR; + if (is_std_error) { + dp->out.flags |= DF_DROP_IF_INVH; /* Just drop messages if stderror + is an invalid handle */ + } + return set_driver_data(dp, opts->ifd, opts->ofd, opts->read_write, 0); +} + +static void fd_stop(ErlDrvData d) +{ + int fd = (int)d; + /* + * I don't know a clean way to terminate the threads + * (TerminateThread() doesn't release the stack), + * so will we'll let the threads live. Normally, the fd + * driver is only used to support the -oldshell option, + * so this shouldn't be a problem in practice. + * + * Since we will not attempt to terminate the threads, + * better not close the input or output files either. + */ + + driver_data[fd].in.thread = (HANDLE) -1; + driver_data[fd].out.thread = (HANDLE) -1; + driver_data[fd].in.fd = INVALID_HANDLE_VALUE; + driver_data[fd].out.fd = INVALID_HANDLE_VALUE; + + /*return */ common_stop(fd); +} + +static ErlDrvData +vanilla_start(ErlDrvPort port_num, char* name, SysDriverOpts* opts) +{ + HANDLE ofd,ifd; + DriverData* dp; + DWORD access; /* Access mode: GENERIC_READ, GENERIC_WRITE. */ + DWORD crFlags; + HANDLE this_process = GetCurrentProcess(); + + access = 0; + if (opts->read_write == DO_READ) + access |= GENERIC_READ; + if (opts->read_write == DO_WRITE) + access |= GENERIC_WRITE; + + if (opts->read_write == DO_READ) + crFlags = OPEN_EXISTING; + else if (opts->read_write == DO_WRITE) + crFlags = CREATE_ALWAYS; + else + crFlags = OPEN_ALWAYS; + + if ((dp = new_driver_data(port_num, opts->packet_bytes, 2, FALSE)) == NULL) + return ERL_DRV_ERROR_GENERAL; + ofd = CreateFile(name, access, FILE_SHARE_READ | FILE_SHARE_WRITE, + NULL, crFlags, FILE_ATTRIBUTE_NORMAL, NULL); + if (!DuplicateHandle(this_process, (HANDLE) ofd, + this_process, &ifd, 0, + FALSE, DUPLICATE_SAME_ACCESS)) { + CloseHandle(ofd); + ofd = INVALID_HANDLE_VALUE; + } + if (ofd == INVALID_HANDLE_VALUE) + return ERL_DRV_ERROR_GENERAL; + return set_driver_data(dp, ifd, ofd, opts->read_write,0); +} + +static void +stop(ErlDrvData index) +{ + common_stop((int)index); +} + +static void common_stop(int index) +{ + DriverData* dp = driver_data+index; + + DEBUGF(("common_stop(%d)\n", index)); + + if (dp->in.ov.hEvent != NULL) { + (void) driver_select(dp->port_num, + (ErlDrvEvent)dp->in.ov.hEvent, + ERL_DRV_READ, 0); + } + if (dp->out.ov.hEvent != NULL) { + (void) driver_select(dp->port_num, + (ErlDrvEvent)dp->out.ov.hEvent, + ERL_DRV_WRITE, 0); + } + + if (dp->out.thread == (HANDLE) -1 && dp->in.thread == (HANDLE) -1) { + release_driver_data(dp); + } else { + /* + * If there are read or write threads, start a thread which will + * wait for them to finish. + */ + HANDLE thread; + DWORD tid; + dp->port_num = PORT_EXITING; + thread = (HANDLE *) _beginthreadex(NULL, 0, threaded_exiter, dp, 0, &tid); + CloseHandle(thread); + } +} + +DWORD WINAPI +threaded_exiter(LPVOID param) +{ + DriverData* dp = (DriverData *) param; + HANDLE handles[2]; + int i; + + /* + * Ask the threads to terminated. + * + * Note that we can't reliable test the state of the ioAllowed event, + * because it is an auto reset event. Therefore, always set the + * exit flag and signal the event. + */ + + i = 0; + if (dp->out.thread != (HANDLE) -1) { + dp->out.flags = DF_EXIT_THREAD; + SetEvent(dp->out.ioAllowed); + handles[i++] = dp->out.thread; + } + if (dp->in.thread != (HANDLE) -1) { + dp->in.flags = DF_EXIT_THREAD; + SetEvent(dp->in.ioAllowed); + handles[i++] = dp->in.thread; + } + + /* + * If we were lucky, the following happened above: + * 1) The output thread terminated (and closed the pipe). + * 2) As a consequence of that, the port program received + * EOF on its standard input. + * 3) Hopefully, because of (2), the port program terminated. + * 4) Because of (3), the input thread terminated. + * + * But this might need some time; therefore, we must wait for + * both threads to terminate. + */ + + if (i > 0) { + switch (WaitForMultipleObjects(i, handles, TRUE, 5000)) { + case WAIT_TIMEOUT: + DEBUGF(("Timeout waiting for %d threads failed\n", i)); + break; + case WAIT_FAILED: + DEBUGF(("Wait for %d threads failed: %s\n", + i, win32_errorstr(GetLastError()))); + break; + default: + break; + } + } + + /* + * Wait for threads to terminate didn't help. Now use some force. + * TerminateThread() is *not* a good idea, because it doesn't clean + * up the thread's stack. + * + * Instead we well terminate the port program and wait for the + * threads to terminate themselves when they receive end of file. + */ + + if (dp->out.thread != (HANDLE) -1) { + int error; + + if (WaitForSingleObject(dp->out.thread, 0) == WAIT_OBJECT_0) { + CloseHandle(dp->out.thread); + dp->out.thread = (HANDLE) -1; + } else if (dp->port_pid != INVALID_HANDLE_VALUE) { + DEBUGF(("Killing port process 0x%x (output thread)\n", dp->port_pid)); + TerminateProcess(dp->port_pid, 0); + if (!CloseHandle(dp->port_pid)) + DEBUGF(("Failed to close output handle!!!\n")); + dp->port_pid = INVALID_HANDLE_VALUE; + DEBUGF(("Waiting for output thread 0x%x to finish\n", dp->out.thread)); + error = WaitForSingleObject(dp->out.thread, INFINITE); + } + } + + if (dp->in.thread != (HANDLE) -1) { + if (WaitForSingleObject(dp->in.thread, 0) == WAIT_OBJECT_0) { + CloseHandle(dp->in.thread); + dp->in.thread = (HANDLE) -1; + } else if (dp->port_pid != INVALID_HANDLE_VALUE) { + DEBUGF(("Killing port process 0x%x (input thread)\n", dp->port_pid)); + TerminateProcess(dp->port_pid, 0); + if (!CloseHandle(dp->port_pid)) + DEBUGF(("Failed to close input handle!!!\n")); + dp->port_pid = INVALID_HANDLE_VALUE; + + DEBUGF(("Waiting for input thread 0x%x to finish\n", dp->in.thread)); + switch (WaitForSingleObject(dp->in.thread, INFINITE)) { + case WAIT_OBJECT_0: + CloseHandle(dp->in.thread); + dp->in.thread = (HANDLE) -1; + break; + default: + DEBUGF(("Wait for input thread to finish failed: %s\n", + win32_errorstr(GetLastError()))); + break; + } + } + } + + release_driver_data(dp); + return 0; +} + +/* ---------------------------------------------------------------------- + * output -- + * Outputs data from Erlang to the port program. + * + * Results: + * Returns the actual number of bytes written (including the + * packet header) or -1 if an error occurred. + * ---------------------------------------------------------------------- + */ + +static void +output(ErlDrvData drv_data, char* buf, int len) +/* long drv_data; /* The slot to use in the driver data table. + * For Windows NT, this is *NOT* a file handle. + * The handle is found in the driver data. + */ +/* char *buf; /* Pointer to data to write to the port program. */ +/* int len; /* Number of bytes to write. */ +{ + DriverData* dp; + int pb; /* The header size for this port. */ + int port_num; /* The actual port number (for diagnostics). */ + char* current; + + dp = driver_data + (int)drv_data; + if ((port_num = dp->port_num) == -1) + return ; /*-1;*/ + + pb = dp->packet_bytes; + + if ((pb+len) == 0) + return ; /* 0; */ + + /* + * Check that the message can be sent with given header length. + */ + + if ((pb == 2 && len > 65535) || (pb == 1 && len > 255)) { + driver_failure_posix(port_num, EINVAL); + return ; /* -1; */ + } + + /* + * Allocate memory for both the message and the header. + */ + + ASSERT(dp->outbuf == NULL); + ASSERT(dp->outBufSize == 0); + + ASSERT(!dp->outbuf); + dp->outbuf = DRV_BUF_ALLOC(pb+len); + if (!dp->outbuf) { + driver_failure_posix(port_num, ENOMEM); + return ; /* -1; */ + } + + dp->outBufSize = pb+len; + erts_smp_atomic_add(&sys_misc_mem_sz, dp->outBufSize); + + /* + * Store header bytes (if any). + */ + + current = dp->outbuf; + switch (pb) { + case 4: + *current++ = (len >> 24) & 255; + *current++ = (len >> 16) & 255; + case 2: + *current++ = (len >> 8) & 255; + case 1: + *current++ = len & 255; + } + + /* + * Start the write. + */ + + if (len) + memcpy(current, buf, len); + + if (!async_write_file(&dp->out, dp->outbuf, pb+len)) { + set_busy_port(port_num, 1); + } else { + dp->out.ov.Offset += pb+len; /* For vanilla driver. */ + /* XXX OffsetHigh should be changed too. */ + ASSERT(erts_smp_atomic_read(&sys_misc_mem_sz) >= dp->outBufSize); + erts_smp_atomic_add(&sys_misc_mem_sz, -1*dp->outBufSize); + DRV_BUF_FREE(dp->outbuf); + dp->outBufSize = 0; + dp->outbuf = NULL; + } + /*return 0;*/ +} + + +/* ---------------------------------------------------------------------- + * ready_input -- + * This function is called (indirectly) from check_io() when an + * event object has been signaled, indicating that there is + * something to read on the corresponding file handle. + * + * If the port is working in the continous stream mode (packet_bytes == 0), + * whatever data read will be sent straight to Erlang. + * + * Results: + * Always 0. + * ---------------------------------------------------------------------- + */ + +static void +ready_input(ErlDrvData drv_data, ErlDrvEvent ready_event) +/* long drv_data; /* Driver data. */ +/* HANDLE ready_event; /* The handle for the ready event. */ +{ + int error = 0; /* The error code (assume initially no errors). */ + DWORD bytesRead; /* Number of bytes read. */ + DriverData* dp; + int pb; + + dp = driver_data+(int)drv_data; + pb = dp->packet_bytes; +#ifdef ERTS_SMP + if(dp->in.thread == (HANDLE) -1) { + dp->in.async_io_active = 0; + } +#endif + DEBUGF(("ready_input: dp %p, event 0x%x\n", dp, ready_event)); + + /* + * Evaluate the result of the overlapped read. + */ + +#ifdef HARD_POLL_DEBUG + poll_debug_read_begin(dp->in.ov.hEvent); +#endif + + error = get_overlapped_result(&dp->in, &bytesRead, TRUE); + +#ifdef HARD_POLL_DEBUG + poll_debug_read_done(dp->in.ov.hEvent,bytesRead); +#endif + + if (error == NO_ERROR) { + if (pb == 0) { /* Continous stream. */ +#ifdef DEBUG + DEBUGF(("ready_input: %d: ", bytesRead)); + erl_bin_write(dp->inbuf, 16, bytesRead); + DEBUGF(("\n")); +#endif + driver_output(dp->port_num, dp->inbuf, bytesRead); + } else { /* Packet mode */ + dp->bytesInBuffer += bytesRead; + + /* + * Loop until we've exhausted the data in the buffer. + */ + + for (;;) { + + /* + * Check for completion of a header read. + */ + + if (dp->bytesInBuffer >= dp->totalNeeded && + dp->totalNeeded == pb) { + + /* + * We have successfully read the packet header + * (and perhaps even the packet). Get the packet size + * from the header and update dp->totalNeeded to include + * the packet size. + */ + + int packet_size = 0; + unsigned char *header = (unsigned char *) dp->inbuf; + + switch (pb) { + case 4: + packet_size = (packet_size << 8) | *header++; + packet_size = (packet_size << 8) | *header++; + case 2: + packet_size = (packet_size << 8) | *header++; + case 1: + packet_size = (packet_size << 8) | *header++; + } + + dp->totalNeeded += packet_size; + + /* + * Make sure that the receive buffer is big enough. + */ + + if (dp->inBufSize < dp->totalNeeded) { + char* new_buf; + + new_buf = DRV_BUF_REALLOC(dp->inbuf, dp->totalNeeded); + if (new_buf == NULL) { + error = ERROR_NOT_ENOUGH_MEMORY; + break; /* Break out of loop into error handler. */ + } + ASSERT(erts_smp_atomic_read(&sys_misc_mem_sz) >= dp->inBufSize); + erts_smp_atomic_add(&sys_misc_mem_sz, + dp->totalNeeded - dp->inBufSize); + dp->inBufSize = dp->totalNeeded; + dp->inbuf = new_buf; + } + } + + /* + * Check for completion of a packet read. + */ + + if (dp->bytesInBuffer < dp->totalNeeded) { + /* + * Not enough bytes in the buffer. Break out of + * the loop and initiate a new read. + */ + + break; + } else { + + /* + * We have successfully read a complete packet, which + * can be passed to Erlang. + */ + + driver_output(dp->port_num, dp->inbuf+pb, dp->totalNeeded-pb); + + /* + * Update the number of bytes remaining in the buffer, + * and move the data remaining (if any) to the beginning + * of the buffer. + */ + + dp->bytesInBuffer -= dp->totalNeeded; + if (dp->bytesInBuffer > 0) { + memmove(dp->inbuf, dp->inbuf+dp->totalNeeded, + dp->bytesInBuffer); + } + + /* + * Indicate that we need the size of a header, and + * go through the loop once more (to either process + * remaining bytes or initiate reading more). + */ + + dp->totalNeeded = pb; + } + } + } + } + + /* + * Start a new overlapped read, or report the error. + */ + + if (error == NO_ERROR) { + async_read_file(&dp->in, dp->inbuf+dp->bytesInBuffer, + dp->inBufSize - dp->bytesInBuffer); + } else { + DEBUGF(("ready_input(): error: %s\n", win32_errorstr(error))); + if (error == ERROR_BROKEN_PIPE || error == ERROR_HANDLE_EOF) { + /* Maybe check exit status */ + if (dp->report_exit) { + DWORD exitcode; + if (GetExitCodeProcess(dp->port_pid, &exitcode) && + exitcode != STILL_ACTIVE) { + driver_report_exit(dp->port_num, exitcode); + } + } + driver_failure_eof(dp->port_num); + } else { /* Report real errors. */ + int error = GetLastError(); + (void) driver_select(dp->port_num, ready_event, ERL_DRV_READ, 0); + _dosmaperr(error); + driver_failure_posix(dp->port_num, errno); + } + } + + /*return 0;*/ +} + +static void +ready_output(ErlDrvData drv_data, ErlDrvEvent ready_event) +{ + DWORD bytesWritten; + DriverData* dp = driver_data + (int)drv_data; + int error; + +#ifdef ERTS_SMP + if(dp->out.thread == (HANDLE) -1) { + dp->out.async_io_active = 0; + } +#endif + DEBUGF(("ready_output(%d, 0x%x)\n", drv_data, ready_event)); + set_busy_port(dp->port_num, 0); + if (!(dp->outbuf)) { + /* Happens because event sometimes get signalled during a succesful + write... */ + return; + } + ASSERT(erts_smp_atomic_read(&sys_misc_mem_sz) >= dp->outBufSize); + erts_smp_atomic_add(&sys_misc_mem_sz, -1*dp->outBufSize); + DRV_BUF_FREE(dp->outbuf); + dp->outBufSize = 0; + dp->outbuf = NULL; +#ifdef HARD_POLL_DEBUG + poll_debug_write_begin(dp->out.ov.hEvent); +#endif + error = get_overlapped_result(&dp->out, &bytesWritten, TRUE); +#ifdef HARD_POLL_DEBUG + poll_debug_write_done(dp->out.ov.hEvent,bytesWritten); +#endif + + if (error == NO_ERROR) { + dp->out.ov.Offset += bytesWritten; /* For vanilla driver. */ + return ; /* 0; */ + } + + (void) driver_select(dp->port_num, ready_event, ERL_DRV_WRITE, 0); + _dosmaperr(error); + driver_failure_posix(dp->port_num, errno); + /* return 0; */ +} + +static void stop_select(ErlDrvEvent e, void* _) +{ + CloseHandle((HANDLE)e); +} + +/* Fills in the systems representation of the beam process identifier. +** The Pid is put in STRING representation in the supplied buffer, +** no interpretation of this should be done by the rest of the +** emulator. The buffer should be at least 21 bytes long. +*/ +void sys_get_pid(char *buffer){ + DWORD p = GetCurrentProcessId(); + /* The pid is scalar and is an unsigned long. */ + sprintf(buffer,"%lu",(unsigned long) p); +} + +void +sys_init_io(void) +{ + + /* Now heres an icky one... This is called before drivers are, so we + can change our view of the number of open files possible. + We estimate the number to twice the amount of ports. + We really dont know on windows, do we? */ + max_files = 2*erts_max_ports; + +#ifdef USE_THREADS +#ifdef ERTS_SMP + if (init_async(-1) < 0) + erl_exit(1, "Failed to initialize async-threads\n"); +#else + { + /* This is special stuff, starting a driver from the + * system routines, but is a nice way of handling stuff + * the erlang way + */ + SysDriverOpts dopts; + int ret; + + sys_memset((void*)&dopts, 0, sizeof(SysDriverOpts)); + add_driver_entry(&async_driver_entry); + ret = erts_open_driver(NULL, NIL, "async", &dopts, NULL); + DEBUGF(("open_driver = %d\n", ret)); + if (ret < 0) + erl_exit(1, "Failed to open async driver\n"); + erts_port[ret].status |= ERTS_PORT_SFLG_IMMORTAL; + } +#endif +#endif +} + +#ifdef ERTS_SMP +void +erts_sys_main_thread(void) +{ + HANDLE dummy; +#ifdef ERTS_ENABLE_LOCK_CHECK + erts_lc_set_thread_name("parent_thread"); +#endif + dummy = CreateEvent(NULL, FALSE, FALSE, NULL); + for(;;) { + WaitForSingleObject(dummy, INFINITE); + } +} +#endif + +void erts_sys_alloc_init(void) +{ + elib_ensure_initialized(); +} + +void *erts_sys_alloc(ErtsAlcType_t t, void *x, Uint sz) +{ + return malloc((size_t) sz); +} + +void *erts_sys_realloc(ErtsAlcType_t t, void *x, void *p, Uint sz) +{ + return realloc(p, (size_t) sz); +} + +void erts_sys_free(ErtsAlcType_t t, void *x, void *p) +{ + free(p); +} + +static Preload* preloaded = NULL; +static unsigned* res_name = NULL; +static int num_preloaded = 0; + +/* Return a pointer to a vector of names of preloaded modules */ + +Preload* sys_preloaded(void) +{ + HRSRC hRes; + unsigned char* data; + +#define GETWORD(p) (0[p] | 1[p] << 8) +#define GETDWORD(p) (GETWORD(p) | GETWORD(p+2) << 16) + + + if (preloaded == NULL) { + int i; + ASSERT(beam_module != NULL); + hRes = FindResource(beam_module, 0, "ERLANG_DICT"); + /* We might have a resource compiler laying out the 0 resource with + "0" as a textual name instead... */ + if (hRes == NULL) { + hRes = FindResource(beam_module, "0", "ERLANG_DICT"); + } + if (hRes == NULL) { + DWORD n = GetLastError(); + fprintf(stderr, "No ERLANG_DICT resource\n"); + exit(1); + } + data = (unsigned char *) LoadResource(beam_module, hRes); + + num_preloaded = GETWORD(data); + if (num_preloaded == 0) { + fprintf(stderr, "No preloaded modules\n"); + exit(1); + } + + data += 2; + preloaded = erts_alloc(ERTS_ALC_T_PRELOADED, + (num_preloaded+1)*sizeof(Preload)); + res_name = erts_alloc(ERTS_ALC_T_PRELOADED, + (num_preloaded+1)*sizeof(unsigned)); + erts_smp_atomic_add(&sys_misc_mem_sz, + (num_preloaded+1)*sizeof(Preload) + + (num_preloaded+1)*sizeof(unsigned)); + for (i = 0; i < num_preloaded; i++) { + int n; + + preloaded[i].size = GETDWORD(data); + data += 4; + res_name[i] = GETWORD(data); + data += 2; + n = GETWORD(data); + data += 2; + preloaded[i].name = erts_alloc(ERTS_ALC_T_PRELOADED, n+1); + erts_smp_atomic_add(&sys_misc_mem_sz, n+1); + sys_memcpy(preloaded[i].name, data, n); + preloaded[i].name[n] = '\0'; + data += n; + DEBUGF(("name: %s; size: %d; resource: %p\n", + preloaded[i].name, preloaded[i].size, res_name[i])); + } + preloaded[i].name = NULL; + } + +#undef GETWORD +#undef GETDWORD + return preloaded; +} + +/* Return a pointer to preloaded code for module "module" */ +unsigned char* sys_preload_begin(Preload* pp) +{ + HRSRC hRes; + unsigned resource; + + ASSERT(beam_module != NULL); + + resource = res_name[pp-preloaded]; + DEBUGF(("Loading name: %s; size: %d; resource: %p\n", + pp->name, pp->size, resource)); + hRes = FindResource(beam_module, (char *) resource, "ERLANG_CODE"); + return pp->code = LoadResource(beam_module, hRes); +} + +/* Clean up if allocated */ +void sys_preload_end(Preload* pp) +{ +} + +/* Read a key from console */ + +int +sys_get_key(int fd) +{ + ASSERT(fd == 0); + + if (win_console) { + return ConGetKey(); + } + + /* + * Black magic follows. (Code stolen from get_overlapped_result()) + */ + + if (fd_driver_input != NULL && fd_driver_input->thread != (HANDLE)-1) { + DWORD error; + int key; + + error = WaitForSingleObject(fd_driver_input->ov.hEvent, INFINITE); + if (error == WAIT_OBJECT_0) { + if (fd_driver_input->bytesTransferred > 0) { + int n; + int i; + char* buf = OV_BUFFER_PTR(fd_driver_input); + + fd_driver_input->bytesTransferred--; + n = fd_driver_input->bytesTransferred; + key = buf[0]; + for (i = n; i > 0; i--) { + buf[i-1] = buf[i]; + } + return key; + } + } + } + return '*'; /* Error! */ +} + +/* + * Returns a human-readable description of the last error. + * The returned pointer will be valid only as long as last-error() + * isn't called again. + */ + +char* win32_errorstr(int error) +{ +#ifdef SMP + LPTSTR lpBufPtr = erts_smp_tsd_get(win32_errstr_key); +#else + static LPTSTR lpBufPtr = NULL; +#endif + if (lpBufPtr) { + LocalFree(lpBufPtr); + } + FormatMessage( + FORMAT_MESSAGE_ALLOCATE_BUFFER | FORMAT_MESSAGE_FROM_SYSTEM | + FORMAT_MESSAGE_IGNORE_INSERTS, + NULL, + error, + MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT), + (LPTSTR) &lpBufPtr, + 0, + NULL); + SetLastError(error); +#ifdef ERTS_SMP + erts_smp_tsd_set(win32_errstr_key,lpBufPtr); +#endif + return lpBufPtr; +} + +char* last_error(void) +{ + return win32_errorstr(GetLastError()); +} + +static void* sys_func_memzero(void* s, size_t n) +{ + return sys_memzero(s, n); +} + +#ifdef DEBUG +static HANDLE hDebugWrite = INVALID_HANDLE_VALUE; + +void erl_debug(char *fmt,...) +{ + char sbuf[1024]; /* Temporary buffer. */ + DWORD written; /* Actual number of chars written. */ + va_list va; + + if (hDebugWrite != INVALID_HANDLE_VALUE) { + va_start(va, fmt); + vsprintf(sbuf, fmt, va); + WriteFile(hDebugWrite, sbuf, strlen(sbuf), &written, NULL); + va_end(va); + } +} + +static void debug_console(void) +{ + HANDLE hRead; /* Handle to read end of pipe. */ + SECURITY_ATTRIBUTES sa; + PROCESS_INFORMATION procInfo; + STARTUPINFO startInfo; + BOOL ok; + + /* + * Create a pipe for communicating with the sub process. + */ + + sa.nLength = sizeof(sa); + sa.lpSecurityDescriptor = NULL; + sa.bInheritHandle = TRUE; + if (!CreatePipe(&hRead, &hDebugWrite, &sa, 0)) { + fprintf(stderr, "Failed to create pipe: %d\n", + GetLastError()); + exit(1); + } + + startInfo.cb = sizeof(STARTUPINFO); + startInfo.lpTitle = "Erlang Debug Log"; + startInfo.lpReserved = NULL; + startInfo.lpReserved2 = NULL; + startInfo.cbReserved2 = 0; + startInfo.lpDesktop = NULL; + startInfo.dwFlags = STARTF_USESTDHANDLES; + startInfo.hStdInput = hRead; + + /* The following handles are not intended to be used. */ + startInfo.hStdOutput = GetStdHandle(STD_OUTPUT_HANDLE); + startInfo.hStdError = GetStdHandle(STD_ERROR_HANDLE); + + ok = CreateProcess(NULL, + "erl_log.exe", /* Application */ + NULL, /* Process security attributes. */ + NULL, /* Thread security attributes. */ + TRUE, /* Handle inheritance flag. */ + CREATE_NEW_CONSOLE, /* Flags. */ + NULL, /* Environment. */ + NULL, /* Current directory. */ + &startInfo,/* Startup info. */ + &procInfo /* Process information. */ + ); + + CloseHandle(hRead); + + if (ok) { + /* + * Since we don't use these, close them at once to avoid a resource + * leak. + */ + CloseHandle(procInfo.hProcess); + CloseHandle(procInfo.hThread); + } else { + fprintf(stderr, "Create process failed: %s\n", last_error()); + exit(1); + } +} + +void +erl_bin_write(buf, sz, max) + unsigned char* buf; + int sz; + int max; +{ + int i, imax; + char comma[5] = ","; + + if (hDebugWrite == INVALID_HANDLE_VALUE) + return; + + if (!sz) + return; + if (sz > max) + imax = max; + else + imax = sz; + + for (i=0; i<imax; i++) { + if (i == imax-1) { + if (sz > max) + strcpy(comma, ",..."); + else + comma[0] = 0; + } + if (isdigit(buf[i])) + erl_debug("%u%s", (int)(buf[i]), comma); + else { + if (isalpha(buf[i])) { + erl_debug("%c%s", buf[i], comma); + } + else + erl_debug("%u%s", (int)(buf[i]), comma); + } + } +} + +void +erl_assert_error(char* expr, char* file, int line) +{ + char message[1024]; + + sprintf(message, "File %hs, line %d: %hs", file, line, expr); + MessageBox(GetActiveWindow(), message, "Assertion failed", + MB_OK | MB_ICONERROR); +#if 0 + erl_crash_dump(file, line, "Assertion failed: %hs\n", expr); +#endif + DebugBreak(); +} + +#endif /* DEBUG */ + +static void +check_supported_os_version(void) +{ +#if defined(_WIN32_WINNT) + { + DWORD major = (_WIN32_WINNT >> 8) & 0xff; + DWORD minor = _WIN32_WINNT & 0xff; + + if (int_os_version.dwPlatformId != VER_PLATFORM_WIN32_NT + || int_os_version.dwMajorVersion < major + || (int_os_version.dwMajorVersion == major + && int_os_version.dwMinorVersion < minor)) + erl_exit(-1, + "Windows version not supported " + "(min required: winnt %d.%d)\n", + major, minor); + } +#else + erl_exit(-1, + "Windows version not supported " + "(min required: win %d.%d)\n", + nt_major, nt_minor); +#endif +} + +#ifdef USE_THREADS +static void *ethr_internal_alloc(size_t size) +{ + return erts_alloc_fnf(ERTS_ALC_T_ETHR_INTERNAL, (Uint) size); +} +static void *ethr_internal_realloc(void *ptr, size_t size) +{ + return erts_realloc_fnf(ERTS_ALC_T_ETHR_INTERNAL, ptr, (Uint) size); +} +static void ethr_internal_free(void *ptr) +{ + erts_free(ERTS_ALC_T_ETHR_INTERNAL, ptr); +} +#endif + +void +erts_sys_pre_init(void) +{ + int_os_version.dwOSVersionInfoSize = sizeof(OSVERSIONINFO); + GetVersionEx(&int_os_version); + check_supported_os_version(); +#ifdef USE_THREADS + { + erts_thr_init_data_t eid = ERTS_THR_INIT_DATA_DEF_INITER; + eid.alloc = ethr_internal_alloc; + eid.realloc = ethr_internal_realloc; + eid.free = ethr_internal_free; + erts_thr_init(&eid); +#ifdef ERTS_ENABLE_LOCK_COUNT + erts_lcnt_init(); +#endif + } +#endif + erts_smp_atomic_init(&sys_misc_mem_sz, 0); + erts_sys_env_init(); +} + +/* + * the last two only used for standalone erlang + * they should are used by sae_main in beam dll to + * enable standalone execution via erl_api-routines + */ + +void noinherit_std_handle(DWORD type) +{ + HANDLE h = GetStdHandle(type); + if (h != 0 && h != INVALID_HANDLE_VALUE) { + SetHandleInformation(h,HANDLE_FLAG_INHERIT,0); + } +} + + +void erl_sys_init(void) +{ + HANDLE handle; + + noinherit_std_handle(STD_OUTPUT_HANDLE); + noinherit_std_handle(STD_INPUT_HANDLE); + noinherit_std_handle(STD_ERROR_HANDLE); + + + erts_smp_mtx_init(&sys_driver_data_lock, "sys_driver_data_lock"); + +#ifdef ERTS_SMP + erts_smp_tsd_key_create(&win32_errstr_key); +#endif + erts_smp_atomic_init(&pipe_creation_counter,0); + /* + * Test if we have named pipes or not. + */ + + switch (int_os_version.dwPlatformId) { + case VER_PLATFORM_WIN32_WINDOWS: + DEBUGF(("Running on Windows 95")); + use_named_pipes = FALSE; + break; + case VER_PLATFORM_WIN32_NT: + DEBUGF(("Running on Windows NT")); +#ifdef DISABLE_NAMED_PIPES + use_named_pipes = FALSE; +#else + use_named_pipes = TRUE; +#endif + break; + default: /* Unsupported platform. */ + exit(1); + } + DEBUGF((" %d.%d, build %d, %s\n", + int_os_version.dwMajorVersion, int_os_version.dwMinorVersion, + int_os_version.dwBuildNumber, int_os_version.szCSDVersion)); + + ASSERT(beam_module != NULL); + init_console(); + + /* + * The following makes sure that the current directory for the current drive + * is remembered (in the environment). + */ + + chdir("."); + + /* + * Make sure that the standard error handle is valid. + */ + handle = GetStdHandle(STD_ERROR_HANDLE); + if (handle == INVALID_HANDLE_VALUE || handle == 0) { + SetStdHandle(STD_ERROR_HANDLE, GetStdHandle(STD_OUTPUT_HANDLE)); + } + erts_sys_init_float(); + erts_init_check_io(); + + /* Suppress windows error message popups */ + SetErrorMode(SetErrorMode(0) | + SEM_FAILCRITICALERRORS | SEM_NOOPENFILEERRORBOX); +} + +#ifdef ERTS_SMP +void +erts_sys_schedule_interrupt(int set) +{ + erts_check_io_interrupt(set); +} + +void +erts_sys_schedule_interrupt_timed(int set, long msec) +{ + erts_check_io_interrupt_timed(set, msec); +} +#endif + +/* + * Called from schedule() when it runs out of runnable processes, + * or when Erlang code has performed INPUT_REDUCTIONS reduction + * steps. runnable == 0 iff there are no runnable Erlang processes. + */ +void +erl_sys_schedule(int runnable) +{ +#ifdef ERTS_SMP + erts_check_io(!runnable); + ERTS_SMP_LC_ASSERT(!ERTS_LC_IS_BLOCKING); +#else + erts_check_io_interrupt(0); + if (runnable) { + erts_check_io(0); /* Poll for I/O */ + check_async_ready(); /* Check async completions */ + } else { + erts_check_io(check_async_ready() ? 0 : 1); + } +#endif +} + +#if defined(USE_THREADS) && !defined(ERTS_SMP) +/* + * Async operation support. + */ + +static ErlDrvEvent async_drv_event; + +void +sys_async_ready(int fd) +{ + SetEvent((HANDLE)async_drv_event); +} + +static int +async_drv_init(void) +{ + async_drv_event = (ErlDrvEvent) NULL; + return 0; +} + +static ErlDrvData +async_drv_start(ErlDrvPort port_num, char* name, SysDriverOpts* opts) +{ + if (async_drv_event != (ErlDrvEvent) NULL) { + return ERL_DRV_ERROR_GENERAL; + } + if ((async_drv_event = (ErlDrvEvent)CreateAutoEvent(FALSE)) == (ErlDrvEvent) NULL) { + return ERL_DRV_ERROR_GENERAL; + } + + driver_select(port_num, async_drv_event, ERL_DRV_READ|ERL_DRV_USE, 1); + if (init_async(async_drv_event) < 0) { + return ERL_DRV_ERROR_GENERAL; + } + return (ErlDrvData)port_num; +} + +static void +async_drv_stop(ErlDrvData port_num) +{ + exit_async(); + driver_select((ErlDrvPort)port_num, async_drv_event, ERL_DRV_READ|ERL_DRV_USE, 0); + /*CloseHandle((HANDLE)async_drv_event);*/ + async_drv_event = (ErlDrvEvent) NULL; +} + + +static void +async_drv_input(ErlDrvData port_num, ErlDrvEvent e) +{ + check_async_ready(); + + /* + * Our event is auto-resetting. + */ +} + +#endif + diff --git a/erts/emulator/sys/win32/sys_env.c b/erts/emulator/sys/win32/sys_env.c new file mode 100644 index 0000000000..ac4be3f316 --- /dev/null +++ b/erts/emulator/sys/win32/sys_env.c @@ -0,0 +1,261 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 2002-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include "sys.h" +#include "erl_sys_driver.h" +#include "erl_alloc.h" + +static char* merge_environment(char *current, char *add); +static char* arg_to_env(char **arg); +static char** env_to_arg(char *env); +static char** find_arg(char **arg, char *str); +static int compare(const void *a, const void *b); + +static erts_smp_rwmtx_t environ_rwmtx; + +void +erts_sys_env_init(void) +{ + erts_smp_rwmtx_init(&environ_rwmtx, "environ"); +} + +int +erts_sys_putenv(char *key_value, int sep_ix) +{ + int res; + char sep = key_value[sep_ix]; + ASSERT(sep == '='); + key_value[sep_ix] = '\0'; + erts_smp_rwmtx_rwlock(&environ_rwmtx); + res = (SetEnvironmentVariable((LPCTSTR) key_value, + (LPCTSTR) &key_value[sep_ix+1]) ? 0 : 1); + erts_smp_rwmtx_rwunlock(&environ_rwmtx); + key_value[sep_ix] = sep; + return res; +} + +int +erts_sys_getenv(char *key, char *value, size_t *size) +{ + size_t req_size = 0; + int res = 0; + DWORD new_size; + + erts_smp_rwmtx_rlock(&environ_rwmtx); + SetLastError(0); + new_size = GetEnvironmentVariable((LPCTSTR) key, + (LPTSTR) value, + (DWORD) *size); + res = !new_size && GetLastError() == ERROR_ENVVAR_NOT_FOUND ? -1 : 0; + erts_smp_rwmtx_runlock(&environ_rwmtx); + if (res < 0) + return res; + res = new_size > *size ? 1 : 0; + *size = new_size; + return res; +} + +struct win32_getenv_state { + char *env; + char *next; +}; + + +void init_getenv_state(GETENV_STATE *state) +{ + erts_smp_rwmtx_rlock(&environ_rwmtx); + state->environment_strings = (char *) GetEnvironmentStrings(); + state->next_string = state->environment_strings; +} + +char *getenv_string(GETENV_STATE *state) +{ + ERTS_SMP_LC_ASSERT(erts_smp_lc_rwmtx_is_rlocked(&environ_rwmtx)); + if (state->next_string[0] == '\0') + return NULL; + else { + char *res = state->next_string; + state->next_string += sys_strlen(res) + 1; + return res; + } +} + +void fini_getenv_state(GETENV_STATE *state) +{ + FreeEnvironmentStrings(state->environment_strings); + state->environment_strings = state->next_string = NULL; + erts_smp_rwmtx_runlock(&environ_rwmtx); +} + +char* +win_build_environment(char* new_env) +{ + if (new_env == NULL) { + return NULL; + } else { + char *tmp, *merged; + + erts_smp_rwmtx_rlock(&environ_rwmtx); + tmp = GetEnvironmentStrings(); + merged = merge_environment(tmp, new_env); + + FreeEnvironmentStrings(tmp); + erts_smp_rwmtx_runlock(&environ_rwmtx); + return merged; + } +} + +static char* +merge_environment(char *old, char *add) +{ + char **a_arg = env_to_arg(add); + char **c_arg = env_to_arg(old); + char *ret; + int i, j; + + for(i = 0; c_arg[i] != NULL; ++i) + ; + + for(j = 0; a_arg[j] != NULL; ++j) + ; + + c_arg = erts_realloc(ERTS_ALC_T_TMP, + c_arg, (i+j+1) * sizeof(char *)); + + for(j = 0; a_arg[j] != NULL; ++j){ + char **tmp; + char *current = a_arg[j]; + + if ((tmp = find_arg(c_arg, current)) != NULL) { + if (current[strlen(current)-1] != '=') { + *tmp = current; + } else { + *tmp = c_arg[--i]; + c_arg[i] = NULL; + } + } else if (current[strlen(current)-1] != '=') { + c_arg[i++] = current; + c_arg[i] = NULL; + } + } + ret = arg_to_env(c_arg); + erts_free(ERTS_ALC_T_TMP, c_arg); + erts_free(ERTS_ALC_T_TMP, a_arg); + return ret; +} + +static char** +find_arg(char **arg, char *str) +{ + char *tmp; + int len; + + if ((tmp = strchr(str, '=')) != NULL) { + tmp++; + len = tmp - str; + while (*arg != NULL){ + if (_strnicmp(*arg, str, len) == 0){ + return arg; + } + ++arg; + } + } + return NULL; +} + +static int +compare(const void *a, const void *b) +{ + char *s1 = *((char **) a); + char *s2 = *((char **) b); + char *e1 = strchr(s1,'='); + char *e2 = strchr(s2,'='); + int ret; + int len; + + if(!e1) + e1 = s1 + strlen(s1); + if(!e2) + e2 = s2 + strlen(s2); + + if((e1 - s1) > (e2 - s2)) + len = (e2 - s2); + else + len = (e1 - s1); + + ret = _strnicmp(s1,s2,len); + if (ret == 0) + return ((e1 - s1) - (e2 - s2)); + else + return ret; +} + +static char** +env_to_arg(char *env) +{ + char **ret; + char *tmp; + int i; + int num_strings = 0; + + for(tmp = env; *tmp != '\0'; tmp += strlen(tmp)+1) { + ++num_strings; + } + ret = erts_alloc(ERTS_ALC_T_TMP, sizeof(char *) * (num_strings + 1)); + i = 0; + for(tmp = env; *tmp != '\0'; tmp += strlen(tmp)+1){ + ret[i++] = tmp; + } + ret[i] = NULL; + return ret; +} + +static char* +arg_to_env(char **arg) +{ + char *block; + char *ptr; + int i; + int totlen = 1; /* extra '\0' */ + + for(i = 0; arg[i] != NULL; ++i) { + totlen += strlen(arg[i])+1; + } + + /* sort the environment vector */ + qsort(arg, i, sizeof(char *), &compare); + + if (totlen == 1){ + block = erts_alloc(ERTS_ALC_T_ENVIRONMENT, 2); + block[0] = block[1] = '\0'; + } else { + block = erts_alloc(ERTS_ALC_T_ENVIRONMENT, totlen); + ptr = block; + for(i=0; arg[i] != NULL; ++i){ + strcpy(ptr, arg[i]); + ptr += strlen(ptr)+1; + } + *ptr = '\0'; + } + return block; +} diff --git a/erts/emulator/sys/win32/sys_float.c b/erts/emulator/sys/win32/sys_float.c new file mode 100644 index 0000000000..9e67ca7f48 --- /dev/null +++ b/erts/emulator/sys/win32/sys_float.c @@ -0,0 +1,145 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 1997-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ +/* Float conversions */ + +#include "sys.h" +#include "signal.h" + +/* global variable for floating point checks, (see sys.h) */ +/* Note! This is part of the interface Machine <---> sys.c */ +volatile int erl_fp_exception = 0; + +static void fpe_exception(int sig); + +void +erts_sys_init_float(void) +{ +} +void erts_thread_init_float(void) +{ +} +void erts_thread_disable_fpe(void) +{ +} + +/* + ** These two functions should maybe use localeconv() to pick up + ** the current radix character, but since it is uncertain how + ** expensive such a system call is, and since no-one has heard + ** of other radix characters than '.' and ',' an ad-hoc + ** low execution time solution is used instead. + */ + +int +sys_chars_to_double(char *buf, double *fp) +{ + char *s = buf, *t, *dp; + + /* Robert says that something like this is what he really wanted: + * (The [.,] radix test is NOT what Robert wanted - it was added later) + * + * 7 == sscanf(Tbuf, "%[+-]%[0-9][.,]%[0-9]%[eE]%[+-]%[0-9]%s", ....); + * if (*s2 == 0 || *s3 == 0 || *s4 == 0 || *s6 == 0 || *s7) + * break; + */ + + /* Scan string to check syntax. */ + if (*s == '+' || *s == '-') s++; + if (!isdigit(*s)) /* Leading digits. */ + return -1; + while (isdigit(*s)) s++; + if (*s != '.' && *s != ',')/* Decimal part. */ + return -1; + dp = s++; /* Remember decimal point pos just in case */ + if (!isdigit(*s)) + return -1; + while (isdigit(*s)) s++; + if (*s == 'e' || *s == 'E') { + /* There is an exponent. */ + s++; + if (*s == '+' || *s == '-') s++; + if (!isdigit(*s)) + return -1; + while (isdigit(*s)) s++; + } + if (*s) /* That should be it */ + return -1; + + errno = 0; + *fp = strtod(buf, &t); + if (t != s) { /* Whole string not scanned */ + /* Try again with other radix char */ + *dp = (*dp == '.') ? ',' : '.'; + errno = 0; + *fp = strtod(buf, &t); + if (t != s) { /* Whole string not scanned */ + return -1; + } + } + if (*fp < -1.0e-307 || 1.0e-307 < *fp) { + if (errno == ERANGE) { + return -1; + } + } else { + if (errno == ERANGE) { + /* Special case: Windows (at least some) regard very small + * i.e non-normalized numbers as a range error for strtod(). + * But not for atof. + */ + *fp = atof(buf); + } + } + + return 0; +} + +/* +** Convert a double to ascii format 0.dddde[+|-]ddd +** return number of characters converted +*/ + +int +sys_double_to_chars(double fp, char *buf) +{ + char *s = buf; + + (void) sprintf(buf, "%.20e", fp); + /* Search upto decimal point */ + if (*s == '+' || *s == '-') s++; + while (isdigit(*s)) s++; + if (*s == ',') *s++ = '.'; /* Replace ',' with '.' */ + /* Scan to end of string */ + while (*s) s++; + return s-buf; /* i.e strlen(buf) */ +} + +int +matherr(struct _exception *exc) +{ + erl_fp_exception = 1; + DEBUGF(("FP exception (matherr) (0x%x) (%d)\n", exc->type, erl_fp_exception)); + return 1; +} + +static void +fpe_exception(int sig) +{ + erl_fp_exception = 1; + DEBUGF(("FP exception\n")); +} diff --git a/erts/emulator/sys/win32/sys_interrupt.c b/erts/emulator/sys/win32/sys_interrupt.c new file mode 100644 index 0000000000..d2449a1bdb --- /dev/null +++ b/erts/emulator/sys/win32/sys_interrupt.c @@ -0,0 +1,142 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 1997-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ +/* + * Purpose: Interrupt handling in windows. + */ +#include "sys.h" +#include "erl_alloc.h" +#include "erl_driver.h" +#include "../../drivers/win32/win_con.h" + +#if defined(__GNUC__) +# define WIN_SYS_INLINE __inline__ +#elif defined(__WIN32__) +# define WIN_SYS_INLINE __forceinline +#endif + +#ifdef ERTS_SMP +erts_smp_atomic_t erts_break_requested; +#define ERTS_SET_BREAK_REQUESTED \ + erts_smp_atomic_set(&erts_break_requested, (long) 1) +#define ERTS_UNSET_BREAK_REQUESTED \ + erts_smp_atomic_set(&erts_break_requested, (long) 0) +#else +volatile int erts_break_requested = 0; +#define ERTS_SET_BREAK_REQUESTED (erts_break_requested = 1) +#define ERTS_UNSET_BREAK_REQUESTED (erts_break_requested = 0) +#endif + +extern int nohup; +HANDLE erts_sys_break_event = NULL; + +void erts_do_break_handling(void) +{ + /* + * Most functions that do_break() calls are intentionally not thread safe; + * therefore, make sure that all threads but this one are blocked before + * proceeding! + */ + erts_smp_block_system(0); + /* call the break handling function, reset the flag */ + do_break(); + + ResetEvent(erts_sys_break_event); + ERTS_UNSET_BREAK_REQUESTED; + + erts_smp_release_system(); +} + + +BOOL WINAPI ctrl_handler_ignore_break(DWORD dwCtrlType) +{ + switch (dwCtrlType) { + case CTRL_C_EVENT: + case CTRL_BREAK_EVENT: + return TRUE; + break; + case CTRL_LOGOFF_EVENT: + if (nohup) + return TRUE; + /* else pour through... */ + case CTRL_CLOSE_EVENT: + case CTRL_SHUTDOWN_EVENT: + erl_exit(0, ""); + break; + } + return TRUE; +} + +void erts_set_ignore_break(void) { + ConSetCtrlHandler(ctrl_handler_ignore_break); + SetConsoleCtrlHandler(ctrl_handler_ignore_break, TRUE); +} + +BOOL WINAPI ctrl_handler_replace_intr(DWORD dwCtrlType) +{ + switch (dwCtrlType) { + case CTRL_C_EVENT: + return FALSE; + case CTRL_BREAK_EVENT: + SetEvent(erts_sys_break_event); + break; + case CTRL_LOGOFF_EVENT: + if (nohup) + return TRUE; + /* else pour through... */ + case CTRL_CLOSE_EVENT: + case CTRL_SHUTDOWN_EVENT: + erl_exit(0, ""); + break; + } + return TRUE; +} + + +/* Don't use ctrl-c for break handler but let it be + used by the shell instead (see user_drv.erl) */ +void erts_replace_intr(void) { + ConSetCtrlHandler(ctrl_handler_replace_intr); + SetConsoleCtrlHandler(ctrl_handler_replace_intr, TRUE); +} + +BOOL WINAPI ctrl_handler(DWORD dwCtrlType) +{ + switch (dwCtrlType) { + case CTRL_C_EVENT: + case CTRL_BREAK_EVENT: + SetEvent(erts_sys_break_event); + break; + case CTRL_LOGOFF_EVENT: + if (nohup) + return TRUE; + /* else pour through... */ + case CTRL_CLOSE_EVENT: + case CTRL_SHUTDOWN_EVENT: + erl_exit(0, ""); + break; + } + return TRUE; +} + +void init_break_handler() +{ + ConSetCtrlHandler(ctrl_handler); + SetConsoleCtrlHandler(ctrl_handler, TRUE); +} + diff --git a/erts/emulator/sys/win32/sys_time.c b/erts/emulator/sys/win32/sys_time.c new file mode 100644 index 0000000000..50e43065b5 --- /dev/null +++ b/erts/emulator/sys/win32/sys_time.c @@ -0,0 +1,96 @@ +/* + * %CopyrightBegin% + * + * Copyright Ericsson AB 1997-2009. All Rights Reserved. + * + * The contents of this file are subject to the Erlang Public License, + * Version 1.1, (the "License"); you may not use this file except in + * compliance with the License. You should have received a copy of the + * Erlang Public License along with this software. If not, it can be + * retrieved online at http://www.erlang.org/. + * + * Software distributed under the License is distributed on an "AS IS" + * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See + * the License for the specific language governing rights and limitations + * under the License. + * + * %CopyrightEnd% + */ +/* + * Purpose: System-dependent time functions. + */ + +#include "sys.h" +#include "assert.h" + +#ifdef __GNUC__ +#define LL_LITERAL(X) X##LL +#else +#define LL_LITERAL(X) X##i64 +#endif + +/******************* Routines for time measurement *********************/ + +#define EPOCH_JULIAN_DIFF LL_LITERAL(11644473600) + +static SysHrTime wrap = 0; +static DWORD last_tick_count = 0; + +int +sys_init_time(void) +{ + return 1; +} + +void +sys_gettimeofday(SysTimeval *tv) +{ + SYSTEMTIME t; + FILETIME ft; + LONGLONG lft; + + GetSystemTime(&t); + SystemTimeToFileTime(&t, &ft); + memcpy(&lft, &ft, sizeof(lft)); + tv->tv_usec = (long) ((lft / LL_LITERAL(10)) % LL_LITERAL(1000000)); + tv->tv_sec = (long) ((lft / LL_LITERAL(10000000)) - EPOCH_JULIAN_DIFF); +} + +SysHrTime +sys_gethrtime(void) +{ + DWORD ticks = (SysHrTime) (GetTickCount() & 0x7FFFFFFF); + if (ticks < (SysHrTime) last_tick_count) { + wrap += LL_LITERAL(1) << 31; + } + last_tick_count = ticks; + return ((((LONGLONG) ticks) + wrap) * LL_LITERAL(1000000)); +} + +clock_t +sys_times(SysTimes *buffer) { + clock_t kernel_ticks = (GetTickCount() / + (1000 / SYS_CLK_TCK)) & 0x7FFFFFFF; + FILETIME dummy; + LONGLONG user; + LONGLONG system; + + buffer->tms_utime = buffer->tms_stime = buffer->tms_cutime = + buffer->tms_cstime = 0; + + if (GetProcessTimes(GetCurrentProcess(), &dummy, &dummy, + (FILETIME *) &system, (FILETIME *) &user) == 0) + return kernel_ticks; + system /= (LONGLONG)(10000000 / SYS_CLK_TCK); + user /= (LONGLONG)(10000000 / SYS_CLK_TCK); + + buffer->tms_utime = (clock_t) (user & LL_LITERAL(0x7FFFFFFF)); + buffer->tms_stime = (clock_t) (system & LL_LITERAL(0x7FFFFFFF)); + return kernel_ticks; +} + + + + + + |