diff options
Diffstat (limited to 'erts/emulator/sys/unix/sys.c')
| -rw-r--r-- | erts/emulator/sys/unix/sys.c | 1955 |
1 files changed, 19 insertions, 1936 deletions
diff --git a/erts/emulator/sys/unix/sys.c b/erts/emulator/sys/unix/sys.c index d94b37430e..6fb86f6dda 100644 --- a/erts/emulator/sys/unix/sys.c +++ b/erts/emulator/sys/unix/sys.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2014. All Rights Reserved. + * Copyright Ericsson AB 1996-2016. All Rights Reserved. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -49,7 +49,6 @@ #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 */ @@ -67,7 +66,7 @@ #include "erl_mseg.h" extern char **environ; -static erts_smp_rwmtx_t environ_rwmtx; +erts_smp_rwmtx_t environ_rwmtx; #define MAX_VSIZE 16 /* Max number of entries allowed in an I/O * vector sock_sendv(). @@ -76,89 +75,12 @@ static erts_smp_rwmtx_t environ_rwmtx; * Don't need global.h, but bif_table.h (included by bif.h), * won't compile otherwise */ -#include "global.h" +#include "global.h" #include "bif.h" -#include "erl_sys_driver.h" #include "erl_check_io.h" #include "erl_cpu_topology.h" -#ifndef DISABLE_VFORK -#define DISABLE_VFORK 0 -#endif - -#if defined IOV_MAX -#define MAXIOV IOV_MAX -#elif defined UIO_MAXIOV -#define MAXIOV UIO_MAXIOV -#else -#define MAXIOV 16 -#endif - -#ifdef USE_THREADS -# ifdef ENABLE_CHILD_WAITER_THREAD -# define CHLDWTHR ENABLE_CHILD_WAITER_THREAD -# else -# define CHLDWTHR 0 -# endif -# define FDBLOCK 1 -#else -# define CHLDWTHR 0 -# define FDBLOCK 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 -}; - -/* Used by the fd driver iff the fd could not be set to non-blocking */ -typedef struct ErtsSysBlocking_ { - ErlDrvPDL pdl; - int res; - int err; - unsigned int pkey; -} ErtsSysBlocking; - - -/* This data is shared by these drivers - initialized by spawn_init() */ -static struct driver_data { - ErlDrvPort port_num; - int ofd, packet_bytes; - ErtsSysReportExit *report_exit; - int pid; - int alive; - int status; - int terminating; - ErtsSysBlocking *blocking; -} *driver_data; /* indexed by fd */ - -static ErtsSysReportExit *report_exit_list; -#if CHLDWTHR && !defined(ERTS_SMP) -static ErtsSysReportExit *report_exit_transit_list; -#endif - extern int driver_interrupt(int, int); extern void do_break(void); @@ -170,33 +92,6 @@ extern void erts_sys_init_float(void); extern void erl_crash_dump(char* file, int line, char* fmt, ...); -#define DIR_SEPARATOR_CHAR '/' - -#if defined(__ANDROID__) -#define SHELL "/system/bin/sh" -#else -#define SHELL "/bin/sh" -#endif /* __ANDROID__ */ - - -#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; @@ -220,58 +115,20 @@ static volatile int have_prepared_crash_dump; (have_prepared_crash_dump++) #endif -static erts_smp_atomic_t sys_misc_mem_sz; +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}; +#if !defined(ETHR_UNUSABLE_SIGUSRX) && defined(ERTS_THR_HAVE_SIG_FUNCS) static int sig_suspend_fds[2] = {-1, -1}; #define ERTS_SYS_SUSPEND_SIGNAL SIGUSR2 - #endif -jmp_buf erts_sys_sigsegv_jmp; - -#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 void note_child_death(int, int); - -#if CHLDWTHR -static void* child_waiter(void *); -#endif +jmp_buf erts_sys_sigsegv_jmp; static int crashdump_companion_cube_fd = -1; @@ -453,9 +310,10 @@ MALLOC_USE_HASH(1); #ifdef USE_THREADS #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 + * guarantee that the main thread will receive all SIGINT, and * SIGUSR1 signals sent to the process, we block these signals in the * parent thread when creating a new thread. */ @@ -551,14 +409,11 @@ erts_sys_pre_init(void) #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 @@ -569,17 +424,6 @@ erts_sys_pre_init(void) #ifdef USE_THREADS -#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_atomic32_init_nob(&erts_break_requested, 0); erts_smp_atomic32_init_nob(&erts_got_sigusr1, 0); @@ -589,9 +433,6 @@ erts_sys_pre_init(void) erts_got_sigusr1 = 0; have_prepared_crash_dump = 0; #endif -#if !CHLDWTHR && !defined(ERTS_SMP) - children_died = 0; -#endif #endif /* USE_THREADS */ @@ -628,39 +469,6 @@ erts_sys_pre_init(void) 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_raw("BINDIR", bindir, &bindirsz); - if (res != 0) { - if (res < 0) - erts_exit(1, - "Environment variable BINDIR is not set\n"); - if (res > 0) - erts_exit(1, - "Value of environment variable BINDIR is too large\n"); - } - if (bindir[0] != DIR_SEPARATOR_CHAR) - erts_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_nob(&sys_misc_mem_sz, csp_path_sz); - erts_snprintf(child_setup_prog, csp_path_sz, - "%s%c%s", - bindir, - DIR_SEPARATOR_CHAR, - CHILD_SETUP_PROG_NAME); - } -#endif #ifdef USE_SETLINEBUF setlinebuf(stdout); @@ -872,7 +680,7 @@ sigusr1_exit(void) #else -#ifdef ERTS_SMP +#ifdef ERTS_SYS_SUSPEND_SIGNAL void sys_thr_suspend(erts_tid_t tid) { erts_thr_kill(tid, ERTS_SYS_SUSPEND_SIGNAL); @@ -900,7 +708,7 @@ static RETSIGTYPE user_signal1(int signum) #endif } -#ifdef ERTS_SMP +#ifdef ERTS_SYS_SUSPEND_SIGNAL #if (defined(SIG_SIGSET) || defined(SIG_SIGNAL)) static RETSIGTYPE suspend_signal(void) #else @@ -913,7 +721,7 @@ static RETSIGTYPE suspend_signal(int signum) res = read(sig_suspend_fds[0], buf, sizeof(int)); } while (res < 0 && errno == EINTR); } -#endif /* #ifdef ERTS_SMP */ +#endif /* #ifdef ERTS_SYS_SUSPEND_SIGNAL */ #endif /* #ifndef ETHR_UNUSABLE_SIGUSRX */ @@ -966,53 +774,20 @@ void init_break_handler(void) sys_signal(SIGINT, request_break); #ifndef ETHR_UNUSABLE_SIGUSRX sys_signal(SIGUSR1, user_signal1); -#ifdef ERTS_SMP - sys_signal(ERTS_SYS_SUSPEND_SIGNAL, suspend_signal); -#endif /* #ifdef ERTS_SMP */ #endif /* #ifndef ETHR_UNUSABLE_SIGUSRX */ sys_signal(SIGQUIT, do_quit); } -int sys_max_files(void) +void sys_init_suspend_handler(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 /* #ifndef ETHR_UNUSABLE_SIGUSRX */ -#endif /* #ifndef ERTS_SMP */ - -#if defined(ERTS_SMP) && !defined(ETHR_UNUSABLE_SIGUSRX) - sys_sigblock(ERTS_SYS_SUSPEND_SIGNAL); +#ifdef ERTS_SYS_SUSPEND_SIGNAL + sys_signal(ERTS_SYS_SUSPEND_SIGNAL, suspend_signal); #endif - } -static void unblock_signals(void) +int sys_max_files(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 /* #ifndef ERTS_SMP */ - -#if defined(ERTS_SMP) && !defined(ETHR_UNUSABLE_SIGUSRX) - sys_sigrelease(ERTS_SYS_SUSPEND_SIGNAL); -#endif - + return(max_files); } /************************** OS info *******************************/ @@ -1102,1502 +877,6 @@ void fini_getenv_state(GETENV_STATE *state) 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) - -/* Driver interfaces */ -static ErlDrvData spawn_start(ErlDrvPort, char*, SysDriverOpts*); -static ErlDrvData fd_start(ErlDrvPort, char*, SysDriverOpts*); -#if FDBLOCK -static void fd_async(void *); -static void fd_ready_async(ErlDrvData drv_data, ErlDrvThreadData thread_data); -#endif -static ErlDrvSSizeT fd_control(ErlDrvData, unsigned int, char *, ErlDrvSizeT, - char **, ErlDrvSizeT); -static ErlDrvData vanilla_start(ErlDrvPort, char*, SysDriverOpts*); -static int spawn_init(void); -static void fd_stop(ErlDrvData); -static void fd_flush(ErlDrvData); -static void stop(ErlDrvData); -static void ready_input(ErlDrvData, ErlDrvEvent); -static void ready_output(ErlDrvData, ErlDrvEvent); -static void output(ErlDrvData, char*, ErlDrvSizeT); -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, -#if FDBLOCK - fd_ready_async, /* ready_async */ -#else - NULL, -#endif - fd_flush, /* 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 -}; - -/* 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_AS_INTR(); /* Make sure we don't sleep in poll */ -#endif -} - -static int set_blocking_data(struct driver_data *dd) { - - dd->blocking = erts_alloc(ERTS_ALC_T_SYS_BLOCKING, sizeof(ErtsSysBlocking)); - - erts_smp_atomic_add_nob(&sys_misc_mem_sz, sizeof(ErtsSysBlocking)); - - dd->blocking->pdl = driver_pdl_create(dd->port_num); - dd->blocking->res = 0; - dd->blocking->err = 0; - dd->blocking->pkey = driver_async_port_key(dd->port_num); - - return 1; -} - -static int set_driver_data(ErlDrvPort port_num, - int ifd, - int ofd, - int packet_bytes, - int read_write, - int exit_status, - int pid, - int is_blocking) -{ - Port *prt; - 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_drvport2id(port_num); - 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; - } - - prt = erts_drvport2port(port_num); - if (prt != ERTS_INVALID_ERL_DRV_PORT) - prt->os_pid = pid; - - 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; - driver_data[ifd].terminating = 0; - driver_data[ifd].blocking = NULL; - if (read_write & DO_WRITE) { - driver_data[ifd].ofd = ofd; - if (is_blocking && FDBLOCK) - if (!set_blocking_data(driver_data+ifd)) - return -1; - 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; - driver_data[ofd].terminating = 0; - driver_data[ofd].blocking = NULL; - if (is_blocking && FDBLOCK) - if (!set_blocking_data(driver_data+ofd)) - return -1; - 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 */ - thr_opts.name = "child_waiter"; -#endif - - sys_signal(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_nob(&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_signal(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, ErlDrvPort port_num) -{ - 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; ) { - size_t last = strlen(cpp[j])-1; - if (cpp[j][last] == '=' && strchr(cpp[j], '=') == cpp[j]+last) { - cpp[j] = cpp[--len]; - if (len < i) { - i--; - } else { - j++; - } - } - else { - j++; - } - } - - 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_raw("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_sched_bind_atfork_prepare(); - -#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 (erts_sched_bind_atfork_child(unbind) != 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; - } - -#if defined(HAVE_CLOSEFROM) - closefrom(opts->use_stdio ? 3 : 5); -#else - for (i = opts->use_stdio ? 3 : 5; i < max_files; i++) - (void) close(i); -#endif - - 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(SHELL, "sh", "-c", cmd_line, (char *) NULL, new_environ); - } - child_error: - _exit(1); - } -#if !DISABLE_VFORK - } -#define ENOUGH_BYTES (44) - else { /* Use vfork() */ - char **cs_argv= erts_alloc(ERTS_ALC_T_TMP,(CS_ARGV_NO_OF_ARGS + 1)* - sizeof(char *)); - char fd_close_range[ENOUGH_BYTES]; /* 44 bytes are enough to */ - char dup2_op[CS_ARGV_NO_OF_DUP2_OPS][ENOUGH_BYTES]; /* 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 */ - erts_snprintf(&dup2_op[i++][0], ENOUGH_BYTES, "%d:%d", ifd[1], 1); - if(opts->redir_stderr) - /* stderr for process */ - erts_snprintf(&dup2_op[i++][0], ENOUGH_BYTES, "%d:%d", ifd[1], 2); - } - if (opts->read_write & DO_WRITE) - /* stdin for process */ - erts_snprintf(&dup2_op[i++][0], ENOUGH_BYTES, "%d:%d", ofd[0], 0); - } else { /* XXX will fail if ofd[0] == 4 (unlikely..) */ - if (opts->read_write & DO_READ) - erts_snprintf(&dup2_op[i++][0], ENOUGH_BYTES, "%d:%d", ifd[1], 4); - if (opts->read_write & DO_WRITE) - erts_snprintf(&dup2_op[i++][0], ENOUGH_BYTES, "%d:%d", ofd[0], 3); - } - for (; i < CS_ARGV_NO_OF_DUP2_OPS; i++) - strcpy(&dup2_op[i][0], "-"); - erts_snprintf(fd_close_range, ENOUGH_BYTES, "%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] = erts_sched_bind_atvfork_child(unbind); - 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); - } -#undef ENOUGH_BYTES -#endif - - erts_sched_bind_atfork_parent(unbind); - - 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, SHELL, SHELL, "-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, 0); - /* 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 ErlDrvSSizeT fd_control(ErlDrvData drv_data, - unsigned int command, - char *buf, ErlDrvSizeT len, - char **rbuf, ErlDrvSizeT 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; - int non_blocking = 0; - - 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... - * - * Added note OTP 18: Some systems seem to use stdout/stderr to log data - * using unix pipes, so we cannot allow the system to block on a write. - * Therefore we use an async thread to write the data to fd's that could - * not be set to non-blocking. When no async threads are available we - * fall back on the old behaviour. - * - * Also the guarantee about what is delivered to the OS has changed. - * Pre 18 the fd driver did no flushing of data before terminating. - * Now it does. This is because we want to be able to guarantee that things - * such as escripts and friends really have outputted all data before - * terminating. This could potentially block the termination of the system - * for a very long time, but if the user wants to terminate fast she should - * use erlang:halt with flush=false. - */ - - 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. */ - - non_blocking = 1; - 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); - non_blocking = 1; - 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, - !non_blocking); - 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_nob(&sys_misc_mem_sz) >= fd_data[fd].sz); - erts_smp_atomic_add_nob(&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(ErlDrvPort prt, int fd) -{ - driver_select(prt,fd,DO_READ|DO_WRITE,0); - clear_fd_data(fd); - SET_BLOCKING(fd); -} - -static void fd_stop(ErlDrvData ev) /* Does not close the fds */ -{ - int ofd; - int fd = (int)(long)ev; - ErlDrvPort prt = driver_data[fd].port_num; - -#if FDBLOCK - if (driver_data[fd].blocking) { - erts_free(ERTS_ALC_T_SYS_BLOCKING,driver_data[fd].blocking); - driver_data[fd].blocking = NULL; - erts_smp_atomic_add_nob(&sys_misc_mem_sz, -1*sizeof(ErtsSysBlocking)); - } -#endif - - nbio_stop_fd(prt, fd); - ofd = driver_data[fd].ofd; - if (ofd != fd && ofd != -1) - nbio_stop_fd(prt, ofd); -} - -static void fd_flush(ErlDrvData fd) -{ - if (!driver_data[(int)(long)fd].terminating) - driver_data[(int)(long)fd].terminating = 1; -} - -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, 0); - 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) -{ - ErlDrvPort prt; - int 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. */ - 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); */ - } -} - -/* used by fd_driver */ -static void outputv(ErlDrvData e, ErlIOVec* ev) -{ - int fd = (int)(long)e; - ErlDrvPort ix = driver_data[fd].port_num; - int pb = driver_data[fd].packet_bytes; - int ofd = driver_data[fd].ofd; - ssize_t n; - ErlDrvSizeT sz; - char lb[4]; - char* lbp; - ErlDrvSizeT len = ev->size; - - /* (len > ((unsigned long)-1 >> (4-pb)*8)) */ - /* if (pb >= 0 && (len & (((ErlDrvSizeT)1 << (pb*8))) - 1) != len) {*/ - if (((pb == 2) && (len > 0xffff)) || (pb == 1 && len > 0xff)) { - driver_failure_posix(ix, EINVAL); - return; /* -1; */ - } - /* Handles 0 <= pb <= 4 only */ - put_int32((Uint32) len, lb); - lbp = lb + (4-pb); - - ev->iov[0].iov_base = lbp; - ev->iov[0].iov_len = pb; - ev->size += pb; - - if (driver_data[fd].blocking && FDBLOCK) - driver_pdl_lock(driver_data[fd].blocking->pdl); - - if ((sz = driver_sizeq(ix)) > 0) { - driver_enqv(ix, ev, 0); - - if (driver_data[fd].blocking && FDBLOCK) - driver_pdl_unlock(driver_data[fd].blocking->pdl); - - if (sz + ev->size >= (1 << 13)) - set_busy_port(ix, 1); - } - else if (!driver_data[fd].blocking || !FDBLOCK) { - /* We try to write directly if the fd in non-blocking */ - 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); - } -#if FDBLOCK - else { - if (ev->size != 0) { - driver_enqv(ix, ev, 0); - driver_pdl_unlock(driver_data[fd].blocking->pdl); - driver_async(ix, &driver_data[fd].blocking->pkey, - fd_async, driver_data+fd, NULL); - } else { - driver_pdl_unlock(driver_data[fd].blocking->pdl); - } - } -#endif - /* return 0;*/ -} - -/* Used by spawn_driver and vanilla driver */ -static void output(ErlDrvData e, char* buf, ErlDrvSizeT len) -{ - int fd = (int)(long)e; - ErlDrvPort ix = driver_data[fd].port_num; - int pb = driver_data[fd].packet_bytes; - int ofd = driver_data[fd].ofd; - ssize_t n; - ErlDrvSizeT 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(ErlDrvPort 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 (driver_data[ready_fd].blocking && FDBLOCK) { - driver_pdl_lock(driver_data[ready_fd].blocking->pdl); - if (driver_sizeq(driver_data[ready_fd].port_num) > 0) { - driver_pdl_unlock(driver_data[ready_fd].blocking->pdl); - /* We have stuff in the output queue, so we just - set the state to terminating and wait for fd_async_ready - to terminate the port */ - if (res == 0) - driver_data[ready_fd].terminating = 2; - else - driver_data[ready_fd].terminating = -err; - return 0; - } - driver_pdl_unlock(driver_data[ready_fd].blocking->pdl); - } - - 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; - ErlDrvPort 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_nob(&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; - ErlDrvPort 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); - if (driver_data[fd].terminating) - driver_failure_atom(driver_data[fd].port_num,"normal"); - 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); -} - -#if FDBLOCK - -static void -fd_async(void *async_data) -{ - int res; - struct driver_data *dd = (struct driver_data*)async_data; - SysIOVec *iov0; - SysIOVec *iov; - int iovlen; - int err = 0; - /* much of this code is stolen from efile_drv:invoke_writev */ - driver_pdl_lock(dd->blocking->pdl); - iov0 = driver_peekq(dd->port_num, &iovlen); - iovlen = iovlen < MAXIOV ? iovlen : MAXIOV; - iov = erts_alloc_fnf(ERTS_ALC_T_SYS_WRITE_BUF, - sizeof(SysIOVec)*iovlen); - if (!iov) { - res = -1; - err = ENOMEM; - driver_pdl_unlock(dd->blocking->pdl); - } else { - memcpy(iov,iov0,iovlen*sizeof(SysIOVec)); - driver_pdl_unlock(dd->blocking->pdl); - - do { - res = writev(dd->ofd, iov, iovlen); - } while (res < 0 && errno == EINTR); - if (res < 0) - err = errno; - - erts_free(ERTS_ALC_T_SYS_WRITE_BUF, iov); - } - dd->blocking->res = res; - dd->blocking->err = err; -} - -void fd_ready_async(ErlDrvData drv_data, - ErlDrvThreadData thread_data) { - struct driver_data *dd = (struct driver_data *)thread_data; - ErlDrvPort port_num = dd->port_num; - - ASSERT(dd->blocking); - ASSERT(dd == (driver_data + (int)(long)drv_data)); - - if (dd->blocking->res > 0) { - driver_pdl_lock(dd->blocking->pdl); - if (driver_deq(port_num, dd->blocking->res) == 0) { - driver_pdl_unlock(dd->blocking->pdl); - set_busy_port(port_num, 0); - if (dd->terminating) { - /* The port is has been ordered to terminate - from either fd_flush or port_inp_failure */ - if (dd->terminating == 1) - driver_failure_atom(port_num, "normal"); - else if (dd->terminating == 2) - driver_failure_eof(port_num); - else if (dd->terminating < 0) - driver_failure_posix(port_num, -dd->terminating); - return; /* -1; */ - } - } else { - driver_pdl_unlock(dd->blocking->pdl); - /* still data left to write in queue */ - driver_async(port_num, &dd->blocking->pkey, fd_async, dd, NULL); - return /* 0; */; - } - } else if (dd->blocking->res < 0) { - if (dd->blocking->err == ERRNO_BLOCK) { - set_busy_port(port_num, 1); - /* still data left to write in queue */ - driver_async(port_num, &dd->blocking->pkey, fd_async, dd, NULL); - } else - driver_failure_posix(port_num, dd->blocking->err); - return; /* -1; */ - } - return; /* 0; */ -} - -#endif - void erts_do_break_handling(void) { struct termios temp_mode; @@ -2738,10 +1017,6 @@ erts_sys_unsetenv(char *key) 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_nob(&sys_misc_mem_sz, - max_files * sizeof(struct fd_data)); } #if (0) /* unused? */ @@ -2935,179 +1210,6 @@ erl_debug(char* fmt, ...) #endif /* DEBUG */ -static ERTS_INLINE void -report_exit_status(ErtsSysReportExit *rep, int status) -{ - Port *pp; -#ifdef ERTS_SMP - CHLD_STAT_UNLOCK; - pp = erts_thr_id2port_sflgs(rep->port, - ERTS_PORT_SFLGS_INVALID_DRIVER_LOOKUP); - CHLD_STAT_LOCK; -#else - pp = erts_id2port_sflgs(rep->port, - NULL, - 0, - ERTS_PORT_SFLGS_INVALID_DRIVER_LOOKUP); -#endif - if (pp) { - if (rep->ifd >= 0) { - driver_data[rep->ifd].alive = 0; - driver_data[rep->ifd].status = status; - (void) driver_select(ERTS_Port2ErlDrvPort(pp), - 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(ERTS_Port2ErlDrvPort(pp), - rep->ofd, - (ERL_DRV_WRITE|ERL_DRV_USE), - 1); - } -#ifdef ERTS_SMP - erts_thr_port_release(pp); -#else - erts_port_release(pp); -#endif - } - 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; - erts_sys_schedule_interrupt(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 @@ -3116,13 +1218,8 @@ child_waiter(void *unused) void erl_sys_schedule(int runnable) { -#ifdef ERTS_SMP ERTS_CHK_IO(!runnable); -#else - ERTS_CHK_IO(runnable ? 0 : !check_children()); -#endif ERTS_SMP_LC_ASSERT(!erts_thr_progress_is_blocking()); - (void) check_children(); } @@ -3150,10 +1247,6 @@ smp_sig_notify(char c) static void * signal_dispatcher_thread_func(void *unused) { -#if !CHLDWTHR - int initialized = 0; - int notify_check_children = 0; -#endif #ifdef ERTS_ENABLE_LOCK_CHECK erts_lc_set_thread_name("signal_dispatcher"); #endif @@ -3191,19 +1284,7 @@ signal_dispatcher_thread_func(void *unused) */ switch (buf[i]) { case 0: /* Emulator initialized */ -#if !CHLDWTHR - initialized = 1; - 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 + break; case 'I': /* SIGINT */ break_requested(); break; @@ -3248,12 +1329,14 @@ init_smp_sig_notify(void) static void init_smp_sig_suspend(void) { +#ifdef ERTS_SYS_SUSPEND_SIGNAL if (pipe(sig_suspend_fds) < 0) { erts_exit(ERTS_ABORT_EXIT, "Failed to create sig_suspend pipe: %s (%d)\n", erl_errno_id(errno), errno); } +#endif } #ifdef __DARWIN__ |