diff options
Diffstat (limited to 'erts/emulator/drivers')
| -rw-r--r-- | erts/emulator/drivers/common/efile_drv.c | 4249 | ||||
| -rw-r--r-- | erts/emulator/drivers/common/erl_efile.h | 176 | ||||
| -rw-r--r-- | erts/emulator/drivers/common/gzio.c | 712 | ||||
| -rw-r--r-- | erts/emulator/drivers/common/gzio.h | 10 | ||||
| -rw-r--r-- | erts/emulator/drivers/common/inet_drv.c | 1724 | ||||
| -rw-r--r-- | erts/emulator/drivers/unix/ttsl_drv.c | 10 | ||||
| -rw-r--r-- | erts/emulator/drivers/unix/unix_efile.c | 1102 | ||||
| -rw-r--r-- | erts/emulator/drivers/win32/ttsl_drv.c | 4 | ||||
| -rw-r--r-- | erts/emulator/drivers/win32/win_efile.c | 2058 |
9 files changed, 1405 insertions, 8640 deletions
diff --git a/erts/emulator/drivers/common/efile_drv.c b/erts/emulator/drivers/common/efile_drv.c deleted file mode 100644 index 3a7b3bb50c..0000000000 --- a/erts/emulator/drivers/common/efile_drv.c +++ /dev/null @@ -1,4249 +0,0 @@ -/* - * %CopyrightBegin% - * - * Copyright Ericsson AB 1996-2017. 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. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - * - * %CopyrightEnd% - */ -/* - * Purpose: Provides file and directory operations. - * - * This file is generic, and does the work of decoding the commands - * and encoding the responses. System-specific functions are found in - * the unix_efile.c and win_efile.c files. - */ - -/* Operations */ - -#define FILE_OPEN 1 /* Essential for startup */ -#define FILE_READ 2 -#define FILE_LSEEK 3 -#define FILE_WRITE 4 -#define FILE_FSTAT 5 /* Essential for startup */ -#define FILE_PWD 6 /* Essential for startup */ -#define FILE_READDIR 7 /* Essential for startup */ -#define FILE_CHDIR 8 -#define FILE_FSYNC 9 -#define FILE_MKDIR 10 -#define FILE_DELETE 11 -#define FILE_RENAME 12 -#define FILE_RMDIR 13 -#define FILE_TRUNCATE 14 -#define FILE_READ_FILE 15 /* Essential for startup */ -#define FILE_WRITE_INFO 16 -#define FILE_LSTAT 19 -#define FILE_READLINK 20 -#define FILE_LINK 21 -#define FILE_SYMLINK 22 -#define FILE_CLOSE 23 -#define FILE_PWRITEV 24 -#define FILE_PREADV 25 -#define FILE_SETOPT 26 -#define FILE_IPREAD 27 -#define FILE_ALTNAME 28 -#define FILE_READ_LINE 29 -#define FILE_FDATASYNC 30 -#define FILE_FADVISE 31 -#define FILE_SENDFILE 32 -#define FILE_FALLOCATE 33 -#define FILE_CLOSE_ON_PORT_EXIT 34 -/* Return codes */ - -#define FILE_RESP_OK 0 -#define FILE_RESP_ERROR 1 -#define FILE_RESP_DATA 2 -#define FILE_RESP_NUMBER 3 -#define FILE_RESP_INFO 4 -#define FILE_RESP_NUMERR 5 -#define FILE_RESP_LDATA 6 -#define FILE_RESP_N2DATA 7 -#define FILE_RESP_EOF 8 -#define FILE_RESP_FNAME 9 -#define FILE_RESP_ALL_DATA 10 -#define FILE_RESP_LFNAME 11 - -/* Options */ - -#define FILE_OPT_DELAYED_WRITE 0 -#define FILE_OPT_READ_AHEAD 1 - -/* IPREAD variants */ - -#define IPREAD_S32BU_P32BU 0 - -/* Limits */ - -#define FILE_SEGMENT_READ (256*1024) -#define FILE_SEGMENT_WRITE (256*1024) - -/* Internal */ - -/* Set to 1 to test having read_ahead implicitly for read_line */ -#define ALWAYS_READ_LINE_AHEAD 0 - - -/* Must not be possible to get from malloc()! */ -#define FILE_FD_INVALID ((Sint)(-1)) - -#ifdef HAVE_CONFIG_H -# include "config.h" -#endif - -#include <ctype.h> -#include <sys/types.h> -#include <stdlib.h> - -/* Need (NON)BLOCKING macros for sendfile */ -#ifndef WANT_NONBLOCKING -#define WANT_NONBLOCKING -#endif - -#include "sys.h" - -#include "erl_driver.h" -#include "erl_efile.h" -#include "erl_threads.h" -#include "gzio.h" -#include "dtrace-wrapper.h" - - -static ErlDrvSysInfo sys_info; - -/* For explanation of this var, see comment for same var in erl_async.c */ -static unsigned gcc_optimizer_hack = 0; - -#ifdef USE_VM_PROBES - -#define DTRACE_EFILE_BUFSIZ 128 - -#define DTRACE_INVOKE_SETUP(op) \ - do { DTRACE3(efile_drv_int_entry, d->sched_i1, d->sched_i2, op); } while (0) -#define DTRACE_INVOKE_SETUP_BY_NAME(op) \ - struct t_data *d = (struct t_data *) data ; \ - DTRACE_INVOKE_SETUP(op) -#define DTRACE_INVOKE_RETURN(op) \ - do { DTRACE3(efile_drv_int_return, d->sched_i1, d->sched_i2, \ - op); } while (0) ; gcc_optimizer_hack++ ; - -/* Assign human-friendlier id numbers to scheduler & I/O worker threads */ -int dt_driver_idnum = 0; -int dt_driver_io_worker_base = 5000; -erts_mtx_t dt_driver_mutex; -pthread_key_t dt_driver_key; - -typedef struct { - int thread_num; - Uint64 tag; -} dt_private; - -dt_private *get_dt_private(int); -#else /* USE_VM_PROBES */ -#define DTRACE_INVOKE_SETUP(op) do {} while (0) -#define DTRACE_INVOKE_SETUP_BY_NAME(op) do {} while (0) -#define DTRACE_INVOKE_RETURN(op) do {} while (0) -#endif /* USE_VM_PROBES */ - -/* #define TRACE 1 */ -#ifdef TRACE -# define TRACE_C(c) do { putchar(c); fflush(stdout); } while (0) -# define TRACE_S(s) do { fputs((s), stdout); fflush(stdout); } while (0) -# define TRACE_F(args) do { printf args ;fflush(stdout); } while (0) -#else -# define TRACE_C(c) ((void)(0)) -# define TRACE_S(s) ((void)(0)) -# define TRACE_F(args) ((void)(0)) -#endif - - -#ifdef USE_THREADS -#define THRDS_AVAILABLE (sys_info.async_threads > 0) -#ifdef HARDDEBUG /* HARDDEBUG in io.c is expected too */ -#define TRACE_DRIVER fprintf(stderr, "Efile: ") -#else -#define TRACE_DRIVER -#endif -#define MUTEX_INIT(m, p) do { IF_THRDS { TRACE_DRIVER; (m = driver_pdl_create(p)); } } while (0) -#define MUTEX_LOCK(m) do { IF_THRDS { TRACE_DRIVER; driver_pdl_lock(m); } } while (0) -#define MUTEX_UNLOCK(m) do { IF_THRDS { TRACE_DRIVER; driver_pdl_unlock(m); } } while (0) -#else -#define THRDS_AVAILABLE (0) -#define MUTEX_INIT(m, p) -#define MUTEX_LOCK(m) -#define MUTEX_UNLOCK(m) -#endif -#define IF_THRDS if (THRDS_AVAILABLE) - - -#define SENDFILE_FLGS_USE_THREADS (1 << 0) -/** - * On DARWIN sendfile can deadlock with close if called in - * different threads. So until Apple fixes so that sendfile - * is not buggy we disable usage of the async pool for - * DARWIN. The testcase t_sendfile_crashduring reproduces - * this error when using +A 10 and enabling SENDFILE_FLGS_USE_THREADS. - */ -#if defined(__APPLE__) && defined(__MACH__) -#define USE_THRDS_FOR_SENDFILE(DATA) 0 -#else -#define USE_THRDS_FOR_SENDFILE(DATA) (DATA->flags & SENDFILE_FLGS_USE_THREADS) -#endif /* defined(__APPLE__) && defined(__MACH__) */ - - - -#if 0 -/* Experimental, for forcing all file operations to use the same thread. */ - static unsigned file_fixed_key = 1; -# define KEY(desc) (&file_fixed_key) -#else -# define KEY(desc) (&(desc)->key) -#endif - -#ifndef MAX -# define MAX(x, y) (((x) > (y)) ? (x) : (y)) -#endif - -#ifdef FILENAMES_16BIT -#ifdef USE_VM_PROBES -#error 16bit characters in filenames and dtrace in combination is not supported. -#endif -# define FILENAME_BYTELEN(Str) filename_len_16bit(Str) -# define FILENAME_COPY(To,From) filename_cpy_16bit((To),(From)) -# define FILENAME_CHARSIZE 2 - - static int filename_len_16bit(char *str) - { - char *p = str; - while(*p != '\0' || p[1] != '\0') { - p += 2; - } - return (p - str); - } - - static void filename_cpy_16bit(char *to, char *from) - { - while(*from != '\0' || from[1] != '\0') { - *to++ = *from++; - *to++ = *from++; - } - *to++ = *from++; - *to++ = *from++; - } - -#else -# define FILENAME_BYTELEN(Str) strlen(Str) -# define FILENAME_COPY(To,From) strcpy(To,From) -# define FILENAME_CHARSIZE 1 -#endif - -#if (MAXPATHLEN+1)*FILENAME_CHARSIZE+1 > BUFSIZ -# define RESBUFSIZE ((MAXPATHLEN+1)*FILENAME_CHARSIZE+1) -#else -# define RESBUFSIZE BUFSIZ -#endif - -#define READDIR_CHUNKS (5) - - - -#if ALWAYS_READ_LINE_AHEAD -#define DEFAULT_LINEBUF_SIZE 2048 -#else -#define DEFAULT_LINEBUF_SIZE 512 /* Small, it's usually discarded anyway */ -#endif - -typedef unsigned char uchar; - -static ErlDrvData file_start(ErlDrvPort port, char* command); -static int file_init(void); -static void file_stop(ErlDrvData); -static void file_output(ErlDrvData, char* buf, ErlDrvSizeT len); -static ErlDrvSSizeT file_control(ErlDrvData, unsigned int command, - char* buf, ErlDrvSizeT len, - char **rbuf, ErlDrvSizeT rlen); -static void file_timeout(ErlDrvData); -static void file_outputv(ErlDrvData, ErlIOVec*); -static void file_async_ready(ErlDrvData, ErlDrvThreadData); -static void file_flush(ErlDrvData); - -#ifdef HAVE_SENDFILE -static void file_ready_output(ErlDrvData data, ErlDrvEvent event); -static void file_stop_select(ErlDrvEvent event, void* _); -#endif /* HAVE_SENDFILE */ - - -enum e_timer {timer_idle, timer_again, timer_write}; -#ifdef HAVE_SENDFILE -enum e_sendfile {sending, not_sending}; -#define SENDFILE_USE_THREADS (1 << 0) -#endif /* HAVE_SENDFILE */ - -struct t_data; - -typedef struct { - SWord fd; - ErlDrvPort port; - unsigned int key; /* Async queue key */ - unsigned flags; /* Original flags from FILE_OPEN. */ - void (*invoke)(void *); - struct t_data *d; - void (*free)(void *); - struct t_data *cq_head; /* Queue of incoming commands */ - struct t_data *cq_tail; /* -""- */ - enum e_timer timer_state; -#ifdef HAVE_SENDFILE - enum e_sendfile sendfile_state; -#endif /* HAVE_SENDFILE */ - size_t read_bufsize; - ErlDrvBinary *read_binp; - size_t read_offset; - size_t read_size; - size_t write_bufsize; - unsigned long write_delay; - int write_error; - Efile_error write_errInfo; - ErlDrvPDL q_mtx; /* Mutex for the driver queue, known by the emulator. Also used for - mutual exclusion when accessing field(s) below. */ - size_t write_buffered; -#ifdef USE_VM_PROBES - int idnum; /* Unique ID # for this driver thread/desc */ - char port_str[DTRACE_TERM_BUF_SIZE]; -#endif -} file_descriptor; - - -static int reply_error(file_descriptor*, Efile_error* errInfo); - -struct erl_drv_entry efile_driver_entry = { - file_init, - file_start, - file_stop, - file_output, - NULL, -#ifdef HAVE_SENDFILE - file_ready_output, -#else - NULL, -#endif /* HAVE_SENDFILE */ - "efile", - NULL, - NULL, - file_control, - file_timeout, - file_outputv, - file_async_ready, - file_flush, - NULL, - NULL, - ERL_DRV_EXTENDED_MARKER, - ERL_DRV_EXTENDED_MAJOR_VERSION, - ERL_DRV_EXTENDED_MINOR_VERSION, - ERL_DRV_FLAG_USE_PORT_LOCKING, - NULL, - NULL, -#ifdef HAVE_SENDFILE - file_stop_select -#else - NULL -#endif /* HAVE_SENDFILE */ -}; - - - -static int thread_short_circuit; - -#define DRIVER_ASYNC(level, desc, f_invoke, data, f_free) \ -if (thread_short_circuit >= (level)) { \ - (*(f_invoke))(data); \ - file_async_ready((ErlDrvData)(desc), (data)); \ -} else { \ - driver_async((desc)->port, KEY(desc), (f_invoke), (data), (f_free)); \ -} - - - -struct t_pbuf_spec { - Sint64 offset; - size_t size; -}; - -struct t_pwritev { - ErlDrvPort port; - ErlDrvPDL q_mtx; - size_t size; - unsigned cnt; - unsigned n; - struct t_pbuf_spec specs[1]; -}; - -struct t_preadv { - ErlIOVec eiov; - unsigned n; - unsigned cnt; - size_t size; - Sint64 offsets[1]; -}; - -#define READDIR_BUFSIZE (8*1024)*READDIR_CHUNKS -#if READDIR_BUFSIZE < (1 + (2 + MAXPATHLEN)*FILENAME_CHARSIZE*READDIR_CHUNKS) -# undef READDIR_BUFSIZE -# define READDIR_BUFSIZE (1 + (2 + MAXPATHLEN)*FILENAME_CHARSIZE*READDIR_CHUNKS) -#endif - -struct t_readdir_buf { - struct t_readdir_buf *next; - size_t n; - char buf[READDIR_BUFSIZE]; -}; - -struct t_data -{ - struct t_data *next; - int command; - int level; - void (*invoke)(void *); - void (*free)(void *); - void *data_to_free; /* used by FILE_CLOSE_ON_PORT_EXIT only */ - int again; - int reply; -#ifdef USE_VM_PROBES - int sched_i1; - Uint64 sched_i2; - char sched_utag[DTRACE_EFILE_BUFSIZ+1]; -#endif - int result_ok; - Efile_error errInfo; - int flags; - SWord fd; - int is_fd_unused; - /**/ - Efile_info info; - EFILE_DIR_HANDLE dir_handle; /* Handle to open directory. */ - ErlDrvBinary *bin; - int drive; - size_t n; - /*off_t offset;*/ - /*size_t bytesRead; Bytes read from the file. */ - /**/ - union { - struct { - Sint64 offset; - int origin; - Sint64 location; - } lseek; - struct { - ErlDrvPort port; - ErlDrvPDL q_mtx; - size_t size; - size_t reply_size; - } writev; - struct t_pwritev pwritev; - struct t_preadv preadv; - struct { - ErlDrvBinary *binp; - size_t bin_offset; - size_t bin_size; - size_t size; - } read; - struct { - ErlDrvBinary *binp; /* in - out */ - size_t read_offset; /* in - out */ - size_t read_size; /* in - out */ - size_t nl_pos; /* out */ - short nl_skip; /* out, 0 or 1 */ -#if !ALWAYS_READ_LINE_AHEAD - short read_ahead; /* in, bool */ -#endif - } read_line; - struct { - ErlDrvBinary *binp; - int size; - int offset; - } read_file; - struct { - struct t_readdir_buf *first_buf; - struct t_readdir_buf *last_buf; - } read_dir; - struct { - Sint64 offset; - Sint64 length; - int advise; - } fadvise; -#ifdef HAVE_SENDFILE - struct { - ErlDrvPort port; - ErlDrvPDL q_mtx; - int out_fd; - off_t offset; - Uint64 nbytes; - Uint64 written; - } sendfile; -#endif /* HAVE_SENDFILE */ - struct { - Sint64 offset; - Sint64 length; - } fallocate; - } c; - char b[1]; -}; - -#define EF_ALLOC(S) driver_alloc((S)) -#define EF_REALLOC(P, S) driver_realloc((P), (S)) -#define EF_SAFE_ALLOC(S) ef_safe_alloc((S)) -#define EF_SAFE_REALLOC(P, S) ef_safe_realloc((P), (S)) -#define EF_FREE(P) do { if((P)) driver_free((P)); } while(0) - -static void *ef_safe_alloc(Uint s) -{ - void *p = EF_ALLOC(s); - if (!p) erts_exit(ERTS_ERROR_EXIT, "efile drv: Can't allocate %lu bytes of memory\n", (unsigned long)s); - return p; -} - -/********************************************************************* - * ErlIOVec manipulation functions. - */ - -/* char EV_CHAR_P(ErlIOVec *ev, int p, int q) */ -#define EV_CHAR_P(ev, p, q) \ - (((char *)(ev)->iov[q].iov_base) + (p)) - -/* int EV_GET_CHAR(ErlIOVec *ev, char *p, int *pp, int *qp) */ -#define EV_GET_CHAR(ev, p, pp, qp) efile_ev_get_char(ev, p ,pp, qp) -static int -efile_ev_get_char(ErlIOVec *ev, char *p, size_t *pp, size_t *qp) { - if (*pp + 1 <= ev->iov[*qp].iov_len) { - *p = *EV_CHAR_P(ev, *pp, *qp); - if (*pp + 1 < ev->iov[*qp].iov_len) - *pp += 1; - else { - *qp += 1; - *pp = 0; - } - return !0; - } - return 0; -} - -/* Uint32 EV_UINT32(ErlIOVec *ev, int p, int q)*/ -#define EV_UINT32(ev, p, q) \ - ((Uint32) ((unsigned char *)(ev)->iov[q].iov_base)[p]) - -/* int EV_GET_UINT32(ErlIOVec *ev, Uint32 *p, int *pp, int *qp) */ -#define EV_GET_UINT32(ev, p, pp, qp) efile_ev_get_uint32(ev, p, pp, qp) -static int -efile_ev_get_uint32(ErlIOVec *ev, Uint32 *p, size_t *pp, size_t *qp) { - if (*pp + 4 <= ev->iov[*qp].iov_len) { - *p = (EV_UINT32(ev, *pp, *qp) << 24) - | (EV_UINT32(ev, *pp + 1, *qp) << 16) - | (EV_UINT32(ev, *pp + 2, *qp) << 8) - | (EV_UINT32(ev, *pp + 3, *qp)); - if (*pp + 4 < ev->iov[*qp].iov_len) - *pp += 4; - else { - *qp += 1; - *pp = 0; - } - return !0; - } - return 0; -} - -/* Uint64 EV_UINT64(ErlIOVec *ev, int p, int q)*/ -#define EV_UINT64(ev, p, q) \ - ((Uint64) ((unsigned char *)(ev)->iov[q].iov_base)[p]) - -/* int EV_GET_UINT64(ErlIOVec *ev, Uint64 *p, int *pp, int *qp) */ -#define EV_GET_UINT64(ev, p, pp, qp) efile_ev_get_uint64(ev, p, pp, qp) -static int -efile_ev_get_uint64(ErlIOVec *ev, Uint64 *p, size_t *pp, size_t *qp) { - if (*pp + 8 <= ev->iov[*qp].iov_len) { - *p = (EV_UINT64(ev, *pp, *qp) << 56) - | (EV_UINT64(ev, *pp + 1, *qp) << 48) - | (EV_UINT64(ev, *pp + 2, *qp) << 40) - | (EV_UINT64(ev, *pp + 3, *qp) << 32) - | (EV_UINT64(ev, *pp + 4, *qp) << 24) - | (EV_UINT64(ev, *pp + 5, *qp) << 16) - | (EV_UINT64(ev, *pp + 6, *qp) << 8) - | (EV_UINT64(ev, *pp + 7, *qp)); - if (*pp + 8 < ev->iov[*qp].iov_len) - *pp += 8; - else { - *qp += 1; - *pp = 0; - } - return !0; - } - return 0; -} - -/* int EV_GET_SINT64(ErlIOVec *ev, Uint64 *p, int *pp, int *qp) */ -#define EV_GET_SINT64(ev, p, pp, qp) efile_ev_get_sint64(ev, p, pp, qp) -static int -efile_ev_get_sint64(ErlIOVec *ev, Sint64 *p, size_t *pp, size_t *qp) { - Uint64 *tmp = (Uint64*)p; - return EV_GET_UINT64(ev, tmp, pp, qp); -} - -#if 0 - -static void ev_clear(ErlIOVec *ev) { - ASSERT(ev); - ev->size = 0; - ev->vsize = 0; - ev->iov = NULL; - ev->binv = NULL; -} - -/* Assumes that ->iov and ->binv were allocated with sys_alloc(). - */ -static void ev_free(ErlIOVec *ev) { - if (! ev) { - return; - } - if (ev->vsize > 0) { - int i; - ASSERT(ev->iov); - ASSERT(ev->binv); - for (i = 0; i < ev->vsize; i++) { - if (ev->binv[i]) { - driver_free_binary(ev->binv[i]); - } - } - EF_FREE(ev->iov); - EF_FREE(ev->binv); - } -} - -/* Copy the contents from source to dest. - * Data in binaries is not copied, just the pointers; - * and refc is incremented. - */ -static ErlIOVec *ev_copy(ErlIOVec *dest, ErlIOVec *source) { - int *ip; - ASSERT(dest); - ASSERT(source); - if (source->vsize == 0) { - /* Empty source */ - ev_clear(dest); - return dest; - } - /* Allocate ->iov and ->binv */ - dest->iov = EF_ALLOC(sizeof(*dest->iov) * source->vsize); - if (! dest->iov) { - return NULL; - } - dest->binv = EF_ALLOC(sizeof(*dest->binv) * source->vsize); - if (! dest->binv) { - EF_FREE(dest->iov); - return NULL; - } - dest->size = source->size; - /* Copy one vector element at the time. - * Use *ip as an alias for dest->vsize to improve readabiliy. - * Keep dest consistent in every iteration by using - * dest->vsize==*ip as loop variable. - */ - for (ip = &dest->vsize, *ip = 0; *ip < source->vsize; (*ip)++) { - if (source->iov[*ip].iov_len == 0) { - /* Empty vector element */ - dest->iov[*ip].iov_len = 0; - dest->iov[*ip].iov_base = NULL; - dest->binv[*ip] = NULL; - } else { - /* Non empty vector element */ - if (source->binv[*ip]) { - /* Contents in binary - copy pointers and increment refc */ - dest->iov[*ip] = source->iov[*ip]; - dest->binv[*ip] = source->binv[*ip]; - driver_binary_inc_refc(source->binv[*ip]); - } else { - /* Contents not in binary - allocate new binary and copy data */ - if (! (dest->binv[*ip] = - driver_alloc_binary(source->iov[*ip].iov_len))) { - goto failed; - } - sys_memcpy(dest->binv[*ip]->orig_bytes, - source->iov[*ip].iov_base, - source->iov[*ip].iov_len); - dest->iov[*ip].iov_base = dest->binv[*ip]->orig_bytes; - dest->iov[*ip].iov_len = source->iov[*ip].iov_len; - } - } - } - return dest; - failed: - ev_free(dest); - return NULL; -} - -#endif - - - -/********************************************************************* - * Command queue functions - */ - -static void cq_enq(file_descriptor *desc, struct t_data *d) { - ASSERT(d); - if (desc->cq_head) { - ASSERT(desc->cq_tail); - ASSERT(!desc->cq_tail->next); - desc->cq_tail = desc->cq_tail->next = d; - } else { - ASSERT(desc->cq_tail == NULL); - desc->cq_head = desc->cq_tail = d; - } - d->next = NULL; -} - -static struct t_data *cq_deq(file_descriptor *desc) { - struct t_data *d = desc->cq_head; - ASSERT(d || (!d && !desc->cq_tail)); - if (d) { - ASSERT(!d->next || (d->next && desc->cq_tail != d)); - if ((desc->cq_head = d->next) == NULL) { - ASSERT(desc->cq_tail == d); - desc->cq_tail = NULL; - } - } - return d; -} - - -/********************************************************************* - * Driver entry point -> init - */ -static int -file_init(void) -{ - char buf[21]; /* enough to hold any 64-bit integer */ - size_t bufsz = sizeof(buf); - thread_short_circuit = (erl_drv_getenv("ERL_EFILE_THREAD_SHORT_CIRCUIT", - buf, - &bufsz) == 0 - ? atoi(buf) - : 0); - driver_system_info(&sys_info, sizeof(ErlDrvSysInfo)); - - /* run initiation of efile_driver if needed */ - efile_init(); - -#ifdef USE_VM_PROBES - erts_mtx_init(&dt_driver_mutex, "efile_drv dtrace mutex", NIL, - ERTS_LOCK_FLAGS_PROPERTY_STATIC | ERTS_LOCK_FLAGS_CATEGORY_IO); - pthread_key_create(&dt_driver_key, NULL); -#endif /* USE_VM_PROBES */ - - return 0; -} - - -/********************************************************************* - * Driver entry point -> start - */ -static ErlDrvData -file_start(ErlDrvPort port, char* command) - -{ - file_descriptor* desc; - - if ((desc = (file_descriptor*) EF_ALLOC(sizeof(file_descriptor))) - == NULL) { - errno = ENOMEM; - return ERL_DRV_ERROR_ERRNO; - } - desc->fd = FILE_FD_INVALID; - desc->port = port; - desc->key = driver_async_port_key(port); - desc->flags = 0; - desc->invoke = NULL; - desc->d = NULL; - desc->free = NULL; - desc->cq_head = NULL; - desc->cq_tail = NULL; - desc->timer_state = timer_idle; -#ifdef HAVE_SENDFILE - desc->sendfile_state = not_sending; -#endif - desc->read_bufsize = 0; - desc->read_binp = NULL; - desc->read_offset = 0; - desc->read_size = 0; - desc->write_delay = 0L; - desc->write_bufsize = 0; - desc->write_error = 0; - MUTEX_INIT(desc->q_mtx, port); /* Refc is one, referenced by emulator now */ - desc->write_buffered = 0; -#ifdef USE_VM_PROBES - dtrace_drvport_str(port, desc->port_str); - get_dt_private(0); /* throw away return value */ -#endif /* USE_VM_PROBES */ - return (ErlDrvData) desc; -} - -static void do_close(int flags, SWord fd) { - if (flags & EFILE_COMPRESSED) { - erts_gzclose((ErtsGzFile)(fd)); - } else { - efile_closefile((int) fd); - } -} - -static void invoke_close(void *data) -{ - struct t_data *d = (struct t_data *) data; - DTRACE_INVOKE_SETUP(FILE_CLOSE); - d->again = 0; - do_close(d->flags, d->fd); - DTRACE_INVOKE_RETURN(FILE_CLOSE); -} - -static void free_data(void *data) -{ - struct t_data *d = (struct t_data *) data; - - switch (d->command) { - case FILE_OPEN: - if (d->is_fd_unused && d->fd != FILE_FD_INVALID) { - /* This is OK to do in scheduler thread because there can be no async op - ongoing for this fd here, as we exited during async open. - Ideally, this close should happen in an async thread too, but that would - require a substantial rewrite, as we are here because of a dead port and - cannot schedule async jobs for that port any more... */ - do_close(d->flags, d->fd); - } - break; - case FILE_CLOSE_ON_PORT_EXIT: - EF_FREE(d->data_to_free); - break; - } - - EF_FREE(data); -} - - -/* - * Sends back an error reply to Erlang. - */ - -static void reply_posix_error(file_descriptor *desc, int posix_errno) { - char response[256]; /* Response buffer. */ - char* s; - char* t; - - /* - * Contents of buffer sent back: - * - * +-----------------------------------------+ - * | FILE_RESP_ERROR | Posix error id string | - * +-----------------------------------------+ - */ - - TRACE_C('E'); - - response[0] = FILE_RESP_ERROR; - for (s = erl_errno_id(posix_errno), t = response+1; *s; s++, t++) - *t = tolower(*s); - driver_output2(desc->port, response, t-response, NULL, 0); -} - -static void reply_Uint_posix_error(file_descriptor *desc, Uint num, - int posix_errno) { - char response[256]; /* Response buffer. */ - char* s; - char* t; - - /* - * Contents of buffer sent back: - * - * +----------------------------------------------------------------------+ - * | FILE_RESP_NUMERR | 64-bit number (big-endian) | Posix error id string | - * +----------------------------------------------------------------------+ - */ - - TRACE_C('N'); - - response[0] = FILE_RESP_NUMERR; -#if SIZEOF_VOID_P == 4 - put_int32(0, response+1); -#else - put_int32(num>>32, response+1); -#endif - put_int32((Uint32)num, response+1+4); - for (s = erl_errno_id(posix_errno), t = response+1+4+4; *s; s++, t++) - *t = tolower(*s); - driver_output2(desc->port, response, t-response, NULL, 0); -} - -#ifdef HAVE_SENDFILE -static void reply_string_error(file_descriptor *desc, char* str) { - char response[256]; /* Response buffer. */ - char* s; - char* t; - - response[0] = FILE_RESP_ERROR; - for (s = str, t = response+1; *s; s++, t++) - *t = tolower(*s); - driver_output2(desc->port, response, t-response, NULL, 0); -} -#endif - -static int reply_error(file_descriptor *desc, - Efile_error *errInfo) /* The error codes. */ -{ - reply_posix_error(desc, errInfo->posix_errno); - return 0; -} - -static int reply_Uint_error(file_descriptor *desc, Uint num, - Efile_error *errInfo) /* The error codes. */ -{ - reply_Uint_posix_error(desc, num, errInfo->posix_errno); - return 0; -} - -static int reply_ok(file_descriptor *desc) { - char c = FILE_RESP_OK; - - driver_output2(desc->port, &c, 1, NULL, 0); - return 0; -} - -static int reply(file_descriptor *desc, int ok, Efile_error *errInfo) { - if (!ok) { - reply_error(desc, errInfo); - } else { - TRACE_C('K'); - reply_ok(desc); - } - return 0; -} - -static int reply_Uint(file_descriptor *desc, Uint result) { - char tmp[1+4+4]; - - /* - * Contents of buffer sent back: - * - * +-----------------------------------------------+ - * | FILE_RESP_NUMBER | 64-bit number (big-endian) | - * +-----------------------------------------------+ - */ - - TRACE_C('R'); - - tmp[0] = FILE_RESP_NUMBER; -#if SIZEOF_VOID_P == 4 - put_int32(0, tmp+1); -#else - put_int32(result>>32, tmp+1); -#endif - put_int32((Uint32)result, tmp+1+4); - driver_output2(desc->port, tmp, sizeof(tmp), NULL, 0); - return 0; -} - -static int reply_Sint64(file_descriptor *desc, Sint64 result) { - char tmp[1+4+4]; - - /* - * Contents of buffer sent back: - * - * +-----------------------------------------------+ - * | FILE_RESP_NUMBER | 64-bit number (big-endian) | - * +-----------------------------------------------+ - */ - - TRACE_C('R'); - - tmp[0] = FILE_RESP_NUMBER; - put_int64(result, tmp+1); - driver_output2(desc->port, tmp, sizeof(tmp), NULL, 0); - return 0; -} - -#if 0 -static void reply_again(file_descriptor *desc) { - char tmp[1]; - tmp[0] = FILE_RESP_AGAIN; - driver_output2(desc->port, tmp, sizeof(tmp), NULL, 0); -} -#endif - -static void reply_ev(file_descriptor *desc, char response, ErlIOVec *ev) { - char tmp[1]; - /* Data arriving at the Erlang process: - * [Response, Binary0, Binary1, .... | BinaryN-1] - */ - tmp[0] = response; - driver_outputv(desc->port, tmp, sizeof(tmp), ev, 0); -} - -static void reply_data(file_descriptor *desc, - ErlDrvBinary *binp, size_t offset, size_t len) { - char header[1+4+4]; - /* Data arriving at the Erlang process: - * [?FILE_RESP_DATA, 64-bit length (big-endian) | Data] - */ - header[0] = FILE_RESP_DATA; -#if SIZEOF_SIZE_T == 4 - put_int32(0, header+1); -#else - put_int32(len>>32, header+1); -#endif - put_int32((Uint32)len, header+1+4); - driver_output_binary(desc->port, header, sizeof(header), - binp, offset, len); -} - -static void reply_buf(file_descriptor *desc, char *buf, size_t len) { - char header[1+4+4]; - /* Data arriving at the Erlang process: - * [?FILE_RESP_DATA, 64-bit length (big-endian) | Data] - */ - header[0] = FILE_RESP_DATA; -#if SIZEOF_SIZE_T == 4 - put_int32(0, header+1); -#else - put_int32(len>>32, header+1); -#endif - put_int32((Uint32)len, header+1+4); - driver_output2(desc->port, header, sizeof(header), buf, len); -} - -static int reply_eof(file_descriptor *desc) { - char c = FILE_RESP_EOF; - - driver_output2(desc->port, &c, 1, NULL, 0); - return 0; -} - -static void invoke_name(void *data, int (*f)(Efile_error *, char *)) -{ - struct t_data *d = (struct t_data *) data; - char *name = (char *) d->b; - - d->again = 0; - d->result_ok = (*f)(&d->errInfo, name); -} - -static void invoke_mkdir(void *data) -{ - DTRACE_INVOKE_SETUP_BY_NAME(FILE_MKDIR); - invoke_name(data, efile_mkdir); - DTRACE_INVOKE_RETURN(FILE_MKDIR); -} - -static void invoke_rmdir(void *data) -{ - DTRACE_INVOKE_SETUP_BY_NAME(FILE_RMDIR); - invoke_name(data, efile_rmdir); - DTRACE_INVOKE_RETURN(FILE_RMDIR); -} - -static void invoke_delete_file(void *data) -{ - DTRACE_INVOKE_SETUP_BY_NAME(FILE_DELETE); - invoke_name(data, efile_delete_file); - DTRACE_INVOKE_RETURN(FILE_DELETE); -} - -static void invoke_chdir(void *data) -{ - DTRACE_INVOKE_SETUP_BY_NAME(FILE_CHDIR); - invoke_name(data, efile_chdir); - DTRACE_INVOKE_RETURN(FILE_CHDIR); -} - -static void invoke_fdatasync(void *data) -{ - struct t_data *d = (struct t_data *) data; - int fd = (int) d->fd; - DTRACE_INVOKE_SETUP(FILE_FDATASYNC); - - d->again = 0; - d->result_ok = efile_fdatasync(&d->errInfo, fd); - DTRACE_INVOKE_RETURN(FILE_FDATASYNC); -} - -static void invoke_fsync(void *data) -{ - struct t_data *d = (struct t_data *) data; - int fd = (int) d->fd; - DTRACE_INVOKE_SETUP(FILE_FSYNC); - - d->again = 0; - d->result_ok = efile_fsync(&d->errInfo, fd); - DTRACE_INVOKE_RETURN(FILE_FSYNC); -} - -static void invoke_truncate(void *data) -{ - struct t_data *d = (struct t_data *) data; - int fd = (int) d->fd; - DTRACE_INVOKE_SETUP(FILE_TRUNCATE); - - d->again = 0; - d->result_ok = efile_truncate_file(&d->errInfo, &fd, d->flags); - DTRACE_INVOKE_RETURN(FILE_TRUNCATE); -} - -static void invoke_read(void *data) -{ - struct t_data *d = (struct t_data *) data; - int status, segment; - size_t size, read_size; - DTRACE_INVOKE_SETUP(FILE_READ); - - segment = d->again && d->c.read.bin_size >= 2*FILE_SEGMENT_READ; - if (segment) { - size = FILE_SEGMENT_READ; - } else { - size = d->c.read.bin_size; - } - read_size = size; - if (d->flags & EFILE_COMPRESSED) { - read_size = erts_gzread((ErtsGzFile)d->fd, - d->c.read.binp->orig_bytes + d->c.read.bin_offset, - size); - status = (read_size != (size_t) -1); - if (!status) { - d->errInfo.posix_errno = EIO; - } - } else { - status = efile_read(&d->errInfo, d->flags, (int) d->fd, - d->c.read.binp->orig_bytes + d->c.read.bin_offset, - size, - &read_size); - } - if ( (d->result_ok = status)) { - ASSERT(read_size <= size); - d->c.read.bin_offset += read_size; - if (read_size < size || !segment) { - d->c.read.bin_size = 0; - d->again = 0; - } else { - d->c.read.bin_size -= read_size; - } - } else { - d->again = 0; - } - DTRACE_INVOKE_RETURN(FILE_READ); -} - -static void free_read(void *data) -{ - struct t_data *d = (struct t_data *) data; - - driver_free_binary(d->c.read.binp); - EF_FREE(d); -} - -static void invoke_read_line(void *data) -{ - struct t_data *d = (struct t_data *) data; - int status; - size_t read_size = 0; - int local_loop = (d->again == 0); - DTRACE_INVOKE_SETUP(FILE_READ_LINE); - - do { - size_t size = (d->c.read_line.binp)->orig_size - - d->c.read_line.read_offset - d->c.read_line.read_size; - if (size == 0) { - /* Need more place */ - ErlDrvSizeT need = (d->c.read_line.read_size >= DEFAULT_LINEBUF_SIZE) ? - d->c.read_line.read_size + DEFAULT_LINEBUF_SIZE : DEFAULT_LINEBUF_SIZE; - ErlDrvBinary *newbin; -#if !ALWAYS_READ_LINE_AHEAD - /* Use read_ahead size if need does not exceed it */ - if (need < (d->c.read_line.binp)->orig_size && - d->c.read_line.read_ahead) - need = (d->c.read_line.binp)->orig_size; -#endif - newbin = driver_alloc_binary(need); - if (newbin == NULL) { - d->result_ok = 0; - d->errInfo.posix_errno = ENOMEM; - d->again = 0; - break; - } - memcpy(newbin->orig_bytes, (d->c.read_line.binp)->orig_bytes + d->c.read_line.read_offset, - d->c.read_line.read_size); - driver_free_binary(d->c.read_line.binp); - d->c.read_line.binp = newbin; - d->c.read_line.read_offset = 0; - size = need - d->c.read_line.read_size; - } - if (d->flags & EFILE_COMPRESSED) { - read_size = erts_gzread((ErtsGzFile)d->fd, - d->c.read_line.binp->orig_bytes + - d->c.read_line.read_offset + d->c.read_line.read_size, - size); - status = (read_size != (size_t) -1); - if (!status) { - d->errInfo.posix_errno = EIO; - } - } else { - status = efile_read(&d->errInfo, d->flags, (int) d->fd, - d->c.read_line.binp->orig_bytes + - d->c.read_line.read_offset + d->c.read_line.read_size, - size, - &read_size); - } - if ( (d->result_ok = status)) { - void *nl_ptr = memchr((d->c.read_line.binp)->orig_bytes + - d->c.read_line.read_offset + d->c.read_line.read_size,'\n',read_size); - ASSERT(read_size <= size); - d->c.read_line.read_size += read_size; - if (nl_ptr != NULL) { - /* If found, we're done */ - d->c.read_line.nl_pos = ((char *) nl_ptr) - - ((char *) ((d->c.read_line.binp)->orig_bytes)) + 1; - if (d->c.read_line.nl_pos > 1 && - *(((char *) nl_ptr) - 1) == '\r') { - --d->c.read_line.nl_pos; - *(((char *) nl_ptr) - 1) = '\n'; - d->c.read_line.nl_skip = 1; - } else { - d->c.read_line.nl_skip = 0; - } - d->again = 0; -#if !ALWAYS_READ_LINE_AHEAD - if (!(d->c.read_line.read_ahead)) { - /* Ouch! Undo buffering... */ - size_t too_much = d->c.read_line.read_size - d->c.read_line.nl_skip - - (d->c.read_line.nl_pos - d->c.read_line.read_offset); - d->c.read_line.read_size -= too_much; - ASSERT(d->c.read_line.read_size >= 0); - if (d->flags & EFILE_COMPRESSED) { - Sint64 location = erts_gzseek((ErtsGzFile)d->fd, - -((Sint64) too_much), EFILE_SEEK_CUR); - if (location == -1) { - d->result_ok = 0; - d->errInfo.posix_errno = errno; - } - } else { - Sint64 location; - d->result_ok = efile_seek(&d->errInfo, (int) d->fd, - -((Sint64) too_much), EFILE_SEEK_CUR, - &location); - } - } -#endif - break; - } else if (read_size == 0) { - d->c.read_line.nl_pos = - d->c.read_line.read_offset + d->c.read_line.read_size; - d->c.read_line.nl_skip = 0; - d->again = 0; - break; - } - } else { - d->again = 0; - break; - } - } while (local_loop); - DTRACE_INVOKE_RETURN(FILE_READ_LINE); -} - -static void free_read_line(void *data) -{ - struct t_data *d = (struct t_data *) data; - - driver_free_binary(d->c.read_line.binp); - EF_FREE(d); -} - -static void invoke_read_file(void *data) -{ - struct t_data *d = (struct t_data *) data; - size_t read_size; - int chop; - DTRACE_INVOKE_SETUP(FILE_READ_FILE); - - if (! d->c.read_file.binp) { /* First invocation only */ - int fd; - Sint64 size; - - if (! (d->result_ok = - efile_openfile(&d->errInfo, d->b, - EFILE_MODE_READ, &fd, &size))) { - goto done; - } - d->fd = fd; - d->c.read_file.size = (int) size; - if (size < 0 || size != d->c.read_file.size || - ! (d->c.read_file.binp = - driver_alloc_binary(d->c.read_file.size))) { - d->result_ok = 0; - d->errInfo.posix_errno = ENOMEM; - goto close; - } - d->c.read_file.offset = 0; - } - /* Invariant: d->c.read_file.size >= d->c.read_file.offset */ - - read_size = (size_t) (d->c.read_file.size - d->c.read_file.offset); - if (! read_size) goto close; - chop = d->again && read_size >= FILE_SEGMENT_READ*2; - if (chop) read_size = FILE_SEGMENT_READ; - d->result_ok = - efile_read(&d->errInfo, - EFILE_MODE_READ, - (int) d->fd, - d->c.read_file.binp->orig_bytes + d->c.read_file.offset, - read_size, - &read_size); - if (d->result_ok) { - d->c.read_file.offset += read_size; - if (chop) goto chop_done; /* again */ - } - close: - efile_closefile((int) d->fd); - done: - d->again = 0; - chop_done: - DTRACE_INVOKE_RETURN(FILE_READ_FILE); -} - -static void free_read_file(void *data) -{ - struct t_data *d = (struct t_data *) data; - - if (d->c.read_file.binp) driver_free_binary(d->c.read_file.binp); - EF_FREE(d); -} - - - -static void invoke_preadv(void *data) -{ - struct t_data *d = (struct t_data *) data; - struct t_preadv *c = &d->c.preadv; - ErlIOVec *ev = &c->eiov; - size_t bytes_read_so_far = 0; - unsigned char *p = (unsigned char *)ev->iov[0].iov_base + 4+4+8*c->cnt; - DTRACE_INVOKE_SETUP(FILE_PREADV); - - while (c->cnt < c->n) { - size_t read_size = ev->iov[1 + c->cnt].iov_len - c->size; - size_t bytes_read = 0; - int chop = d->again - && bytes_read_so_far + read_size >= 2*FILE_SEGMENT_READ; - if (chop) { - ASSERT(bytes_read_so_far < FILE_SEGMENT_READ); - read_size = FILE_SEGMENT_READ + FILE_SEGMENT_READ/2 - - bytes_read_so_far; - } - if ( (d->result_ok - = efile_pread(&d->errInfo, - (int) d->fd, - c->offsets[c->cnt] + c->size, - ((char *)ev->iov[1 + c->cnt].iov_base) + c->size, - read_size, - &bytes_read))) { - bytes_read_so_far += bytes_read; - if (chop && bytes_read == read_size) { - c->size += bytes_read; - goto done; - } - ASSERT(bytes_read <= read_size); - ev->iov[1 + c->cnt].iov_len = bytes_read + c->size; - ev->size += bytes_read + c->size; - put_int64(bytes_read + c->size, p); p += 8; - c->size = 0; - c->cnt++; - if (d->again - && bytes_read_so_far >= FILE_SEGMENT_READ - && c->cnt < c->n) { - goto done; - } - } else { - /* In case of a read error, ev->size will not be correct, - * which does not matter since no read data is returned - * to Erlang. - */ - break; - } - } - d->again = 0; - done: - DTRACE_INVOKE_RETURN(FILE_PREADV); -} - -static void free_preadv(void *data) { - struct t_data *d = data; - int i; - ErlIOVec *ev = &d->c.preadv.eiov; - - for(i = 0; i < ev->vsize; i++) { - driver_free_binary(ev->binv[i]); - } - EF_FREE(d); -} - -static void invoke_ipread(void *data) -{ - struct t_data *d = data; - struct t_preadv *c = &d->c.preadv; - ErlIOVec *ev = &c->eiov; - size_t bytes_read = 0; - char buf[2*sizeof(Uint32)]; - Uint32 offset, size; - DTRACE_INVOKE_SETUP(FILE_IPREAD); - - /* Read indirection header */ - if (! efile_pread(&d->errInfo, (int) d->fd, c->offsets[0], - buf, sizeof(buf), &bytes_read)) { - goto error; - } - if (bytes_read != sizeof(buf)) goto done; /* eof */ - size = get_int32(buf); - offset = get_int32(buf+4); - if (size > c->size) goto done; /* eof */ - c->n = 1; - c->cnt = 0; - c->size = 0; - c->offsets[0] = offset; - if (! (ev->binv[0] = driver_alloc_binary(3*8))) { - d->errInfo.posix_errno = ENOMEM; - goto error; - } - ev->vsize = 1; - ev->iov[0].iov_len = 3*8; - ev->iov[0].iov_base = ev->binv[0]->orig_bytes; - ev->size = ev->iov[0].iov_len; - put_int64(offset, ev->iov[0].iov_base); - put_int64(size, ((char *)ev->iov[0].iov_base) + 2*8); - if (size == 0) { - put_int64(size, ((char *)ev->iov[0].iov_base) + 8); - goto done; - } - if (! (ev->binv[1] = driver_alloc_binary(size))) { - d->errInfo.posix_errno = ENOMEM; - goto error; - } - ev->vsize = 2; - ev->iov[1].iov_len = size; - ev->iov[1].iov_base = ev->binv[1]->orig_bytes; - /* Read data block */ - d->invoke = invoke_preadv; - invoke_preadv(data); - DTRACE_INVOKE_RETURN(FILE_IPREAD); - return; - error: - d->result_ok = 0; - d->again = 0; - DTRACE_INVOKE_RETURN(FILE_IPREAD); - return; - done: - d->result_ok = !0; - d->again = 0; - DTRACE_INVOKE_RETURN(FILE_IPREAD); -} - -/* invoke_writev and invoke_pwritev are the only thread functions that - * access non-thread data i.e the port queue and a mutex in the port - * structure that is used to lock the port queue. - * - * The port will normally not be terminated until the port queue is - * empty, but if the port is killed, i.e., exit(Port, kill) is called, - * it will terminate regardless of the port queue state. When the - * port is invalid driver_peekq() returns NULL and set the size to -1, - * and driver_sizeq() returns -1. - */ - -static void invoke_writev(void *data) { - struct t_data *d = (struct t_data *) data; - SysIOVec *iov0; - SysIOVec *iov; - int iovlen; - int iovcnt; - size_t size; - size_t p; - int segment; - DTRACE_INVOKE_SETUP(FILE_WRITE); - - segment = d->again && d->c.writev.size >= 2*FILE_SEGMENT_WRITE; - if (segment) { - size = FILE_SEGMENT_WRITE; - } else { - size = d->c.writev.size; - } - - /* Copy the io vector to avoid locking the port que while writing, - * also, both we and efile_writev might/will change the SysIOVec - * when segmenting or due to partial write and we do not want to - * tamper with the actual queue that we get from driver_peekq - */ - MUTEX_LOCK(d->c.writev.q_mtx); /* Lock before accessing the port queue */ - iov0 = driver_peekq(d->c.writev.port, &iovlen); - - /* Calculate iovcnt */ - for (p = 0, iovcnt = 0; - p < size && iovcnt < iovlen; - p += iov0[iovcnt++].iov_len) - ; - iov = EF_SAFE_ALLOC(sizeof(SysIOVec)*iovcnt); - memcpy(iov,iov0,iovcnt*sizeof(SysIOVec)); - MUTEX_UNLOCK(d->c.writev.q_mtx); - /* Let go of lock until we deque from original vector */ - - if (iovlen > 0) { - ASSERT(iov[iovcnt-1].iov_len > p - size); - iov[iovcnt-1].iov_len -= p - size; - if (d->flags & EFILE_COMPRESSED) { - int i, status = 1; - for (i = 0; i < iovcnt; i++) { - if (iov[i].iov_base && iov[i].iov_len > 0) { - /* Just in case, I do not know what gzwrite does - * with errno. - */ - errno = EINVAL; - status = erts_gzwrite((ErtsGzFile)d->fd, - iov[i].iov_base, - iov[i].iov_len) == iov[i].iov_len; - if (! status) { - d->errInfo.posix_errno = - d->errInfo.os_errno = errno; /* XXX Correct? */ - break; - } - } - } - d->result_ok = status; - } else { - d->result_ok = efile_writev(&d->errInfo, - d->flags, (int) d->fd, - iov, iovcnt); - } - } else if (iovlen == 0) { - d->result_ok = 1; - } - else { /* Port has terminated */ - d->result_ok = 0; - d->errInfo.posix_errno = d->errInfo.os_errno = EINVAL; - } - EF_FREE(iov); - - if (! d->result_ok) { - d->again = 0; - MUTEX_LOCK(d->c.writev.q_mtx); - driver_deq(d->c.writev.port, d->c.writev.size); - MUTEX_UNLOCK(d->c.writev.q_mtx); - } else { - if (! segment) { - d->again = 0; - } - d->c.writev.size -= size; - TRACE_F(("w%lu", (unsigned long)size)); - MUTEX_LOCK(d->c.writev.q_mtx); - driver_deq(d->c.writev.port, size); - MUTEX_UNLOCK(d->c.writev.q_mtx); - } - - - DTRACE_INVOKE_RETURN(FILE_WRITE); -} - -static void invoke_pwd(void *data) -{ - struct t_data *d = (struct t_data *) data; - DTRACE_INVOKE_SETUP(FILE_PWD); - - d->again = 0; - d->result_ok = efile_getdcwd(&d->errInfo,d->drive, d->b+1, - RESBUFSIZE-1); - DTRACE_INVOKE_RETURN(FILE_PWD); -} - -static void invoke_readlink(void *data) -{ - struct t_data *d = (struct t_data *) data; - char resbuf[RESBUFSIZE]; /* Result buffer. */ - DTRACE_INVOKE_SETUP(FILE_READLINK); - - d->again = 0; - d->result_ok = efile_readlink(&d->errInfo, d->b, resbuf+1, - RESBUFSIZE-1); - if (d->result_ok != 0) - FILENAME_COPY((char *) d->b + 1, resbuf+1); - DTRACE_INVOKE_RETURN(FILE_READLINK); -} - -static void invoke_altname(void *data) -{ - struct t_data *d = (struct t_data *) data; - char resbuf[RESBUFSIZE]; /* Result buffer. */ - DTRACE_INVOKE_SETUP(FILE_ALTNAME); - - d->again = 0; - d->result_ok = efile_altname(&d->errInfo, d->b, resbuf+1, - RESBUFSIZE-1); - if (d->result_ok != 0) - FILENAME_COPY((char *) d->b + 1, resbuf+1); - DTRACE_INVOKE_RETURN(FILE_ALTNAME); -} - -static void invoke_pwritev(void *data) { - struct t_data* const d = (struct t_data *) data; - struct t_pwritev * const c = &d->c.pwritev; - SysIOVec *iov0; - SysIOVec *iov; - int iovlen; - int iovcnt; - size_t p; - int segment; - size_t size, write_size, written; - DTRACE_INVOKE_SETUP(FILE_PWRITEV); - - segment = d->again && c->size >= 2*FILE_SEGMENT_WRITE; - if (segment) { - size = FILE_SEGMENT_WRITE; - } else { - size = c->size; - } - d->result_ok = !0; - p = 0; - /* Lock the queue just for a while, we don't want it locked during write */ - MUTEX_LOCK(c->q_mtx); - iov0 = driver_peekq(c->port, &iovlen); - iov = EF_SAFE_ALLOC(sizeof(SysIOVec)*iovlen); - memcpy(iov,iov0,sizeof(SysIOVec)*iovlen); - MUTEX_UNLOCK(c->q_mtx); - - if (iovlen < 0) - goto error; /* Port terminated */ - for (iovcnt = 0, written = 0; - c->cnt < c->n && iovcnt < iovlen && written < size; - c->cnt++) { - int chop; - write_size = c->specs[c->cnt].size; - if (iov[iovcnt].iov_len - p < write_size) { - goto error; - } - chop = segment && written + write_size >= 2*FILE_SEGMENT_WRITE; - if (chop) { - ASSERT(written < FILE_SEGMENT_WRITE); - write_size = FILE_SEGMENT_WRITE + FILE_SEGMENT_WRITE/2 - - written; - } - d->result_ok = efile_pwrite(&d->errInfo, (int) d->fd, - (char *)(iov[iovcnt].iov_base) + p, - write_size, - c->specs[c->cnt].offset); - if (! d->result_ok) { - d->again = 0; - goto deq_error; - } - written += write_size; - c->size -= write_size; - if (chop) { - c->specs[c->cnt].offset += write_size; - c->specs[c->cnt].size -= write_size; - /* Schedule out (d->again != 0) */ - break; - } - /* Move forward in buffer */ - p += write_size; - ASSERT(iov[iovcnt].iov_len >= p); - if (iov[iovcnt].iov_len == p) { - /* Move to next iov[], we trust that it is not a - * zero length vector, and thereby depend on that - * such are not queued. - */ - iovcnt++; p = 0; - } - } - if (! segment) { - if (c->cnt != c->n) { - /* Mismatch between number of - * pos/size specs vs number of queued buffers . - */ - error: - d->errInfo.posix_errno = EINVAL; - d->result_ok = 0; - d->again = 0; - deq_error: - MUTEX_LOCK(c->q_mtx); - driver_deq(c->port, c->size); - MUTEX_UNLOCK(c->q_mtx); - - goto done; - } else { - ASSERT(written == size); - d->again = 0; - } - } else { - ASSERT(written >= FILE_SEGMENT_WRITE); - } - - MUTEX_LOCK(c->q_mtx); - driver_deq(c->port, written); - MUTEX_UNLOCK(c->q_mtx); - done: - EF_FREE(iov); /* Free our copy of the vector, nothing to restore */ - - DTRACE_INVOKE_RETURN(FILE_PWRITEV); -} - -static void invoke_flstat(void *data) -{ - struct t_data *d = (struct t_data *) data; - - DTRACE3(efile_drv_int_entry, d->sched_i1, d->sched_i2, - d->command == FILE_LSTAT ? FILE_LSTAT : FILE_FSTAT); - d->again = 0; - d->result_ok = efile_fileinfo(&d->errInfo, &d->info, - d->b, d->command == FILE_LSTAT); - DTRACE3(efile_drv_int_entry, d->sched_i1, d->sched_i2, - d->command == FILE_LSTAT ? FILE_LSTAT : FILE_FSTAT); - gcc_optimizer_hack++; -} - -static void invoke_link(void *data) -{ - struct t_data *d = (struct t_data *) data; - char *name = d->b; - char *new_name; - DTRACE_INVOKE_SETUP(FILE_LINK); - - d->again = 0; - new_name = name+FILENAME_BYTELEN(name)+FILENAME_CHARSIZE; - d->result_ok = efile_link(&d->errInfo, name, new_name); - DTRACE_INVOKE_RETURN(FILE_LINK); -} - -static void invoke_symlink(void *data) -{ - struct t_data *d = (struct t_data *) data; - char *name = d->b; - char *new_name; - DTRACE_INVOKE_SETUP(FILE_SYMLINK); - - d->again = 0; - new_name = name+FILENAME_BYTELEN(name)+FILENAME_CHARSIZE; - d->result_ok = efile_symlink(&d->errInfo, name, new_name); - DTRACE_INVOKE_RETURN(FILE_SYMLINK); -} - -static void invoke_rename(void *data) -{ - struct t_data *d = (struct t_data *) data; - char *name = d->b; - char *new_name; - DTRACE_INVOKE_SETUP(FILE_RENAME); - - d->again = 0; - new_name = name+FILENAME_BYTELEN(name)+FILENAME_CHARSIZE; - d->result_ok = efile_rename(&d->errInfo, name, new_name); - DTRACE_INVOKE_RETURN(FILE_RENAME); -} - -static void invoke_write_info(void *data) -{ - struct t_data *d = (struct t_data *) data; - DTRACE_INVOKE_SETUP(FILE_WRITE_INFO); - - d->again = 0; - d->result_ok = efile_write_info(&d->errInfo, &d->info, d->b); - DTRACE_INVOKE_RETURN(FILE_WRITE_INFO); -} - -static void invoke_lseek(void *data) -{ - struct t_data *d = (struct t_data *) data; - int status; - DTRACE_INVOKE_SETUP(FILE_LSEEK); - - d->again = 0; - if (d->flags & EFILE_COMPRESSED) { - int offset = (int) d->c.lseek.offset; - - if (offset != d->c.lseek.offset) { - d->errInfo.posix_errno = EINVAL; - status = 0; - } else { - d->c.lseek.location = erts_gzseek((ErtsGzFile)d->fd, - offset, d->c.lseek.origin); - if (d->c.lseek.location == -1) { - d->errInfo.posix_errno = errno; - status = 0; - } else { - status = 1; - } - } - } else { - status = efile_seek(&d->errInfo, (int) d->fd, - d->c.lseek.offset, d->c.lseek.origin, - &d->c.lseek.location); - } - d->result_ok = status; - DTRACE_INVOKE_RETURN(FILE_LSEEK); -} - -static void invoke_readdir(void *data) -{ - struct t_data *d = (struct t_data *) data; - char *p = NULL; - size_t file_bs; - size_t n = 0, total = 0; - struct t_readdir_buf *b = NULL; - int res = 0; - DTRACE_INVOKE_SETUP(FILE_READDIR); - - d->again = 0; - d->errInfo.posix_errno = 0; - - do { - total = READDIR_BUFSIZE; - n = 1; - b = EF_SAFE_ALLOC(sizeof(struct t_readdir_buf)); - b->next = NULL; - - if (d->c.read_dir.last_buf) { - d->c.read_dir.last_buf->next = b; - } else { - d->c.read_dir.first_buf = b; - } - d->c.read_dir.last_buf = b; - - p = &b->buf[0]; - p[0] = FILE_RESP_LFNAME; - file_bs = READDIR_BUFSIZE - n; - - do { - res = efile_readdir(&d->errInfo, d->b, &d->dir_handle, p + n + 2, &file_bs); - - if (res) { - put_int16((Uint16)file_bs, p + n); - n += 2 + file_bs; - file_bs = READDIR_BUFSIZE - n; - } - } while( res && ((total - n - 2) >= MAXPATHLEN*FILENAME_CHARSIZE)); - - b->n = n; - } while(res); - - d->result_ok = (d->errInfo.posix_errno == 0); - DTRACE_INVOKE_RETURN(FILE_READDIR); -} - -static void invoke_open(void *data) -{ - struct t_data *d = (struct t_data *) data; - int status = 1; /* Status of open call. */ - DTRACE_INVOKE_SETUP(FILE_OPEN); - - d->again = 0; - if ((d->flags & EFILE_COMPRESSED) == 0) { - int fd; - status = efile_openfile(&d->errInfo, d->b, d->flags, &fd, NULL); - d->fd = fd; - } else { - char* mode = NULL; - - if (((d->flags & (EFILE_MODE_READ_WRITE)) == EFILE_MODE_READ_WRITE) || - (d->flags & EFILE_MODE_APPEND)) { - status = 0; - d->errInfo.posix_errno = EINVAL; - } else { - status = efile_may_openfile(&d->errInfo, d->b); - if (status || (d->errInfo.posix_errno != EISDIR)) { - mode = (d->flags & EFILE_MODE_READ) ? "rb" : "wb"; - d->fd = (SWord) erts_gzopen(d->b, mode); - if ((ErtsGzFile)d->fd) { - status = 1; - } else { - if (errno == 0) { - errno = ENOMEM; - } - d->errInfo.posix_errno = errno; - status = 0; - } - } - } - } - - d->result_ok = status; - if (!status) { - d->fd = FILE_FD_INVALID; - } - DTRACE_INVOKE_RETURN(FILE_OPEN); -} - -static void invoke_fadvise(void *data) -{ - struct t_data *d = (struct t_data *) data; - int fd = (int) d->fd; - off_t offset = (off_t) d->c.fadvise.offset; - off_t length = (off_t) d->c.fadvise.length; - int advise = (int) d->c.fadvise.advise; - DTRACE_INVOKE_SETUP(FILE_FADVISE); - - d->again = 0; - d->result_ok = efile_fadvise(&d->errInfo, fd, offset, length, advise); - DTRACE_INVOKE_RETURN(FILE_FADVISE); -} - -#ifdef HAVE_SENDFILE -static void invoke_sendfile(void *data) -{ - struct t_data *d = (struct t_data *)data; - int fd = d->fd; - int out_fd = (int)d->c.sendfile.out_fd; - Uint64 nbytes = d->c.sendfile.nbytes; - int result = 0; - d->again = 0; - - result = efile_sendfile(&d->errInfo, fd, out_fd, &d->c.sendfile.offset, &nbytes, NULL); - - d->c.sendfile.written += nbytes; - - if (result == 1 || (result == 0 && USE_THRDS_FOR_SENDFILE(d))) { - d->result_ok = 0; - } else if (result == 0 && (d->errInfo.posix_errno == EAGAIN - || d->errInfo.posix_errno == EINTR)) { - if ((d->c.sendfile.nbytes - nbytes) != 0) { - d->result_ok = 1; - if (d->c.sendfile.nbytes != 0) - d->c.sendfile.nbytes -= nbytes; - } else if (nbytes == 0 && d->c.sendfile.nbytes == 0) { - d->result_ok = 1; - } else - d->result_ok = 0; - } else { - d->result_ok = -1; - } -} - -static void free_sendfile(void *data) { - struct t_data *d = (struct t_data *)data; - if (USE_THRDS_FOR_SENDFILE(d)) { - SET_NONBLOCKING(d->c.sendfile.out_fd); - } else { - MUTEX_LOCK(d->c.sendfile.q_mtx); - driver_deq(d->c.sendfile.port,1); - MUTEX_UNLOCK(d->c.sendfile.q_mtx); - driver_select(d->c.sendfile.port, (ErlDrvEvent)(long)d->c.sendfile.out_fd, - ERL_DRV_USE_NO_CALLBACK|ERL_DRV_WRITE, 0); - } - EF_FREE(data); -} - -static void file_ready_output(ErlDrvData data, ErlDrvEvent event) -{ - file_descriptor* fd = (file_descriptor*) data; - - switch (fd->d->command) { - case FILE_SENDFILE: - driver_select(fd->d->c.sendfile.port, event, - (int)ERL_DRV_WRITE,(int) 0); - invoke_sendfile((void *)fd->d); - file_async_ready(data, (ErlDrvThreadData)fd->d); - break; - default: - break; - } -} - -static void file_stop_select(ErlDrvEvent event, void* _) -{ - -} - -static int flush_sendfile(file_descriptor *desc,void *_) { - if (desc->sendfile_state == sending) { - desc->d->result_ok = -1; - desc->d->errInfo.posix_errno = ECONNABORTED; - file_async_ready((ErlDrvData)desc,(ErlDrvThreadData)desc->d); - } - return 1; -} -#endif /* HAVE_SENDFILE */ - - -static void invoke_fallocate(void *data) -{ - struct t_data *d = (struct t_data *) data; - int fd = (int) d->fd; - Sint64 offset = d->c.fallocate.offset; - Sint64 length = d->c.fallocate.length; - - d->again = 0; - d->result_ok = efile_fallocate(&d->errInfo, fd, offset, length); -} - -static void free_readdir(void *data) -{ - struct t_data *d = (struct t_data *) data; - struct t_readdir_buf *b1 = d->c.read_dir.first_buf; - - while (b1) { - struct t_readdir_buf *b2 = b1; - b1 = b1->next; - EF_FREE(b2); - } - EF_FREE(d); -} - - - -static void try_free_read_bin(file_descriptor *desc) { - if ((desc->read_size == 0) - && (desc->read_offset >= desc->read_binp->orig_size)) { - ASSERT(desc->read_offset == desc->read_binp->orig_size); - driver_free_binary(desc->read_binp); - desc->read_binp = NULL; - desc->read_offset = 0; - desc->read_size = 0; - } -} - - - -static int try_again(file_descriptor *desc, struct t_data *d) { - if (! d->again) - return 0; - if (desc->timer_state != timer_idle) { - driver_cancel_timer(desc->port); - } - desc->timer_state = timer_again; - desc->invoke = d->invoke; - desc->d = d; - desc->free = d->free; - driver_set_timer(desc->port, 0L); - return !0; -} - - - -static void cq_execute(file_descriptor *desc) { - struct t_data *d; - register void *void_ptr; /* Soft cast variable */ - if (desc->timer_state == timer_again) - return; -#ifdef HAVE_SENDFILE - if (desc->sendfile_state == sending) - return; -#endif - if (! (d = cq_deq(desc))) - return; - TRACE_F(("x%i", (int) d->command)); - d->again = sys_info.async_threads == 0; - DRIVER_ASYNC(d->level, desc, d->invoke, void_ptr=d, d->free); -} - -static struct t_data *async_write(file_descriptor *desc, int *errp, - int reply, Uint32 reply_size -#ifdef USE_VM_PROBES - ,Sint64 *dt_i1, Sint64 *dt_i2, Sint64 *dt_i3 -#endif -) { - struct t_data *d; - if (! (d = EF_ALLOC(sizeof(struct t_data) - 1))) { - if (errp) *errp = ENOMEM; - return NULL; - } - TRACE_F(("w%lu", (unsigned long)desc->write_buffered)); - d->command = FILE_WRITE; - d->fd = desc->fd; - d->flags = desc->flags; - d->c.writev.port = desc->port; - d->c.writev.q_mtx = desc->q_mtx; - d->c.writev.size = desc->write_buffered; -#ifdef USE_VM_PROBES - if (dt_i1 != NULL) { - *dt_i1 = d->fd; - *dt_i2 = d->flags; - *dt_i3 = d->c.writev.size; - } -#endif - d->reply = reply; - d->c.writev.reply_size = reply_size; - d->invoke = invoke_writev; - d->free = free_data; - d->level = 1; - cq_enq(desc, d); - desc->write_buffered = 0; - return d; -} - -static int flush_write(file_descriptor *desc, int *errp -#ifdef USE_VM_PROBES - , dt_private *dt_priv, char *dt_utag -#endif -) { - int result = 0; -#ifdef USE_VM_PROBES - Sint64 dt_i1 = 0, dt_i2 = 0, dt_i3 = 0; -#endif - struct t_data *d = NULL; - - MUTEX_LOCK(desc->q_mtx); - if (desc->write_buffered > 0) { - if ((d = async_write(desc, errp, 0, 0 -#ifdef USE_VM_PROBES - ,&dt_i1, &dt_i2, &dt_i3 -#endif - )) == NULL) { - result = -1; - } - } - MUTEX_UNLOCK(desc->q_mtx); -#ifdef USE_VM_PROBES - if (d != NULL) { - d->sched_i1 = dt_priv->thread_num; - d->sched_i2 = dt_priv->tag; - d->sched_utag[0] = '\0'; - if (dt_utag != NULL) { - if (dt_utag[0] == '\0') { - dt_utag = NULL; - } else { - strncpy(d->sched_utag, dt_utag, sizeof(d->sched_utag) - 1); - d->sched_utag[sizeof(d->sched_utag) - 1] = '\0'; - } - } - DTRACE11(efile_drv_entry, dt_priv->thread_num, dt_priv->tag++, - dt_utag, FILE_WRITE, - NULL, NULL, dt_i1, dt_i2, dt_i3, 0, desc->port_str); - } -#endif /* USE_VM_PROBES */ - return result; -} - -static int check_write_error(file_descriptor *desc, int *errp) { - if (desc->write_error) { - if (errp) *errp = desc->write_errInfo.posix_errno; - desc->write_error = 0; - return -1; - } - return 0; -} - -static int flush_write_check_error(file_descriptor *desc, int *errp -#ifdef USE_VM_PROBES - , dt_private *dt_priv, char *dt_utag -#endif - ) { - int r; - if ( (r = flush_write(desc, errp -#ifdef USE_VM_PROBES - , dt_priv, dt_utag -#endif - )) != 0) { - check_write_error(desc, NULL); - return r; - } else { - return check_write_error(desc, errp); - } -} - -static struct t_data *async_lseek(file_descriptor *desc, int *errp, int reply, - Sint64 offset, int origin -#ifdef USE_VM_PROBES - , Sint64 *dt_i1, Sint64 *dt_i2, Sint64 *dt_i3 -#endif - ) { - struct t_data *d; - if (! (d = EF_ALLOC(sizeof(struct t_data)))) { - *errp = ENOMEM; - return NULL; - } - d->flags = desc->flags; - d->fd = desc->fd; - d->command = FILE_LSEEK; - d->reply = reply; - d->c.lseek.offset = offset; - d->c.lseek.origin = origin; -#ifdef USE_VM_PROBES - if (dt_i1 != NULL) { - *dt_i1 = d->fd; - *dt_i2 = d->c.lseek.offset; - *dt_i3 = d->c.lseek.origin; - } -#endif - d->invoke = invoke_lseek; - d->free = free_data; - d->level = 1; - cq_enq(desc, d); - return d; -} - -static void flush_read(file_descriptor *desc) { - desc->read_offset = 0; - desc->read_size = 0; - if (desc->read_binp) { - driver_free_binary(desc->read_binp); - desc->read_binp = NULL; - } -} - -static int lseek_flush_read(file_descriptor *desc, int *errp -#ifdef USE_VM_PROBES - ,dt_private *dt_priv, char *dt_utag -#endif - ) { - int r = 0; - size_t read_size = desc->read_size; -#ifdef USE_VM_PROBES - Sint64 dt_i1 = 0, dt_i2 = 0, dt_i3 = 0; -#endif - struct t_data *d; - - flush_read(desc); - if (read_size != 0) { - if ((d = async_lseek(desc, errp, 0, - -((ssize_t)read_size), EFILE_SEEK_CUR -#ifdef USE_VM_PROBES - , &dt_i1, &dt_i2, &dt_i3 -#endif - )) == NULL) { - r = -1; - } else { -#ifdef USE_VM_PROBES - d->sched_i1 = dt_priv->thread_num; - d->sched_i2 = dt_priv->tag; - d->sched_utag[0] = '\0'; - if (dt_utag != NULL) { - if (dt_utag[0] == '\0') { - dt_utag = NULL; - } else { - strncpy(d->sched_utag, dt_utag, sizeof(d->sched_utag) - 1); - d->sched_utag[sizeof(d->sched_utag) - 1] = '\0'; - } - } - DTRACE11(efile_drv_entry, dt_priv->thread_num, dt_priv->tag++, - dt_utag, FILE_LSEEK, - NULL, NULL, dt_i1, dt_i2, dt_i3, 0, desc->port_str); -#endif /* USE_VM_PROBES */ - } - } - return r; -} - - -/********************************************************************* - * Driver entry point -> stop - * The close has to be scheduled on async thread, so that currently active - * async operation does not suddenly have the ground disappearing under their feet... - */ -static void -file_stop(ErlDrvData e) -{ - file_descriptor* desc = (file_descriptor*)e; - - TRACE_C('p'); - - IF_THRDS { - flush_read(desc); - if (desc->fd != FILE_FD_INVALID) { - struct t_data *d = EF_SAFE_ALLOC(sizeof(struct t_data)); - d->command = FILE_CLOSE_ON_PORT_EXIT; - d->reply = !0; - d->fd = desc->fd; - d->flags = desc->flags; - d->invoke = invoke_close; - d->free = free_data; - d->level = 2; - d->data_to_free = (void *) desc; - cq_enq(desc, d); - desc->fd = FILE_FD_INVALID; - desc->flags = 0; - cq_execute(desc); - } else { - EF_FREE(desc); - } - } else { - if (desc->fd != FILE_FD_INVALID) { - do_close(desc->flags, desc->fd); - desc->fd = FILE_FD_INVALID; - desc->flags = 0; - } - if (desc->read_binp) { - driver_free_binary(desc->read_binp); - } - EF_FREE(desc); - } -} - -/********************************************************************* - * Driver entry point -> ready_async - */ -static void -file_async_ready(ErlDrvData e, ErlDrvThreadData data) -{ - file_descriptor *desc = (file_descriptor*)e; - struct t_data *d = (struct t_data *) data; - char header[5]; /* result code + count */ - char resbuf[RESBUFSIZE]; /* Result buffer. */ -#ifdef USE_VM_PROBES - int sched_i1 = d->sched_i1, sched_i2 = d->sched_i2, command = d->command, - result_ok = d->result_ok, - posix_errno = d->result_ok ? 0 : d->errInfo.posix_errno; - DTRACE_CHARBUF(sched_utag, DTRACE_EFILE_BUFSIZ+1); - - sched_utag[0] = '\0'; - if (DTRACE_ENABLED(efile_drv_return)) { - strncpy(sched_utag, d->sched_utag, DTRACE_EFILE_BUFSIZ); - sched_utag[DTRACE_EFILE_BUFSIZ] = '\0'; - } -#endif /* USE_VM_PROBES */ - - TRACE_C('r'); - - if (try_again(desc, d)) { - /* DTRACE TODO: what kind of probe makes sense here? */ - return; - } - - switch (d->command) - { - case FILE_READ: - if (!d->result_ok) { - reply_error(desc, &d->errInfo); - } else { - size_t available_bytes = - d->c.read.bin_offset + d->c.read.bin_size - desc->read_offset; - if (available_bytes < d->c.read.size) { - d->c.read.size = available_bytes; - } - TRACE_C('D'); - reply_data(desc, d->c.read.binp, - desc->read_offset, d->c.read.size); - desc->read_offset += d->c.read.size; - desc->read_size = - d->c.read.bin_offset + d->c.read.bin_size - desc->read_offset; - try_free_read_bin(desc); - } - free_read(data); - break; - case FILE_READ_LINE: - /* The read_line structure differs from the read structure. - The data->read_offset and d->c.read_line.read_offset are copies, as are - data->read_size and d->c.read_line.read_size - The read_line function does not kniow in advance how large the binary has to be, - why new allocation (but not reallocation of the old binary, for obvious reasons) - may happen in the worker thread. */ - if (!d->result_ok) { - reply_error(desc, &d->errInfo); - } else { - size_t len = d->c.read_line.nl_pos - d->c.read_line.read_offset; - TRACE_C('L'); - reply_data(desc, d->c.read_line.binp, - d->c.read_line.read_offset, len); - desc->read_offset = d->c.read_line.read_offset + d->c.read_line.nl_skip + len; - desc->read_size = - d->c.read_line.read_size - d->c.read_line.nl_skip - len; - if (desc->read_binp != d->c.read_line.binp) { /* New binary allocated */ - driver_free_binary(desc->read_binp); - desc->read_binp = d->c.read_line.binp; - driver_binary_inc_refc(desc->read_binp); - } -#if !ALWAYS_READ_LINE_AHEAD - ASSERT(desc->read_bufsize > 0 || desc->read_size == 0); - if (desc->read_bufsize == 0) { - desc->read_offset = desc->read_binp->orig_size; /* triggers cleanup */ - } -#endif - try_free_read_bin(desc); - } - free_read_line(data); - break; - case FILE_READ_FILE: - if (!d->result_ok) - reply_error(desc, &d->errInfo); - else { - header[0] = FILE_RESP_ALL_DATA; - TRACE_C('R'); - driver_output_binary(desc->port, header, 1, - d->c.read_file.binp, - 0, d->c.read_file.offset); - } - free_read_file(data); - break; - case FILE_WRITE: - if (d->reply) { - if (! d->result_ok) { - reply_error(desc, &d->errInfo); - } else { - reply_Uint(desc, d->c.writev.reply_size); - } - } else { - if (! d->result_ok) { - desc->write_error = !0; - desc->write_errInfo = d->errInfo; - } - } - free_data(data); - break; - case FILE_LSEEK: - if (d->reply) { - if (d->result_ok) - reply_Sint64(desc, d->c.lseek.location); - else - reply_error(desc, &d->errInfo); - } - free_data(data); - break; - case FILE_MKDIR: - case FILE_RMDIR: - case FILE_CHDIR: - case FILE_DELETE: - case FILE_FDATASYNC: - case FILE_FSYNC: - case FILE_TRUNCATE: - case FILE_LINK: - case FILE_SYMLINK: - case FILE_RENAME: - case FILE_WRITE_INFO: - case FILE_FADVISE: - case FILE_FALLOCATE: - reply(desc, d->result_ok, &d->errInfo); - free_data(data); - break; - case FILE_ALTNAME: - case FILE_PWD: - case FILE_READLINK: - { - int length; - char *resbuf = d->b; - - if (!d->result_ok) - reply_error(desc, &d->errInfo); - else { - resbuf[0] = FILE_RESP_FNAME; - length = 1+FILENAME_BYTELEN((char*) resbuf+1); - TRACE_C('R'); - driver_output2(desc->port, resbuf, 1, resbuf+1, length-1); - } - free_data(data); - break; - } - case FILE_OPEN: - if (!d->result_ok) { - reply_error(desc, &d->errInfo); - } else { - ASSERT(d->is_fd_unused); - desc->fd = d->fd; - desc->flags = d->flags; - d->is_fd_unused = 0; - reply_Uint(desc, d->fd); - } - free_data(data); - break; - case FILE_FSTAT: - case FILE_LSTAT: - { - if (d->result_ok) { - resbuf[0] = FILE_RESP_INFO; - - put_int32(d->info.size_high, &resbuf[1 + ( 0 * 4)]); - put_int32(d->info.size_low, &resbuf[1 + ( 1 * 4)]); - put_int32(d->info.type, &resbuf[1 + ( 2 * 4)]); - - /* Note 64 bit indexing in resbuf here */ - put_int64(d->info.accessTime, &resbuf[1 + ( 3 * 4)]); - put_int64(d->info.modifyTime, &resbuf[1 + ( 5 * 4)]); - put_int64(d->info.cTime, &resbuf[1 + ( 7 * 4)]); - - put_int32(d->info.mode, &resbuf[1 + ( 9 * 4)]); - put_int32(d->info.links, &resbuf[1 + (10 * 4)]); - put_int32(d->info.major_device, &resbuf[1 + (11 * 4)]); - put_int32(d->info.minor_device, &resbuf[1 + (12 * 4)]); - put_int32(d->info.inode, &resbuf[1 + (13 * 4)]); - put_int32(d->info.uid, &resbuf[1 + (14 * 4)]); - put_int32(d->info.gid, &resbuf[1 + (15 * 4)]); - put_int32(d->info.access, &resbuf[1 + (16 * 4)]); - -#define RESULT_SIZE (1 + (17 * 4)) - TRACE_C('R'); - driver_output2(desc->port, resbuf, RESULT_SIZE, NULL, 0); -#undef RESULT_SIZE - } else - reply_error(desc, &d->errInfo); - } - free_data(data); - break; - case FILE_READDIR: - if (!d->result_ok) { - reply_error(desc, &d->errInfo); - } else { - struct t_readdir_buf *b1 = d->c.read_dir.first_buf; - char op = FILE_RESP_LFNAME; - - TRACE_C('R'); - ASSERT(b1); - - while (b1) { - struct t_readdir_buf *b2 = b1; - char *p = &b1->buf[0]; - driver_output2(desc->port, p, 1, p + 1, b1->n - 1); - b1 = b1->next; - EF_FREE(b2); - } - driver_output2(desc->port, &op, 1, NULL, 0); - - d->c.read_dir.first_buf = NULL; - d->c.read_dir.last_buf = NULL; - } - free_readdir(data); - break; - case FILE_CLOSE: - if (d->reply) { - TRACE_C('K'); - reply_ok(desc); -#ifdef USE_VM_PROBES - result_ok = 1; -#endif - } - free_data(data); - break; - case FILE_PWRITEV: - if (!d->result_ok) { - reply_Uint_error(desc, d->c.pwritev.cnt, &d->errInfo); - } else { - reply_Uint(desc, d->c.pwritev.n); - } - free_data(data); - break; - case FILE_PREADV: - if (!d->result_ok) { - reply_error(desc, &d->errInfo); - } else { - reply_ev(desc, FILE_RESP_LDATA, &d->c.preadv.eiov); - } - free_preadv(data); - break; - case FILE_IPREAD: - if (!d->result_ok) { - reply_error(desc, &d->errInfo); - } else if (!d->c.preadv.eiov.vsize) { - reply_eof(desc); - } else { - reply_ev(desc, FILE_RESP_N2DATA, &d->c.preadv.eiov); - } - free_preadv(data); - break; -#ifdef HAVE_SENDFILE - case FILE_SENDFILE: - if (d->result_ok == -1) { - if (d->errInfo.posix_errno == ECONNRESET || - d->errInfo.posix_errno == ENOTCONN || - d->errInfo.posix_errno == EPIPE) - reply_string_error(desc,"closed"); - else - reply_error(desc, &d->errInfo); - desc->sendfile_state = not_sending; - free_sendfile(data); - } else if (d->result_ok == 0) { - reply_Sint64(desc, d->c.sendfile.written); - desc->sendfile_state = not_sending; - free_sendfile(data); - } else if (d->result_ok == 1) { /* If we are using select to send the rest of the data */ - desc->sendfile_state = sending; - desc->d = d; - driver_select(desc->port, (ErlDrvEvent)(long)d->c.sendfile.out_fd, - ERL_DRV_USE|ERL_DRV_WRITE, 1); - } - break; -#endif - case FILE_CLOSE_ON_PORT_EXIT: - /* See file_stop. However this is never invoked after the port is killed. */ - free_data(data); - desc = NULL; - /* This is it for this port, so just send dtrace and return, avoid doing anything to the freed data */ - DTRACE6(efile_drv_return, sched_i1, sched_i2, sched_utag, - command, result_ok, posix_errno); - return; - default: - abort(); - } - DTRACE6(efile_drv_return, sched_i1, sched_i2, sched_utag, - command, result_ok, posix_errno); - if (desc->write_buffered != 0 && desc->timer_state == timer_idle ) { - desc->timer_state = timer_write; - driver_set_timer(desc->port, desc->write_delay); - } - cq_execute(desc); - -} - - -/********************************************************************* - * Driver entry point -> output - */ -static void -file_output(ErlDrvData e, char* buf, ErlDrvSizeT count) -{ - file_descriptor* desc = (file_descriptor*)e; - Efile_error errInfo; /* The error codes for the last operation. */ - Sint fd; /* The file descriptor for this port, if any, - * -1 if none. - */ - char* name; /* Points to the filename in buf. */ - int command; - struct t_data *d = NULL; -#ifdef USE_VM_PROBES - char *dt_utag = NULL; - char *dt_s1 = NULL, *dt_s2 = NULL; - Sint64 dt_i1 = 0; - Sint64 dt_i2 = 0; - Sint64 dt_i3 = 0; - Sint64 dt_i4 = 0; - dt_private *dt_priv = get_dt_private(0); -#endif /* USE_VM_PROBES */ - - TRACE_C('o'); - - fd = desc->fd; - name = buf+1; - command = *(uchar*)buf++; - - switch(command) { - - case FILE_MKDIR: - { - d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE); - - FILENAME_COPY(d->b, name); -#ifdef USE_VM_PROBES - dt_s1 = d->b; - dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE; -#endif - d->command = command; - d->invoke = invoke_mkdir; - d->free = free_data; - d->level = 2; - goto done; - } - case FILE_RMDIR: - { - d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE); - - FILENAME_COPY(d->b, name); -#ifdef USE_VM_PROBES - dt_s1 = d->b; - dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE; -#endif - d->command = command; - d->invoke = invoke_rmdir; - d->free = free_data; - d->level = 2; - goto done; - } - case FILE_DELETE: - { - d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE); - - FILENAME_COPY(d->b, name); -#ifdef USE_VM_PROBES - dt_s1 = d->b; - dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE; -#endif - d->command = command; - d->invoke = invoke_delete_file; - d->free = free_data; - d->level = 2; - goto done; - } - case FILE_RENAME: - { - char* new_name; - int namelen = FILENAME_BYTELEN(name)+FILENAME_CHARSIZE; - new_name = name+namelen; - d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 - + namelen - + FILENAME_BYTELEN(new_name) + FILENAME_CHARSIZE); - - FILENAME_COPY(d->b, name); - FILENAME_COPY(d->b + namelen, new_name); -#ifdef USE_VM_PROBES - dt_s1 = d->b; - dt_s2 = d->b + namelen; - dt_utag = buf + namelen + FILENAME_BYTELEN(new_name) + FILENAME_CHARSIZE; -#endif - d->flags = desc->flags; - d->fd = fd; - d->command = command; - d->invoke = invoke_rename; - d->free = free_data; - d->level = 2; - goto done; - } - case FILE_CHDIR: - { - d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE); - - FILENAME_COPY(d->b, name); -#ifdef USE_VM_PROBES - dt_s1 = d->b; - dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE; -#endif - d->command = command; - d->invoke = invoke_chdir; - d->free = free_data; - d->level = 2; - goto done; - } - case FILE_PWD: - { - d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + RESBUFSIZE + 1); - - d->drive = *(uchar*)buf; -#ifdef USE_VM_PROBES - dt_utag = buf + 1; -#endif - d->command = command; - d->invoke = invoke_pwd; - d->free = free_data; - d->level = 2; - goto done; - } - - case FILE_READDIR: -#ifdef USE_THREADS - if (sys_info.async_threads > 0) - { - d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(name) + - FILENAME_CHARSIZE); - - FILENAME_COPY(d->b, name); -#ifdef USE_VM_PROBES - dt_s1 = d->b; - dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE; -#endif - d->dir_handle = NULL; - d->command = command; - d->invoke = invoke_readdir; - d->free = free_readdir; - d->level = 2; - d->c.read_dir.first_buf = NULL; - d->c.read_dir.last_buf = NULL; - goto done; - } - else -#endif - { - size_t resbufsize; - size_t n = 0, total = 0; - int res = 0; - char resbuf[READDIR_BUFSIZE]; - - EFILE_DIR_HANDLE dir_handle; /* Handle to open directory. */ - - total = READDIR_BUFSIZE; - errInfo.posix_errno = 0; - dir_handle = NULL; - resbuf[0] = FILE_RESP_LFNAME; - -#ifdef USE_VM_PROBES - dt_s1 = name; - dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE; -#endif - /* Fill the buffer with multiple directory listings before sending it to the - * receiving process. READDIR_CHUNKS is minimum number of files sent to the - * receiver. - * Format for each driver_output2: - * ------------------------------------ - * | Type | Len | Filename | ... - * | 1 byte | 2 bytes | Len bytes | ... - * ------------------------------------ - */ - - do { - n = 1; - resbufsize = READDIR_BUFSIZE - n; - - do { - res = efile_readdir(&errInfo, name, &dir_handle, resbuf + n + 2, &resbufsize); - - if (res) { - put_int16((Uint16)resbufsize, resbuf + n); - n += 2 + resbufsize; - resbufsize = READDIR_BUFSIZE - n; - } - } while( res && ((total - n - 2) >= MAXPATHLEN*FILENAME_CHARSIZE)); - - if (n > 1) { - driver_output2(desc->port, resbuf, 1, resbuf + 1, n - 1); - } - } while(res); - - if (errInfo.posix_errno != 0) { - reply_error(desc, &errInfo); - return; - } -#ifdef USE_VM_PROBES - if (dt_utag != NULL && dt_utag[0] == '\0') { - dt_utag = NULL; - } - - DTRACE11(efile_drv_entry, dt_priv->thread_num, dt_priv->tag, - dt_utag, command, name, dt_s2, - dt_i1, dt_i2, dt_i3, dt_i4, desc->port_str); - DTRACE6(efile_drv_return, dt_priv->thread_num, dt_priv->tag++, - dt_utag, command, 1, 0); -#endif - TRACE_C('R'); - driver_output2(desc->port, resbuf, 1, NULL, 0); - return; - } - case FILE_OPEN: - { - d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(buf+4) + - FILENAME_CHARSIZE); - - d->flags = get_int32((uchar*)buf); - name = buf+4; - FILENAME_COPY(d->b, name); -#ifdef USE_VM_PROBES - dt_i1 = d->flags; - dt_s1 = d->b; - dt_utag = name + FILENAME_BYTELEN(d->b) + FILENAME_CHARSIZE; -#endif - d->command = command; - d->invoke = invoke_open; - d->free = free_data; - d->level = 2; - d->is_fd_unused = 1; - goto done; - } - - case FILE_FDATASYNC: - { - d = EF_SAFE_ALLOC(sizeof(struct t_data)); - - d->fd = fd; -#ifdef USE_VM_PROBES - dt_utag = name; - dt_i1 = fd; -#endif - d->command = command; - d->invoke = invoke_fdatasync; - d->free = free_data; - d->level = 2; - goto done; - } - - case FILE_FSYNC: - { - d = EF_SAFE_ALLOC(sizeof(struct t_data)); - - d->fd = fd; -#ifdef USE_VM_PROBES - dt_utag = name; - dt_i1 = fd; -#endif - d->command = command; - d->invoke = invoke_fsync; - d->free = free_data; - d->level = 2; - goto done; - } - - - case FILE_FSTAT: - case FILE_LSTAT: - { - d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(name) + - FILENAME_CHARSIZE); - - FILENAME_COPY(d->b, name); - d->fd = fd; -#ifdef USE_VM_PROBES - dt_utag = name + FILENAME_BYTELEN(d->b) + FILENAME_CHARSIZE; - if (command == FILE_LSTAT) { - dt_s1 = d->b; - } else { - dt_i1 = fd; - } -#endif - d->command = command; - d->invoke = invoke_flstat; - d->free = free_data; - d->level = 2; - goto done; - } - - case FILE_TRUNCATE: - { - d = EF_SAFE_ALLOC(sizeof(struct t_data)); - - d->flags = desc->flags; - d->fd = fd; -#ifdef USE_VM_PROBES - dt_utag = name; - dt_i1 = fd; - dt_i2 = d->flags; -#endif - d->command = command; - d->invoke = invoke_truncate; - d->free = free_data; - d->level = 2; - goto done; - } - - case FILE_WRITE_INFO: - { - d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 - + FILENAME_BYTELEN(buf + 9*4) + FILENAME_CHARSIZE); - - d->info.mode = get_int32(buf + 0 * 4); - d->info.uid = get_int32(buf + 1 * 4); - d->info.gid = get_int32(buf + 2 * 4); - d->info.accessTime = get_int64(buf + 3 * 4); - d->info.modifyTime = get_int64(buf + 5 * 4); - d->info.cTime = get_int64(buf + 7 * 4); - - FILENAME_COPY(d->b, buf + 9*4); -#ifdef USE_VM_PROBES - dt_i1 = d->info.mode; - dt_i2 = d->info.uid; - dt_i3 = d->info.gid; - dt_s1 = d->b; - dt_utag = buf + 9 * 4 + FILENAME_BYTELEN(d->b) + FILENAME_CHARSIZE; -#endif - d->command = command; - d->invoke = invoke_write_info; - d->free = free_data; - d->level = 2; - goto done; - } - - case FILE_READLINK: - { - d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + - MAX(RESBUFSIZE, (FILENAME_BYTELEN(name) + - FILENAME_CHARSIZE)) + 1); - FILENAME_COPY(d->b, name); -#ifdef USE_VM_PROBES - dt_s1 = d->b; - dt_utag = name + FILENAME_BYTELEN(d->b) + FILENAME_CHARSIZE; -#endif - d->command = command; - d->invoke = invoke_readlink; - d->free = free_data; - d->level = 2; - goto done; - } - - case FILE_ALTNAME: - { - d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + - MAX(RESBUFSIZE, (FILENAME_BYTELEN(name) + - FILENAME_CHARSIZE)) + 1); - FILENAME_COPY(d->b, name); -#ifdef USE_VM_PROBES - dt_s1 = d->b; - dt_utag = name + FILENAME_BYTELEN(d->b) + FILENAME_CHARSIZE; -#endif - d->command = command; - d->invoke = invoke_altname; - d->free = free_data; - d->level = 2; - goto done; - } - - - case FILE_LINK: - { - char* new_name; - int namelen = FILENAME_BYTELEN(name) + FILENAME_CHARSIZE; - - new_name = name+namelen; - d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 - + namelen - + FILENAME_BYTELEN(new_name) + FILENAME_CHARSIZE); - - FILENAME_COPY(d->b, name); - FILENAME_COPY(d->b + namelen, new_name); -#ifdef USE_VM_PROBES - dt_s1 = d->b; - dt_s2 = d->b + namelen; - dt_utag = buf + namelen + FILENAME_BYTELEN(dt_s2) + FILENAME_CHARSIZE; -#endif - d->flags = desc->flags; - d->fd = fd; - d->command = command; - d->invoke = invoke_link; - d->free = free_data; - d->level = 2; - goto done; - } - - case FILE_SYMLINK: - { - char* new_name; - int namelen = FILENAME_BYTELEN(name) + FILENAME_CHARSIZE; - - new_name = name+namelen; - d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 - + namelen - + FILENAME_BYTELEN(new_name) + FILENAME_CHARSIZE); - - FILENAME_COPY(d->b, name); - FILENAME_COPY(d->b + namelen, new_name); -#ifdef USE_VM_PROBES - dt_s1 = d->b; - dt_s2 = d->b + namelen; - dt_utag = buf + namelen + FILENAME_BYTELEN(dt_s2) + FILENAME_CHARSIZE; -#endif - d->flags = desc->flags; - d->fd = fd; - d->command = command; - d->invoke = invoke_symlink; - d->free = free_data; - d->level = 2; - goto done; - } - - case FILE_FADVISE: - { - d = EF_SAFE_ALLOC(sizeof(struct t_data)); - - d->fd = fd; - d->command = command; - d->invoke = invoke_fadvise; - d->free = free_data; - d->level = 2; - d->c.fadvise.offset = get_int64((uchar*) buf); - d->c.fadvise.length = get_int64(((uchar*) buf) + sizeof(Sint64)); - d->c.fadvise.advise = get_int32(((uchar*) buf) + 2 * sizeof(Sint64)); -#ifdef USE_VM_PROBES - dt_i1 = d->fd; - dt_i2 = d->c.fadvise.offset; - dt_i3 = d->c.fadvise.length; - dt_i4 = d->c.fadvise.advise; - dt_utag = buf + 3 * sizeof(Sint64); -#endif - goto done; - } - - case FILE_FALLOCATE: - { - d = EF_SAFE_ALLOC(sizeof(struct t_data)); - - d->fd = fd; - d->command = command; - d->invoke = invoke_fallocate; - d->free = free_data; - d->level = 2; - d->c.fallocate.offset = get_int64((uchar*) buf); - d->c.fallocate.length = get_int64(((uchar*) buf) + sizeof(Sint64)); - goto done; - } - - } - - /* - * Ignore anything else -- let the caller hang. - */ - - return; - - done: - if (d) { -#ifdef USE_VM_PROBES - d->sched_i1 = dt_priv->thread_num; - d->sched_i2 = dt_priv->tag; - d->sched_utag[0] = '\0'; - if (dt_utag != NULL) { - if (dt_utag[0] == '\0') { - dt_utag = NULL; - } else { - strncpy(d->sched_utag, dt_utag, sizeof(d->sched_utag) - 1); - d->sched_utag[sizeof(d->sched_utag) - 1] = '\0'; - } - } - DTRACE11(efile_drv_entry, dt_priv->thread_num, dt_priv->tag++, - dt_utag, command, dt_s1, dt_s2, - dt_i1, dt_i2, dt_i3, dt_i4, desc->port_str); -#endif - cq_enq(desc, d); - } -} - -/********************************************************************* - * Driver entry point -> flush - */ -static void -file_flush(ErlDrvData e) { - file_descriptor *desc = (file_descriptor *)e; -#ifdef DEBUG - int r; -#endif -#ifdef USE_VM_PROBES - dt_private *dt_priv = get_dt_private(dt_driver_io_worker_base); -#endif - - TRACE_C('f'); - -#ifdef HAVE_SENDFILE - flush_sendfile(desc, NULL); -#endif - -#ifdef DEBUG - r = -#endif - flush_write(desc, NULL -#ifdef USE_VM_PROBES - , dt_priv, (desc->d == NULL) ? NULL : desc->d->sched_utag -#endif - ); - /* Only possible reason for bad return value is ENOMEM, and - * there is nobody to tell... - */ -#ifdef DEBUG - ASSERT(r == 0); -#endif - cq_execute(desc); -} - - - -/********************************************************************* - * Driver entry point -> control - * Only debug functionality... - */ -static ErlDrvSSizeT -file_control(ErlDrvData e, unsigned int command, - char* buf, ErlDrvSizeT len, char **rbuf, ErlDrvSizeT rlen) { - file_descriptor *desc = (file_descriptor *)e; - switch (command) { - case 'K' : - if (rlen < 4) { - *rbuf = EF_ALLOC(4); - } - (*rbuf)[0] = ((desc->key) >> 24) & 0xFF; - (*rbuf)[1] = ((desc->key) >> 16) & 0xFF; - (*rbuf)[2] = ((desc->key) >> 8) & 0xFF; - (*rbuf)[3] = (desc->key) & 0xFF; - return 4; - default: - return 0; - } -} - -/********************************************************************* - * Driver entry point -> timeout - */ -static void -file_timeout(ErlDrvData e) { - file_descriptor *desc = (file_descriptor *)e; - enum e_timer timer_state = desc->timer_state; -#ifdef USE_VM_PROBES - dt_private *dt_priv = get_dt_private(dt_driver_io_worker_base); -#endif - - TRACE_C('t'); - - desc->timer_state = timer_idle; - switch (timer_state) { - case timer_idle: - ASSERT(0); - break; - case timer_again: - ASSERT(desc->invoke); - ASSERT(desc->free); - driver_async(desc->port, KEY(desc), desc->invoke, desc->d, desc->free); - break; - case timer_write: { -#ifdef DEBUG - int r = -#endif - flush_write(desc, NULL -#ifdef USE_VM_PROBES - , dt_priv, (desc->d == NULL) ? NULL : desc->d->sched_utag -#endif - ); - /* Only possible reason for bad return value is ENOMEM, and - * there is nobody to tell... - */ - ASSERT(r == 0); - cq_execute(desc); - } break; - } /* case */ -} - - - -/********************************************************************* - * Driver entry point -> outputv - */ -static void -file_outputv(ErlDrvData e, ErlIOVec *ev) { - file_descriptor* desc = (file_descriptor*)e; - char command; - size_t p, q; - int err; - struct t_data *d = NULL; -#ifdef USE_VM_PROBES - Sint64 dt_i1 = 0, dt_i2 = 0, dt_i3 = 0; - Sint64 dt_i4 = 0; - char *dt_utag = NULL; - char *dt_s1 = NULL; - dt_private *dt_priv = get_dt_private(dt_driver_io_worker_base); -#endif - - TRACE_C('v'); - - p = 0; q = 1; - if (! EV_GET_CHAR(ev, &command, &p, &q)) { - /* Empty command */ - reply_posix_error(desc, EINVAL); - goto done; - } - /* 'command' contains the decoded command number, - * 'p' and 'q' point out the next byte in the command: - * ((char *)ev->iov[q].iov_base) + p; - */ - - TRACE_F(("%i", (int) command)); - - switch (command) { - - case FILE_CLOSE: { -#ifdef USE_VM_PROBES - dt_utag = EV_CHAR_P(ev, p, q); -#endif - flush_read(desc); - if (flush_write_check_error(desc, &err -#ifdef USE_VM_PROBES - , dt_priv, dt_utag -#endif - ) < 0) { - reply_posix_error(desc, err); - goto done; - } - if (desc->fd != FILE_FD_INVALID) { - if (! (d = EF_ALLOC(sizeof(struct t_data)))) { - reply_posix_error(desc, ENOMEM); - } else { - d->command = command; - d->reply = !0; - d->fd = desc->fd; - d->flags = desc->flags; -#ifdef USE_VM_PROBES - dt_i1 = d->fd; - dt_i2 = d->flags; -#endif - d->invoke = invoke_close; - d->free = free_data; - d->level = 2; - cq_enq(desc, d); - desc->fd = FILE_FD_INVALID; - desc->flags = 0; - } - } else { - reply_posix_error(desc, EBADF); - } - } goto done; - - case FILE_READ: { - Uint32 sizeH, sizeL; - size_t size, alloc_size; - - if (!EV_GET_UINT32(ev, &sizeH, &p, &q) - || !EV_GET_UINT32(ev, &sizeL, &p, &q)) { - /* Wrong buffer length to contain the read count */ - reply_posix_error(desc, EINVAL); - goto done; - } -#ifdef USE_VM_PROBES - dt_utag = EV_CHAR_P(ev, p, q); -#endif - if (flush_write_check_error(desc, &err -#ifdef USE_VM_PROBES - , dt_priv, dt_utag -#endif - ) < 0) { - reply_posix_error(desc, err); - goto done; - } -#if ALWAYS_READ_LINE_AHEAD - if (desc->read_bufsize == 0 && desc->read_binp != NULL && desc->read_size > 0) { - /* We have allocated a buffer for line mode but should not really have a - read-ahead buffer... */ - if (lseek_flush_read(desc, &err -#ifdef USE_VM_PROBES - , dt_priv, dt_utag -#endif - ) < 0) { - reply_posix_error(desc, err); - goto done; - } - } -#endif -#if SIZEOF_SIZE_T == 4 - if (sizeH != 0) { - reply_posix_error(desc, EINVAL); - goto done; - } - size = sizeL; -#else - size = ((size_t)sizeH << 32) | sizeL; -#endif - if ((desc->fd == FILE_FD_INVALID) - || (! (desc->flags & EFILE_MODE_READ)) ) { - reply_posix_error(desc, EBADF); - goto done; - } - if (size == 0) { - reply_buf(desc, &command, 0); - goto done; - } - if (desc->read_size >= size) { - /* We already have all data */ - TRACE_C('D'); - reply_data(desc, desc->read_binp, desc->read_offset, size); - desc->read_offset += size; - desc->read_size -= size; - try_free_read_bin(desc); - goto done; - } - /* We may have some of the data - */ - /* Justification for the following strange formula: - * If the read request is for such a large block as more than - * half the buffer size it may lead to a lot of unnecessary copying, - * since the tail of the old buffer is copied to the head of the - * new, and if the tail is almost half the buffer it is a lot - * to copy. Therefore allocate the exact amount needed in - * this case, giving no lingering tail. */ - alloc_size = - size > (desc->read_bufsize>>1) ? - size : desc->read_bufsize; - if (! desc->read_binp) { - /* Need to allocate a new binary for the result */ - if (! (desc->read_binp = driver_alloc_binary(alloc_size))) { - reply_posix_error(desc, ENOMEM); - goto done; - } - } else { - /* We already have a buffer */ - if (desc->read_binp->orig_size - desc->read_offset < size) { - /* Need to allocate a new binary for the result */ - ErlDrvBinary *binp; - if (! (binp = driver_alloc_binary(alloc_size))) { - reply_posix_error(desc, ENOMEM); - goto done; - } - /* Move data we already have to the new binary */ - sys_memcpy(binp->orig_bytes, - desc->read_binp->orig_bytes + desc->read_offset, - desc->read_size); - driver_free_binary(desc->read_binp); - desc->read_offset = 0; - desc->read_binp = binp; - } - } - if (! (d = EF_ALLOC(sizeof(struct t_data)))) { - reply_posix_error(desc, ENOMEM); - goto done; - } - d->command = command; - d->reply = !0; - d->fd = desc->fd; - d->flags = desc->flags; - d->c.read.binp = desc->read_binp; - d->c.read.bin_offset = desc->read_offset + desc->read_size; - d->c.read.bin_size = desc->read_binp->orig_size - d->c.read.bin_offset; - d->c.read.size = size; -#ifdef USE_VM_PROBES - dt_i1 = d->fd; - dt_i2 = d->flags; - dt_i3 = d->c.read.size; -#endif - driver_binary_inc_refc(d->c.read.binp); - d->invoke = invoke_read; - d->free = free_read; - d->level = 1; - cq_enq(desc, d); - } goto done; /* case FILE_READ: */ - - case FILE_READ_LINE: { - /* - * Icky little creature... We do mostly as ordinary file read, but with a few differences. - * 1) We have to scan for proper newline sequence if there is a buffer already, we cannot know - * in advance if the buffer contains a whole line without scanning. - * 2) We do not know how large the buffer needs to be in advance. We give a default buffer, - * but the worker may need to allocate a new one. Freeing the old and rereferencing a newly - * allocated binary + dealing with offsets and lengts are done in file_async ready - * for this OP. - */ -#ifdef USE_VM_PROBES - dt_utag = EV_CHAR_P(ev, p, q); -#endif - if (flush_write_check_error(desc, &err -#ifdef USE_VM_PROBES - , dt_priv, dt_utag -#endif - ) < 0) { - reply_posix_error(desc, err); - goto done; - } - if (ev->size != 1 -#ifdef USE_VM_PROBES - + FILENAME_BYTELEN(dt_utag) + FILENAME_CHARSIZE -#endif - ) { - /* Wrong command length */ - reply_posix_error(desc, EINVAL); - goto done; - } - if ((desc->fd == FILE_FD_INVALID) - || (! (desc->flags & EFILE_MODE_READ)) ) { - reply_posix_error(desc, EBADF); - goto done; - } - if (desc->read_size > 0) { - /* look for '\n' in what we'we already got */ - void *nl_ptr = memchr(desc->read_binp->orig_bytes + desc->read_offset,'\n',desc->read_size); - if (nl_ptr != NULL) { - /* If found, we're done */ - int skip = 0; - size_t size = ((char *) nl_ptr) - - ((char *) (desc->read_binp->orig_bytes + desc->read_offset)) + 1; - if (size > 1 && - *(((char *) nl_ptr) - 1) == '\r') { - *(((char *) nl_ptr) - 1) = '\n'; - skip = 1; - --size; - } - reply_data(desc, desc->read_binp, desc->read_offset, size); - desc->read_offset += (size + skip); - desc->read_size -= (size + skip); - try_free_read_bin(desc); - goto done; - } - } - /* Now, it's up to the thread to work out the need for more buffers and such, it's - no use doing it in this thread as we do not have the information required anyway. - Even a NULL buffer could be handled by the thread, but code is simplified by us - allocating it */ - if (! desc->read_binp) { - int alloc_size = (desc->read_bufsize > DEFAULT_LINEBUF_SIZE) ? desc->read_bufsize : - DEFAULT_LINEBUF_SIZE; - /* Allocate a new binary for the result */ - if (! (desc->read_binp = driver_alloc_binary(alloc_size))) { - reply_posix_error(desc, ENOMEM); - goto done; - } - } - if (! (d = EF_ALLOC(sizeof(struct t_data)))) { - reply_posix_error(desc, ENOMEM); - goto done; - } - - d->command = command; - d->reply = !0; - d->fd = desc->fd; - d->flags = desc->flags; - d->c.read_line.binp = desc->read_binp; - d->c.read_line.read_offset = desc->read_offset; - d->c.read_line.read_size = desc->read_size; -#ifdef USE_VM_PROBES - dt_i1 = d->fd; - dt_i2 = d->flags; - dt_i3 = d->c.read_line.read_offset; -#endif -#if !ALWAYS_READ_LINE_AHEAD - d->c.read_line.read_ahead = (desc->read_bufsize > 0); -#ifdef USE_VM_PROBES - dt_i4 = d->c.read_line.read_ahead; -#endif -#endif - driver_binary_inc_refc(d->c.read.binp); - d->invoke = invoke_read_line; - d->free = free_read_line; - d->level = 1; - cq_enq(desc, d); - } goto done; - case FILE_WRITE: { /* Dtrace: The dtrace user tag is not last in message, - but follows the message tag directly. - This is handled specially in prim_file.erl */ - ErlDrvSizeT skip = 1; - ErlDrvSizeT size = ev->size - skip; - -#ifdef USE_VM_PROBES - dt_utag = EV_CHAR_P(ev, p, q); - skip += FILENAME_BYTELEN(dt_utag) + FILENAME_CHARSIZE; - size = ev->size - skip; -#endif - if (lseek_flush_read(desc, &err -#ifdef USE_VM_PROBES - , dt_priv, dt_utag -#endif - ) < 0) { - reply_posix_error(desc, err); - goto done; - } - if (! (desc->flags & EFILE_MODE_WRITE)) { - reply_posix_error(desc, EBADF); - goto done; - } - if (size == 0) { - reply_Uint(desc, size); - goto done; - } - MUTEX_LOCK(desc->q_mtx); - if (driver_enqv(desc->port, ev, skip)) { - MUTEX_UNLOCK(desc->q_mtx); - reply_posix_error(desc, ENOMEM); - goto done; - } - desc->write_buffered += size; - if (desc->write_buffered < desc->write_bufsize) { - MUTEX_UNLOCK(desc->q_mtx); - reply_Uint(desc, size); - if (desc->timer_state == timer_idle) { - desc->timer_state = timer_write; - driver_set_timer(desc->port, desc->write_delay); - } - } else { - if ((d = async_write(desc, &err, !0, size -#ifdef USE_VM_PROBES - , &dt_i1, &dt_i2, &dt_i3 -#endif - )) == NULL) { - MUTEX_UNLOCK(desc->q_mtx); - reply_posix_error(desc, err); - goto done; - } else { - MUTEX_UNLOCK(desc->q_mtx); - } - } - } goto done; /* case FILE_WRITE */ - - case FILE_PWRITEV: { /* Dtrace: The dtrace user tag is not last in message, - but follows the message tag directly. - This is handled specially in prim_file.erl */ - Uint32 i, j, n; - size_t total; -#ifdef USE_VM_PROBES - char dt_tmp; - int dt_utag_bytes = 1; - - dt_utag = EV_CHAR_P(ev, p, q); - /* This will work for UTF-8, but not for UTF-16 - extra reminder here */ -#ifdef FILENAMES_16BIT -#error 16bit characters in filenames and dtrace in combination is not supported. -#endif - while (EV_GET_CHAR(ev, &dt_tmp, &p, &q) && dt_tmp != '\0') { - dt_utag_bytes++; - } -#endif - if (ev->size < 1+4 -#ifdef USE_VM_PROBES - + dt_utag_bytes -#endif - || !EV_GET_UINT32(ev, &n, &p, &q)) { - /* Buffer too short to contain even the number of pos/size specs */ - reply_Uint_posix_error(desc, 0, EINVAL); - goto done; - } - if (lseek_flush_read(desc, &err -#ifdef USE_VM_PROBES - , dt_priv, dt_utag -#endif - ) < 0) { - reply_Uint_posix_error(desc, 0, err); - goto done; - } - if (flush_write_check_error(desc, &err -#ifdef USE_VM_PROBES - , dt_priv, dt_utag -#endif - ) < 0) { - reply_Uint_posix_error(desc, 0, err); - goto done; - } - if (n == 0) { - /* Trivial case - nothing to write */ - if (ev->size != 1+4) { - reply_posix_error(desc, err); - } else { - reply_Uint(desc, 0); - } - goto done; - } - if (ev->size < 1+4+8*(2*n) -#ifdef USE_VM_PROBES - + dt_utag_bytes -#endif - ) { - /* Buffer too short to contain even the pos/size specs */ - reply_Uint_posix_error(desc, 0, EINVAL); - goto done; - } - d = EF_ALLOC(sizeof(struct t_data) - + (n * sizeof(struct t_pbuf_spec))); - if (! d) { - reply_Uint_posix_error(desc, 0, ENOMEM); - goto done; - } - d->command = command; - d->reply = !0; - d->fd = desc->fd; - d->flags = desc->flags; -#ifdef USE_VM_PROBES - dt_i1 = d->fd; - dt_i2 = d->flags; -#endif - d->c.pwritev.port = desc->port; - d->c.pwritev.q_mtx = desc->q_mtx; - d->c.pwritev.n = n; - d->c.pwritev.cnt = 0; - total = 0; - j = 0; - /* Create pos/size specs in the thread data structure - * for all non-zero size binaries. Calculate total size. - */ - for(i = 0; i < n; i++) { - Uint32 sizeH, sizeL; - size_t size; - if ( !EV_GET_SINT64(ev, &d->c.pwritev.specs[i].offset, &p, &q) - || !EV_GET_UINT32(ev, &sizeH, &p, &q) - || !EV_GET_UINT32(ev, &sizeL, &p, &q)) { - /* Misalignment in buffer */ - reply_Uint_posix_error(desc, 0, EINVAL); - EF_FREE(d); - goto done; - } -#if SIZEOF_SIZE_T == 4 - if (sizeH != 0) { - reply_Uint_posix_error(desc, 0, EINVAL); - EF_FREE(d); - goto done; - } - size = sizeL; -#else - size = ((size_t)sizeH<<32) | sizeL; -#endif - if (size > 0) { - total += size; - d->c.pwritev.specs[j].size = size; - j++; - } - } - d->c.pwritev.size = total; -#ifdef USE_VM_PROBES - dt_i3 = d->c.pwritev.size; -#endif - if (j == 0) { - /* Trivial case - nothing to write */ - EF_FREE(d); - reply_Uint(desc, 0); - } else { - ErlDrvSizeT skip = 1 + 4 + 8 * (2*n) -#ifdef USE_VM_PROBES - + dt_utag_bytes -#endif - ; - if (skip + total != ev->size) { - /* Actual amount of data does not match - * total of all pos/size specs - */ - EF_FREE(d); - reply_Uint_posix_error(desc, 0, EINVAL); - } else { - /* Enqueue the data */ - MUTEX_LOCK(desc->q_mtx); - driver_enqv(desc->port, ev, skip); - MUTEX_UNLOCK(desc->q_mtx); - /* Execute the command */ - d->invoke = invoke_pwritev; - d->free = free_data; - d->level = 1; - cq_enq(desc, d); - } - } - } goto done; /* case FILE_PWRITEV: */ - - case FILE_PREADV: { /* Dtrace: The dtrace user tag is not last in message, - but follows the message tag directly. - This is handled specially in prim_file.erl */ - register void * void_ptr; - Uint32 i, n; - ErlIOVec *res_ev; -#ifdef USE_VM_PROBES - char dt_tmp; - int dt_utag_bytes = 1; - /* This will work for UTF-8, but not for UTF-16 - extra reminder here */ -#ifdef FILENAMES_16BIT -#error 16bit characters in filenames and dtrace in combination is not supported. -#endif - dt_utag = EV_CHAR_P(ev, p, q); - while (EV_GET_CHAR(ev, &dt_tmp, &p, &q) && dt_tmp != '\0') { - dt_utag_bytes++; - } -#endif - if (lseek_flush_read(desc, &err -#ifdef USE_VM_PROBES - , dt_priv, dt_utag -#endif - ) < 0) { - reply_posix_error(desc, err); - goto done; - } - if (flush_write_check_error(desc, &err -#ifdef USE_VM_PROBES - , dt_priv, dt_utag -#endif - ) < 0) { - reply_posix_error(desc, err); - goto done; - } - if (ev->size < 1+8 -#ifdef USE_VM_PROBES - + dt_utag_bytes -#endif - || !EV_GET_UINT32(ev, &n, &p, &q) - || !EV_GET_UINT32(ev, &n, &p, &q)) { - /* Buffer too short to contain even the number of pos/size specs */ - reply_posix_error(desc, EINVAL); - goto done; - } - if (ev->size < 1+8+8*(2*n) -#ifdef USE_VM_PROBES - + dt_utag_bytes -#endif - ) { - /* Buffer wrong length to contain the pos/size specs */ - reply_posix_error(desc, EINVAL); - goto done; - } - /* Create the thread data structure with the contained ErlIOVec - * and corresponding binaries for the response - */ - d = EF_ALLOC(sizeof(*d) - + (n * sizeof(*d->c.preadv.offsets)) - + ((1+n) * (sizeof(*res_ev->iov) - + sizeof(*res_ev->binv)))); - if (! d) { - reply_posix_error(desc, ENOMEM); - goto done; - } - d->command = command; - d->reply = !0; - d->fd = desc->fd; - d->flags = desc->flags; -#ifdef USE_VM_PROBES - dt_i1 = d->fd; - dt_i2 = d->flags; -#endif - d->c.preadv.n = n; - d->c.preadv.cnt = 0; - d->c.preadv.size = 0; - res_ev = &d->c.preadv.eiov; - /* XXX possible alignment problems here for weird machines */ - res_ev->vsize = 1+d->c.preadv.n; - res_ev->iov = void_ptr = &d->c.preadv.offsets[d->c.preadv.n]; - res_ev->binv = void_ptr = &res_ev->iov[res_ev->vsize]; - /* Read in the pos/size specs and allocate binaries for the results */ - for (i = 1; i < 1+n; i++) { - Uint32 sizeH, sizeL; - size_t size; - if ( !EV_GET_SINT64(ev, &d->c.preadv.offsets[i-1], &p, &q) - || !EV_GET_UINT32(ev, &sizeH, &p, &q) - || !EV_GET_UINT32(ev, &sizeL, &p, &q)) { - reply_posix_error(desc, EINVAL); - break; - } -#if SIZEOF_SIZE_T == 4 - if (sizeH != 0) { - reply_posix_error(desc, EINVAL); - break; - } - size = sizeL; -#else - size = ((size_t)sizeH<<32) | sizeL; -#endif -#ifdef USE_VM_PROBES - dt_i3 += size; -#endif - if (! (res_ev->binv[i] = driver_alloc_binary(size))) { - reply_posix_error(desc, ENOMEM); - break; - } else { - res_ev->iov[i].iov_len = size; - res_ev->iov[i].iov_base = res_ev->binv[i]->orig_bytes; - } - } - if (i < 1+n) { - for (i--; i > 0; i--) { - driver_free_binary(res_ev->binv[i]); - } - EF_FREE(d); - goto done; - } - /* Allocate the header binary (index 0) */ - res_ev->binv[0] = driver_alloc_binary(4+4+8*n); - if (! res_ev->binv[0]) { - reply_posix_error(desc, ENOMEM); - for (i = 1; i < 1+n; i++) { - driver_free_binary(res_ev->binv[i]); - } - EF_FREE(d); - goto done; - } - res_ev->iov[0].iov_len = 4+4+8*n; - res_ev->iov[0].iov_base = res_ev->binv[0]->orig_bytes; - /* Fill in the number of buffers in the header */ - put_int32(0, res_ev->iov[0].iov_base); - put_int32(n, (char *)(res_ev->iov[0].iov_base) + 4); - /**/ - res_ev->size = res_ev->iov[0].iov_len; - if (n == 0) { - /* Trivial case - nothing to read */ - reply_ev(desc, FILE_RESP_LDATA, res_ev); - free_preadv(d); - goto done; - } else { - d->invoke = invoke_preadv; - d->free = free_preadv; - d->level = 1; - cq_enq(desc, d); - } - } goto done; /* case FILE_PREADV: */ - - case FILE_LSEEK: { - Sint64 offset; /* Offset for seek */ - Uint32 origin; /* Origin of seek. */ - - if (ev->size < 1+8+4 - || !EV_GET_SINT64(ev, &offset, &p, &q) - || !EV_GET_UINT32(ev, &origin, &p, &q)) { - /* Wrong length of buffer to contain offset and origin */ - reply_posix_error(desc, EINVAL); - goto done; - } -#ifdef USE_VM_PROBES - dt_utag = EV_CHAR_P(ev, p, q); -#endif - if (lseek_flush_read(desc, &err -#ifdef USE_VM_PROBES - , dt_priv, dt_utag -#endif - ) < 0) { - reply_posix_error(desc, err); - goto done; - } - if (flush_write_check_error(desc, &err -#ifdef USE_VM_PROBES - , dt_priv, dt_utag -#endif - ) < 0) { - reply_posix_error(desc, err); - goto done; - } - if ((d = async_lseek(desc, &err, !0, offset, origin -#ifdef USE_VM_PROBES - , &dt_i1, &dt_i2, &dt_i3 -#endif - )) == NULL) { - reply_posix_error(desc, err); - goto done; - } - } goto done; - - case FILE_READ_FILE: { - char *filename; - if (ev->size < 1+1) { - /* Buffer contains empty name */ - reply_posix_error(desc, ENOENT); - goto done; - } -#ifndef USE_VM_PROBES - /* In the dtrace case, the iov has an extra element, the dtrace utag - we will need - another test to see that - the filename is in a single buffer: */ - if (ev->size-1 != ev->iov[q].iov_len-p) { - /* Name not in one single buffer */ - reply_posix_error(desc, EINVAL); - goto done; - } -#else - if (((byte *)ev->iov[q].iov_base)[ev->iov[q].iov_len-1] != '\0') { - /* Name not in one single buffer */ - reply_posix_error(desc, EINVAL); - goto done; - } -#endif - filename = EV_CHAR_P(ev, p, q); - d = EF_ALLOC(sizeof(struct t_data) -1 + FILENAME_BYTELEN(filename) + FILENAME_CHARSIZE); - if (! d) { - reply_posix_error(desc, ENOMEM); - goto done; - } - d->command = command; - d->reply = !0; - /* Copy name */ - FILENAME_COPY(d->b, filename); -#ifdef USE_VM_PROBES - { - char dt_tmp; - - /* This will work for UTF-8, but not for UTF-16 - extra reminder here */ -#ifdef FILENAMES_16BIT -#error 16bit characters in filenames and dtrace in combination is not supported. -#endif - while (EV_GET_CHAR(ev, &dt_tmp, &p, &q) && dt_tmp != '\0') - ; - dt_s1 = d->b; - dt_utag = EV_CHAR_P(ev, p, q); - } -#endif - d->c.read_file.binp = NULL; - d->invoke = invoke_read_file; - d->free = free_read_file; - d->level = 2; - cq_enq(desc, d); - } goto done; - - case FILE_IPREAD: { - /* This operation cheets by using invoke_preadv() and free_preadv() - * plus its own invoke_ipread. Therefore the result format is - * a bit awkward - the header binary contains one extra 64 bit - * field that invoke_preadv() fortunately ignores, - * and the first 64 bit field does not contain the number of - * data binaries which invoke_preadv() also ignores. - */ - register void * void_ptr; - char mode; - Sint64 hdr_offset; - Uint32 max_size; - ErlIOVec *res_ev; - int vsize; - if (! EV_GET_CHAR(ev, &mode, &p, &q)) { - /* Empty command */ - reply_posix_error(desc, EINVAL); - goto done; - } - if (mode != IPREAD_S32BU_P32BU) { - reply_posix_error(desc, EINVAL); - goto done; - } - if (ev->size < 1+1+8+4 - || !EV_GET_SINT64(ev, &hdr_offset, &p, &q) - || !EV_GET_UINT32(ev, &max_size, &p, &q)) { - /* Buffer too short to contain - * the header offset and max size spec */ - reply_posix_error(desc, EINVAL); - goto done; - } -#ifdef USE_VM_PROBES - dt_utag = EV_CHAR_P(ev, p, q); -#endif - if (lseek_flush_read(desc, &err -#ifdef USE_VM_PROBES - , dt_priv, dt_utag -#endif - ) < 0) { - reply_posix_error(desc, err); - goto done; - } - if (flush_write_check_error(desc, &err -#ifdef USE_VM_PROBES - , dt_priv, dt_utag -#endif - ) < 0) { - reply_posix_error(desc, err); - goto done; - } - /* Create the thread data structure with the contained ErlIOVec - * and corresponding binaries for the response - */ - vsize = 2; - d = EF_ALLOC(sizeof(*d) + - vsize*(sizeof(*res_ev->iov) + sizeof(*res_ev->binv))); - if (! d) { - reply_posix_error(desc, ENOMEM); - goto done; - } - d->command = command; - d->reply = !0; - d->fd = desc->fd; - d->flags = desc->flags; - d->c.preadv.offsets[0] = hdr_offset; - d->c.preadv.size = max_size; -#ifdef USE_VM_PROBES - dt_i1 = d->fd; - dt_i2 = d->flags; - dt_i3 = d->c.preadv.offsets[0]; - dt_i4 = d->c.preadv.size; -#endif - res_ev = &d->c.preadv.eiov; - /* XXX possible alignment problems here for weird machines */ - res_ev->iov = void_ptr = d + 1; - res_ev->binv = void_ptr = res_ev->iov + vsize; - res_ev->size = 0; - res_ev->vsize = 0; - d->invoke = invoke_ipread; - d->free = free_preadv; - d->level = 1; - cq_enq(desc, d); - } goto done; /* case FILE_IPREAD: */ - - case FILE_SETOPT: { - char opt; - - if (ev->size < 1+1 - || !EV_GET_CHAR(ev, &opt, &p, &q)) { - /* Buffer too short to contain even the option type */ - reply_posix_error(desc, EINVAL); - goto done; - } -#ifdef USE_VM_PROBES - dt_i1 = opt; - dt_utag = EV_CHAR_P(ev, p, q); -#endif - switch (opt) { - case FILE_OPT_DELAYED_WRITE: { - Uint32 sizeH, sizeL, delayH, delayL; - if (ev->size != 1+1+4*sizeof(Uint32) -#ifdef USE_VM_PROBES - + FILENAME_BYTELEN(dt_utag) + FILENAME_CHARSIZE -#endif - || !EV_GET_UINT32(ev, &sizeH, &p, &q) - || !EV_GET_UINT32(ev, &sizeL, &p, &q) - || !EV_GET_UINT32(ev, &delayH, &p, &q) - || !EV_GET_UINT32(ev, &delayL, &p, &q)) { - /* Buffer has wrong length to contain the option values */ - reply_posix_error(desc, EINVAL); - goto done; - } -#if SIZEOF_SIZE_T == 4 - if (sizeH != 0) { - reply_posix_error(desc, EINVAL); - goto done; - } - desc->write_bufsize = sizeL; -#else - desc->write_bufsize = ((size_t)sizeH << 32) | sizeL; -#endif -#if SIZEOF_LONG == 4 - if (delayH != 0) { - reply_posix_error(desc, EINVAL); - goto done; - } - desc->write_delay = delayL; -#else - desc->write_delay = ((unsigned long)delayH << 32) | delayL; -#endif -#ifdef USE_VM_PROBES - dt_i2 = desc->write_delay; -#endif - TRACE_C('K'); - reply_ok(desc); - } goto done; - case FILE_OPT_READ_AHEAD: { - Uint32 sizeH, sizeL; - if (ev->size != 1+1+2*sizeof(Uint32) -#ifdef USE_VM_PROBES - + FILENAME_BYTELEN(dt_utag)+FILENAME_CHARSIZE -#endif - || !EV_GET_UINT32(ev, &sizeH, &p, &q) - || !EV_GET_UINT32(ev, &sizeL, &p, &q)) { - /* Buffer has wrong length to contain the option values */ - reply_posix_error(desc, EINVAL); - goto done; - } -#if SIZEOF_SIZE_T == 4 - if (sizeH != 0) { - reply_posix_error(desc, EINVAL); - goto done; - } - desc->read_bufsize = sizeL; -#else - desc->read_bufsize = ((size_t)sizeH << 32) | sizeL; -#endif -#ifdef USE_VM_PROBES - dt_i2 = desc->read_bufsize; -#endif - TRACE_C('K'); - reply_ok(desc); - } goto done; - default: - reply_posix_error(desc, EINVAL); - goto done; - } /* case FILE_OPT_DELAYED_WRITE: */ - } ASSERT(0); goto done; /* case FILE_SETOPT: */ - - case FILE_SENDFILE: { - -#ifdef HAVE_SENDFILE - struct t_data *d; - Uint32 out_fd, offsetH, offsetL, hd_len, tl_len; - Uint64 nbytes; - char flags; - - if (ev->size < 1 + 7 * sizeof(Uint32) + sizeof(char) - || !EV_GET_UINT32(ev, &out_fd, &p, &q) - || !EV_GET_CHAR(ev, &flags, &p, &q) - || !EV_GET_UINT32(ev, &offsetH, &p, &q) - || !EV_GET_UINT32(ev, &offsetL, &p, &q) - || !EV_GET_UINT64(ev, &nbytes, &p, &q) - || !EV_GET_UINT32(ev, &hd_len, &p, &q) - || !EV_GET_UINT32(ev, &tl_len, &p, &q)) { - /* Buffer has wrong length to contain all the needed values */ - reply_posix_error(desc, EINVAL); - goto done; - } - - if (hd_len != 0 || tl_len != 0) { - /* We do not allow header, trailers */ - reply_posix_error(desc, EINVAL); - goto done; - } - - - if (flags & SENDFILE_FLGS_USE_THREADS && !THRDS_AVAILABLE) { - /* We do not allow use_threads flag on a system where - no threads are available. */ - reply_posix_error(desc, EINVAL); - goto done; - } - - d = EF_SAFE_ALLOC(sizeof(struct t_data)); - d->fd = desc->fd; - d->command = command; - d->invoke = invoke_sendfile; - d->free = free_sendfile; - d->flags = flags; - d->level = 2; - - d->c.sendfile.out_fd = (int) out_fd; - d->c.sendfile.written = 0; - d->c.sendfile.port = desc->port; - d->c.sendfile.q_mtx = desc->q_mtx; - - #if SIZEOF_OFF_T == 4 - if (offsetH != 0) { - reply_posix_error(desc, EINVAL); - goto done; - } - d->c.sendfile.offset = (off_t) offsetL; - #else - d->c.sendfile.offset = ((off_t) offsetH << 32) | offsetL; - #endif - - d->c.sendfile.nbytes = nbytes; - - if (USE_THRDS_FOR_SENDFILE(d)) { - SET_BLOCKING(d->c.sendfile.out_fd); - } else { - /** - * Write a place holder to queue in order to force file_flush - * to be called before the driver is closed. - */ - char tmp[1] = ""; - MUTEX_LOCK(d->c.sendfile.q_mtx); - if (driver_enq(d->c.sendfile.port, tmp, 1)) { - MUTEX_UNLOCK(d->c.sendfile.q_mtx); - reply_posix_error(desc, ENOMEM); - goto done; - } - MUTEX_UNLOCK(d->c.sendfile.q_mtx); - } - - cq_enq(desc, d); -#else - reply_posix_error(desc, ENOTSUP); -#endif - goto done; - } /* case FILE_SENDFILE: */ - - } /* switch(command) */ - - if (lseek_flush_read(desc, &err -#ifdef USE_VM_PROBES - , dt_priv, dt_utag -#endif - ) < 0) { - reply_posix_error(desc, err); - goto done; - } - if (flush_write_check_error(desc, &err -#ifdef USE_VM_PROBES - , dt_priv, dt_utag -#endif - ) < 0) { - reply_posix_error(desc, err); - goto done; - } else { - /* Flatten buffer and send it to file_output(desc, buf, len) */ - int len = ev->size; - char *buf = EF_ALLOC(len); - if (! buf) { - reply_posix_error(desc, ENOMEM); - goto done; - } - driver_vec_to_buf(ev, buf, len); - file_output((ErlDrvData) desc, buf, len); - EF_FREE(buf); - goto done; - } - - done: - if (d != NULL) { -#ifdef USE_VM_PROBES - /* - * If d == NULL, then either: - * 1). There was an error of some sort, or - * 2). The command given to us is actually implemented - * by file_output() instead. - * - * Case #1 is probably a TODO item, perhaps? - * Case #2 we definitely don't want to activate a probe. - */ - d->sched_i1 = dt_priv->thread_num; - d->sched_i2 = dt_priv->tag; - d->sched_utag[0] = '\0'; - if (dt_utag != NULL) { - if (dt_utag[0] == '\0') { - dt_utag = NULL; - } else { - strncpy(d->sched_utag, dt_utag, sizeof(d->sched_utag) - 1); - d->sched_utag[sizeof(d->sched_utag) - 1] = '\0'; - } - } - DTRACE11(efile_drv_entry, dt_priv->thread_num, dt_priv->tag++, - dt_utag, command, dt_s1, NULL, dt_i1, dt_i2, dt_i3, dt_i4, - desc->port_str); -#endif - } - cq_execute(desc); -} - -#ifdef USE_VM_PROBES -dt_private * -get_dt_private(int base) -{ - dt_private *dt_priv = (dt_private *) pthread_getspecific(dt_driver_key); - - if (dt_priv == NULL) { - dt_priv = EF_SAFE_ALLOC(sizeof(dt_private)); - erts_mtx_lock(&dt_driver_mutex); - dt_priv->thread_num = (base + dt_driver_idnum++); - erts_mtx_unlock(&dt_driver_mutex); - dt_priv->tag = 0; - pthread_setspecific(dt_driver_key, dt_priv); - } - return dt_priv; -} -#endif /* USE_VM_PROBES */ diff --git a/erts/emulator/drivers/common/erl_efile.h b/erts/emulator/drivers/common/erl_efile.h deleted file mode 100644 index b7f063b4f2..0000000000 --- a/erts/emulator/drivers/common/erl_efile.h +++ /dev/null @@ -1,176 +0,0 @@ -/* - * %CopyrightBegin% - * - * Copyright Ericsson AB 1997-2016. 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. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - * - * %CopyrightEnd% - */ -/* - * Defines the interfaces between the generic efile driver and its - * operating-system dependent helpers. - */ - -#include "sys.h" -#include "erl_driver.h" - -/* - * Open modes for efile_openfile(). - */ -#define EFILE_MODE_READ 1 -#define EFILE_MODE_WRITE 2 /* Implies truncating file when used alone. */ -#define EFILE_MODE_READ_WRITE 3 -#define EFILE_MODE_APPEND 4 -#define EFILE_COMPRESSED 8 -#define EFILE_MODE_EXCL 16 -#define EFILE_NO_TRUNCATE 32 /* Special for reopening on VxWorks */ -#define EFILE_MODE_SYNC 64 - -/* - * Seek modes for efile_seek(). - */ -#define EFILE_SEEK_SET 0 -#define EFILE_SEEK_CUR 1 -#define EFILE_SEEK_END 2 - -/* - * File types returned by efile_fileinfo(). - */ -#define FT_DEVICE 1 -#define FT_DIRECTORY 2 -#define FT_REGULAR 3 -#define FT_SYMLINK 4 -#define FT_OTHER 5 - -/* - * Access attributes returned by efile_fileinfo() (the bits can be ORed - * together). - */ -#define FA_NONE 0 -#define FA_WRITE 1 -#define FA_READ 2 - -/* Some OS'es (i.e. Windows) has filenames in wide charaqcters. That requires special handling */ -/* Note that we do *not* honor alignment in the communication to the OS specific driver, */ -/* which is not a problem on x86, but might be on other platforms. The OS specific efile */ -/* implementation is expected to align if needed */ -#ifdef __WIN32__ -#define FILENAMES_16BIT 1 -#endif - -/* We use sendfilev if it exist on solaris */ -#if !defined(HAVE_SENDFILE) && defined(HAVE_SENDFILEV) -#define HAVE_SENDFILE -#endif - -/* - * An handle to an open directory. To be cast to the correct type - * in the system-dependent directory functions. - */ - -typedef struct _Efile_Dir_Handle* EFILE_DIR_HANDLE; - -/* - * Error information from the last call. - */ -typedef struct _Efile_error { - int posix_errno; /* Posix error number, as in <errno.h>. */ - int os_errno; /* Os-dependent error number (not used). */ -} Efile_error; - -/* - * Describes what is returned by file:file_info/1. - */ - -typedef struct _Efile_info { - Uint32 size_low; /* Size of file, lower 32 bits.. */ - Uint32 size_high; /* Size of file, higher 32 bits. */ - Uint32 type; /* Type of file -- one of FT_*. */ - Uint32 access; /* Access to file -- one of FA_*. */ - Uint32 mode; /* Access permissions -- bit field. */ - Uint32 links; /* Number of links to file. */ - Uint32 major_device; /* Major device or file system. */ - Uint32 minor_device; /* Minor device (for devices). */ - Uint32 inode; /* Inode number. */ - Uint32 uid; /* User id of owner. */ - Uint32 gid; /* Group id of owner. */ - Sint64 accessTime; /* Last time the file was accessed. */ - Sint64 modifyTime; /* Last time the file was modified. */ - Sint64 cTime; /* Creation time (Windows) or last - * inode change (Unix). - */ -} Efile_info; - - -#ifdef HAVE_SENDFILE -/* - * Describes the structure of headers/trailers for sendfile - */ -struct t_sendfile_hdtl { - SysIOVec *headers; - int hdr_cnt; - SysIOVec *trailers; - int trl_cnt; -}; -#endif /* HAVE_SENDFILE */ - -/* - * Functions. - */ -int efile_init(void); -int efile_mkdir(Efile_error* errInfo, char* name); -int efile_rmdir(Efile_error* errInfo, char* name); -int efile_delete_file(Efile_error* errInfo, char* name); -int efile_rename(Efile_error* errInfo, char* src, char* dst); -int efile_chdir(Efile_error* errInfo, char* name); -int efile_getdcwd(Efile_error* errInfo, int drive, - char* buffer, size_t size); -int efile_readdir(Efile_error* errInfo, char* name, - EFILE_DIR_HANDLE* dir_handle, - char* buffer, size_t *size); -int efile_openfile(Efile_error* errInfo, char* name, int flags, - int* pfd, Sint64* pSize); -void efile_closefile(int fd); -int efile_fdatasync(Efile_error* errInfo, int fd); -int efile_fsync(Efile_error* errInfo, int fd); -int efile_fileinfo(Efile_error* errInfo, Efile_info* pInfo, - char *name, int info_for_link); -int efile_write_info(Efile_error* errInfo, Efile_info* pInfo, char *name); -int efile_write(Efile_error* errInfo, int flags, int fd, - char* buf, size_t count); -int efile_writev(Efile_error* errInfo, int flags, int fd, - SysIOVec* iov, int iovcnt); -int efile_read(Efile_error* errInfo, int flags, int fd, - char* buf, size_t count, size_t* pBytesRead); -int efile_seek(Efile_error* errInfo, int fd, - Sint64 offset, int origin, Sint64* new_location); -int efile_truncate_file(Efile_error* errInfo, int *fd, int flags); -int efile_pwrite(Efile_error* errInfo, int fd, - char* buf, size_t count, Sint64 offset); -int efile_pread(Efile_error* errInfo, int fd, - Sint64 offset, char* buf, size_t count, size_t* pBytesRead); -int efile_readlink(Efile_error* errInfo, char *name, - char* buffer, size_t size); -int efile_altname(Efile_error* errInfo, char *name, - char* buffer, size_t size); -int efile_link(Efile_error* errInfo, char* old, char* new); -int efile_symlink(Efile_error* errInfo, char* old, char* new); -int efile_may_openfile(Efile_error* errInfo, char *name); -int efile_fadvise(Efile_error* errInfo, int fd, Sint64 offset, Sint64 length, - int advise); -#ifdef HAVE_SENDFILE -int efile_sendfile(Efile_error* errInfo, int in_fd, int out_fd, - off_t *offset, Uint64 *nbytes, struct t_sendfile_hdtl *hdtl); -#endif /* HAVE_SENDFILE */ -int efile_fallocate(Efile_error* errInfo, int fd, Sint64 offset, Sint64 length); diff --git a/erts/emulator/drivers/common/gzio.c b/erts/emulator/drivers/common/gzio.c index f60c781894..86c3b07cea 100644 --- a/erts/emulator/drivers/common/gzio.c +++ b/erts/emulator/drivers/common/gzio.c @@ -19,726 +19,16 @@ #include <unistd.h> #endif #include <ctype.h> + #include "erl_driver.h" -#include "erl_efile.h" #include "sys.h" -#ifdef __WIN32__ -#ifndef HAVE_CONFLICTING_FREAD_DECLARATION -#define HAVE_CONFLICTING_FREAD_DECLARATION -#endif -#define FILENAMES_16BIT 1 -#endif - -#ifdef STDC -# define zstrerror(errnum) strerror(errnum) -#else -# define zstrerror(errnum) "" -#endif - #include "gzio_zutil.h" #include "erl_zlib.h" #include "gzio.h" -/********struct internal_state {int dummy;}; / * for buggy compilers */ - -#define Z_BUFSIZE 4096 - -#define ALLOC(size) driver_alloc(size) -#define TRYFREE(p) {if (p) driver_free(p);} - static int gz_magic[2] = {0x1f, 0x8b}; /* gzip magic header */ -/* gzip flag byte */ -#define ASCII_FLAG 0x01 /* bit 0 set: file probably ascii text */ -#define HEAD_CRC 0x02 /* bit 1 set: header CRC present */ -#define EXTRA_FIELD 0x04 /* bit 2 set: extra field present */ -#define ORIG_NAME 0x08 /* bit 3 set: original file name present */ -#define COMMENT 0x10 /* bit 4 set: file comment present */ -#define RESERVED 0xE0 /* bits 5..7: reserved */ - -typedef struct gz_stream { - z_stream stream; - int z_err; /* error code for last stream operation */ - int z_eof; /* set if end of input file */ -#ifdef UNIX - int file; /* .gz file descriptor */ -#else - FILE *file; /* .gz file */ -#endif - Byte *inbuf; /* input buffer */ - Byte *outbuf; /* output buffer */ - uLong crc; /* crc32 of uncompressed data */ - char *msg; /* error message */ - char *path; /* path name for debugging only */ - int transparent; /* 1 if input file is not a .gz file */ - char mode; /* 'w' or 'r' */ - int position; /* Position (for seek) */ - int (*destroy)(struct gz_stream*); /* Function to destroy - * this structure. */ -} gz_stream; - -local ErtsGzFile gz_open (const char *path, const char *mode); -local int get_byte (gz_stream *s); -local void check_header (gz_stream *s); -local int destroy (gz_stream *s); -local uLong getLong (gz_stream *s); - -#ifdef UNIX -/* - * In Solaris 8 and earlier, fopen() and its friends cannot handle - * file descriptors larger than 255. Therefore, we use read()/write() - * on all Unix systems. - */ -# define ERTS_GZWRITE(File, Buf, Count) write((File), (Buf), (Count)) -# define ERTS_GZREAD(File, Buf, Count) read((File), (Buf), (Count)) -#else -/* - * On all other operating systems, using fopen(), fread()/fwrite(), since - * there is not guaranteed to exist any read()/write() (not part of - * ANSI/ISO-C). - */ -# define ERTS_GZWRITE(File, Buf, Count) fwrite((Buf), 1, (Count), (File)) -# define ERTS_GZREAD(File, Buf, Count) fread((Buf), 1, (Count), (File)) -#endif - -/* - * Ripped from efile_drv.c - */ - -#ifdef FILENAMES_16BIT -# define FILENAME_BYTELEN(Str) filename_len_16bit(Str) -# define FILENAME_COPY(To,From) filename_cpy_16bit((To),(From)) -# define FILENAME_CHARSIZE 2 - - static int filename_len_16bit(const char *str) - { - const char *p = str; - while(*p != '\0' || p[1] != '\0') { - p += 2; - } - return (p - str); - } - - static void filename_cpy_16bit(char *to, const char *from) - { - while(*from != '\0' || from[1] != '\0') { - *to++ = *from++; - *to++ = *from++; - } - *to++ = *from++; - *to++ = *from++; - } - -#else -# define FILENAME_BYTELEN(Str) strlen(Str) -# define FILENAME_COPY(To,From) strcpy(To,From) -# define FILENAME_CHARSIZE 1 -#endif - -/* =========================================================================== - Opens a gzip (.gz) file for reading or writing. The mode parameter - is as in fopen ("rb" or "wb"). The file is given either by file descriptor - or path name (if fd == -1). - gz_open return NULL if the file could not be opened or if there was - insufficient memory to allocate the (de)compression state; errno - can be checked to distinguish the two cases (if errno is zero, the - zlib error is Z_MEM_ERROR). -*/ -local ErtsGzFile gz_open (path, mode) - const char *path; - const char *mode; -{ - int err; - int level = Z_DEFAULT_COMPRESSION; /* compression level */ - char *p = (char*)mode; - gz_stream *s; - char fmode[80]; /* copy of mode, without the compression level */ - char *m = fmode; - - if (!path || !mode) return Z_NULL; - - s = (gz_stream *)ALLOC(sizeof(gz_stream)); - if (!s) return Z_NULL; - - erl_zlib_alloc_init(&s->stream); - s->stream.next_in = s->inbuf = Z_NULL; - s->stream.next_out = s->outbuf = Z_NULL; - s->stream.avail_in = s->stream.avail_out = 0; -#ifdef UNIX - s->file = -1; -#else - s->file = NULL; -#endif - s->z_err = Z_OK; - s->z_eof = 0; - s->crc = crc32(0L, Z_NULL, 0); - s->msg = NULL; - s->transparent = 0; - s->position = 0; - s->destroy = destroy; - - s->path = (char*)ALLOC(FILENAME_BYTELEN(path)+FILENAME_CHARSIZE); - if (s->path == NULL) { - return s->destroy(s), (ErtsGzFile)Z_NULL; - } - FILENAME_COPY(s->path, path); /* do this early for debugging */ - - s->mode = '\0'; - do { - if (*p == 'r') - s->mode = 'r'; - if (*p == 'w' || *p == 'a') - s->mode = 'w'; - if (isdigit((int)*p)) { - level = *p - '0'; - } else { - *m++ = *p; /* Copy the mode */ - } - } while (*p++ && m < fmode + sizeof(fmode) - 1); - *m = '\0'; - if (s->mode == '\0') - return s->destroy(s), (ErtsGzFile)Z_NULL; - - if (s->mode == 'w') { - err = deflateInit2(&(s->stream), level, - Z_DEFLATED, MAX_WBITS+16, DEF_MEM_LEVEL, 0); - /* windowBits is passed < 0 to suppress zlib header */ - - s->stream.next_out = s->outbuf = (Byte*)ALLOC(Z_BUFSIZE); - - if (err != Z_OK || s->outbuf == Z_NULL) { - return s->destroy(s), (ErtsGzFile)Z_NULL; - } - } else { - /* - * It is tempting to use the built-in support in zlib - * for handling GZIP headers, but unfortunately it - * cannot handle multiple GZIP headers (which occur when - * several GZIP files have been concatenated). - */ - - err = inflateInit2(&(s->stream), -MAX_WBITS); - s->stream.next_in = s->inbuf = (Byte*)ALLOC(Z_BUFSIZE); - - if (err != Z_OK || s->inbuf == Z_NULL) { - return s->destroy(s), (ErtsGzFile)Z_NULL; - } - } - s->stream.avail_out = Z_BUFSIZE; - - errno = 0; -#if defined(FILENAMES_16BIT) - { - FILE* efile_wfopen(const WCHAR* name, const WCHAR* mode); - WCHAR wfmode[80]; - int i = 0; - int j; - for(j = 0; fmode[j] != '\0'; ++j) { - wfmode[i++] = (WCHAR) fmode[j]; - } - wfmode[i++] = L'\0'; - s->file = efile_wfopen((WCHAR *)path, wfmode); - if (s->file == NULL) { - return s->destroy(s), (ErtsGzFile)Z_NULL; - } - } -#elif defined(UNIX) - if (s->mode == 'r') { - s->file = open(path, O_RDONLY); - } else { - s->file = open(path, O_WRONLY | O_CREAT | O_TRUNC, 0666); - } - if (s->file == -1) { - return s->destroy(s), (ErtsGzFile)Z_NULL; - } -#else - s->file = fopen(path, fmode); - if (s->file == NULL) { - return s->destroy(s), (ErtsGzFile)Z_NULL; - } -#endif - if (s->mode == 'r') { - check_header(s); /* skip the .gz header */ - } - return (ErtsGzFile)s; -} - -/* =========================================================================== - Rewind a gzfile back to the beginning. -*/ - -local int gz_rewind (gz_stream *s) -{ - TRYFREE(s->msg); - -#ifdef UNIX - lseek(s->file, 0L, SEEK_SET); -#else - fseek(s->file, 0L, SEEK_SET); -#endif - inflateReset(&(s->stream)); - s->stream.next_in = Z_NULL; - s->stream.next_out = Z_NULL; - s->stream.avail_in = s->stream.avail_out = 0; - s->z_err = Z_OK; - s->z_eof = 0; - s->crc = crc32(0L, Z_NULL, 0); - s->msg = NULL; - s->position = 0; - s->stream.next_in = s->inbuf; - - s->stream.avail_out = Z_BUFSIZE; - - check_header(s); /* skip the .gz header */ - return 1; -} - -/* =========================================================================== - Opens a gzip (.gz) file for reading or writing. -*/ -ErtsGzFile erts_gzopen (path, mode) - const char *path; - const char *mode; -{ - return gz_open (path, mode); -} - - -/* =========================================================================== - Read a byte from a gz_stream; update next_in and avail_in. Return EOF - for end of file. - IN assertion: the stream s has been successfully opened for reading. -*/ -local int get_byte(s) - gz_stream *s; -{ - if (s->z_eof) return EOF; - if (s->stream.avail_in == 0) { -#ifdef UNIX - ssize_t res; - errno = 0; - res = ERTS_GZREAD(s->file, s->inbuf, Z_BUFSIZE); - if (res == 0) { - s->stream.avail_in = 0; - s->z_eof = 1; - return EOF; - } else if (res < 0) { - s->stream.avail_in = 0; - s->z_eof = 1; - s->z_err = Z_ERRNO; - return EOF; - } else { - s->stream.avail_in = (uInt) res; - } -#else - errno = 0; - s->stream.avail_in = ERTS_GZREAD(s->file, s->inbuf, Z_BUFSIZE); - if (s->stream.avail_in == 0) { - s->z_eof = 1; - if (s->file && ferror(s->file)) - s->z_err = Z_ERRNO; - return EOF; - } -#endif - s->stream.next_in = s->inbuf; - } - s->stream.avail_in--; - return *(s->stream.next_in)++; -} - -/* =========================================================================== - Check the gzip header of a gz_stream opened for reading. Set the stream - mode to transparent if the gzip magic header is not present; set s->err - to Z_DATA_ERROR if the magic header is present but the rest of the header - is incorrect. - IN assertion: the stream s has already been created sucessfully; - s->stream.avail_in is zero for the first time, but may be non-zero - for concatenated .gz files. -*/ -local void check_header(s) - gz_stream *s; -{ - int method; /* method byte */ - int flags; /* flags byte */ - uInt len; - int c; - - /* Check the gzip magic header */ - for (len = 0; len < 2; len++) { - c = get_byte(s); - if (c != gz_magic[len]) { - if (len != 0) s->stream.avail_in++, s->stream.next_in--; - if (c != EOF) { - s->stream.avail_in++, s->stream.next_in--; - s->transparent = 1; - } - s->z_err = s->stream.avail_in != 0 ? Z_OK : Z_STREAM_END; - return; - } - } - method = get_byte(s); - flags = get_byte(s); - if (method != Z_DEFLATED || (flags & RESERVED) != 0) { - s->z_err = Z_DATA_ERROR; - return; - } - - /* Discard time, xflags and OS code: */ - for (len = 0; len < 6; len++) (void)get_byte(s); - - if ((flags & EXTRA_FIELD) != 0) { /* skip the extra field */ - len = (uInt)get_byte(s); - len += ((uInt)get_byte(s))<<8; - /* len is garbage if EOF but the loop below will quit anyway */ - while (len-- != 0 && get_byte(s) != EOF) ; - } - if ((flags & ORIG_NAME) != 0) { /* skip the original file name */ - while ((c = get_byte(s)) != 0 && c != EOF) ; - } - if ((flags & COMMENT) != 0) { /* skip the .gz file comment */ - while ((c = get_byte(s)) != 0 && c != EOF) ; - } - if ((flags & HEAD_CRC) != 0) { /* skip the header crc */ - for (len = 0; len < 2; len++) (void)get_byte(s); - } - s->z_err = s->z_eof ? Z_DATA_ERROR : Z_OK; -} - - /* =========================================================================== - * Cleanup then free the given gz_stream. Return a zlib error code. - Try freeing in the reverse order of allocations. - */ -local int destroy (s) - gz_stream *s; -{ - int err = Z_OK; - - if (!s) return Z_STREAM_ERROR; - - TRYFREE(s->msg); - - if (s->stream.state != NULL) { - if (s->mode == 'w') { - err = deflateEnd(&(s->stream)); - } else if (s->mode == 'r') { - err = inflateEnd(&(s->stream)); - } - } -#ifdef UNIX - if (s->file != -1 && close(s->file)) { - err = Z_ERRNO; - } -#else - if (s->file != NULL && fclose(s->file)) { - err = Z_ERRNO; - } -#endif - if (s->z_err < 0) err = s->z_err; - - TRYFREE(s->inbuf); - TRYFREE(s->outbuf); - TRYFREE(s->path); - TRYFREE(s); - return err; -} - -/* =========================================================================== - Reads the given number of uncompressed bytes from the compressed file. - gzread returns the number of bytes actually read (0 for end of file). -*/ -int -erts_gzread(ErtsGzFile file, voidp buf, unsigned len) -{ - gz_stream *s = (gz_stream*)file; - Bytef *start = buf; /* starting point for crc computation */ - Byte *next_out; /* == stream.next_out but not forced far (for MSDOS) */ - - if (s == NULL || s->mode != 'r') return Z_STREAM_ERROR; - - if (s->z_err == Z_DATA_ERROR || s->z_err == Z_ERRNO) return -1; - if (s->z_err == Z_STREAM_END) return 0; /* EOF */ - - s->stream.next_out = next_out = buf; - s->stream.avail_out = len; - - while (s->stream.avail_out != 0) { - - if (s->transparent) { - /* Copy first the lookahead bytes: */ - uInt n = s->stream.avail_in; - if (n > s->stream.avail_out) n = s->stream.avail_out; - if (n > 0) { - zmemcpy(s->stream.next_out, s->stream.next_in, n); - next_out += n; - s->stream.next_out = next_out; - s->stream.next_in += n; - s->stream.avail_out -= n; - s->stream.avail_in -= n; - } - if (s->stream.avail_out > 0) { - s->stream.avail_out -= ERTS_GZREAD(s->file, next_out, - s->stream.avail_out); - } - len -= s->stream.avail_out; - s->stream.total_in += (uLong)len; - s->stream.total_out += (uLong)len; - if (len == 0) s->z_eof = 1; - s->position += (int)len; - return (int)len; - } - if (s->stream.avail_in == 0 && !s->z_eof) { -#ifdef UNIX - ssize_t res; - errno = 0; - res = ERTS_GZREAD(s->file, s->inbuf, Z_BUFSIZE); - if (res == 0) { - s->stream.avail_in = 0; - s->z_eof = 1; - return EOF; - } else if (res < 0) { - s->stream.avail_in = 0; - s->z_eof = 1; - s->z_err = Z_ERRNO; - return EOF; - } else { - s->stream.avail_in = (uInt) res; - } -#else - errno = 0; - s->stream.avail_in = ERTS_GZREAD(s->file, s->inbuf, Z_BUFSIZE); - if (s->stream.avail_in == 0) { - s->z_eof = 1; - if (s->file && ferror(s->file)) { - s->z_err = Z_ERRNO; - break; - } - } -#endif - s->stream.next_in = s->inbuf; - } - s->z_err = inflate(&(s->stream), Z_NO_FLUSH); - - if (s->z_err == Z_STREAM_END) { - /* Check CRC and original size */ - s->crc = crc32(s->crc, start, (uInt)(s->stream.next_out - start)); - start = s->stream.next_out; - - if (getLong(s) != s->crc) { - s->z_err = Z_DATA_ERROR; - } else { - (void)getLong(s); - /* The uncompressed length returned by above getlong() may - * be different from s->stream.total_out) in case of - * concatenated .gz files. Check for such files: - */ - check_header(s); - if (s->z_err == Z_OK) { - uLong total_in = s->stream.total_in; - uLong total_out = s->stream.total_out; - - inflateReset(&(s->stream)); - s->stream.total_in = total_in; - s->stream.total_out = total_out; - s->crc = crc32(0L, Z_NULL, 0); - } - } - } - if (s->z_err != Z_OK || s->z_eof) break; - } - s->crc = crc32(s->crc, start, (uInt)(s->stream.next_out - start)); - - s->position += (int)(len - s->stream.avail_out); - - return (int)(len - s->stream.avail_out); -} - -/* =========================================================================== - Writes the given number of uncompressed bytes into the compressed file. - gzwrite returns the number of bytes actually written (0 in case of error). -*/ -int -erts_gzwrite(ErtsGzFile file, voidp buf, unsigned len) -{ - gz_stream *s = (gz_stream*)file; - - if (s == NULL || s->mode != 'w') return Z_STREAM_ERROR; - - s->stream.next_in = buf; - s->stream.avail_in = len; - - while (s->stream.avail_in != 0) { - - if (s->stream.avail_out == 0) { - - s->stream.next_out = s->outbuf; - if (ERTS_GZWRITE(s->file, s->outbuf, Z_BUFSIZE) != Z_BUFSIZE) { - s->z_err = Z_ERRNO; - break; - } - s->stream.avail_out = Z_BUFSIZE; - } - s->z_err = deflate(&(s->stream), Z_NO_FLUSH); - if (s->z_err != Z_OK) break; - } - s->position += (int)(len - s->stream.avail_in); - return (int)(len - s->stream.avail_in); -} - -/* - * For use by Erlang file driver. - * - * XXX Limitations: - * - SEEK_END is not allowed (length of file is not known). - * - When writing, only forward seek is supported. - */ - -int -erts_gzseek(ErtsGzFile file, int offset, int whence) -{ - int pos; - gz_stream* s = (gz_stream *) file; - - switch (whence) { - case EFILE_SEEK_SET: whence = SEEK_SET; break; - case EFILE_SEEK_CUR: whence = SEEK_CUR; break; - case EFILE_SEEK_END: whence = SEEK_END; break; - default: - errno = EINVAL; - return -1; - } - - if (s == NULL) { - errno = EINVAL; - return -1; - } - if (s->z_err == Z_DATA_ERROR || s->z_err == Z_ERRNO) { - errno = EIO; - return -1; - } - - switch (whence) { - case SEEK_SET: pos = offset; break; - case SEEK_CUR: pos = s->position+offset; break; - case SEEK_END: - default: - errno = EINVAL; return -1; - } - - if (pos == s->position) { - return pos; - } - - if (pos < s->position) { - if (s->mode == 'w') { - errno = EINVAL; - return -1; - } - gz_rewind(s); - } - - while (s->position < pos) { - char buf[512]; - int n; - int save_pos = s->position; - - n = pos - s->position; - if (n > sizeof(buf)) - n = sizeof(buf); - - if (s->mode == 'r') { - erts_gzread(file, buf, n); - } else { - memset(buf, '\0', n); - erts_gzwrite(file, buf, n); - } - if (save_pos == s->position) break; - } - - return s->position; -} - -/* =========================================================================== - Flushes all pending output into the compressed file. The parameter - flush is as in the deflate() function. - gzflush should be called only when strictly necessary because it can - degrade compression. -*/ -int -erts_gzflush(ErtsGzFile file, int flush) -{ - uInt len; - int done = 0; - gz_stream *s = (gz_stream*)file; - - if (s == NULL || s->mode != 'w') return Z_STREAM_ERROR; - - s->stream.avail_in = 0; /* should be zero already anyway */ - - for (;;) { - len = Z_BUFSIZE - s->stream.avail_out; - - if (len != 0) { - if ((uInt)ERTS_GZWRITE(s->file, s->outbuf, len) != len) { - s->z_err = Z_ERRNO; - return Z_ERRNO; - } - s->stream.next_out = s->outbuf; - s->stream.avail_out = Z_BUFSIZE; - } - if (done) break; - s->z_err = deflate(&(s->stream), flush); - - /* deflate has finished flushing only when it hasn't used up - * all the available space in the output buffer: - */ - done = (s->stream.avail_out != 0 || s->z_err == Z_STREAM_END); - - if (s->z_err != Z_OK && s->z_err != Z_STREAM_END) break; - } -#ifndef UNIX - fflush(s->file); -#endif - return s->z_err == Z_STREAM_END ? Z_OK : s->z_err; -} - -/* =========================================================================== - Reads a long in LSB order from the given gz_stream. Sets -*/ -local uLong getLong (s) - gz_stream *s; -{ - uLong x = (uLong)get_byte(s); - int c; - - x += ((uLong)get_byte(s))<<8; - x += ((uLong)get_byte(s))<<16; - c = get_byte(s); - if (c == EOF) s->z_err = Z_DATA_ERROR; - x += ((uLong)c)<<24; - return x; -} - -/* =========================================================================== - Flushes all pending output if necessary, closes the compressed file - and deallocates all the (de)compression state. -*/ -int -erts_gzclose(ErtsGzFile file) -{ - int err; - gz_stream *s = (gz_stream*)file; - - if (s == NULL) return Z_STREAM_ERROR; - - if (s->mode == 'w') { - err = erts_gzflush (file, Z_FINISH); - if (err != Z_OK) return s->destroy(s); - } - return s->destroy(s); -} - - /* =========================================================================== Uncompresses the buffer given and returns a pointer to a binary. If the buffer was not compressed with gzip, the buffer contents diff --git a/erts/emulator/drivers/common/gzio.h b/erts/emulator/drivers/common/gzio.h index ee0ebe7bd8..20433a1a17 100644 --- a/erts/emulator/drivers/common/gzio.h +++ b/erts/emulator/drivers/common/gzio.h @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1999-2016. All Rights Reserved. + * Copyright Ericsson AB 1999-2018. 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. @@ -20,13 +20,5 @@ #include "zlib.h" -typedef struct erts_gzFile* ErtsGzFile; - -ErtsGzFile erts_gzopen (const char *path, const char *mode); -int erts_gzread(ErtsGzFile file, voidp buf, unsigned len); -int erts_gzwrite(ErtsGzFile file, voidp buf, unsigned len); -int erts_gzseek(ErtsGzFile, int, int); -int erts_gzflush(ErtsGzFile file, int flush); -int erts_gzclose(ErtsGzFile file); ErlDrvBinary* erts_gzinflate_buffer(char*, uLong); ErlDrvBinary* erts_gzdeflate_buffer(char*, uLong); diff --git a/erts/emulator/drivers/common/inet_drv.c b/erts/emulator/drivers/common/inet_drv.c index 554c48059f..b71ce0389d 100644 --- a/erts/emulator/drivers/common/inet_drv.c +++ b/erts/emulator/drivers/common/inet_drv.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1997-2017. All Rights Reserved. + * Copyright Ericsson AB 1997-2018. 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. @@ -38,6 +38,7 @@ #include <ctype.h> #include <sys/types.h> #include <errno.h> +#include <stdint.h> #define IDENTITY(c) c #define STRINGIFY_1(b) IDENTITY(#b) @@ -63,6 +64,20 @@ #include <sys/un.h> #endif +#ifdef HAVE_SENDFILE +#if defined(__linux__) || (defined(__sun) && defined(__SVR4)) + #include <sys/sendfile.h> +#elif defined(__FreeBSD__) || defined(__DragonFly__) + /* Need to define __BSD_VISIBLE in order to expose prototype of sendfile */ + #define __BSD_VISIBLE 1 + #include <sys/socket.h> +#endif +#endif + +#if defined(__APPLE__) && defined(__MACH__) && !defined(__DARWIN__) + #define __DARWIN__ 1 +#endif + /* All platforms fail on malloc errors. */ #define FATAL_MALLOC @@ -559,7 +574,7 @@ static int my_strncasecmp(const char *s1, const char *s2, size_t n) driver_select(port, e, mode | (on?ERL_DRV_USE:0), on) #define sock_select(d, flags, onoff) do { \ - ASSERT(!(d)->is_ignored); \ + ASSERT(!INET_IGNORED(d)); \ (d)->event_mask = (onoff) ? \ ((d)->event_mask | (flags)) : \ ((d)->event_mask & ~(flags)); \ @@ -610,6 +625,30 @@ static size_t my_strnlen(const char *s, size_t maxlen) # define VALGRIND_MAKE_MEM_DEFINED(ptr,size) #endif +#ifndef __WIN32__ +/* Calculate CMSG_NXTHDR without having a struct msghdr*. + * CMSG_LEN only caters for alignment for start of data. + * To get how much to advance we need to use CMSG_SPACE + * on the payload length. To get the payload length we + * take the calculated cmsg->cmsg_len and subtract the + * header length. To get the header length we use + * the pointer difference from the cmsg start pointer + * to the CMSG_DATA(cmsg) pointer. + * + * Some platforms (seen on ppc Linux 2.6.29-3.ydl61.3) + * may return 0 as the cmsg_len if the cmsg is to be ignored. + */ +#define LEN_CMSG_DATA(cmsg) \ + ((cmsg)->cmsg_len < sizeof (struct cmsghdr) ? 0 : \ + (cmsg)->cmsg_len - ((char*)CMSG_DATA(cmsg) - (char*)(cmsg))) +#define NXT_CMSG_HDR(cmsg) \ + ((struct cmsghdr*)(((char*)(cmsg)) + CMSG_SPACE(LEN_CMSG_DATA(cmsg)))) +#endif + +#if !defined(IPV6_PKTOPTIONS) && defined(IPV6_2292PKTOPTIONS) +#define IPV6_PKTOPTIONS IPV6_2292PKTOPTIONS +#endif + /* Magic errno value used locally for return of {error, system_limit} - the emulator definition of SYSTEM_LIMIT is not available here. @@ -701,6 +740,7 @@ static size_t my_strnlen(const char *s, size_t maxlen) #define TCP_REQ_RECV 42 #define TCP_REQ_UNRECV 43 #define TCP_REQ_SHUTDOWN 44 +#define TCP_REQ_SENDFILE 45 /* UDP and SCTP requests */ #define PACKET_REQ_RECV 60 /* Common for UDP and SCTP */ /* #define SCTP_REQ_LISTEN 61 MERGED Different from TCP; not for UDP */ @@ -723,6 +763,7 @@ static size_t my_strnlen(const char *s, size_t maxlen) #define TCP_ADDF_DELAYED_ECONNRESET 128 /* An ECONNRESET error occurred on send or shutdown */ #define TCP_ADDF_SHUTDOWN_WR_DONE 256 /* A shutdown(sock, SHUT_WR) or SHUT_RDWR was made */ #define TCP_ADDF_LINGER_ZERO 512 /* Discard driver queue on port close */ +#define TCP_ADDF_SENDFILE 1024 /* Send from an fd instead of the driver queue */ /* *_REQ_* replies */ #define INET_REP_ERROR 0 @@ -771,6 +812,12 @@ static size_t my_strnlen(const char *s, size_t maxlen) #define INET_LOPT_LINE_DELIM 40 /* Line delimiting char */ #define INET_OPT_TCLASS 41 /* IPv6 transport class */ #define INET_OPT_BIND_TO_DEVICE 42 /* get/set network device the socket is bound to */ +#define INET_OPT_RECVTOS 43 /* IP_RECVTOS ancillary data */ +#define INET_OPT_RECVTCLASS 44 /* IPV6_RECVTCLASS ancillary data */ +#define INET_OPT_PKTOPTIONS 45 /* IP(V6)_PKTOPTIONS get ancillary data */ +#define INET_OPT_TTL 46 /* IP_TTL */ +#define INET_OPT_RECVTTL 47 /* IP_RECVTTL ancillary data */ +#define TCP_OPT_NOPUSH 48 /* super-Nagle, aka TCP_CORK */ /* SCTP options: a separate range, from 100: */ #define SCTP_OPT_RTOINFO 100 #define SCTP_OPT_ASSOCINFO 101 @@ -846,6 +893,20 @@ static size_t my_strnlen(const char *s, size_t maxlen) #define SCTP_FLAG_SACDELAY_ENABLE (32 /* am_sackdelay_enable */) #define SCTP_FLAG_SACDELAY_DISABLE (64 /* am_sackdelay_disable */) +/* Flags for recv_cmsgflags */ +#define INET_CMSG_RECVTOS (1 << 0) /* am_recvtos, am_tos */ +#define INET_CMSG_RECVTCLASS (1 << 1) /* am_recvtclass, am_tclass */ +#define INET_CMSG_RECVTTL (1 << 2) /* am_recvttl, am_ttl */ + +/* Inet flags */ +#define INET_FLG_BUFFER_SET (1 << 0) /* am_buffer has been set by user */ +#define INET_FLG_IS_IGNORED (1 << 1) /* If a fd is ignored by the inet_drv. + This flag should be set to true when + the fd is used outside of inet_drv. */ +#define INET_FLG_IS_IGNORED_RD (1 << 2) +#define INET_FLG_IS_IGNORED_WR (1 << 3) +#define INET_FLG_IS_IGNORED_PASS (1 << 4) + /* ** End of interface constants. **--------------------------------------------------------------------------*/ @@ -891,18 +952,31 @@ static size_t my_strnlen(const char *s, size_t maxlen) #define INET_IFNAMSIZ 16 /* INET Ignore states */ -#define INET_IGNORE_NONE 0 -#define INET_IGNORE_READ (1 << 0) -#define INET_IGNORE_WRITE (1 << 1) -#define INET_IGNORE_PASSIVE (1 << 2) +#define INET_IGNORE_CLEAR(desc) ((desc)->flags & ~(INET_IGNORE_READ|INET_IGNORE_WRITE|INET_IGNORE_PASSIVE)) +#define INET_IGNORED(desc) ((desc)->flags & INET_FLG_IS_IGNORED) +#define INET_IGNORE_READ (INET_FLG_IS_IGNORED|INET_FLG_IS_IGNORED_RD) +#define INET_IGNORE_WRITE (INET_FLG_IS_IGNORED|INET_FLG_IS_IGNORED_WR) +#define INET_IGNORE_PASSIVE (INET_FLG_IS_IGNORED|INET_FLG_IS_IGNORED_PASS) /* Max length of Erlang Term Buffer (for outputting structured terms): */ #ifdef HAVE_SCTP #define PACKET_ERL_DRV_TERM_DATA_LEN 512 #else +#ifndef __WIN32__ +/* Assume we have recvmsg() and might need room for ancillary data */ +#define PACKET_ERL_DRV_TERM_DATA_LEN 64 +#else #define PACKET_ERL_DRV_TERM_DATA_LEN 32 #endif +#endif +typedef struct _tcp_descriptor tcp_descriptor; + +#if defined(TCP_CORK) +#define INET_TCP_NOPUSH TCP_CORK +#elif defined(TCP_NOPUSH) && !defined(__DARWIN__) +#define INET_TCP_NOPUSH TCP_NOPUSH +#endif #define BIN_REALLOC_MARGIN(x) ((x)/4) /* 25% */ @@ -952,16 +1026,19 @@ typedef struct _multi_timer_data { struct _multi_timer_data *prev; } MultiTimerData; -static MultiTimerData *add_multi_timer(MultiTimerData **first, ErlDrvPort port, - ErlDrvTermData caller, unsigned timeout, - void (*timeout_fun)(ErlDrvData drv_data, - ErlDrvTermData caller)); -static void fire_multi_timers(MultiTimerData **first, ErlDrvPort port, +static MultiTimerData *add_multi_timer(tcp_descriptor *desc, ErlDrvPort port, + ErlDrvTermData caller, unsigned timeout, + void (*timeout_fun)(ErlDrvData drv_data, + ErlDrvTermData caller)); +static void fire_multi_timers(tcp_descriptor *desc, ErlDrvPort port, ErlDrvData data); -static void remove_multi_timer(MultiTimerData **first, ErlDrvPort port, MultiTimerData *p); +static void remove_multi_timer(tcp_descriptor *desc, ErlDrvPort port, MultiTimerData *p); +static void cancel_multi_timer(tcp_descriptor *desc, ErlDrvPort port, + void (*timeout_fun)(ErlDrvData drv_data, + ErlDrvTermData caller)); static void tcp_inet_multi_timeout(ErlDrvData e, ErlDrvTermData caller); -static void clean_multi_timers(MultiTimerData **first, ErlDrvPort port); +static void clean_multi_timers(tcp_descriptor *desc, ErlDrvPort port); typedef struct { int id; /* id used to identify reply */ @@ -1018,6 +1095,7 @@ typedef struct { inet_async_op* oph; /* queue head or NULL */ inet_async_op* opt; /* queue tail or NULL */ inet_async_op op_queue[INET_MAX_ASYNC]; /* call queue */ + int op_ref; /* queue reference generator */ int active; /* 0 = passive, 1 = active, 2 = active once */ Sint16 active_count; /* counter for {active,N} */ @@ -1060,13 +1138,12 @@ typedef struct { double send_avg; /* average packet size sent */ subs_list empty_out_q_subs; /* Empty out queue subscribers */ - int is_ignored; /* if a fd is ignored by the inet_drv. - This flag should be set to true when - the fd is used outside of inet_drv. */ + int flags; #ifdef HAVE_SETNS char *netns; /* Socket network namespace name as full file path */ #endif + int recv_cmsgflags; /* Which ancillary data to expect */ } inet_descriptor; @@ -1136,7 +1213,6 @@ static int packet_inet_init(void); static void packet_inet_stop(ErlDrvData); static void packet_inet_command(ErlDrvData, char*, ErlDrvSizeT); static void packet_inet_drv_input(ErlDrvData data, ErlDrvEvent event); -static void packet_inet_drv_output(ErlDrvData data, ErlDrvEvent event); static ErlDrvData udp_inet_start(ErlDrvPort, char* command); #ifdef HAVE_SCTP static ErlDrvData sctp_inet_start(ErlDrvPort, char* command); @@ -1161,7 +1237,7 @@ static struct erl_drv_entry udp_inet_driver_entry = NULL, #else packet_inet_drv_input, - packet_inet_drv_output, + NULL, #endif "udp_inet", NULL, @@ -1196,7 +1272,7 @@ static struct erl_drv_entry sctp_inet_driver_entry = NULL, #else packet_inet_drv_input, - packet_inet_drv_output, + NULL, #endif "sctp_inet", NULL, @@ -1219,7 +1295,7 @@ static struct erl_drv_entry sctp_inet_driver_entry = }; #endif -typedef struct { +struct _tcp_descriptor { inet_descriptor inet; /* common data structure (DON'T MOVE) */ int high; /* high watermark */ int low; /* low watermark */ @@ -1235,8 +1311,24 @@ typedef struct { int http_state; /* 0 = response|request 1=headers fields */ inet_async_multi_op *multi_first;/* NULL == no multi-accept-queue, op is in ordinary queue */ inet_async_multi_op *multi_last; - MultiTimerData *mtd; /* Timer structures for multiple accept */ -} tcp_descriptor; + MultiTimerData *mtd; /* Timer structures for multiple accept */ + MultiTimerData *mtd_cache; /* A cache for timer allocations */ +#ifdef HAVE_SENDFILE + struct { + ErlDrvSizeT ioq_skip; /* The number of bytes in the queue at the time + * sendfile was issued, which must be sent + * before issuing the sendfile call itself. */ + int dup_file_fd; /* The file handle to send from; this is + * duplicated when sendfile is issued to + * reduce (but not eliminate) the impact of a + * nasty race, so we have to remember to close + * it. */ + Uint64 bytes_sent; + Uint64 offset; + Uint64 length; + } sendfile; +#endif +}; /* send function */ static int tcp_send(tcp_descriptor* desc, char* ptr, ErlDrvSizeT len); @@ -1246,6 +1338,11 @@ static int tcp_deliver(tcp_descriptor* desc, int len); static int tcp_shutdown_error(tcp_descriptor* desc, int err); +#ifdef HAVE_SENDFILE +static int tcp_inet_sendfile(tcp_descriptor* desc); +static int tcp_sendfile_aborted(tcp_descriptor* desc, int socket_error); +#endif + static int tcp_inet_output(tcp_descriptor* desc, HANDLE event); static int tcp_inet_input(tcp_descriptor* desc, HANDLE event); @@ -1262,14 +1359,12 @@ typedef struct { static int packet_inet_input(udp_descriptor* udesc, HANDLE event); -static int packet_inet_output(udp_descriptor* udesc, HANDLE event); #endif /* convert descriptor pointer to inet_descriptor pointer */ #define INETP(d) (&(d)->inet) -static int async_ref = 0; /* async reference id generator */ -#define NEW_ASYNC_ID() ((async_ref++) & 0xffff) +#define NEW_ASYNC_ID(desc) ((desc)->op_ref++ & 0xffff) /* check for transition from active to passive */ #define INET_CHECK_ACTIVE_TO_PASSIVE(inet) \ @@ -1305,6 +1400,11 @@ static ErlDrvTermData am_udp_error; #ifdef HAVE_SYS_UN_H static ErlDrvTermData am_local; #endif +#ifndef __WIN32__ +static ErlDrvTermData am_tos; +static ErlDrvTermData am_tclass; +static ErlDrvTermData am_ttl; +#endif #ifdef HAVE_SCTP static ErlDrvTermData am_sctp; static ErlDrvTermData am_sctp_passive; @@ -1325,12 +1425,16 @@ static ErlDrvTermData am_sndbuf; static ErlDrvTermData am_reuseaddr; static ErlDrvTermData am_dontroute; static ErlDrvTermData am_priority; -static ErlDrvTermData am_tos; -static ErlDrvTermData am_tclass; +static ErlDrvTermData am_recvtos; +static ErlDrvTermData am_recvtclass; +static ErlDrvTermData am_recvttl; static ErlDrvTermData am_ipv6_v6only; static ErlDrvTermData am_netns; static ErlDrvTermData am_bind_to_device; #endif +#ifdef HAVE_SENDFILE +static ErlDrvTermData am_sendfile; +#endif static char str_eafnosupport[] = "eafnosupport"; static char str_einval[] = "einval"; @@ -1514,8 +1618,10 @@ static void *realloc_wrapper(void *current, ErlDrvSizeT size){ # define ASSOC_ID_LEN 4 # define LOAD_ASSOC_ID LOAD_UINT # define LOAD_ASSOC_ID_CNT LOAD_UINT_CNT -# define SCTP_ANC_BUFF_SIZE INET_DEF_BUFFER/2 /* XXX: not very good... */ +#else +# define IS_SCTP(desc) 0 #endif +# define ANC_BUFF_SIZE INET_DEF_BUFFER/2 /* XXX: not very good... */ #ifdef HAVE_UDP static int load_address(ErlDrvTermData* spec, int i, char* buf) @@ -1751,6 +1857,7 @@ static void release_buffer(ErlDrvBinary* buf) #ifdef HAVE_UDP static ErlDrvBinary* realloc_buffer(ErlDrvBinary* buf, ErlDrvSizeT newsz) { + DEBUGF(("realloc_buffer: %ld -> %ld\r\n", (buf==NULL) ? 0 : buf->orig_size, newsz)); return driver_realloc_binary(buf, newsz); } #endif @@ -1921,7 +2028,7 @@ static void enq_multi_op(tcp_descriptor *desc, char *buf, int req, ErlDrvTermData caller, MultiTimerData *timeout, ErlDrvMonitor *monitorp) { - int id = NEW_ASYNC_ID(); + int id = NEW_ASYNC_ID(INETP(desc)); enq_old_multi_op(desc,id,req,caller,timeout,monitorp); if (buf != NULL) put_int16(id, buf); @@ -1990,7 +2097,7 @@ static int remove_multi_op(tcp_descriptor *desc, int *id_p, int *req_p, static int enq_async_w_tmo(inet_descriptor* desc, char* buf, int req, unsigned timeout, ErlDrvMonitor *monitorp) { - int id = NEW_ASYNC_ID(); + int id = NEW_ASYNC_ID(desc); inet_async_op* opp; if ((opp = desc->oph) == NULL) /* queue empty */ @@ -2715,6 +2822,66 @@ static int inet_async_data(inet_descriptor* desc, const char* buf, int len) } } +#ifndef __WIN32__ +static int load_cmsg_int(ErlDrvTermData *spec, int i, + struct cmsghdr *cmsg) { + union u { + byte uint8; + Uint16 uint16; + Uint32 uint32; + Uint64 uint64; + } *p; + p = (union u*) CMSG_DATA(cmsg); + switch (LEN_CMSG_DATA(cmsg) * CHAR_BIT) { + case 8: + return LOAD_INT(spec, i, p->uint8); + case 16: + return LOAD_INT(spec, i, p->uint16); + + case 32: + return LOAD_INT(spec, i, p->uint32); + + case 64: + return LOAD_INT(spec, i, p->uint64); + } + return LOAD_INT(spec, i, 0); +} + +static int parse_ancillary_data_item(ErlDrvTermData *spec, int i, + struct cmsghdr *cmsg, int *n) { +#define LOAD_CMSG_INT(proto, type, am) \ + if (cmsg->cmsg_level == (proto) && \ + cmsg->cmsg_type == (type)) { \ + i = LOAD_ATOM(spec, i, (am)); \ + i = load_cmsg_int(spec, i, cmsg); \ + i = LOAD_TUPLE(spec, i, 2); \ + (*n)++; \ + return i; \ + } +#if defined(IPPROTO_IP) && defined(IP_TOS) + LOAD_CMSG_INT(IPPROTO_IP, IP_TOS, am_tos); +#endif +#if defined(IPPROTO_IPV6) && defined(IPV6_TCLASS) + LOAD_CMSG_INT(IPPROTO_IPV6, IPV6_TCLASS, am_tclass); +#endif +#if defined(IPPROTO_IP) && defined(IP_TTL) + LOAD_CMSG_INT(IPPROTO_IP, IP_TTL, am_ttl); +#endif + /* BSD uses the RECV* names in CMSG fields */ +#if defined(IPPROTO_IP) && defined(IP_RECVTOS) + LOAD_CMSG_INT(IPPROTO_IP, IP_RECVTOS, am_tos); +#endif +#if defined(IPPROTO_IPV6) && defined(IPV6_RECVTCLASS) + LOAD_CMSG_INT(IPPROTO_IPV6, IPV6_RECVTCLASS, am_tclass); +#endif +#if defined(IPPROTO_IP) && defined(IP_RECVTTL) + LOAD_CMSG_INT(IPPROTO_IP, IP_RECVTTL, am_ttl); +#endif +#undef LOAD_CMSG_INT + return i; +} +#endif /* #ifndef __WIN32__ */ + #ifdef HAVE_SCTP /* ** SCTP-related atoms: @@ -2846,11 +3013,18 @@ static int sctp_parse_ancillary_data for (cmsg = frst_msg; cmsg != NULL; cmsg = CMSG_NXTHDR(mptr,cmsg)) { struct sctp_sndrcvinfo * sri; - +#ifndef __WIN32 + int old_s; + + /* Parse ancillary data common to UDP */ + old_s = s; + i = parse_ancillary_data_item(spec, i, cmsg, &s); + if (s > old_s) continue; /* Skip other possible ancillary data, e.g. from IPv6: */ if (cmsg->cmsg_level != IPPROTO_SCTP || cmsg->cmsg_type != SCTP_SNDRCV) continue; +#endif if (((char*)cmsg + cmsg->cmsg_len) - (char*)frst_msg > mptr->msg_controllen) @@ -3190,6 +3364,23 @@ static int sctp_parse_async_event } #endif /* HAVE_SCTP */ +#ifndef __WIN32__ +static int udp_parse_ancillary_data(ErlDrvTermData *spec, int i, + struct msghdr *mptr) { + struct cmsghdr *cmsg; + int n; + + n = 0; + for (cmsg = CMSG_FIRSTHDR(mptr); + cmsg != NULL; + cmsg = CMSG_NXTHDR(mptr, cmsg)) { + i = parse_ancillary_data_item(spec, i, cmsg, &n); + } + i = LOAD_NIL(spec, i); + return LOAD_LIST(spec, i, n+1); +} +#endif /* ifndef __WIN32__ */ + /* ** passive mode reply: ** for UDP: @@ -3212,7 +3403,7 @@ static int sctp_parse_async_event static int inet_async_binary_data (inet_descriptor* desc, unsigned int phsz, - ErlDrvBinary * bin, int offs, int len, void * extra) + ErlDrvBinary * bin, int offs, int len, void *mp) { unsigned int hsz = desc->hsz + phsz; ErlDrvTermData spec [PACKET_ERL_DRV_TERM_DATA_LEN]; @@ -3245,9 +3436,10 @@ inet_async_binary_data if (IS_SCTP(desc)) { /* For SCTP we always have desc->hsz==0 (i.e., no application-level headers are used), so hsz==phsz (see above): */ - struct msghdr* mptr; int sz; - + struct msghdr *mptr; + + mptr = mp; ASSERT (hsz == phsz && hsz != 0); sz = len - hsz; /* Size of the msg data proper, w/o the addr */ @@ -3255,7 +3447,6 @@ inet_async_binary_data i = LOAD_STRING(spec, i, bin->orig_bytes+offs, hsz); /* Put in the list (possibly empty) of Ancillary Data: */ - mptr = (struct msghdr *) extra; i = sctp_parse_ancillary_data (spec, i, mptr); /* Then: Data or Event (Notification)? */ @@ -3284,20 +3475,32 @@ inet_async_binary_data } else #endif /* HAVE_SCTP */ - /* Generic case. Both Addr and Data (or a single list of them together) are - returned: */ + { + /* Generic case. Both Addr and Data + * (or a single list of them together) are returned: */ - if ((desc->mode == INET_MODE_LIST) || (hsz > len)) { - /* INET_MODE_LIST => [H1,H2,...Hn] */ - i = LOAD_STRING(spec, i, bin->orig_bytes+offs, len); - } - else { - /* INET_MODE_BINARY => [H1,H2,...HSz | Binary] or [Binary]: */ - int sz = len - hsz; - i = LOAD_BINARY(spec, i, bin, offs+hsz, sz); - if (hsz > 0) - i = LOAD_STRING_CONS(spec, i, bin->orig_bytes+offs, hsz); + if ((desc->mode == INET_MODE_LIST) || (hsz > len)) { + /* INET_MODE_LIST => [H1,H2,...Hn] */ + i = LOAD_STRING(spec, i, bin->orig_bytes+offs, len); + } + else { + /* INET_MODE_BINARY => [H1,H2,...HSz | Binary] or [Binary]: */ + int sz = len - hsz; + i = LOAD_BINARY(spec, i, bin, offs+hsz, sz); + if (hsz > 0) + i = LOAD_STRING_CONS(spec, i, bin->orig_bytes+offs, hsz); + } + +#ifndef __WIN32__ + if (mp) { + /* We got ancillary data from an UDP recvmsg. + * Insert an additional tuple level {[F|AddrData],AncData} */ + i = udp_parse_ancillary_data(spec, i, (struct msghdr*)mp); + i = LOAD_TUPLE(spec, i, 2); + } +#endif } + /* Close up the {ok, ...} or {error, ...} tuple: */ i = LOAD_TUPLE(spec, i, 2); @@ -3429,8 +3632,9 @@ static int tcp_error_message(tcp_descriptor* desc, int err) ** [AddrLen, H2,...,HSz] are msg headers for UDP AF_UNIX only ** Data : List() | Binary() */ -static int packet_binary_message - (inet_descriptor* desc, ErlDrvBinary* bin, int offs, int len, void* extra) +static int packet_binary_message(inet_descriptor* desc, + ErlDrvBinary* bin, int offs, int len, + void *mp) { unsigned int hsz = desc->hsz; ErlDrvTermData spec [PACKET_ERL_DRV_TERM_DATA_LEN]; @@ -3455,8 +3659,14 @@ static int packet_binary_message # ifdef HAVE_SCTP if (!IS_SCTP(desc)) - { # endif + { +#ifndef __WIN32__ + if (mp) i = udp_parse_ancillary_data(spec, i, (struct msghdr*)mp); +#endif + /* We got ancillary data from an UDP recvmsg. + * Insert an additional tuple level {AncData,[F|AddrData]} + */ if ((desc->mode == INET_MODE_LIST) || (hsz > len)) /* INET_MODE_LIST, or only headers => [H1,H2,...Hn] */ i = LOAD_STRING(spec, i, bin->orig_bytes+offs, len); @@ -3468,16 +3678,24 @@ static int packet_binary_message if (hsz > 0) i = LOAD_STRING_CONS(spec, i, bin->orig_bytes+offs, hsz); } -# ifdef HAVE_SCTP + /* Close up the outer 5-or-6-tuple */ +#ifndef __WIN32__ + if (mp) i = LOAD_TUPLE(spec, i, 6); + else +#endif + i = LOAD_TUPLE(spec, i, 5); } +# ifdef HAVE_SCTP else - { /* For SCTP we always have desc->hsz==0 (i.e., no application-level + { + struct msghdr *mptr; + + mptr = mp; + /* For SCTP we always have desc->hsz==0 (i.e., no application-level headers are used): */ - struct msghdr* mptr; ASSERT(hsz == 0); /* Put in the list (possibly empty) of Ancillary Data: */ - mptr = (struct msghdr *) extra; i = sctp_parse_ancillary_data (spec, i, mptr); /* Then: Data or Event (Notification)? */ @@ -3503,11 +3721,11 @@ static int packet_binary_message /* Close up the {[AncilData], Event_OR_Data} tuple: */ i = LOAD_TUPLE (spec, i, 2); + /* Close up the outer 5-tuple: */ + i = LOAD_TUPLE(spec, i, 5); } # endif /* HAVE_SCTP */ - /* Close up the outer 5-tuple: */ - i = LOAD_TUPLE(spec, i, 5); ASSERT(i <= PACKET_ERL_DRV_TERM_DATA_LEN); return erl_drv_output_term(desc->dport, spec, i); } @@ -3641,19 +3859,19 @@ tcp_reply_binary_data(tcp_descriptor* desc, ErlDrvBinary* bin, int offs, int len static int packet_reply_binary_data(inet_descriptor* desc, unsigned int hsz, ErlDrvBinary * bin, int offs, int len, - void * extra) + void *mp) { int code; if (desc->active == INET_PASSIVE) /* "inet" is actually for both UDP and SCTP, as well as TCP! */ - return inet_async_binary_data(desc, hsz, bin, offs, len, extra); + return inet_async_binary_data(desc, hsz, bin, offs, len, mp); else { /* INET_ACTIVE or INET_ONCE: */ if (desc->deliver == INET_DELIVER_PORT) code = inet_port_binary_data(desc, bin, offs, len); else - code = packet_binary_message(desc, bin, offs, len, extra); + code = packet_binary_message(desc, bin, offs, len, mp); if (code < 0) return code; INET_CHECK_ACTIVE_TO_PASSIVE(desc); @@ -3720,8 +3938,9 @@ static void inet_init_sctp(void) { INIT_ATOM(reuseaddr); INIT_ATOM(dontroute); INIT_ATOM(priority); - INIT_ATOM(tos); - INIT_ATOM(tclass); + INIT_ATOM(recvtos); + INIT_ATOM(recvtclass); + INIT_ATOM(recvttl); INIT_ATOM(ipv6_v6only); INIT_ATOM(netns); INIT_ATOM(bind_to_device); @@ -3864,6 +4083,11 @@ static int inet_init() #endif INIT_ATOM(empty_out_q); INIT_ATOM(ssl_tls); +#ifndef __WIN32__ + INIT_ATOM(tos); + INIT_ATOM(tclass); + INIT_ATOM(ttl); +#endif INIT_ATOM(http_eoh); INIT_ATOM(http_header); @@ -3877,6 +4101,10 @@ static int inet_init() INIT_ATOM(https); INIT_ATOM(scheme); +#ifdef HAVE_SENDFILE + INIT_ATOM(sendfile); +#endif + /* add TCP, UDP and SCTP drivers */ add_driver_entry(&tcp_inet_driver_entry); #ifdef HAVE_UDP @@ -4331,7 +4559,7 @@ static void desc_close(inet_descriptor* desc) * We should close the fd here, but the other driver might still * be selecting on it. */ - if (!desc->is_ignored) + if (!INET_IGNORED(desc)) driver_select(desc->port,(ErlDrvEvent)(long)desc->event, ERL_DRV_USE, 0); else @@ -4978,6 +5206,71 @@ static int hwaddr_libdlpi_lookup(const char *ifnm, } #endif +#ifdef HAVE_GETIFADDRS +/* Returns 0 for success and errno() for failure */ +static int call_getifaddrs(inet_descriptor* desc_p, struct ifaddrs **ifa_pp) +{ + int result, save_errno; +#ifdef HAVE_SETNS + int current_ns; + + current_ns = 0; + if (desc_p->netns != NULL) { + int new_ns; + /* Temporarily change network namespace for this thread + * over the getifaddrs() call + */ + current_ns = open("/proc/self/ns/net", O_RDONLY); + if (current_ns == INVALID_SOCKET) + return sock_errno(); + new_ns = open(desc_p->netns, O_RDONLY); + if (new_ns == INVALID_SOCKET) { + save_errno = sock_errno(); + while (close(current_ns) == INVALID_SOCKET && + sock_errno() == EINTR); + return save_errno; + } + if (setns(new_ns, CLONE_NEWNET) != 0) { + save_errno = sock_errno(); + while (close(new_ns) == INVALID_SOCKET && + sock_errno() == EINTR); + while (close(current_ns) == INVALID_SOCKET && + sock_errno() == EINTR); + return save_errno; + } + else { + while (close(new_ns) == INVALID_SOCKET && + sock_errno() == EINTR); + } + } +#endif + save_errno = 0; + result = getifaddrs(ifa_pp); + if (result < 0) + save_errno = sock_errno(); +#ifdef HAVE_SETNS + if (desc_p->netns != NULL) { + /* Restore network namespace */ + if (setns(current_ns, CLONE_NEWNET) != 0) { + /* XXX Failed to restore network namespace. + * What to do? Tidy up and return an error... + * Note that the thread now might still be in the set namespace. + * Can this even happen? Should the emulator be aborted? + */ + if (result >= 0) { + /* We got a result but have to waste it */ + save_errno = sock_errno(); + freeifaddrs(*ifa_pp); + } + } + while (close(current_ns) == INVALID_SOCKET && + sock_errno() == EINTR); + } +#endif + return save_errno; +} +#endif /* #ifdef HAVE_GETIFADDRS */ + /* FIXME: temporary hack */ #ifndef IFHWADDRLEN #define IFHWADDRLEN 6 @@ -5055,8 +5348,8 @@ static ErlDrvSSizeT inet_ctl_ifget(inet_descriptor* desc, struct sockaddr_dl *sdlp; int found = 0; - if (getifaddrs(&ifa) == -1) - goto error; + if (call_getifaddrs(desc, &ifa) != 0) + goto error; for (ifp = ifa; ifp; ifp = ifp->ifa_next) { if ((ifp->ifa_addr->sa_family == AF_LINK) && @@ -5078,8 +5371,8 @@ static ErlDrvSSizeT inet_ctl_ifget(inet_descriptor* desc, sys_memcpy(sptr, sdlp->sdl_data + sdlp->sdl_nlen, sdlp->sdl_alen); - freeifaddrs(ifa); sptr += sdlp->sdl_alen; + freeifaddrs(ifa); #endif break; } @@ -5774,6 +6067,7 @@ static ErlDrvSSizeT inet_ctl_getifaddrs(inet_descriptor* desc_p, ErlDrvSizeT buf_size; char *buf_p; char *buf_alloc_p; + int save_errno; buf_size = GETIFADDRS_BUFSZ; buf_alloc_p = ALLOC(GETIFADDRS_BUFSZ); @@ -5808,9 +6102,9 @@ static ErlDrvSSizeT inet_ctl_getifaddrs(inet_descriptor* desc_p, } \ } while (0) - if (getifaddrs(&ifa_p) < 0) { - return ctl_error(sock_errno(), rbuf_pp, rsize); - } + if ((save_errno = call_getifaddrs(desc_p, &ifa_p)) != 0) + return ctl_error(save_errno, rbuf_pp, rsize); + ifa_free_p = ifa_p; *buf_p++ = INET_REP_OK; for (; ifa_p; ifa_p = ifa_p->ifa_next) { @@ -5892,7 +6186,8 @@ static ErlDrvSSizeT inet_ctl_getifaddrs(inet_descriptor* desc_p, but ditto for the other worked and that was actually the requested option, failure was still reported to erlang. */ -#if defined(IP_TOS) && defined(SOL_IP) && defined(SO_PRIORITY) +#if defined(IP_TOS) && defined(IPPROTO_IP) \ + && defined(SO_PRIORITY) && !defined(__WIN32__) static int setopt_prio_tos_trick (int fd, int proto, int type, char* arg_ptr, int arg_sz, int propagate) { @@ -5914,14 +6209,14 @@ static int setopt_prio_tos_trick res_prio = sock_getopt(fd, SOL_SOCKET, SO_PRIORITY, (char *) &tmp_ival_prio, &tmp_arg_sz_prio); - res_tos = sock_getopt(fd, SOL_IP, IP_TOS, + res_tos = sock_getopt(fd, IPPROTO_IP, IP_TOS, (char *) &tmp_ival_tos, &tmp_arg_sz_tos); res = sock_setopt(fd, proto, type, arg_ptr, arg_sz); if (res == 0) { if (type != SO_PRIORITY) { if (type != IP_TOS && res_tos == 0) { res_tos = sock_setopt(fd, - SOL_IP, + IPPROTO_IP, IP_TOS, (char *) &tmp_ival_tos, tmp_arg_sz_tos); @@ -5965,7 +6260,7 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) int proto; int opt; struct linger li_val; -#ifdef HAVE_MULTICAST_SUPPORT +#if defined(HAVE_MULTICAST_SUPPORT) && defined(IPPROTO_IP) struct ip_mreq mreq_val; #endif int ival; @@ -5988,6 +6283,8 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) /* XXX { int i; for(i=0;i<len;++i) fprintf(stderr,"0x%02X, ", (unsigned) ptr[i]); fprintf(stderr,"\r\n");} */ while(len >= 5) { + int recv_cmsgflags; + opt = *ptr++; ival = get_int32(ptr); ptr += 4; @@ -5996,6 +6293,7 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) arg_sz = sizeof(ival); proto = SOL_SOCKET; propagate = 0; + recv_cmsgflags = desc->recv_cmsgflags; switch(opt) { case INET_LOPT_HEADER: @@ -6022,6 +6320,7 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) (long)desc->port, desc->s, ival)); if (ival < INET_MIN_BUFFER) ival = INET_MIN_BUFFER; desc->bufsz = ival; + desc->flags |= INET_FLG_BUFFER_SET; continue; case INET_LOPT_ACTIVE: @@ -6217,6 +6516,16 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) case INET_OPT_RCVBUF: type = SO_RCVBUF; DEBUGF(("inet_set_opts(%ld): s=%d, SO_RCVBUF=%d\r\n", (long)desc->port, desc->s, ival)); + if (!(desc->flags & INET_FLG_BUFFER_SET)) { + /* make sure we have desc->bufsz >= SO_RCVBUF */ + if (ival > (1 << 16) && desc->stype == SOCK_DGRAM && !IS_SCTP(desc)) + /* For UDP we don't want to automatically + set the buffer size to be larger than + the theoretical max MTU */ + desc->bufsz = 1 << 16; + else if (ival > desc->bufsz) + desc->bufsz = ival; + } break; case INET_OPT_LINGER: type = SO_LINGER; if (len < 4) @@ -6246,28 +6555,80 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) (long)desc->port, desc->s, ival)); break; #else + /* inet_fill_opts always returns a value for this option, + * so we need to ignore it if not implemented */ continue; #endif case INET_OPT_TOS: -#if defined(IP_TOS) && defined(SOL_IP) - proto = SOL_IP; +#if defined(IP_TOS) && defined(IPPROTO_IP) + proto = IPPROTO_IP; type = IP_TOS; propagate = 1; DEBUGF(("inet_set_opts(%ld): s=%d, IP_TOS=%d\r\n", (long)desc->port, desc->s, ival)); break; #else + /* inet_fill_opts always returns a value for this option, + * so we need to ignore it if not implemented. */ continue; #endif -#if defined(IPV6_TCLASS) && defined(SOL_IPV6) +#if defined(IPV6_TCLASS) && defined(IPPROTO_IPV6) case INET_OPT_TCLASS: - proto = SOL_IPV6; + proto = IPPROTO_IPV6; type = IPV6_TCLASS; propagate = 1; DEBUGF(("inet_set_opts(%ld): s=%d, IPV6_TCLASS=%d\r\n", (long)desc->port, desc->s, ival)); break; #endif +#if defined(IP_TTL) && defined(IPPROTO_IP) + case INET_OPT_TTL: + proto = IPPROTO_IP; + type = IP_TTL; + propagate = 1; + DEBUGF(("inet_set_opts(%ld): s=%d, IP_TTL=%d\r\n", + (long)desc->port, desc->s, ival)); + break; +#endif +#if defined(IP_RECVTOS) && defined(IPPROTO_IP) + case INET_OPT_RECVTOS: + proto = IPPROTO_IP; + type = IP_RECVTOS; + propagate = 1; + recv_cmsgflags = + ival ? + (desc->recv_cmsgflags | INET_CMSG_RECVTOS) : + (desc->recv_cmsgflags & ~INET_CMSG_RECVTOS); + DEBUGF(("inet_set_opts(%ld): s=%d, IP_RECVTOS=%d\r\n", + (long)desc->port, desc->s, ival)); + break; +#endif +#if defined(IPV6_RECVTCLASS) && defined(IPPROTO_IPV6) + case INET_OPT_RECVTCLASS: + proto = IPPROTO_IPV6; + type = IPV6_RECVTCLASS; + propagate = 1; + recv_cmsgflags = + ival ? + (desc->recv_cmsgflags | INET_CMSG_RECVTCLASS) : + (desc->recv_cmsgflags & ~INET_CMSG_RECVTCLASS); + DEBUGF(("inet_set_opts(%ld): s=%d, IPV6_RECVTCLASS=%d\r\n", + (long)desc->port, desc->s, ival)); + break; +#endif +#if defined(IP_RECVTTL) && defined(IPPROTO_IP) + case INET_OPT_RECVTTL: + proto = IPPROTO_IP; + type = IP_RECVTTL; + propagate = 1; + recv_cmsgflags = + ival ? + (desc->recv_cmsgflags | INET_CMSG_RECVTTL) : + (desc->recv_cmsgflags & ~INET_CMSG_RECVTTL); + DEBUGF(("inet_set_opts(%ld): s=%d, IP_RECVTTL=%d\r\n", + (long)desc->port, desc->s, ival)); + break; +#endif case TCP_OPT_NODELAY: proto = IPPROTO_TCP; @@ -6276,7 +6637,20 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) (long)desc->port, desc->s, ival)); break; -#ifdef HAVE_MULTICAST_SUPPORT + case TCP_OPT_NOPUSH: +#if defined(INET_TCP_NOPUSH) + proto = IPPROTO_TCP; + type = INET_TCP_NOPUSH; + DEBUGF(("inet_set_opts(%ld): s=%d, t=%d TCP_NOPUSH=%d\r\n", + (long)desc->port, desc->s, type, ival)); + break; +#else + /* inet_fill_opts always returns a value for this option, + * so we need to ignore it if not implemented, just in case */ + continue; +#endif + +#if defined(HAVE_MULTICAST_SUPPORT) && defined(IPPROTO_IP) case UDP_OPT_MULTICAST_TTL: proto = IPPROTO_IP; @@ -6322,10 +6696,10 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) arg_sz = sizeof(mreq_val); break; -#endif /* HAVE_MULTICAST_SUPPORT */ +#endif /* defined(HAVE_MULTICAST_SUPPORT) && defined(IPPROTO_IP) */ case INET_OPT_IPV6_V6ONLY: -#if HAVE_DECL_IPV6_V6ONLY +#if HAVE_DECL_IPV6_V6ONLY && defined(IPPROTO_IPV6) proto = IPPROTO_IPV6; type = IPV6_V6ONLY; propagate = 1; @@ -6385,28 +6759,29 @@ static int inet_set_opts(inet_descriptor* desc, char* ptr, int len) default: return -1; } -#if defined(IP_TOS) && defined(SOL_IP) && defined(SO_PRIORITY) +#if defined(IP_TOS) && defined(IPPROTO_IP) \ + && defined(SO_PRIORITY) && !defined(__WIN32__) res = setopt_prio_tos_trick (desc->s, proto, type, arg_ptr, arg_sz, propagate); #else res = sock_setopt (desc->s, proto, type, arg_ptr, arg_sz); #endif + if (res == 0) desc->recv_cmsgflags = recv_cmsgflags; if (propagate && res != 0) { return -1; } DEBUGF(("inet_set_opts(%ld): s=%d returned %d\r\n", (long)desc->port, desc->s, res)); - if (type == SO_RCVBUF) { - /* make sure we have desc->bufsz >= SO_RCVBUF */ - if (ival > desc->bufsz) - desc->bufsz = ival; - } } if ( ((desc->stype == SOCK_STREAM) && IS_CONNECTED(desc)) || ((desc->stype == SOCK_DGRAM) && IS_OPEN(desc))) { - if (desc->active != old_active) + if (desc->active != old_active) { + /* Need to cancel the read_packet timer if we go from active to passive. */ + if (desc->active == INET_PASSIVE && desc->stype == SOCK_DGRAM) + driver_cancel_timer(desc->port); sock_select(desc, (FD_READ|FD_CLOSE), (desc->active>0)); + } /* XXX: UDP sockets could also trigger immediate read here NIY */ if ((desc->stype==SOCK_STREAM) && desc->active) { @@ -6526,10 +6901,14 @@ static int sctp_set_opts(inet_descriptor* desc, char* ptr, int len) while (curr < ptr + len) { + int recv_cmsgflags; /* Get the Erlang-encoded option type -- always 1 byte: */ - int eopt = *curr; + int eopt; + + eopt = *curr; curr++; + recv_cmsgflags = desc->recv_cmsgflags; /* Get the option value. XXX: The condition (curr < ptr + len) does not preclude us from reading from beyond the buffer end, if the Erlang part of the driver specifies its input wrongly! @@ -6544,6 +6923,7 @@ static int sctp_set_opts(inet_descriptor* desc, char* ptr, int len) if (desc->bufsz < INET_MIN_BUFFER) desc->bufsz = INET_MIN_BUFFER; + desc->flags |= INET_FLG_BUFFER_SET; res = 0; /* This does not affect the kernel buffer size */ continue; @@ -6665,6 +7045,7 @@ static int sctp_set_opts(inet_descriptor* desc, char* ptr, int len) smaller than the kernel one: */ if (desc->bufsz <= arg.ival) desc->bufsz = arg.ival; + desc->flags |= INET_FLG_BUFFER_SET; break; } case INET_OPT_SNDBUF: @@ -6675,10 +7056,6 @@ static int sctp_set_opts(inet_descriptor* desc, char* ptr, int len) arg_ptr = (char*) (&arg.ival); arg_sz = sizeof ( arg.ival); - /* Adjust the size of the user-level recv buffer, so it's not - smaller than the kernel one: */ - if (desc->bufsz <= arg.ival) - desc->bufsz = arg.ival; break; } case INET_OPT_REUSEADDR: @@ -6710,28 +7087,32 @@ static int sctp_set_opts(inet_descriptor* desc, char* ptr, int len) break; } # else - continue; /* Option not supported -- ignore it */ + /* inet_fill_opts always returns a value for this option, + * so we need to ignore it if not implemented, just in case */ + continue; # endif case INET_OPT_TOS: -# if defined(IP_TOS) && defined(SOL_IP) +# if defined(IP_TOS) && defined(IPPROTO_IP) { arg.ival= get_int32 (curr); curr += 4; - proto = SOL_IP; + proto = IPPROTO_IP; type = IP_TOS; arg_ptr = (char*) (&arg.ival); arg_sz = sizeof ( arg.ival); break; } # else - continue; /* Option not supported -- ignore it */ + /* inet_fill_opts always returns a value for this option, + * so we need to ignore it if not implemented, just in case */ + continue; # endif -# if defined(IPV6_TCLASS) && defined(SOL_IPV6) +# if defined(IPV6_TCLASS) && defined(IPPROTO_IPV6) case INET_OPT_TCLASS: { arg.ival= get_int32 (curr); curr += 4; - proto = SOL_IPV6; + proto = IPPROTO_IPV6; type = IPV6_TCLASS; arg_ptr = (char*) (&arg.ival); arg_sz = sizeof ( arg.ival); @@ -6739,9 +7120,69 @@ static int sctp_set_opts(inet_descriptor* desc, char* ptr, int len) } # endif +# if defined(IP_TTL) && defined(IPPROTO_IP) + case INET_OPT_TTL: + { + arg.ival= get_int32 (curr); curr += 4; + proto = IPPROTO_IP; + type = IP_TTL; + arg_ptr = (char*) (&arg.ival); + arg_sz = sizeof ( arg.ival); + break; + } +# endif + +# if defined(IP_RECVTOS) && defined(IPPROTO_IP) + case INET_OPT_RECVTOS: + { + arg.ival= get_int32 (curr); curr += 4; + proto = IPPROTO_IP; + type = IP_RECVTOS; + arg_ptr = (char*) (&arg.ival); + arg_sz = sizeof ( arg.ival); + recv_cmsgflags = + arg.ival ? + (desc->recv_cmsgflags | INET_CMSG_RECVTOS) : + (desc->recv_cmsgflags & ~INET_CMSG_RECVTOS); + break; + } +# endif + +# if defined(IPV6_RECVTCLASS) && defined(IPPROTO_IPV6) + case INET_OPT_RECVTCLASS: + { + arg.ival= get_int32 (curr); curr += 4; + proto = IPPROTO_IPV6; + type = IPV6_RECVTCLASS; + arg_ptr = (char*) (&arg.ival); + arg_sz = sizeof ( arg.ival); + recv_cmsgflags = + arg.ival ? + (desc->recv_cmsgflags | INET_CMSG_RECVTCLASS) : + (desc->recv_cmsgflags & ~INET_CMSG_RECVTCLASS); + break; + } +# endif + +# if defined(IP_RECVTTL) && defined(IPPROTO_IP) + case INET_OPT_RECVTTL: + { + arg.ival= get_int32 (curr); curr += 4; + proto = IPPROTO_IP; + type = IP_RECVTTL; + arg_ptr = (char*) (&arg.ival); + arg_sz = sizeof ( arg.ival); + recv_cmsgflags = + arg.ival ? + (desc->recv_cmsgflags | INET_CMSG_RECVTTL) : + (desc->recv_cmsgflags & ~INET_CMSG_RECVTTL); + break; + } +# endif + case INET_OPT_IPV6_V6ONLY: -# if HAVE_DECL_IPV6_V6ONLY +# if HAVE_DECL_IPV6_V6ONLY && defined(IPPROTO_IPV6) { arg.ival= get_int32 (curr); curr += 4; proto = IPPROTO_IPV6; @@ -6991,13 +7432,15 @@ static int sctp_set_opts(inet_descriptor* desc, char* ptr, int len) */ return -1; } -#if defined(IP_TOS) && defined(SOL_IP) && defined(SO_PRIORITY) +#if defined(IP_TOS) && defined(IPPROTO_IP) \ + && defined(SO_PRIORITY) && !defined(__WIN32__) res = setopt_prio_tos_trick (desc->s, proto, type, arg_ptr, arg_sz, 1); #else res = sock_setopt (desc->s, proto, type, arg_ptr, arg_sz); #endif /* The return values of "sock_setopt" can only be 0 or -1: */ ASSERT(res == 0 || res == -1); + if (res == 0) desc->recv_cmsgflags = recv_cmsgflags; if (res == -1) { /* Got an error, DO NOT continue with other options. However, on Solaris 10, we DO allow SO_SNDBUF and SO_RCVBUF to fail, assu- @@ -7018,6 +7461,35 @@ static int sctp_set_opts(inet_descriptor* desc, char* ptr, int len) } #endif /* HAVE_SCTP */ +#ifndef __WIN32__ +static void put_cmsg_int32(struct cmsghdr *cmsg, char *ptr) { + union u { + byte uint8; + Uint16 uint16; + Uint32 uint32; + Uint64 uint64; + } *p; + p = (union u*) CMSG_DATA(cmsg); + switch (LEN_CMSG_DATA(cmsg) * CHAR_BIT) { + case 8: + put_int32((Uint32) p->uint8, ptr); + break; + case 16: + put_int32((Uint32) p->uint16, ptr); + break; + case 32: + put_int32(p->uint32, ptr); + break; + case 64: + put_int32((Uint32) p->uint64, ptr); + break; + default: + put_int32(0, ptr); + } + return; +} +#endif + /* load all option values into the buf and reply ** return total length of reply filled into ptr ** ptr should point to a buffer with 9*len +1 to be safe!! @@ -7252,8 +7724,8 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc, continue; #endif case INET_OPT_TOS: -#if defined(IP_TOS) && defined(SOL_IP) - proto = SOL_IP; +#if defined(IP_TOS) && defined(IPPROTO_IP) + proto = IPPROTO_IP; type = IP_TOS; break; #else @@ -7262,14 +7734,50 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc, continue; #endif case INET_OPT_TCLASS: -#if defined(IPV6_TCLASS) && defined(SOL_IPV6) - proto = SOL_IPV6; +#if defined(IPV6_TCLASS) && defined(IPPROTO_IPV6) + proto = IPPROTO_IPV6; type = IPV6_TCLASS; break; #else TRUNCATE_TO(0,ptr); continue; #endif + case INET_OPT_TTL: +#if defined(IP_TTL) && defined(IPPROTO_IP) + proto = IPPROTO_IP; + type = IP_TTL; + break; +#else + TRUNCATE_TO(0,ptr); + continue; +#endif + case INET_OPT_RECVTOS: +#if defined(IP_RECVTOS) && defined(IPPROTO_IP) + proto = IPPROTO_IP; + type = IP_RECVTOS; + break; +#else + TRUNCATE_TO(0,ptr); + continue; +#endif + case INET_OPT_RECVTCLASS: +#if defined(IPV6_RECVTCLASS) && defined(IPPROTO_IPV6) + proto = IPPROTO_IPV6; + type = IPV6_RECVTCLASS; + break; +#else + TRUNCATE_TO(0,ptr); + continue; +#endif + case INET_OPT_RECVTTL: +#if defined(IP_RECVTTL) && defined(IPPROTO_IP) + proto = IPPROTO_IP; + type = IP_RECVTTL; + break; +#else + TRUNCATE_TO(0,ptr); + continue; +#endif case INET_OPT_REUSEADDR: type = SO_REUSEADDR; break; @@ -7295,8 +7803,18 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc, proto = IPPROTO_TCP; type = TCP_NODELAY; break; + case TCP_OPT_NOPUSH: +#if defined(INET_TCP_NOPUSH) + proto = IPPROTO_TCP; + type = INET_TCP_NOPUSH; + break; +#else + *ptr++ = opt; + put_int32(0, ptr); + continue; +#endif -#ifdef HAVE_MULTICAST_SUPPORT +#if defined(HAVE_MULTICAST_SUPPORT) && defined(IPPROTO_IP) case UDP_OPT_MULTICAST_TTL: proto = IPPROTO_IP; type = IP_MULTICAST_TTL; @@ -7315,10 +7833,10 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc, arg_ptr = (char*) &li_val; type = SO_LINGER; break; -#endif /* HAVE_MULTICAST_SUPPORT */ +#endif /* defined(HAVE_MULTICAST_SUPPORT) && defined(IPPROTO_IP) */ case INET_OPT_IPV6_V6ONLY: -#if HAVE_DECL_IPV6_V6ONLY +#if HAVE_DECL_IPV6_V6ONLY && defined(IPPROTO_IPV6) proto = IPPROTO_IPV6; type = IPV6_V6ONLY; break; @@ -7396,6 +7914,93 @@ static ErlDrvSSizeT inet_fill_opts(inet_descriptor* desc, continue; #endif +#ifndef __WIN32__ + /* Winsock does not have struct cmsghdr */ + case INET_OPT_PKTOPTIONS: { + struct cmsghdr *cmsg, *cmsg_top; + SOCKLEN_T cmsg_sz; + union { + /* Ensure alignment */ + struct cmsghdr hdr; + /* Room for (IP_TOS | IPV6_TCLASS) + IP_TTL */ + char buf[2*CMSG_SPACE(sizeof(int))]; + } cmsgbuf; + /* Select between IPv4 or IPv6 PKTOPTIONS + * depending on the socket protocol family + */ + switch (desc->sfamily) { +#if defined(IPPROTO_IP) && defined(IP_PKTOPTIONS) + case AF_INET: { + proto = IPPROTO_IP; + type = IP_PKTOPTIONS; + } + break; +#endif +#if defined(IPPROTO_IPV6) && defined(IPV6_PKTOPTIONS) && defined(AF_INET6) + case AF_INET6: { + proto = IPPROTO_IPV6; + type = IPV6_PKTOPTIONS; + } + break; +#endif + default: { + RETURN_ERROR(); + } + } /* switch */ + TRUNCATE_TO(0, ptr); + /* Fetch a cmsg buffer from the socket */ + cmsg_sz = sizeof(cmsgbuf.buf); + if (IS_SOCKET_ERROR(sock_getopt(desc->s, proto, type, + cmsgbuf.buf, &cmsg_sz))) { + continue; + } + /* Reply with Opt/8, Length/32, [COpt/8, Value/32]* + * i.e opt, total length and then all returned + * cmsg options and values + */ + PLACE_FOR(1+4, ptr); + *ptr++ = opt; + arg_ptr = ptr; /* Where to put total length */ + arg_sz = 0; /* Total length */ + for (cmsg_top = (struct cmsghdr*)(cmsgbuf.buf + cmsg_sz), + cmsg = (struct cmsghdr*)cmsgbuf.buf; + cmsg < cmsg_top; + cmsg = NXT_CMSG_HDR(cmsg)) { +#define PUT_CMSG_INT32(CMSG_LEVEL, CMSG_TYPE, OPT) \ + if ((cmsg->cmsg_level == CMSG_LEVEL) && \ + (cmsg->cmsg_type == CMSG_TYPE)) { \ + PLACE_FOR(1+4, ptr); \ + *ptr++ = OPT; \ + put_cmsg_int32(cmsg, ptr); \ + arg_sz += 1+4; \ + continue; \ + } +#if defined(IPPROTO_IP) && defined(IP_TOS) + PUT_CMSG_INT32(IPPROTO_IP, IP_TOS, INET_OPT_TOS); +#endif +#if defined(IPPROTO_IPV6) && defined(IPV6_TCLASS) + PUT_CMSG_INT32(IPPROTO_IPV6, IPV6_TCLASS, INET_OPT_TCLASS); +#endif +#if defined(IPPROTO_IP) && defined(IP_TTL) + PUT_CMSG_INT32(IPPROTO_IP, IP_TTL, INET_OPT_TTL); +#endif + /* BSD uses the RECV* names in CMSG fields */ +#if defined(IPPROTO_IP) && defined(IP_RECVTOS) + PUT_CMSG_INT32(IPPROTO_IP, IP_RECVTOS, INET_OPT_TOS); +#endif +#if defined(IPPROTO_IPV6) && defined(IPV6_RECVTCLASS) + PUT_CMSG_INT32(IPPROTO_IPV6, IPV6_RECVTCLASS, INET_OPT_TCLASS); +#endif +#if defined(IPPROTO_IP) && defined(IP_RECVTTL) + PUT_CMSG_INT32(IPPROTO_IP, IP_RECVTTL, INET_OPT_TTL); +#endif +#undef PUT_CMSG_INT32 + } + put_int32(arg_sz, arg_ptr); /* Put total length */ + continue; + } +#endif /* #ifdef __WIN32__ */ + default: RETURN_ERROR(); } @@ -7455,6 +8060,7 @@ static int load_paddrinfo (ErlDrvTermData * spec, int i, return i; } + /* ** "sctp_fill_opts": Returns {ok, Results}, or an error: */ @@ -7704,6 +8310,7 @@ static ErlDrvSSizeT sctp_fill_opts(inet_descriptor* desc, case INET_OPT_PRIORITY : case INET_OPT_TOS : case INET_OPT_TCLASS : + case INET_OPT_TTL : case INET_OPT_IPV6_V6ONLY: case SCTP_OPT_AUTOCLOSE: case SCTP_OPT_MAXSEG : @@ -7711,6 +8318,9 @@ static ErlDrvSSizeT sctp_fill_opts(inet_descriptor* desc, case SCTP_OPT_NODELAY : case SCTP_OPT_DISABLE_FRAGMENTS: case SCTP_OPT_I_WANT_MAPPED_V4_ADDR: + case INET_OPT_RECVTOS : + case INET_OPT_RECVTCLASS : + case INET_OPT_RECVTTL : { int res = 0; unsigned int sz = sizeof(res); @@ -7766,8 +8376,8 @@ static ErlDrvSSizeT sctp_fill_opts(inet_descriptor* desc, } case INET_OPT_TOS: { -# if defined(IP_TOS) && defined(SOL_IP) - proto = SOL_IP; +# if defined(IP_TOS) && defined(IPPROTO_IP) + proto = IPPROTO_IP; type = IP_TOS; is_int = 1; tag = am_tos; @@ -7779,8 +8389,8 @@ static ErlDrvSSizeT sctp_fill_opts(inet_descriptor* desc, } case INET_OPT_TCLASS: { -# if defined(IPV6_TCLASS) && defined(SOL_IPV6) - proto = SOL_IPV6; +# if defined(IPV6_TCLASS) && defined(IPPROTO_IPV6) + proto = IPPROTO_IPV6; type = IPV6_TCLASS; is_int = 1; tag = am_tclass; @@ -7790,8 +8400,60 @@ static ErlDrvSSizeT sctp_fill_opts(inet_descriptor* desc, continue; # endif } + case INET_OPT_TTL: + { +# if defined(IP_TTL) && defined(IPPROTO_IP) + proto = IPPROTO_IP; + type = IP_TTL; + is_int = 1; + tag = am_ttl; + break; +# else + /* Not supported -- ignore */ + continue; +# endif + } + case INET_OPT_RECVTOS: + { +# if defined(IP_RECVTOS) && defined(IPPROTO_IP) + proto = IPPROTO_IP; + type = IP_RECVTOS; + is_int = 0; + tag = am_recvtos; + break; +# else + /* Not supported -- ignore */ + continue; +# endif + } + case INET_OPT_RECVTCLASS: + { +# if defined(IPV6_RECVTCLASS) && defined(IPPROTO_IPV6) + proto = IPPROTO_IPV6; + type = IPV6_RECVTCLASS; + is_int = 0; + tag = am_recvtclass; + break; +# else + /* Not supported -- ignore */ + continue; +# endif + } + case INET_OPT_RECVTTL: + { +# if defined(IP_RECVTTL) && defined(IPPROTO_IP) + proto = IPPROTO_IP; + type = IP_RECVTTL; + is_int = 0; + tag = am_recvttl; + break; +# else + /* Not supported -- ignore */ + continue; +# endif + } case INET_OPT_IPV6_V6ONLY: -# if HAVE_DECL_IPV6_V6ONLY +# if HAVE_DECL_IPV6_V6ONLY && defined(IPPROTO_IPV6) { proto = IPPROTO_IPV6; type = IPV6_V6ONLY; @@ -8423,6 +9085,7 @@ static ErlDrvData inet_start(ErlDrvPort port, int size, int protocol) desc->delimiter = '\n'; /* line delimiting char */ desc->oph = NULL; desc->opt = NULL; + desc->op_ref = 0; desc->peer_ptr = NULL; desc->name_ptr = NULL; @@ -8449,12 +9112,14 @@ static ErlDrvData inet_start(ErlDrvPort port, int size, int protocol) sys_memzero((char *)&desc->remote,sizeof(desc->remote)); - desc->is_ignored = 0; + desc->flags = 0; #ifdef HAVE_SETNS desc->netns = NULL; #endif + desc->recv_cmsgflags = 0; + return (ErlDrvData)desc; } @@ -8849,19 +9514,19 @@ static ErlDrvSSizeT inet_ctl(inet_descriptor* desc, int cmd, char* buf, if (desc->stype != SOCK_STREAM) return ctl_error(EINVAL, rbuf, rsize); - if (*buf == 1 && !desc->is_ignored) { + if (*buf == 1 && !INET_IGNORED(desc)) { sock_select(desc, (FD_READ|FD_WRITE|FD_CLOSE|ERL_DRV_USE_NO_CALLBACK), 0); if (desc->active) - desc->is_ignored = INET_IGNORE_READ; + desc->flags |= INET_IGNORE_READ; else - desc->is_ignored = INET_IGNORE_PASSIVE; - } else if (*buf == 0 && desc->is_ignored) { + desc->flags |= INET_IGNORE_PASSIVE; + } else if (*buf == 0 && INET_IGNORED(desc)) { int flags = FD_CLOSE; - if (desc->is_ignored & INET_IGNORE_READ) + if (desc->flags & INET_IGNORE_READ) flags |= FD_READ; - if (desc->is_ignored & INET_IGNORE_WRITE) + if (desc->flags & INET_IGNORE_WRITE) flags |= FD_WRITE; - desc->is_ignored = INET_IGNORE_NONE; + desc->flags = INET_IGNORE_CLEAR(desc); if (flags != FD_CLOSE) sock_select(desc, flags, 1); } else @@ -9131,6 +9796,7 @@ static ErlDrvData prep_tcp_inet_start(ErlDrvPort port, char* args) desc->tcp_add_flags = 0; desc->http_state = 0; desc->mtd = NULL; + desc->mtd_cache = NULL; desc->multi_first = desc->multi_last = NULL; DEBUGF(("tcp_inet_start(%ld) }\r\n", (long)port)); return (ErlDrvData) desc; @@ -9234,15 +9900,14 @@ static void tcp_close_check(tcp_descriptor* desc) driver_demonitor_process(desc->inet.port, &monitor); send_async_error(desc->inet.dport, id, caller, am_closed); } - clean_multi_timers(&(desc->mtd), desc->inet.port); } - else if (desc->inet.state == INET_STATE_CONNECTING) { async_error_am(INETP(desc), am_closed); } else if (desc->inet.state == INET_STATE_CONNECTED) { async_error_am_all(INETP(desc), am_closed); } + clean_multi_timers(desc, desc->inet.port); } /* @@ -9272,12 +9937,28 @@ static void tcp_inet_stop(ErlDrvData e) * will be freed through tcp_inet_stop later on. */ static void tcp_desc_close(tcp_descriptor* desc) { +#ifdef HAVE_SENDFILE + if(desc->tcp_add_flags & TCP_ADDF_SENDFILE) { + desc->tcp_add_flags &= ~TCP_ADDF_SENDFILE; + close(desc->sendfile.dup_file_fd); + } +#endif + tcp_clear_input(desc); tcp_clear_output(desc); erl_inet_close(INETP(desc)); } +static void tcp_inet_recv_timeout(ErlDrvData e, ErlDrvTermData dummy) +{ + tcp_descriptor* desc = (tcp_descriptor*)e; + ASSERT(!desc->inet.active); + sock_select(INETP(desc),(FD_READ|FD_CLOSE),0); + desc->i_remain = 0; + async_error_am(INETP(desc), am_timeout); +} + /* TCP requests from Erlang */ static ErlDrvSSizeT tcp_inet_ctl(ErlDrvData e, unsigned int cmd, char* buf, ErlDrvSizeT len, @@ -9285,6 +9966,7 @@ static ErlDrvSSizeT tcp_inet_ctl(ErlDrvData e, unsigned int cmd, { tcp_descriptor* desc = (tcp_descriptor*)e; + cmd -= ERTS_INET_DRV_CONTROL_MAGIC_NUMBER; switch(cmd) { case INET_REQ_OPEN: { /* open socket and return internal index */ int domain; @@ -9448,12 +10130,12 @@ static ErlDrvSSizeT tcp_inet_ctl(ErlDrvData e, unsigned int cmd, if (time_left <= 0) { time_left = 1; } - omtd = add_multi_timer(&(desc->mtd), desc->inet.port, ocaller, + omtd = add_multi_timer(desc, desc->inet.port, ocaller, time_left, &tcp_inet_multi_timeout); } enq_old_multi_op(desc, oid, oreq, ocaller, omtd, &omonitor); if (timeout != INET_INFINITY) { - mtd = add_multi_timer(&(desc->mtd), desc->inet.port, caller, + mtd = add_multi_timer(desc, desc->inet.port, caller, timeout, &tcp_inet_multi_timeout); } enq_multi_op(desc, tbuf, INET_REQ_ACCEPT, caller, mtd, &monitor); @@ -9468,7 +10150,7 @@ static ErlDrvSSizeT tcp_inet_ctl(ErlDrvData e, unsigned int cmd, return ctl_xerror("noproc", rbuf, rsize); } if (timeout != INET_INFINITY) { - mtd = add_multi_timer(&(desc->mtd), desc->inet.port, caller, + mtd = add_multi_timer(desc, desc->inet.port, caller, timeout, &tcp_inet_multi_timeout); } enq_multi_op(desc, tbuf, INET_REQ_ACCEPT, caller, mtd, &monitor); @@ -9560,16 +10242,17 @@ static ErlDrvSSizeT tcp_inet_ctl(ErlDrvData e, unsigned int cmd, if (enq_async(INETP(desc), tbuf, TCP_REQ_RECV) < 0) return ctl_error(EALREADY, rbuf, rsize); - if (INETP(desc)->is_ignored || tcp_recv(desc, n) == 0) { + if (INET_IGNORED(INETP(desc)) || tcp_recv(desc, n) == 0) { if (timeout == 0) async_error_am(INETP(desc), am_timeout); else { if (timeout != INET_INFINITY) - driver_set_timer(desc->inet.port, timeout); - if (!INETP(desc)->is_ignored) + add_multi_timer(desc, INETP(desc)->port, 0, + timeout, &tcp_inet_recv_timeout); + if (!INET_IGNORED(INETP(desc))) sock_select(INETP(desc),(FD_READ|FD_CLOSE),1); else - INETP(desc)->is_ignored |= INET_IGNORE_READ; + INETP(desc)->flags |= INET_IGNORE_READ; } } return ctl_reply(INET_REP_OK, tbuf, 2, rbuf, rsize); @@ -9610,6 +10293,59 @@ static ErlDrvSSizeT tcp_inet_ctl(ErlDrvData e, unsigned int cmd, return ctl_reply(INET_REP_OK, NULL, 0, rbuf, rsize); } } + + case TCP_REQ_SENDFILE: { +#ifdef HAVE_SENDFILE + const ErlDrvSizeT required_len = + sizeof(desc->sendfile.dup_file_fd) + + sizeof(Uint64) * 2; + + int raw_file_fd; + + DEBUGF(("tcp_inet_ctl(%ld): SENDFILE\r\n", (long)desc->inet.port)); + + if (len != required_len) { + return ctl_error(EINVAL, rbuf, rsize); + } else if (!IS_CONNECTED(INETP(desc))) { + return ctl_error(ENOTCONN, rbuf, rsize); + } + + sys_memcpy(&raw_file_fd, buf, sizeof(raw_file_fd)); + buf += sizeof(raw_file_fd); + + desc->sendfile.dup_file_fd = dup(raw_file_fd); + + if(desc->sendfile.dup_file_fd == -1) { + return ctl_error(errno, rbuf, rsize); + } + + desc->sendfile.offset = get_int64(buf); + buf += sizeof(Uint64); + + desc->sendfile.length = get_int64(buf); + buf += sizeof(Uint64); + + ASSERT(desc->sendfile.offset >= 0); + ASSERT(desc->sendfile.length >= 0); + + desc->sendfile.ioq_skip = driver_sizeq(desc->inet.port); + desc->sendfile.bytes_sent = 0; + + desc->inet.caller = driver_caller(desc->inet.port); + desc->tcp_add_flags |= TCP_ADDF_SENDFILE; + + /* See if we can finish sending without selecting & rescheduling. */ + if (tcp_inet_sendfile(desc) == 0) { + if(desc->sendfile.length > 0) { + sock_select(INETP(desc), FD_WRITE, 1); + } + } + return ctl_reply(INET_REP_OK, NULL, 0, rbuf, rsize); +#else + return ctl_error(ENOTSUP, rbuf, rsize); +#endif + } + default: DEBUGF(("tcp_inet_ctl(%ld): %u\r\n", (long)desc->inet.port, cmd)); return inet_ctl(INETP(desc), cmd, buf, len, rbuf, rsize); @@ -9617,12 +10353,27 @@ static ErlDrvSSizeT tcp_inet_ctl(ErlDrvData e, unsigned int cmd, } +static void tcp_inet_send_timeout(ErlDrvData e, ErlDrvTermData dummy) +{ + tcp_descriptor* desc = (tcp_descriptor*)e; + ASSERT(IS_BUSY(INETP(desc))); + ASSERT(desc->busy_on_send); + desc->inet.caller = desc->inet.busy_caller; + desc->inet.state &= ~INET_F_BUSY; + desc->busy_on_send = 0; + set_busy_port(desc->inet.port, 0); + inet_reply_error_am(INETP(desc), am_timeout); + if (desc->send_timeout_close) { + tcp_desc_close(desc); + } +} + /* ** tcp_inet_timeout: ** called when timer expire: ** TCP socket may be: ** -** a) receiving -- deselect +** a) receiving -- send timeout ** b) connecting -- close socket ** c) accepting -- reset listener ** @@ -9636,26 +10387,9 @@ static void tcp_inet_timeout(ErlDrvData e) DEBUGF(("tcp_inet_timeout(%ld) {s=%d\r\n", (long)desc->inet.port, desc->inet.s)); if ((state & INET_F_MULTI_CLIENT)) { /* Multi-client always means multi-timers */ - fire_multi_timers(&(desc->mtd), desc->inet.port, e); + fire_multi_timers(desc, desc->inet.port, e); } else if ((state & INET_STATE_CONNECTED) == INET_STATE_CONNECTED) { - if (desc->busy_on_send) { - ASSERT(IS_BUSY(INETP(desc))); - desc->inet.caller = desc->inet.busy_caller; - desc->inet.state &= ~INET_F_BUSY; - desc->busy_on_send = 0; - set_busy_port(desc->inet.port, 0); - inet_reply_error_am(INETP(desc), am_timeout); - if (desc->send_timeout_close) { - tcp_desc_close(desc); - } - } - else { - /* assume recv timeout */ - ASSERT(!desc->inet.active); - sock_select(INETP(desc),(FD_READ|FD_CLOSE),0); - desc->i_remain = 0; - async_error_am(INETP(desc), am_timeout); - } + fire_multi_timers(desc, desc->inet.port, e); } else if ((state & INET_STATE_CONNECTING) == INET_STATE_CONNECTING) { /* assume connect timeout */ @@ -9749,12 +10483,27 @@ static void tcp_inet_commandv(ErlDrvData e, ErlIOVec* ev) static void tcp_inet_flush(ErlDrvData e) { tcp_descriptor* desc = (tcp_descriptor*)e; - if (!(desc->inet.event_mask & FD_WRITE)) { - /* Discard send queue to avoid hanging port (OTP-7615) */ - tcp_clear_output(desc); + int discard_output; + + /* Discard send queue to avoid hanging port (OTP-7615) */ + discard_output = !(desc->inet.event_mask & FD_WRITE); + + discard_output |= desc->tcp_add_flags & TCP_ADDF_LINGER_ZERO; + +#ifdef HAVE_SENDFILE + /* The old file driver aborted when it was stopped during sendfile, so + * we'll clear the flag and discard all output. */ + if(desc->tcp_add_flags & TCP_ADDF_SENDFILE) { + desc->tcp_add_flags &= ~TCP_ADDF_SENDFILE; + close(desc->sendfile.dup_file_fd); + + discard_output = 1; + } +#endif + + if (discard_output) { + tcp_clear_output(desc); } - if (desc->tcp_add_flags & TCP_ADDF_LINGER_ZERO) - tcp_clear_output(desc); } static void tcp_inet_process_exit(ErlDrvData e, ErlDrvMonitor *monitorp) @@ -9770,7 +10519,7 @@ static void tcp_inet_process_exit(ErlDrvData e, ErlDrvMonitor *monitorp) return; } if (timeout != NULL) { - remove_multi_timer(&(desc->mtd), desc->inet.port, timeout); + remove_multi_timer(desc, desc->inet.port, timeout); } if (desc->multi_first == NULL) { sock_select(INETP(desc),FD_ACCEPT,0); @@ -9801,6 +10550,7 @@ static int tcp_recv_closed(tcp_descriptor* desc) #ifdef DEBUG long port = (long) desc->inet.port; /* Used after driver_exit() */ #endif + int blocking_send = 0; DEBUGF(("tcp_recv_closed(%ld): s=%d, in %s, line %d\r\n", port, desc->inet.s, __FILE__, __LINE__)); if (IS_BUSY(INETP(desc))) { @@ -9808,7 +10558,7 @@ static int tcp_recv_closed(tcp_descriptor* desc) desc->inet.caller = desc->inet.busy_caller; tcp_clear_output(desc); if (desc->busy_on_send) { - driver_cancel_timer(desc->inet.port); + cancel_multi_timer(desc, INETP(desc)->port, &tcp_inet_send_timeout); desc->busy_on_send = 0; DEBUGF(("tcp_recv_closed(%ld): busy on send\r\n", port)); } @@ -9816,10 +10566,25 @@ static int tcp_recv_closed(tcp_descriptor* desc) set_busy_port(desc->inet.port, 0); inet_reply_error_am(INETP(desc), am_closed); DEBUGF(("tcp_recv_closed(%ld): busy reply 'closed'\r\n", port)); + blocking_send = 1; } +#ifdef HAVE_SENDFILE + if (desc->tcp_add_flags & TCP_ADDF_SENDFILE) { + tcp_sendfile_aborted(desc, ENOTCONN); + blocking_send = 1; + } +#endif + if (!blocking_send) { + /* No blocking send op to reply to right now. + * If next op is a send, make sure it returns {error,closed} + * rather than {error,enotconn}. + */ + desc->tcp_add_flags |= TCP_ADDF_DELAYED_CLOSE_SEND; + } + if (!desc->inet.active) { - /* We must cancel any timer here ! */ - driver_cancel_timer(desc->inet.port); + /* We must cancel any timer here ! */ + clean_multi_timers(desc, INETP(desc)->port); /* passive mode do not terminate port ! */ tcp_clear_input(desc); if (desc->inet.exitf) { @@ -9854,16 +10619,21 @@ static int tcp_recv_error(tcp_descriptor* desc, int err) desc->inet.caller = desc->inet.busy_caller; tcp_clear_output(desc); if (desc->busy_on_send) { - driver_cancel_timer(desc->inet.port); + cancel_multi_timer(desc, INETP(desc)->port, &tcp_inet_send_timeout); desc->busy_on_send = 0; } desc->inet.state &= ~INET_F_BUSY; set_busy_port(desc->inet.port, 0); inet_reply_error_am(INETP(desc), am_closed); } +#ifdef HAVE_SENDFILE + if (desc->tcp_add_flags & TCP_ADDF_SENDFILE) { + tcp_sendfile_aborted(desc, err); + } +#endif if (!desc->inet.active) { /* We must cancel any timer here ! */ - driver_cancel_timer(desc->inet.port); + clean_multi_timers(desc, INETP(desc)->port); tcp_clear_input(desc); if (desc->inet.exitf) { tcp_desc_close(desc); @@ -9968,13 +10738,13 @@ static int tcp_deliver(tcp_descriptor* desc, int len) if (len == 0) { /* empty buffer or waiting for more input */ if ((desc->i_buf == NULL) || (desc->i_remain > 0)) - return count; + return 0; if ((n = tcp_remain(desc, &len)) != 0) { if (n < 0) /* packet error */ return n; if (len > 0) /* more data pending */ desc->i_remain = len; - return count; + return 0; } } @@ -10026,9 +10796,7 @@ static int tcp_deliver(tcp_descriptor* desc, int len) len = 0; if (!desc->inet.active) { - if (!desc->busy_on_send) { - driver_cancel_timer(desc->inet.port); - } + cancel_multi_timer(desc, INETP(desc)->port, &tcp_inet_recv_timeout); sock_select(INETP(desc),(FD_READ|FD_CLOSE),0); if (desc->i_buf != NULL) tcp_restart_input(desc); @@ -10054,7 +10822,7 @@ static int tcp_recv(tcp_descriptor* desc, int request_len) int len; int nread; - if (desc->i_buf == NULL) { /* allocte a read buffer */ + if (desc->i_buf == NULL) { /* allocate a read buffer */ int sz = (request_len > 0) ? request_len : desc->inet.bufsz; if ((desc->i_buf = alloc_buffer(sz)) == NULL) @@ -10127,10 +10895,11 @@ static int tcp_recv(tcp_descriptor* desc, int request_len) return tcp_deliver(desc, desc->i_ptr - desc->i_ptr_start); } else { - if ((nread = tcp_remain(desc, &len)) < 0) + nread = tcp_remain(desc, &len); + if (nread < 0) return tcp_recv_error(desc, EMSGSIZE); else if (nread == 0) - return tcp_deliver(desc, len); + return tcp_deliver(desc, len); else if (len > 0) desc->i_remain = len; /* set remain */ } @@ -10350,7 +11119,7 @@ static int tcp_inet_input(tcp_descriptor* desc, HANDLE event) #ifdef DEBUG long port = (long) desc->inet.port; /* Used after driver_exit() */ #endif - ASSERT(!INETP(desc)->is_ignored); + ASSERT(!INET_IGNORED(INETP(desc))); DEBUGF(("tcp_inet_input(%ld) {s=%d\r\n", port, desc->inet.s)); /* XXX fprintf(stderr,"tcp_inet_input(%ld) {s=%d}\r\n",(long) desc->inet.port, desc->inet.s); */ if (desc->inet.state == INET_STATE_ACCEPTING) { @@ -10449,7 +11218,7 @@ static int tcp_inet_input(tcp_descriptor* desc, HANDLE event) } if (timeout != NULL) { - remove_multi_timer(&(desc->mtd), desc->inet.port, timeout); + remove_multi_timer(desc, desc->inet.port, timeout); } driver_demonitor_process(desc->inet.port, &monitor); @@ -10508,8 +11277,8 @@ static int tcp_send_or_shutdown_error(tcp_descriptor* desc, int err) if (IS_BUSY(INETP(desc))) { desc->inet.caller = desc->inet.busy_caller; if (desc->busy_on_send) { - driver_cancel_timer(desc->inet.port); - desc->busy_on_send = 0; + cancel_multi_timer(desc, INETP(desc)->port, &tcp_inet_send_timeout); + desc->busy_on_send = 0; } desc->inet.state &= ~INET_F_BUSY; set_busy_port(desc->inet.port, 0); @@ -10524,27 +11293,31 @@ static int tcp_send_or_shutdown_error(tcp_descriptor* desc, int err) DEBUGF(("driver_failure_eof(%ld) in %s, line %d\r\n", (long)desc->inet.port, __FILE__, __LINE__)); if (desc->inet.active) { + ErlDrvTermData err_atom; if (show_econnreset) { tcp_error_message(desc, err); - tcp_closed_message(desc); - inet_reply_error(INETP(desc), err); + err_atom = error_atom(err); } else { - tcp_closed_message(desc); - inet_reply_error_am(INETP(desc), am_closed); + err_atom = am_closed; } + tcp_closed_message(desc); + if (!(desc->tcp_add_flags & TCP_ADDF_SENDFILE)) + inet_reply_error_am(INETP(desc), err_atom); + if (desc->inet.exitf) driver_exit(desc->inet.port, 0); else tcp_desc_close(desc); } else { tcp_close_check(desc); - tcp_desc_close(desc); if (desc->inet.caller) { - if (show_econnreset) - inet_reply_error(INETP(desc), err); - else - inet_reply_error_am(INETP(desc), am_closed); + if (!(desc->tcp_add_flags & TCP_ADDF_SENDFILE)) { + if (show_econnreset) + inet_reply_error(INETP(desc), err); + else + inet_reply_error_am(INETP(desc), am_closed); + } } else { /* No blocking send op to reply to right now. @@ -10553,6 +11326,7 @@ static int tcp_send_or_shutdown_error(tcp_descriptor* desc, int err) */ desc->tcp_add_flags |= TCP_ADDF_DELAYED_CLOSE_SEND; } + tcp_desc_close(desc); /* * Make sure that the next receive operation gets an {error,closed} @@ -10609,6 +11383,12 @@ static int tcp_shutdown_error(tcp_descriptor* desc, int err) return tcp_send_or_shutdown_error(desc, err); } +static void tcp_inet_delay_send(ErlDrvData data, ErlDrvTermData dummy) +{ + tcp_descriptor *desc = (tcp_descriptor*)data; + (void)tcp_inet_output(desc, INETP(desc)->s); +} + /* ** Send non-blocking vector data */ @@ -10649,7 +11429,9 @@ static int tcp_sendv(tcp_descriptor* desc, ErlIOVec* ev) ev->size += h_len; } - if ((sz = driver_sizeq(ix)) > 0) { + sz = driver_sizeq(ix); + + if ((desc->tcp_add_flags & TCP_ADDF_SENDFILE) || sz > 0) { driver_enqv(ix, ev, 0); if (sz+ev->size >= desc->high) { DEBUGF(("tcp_sendv(%ld): s=%d, sender forced busy\r\n", @@ -10659,7 +11441,9 @@ static int tcp_sendv(tcp_descriptor* desc, ErlIOVec* ev) set_busy_port(desc->inet.port, 1); if (desc->send_timeout != INET_INFINITY) { desc->busy_on_send = 1; - driver_set_timer(desc->inet.port, desc->send_timeout); + add_multi_timer(desc, INETP(desc)->port, + 0 /* arg */, desc->send_timeout /* timeout */, + &tcp_inet_send_timeout); } return 1; } @@ -10670,11 +11454,14 @@ static int tcp_sendv(tcp_descriptor* desc, ErlIOVec* ev) DEBUGF(("tcp_sendv(%ld): s=%d, about to send "LLU","LLU" bytes\r\n", (long)desc->inet.port, desc->inet.s, (llu_t)h_len, (llu_t)len)); - if (INETP(desc)->is_ignored) { - INETP(desc)->is_ignored |= INET_IGNORE_WRITE; + if (INET_IGNORED(INETP(desc))) { + INETP(desc)->flags |= INET_IGNORE_WRITE; n = 0; } else if (desc->tcp_add_flags & TCP_ADDF_DELAY_SEND) { - n = 0; + driver_enqv(ix, ev, 0); + add_multi_timer(desc, INETP(desc)->port, 0, + 0, &tcp_inet_delay_send); + return 0; } else if (IS_SOCKET_ERROR(sock_sendv(desc->inet.s, ev->iov, vsize, &n, 0))) { if ((sock_errno() != ERRNO_BLOCK) && (sock_errno() != EINTR)) { @@ -10703,7 +11490,7 @@ static int tcp_sendv(tcp_descriptor* desc, ErlIOVec* ev) DEBUGF(("tcp_sendv(%ld): s=%d, Send failed, queuing\r\n", (long)desc->inet.port, desc->inet.s)); driver_enqv(ix, ev, n); - if (!INETP(desc)->is_ignored) + if (!INET_IGNORED(INETP(desc))) sock_select(INETP(desc),(FD_WRITE|FD_CLOSE), 1); } return 0; @@ -10743,8 +11530,9 @@ static int tcp_send(tcp_descriptor* desc, char* ptr, ErlDrvSizeT len) inet_output_count(INETP(desc), len+h_len); + sz = driver_sizeq(ix); - if ((sz = driver_sizeq(ix)) > 0) { + if ((desc->tcp_add_flags & TCP_ADDF_SENDFILE) || sz > 0) { if (h_len > 0) driver_enq(ix, buf, h_len); driver_enq(ix, ptr, len); @@ -10756,7 +11544,9 @@ static int tcp_send(tcp_descriptor* desc, char* ptr, ErlDrvSizeT len) set_busy_port(desc->inet.port, 1); if (desc->send_timeout != INET_INFINITY) { desc->busy_on_send = 1; - driver_set_timer(desc->inet.port, desc->send_timeout); + add_multi_timer(desc, INETP(desc)->port, + 0 /* arg */, desc->send_timeout /* timeout */, + &tcp_inet_send_timeout); } return 1; } @@ -10769,8 +11559,8 @@ static int tcp_send(tcp_descriptor* desc, char* ptr, ErlDrvSizeT len) DEBUGF(("tcp_send(%ld): s=%d, about to send "LLU","LLU" bytes\r\n", (long)desc->inet.port, desc->inet.s, (llu_t)h_len, (llu_t)len)); - if (INETP(desc)->is_ignored) { - INETP(desc)->is_ignored |= INET_IGNORE_WRITE; + if (INET_IGNORED(INETP(desc))) { + INETP(desc)->flags |= INET_IGNORE_WRITE; n = 0; } else if (desc->tcp_add_flags & TCP_ADDF_DELAY_SEND) { sock_send(desc->inet.s, buf, 0, 0); @@ -10803,7 +11593,7 @@ static int tcp_send(tcp_descriptor* desc, char* ptr, ErlDrvSizeT len) n -= h_len; driver_enq(ix, ptr+n, len-n); } - if (!INETP(desc)->is_ignored) + if (!INET_IGNORED(INETP(desc))) sock_select(INETP(desc),(FD_WRITE|FD_CLOSE), 1); } return 0; @@ -10834,6 +11624,247 @@ static void tcp_inet_drv_input(ErlDrvData data, ErlDrvEvent event) (void)tcp_inet_input((tcp_descriptor*)data, (HANDLE)event); } +#ifdef HAVE_SENDFILE +static int tcp_sendfile_completed(tcp_descriptor* desc) { + ErlDrvTermData spec[LOAD_PORT_CNT + LOAD_TUPLE_CNT * 2 + + LOAD_ATOM_CNT * 2 + LOAD_UINT_CNT * 2]; + Uint32 sent_low, sent_high; + int i; + + desc->tcp_add_flags &= ~TCP_ADDF_SENDFILE; + close(desc->sendfile.dup_file_fd); + + /* While we flushed the output queue prior to sending the file, we've + * deferred clearing busy status until now as there's no point in doing so + * while we still have a file to send. + * + * The watermark is checked since more data may have been added while we + * were sending the file. */ + + if (driver_sizeq(desc->inet.port) <= desc->low) { + if (IS_BUSY(INETP(desc))) { + desc->inet.caller = desc->inet.busy_caller; + desc->inet.state &= ~INET_F_BUSY; + + set_busy_port(desc->inet.port, 0); + + /* if we have a timer then cancel and send ok to client */ + if (desc->busy_on_send) { + cancel_multi_timer(desc, INETP(desc)->port, + &tcp_inet_send_timeout); + desc->busy_on_send = 0; + } + + inet_reply_ok(INETP(desc)); + } + } + + if (driver_sizeq(desc->inet.port) == 0) { + sock_select(INETP(desc), FD_WRITE, 0); + send_empty_out_q_msgs(INETP(desc)); + + if (desc->tcp_add_flags & TCP_ADDF_PENDING_SHUTDOWN) { + tcp_shutdown_async(desc); + } + } + + sent_low = ((Uint64)desc->sendfile.bytes_sent >> 0) & 0xFFFFFFFF; + sent_high = ((Uint64)desc->sendfile.bytes_sent >> 32) & 0xFFFFFFFF; + + i = LOAD_ATOM(spec, 0, am_sendfile); + i = LOAD_PORT(spec, i, desc->inet.dport); + i = LOAD_ATOM(spec, i, am_ok); + i = LOAD_UINT(spec, i, sent_low); + i = LOAD_UINT(spec, i, sent_high); + i = LOAD_TUPLE(spec, i, 3); + i = LOAD_TUPLE(spec, i, 3); + + ASSERT(i == sizeof(spec)/sizeof(*spec)); + + return erl_drv_output_term(desc->inet.dport, spec, i); +} + +static int tcp_sendfile_aborted(tcp_descriptor* desc, int socket_error) { + ErlDrvTermData spec[LOAD_PORT_CNT + LOAD_TUPLE_CNT * 2 + LOAD_ATOM_CNT * 3]; + int i; + + /* We don't clean up sendfile state here, as that's done in tcp_desc_close + * following normal error handling. All we do here is report the failure. */ + + i = LOAD_ATOM(spec, 0, am_sendfile); + i = LOAD_PORT(spec, i, desc->inet.dport); + i = LOAD_ATOM(spec, i, am_error); + + switch (socket_error) { + case ECONNRESET: + case ENOTCONN: + case EPIPE: + i = LOAD_ATOM(spec, i, am_closed); + break; + default: + i = LOAD_ATOM(spec, i, error_atom(socket_error)); + } + + i = LOAD_TUPLE(spec, i, 2); + i = LOAD_TUPLE(spec, i, 3); + + ASSERT(i == sizeof(spec)/sizeof(*spec)); + + return erl_drv_output_term(desc->inet.dport, spec, i); +} + +static int tcp_inet_sendfile(tcp_descriptor* desc) { + ErlDrvPort ix = desc->inet.port; + int result = 0; + ssize_t n; + + DEBUGF(("tcp_inet_sendfile(%ld) {s=%d\r\n", (long)ix, desc->inet.s)); + + /* If there was any data in the queue by the time sendfile was issued, + * we'll need to skip it first. Note that we don't clear busy status until + * we're finished sending the file. */ + while (desc->sendfile.ioq_skip > 0) { + ssize_t bytes_to_send; + SysIOVec* iov; + int vsize; + + ASSERT(driver_sizeq(ix) >= desc->sendfile.ioq_skip); + + if ((iov = driver_peekq(ix, &vsize)) == NULL) { + ERTS_INTERNAL_ERROR("ioq empty when sendfile.ioq_skip > 0"); + } + + bytes_to_send = MIN(desc->sendfile.ioq_skip, iov[0].iov_len); + n = sock_send(desc->inet.s, iov[0].iov_base, bytes_to_send, 0); + + if (!IS_SOCKET_ERROR(n)) { + desc->sendfile.ioq_skip -= n; + driver_deq(ix, n); + } else if (sock_errno() == ERRNO_BLOCK) { +#ifdef __WIN32__ + desc->inet.send_would_block = 1; +#endif + goto done; + } else if (sock_errno() != EINTR) { + goto socket_error; + } + } + + while (desc->sendfile.length > 0) { + /* For some reason the maximum ssize_t cannot be used as the max size. + * 1GB seems to work on all platforms */ + const Sint64 SENDFILE_CHUNK_SIZE = ((1UL << 30) - 1); + + ssize_t bytes_to_send = MIN(SENDFILE_CHUNK_SIZE, desc->sendfile.length); + off_t offset = desc->sendfile.offset; + +#if defined(__linux__) + n = sendfile(desc->inet.s, desc->sendfile.dup_file_fd, &offset, + bytes_to_send); +#elif defined(__FreeBSD__) || defined(__DragonFly__) || defined(__DARWIN__) + { + off_t bytes_sent; + int error; + + #if defined(__DARWIN__) + bytes_sent = bytes_to_send; + + error = sendfile(desc->sendfile.dup_file_fd, desc->inet.s, offset, + &bytes_sent, NULL, 0); + n = bytes_sent; + #else + error = sendfile(desc->sendfile.dup_file_fd, desc->inet.s, offset, + bytes_to_send, NULL, &bytes_sent, 0); + n = bytes_sent; + #endif + + if(error < 0) { + /* EAGAIN/EINTR report partial success by setting bytes_sent, + * so we have to skip error handling if nonzero, and skip EOF + * handling if zero, as it's possible that we didn't manage to + * send anything at all before being interrupted by a + * signal. */ + if((errno != EAGAIN && errno != EINTR) || bytes_sent == 0) { + n = -1; + } + } + } +#elif defined(__sun) && defined(__SVR4) && defined(HAVE_SENDFILEV) + { + sendfilevec_t sfvec[1]; + size_t bytes_sent; + ssize_t error; + + sfvec[0].sfv_fd = desc->sendfile.dup_file_fd; + sfvec[0].sfv_len = bytes_to_send; + sfvec[0].sfv_off = offset; + sfvec[0].sfv_flag = 0; + + error = sendfilev(desc->inet.s, sfvec, 1, &bytes_sent); + n = bytes_sent; + + if(error < 0) { + if(errno == EINVAL) { + /* On some solaris versions (I've seen it on SunOS 5.10), + * using a sfv_len larger than the filesize will result in + * a (-1 && errno == EINVAL). We translate this to a + * successful send of the data.*/ + } else { + /* EAGAIN/EINTR behavior is identical to *BSD. */ + if((errno != EAGAIN && errno != EINTR) || bytes_sent == 0) { + n = -1; + } + } + } + } +#else + #error "Unsupported sendfile syscall; update configure test." +#endif + + if (n > 0) { + desc->sendfile.bytes_sent += n; + desc->sendfile.offset += n; + desc->sendfile.length -= n; + } else if (n == 0) { + /* EOF. */ + desc->sendfile.length = 0; + break; + } else if (IS_SOCKET_ERROR(n) && sock_errno() != EINTR) { + if (sock_errno() != ERRNO_BLOCK) { + goto socket_error; + } + +#ifdef __WIN32__ + desc->inet.send_would_block = 1; +#endif + break; + } + } + + if (desc->sendfile.length == 0) { + tcp_sendfile_completed(desc); + } + + goto done; + +socket_error: { + int socket_errno = sock_errno(); + + DEBUGF(("tcp_inet_sendfile(%ld): send errno = %d (errno %d)\r\n", + (long)desc->inet.port, socket_errno, errno)); + + tcp_sendfile_aborted(desc, socket_errno); + result = tcp_send_error(desc, socket_errno); + + goto done; + } + +done: + DEBUGF(("tcp_inet_sendfile(%ld) }\r\n", (long)desc->inet.port)); + return result; +} +#endif /* HAVE_SENDFILE */ + /* socket ready for ouput: ** 1. INET_STATE_CONNECTING => non block connect ? ** 2. INET_STATE_CONNECTED => write output @@ -10843,7 +11874,7 @@ static int tcp_inet_output(tcp_descriptor* desc, HANDLE event) int ret = 0; ErlDrvPort ix = desc->inet.port; - ASSERT(!INETP(desc)->is_ignored); + ASSERT(!INET_IGNORED(INETP(desc))); DEBUGF(("tcp_inet_output(%ld) {s=%d\r\n", (long)desc->inet.port, desc->inet.s)); if (desc->inet.state == INET_STATE_CONNECTING) { @@ -10894,7 +11925,14 @@ static int tcp_inet_output(tcp_descriptor* desc, HANDLE event) async_ok(INETP(desc)); } else if (IS_CONNECTED(INETP(desc))) { - for (;;) { + +#ifdef HAVE_SENDFILE + if(desc->tcp_add_flags & TCP_ADDF_SENDFILE) { + return tcp_inet_sendfile(desc); + } +#endif + + for (;;) { int vsize; ssize_t n; SysIOVec* iov; @@ -10920,6 +11958,12 @@ static int tcp_inet_output(tcp_descriptor* desc, HANDLE event) #ifdef __WIN32__ desc->inet.send_would_block = 1; #endif + /* If DELAY_SEND is set ready_output may have + been called without doing select so we do + a select in order to get into the correct + state */ + if (desc->tcp_add_flags & TCP_ADDF_DELAY_SEND) + sock_select(INETP(desc), FD_WRITE, 1); goto done; } else if (n == 0) { /* Workaround for redhat/CentOS 6.3 returning 0 when sending packets with @@ -10945,7 +11989,7 @@ static int tcp_inet_output(tcp_descriptor* desc, HANDLE event) set_busy_port(desc->inet.port, 0); /* if we have a timer then cancel and send ok to client */ if (desc->busy_on_send) { - driver_cancel_timer(desc->inet.port); + cancel_multi_timer(desc, INETP(desc)->port, &tcp_inet_send_timeout); desc->busy_on_send = 0; } inet_reply_ok(INETP(desc)); @@ -11152,6 +12196,7 @@ static ErlDrvSSizeT packet_inet_ctl(ErlDrvData e, unsigned int cmd, char* buf, int type = SOCK_DGRAM; int af = AF_INET; + cmd -= ERTS_INET_DRV_CONTROL_MAGIC_NUMBER; switch(cmd) { case INET_REQ_OPEN: /* open socket and return internal index */ DEBUGF(("packet_inet_ctl(%ld): OPEN\r\n", (long)desc->port)); @@ -11300,24 +12345,20 @@ static ErlDrvSSizeT packet_inet_ctl(ErlDrvData e, unsigned int cmd, char* buf, (desc->sfamily, &remote, &buf, &len)) != NULL) return ctl_xerror(xerror, rbuf, rsize); - sock_select(desc, FD_CONNECT, 1); code = sock_connect(desc->s, &remote.sa, len); if (IS_SOCKET_ERROR(code) && (sock_errno() == EINPROGRESS)) { /* XXX: Unix only -- WinSock would have a different cond! */ - desc->state = INET_STATE_CONNECTING; if (timeout != INET_INFINITY) driver_set_timer(desc->port, timeout); enq_async(desc, tbuf, INET_REQ_CONNECT); + async_ok(desc); } else if (code == 0) { /* OK we are connected */ - sock_select(desc, FD_CONNECT, 0); - desc->state = INET_STATE_CONNECTED; enq_async(desc, tbuf, INET_REQ_CONNECT); async_ok(desc); } else { - sock_select(desc, FD_CONNECT, 0); return ctl_error(sock_errno(), rbuf, rsize); } return ctl_reply(INET_REP_OK, tbuf, 2, rbuf, rsize); @@ -11508,9 +12549,12 @@ static void packet_inet_timeout(ErlDrvData e) { udp_descriptor * udesc = (udp_descriptor*) e; inet_descriptor * desc = INETP(udesc); - if (!(desc->active)) + if (!(desc->active)) { sock_select(desc, FD_READ, 0); - async_error_am (desc, am_timeout); + async_error_am (desc, am_timeout); + } else { + (void)packet_inet_input(udesc, desc->s); + } } @@ -11545,10 +12589,10 @@ static void packet_inet_command(ErlDrvData e, char* buf, ErlDrvSizeT len) if (IS_SCTP(desc)) { ErlDrvSizeT data_len; - struct iovec iov[1]; /* For real data */ - struct msghdr mhdr; /* Message wrapper */ - struct sctp_sndrcvinfo *sri; /* The actual ancilary data */ - union { /* For ancilary data */ + struct iovec iov[1]; /* For real data */ + struct msghdr mhdr; /* Message wrapper */ + struct sctp_sndrcvinfo *sri; /* The actual ancillary data */ + union { /* For ancillary data */ struct cmsghdr hdr; char ancd[CMSG_SPACE(sizeof(*sri))]; } cmsg; @@ -11558,12 +12602,12 @@ static void packet_inet_command(ErlDrvData e, char* buf, ErlDrvSizeT len) return; } - /* The ancilary data */ + /* The ancillary data */ sri = (struct sctp_sndrcvinfo *) (CMSG_DATA(&cmsg.hdr)); /* Get the "sndrcvinfo" from the buffer, advancing the "ptr": */ ptr = sctp_get_sendparams(sri, ptr); - /* The ancilary data wrapper */ + /* The ancillary data wrapper */ cmsg.hdr.cmsg_level = IPPROTO_SCTP; cmsg.hdr.cmsg_type = SCTP_SNDRCV; cmsg.hdr.cmsg_len = CMSG_LEN(sizeof(*sri)); @@ -11578,7 +12622,7 @@ static void packet_inet_command(ErlDrvData e, char* buf, ErlDrvSizeT len) iov[0].iov_base = ptr; /* The real data */ mhdr.msg_iov = iov; mhdr.msg_iovlen = 1; - mhdr.msg_control = cmsg.ancd; /* For ancilary data */ + mhdr.msg_control = cmsg.ancd; /* For ancillary data */ mhdr.msg_controllen = cmsg.hdr.cmsg_len; VALGRIND_MAKE_MEM_DEFINED(mhdr.msg_control, mhdr.msg_controllen); /*suppress "uninitialised bytes"*/ mhdr.msg_flags = 0; /* Not used with "sendmsg" */ @@ -11662,10 +12706,12 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) char abuf[sizeof(inet_address)]; /* buffer address; enough??? */ int packet_count = udesc->read_packets; int count = 0; /* number of packets delivered to owner */ -#ifdef HAVE_SCTP +#ifndef __WIN32__ struct msghdr mhdr; /* Top-level msg structure */ struct iovec iov[1]; /* Data or Notification Event */ - char ancd[SCTP_ANC_BUFF_SIZE]; /* Ancillary Data */ + char ancd[ANC_BUFF_SIZE]; /* Ancillary Data */ +#endif +#ifdef HAVE_SCTP int short_recv = 0; #endif @@ -11675,15 +12721,11 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) sys_memzero((char *) &other, sizeof(other)); /* udesc->i_buf is only kept between SCTP fragments */ - if (udesc->i_buf == NULL) { - udesc->i_bufsz = desc->bufsz + len; - if ((udesc->i_buf = alloc_buffer(udesc->i_bufsz)) == NULL) - return packet_error(udesc, ENOMEM); - /* pointer to message start */ - udesc->i_ptr = udesc->i_buf->orig_bytes + len; - } else { - ErlDrvBinary* tmp; +#ifdef HAVE_SCTP + if (udesc->i_buf != NULL) { + ErlDrvBinary* tmp; int bufsz; + ASSERT(IS_SCTP(desc)); bufsz = desc->bufsz + (udesc->i_ptr - udesc->i_buf->orig_bytes); if ((tmp = realloc_buffer(udesc->i_buf, bufsz)) == NULL) { release_buffer(udesc->i_buf); @@ -11695,6 +12737,15 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) udesc->i_buf = tmp; udesc->i_bufsz = bufsz; } + } else +#endif + { + ASSERT(udesc->i_buf == NULL); + udesc->i_bufsz = desc->bufsz + len; + if ((udesc->i_buf = alloc_buffer(udesc->i_bufsz)) == NULL) + return packet_error(udesc, ENOMEM); + /* pointer to message start */ + udesc->i_ptr = udesc->i_buf->orig_bytes + len; } /* Note: On Windows NT, recvfrom() fails if the socket is connected. */ @@ -11709,7 +12760,7 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) mhdr.msg_iov = iov; mhdr.msg_iovlen = 1; mhdr.msg_control = ancd; - mhdr.msg_controllen = SCTP_ANC_BUFF_SIZE; + mhdr.msg_controllen = ANC_BUFF_SIZE; mhdr.msg_flags = 0; /* To be filled by "recvmsg" */ /* Do the actual SCTP receive: */ @@ -11724,6 +12775,24 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) other = desc->remote; goto check_result; } +#ifndef __WIN32__ + /* recvmsg() does not exist in the Winsock API */ + if (desc->recv_cmsgflags) { + /* Use recvmsg() */ + iov->iov_base = udesc->i_ptr; + iov->iov_len = desc->bufsz; + mhdr.msg_name = &other; + mhdr.msg_namelen = len; + mhdr.msg_iov = iov; + mhdr.msg_iovlen = 1; + mhdr.msg_control = ancd; + mhdr.msg_controllen = ANC_BUFF_SIZE; + mhdr.msg_flags = 0; + n = sock_recvmsg(desc->s, &mhdr, 0); + len = mhdr.msg_namelen; + goto check_result; + } +#endif n = sock_recvfrom(desc->s, udesc->i_ptr, desc->bufsz, 0, &other.sa, &len); check_result: @@ -11736,7 +12805,7 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) udesc->i_buf = NULL; if (!desc->active) { async_error(desc, err); - driver_cancel_timer(desc->port); + driver_cancel_timer(desc->port); sock_select(desc,FD_READ,0); } else { @@ -11753,6 +12822,14 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) ) { sock_select(desc,FD_READ,1); } +#ifdef HAVE_SCTP + if (!short_recv) { +#endif + release_buffer(udesc->i_buf); + udesc->i_buf = NULL; +#ifdef HAVE_SCTP + } +#endif return count; /* strange, not ready */ } @@ -11768,7 +12845,7 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) { /* message received */ int code; - void * extra = NULL; + void *mp; char * ptr; int nsz; @@ -11799,21 +12876,25 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) udesc->i_ptr = NULL; /* not used from here */ } } + mp = NULL; #ifdef HAVE_SCTP - if (IS_SCTP(desc)) extra = &mhdr; + if (IS_SCTP(desc)) mp = &mhdr; +#endif +#ifndef __WIN32__ + if (desc->recv_cmsgflags) mp = &mhdr; #endif /* Actual parsing and return of the data received, occur here: */ code = packet_reply_binary_data(desc, len, udesc->i_buf, (sizeof(other) - len), nsz, - extra); + mp); free_buffer(udesc->i_buf); udesc->i_buf = NULL; if (code < 0) return count; count++; if (!desc->active) { - driver_cancel_timer(desc->port); /* possibly cancel */ + driver_cancel_timer(desc->port); sock_select(desc,FD_READ,0); return count; /* passive mode (read one packet only) */ } @@ -11829,80 +12910,18 @@ static int packet_inet_input(udp_descriptor* udesc, HANDLE event) sock_select(desc, FD_READ, 1); } #endif - return count; -} - -static void packet_inet_drv_output(ErlDrvData e, ErlDrvEvent event) -{ - (void) packet_inet_output((udp_descriptor*)e, (HANDLE)event); -} -/* UDP/SCTP socket ready for output: -** This is a Back-End for Non-Block SCTP Connect (INET_STATE_CONNECTING) -*/ -static int packet_inet_output(udp_descriptor* udesc, HANDLE event) -{ - inet_descriptor* desc = INETP(udesc); - int ret = 0; - ErlDrvPort ix = desc->port; + /* We set a timer on the port to trigger now. + This emulates a "yield" operation as that is + what we want to do here. We do *NOT* do a deselect + as that is expensive, instead we check if the + socket it still active when the timeout triggers + and if it is not, then we just ignore the timeout */ + driver_set_timer(desc->port, 0); - DEBUGF(("packet_inet_output(%ld) {s=%d\r\n", - (long)desc->port, desc->s)); - - if (desc->state == INET_STATE_CONNECTING) { - sock_select(desc, FD_CONNECT, 0); - - driver_cancel_timer(ix); /* posssibly cancel a timer */ -#ifndef __WIN32__ - /* - * XXX This is strange. This *should* work on Windows NT too, - * but doesn't. An bug in Winsock 2.0 for Windows NT? - * - * See "Unix Netwok Programming", W.R.Stevens, p 412 for a - * discussion about Unix portability and non blocking connect. - */ - -#ifndef SO_ERROR - { - int sz = sizeof(desc->remote); - int code = sock_peer(desc->s, - (struct sockaddr*) &desc->remote, &sz); - - if (IS_SOCKET_ERROR(code)) { - desc->state = INET_STATE_OPEN; /* restore state */ - ret = async_error(desc, sock_errno()); - goto done; - } - } -#else - { - int error = 0; /* Has to be initiated, we check it */ - unsigned int sz = sizeof(error); /* even if we get -1 */ - int code = sock_getopt(desc->s, SOL_SOCKET, SO_ERROR, - (void *)&error, &sz); - - if ((code < 0) || error) { - desc->state = INET_STATE_OPEN; /* restore state */ - ret = async_error(desc, error); - goto done; - } - } -#endif /* SO_ERROR */ -#endif /* !__WIN32__ */ - - desc->state = INET_STATE_CONNECTED; - async_ok(desc); - } - else { - sock_select(desc,FD_CONNECT,0); - - DEBUGF(("packet_inet_output(%ld): bad state: %04x\r\n", - (long)desc->port, desc->state)); - } - done: - DEBUGF(("packet_inet_output(%ld) }\r\n", (long)desc->port)); - return ret; + return count; } + #endif /*---------------------------------------------------------------------------*/ @@ -11963,55 +12982,71 @@ make_noninheritable_handle(SOCKET s) * Multi-timers */ -static void fire_multi_timers(MultiTimerData **first, ErlDrvPort port, +static void fire_multi_timers(tcp_descriptor *desc, ErlDrvPort port, ErlDrvData data) { - ErlDrvTime next_timeout; - if (!*first) { - ASSERT(0); - return; + ErlDrvTime next_timeout = 0; + if (!desc->mtd) { + ASSERT(0); + return; } #ifdef DEBUG { ErlDrvTime chk = erl_drv_monotonic_time(ERL_DRV_MSEC); - ASSERT(chk >= (*first)->when); + ASSERT(chk >= desc->mtd->when); } #endif do { - MultiTimerData *save = *first; - *first = save->next; - (*(save->timeout_function))(data,save->caller); - FREE(save); - if (*first == NULL) { + MultiTimerData save = *desc->mtd; + + /* We first remove the timer so that the timeout_functions has + can call clean_multi_timers without breaking anything */ + if (desc->mtd_cache == NULL) { + desc->mtd_cache = desc->mtd; + } else { + FREE(desc->mtd); + } + + desc->mtd = save.next; + if (desc->mtd != NULL) + desc->mtd->prev = NULL; + + (*(save.timeout_function))(data,save.caller); + + if (desc->mtd == NULL) return; - } - (*first)->prev = NULL; - next_timeout = (*first)->when - erl_drv_monotonic_time(ERL_DRV_MSEC); + + next_timeout = desc->mtd->when - erl_drv_monotonic_time(ERL_DRV_MSEC); } while (next_timeout <= 0); + driver_set_timer(port, (unsigned long) next_timeout); } -static void clean_multi_timers(MultiTimerData **first, ErlDrvPort port) +static void clean_multi_timers(tcp_descriptor *desc, ErlDrvPort port) { - MultiTimerData *p; - if (*first) { + if (desc->mtd) { driver_cancel_timer(port); } - while (*first) { - p = *first; - *first = p->next; - FREE(p); + while (desc->mtd) { + MultiTimerData *p = desc->mtd; + desc->mtd = p->next; + FREE(p); + } + desc->mtd = NULL; + if (desc->mtd_cache) { + FREE(desc->mtd_cache); + desc->mtd_cache = NULL; } } -static void remove_multi_timer(MultiTimerData **first, ErlDrvPort port, MultiTimerData *p) +static void remove_multi_timer(tcp_descriptor *desc, ErlDrvPort port, MultiTimerData *p) { if (p->prev != NULL) { p->prev->next = p->next; } else { driver_cancel_timer(port); - *first = p->next; - if (*first) { - ErlDrvTime ntmo = (*first)->when - erl_drv_monotonic_time(ERL_DRV_MSEC); + desc->mtd = p->next; + if (desc->mtd) { + ErlDrvTime ntmo = desc->mtd->when - erl_drv_monotonic_time(ERL_DRV_MSEC); if (ntmo < 0) ntmo = 0; driver_set_timer(port, (unsigned long) ntmo); @@ -12020,36 +13055,67 @@ static void remove_multi_timer(MultiTimerData **first, ErlDrvPort port, MultiTim if (p->next != NULL) { p->next->prev = p->prev; } - FREE(p); + if (desc->mtd_cache == NULL) + desc->mtd_cache = p; + else + FREE(p); +} + +/* Cancel a timer based on the timeout_fun */ +static void cancel_multi_timer(tcp_descriptor *desc, ErlDrvPort port, + void (*timeout_fun)(ErlDrvData drv_data, + ErlDrvTermData caller)) +{ + MultiTimerData *timer = desc->mtd; + while(timer && timer->timeout_function != timeout_fun) { + timer = timer->next; + } + if (timer) { + remove_multi_timer(desc, port, timer); + } } -static MultiTimerData *add_multi_timer(MultiTimerData **first, ErlDrvPort port, +static MultiTimerData *add_multi_timer(tcp_descriptor *desc, ErlDrvPort port, ErlDrvTermData caller, unsigned timeout, void (*timeout_fun)(ErlDrvData drv_data, ErlDrvTermData caller)) { MultiTimerData *mtd, *p, *s; - mtd = ALLOC(sizeof(MultiTimerData)); - mtd->when = erl_drv_monotonic_time(ERL_DRV_MSEC) + ((ErlDrvTime) timeout) + 1; + + /* Use cached timer if available */ + if (desc->mtd_cache != NULL) { + mtd = desc->mtd_cache; + desc->mtd_cache = NULL; + } else + mtd = ALLOC(sizeof(MultiTimerData)); + + if (timeout) + mtd->when = erl_drv_monotonic_time(ERL_DRV_MSEC) + ((ErlDrvTime) timeout); + else + mtd->when = INT64_MIN; /* Don't have to get the time for 0 msec timeouts */ + mtd->timeout_function = timeout_fun; mtd->caller = caller; mtd->next = mtd->prev = NULL; - for(p = *first,s = NULL; p != NULL; s = p, p = p->next) { + + /* Find correct slot in timer linked list */ + for(p = desc->mtd,s = NULL; p != NULL; s = p, p = p->next) { if (p->when >= mtd->when) { break; } } + /* Insert in linked list */ if (!p) { if (!s) { - *first = mtd; + desc->mtd = mtd; } else { s->next = mtd; mtd->prev = s; } } else { if (!s) { - *first = mtd; + desc->mtd = mtd; } else { s->next = mtd; mtd->prev = s; @@ -12057,10 +13123,8 @@ static MultiTimerData *add_multi_timer(MultiTimerData **first, ErlDrvPort port, mtd->next = p; p->prev = mtd; } + /* Possibly set new timer */ if (!s) { - if (mtd->next) { - driver_cancel_timer(port); - } driver_set_timer(port,timeout); } return mtd; diff --git a/erts/emulator/drivers/unix/ttsl_drv.c b/erts/emulator/drivers/unix/ttsl_drv.c index 2a508b02eb..d2a524cb6c 100644 --- a/erts/emulator/drivers/unix/ttsl_drv.c +++ b/erts/emulator/drivers/unix/ttsl_drv.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2016. All Rights Reserved. + * Copyright Ericsson AB 1996-2018. 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. @@ -394,6 +394,8 @@ static ErlDrvSSizeT ttysl_control(ErlDrvData drv_data, { char resbuff[2*sizeof(Uint32)]; ErlDrvSizeT res_size; + + command -= ERTS_TTYSL_DRV_CONTROL_MAGIC_NUMBER; switch (command) { case CTRL_OP_GET_WINSIZE: { @@ -419,7 +421,7 @@ static ErlDrvSSizeT ttysl_control(ErlDrvData drv_data, } break; default: - return 0; + return -1; } if (rlen < res_size) { *rbuf = driver_alloc(res_size); @@ -892,8 +894,8 @@ static void ttysl_from_tty(ErlDrvData ttysl_data, ErlDrvEvent fd) tpos = 0; } } - } else { - DEBUGLOG(("ttysl_from_tty: driver failure in read(%d,..) = %d\n", (int)(SWord)fd, i)); + } else if (errno != EAGAIN && errno != EWOULDBLOCK) { + DEBUGLOG(("ttysl_from_tty: driver failure in read(%d,..) = %d (errno = %d)\n", (int)(SWord)fd, i, errno)); driver_failure(ttysl_port, -1); } } diff --git a/erts/emulator/drivers/unix/unix_efile.c b/erts/emulator/drivers/unix/unix_efile.c deleted file mode 100644 index 33e4d75ef7..0000000000 --- a/erts/emulator/drivers/unix/unix_efile.c +++ /dev/null @@ -1,1102 +0,0 @@ -/* - * %CopyrightBegin% - * - * Copyright Ericsson AB 1997-2017. 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. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - * - * %CopyrightEnd% - */ -/* - * Purpose: Provides file and directory operations for Unix. - */ -#ifdef HAVE_CONFIG_H -# include "config.h" -#endif -#if defined(HAVE_POSIX_FALLOCATE) && !defined(__sun) && !defined(__sun__) -#define _XOPEN_SOURCE 600 -#endif -#if !defined(_GNU_SOURCE) && defined(HAVE_LINUX_FALLOC_H) -#define _GNU_SOURCE -#endif -#include "sys.h" -#include "erl_driver.h" -#include "erl_efile.h" -#include <utime.h> -#ifdef HAVE_UNISTD_H -#include <unistd.h> -#endif -#ifdef HAVE_SYS_UIO_H -#include <sys/types.h> -#include <sys/uio.h> -#if defined(HAVE_SENDFILE) && (defined(__FreeBSD__) || defined(__DragonFly__)) -/* Need to define __BSD_VISIBLE in order to expose prototype of sendfile */ -#define __BSD_VISIBLE 1 -#include <sys/socket.h> -#endif -#endif -#if defined(HAVE_SENDFILE) && (defined(__linux__) || (defined(__sun) && defined(__SVR4))) -#include <sys/sendfile.h> -#endif - -#if defined(__APPLE__) && defined(__MACH__) && !defined(__DARWIN__) -#define __DARWIN__ 1 -#endif - -#if defined(__DARWIN__) || defined(HAVE_LINUX_FALLOC_H) || defined(HAVE_POSIX_FALLOCATE) -#include <fcntl.h> -#endif - -#ifdef HAVE_LINUX_FALLOC_H -#include <linux/falloc.h> -#endif - -#ifdef SUNOS4 -# define getcwd(buf, size) getwd(buf) -#endif - -/* Find a definition of MAXIOV, that is used in the code later. */ -#if defined IOV_MAX -#define MAXIOV IOV_MAX -#elif defined UIO_MAXIOV -#define MAXIOV UIO_MAXIOV -#else -#define MAXIOV 16 -#endif - - -/* - * Macros for testing file types. - */ - -#define ISDIR(st) (S_ISDIR((st).st_mode)) -#define ISREG(st) (S_ISREG((st).st_mode)) -#define ISDEV(st) (S_ISCHR((st).st_mode) || S_ISBLK((st).st_mode)) -#define ISLNK(st) (S_ISLNK((st).st_mode)) -#ifdef NO_UMASK -#define FILE_MODE 0644 -#define DIR_MODE 0755 -#else -#define FILE_MODE 0666 -#define DIR_MODE 0777 -#endif - -#define IS_DOT_OR_DOTDOT(s) \ - (s[0] == '.' && (s[1] == '\0' || (s[1] == '.' && s[2] == '\0'))) - -static int check_error(int result, Efile_error* errInfo); - -static int -check_error(int result, Efile_error *errInfo) -{ - if (result < 0) { - errInfo->posix_errno = errInfo->os_errno = errno; - return 0; - } - return 1; -} - -int -efile_init() { - return 1; -} - -int -efile_mkdir(Efile_error* errInfo, /* Where to return error codes. */ - char* name) /* Name of directory to create. */ -{ -#ifdef NO_MKDIR_MODE - return check_error(mkdir(name), errInfo); -#else - return check_error(mkdir(name, DIR_MODE), errInfo); -#endif -} - -int -efile_rmdir(Efile_error* errInfo, /* Where to return error codes. */ - char* name) /* Name of directory to delete. */ -{ - if (rmdir(name) == 0) { - return 1; - } - if (errno == ENOTEMPTY) { - errno = EEXIST; - } - if (errno == EEXIST) { - int saved_errno = errno; - struct stat file_stat; - struct stat cwd_stat; - - /* - * The error code might be wrong if this is the current directory. - */ - - if (stat(name, &file_stat) == 0 && stat(".", &cwd_stat) == 0 && - file_stat.st_ino == cwd_stat.st_ino && - file_stat.st_dev == cwd_stat.st_dev) { - saved_errno = EINVAL; - } - errno = saved_errno; - } - return check_error(-1, errInfo); -} - -int -efile_delete_file(Efile_error* errInfo, /* Where to return error codes. */ - char* name) /* Name of file to delete. */ -{ - if (unlink(name) == 0) { - return 1; - } - if (errno == EISDIR) { /* Linux sets the wrong error code. */ - errno = EPERM; - } - return check_error(-1, errInfo); -} - -/* - *--------------------------------------------------------------------------- - * - * Changes the name of an existing file or directory, from src to dst. - * If src and dst refer to the same file or directory, does nothing - * and returns success. Otherwise if dst already exists, it will be - * deleted and replaced by src subject to the following conditions: - * If src is a directory, dst may be an empty directory. - * If src is a file, dst may be a file. - * In any other situation where dst already exists, the rename will - * fail. - * - * Results: - * If the directory was successfully created, returns 1. - * Otherwise the return value is 0 and errno is set to - * indicate the error. Some possible values for errno are: - * - * EACCES: src or dst parent directory can't be read and/or written. - * EEXIST: dst is a non-empty directory. - * EINVAL: src is a root directory or dst is a subdirectory of src. - * EISDIR: dst is a directory, but src is not. - * ENOENT: src doesn't exist, or src or dst is "". - * ENOTDIR: src is a directory, but dst is not. - * EXDEV: src and dst are on different filesystems. - * - * Side effects: - * The implementation of rename may allow cross-filesystem renames, - * but the caller should be prepared to emulate it with copy and - * delete if errno is EXDEV. - * - *--------------------------------------------------------------------------- - */ - -int -efile_rename(Efile_error* errInfo, /* Where to return error codes. */ - char* src, /* Original name. */ - char* dst) /* New name. */ -{ - if (rename(src, dst) == 0) { - return 1; - } - if (errno == ENOTEMPTY) { - errno = EEXIST; - } -#if defined (sparc) - /* - * SunOS 4.1.4 reports overwriting a non-empty directory with a - * directory as EINVAL instead of EEXIST (first rule out the correct - * EINVAL result code for moving a directory into itself). Must be - * conditionally compiled because realpath() is only defined on SunOS. - */ - - if (errno == EINVAL) { - char srcPath[MAXPATHLEN], dstPath[MAXPATHLEN]; - DIR *dirPtr; - struct dirent *dirEntPtr; - -#ifdef PURIFY - memset(srcPath, '\0', sizeof(srcPath)); - memset(dstPath, '\0', sizeof(dstPath)); -#endif - - if ((realpath(src, srcPath) != NULL) - && (realpath(dst, dstPath) != NULL) - && (strncmp(srcPath, dstPath, strlen(srcPath)) != 0)) { - dirPtr = opendir(dst); - if (dirPtr != NULL) { - while ((dirEntPtr = readdir(dirPtr)) != NULL) { - if ((strcmp(dirEntPtr->d_name, ".") != 0) && - (strcmp(dirEntPtr->d_name, "..") != 0)) { - errno = EEXIST; - closedir(dirPtr); - return check_error(-1, errInfo); - } - } - closedir(dirPtr); - } - } - errno = EINVAL; - } -#endif /* sparc */ - - if (strcmp(src, "/") == 0) { - /* - * Alpha reports renaming / as EBUSY and Linux reports it as EACCES, - * instead of EINVAL. - */ - - errno = EINVAL; - } - - /* - * DEC Alpha OSF1 V3.0 returns EACCES when attempting to move a - * file across filesystems and the parent directory of that file is - * not writable. Most other systems return EXDEV. Does nothing to - * correct this behavior. - */ - - return check_error(-1, errInfo); -} - -int -efile_chdir(Efile_error* errInfo, /* Where to return error codes. */ - char* name) /* Name of directory to make current. */ -{ - return check_error(chdir(name), errInfo); -} - - -int -efile_getdcwd(Efile_error* errInfo, /* Where to return error codes. */ - int drive, /* 0 - current, 1 - A, 2 - B etc. */ - char* buffer, /* Where to return the current - directory. */ - size_t size) /* Size of buffer. */ -{ - if (drive == 0) { - if (getcwd(buffer, size) == NULL) - return check_error(-1, errInfo); - -#ifdef SIMSPARCSOLARIS - /* We get "host:" prepended to the dirname - remove!. */ - { - int i = 0; - int j = 0; - while ((buffer[i] != ':') && (buffer[i] != '\0')) i++; - if (buffer[i] == ':') { - i++; - while ((buffer[j++] = buffer[i++]) != '\0'); - } - } -#endif - return 1; - } - - /* - * Drives other than 0 is not supported on Unix. - */ - - errno = ENOTSUP; - return check_error(-1, errInfo); -} - -int -efile_readdir(Efile_error* errInfo, /* Where to return error codes. */ - char* name, /* Name of directory to open. */ - EFILE_DIR_HANDLE* p_dir_handle, /* Pointer to directory - handle of - open directory.*/ - char* buffer, /* Pointer to buffer for - one filename. */ - size_t *size) /* in-out Size of buffer, length - of name. */ -{ - DIR *dp; /* Pointer to directory structure. */ - struct dirent* dirp; /* Pointer to directory entry. */ - - /* - * If this is the first call, we must open the directory. - */ - - if (*p_dir_handle == NULL) { - dp = opendir(name); - if (dp == NULL) - return check_error(-1, errInfo); - *p_dir_handle = (EFILE_DIR_HANDLE) dp; - } - - /* - * Retrieve the name of the next file using the directory handle. - */ - - dp = *((DIR **)((void *)p_dir_handle)); - for (;;) { - dirp = readdir(dp); - if (dirp == NULL) { - closedir(dp); - return 0; - } - if (IS_DOT_OR_DOTDOT(dirp->d_name)) - continue; - buffer[0] = '\0'; - strncat(buffer, dirp->d_name, (*size)-1); - *size = strlen(dirp->d_name); - return 1; - } -} - -int -efile_openfile(Efile_error* errInfo, /* Where to return error codes. */ - char* name, /* Name of directory to open. */ - int flags, /* Flags to user for opening. */ - int* pfd, /* Where to store the file - descriptor. */ - Sint64 *pSize) /* Where to store the size of the - file. */ -{ - struct stat statbuf; - int fd; - int mode; /* Open mode. */ - - switch (flags & (EFILE_MODE_READ|EFILE_MODE_WRITE)) { - case EFILE_MODE_READ: - mode = O_RDONLY; - break; - case EFILE_MODE_WRITE: - if (flags & EFILE_NO_TRUNCATE) - mode = O_WRONLY | O_CREAT; - else - mode = O_WRONLY | O_CREAT | O_TRUNC; - break; - case EFILE_MODE_READ_WRITE: - mode = O_RDWR | O_CREAT; - break; - default: - errno = EINVAL; - return check_error(-1, errInfo); - } - - if (flags & EFILE_MODE_APPEND) { - mode &= ~O_TRUNC; - mode |= O_APPEND; - } - if (flags & EFILE_MODE_EXCL) { - mode |= O_EXCL; - } - if (flags & EFILE_MODE_SYNC) { -#ifdef O_SYNC - mode |= O_SYNC; -#else - errno = ENOTSUP; - return check_error(-1, errInfo); -#endif - } - -#ifdef HAVE_FSTAT - while (((fd = open(name, mode, FILE_MODE)) < 0) && (errno == EINTR)); - if (!check_error(fd, errInfo)) return 0; -#endif - - if ( -#ifdef HAVE_FSTAT - fstat(fd, &statbuf) < 0 -#else - stat(name, &statbuf) < 0 -#endif - ) { - /* statbuf is undefined: if the caller depends on it, - i.e. invoke_read_file(), fail the call immediately */ - if (pSize && flags == EFILE_MODE_READ) { - check_error(-1, errInfo); -#ifdef HAVE_FSTAT - efile_closefile(fd); -#endif - return 0; - } - } - else if (! ISREG(statbuf)) { - struct stat nullstatbuf; - /* - * For UNIX only, here is some ugly code to allow - * /dev/null to be opened as a file. - */ - if ( (stat("/dev/null", &nullstatbuf) < 0) - || (statbuf.st_ino != nullstatbuf.st_ino) - || (statbuf.st_dev != nullstatbuf.st_dev) ) { -#ifdef HAVE_FSTAT - efile_closefile(fd); -#endif - errno = EISDIR; - return check_error(-1, errInfo); - } - } - -#ifndef HAVE_FSTAT - while (((fd = open(name, mode, FILE_MODE)) < 0) && (errno == EINTR)); - if (!check_error(fd, errInfo)) return 0; -#endif - - *pfd = fd; - if (pSize) *pSize = statbuf.st_size; - return 1; -} - -int -efile_may_openfile(Efile_error* errInfo, char *name) { - struct stat statbuf; /* Information about the file */ - int result; - - result = stat(name, &statbuf); - if (!check_error(result, errInfo)) - return 0; - if (!ISREG(statbuf)) { - errno = EISDIR; - return check_error(-1, errInfo); - } - return 1; -} - -void -efile_closefile(int fd) -{ - close(fd); -} - -int -efile_fdatasync(Efile_error *errInfo, /* Where to return error codes. */ - int fd) /* File descriptor for file to sync data. */ -{ -#if defined(HAVE_FDATASYNC) && !defined(__DARWIN__) - return check_error(fdatasync(fd), errInfo); -#else - return efile_fsync(errInfo, fd); -#endif -} - -int -efile_fsync(Efile_error *errInfo, /* Where to return error codes. */ - int fd) /* File descriptor for file to sync. */ -{ -#ifdef NO_FSYNC - undefined fsync /* XXX: Really? */ -#else -#if defined(__DARWIN__) && defined(F_FULLFSYNC) - return check_error(fcntl(fd, F_FULLFSYNC), errInfo); -#else - return check_error(fsync(fd), errInfo); -#endif /* __DARWIN__ */ -#endif /* NO_FSYNC */ -} - -int -efile_fileinfo(Efile_error* errInfo, Efile_info* pInfo, - char* name, int info_for_link) -{ - struct stat statbuf; /* Information about the file */ - int result; - - if (info_for_link) { - result = lstat(name, &statbuf); - } else { - result = stat(name, &statbuf); - } - if (!check_error(result, errInfo)) { - return 0; - } - -#if SIZEOF_OFF_T == 4 - pInfo->size_high = 0; -#else - pInfo->size_high = (Uint32)(statbuf.st_size >> 32); -#endif - pInfo->size_low = (Uint32)statbuf.st_size; - -#ifdef NO_ACCESS - /* Just look at read/write access for owner. */ - - pInfo->access = ((statbuf.st_mode >> 6) & 07) >> 1; - -#else - pInfo->access = FA_NONE; - if (access(name, R_OK) == 0) - pInfo->access |= FA_READ; - if (access(name, W_OK) == 0) - pInfo->access |= FA_WRITE; - -#endif - - if (ISDEV(statbuf)) - pInfo->type = FT_DEVICE; - else if (ISDIR(statbuf)) - pInfo->type = FT_DIRECTORY; - else if (ISREG(statbuf)) - pInfo->type = FT_REGULAR; - else if (ISLNK(statbuf)) - pInfo->type = FT_SYMLINK; - else - pInfo->type = FT_OTHER; - - pInfo->accessTime = (Sint64)statbuf.st_atime; - pInfo->modifyTime = (Sint64)statbuf.st_mtime; - pInfo->cTime = (Sint64)statbuf.st_ctime; - - pInfo->mode = statbuf.st_mode; - pInfo->links = statbuf.st_nlink; - pInfo->major_device = statbuf.st_dev; - pInfo->minor_device = statbuf.st_rdev; - pInfo->inode = statbuf.st_ino; - pInfo->uid = statbuf.st_uid; - pInfo->gid = statbuf.st_gid; - - return 1; -} - -int -efile_write_info(Efile_error *errInfo, Efile_info *pInfo, char *name) -{ - struct utimbuf tval; - - /* - * On some systems chown will always fail for a non-root user unless - * POSIX_CHOWN_RESTRICTED is not set. Others will succeed as long as - * you don't try to chown a file to someone besides youself. - */ - - if (chown(name, pInfo->uid, pInfo->gid) && errno != EPERM) { - return check_error(-1, errInfo); - } - - if (pInfo->mode != -1) { - mode_t newMode = pInfo->mode & (S_ISUID | S_ISGID | - S_IRWXU | S_IRWXG | S_IRWXO); - if (chmod(name, newMode)) { - newMode &= ~(S_ISUID | S_ISGID); - if (chmod(name, newMode)) { - return check_error(-1, errInfo); - } - } - } - - tval.actime = (time_t)pInfo->accessTime; - tval.modtime = (time_t)pInfo->modifyTime; - - return check_error(utime(name, &tval), errInfo); -} - - -int -efile_write(Efile_error* errInfo, /* Where to return error codes. */ - int flags, /* Flags given when file was - opened. */ - int fd, /* File descriptor to write to. */ - char* buf, /* Buffer to write. */ - size_t count) /* Number of bytes to write. */ -{ - ssize_t written; /* Bytes written in last operation. */ - - while (count > 0) { - if ((written = write(fd, buf, count)) < 0) { - if (errno != EINTR) - return check_error(-1, errInfo); - else - written = 0; - } - ASSERT(written <= count); - buf += written; - count -= written; - } - return 1; -} - -int -efile_writev(Efile_error* errInfo, /* Where to return error codes */ - int flags, /* Flags given when file was - * opened */ - int fd, /* File descriptor to write to */ - SysIOVec* iov, /* Vector of buffer structs. - * The structs may be changed i.e. - * due to incomplete writes */ - int iovcnt) /* Number of structs in vector */ -{ - int cnt = 0; /* Buffers so far written */ - - ASSERT(iovcnt >= 0); - - while (cnt < iovcnt) { - if ((! iov[cnt].iov_base) || (iov[cnt].iov_len <= 0)) { - /* Empty buffer - skip */ - cnt++; - } else { /* Non-empty buffer */ - ssize_t w; /* Bytes written in this call */ -#ifdef HAVE_WRITEV - int b = iovcnt - cnt; /* Buffers to write */ - /* Use as many buffers as MAXIOV allows */ - if (b > MAXIOV) - b = MAXIOV; - if (b > 1) { - do { - w = writev(fd, &iov[cnt], b); - } while (w < 0 && errno == EINTR); - if (w < 0 && errno == EINVAL) { - goto single_write; - } - } else - single_write: - /* Degenerated io vector - use regular write */ -#endif - { - do { - size_t iov_len = iov[cnt].iov_len; - size_t limit = 1024*1024*1024; /* 1GB */ - if (iov_len > limit) { - iov_len = limit; - } - w = write(fd, iov[cnt].iov_base, iov_len); - } while (w < 0 && errno == EINTR); - ASSERT(w <= iov[cnt].iov_len || - (w == -1 && errno != EINTR)); - } - if (w < 0) return check_error(-1, errInfo); - /* Move forward to next buffer to write */ - for (; cnt < iovcnt && w > 0; cnt++) { - if (iov[cnt].iov_base && iov[cnt].iov_len > 0) { - if (w < iov[cnt].iov_len) { - /* Adjust the buffer for next write */ - iov[cnt].iov_len -= w; - iov[cnt].iov_base = ((char *)iov[cnt].iov_base) + w; - w = 0; - break; - } else { - w -= iov[cnt].iov_len; - } - } - } - ASSERT(w == 0); - } /* else Non-empty buffer */ - } /* while (cnt< iovcnt) */ - return 1; -} - -int -efile_read(Efile_error* errInfo, /* Where to return error codes. */ - int flags, /* Flags given when file was opened. */ - int fd, /* File descriptor to read from. */ - char* buf, /* Buffer to read into. */ - size_t count, /* Number of bytes to read. */ - size_t *pBytesRead) /* Where to return number of - bytes read. */ -{ - ssize_t n; - - for (;;) { - if ((n = read(fd, buf, count)) >= 0) - break; - else if (errno != EINTR) - return check_error(-1, errInfo); - } - *pBytesRead = (size_t) n; - return 1; -} - - -/* pread() and pwrite() */ -/* Some unix systems, notably Solaris has these syscalls */ -/* It is especially nice for i.e. the dets module to have support */ -/* for this, even if the underlying OS dosn't support it, it is */ -/* reasonably easy to work around by first calling seek, and then */ -/* calling read(). */ -/* This later strategy however changes the file pointer, which pread() */ -/* does not do. We choose to ignore this and say that the location */ -/* of the file pointer is undefined after a call to any of the p functions*/ - - -int -efile_pread(Efile_error* errInfo, /* Where to return error codes. */ - int fd, /* File descriptor to read from. */ - Sint64 offset, /* Offset in bytes from BOF. */ - char* buf, /* Buffer to read into. */ - size_t count, /* Number of bytes to read. */ - size_t *pBytesRead) /* Where to return - number of bytes read. */ -{ -#if defined(HAVE_PREAD) && defined(HAVE_PWRITE) - ssize_t n; - off_t off = (off_t) offset; - if (off != offset) { - errno = EINVAL; - return check_error(-1, errInfo); - } - for (;;) { - if ((n = pread(fd, buf, count, offset)) >= 0) - break; - else if (errno != EINTR) - return check_error(-1, errInfo); - } - *pBytesRead = (size_t) n; - return 1; -#else - { - int res = efile_seek(errInfo, fd, offset, EFILE_SEEK_SET, NULL); - if (res) { - return efile_read(errInfo, 0, fd, buf, count, pBytesRead); - } else { - return res; - } - } -#endif -} - - - -int -efile_pwrite(Efile_error* errInfo, /* Where to return error codes. */ - int fd, /* File descriptor to write to. */ - char* buf, /* Buffer to write. */ - size_t count, /* Number of bytes to write. */ - Sint64 offset) /* where to write it */ -{ -#if defined(HAVE_PREAD) && defined(HAVE_PWRITE) - ssize_t written; /* Bytes written in last operation. */ - off_t off = (off_t) offset; - if (off != offset) { - errno = EINVAL; - return check_error(-1, errInfo); - } - - while (count > 0) { - if ((written = pwrite(fd, buf, count, offset)) < 0) { - if (errno != EINTR) - return check_error(-1, errInfo); - else - written = 0; - } - ASSERT(written <= count); - buf += written; - count -= written; - offset += written; - } - return 1; -#else /* For unix systems that don't support pread() and pwrite() */ - { - int res = efile_seek(errInfo, fd, offset, EFILE_SEEK_SET, NULL); - - if (res) { - return efile_write(errInfo, 0, fd, buf, count); - } else { - return res; - } - } -#endif -} - - -int -efile_seek(Efile_error* errInfo, /* Where to return error codes. */ - int fd, /* File descriptor to do the seek on. */ - Sint64 offset, /* Offset in bytes from the given - origin. */ - int origin, /* Origin of seek (SEEK_SET, SEEK_CUR, - SEEK_END). */ - Sint64 *new_location) /* Resulting new location in file. */ -{ - off_t off, result; - - switch (origin) { - case EFILE_SEEK_SET: origin = SEEK_SET; break; - case EFILE_SEEK_CUR: origin = SEEK_CUR; break; - case EFILE_SEEK_END: origin = SEEK_END; break; - default: - errno = EINVAL; - return check_error(-1, errInfo); - } - off = (off_t) offset; - if (off != offset) { - errno = EINVAL; - return check_error(-1, errInfo); - } - - errno = 0; - result = lseek(fd, off, origin); - - /* - * Note that the man page for lseek (on SunOs 5) says: - * - * "if fildes is a remote file descriptor and offset is - * negative, lseek() returns the file pointer even if it is - * negative." - */ - - if (result < 0 && errno == 0) - errno = EINVAL; - if (result < 0) - return check_error(-1, errInfo); - if (new_location) { - *new_location = result; - } - return 1; -} - - -int -efile_truncate_file(Efile_error* errInfo, int *fd, int flags) -{ -#ifndef NO_FTRUNCATE - off_t offset; - - return check_error((offset = lseek(*fd, 0, 1)) >= 0 && - ftruncate(*fd, offset) == 0 ? 1 : -1, - errInfo); -#else - return 1; -#endif -} - -int -efile_readlink(Efile_error* errInfo, char* name, char* buffer, size_t size) -{ - int len; - ASSERT(size > 0); - len = readlink(name, buffer, size-1); - if (len == -1) { - return check_error(-1, errInfo); - } - buffer[len] = '\0'; - return 1; -} - -int -efile_altname(Efile_error* errInfo, char* name, char* buffer, size_t size) -{ - errno = ENOTSUP; - return check_error(-1, errInfo); -} - -int -efile_link(Efile_error* errInfo, char* old, char* new) -{ - return check_error(link(old, new), errInfo); -} - -int -efile_symlink(Efile_error* errInfo, char* old, char* new) -{ - return check_error(symlink(old, new), errInfo); -} - -int -efile_fadvise(Efile_error* errInfo, int fd, Sint64 offset, - Sint64 length, int advise) -{ -#ifdef HAVE_POSIX_FADVISE - return check_error(posix_fadvise(fd, offset, length, advise), errInfo); -#else - return check_error(0, errInfo); -#endif -} - -#ifdef HAVE_SENDFILE -/* For some reason the maximum size_t cannot be used as the max size - 3GB seems to work on all platforms */ -#define SENDFILE_CHUNK_SIZE ((1UL << 30) -1) - -/* - * sendfile: The implementation of the sendfile system call varies - * a lot on different *nix platforms so to make the api similar in all - * we have to emulate some things in linux and play with variables on - * bsd/darwin. - * - * All of the calls will split a command which tries to send more than - * SENDFILE_CHUNK_SIZE of data at once. - * - * On platforms where *nbytes of 0 does not mean the entire file, this is - * simulated. - * - * It could be possible to implement header/trailer in sendfile. Though - * you would have to emulate it in linux and on BSD/Darwin some complex - * calculations have to be made when using a non blocking socket to figure - * out how much of the header/file/trailer was sent in each command. - * - * The semantics of the API is this: - * Return value: 1 if all data was sent and the function does not need to - * be called again. 0 if an error occures OR if there is more data which - * has to be sent (EAGAIN or EINTR will be set appropriately) - * - * The amount of data written in a call is returned through nbytes. - * - */ - -int -efile_sendfile(Efile_error* errInfo, int in_fd, int out_fd, - off_t *offset, Uint64 *nbytes, struct t_sendfile_hdtl* hdtl) -{ - Uint64 written = 0; -#if defined(__linux__) - ssize_t retval; - do { - /* check if *nbytes is 0 or greater than chunk size */ - if (*nbytes == 0 || *nbytes > SENDFILE_CHUNK_SIZE) - retval = sendfile(out_fd, in_fd, offset, SENDFILE_CHUNK_SIZE); - else - retval = sendfile(out_fd, in_fd, offset, *nbytes); - if (retval > 0) { - written += retval; - *nbytes -= retval; - } - } while (retval == SENDFILE_CHUNK_SIZE); - if (written != 0) { - /* -1 is not returned by the linux API so we have to simulate it */ - retval = -1; - errno = EAGAIN; - } -#elif defined(__sun) && defined(__SVR4) && defined(HAVE_SENDFILEV) - ssize_t retval; - size_t len; - sendfilevec_t fdrec; - fdrec.sfv_fd = in_fd; - fdrec.sfv_flag = 0; - do { - fdrec.sfv_off = *offset; - len = 0; - /* check if *nbytes is 0 or greater than chunk size */ - if (*nbytes == 0 || *nbytes > SENDFILE_CHUNK_SIZE) - fdrec.sfv_len = SENDFILE_CHUNK_SIZE; - else - fdrec.sfv_len = *nbytes; - - retval = sendfilev(out_fd, &fdrec, 1, &len); - - if (retval == -1 && errno == EINVAL) { - /* On some solaris versions (I've seen it on SunOS 5.10), - using a sfv_len larger then a filesize will result in - a -1 && errno == EINVAL return. We translate this so - a successful send of the data.*/ - retval = len; - } - - if (retval != -1 || errno == EAGAIN || errno == EINTR) { - *offset += len; - *nbytes -= len; - written += len; - } - } while (len == SENDFILE_CHUNK_SIZE); -#elif defined(__DARWIN__) - int retval; - off_t len; - do { - /* check if *nbytes is 0 or greater than chunk size */ - if(*nbytes > SENDFILE_CHUNK_SIZE) - len = SENDFILE_CHUNK_SIZE; - else - len = *nbytes; - retval = sendfile(in_fd, out_fd, *offset, &len, NULL, 0); - if (retval != -1 || errno == EAGAIN || errno == EINTR) { - *offset += len; - *nbytes -= len; - written += len; - } - } while (len == SENDFILE_CHUNK_SIZE); -#elif defined(__FreeBSD__) || defined(__DragonFly__) - off_t len; - int retval; - do { - if (*nbytes > SENDFILE_CHUNK_SIZE) - retval = sendfile(in_fd, out_fd, *offset, SENDFILE_CHUNK_SIZE, - NULL, &len, 0); - else - retval = sendfile(in_fd, out_fd, *offset, *nbytes, NULL, &len, 0); - if (retval != -1 || errno == EAGAIN || errno == EINTR) { - *offset += len; - *nbytes -= len; - written += len; - } - } while(len == SENDFILE_CHUNK_SIZE); -#endif - *nbytes = written; - return check_error(retval, errInfo); -} -#endif /* HAVE_SENDFILE */ - -#ifdef HAVE_POSIX_FALLOCATE -static int -call_posix_fallocate(int fd, Sint64 offset, Sint64 length) -{ - int ret; - - /* - * On Linux and Solaris for example, posix_fallocate() returns - * a positive error number on error and it does not set errno. - * On FreeBSD however (9.0 at least), it returns -1 on error - * and it sets errno. - */ - do { - ret = posix_fallocate(fd, (off_t) offset, (off_t) length); - if (ret > 0) { - errno = ret; - ret = -1; - } - } while (ret != 0 && errno == EINTR); - - return ret; -} -#endif /* HAVE_POSIX_FALLOCATE */ - -int -efile_fallocate(Efile_error* errInfo, int fd, Sint64 offset, Sint64 length) -{ -#if defined HAVE_FALLOCATE - /* Linux specific, more efficient than posix_fallocate. */ - int ret; - - do { - ret = fallocate(fd, FALLOC_FL_KEEP_SIZE, (off_t) offset, (off_t) length); - } while (ret != 0 && errno == EINTR); - -#if defined HAVE_POSIX_FALLOCATE - /* Fallback to posix_fallocate if available. */ - if (ret != 0) { - ret = call_posix_fallocate(fd, offset, length); - } -#endif - - return check_error(ret, errInfo); -#elif defined F_PREALLOCATE - /* Mac OS X specific, equivalent to posix_fallocate. */ - int ret; - fstore_t fs; - - memset(&fs, 0, sizeof(fs)); - fs.fst_flags = F_ALLOCATECONTIG; - fs.fst_posmode = F_VOLPOSMODE; - fs.fst_offset = (off_t) offset; - fs.fst_length = (off_t) length; - - ret = fcntl(fd, F_PREALLOCATE, &fs); - - if (-1 == ret) { - fs.fst_flags = F_ALLOCATEALL; - ret = fcntl(fd, F_PREALLOCATE, &fs); - -#if defined HAVE_POSIX_FALLOCATE - /* Fallback to posix_fallocate if available. */ - if (-1 == ret) { - ret = call_posix_fallocate(fd, offset, length); - } -#endif - } - - return check_error(ret, errInfo); -#elif defined HAVE_POSIX_FALLOCATE - /* Other Unixes, use posix_fallocate if available. */ - return check_error(call_posix_fallocate(fd, offset, length), errInfo); -#else - errno = ENOTSUP; - return check_error(-1, errInfo); -#endif -} diff --git a/erts/emulator/drivers/win32/ttsl_drv.c b/erts/emulator/drivers/win32/ttsl_drv.c index 99e7fb25a4..d19bfa3079 100644 --- a/erts/emulator/drivers/win32/ttsl_drv.c +++ b/erts/emulator/drivers/win32/ttsl_drv.c @@ -176,6 +176,8 @@ static ErlDrvSSizeT ttysl_control(ErlDrvData drv_data, { char resbuff[2*sizeof(Uint32)]; ErlDrvSizeT res_size; + + command -= ERTS_TTYSL_DRV_CONTROL_MAGIC_NUMBER; switch (command) { case CTRL_OP_GET_WINSIZE: { @@ -201,7 +203,7 @@ static ErlDrvSSizeT ttysl_control(ErlDrvData drv_data, } break; default: - return 0; + return -1; } if (rlen < res_size) { *rbuf = driver_alloc(res_size); diff --git a/erts/emulator/drivers/win32/win_efile.c b/erts/emulator/drivers/win32/win_efile.c deleted file mode 100644 index 2d366b5833..0000000000 --- a/erts/emulator/drivers/win32/win_efile.c +++ /dev/null @@ -1,2058 +0,0 @@ -/* - * %CopyrightBegin% - * - * Copyright Ericsson AB 1997-2016. 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. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - * - * %CopyrightEnd% - */ -/* - * Purpose: Provides file and directory operations for Windows. - */ - -#include <windows.h> -#ifdef HAVE_CONFIG_H -# include "config.h" -#endif -#include "sys.h" -#include <ctype.h> -#include <wchar.h> -#include "erl_efile.h" - -#define DBG_TRACE_MASK 0 -/* 1 = file name ops - * 2 = file descr ops - * 4 = errors - * 8 = path name conversion - */ -#if !DBG_TRACE_MASK -# define DBG_TRACE(M,S) -# define DBG_TRACE1(M,FMT,A) -# define DBG_TRACE2(M,FMT,A,B) -#else -# define DBG_TRACE(M,S) do { if ((M)&DBG_TRACE_MASK) fwprintf(stderr, L"DBG_TRACE %d: %s\r\n", __LINE__, (WCHAR*)(S)); }while(0) -# define DBG_TRACE1(M,FMT,A) do { if ((M)&DBG_TRACE_MASK) fwprintf(stderr, L"DBG_TRACE %d: " L##FMT L"\r\n", __LINE__, (A)); }while(0) -# define DBG_TRACE2(M,FMT,A,B) do { if ((M)&DBG_TRACE_MASK) fwprintf(stderr, L"DBG_TRACE %d: " L##FMT L"\r\n", __LINE__, (A), (B)); }while(0) -#endif - -/* - * Microsoft-specific function to map a WIN32 error code to a Posix errno. - */ - -#define ISSLASH(a) ((a) == L'\\' || (a) == L'/') -#define ISDIR(st) (((st).st_mode&S_IFMT) == S_IFDIR) -#define ISREG(st) (((st).st_mode&S_IFMT) == S_IFREG) - -#define IS_DOT_OR_DOTDOT(s) \ - ((s)[0] == L'.' && ((s)[1] == L'\0' || ((s)[1] == L'.' && (s)[2] == L'\0'))) - -#define FILE_SHARE_FLAGS (FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE) - -#ifndef INVALID_FILE_ATTRIBUTES -#define INVALID_FILE_ATTRIBUTES ((DWORD) 0xFFFFFFFF) -#endif - -#define TICKS_PER_SECOND (10000000ULL) -#define EPOCH_DIFFERENCE (11644473600LL) - -#define FILETIME_TO_EPOCH(epoch, ft) \ - do { \ - ULARGE_INTEGER ull; \ - ull.LowPart = (ft).dwLowDateTime; \ - ull.HighPart = (ft).dwHighDateTime; \ - (epoch) = ((ull.QuadPart / TICKS_PER_SECOND) - EPOCH_DIFFERENCE); \ - } while(0) - -#define EPOCH_TO_FILETIME(ft, epoch) \ - do { \ - ULARGE_INTEGER ull; \ - ull.QuadPart = (((epoch) + EPOCH_DIFFERENCE) * TICKS_PER_SECOND); \ - (ft).dwLowDateTime = ull.LowPart; \ - (ft).dwHighDateTime = ull.HighPart; \ - } while(0) - - -static int check_error(int result, Efile_error* errInfo); -static int set_error(Efile_error* errInfo); -static int set_os_errno(Efile_error* errInfo, DWORD os_errno); -static int is_root_unc_name(const WCHAR *path); -static int extract_root(WCHAR *name); -static unsigned short dos_to_posix_mode(int attr, const WCHAR *name); - - -struct wpath_tmp_buffer { - struct wpath_tmp_buffer* next; - WCHAR buffer[1]; -}; - -typedef struct { - Efile_error* errInfo; - struct wpath_tmp_buffer* buf_list; -}Efile_call_state; - -static void call_state_init(Efile_call_state* state, Efile_error* errInfo) -{ - state->errInfo = errInfo; - state->buf_list = NULL; -} -static WCHAR* wpath_tmp_alloc(Efile_call_state* state, size_t len) -{ - size_t sz = offsetof(struct wpath_tmp_buffer, buffer) - + (len+1)*sizeof(WCHAR); - struct wpath_tmp_buffer* p = driver_alloc(sz); - p->next = state->buf_list; - state->buf_list = p; - return p->buffer; -} -static void call_state_free(Efile_call_state* state) -{ - while(state->buf_list) { - struct wpath_tmp_buffer* next = state->buf_list->next; - driver_free(state->buf_list); - state->buf_list = next; - } -} -static WCHAR* get_cwd_wpath_tmp(Efile_call_state* state) -{ - WCHAR dummy; - DWORD size = GetCurrentDirectoryW(0, &dummy); - WCHAR* ret = NULL; - - if (size) { - ret = wpath_tmp_alloc(state, size); - if (!GetCurrentDirectoryW(size, ret)) { - ret = NULL; - } - } - return ret; -} -static WCHAR* get_full_wpath_tmp(Efile_call_state* state, - const WCHAR* file, - WCHAR** file_part, - DWORD extra) -{ - WCHAR dummy; - DWORD size = GetFullPathNameW(file, 0, &dummy, NULL); - WCHAR* ret = NULL; - - if (size) { - int ok; - ret = wpath_tmp_alloc(state, size + extra); - if (file_part) { - ok = (GetFullPathNameW(file, size, ret, file_part) != 0); - } - else { - ok = (_wfullpath(ret, file, size) != NULL); - } - if (!ok) { - ret = NULL; - } - } - return ret; -} - -static void ensure_wpath_max(Efile_call_state* state, WCHAR** pathp, size_t max); -static int do_rmdir(Efile_call_state*, char* name); -static int do_rename(Efile_call_state*, char* src, char* dst); -static int do_readdir(Efile_call_state*, char* name, EFILE_DIR_HANDLE*, char* buffer, size_t *size); -static int do_fileinfo(Efile_call_state*, Efile_info*, char* orig_name, int info_for_link); -static char* do_readlink(Efile_call_state*, char* name, char* buffer, size_t size); -static int do_altname(Efile_call_state*, char* orig_name, char* buffer, size_t size); - - -static int errno_map(DWORD last_error) { - - switch (last_error) { - case ERROR_SUCCESS: - return 0; - case ERROR_INVALID_FUNCTION: - case ERROR_INVALID_DATA: - case ERROR_INVALID_PARAMETER: - case ERROR_INVALID_TARGET_HANDLE: - case ERROR_INVALID_CATEGORY: - case ERROR_NEGATIVE_SEEK: - return EINVAL; - case ERROR_DIR_NOT_EMPTY: - return EEXIST; - case ERROR_BAD_FORMAT: - return ENOEXEC; - case ERROR_PATH_NOT_FOUND: - case ERROR_FILE_NOT_FOUND: - case ERROR_NO_MORE_FILES: - return ENOENT; - case ERROR_TOO_MANY_OPEN_FILES: - return EMFILE; - case ERROR_ACCESS_DENIED: - case ERROR_INVALID_ACCESS: - case ERROR_CURRENT_DIRECTORY: - case ERROR_SHARING_VIOLATION: - case ERROR_LOCK_VIOLATION: - case ERROR_INVALID_PASSWORD: - case ERROR_DRIVE_LOCKED: - return EACCES; - case ERROR_INVALID_HANDLE: - return EBADF; - case ERROR_NOT_ENOUGH_MEMORY: - case ERROR_OUTOFMEMORY: - case ERROR_OUT_OF_STRUCTURES: - return ENOMEM; - case ERROR_INVALID_DRIVE: - case ERROR_BAD_UNIT: - case ERROR_NOT_READY: - case ERROR_REM_NOT_LIST: - case ERROR_DUP_NAME: - case ERROR_BAD_NETPATH: - case ERROR_NETWORK_BUSY: - case ERROR_DEV_NOT_EXIST: - case ERROR_BAD_NET_NAME: - return ENXIO; - case ERROR_NOT_SAME_DEVICE: - return EXDEV; - case ERROR_WRITE_PROTECT: - return EROFS; - case ERROR_BAD_LENGTH: - case ERROR_BUFFER_OVERFLOW: - return E2BIG; - case ERROR_SEEK: - case ERROR_SECTOR_NOT_FOUND: - return ESPIPE; - case ERROR_NOT_DOS_DISK: - return ENODEV; - case ERROR_GEN_FAILURE: - return ENODEV; - case ERROR_SHARING_BUFFER_EXCEEDED: - case ERROR_NO_MORE_SEARCH_HANDLES: - return EMFILE; - case ERROR_HANDLE_EOF: - case ERROR_BROKEN_PIPE: - return EPIPE; - case ERROR_HANDLE_DISK_FULL: - case ERROR_DISK_FULL: - return ENOSPC; - case ERROR_NOT_SUPPORTED: - return ENOTSUP; - case ERROR_FILE_EXISTS: - case ERROR_ALREADY_EXISTS: - case ERROR_CANNOT_MAKE: - return EEXIST; - case ERROR_ALREADY_ASSIGNED: - return EBUSY; - case ERROR_NO_PROC_SLOTS: - return EAGAIN; - case ERROR_CANT_RESOLVE_FILENAME: - return EMLINK; - case ERROR_PRIVILEGE_NOT_HELD: - return EPERM; - case ERROR_ARENA_TRASHED: - case ERROR_INVALID_BLOCK: - case ERROR_BAD_ENVIRONMENT: - case ERROR_BAD_COMMAND: - case ERROR_CRC: - case ERROR_OUT_OF_PAPER: - case ERROR_READ_FAULT: - case ERROR_WRITE_FAULT: - case ERROR_WRONG_DISK: - case ERROR_NET_WRITE_FAULT: - return EIO; - default: /* not to do with files I expect. */ - return EIO; - } -} - -static int -check_error(int result, Efile_error* errInfo) -{ - if (result < 0) { - errInfo->posix_errno = errno; - errInfo->os_errno = GetLastError(); - DBG_TRACE2(4, "ERROR os_error=%d errno=%d @@@@@@@@@@@@@@@@@@@@@@@@@@@@", - errInfo->os_errno, errInfo->posix_errno); - return 0; - } - return 1; -} - -static void -save_last_error(Efile_error* errInfo) -{ - errInfo->posix_errno = errno; - errInfo->os_errno = GetLastError(); - DBG_TRACE2(4, "ERROR os_error=%d errno=%d $$$$$$$$$$$$$$$$$$$$$$$$$$$$$", - errInfo->os_errno, errInfo->posix_errno); -} - - -/* - * Fills the provided error information structure with information - * with the error code given by GetLastError() and its corresponding - * Posix error number. - * - * Returns 0. - */ - -static int -set_error(Efile_error* errInfo) -{ - set_os_errno(errInfo, GetLastError()); - return 0; -} - - -static int -set_os_errno(Efile_error* errInfo, DWORD os_errno) -{ - errInfo->os_errno = os_errno; - errInfo->posix_errno = errno_map(os_errno); - DBG_TRACE2(4, "ERROR os_error=%d errno=%d ############################", - errInfo->os_errno, errInfo->posix_errno); - return 0; -} - -int -efile_init() { - return 1; -} - -/* - * A writev with Unix semantics, but with Windows arguments - */ -static int -win_writev(Efile_error* errInfo, - HANDLE fd, /* handle to file */ - FILE_SEGMENT_ELEMENT iov[], /* array of buffer pointers */ - DWORD *size) /* number of bytes to write */ -{ - OVERLAPPED ov; - ov.Offset = 0L; - ov.OffsetHigh = 0L; - ov.hEvent = CreateEvent(NULL, TRUE, FALSE, NULL); - if (ov.hEvent == NULL) - return set_error(errInfo); - if (! write_file_gather(fd, iov, *size, NULL, &ov)) - return set_error(errInfo); - if (WaitForSingleObject(ov.hEvent, INFINITE) != WAIT_OBJECT_0) - return set_error(errInfo); - if (! GetOverlappedResult(fd, &ov, size, FALSE)) - return set_error(errInfo); - return 1; -} - - -/* Check '*pathp' and convert it if needed to something that windows will accept. - * Typically use UNC path with \\?\ prefix if absolute path is longer than 260. - */ -static void ensure_wpath(Efile_call_state* state, WCHAR** pathp) -{ - ensure_wpath_max(state, pathp, MAX_PATH); -} - -static void ensure_wpath_max(Efile_call_state* state, WCHAR** pathp, size_t max) -{ - WCHAR* path = *pathp; - WCHAR* p; - size_t len = wcslen(path); - int unc_fixup = 0; - - if (path[0] == 0) { - DBG_TRACE(8, L"Let empty path pass through"); - return; - } - - DBG_TRACE1(8,"IN: %s", path); - - if (path[1] == L':' && ISSLASH(path[2])) { /* absolute path */ - if (len >= max) { - WCHAR *src, *dst; - - *pathp = wpath_tmp_alloc(state, 4+len+1); - dst = *pathp; - wcscpy(dst, L"\\\\?\\"); - for (src=path,dst+=4; *src; src++) { - if (*src == L'/') { - if (dst[-1] != L'\\') { - *dst++ = L'\\'; - } - /*else ignore redundant slashes */ - } - else - *dst++ = *src; - } - *dst = 0; - unc_fixup = 1; - } - } - else if (!(ISSLASH(path[0]) && ISSLASH(path[1]))) { /* relative path */ - DWORD cwdLen = GetCurrentDirectoryW(0, NULL); - DWORD absLen = cwdLen + 1 + len; - if (absLen >= max) { - WCHAR *fullPath = wpath_tmp_alloc(state, 4+4+absLen); - DWORD fullLen; - - fullLen = GetFullPathNameW(path, 4 + absLen, fullPath+4, NULL); - if (fullLen >= 4+absLen) { - *pathp = path; - DBG_TRACE2(8,"ensure_wpath FAILED absLen=%u %s", (int)absLen, path); - return; - } - /* GetFullPathNameW can return paths longer than MAX_PATH without the \\?\ prefix. - * At least seen on Windows 7. Go figure... - */ - if (fullLen >= max && wcsncmp(fullPath+4, L"\\\\?\\", 4) != 0) { - wcsncpy(fullPath, L"\\\\?\\", 4); - *pathp = fullPath; - } - else { - *pathp = fullPath + 4; - } - } - } - - if (unc_fixup) { - WCHAR* endp; - - p = *pathp; - len = wcslen(p); - endp = p + len; - if (len > 4) { - p += 4; - while (*p) { - if (p[0] == L'\\' && p[1] == L'.') { - if (p[2] == L'\\' || !p[2]) { /* single dot */ - wmemmove(p, p+2, (&endp[1] - &p[2])); - endp -= 2; - } - else if (p[2] == L'.' && (p[3] == L'\\' || !p[3])) { /* double dot */ - WCHAR* r; - for (r=p-1; *r == L'\\'; --r) - /*skip redundant slashes*/; - for (; *r != L'\\'; --r) - /*find start of prev directory*/; - if (r < *pathp + 6) - break; - wmemmove(r, p+3, (&endp[1] - &p[3])); - p = r; - } - else p += 3; - } - else ++p; - } - } - } - DBG_TRACE1(8,"OUT: %s", *pathp); -} - -int -efile_mkdir(Efile_error* errInfo, /* Where to return error codes. */ - char* name) /* Name of directory to create. */ -{ - Efile_call_state state; - WCHAR* wname = (WCHAR*)name; - int ret; - - DBG_TRACE(1, name); - call_state_init(&state, errInfo); - ensure_wpath_max(&state, &wname, 248); /* Yes, 248 limit for normal paths */ - - ret = (int) CreateDirectoryW(wname, NULL); - if (!ret) - set_error(errInfo); - - call_state_free(&state); - return ret; -} - -int -efile_rmdir(Efile_error* errInfo, /* Where to return error codes. */ - char* name) /* Name of directory to delete. */ -{ - Efile_call_state state; - int ret; - - DBG_TRACE(1, name); - call_state_init(&state, errInfo); - ret = do_rmdir(&state, name); - call_state_free(&state); - return ret; -} - -static int do_rmdir(Efile_call_state* state, char* name) -{ - OSVERSIONINFO os; - DWORD attr; - WCHAR *wname = (WCHAR *) name; - WCHAR *buffer = NULL; - - ensure_wpath(state, &wname); - - if (RemoveDirectoryW(wname) != FALSE) { - return 1; - } - errno = errno_map(GetLastError()); - if (errno == EACCES) { - attr = GetFileAttributesW(wname); - if (attr != (DWORD) -1) { - if ((attr & FILE_ATTRIBUTE_DIRECTORY) == 0) { - /* - * Windows 95 reports calling RemoveDirectory on a file as an - * EACCES, not an ENOTDIR. - */ - - errno = ENOTDIR; - goto end; - } - - /* - * Windows 95 reports removing a non-empty directory as - * an EACCES, not an EEXIST. If the directory is not empty, - * change errno so caller knows what's going on. - */ - - os.dwOSVersionInfoSize = sizeof(os); - GetVersionEx(&os); - if (os.dwPlatformId == VER_PLATFORM_WIN32_WINDOWS) { - HANDLE handle; - WIN32_FIND_DATAW data; - int len = wcslen(wname); - - buffer = wpath_tmp_alloc(state, len + 4); - wcscpy(buffer, wname); - if (buffer[0] && buffer[len-1] != L'\\' && buffer[len-1] != L'/') { - wcscat(buffer, L"\\"); - } - wcscat(buffer, L"*.*"); - handle = FindFirstFileW(buffer, &data); - if (handle != INVALID_HANDLE_VALUE) { - while (1) { - if ((wcscmp(data.cFileName, L".") != 0) - && (wcscmp(data.cFileName, L"..") != 0)) { - /* - * Found something in this directory. - */ - - errno = EEXIST; - break; - } - if (FindNextFileW(handle, &data) == FALSE) { - break; - } - } - FindClose(handle); - } - } - } - } - - if (errno == ENOTEMPTY) { - /* - * Posix allows both EEXIST or ENOTEMPTY, but we'll always - * return EEXIST to allow easy matching in Erlang code. - */ - - errno = EEXIST; - } - - end: - save_last_error(state->errInfo); - return 0; -} - -int -efile_delete_file(Efile_error* errInfo, /* Where to return error codes. */ - char* name) /* Name of file to delete. */ -{ - Efile_call_state state; - int ret; - DBG_TRACE(1, name); - call_state_init(&state, errInfo); - ret = do_delete_file(&state, name); - call_state_free(&state); - return ret; -} - -static int do_delete_file(Efile_call_state* state, char* name) -{ - DWORD attr; - WCHAR *wname = (WCHAR *) name; - - ensure_wpath(state, &wname); - - if (DeleteFileW(wname) != FALSE) { - return 1; - } - - errno = errno_map(GetLastError()); - if (errno == EACCES) { - attr = GetFileAttributesW(wname); - if (attr != (DWORD) -1) { - if (attr & FILE_ATTRIBUTE_DIRECTORY) { - /* - * Windows NT reports removing a directory as EACCES instead - * of EPERM. - */ - - errno = EPERM; - } - } - } else if (errno == ENOENT) { - attr = GetFileAttributesW(wname); - if (attr != (DWORD) -1) { - if (attr & FILE_ATTRIBUTE_DIRECTORY) { - /* - * Windows 95 reports removing a directory as ENOENT instead - * of EPERM. - */ - - errno = EPERM; - } - } - } else if (errno == EINVAL) { - /* - * Windows NT reports removing a char device as EINVAL instead of - * EACCES. - */ - - errno = EACCES; - } - - return check_error(-1, state->errInfo); -} - -/* - *--------------------------------------------------------------------------- - * - * Changes the name of an existing file or directory, from src to dst. - * If src and dst refer to the same file or directory, does nothing - * and returns success. Otherwise if dst already exists, it will be - * deleted and replaced by src subject to the following conditions: - * If src is a directory, dst may be an empty directory. - * If src is a file, dst may be a file. - * In any other situation where dst already exists, the rename will - * fail. - * - * Some possible error codes: - * - * EACCES: src or dst parent directory can't be read and/or written. - * EEXIST: dst is a non-empty directory. - * EINVAL: src is a root directory or dst is a subdirectory of src. - * EISDIR: dst is a directory, but src is not. - * ENOENT: src doesn't exist, or src or dst is "". - * ENOTDIR: src is a directory, but dst is not. - * EXDEV: src and dst are on different filesystems. - * - * Side effects: - * The implementation of rename may allow cross-filesystem renames, - * but the caller should be prepared to emulate it with copy and - * delete if errno is EXDEV. - * - *--------------------------------------------------------------------------- - */ - -int -efile_rename(Efile_error* errInfo, char* src, char* dst) -{ - Efile_call_state state; - int ret; - DBG_TRACE(1, src); - call_state_init(&state, errInfo); - ret = do_rename(&state, src, dst); - call_state_free(&state); - return ret; -} - -static int -do_rename(Efile_call_state* state, - char* src, /* Original name. */ - char* dst) /* New name. */ -{ - DWORD srcAttr, dstAttr; - WCHAR *wsrc = (WCHAR *) src; - WCHAR *wdst = (WCHAR *) dst; - - ensure_wpath(state, &wsrc); - ensure_wpath(state, &wdst); - - if (MoveFileW(wsrc, wdst) != FALSE) { - return 1; - } - - errno = errno_map(GetLastError()); - srcAttr = GetFileAttributesW(wsrc); - dstAttr = GetFileAttributesW(wdst); - if (srcAttr == (DWORD) -1) { - srcAttr = 0; - } - if (dstAttr == (DWORD) -1) { - dstAttr = 0; - } - - if (errno == EBADF) { - errno = EACCES; - return check_error(-1, state->errInfo); - } - if (errno == EACCES) { - decode: - if (srcAttr & FILE_ATTRIBUTE_DIRECTORY) { - WCHAR *srcPath, *dstPath; - WCHAR *srcRest, *dstRest; - int size; - - srcPath = get_full_wpath_tmp(state, wsrc, &srcRest, 0); - if (!srcPath) { - save_last_error(state->errInfo); - return 0; - } - - dstPath = get_full_wpath_tmp(state, wdst, &dstRest, 0); - if (!dstPath) { - save_last_error(state->errInfo); - return 0; - } - - if (srcRest == NULL) { - srcRest = srcPath + wcslen(srcPath); - } - if (_wcsnicmp(srcPath, dstPath, srcRest - srcPath) == 0) { - /* - * Trying to move a directory into itself. - */ - - errno = EINVAL; - } - if (extract_root(srcPath)) { - /* - * Attempt to move a root directory. Never allowed. - */ - errno = EINVAL; - } - - (void) extract_root(dstPath); - if (dstPath[0] == L'\0') { - /* - * The filename was invalid. (Don't know why, - * but play it safe.) - */ - errno = EINVAL; - } - if (_wcsicmp(srcPath, dstPath) != 0) { - /* - * If src is a directory and dst filesystem != src - * filesystem, errno should be EXDEV. It is very - * important to get this behavior, so that the caller - * can respond to a cross filesystem rename by - * simulating it with copy and delete. The MoveFile - * system call already handles the case of moving a - * *file* between filesystems. - */ - - errno = EXDEV; - } - } - - /* - * Other types of access failure is that dst is a read-only - * filesystem, that an open file referred to src or dest, or that - * src or dest specified the current working directory on the - * current filesystem. EACCES is returned for those cases. - */ - - } else if (errno == EEXIST) { - /* - * Reports EEXIST any time the target already exists. If it makes - * sense, remove the old file and try renaming again. - */ - - if (srcAttr & FILE_ATTRIBUTE_DIRECTORY) { - if (dstAttr & FILE_ATTRIBUTE_DIRECTORY) { - /* - * Overwrite empty dst directory with src directory. The - * following call will remove an empty directory. If it - * fails, it's because it wasn't empty. - */ - - if (RemoveDirectoryW(wdst)) { - /* - * Now that that empty directory is gone, we can try - * renaming again. If that fails, we'll put this empty - * directory back, for completeness. - */ - - if (MoveFileW(wsrc, wdst) != FALSE) { - return 1; - } - - /* - * Some new error has occurred. Don't know what it - * could be, but report this one. - */ - - errno = errno_map(GetLastError()); - CreateDirectoryW(wdst, NULL); - SetFileAttributesW(wdst, dstAttr); - if (errno == EACCES) { - /* - * Decode the EACCES to a more meaningful error. - */ - - goto decode; - } - } - } else { /* (dstAttr & FILE_ATTRIBUTE_DIRECTORY) == 0 */ - errno = ENOTDIR; - } - } else { /* (srcAttr & FILE_ATTRIBUTE_DIRECTORY) == 0 */ - if (dstAttr & FILE_ATTRIBUTE_DIRECTORY) { - errno = EISDIR; - } else { - /* - * Overwrite existing file by: - * - * 1. Rename existing file to temp name. - * 2. Rename old file to new name. - * 3. If success, delete temp file. If failure, - * put temp file back to old name. - */ - - WCHAR *tempName; - int result; - WCHAR *rest; - - tempName = get_full_wpath_tmp(state, wdst, &rest, 14); - if (!tempName || !rest) { - save_last_error(state->errInfo); - return 0; - } - - *rest = L'\0'; - result = -1; - if (GetTempFileNameW(tempName, L"erlr", 0, tempName) != 0) { - /* - * Strictly speaking, need the following DeleteFile and - * MoveFile to be joined as an atomic operation so no - * other app comes along in the meantime and creates the - * same temp file. - */ - - DeleteFileW(tempName); - if (MoveFileW(wdst, tempName) != FALSE) { - if (MoveFileW(wsrc, wdst) != FALSE) { - SetFileAttributesW(tempName, FILE_ATTRIBUTE_NORMAL); - DeleteFileW(tempName); - return 1; - } else { - DeleteFileW(wdst); - MoveFileW(tempName, wdst); - } - } - - /* - * Can't backup dst file or move src file. Return that - * error. Could happen if an open file refers to dst. - */ - - errno = errno_map(GetLastError()); - if (errno == EACCES) { - /* - * Decode the EACCES to a more meaningful error. - */ - goto decode; - } - } - return result; - } - } - } - return check_error(-1, state->errInfo); -} - -int -efile_chdir(Efile_error* errInfo, /* Where to return error codes. */ - char* name) /* Name of directory to make current. */ -{
- /* We don't even try to handle long paths here
- * as current working directory is always limited to MAX_PATH
- * even if we use UNC paths and SetCurrentDirectoryW()
- */
- int success = check_error(_wchdir((WCHAR *) name), errInfo);
- if (!success && errInfo->posix_errno == EINVAL)
- /* POSIXification of errno */
- errInfo->posix_errno = ENOENT; - return success; -} - -int -efile_getdcwd(Efile_error* errInfo, /* Where to return error codes. */ - int drive, /* 0 - current, 1 - A, 2 - B etc. */ - char* buffer, /* Where to return the current directory. */ - size_t size) /* Size of buffer. */ -{ - WCHAR *wbuffer = (WCHAR *) buffer; - size_t wbuffer_size = size / 2; - DBG_TRACE(1, L"#getdcwd#"); - if (_wgetdcwd(drive, wbuffer, wbuffer_size) == NULL) { - return check_error(-1, errInfo); - } - DBG_TRACE1(8, "getdcwd OS=%s", wbuffer); - if (wcsncmp(wbuffer, L"\\\\?\\", 4) == 0) { - wmemmove(wbuffer, wbuffer+4, wcslen(wbuffer+4)+1); - } - for ( ; *wbuffer; wbuffer++) - if (*wbuffer == L'\\') - *wbuffer = L'/'; - DBG_TRACE1(8, "getdcwd ERLANG=%s", (WCHAR*)buffer); - return 1; -} - -int -efile_readdir(Efile_error* errInfo, char* name, EFILE_DIR_HANDLE* dir_handle, - char* buffer, size_t *size) -{ - Efile_call_state state; - int ret; - DBG_TRACE(dir_handle?2:1, name); - call_state_init(&state, errInfo); - ret = do_readdir(&state, name, dir_handle, buffer, size); - call_state_free(&state); - return ret; -} - -static int do_readdir(Efile_call_state* state, - char* name, /* Name of directory to list */ - EFILE_DIR_HANDLE* dir_handle, /* Handle of opened directory or NULL */ - char* buffer, /* Buffer to put one filename in */ - size_t *size) /* in-out size of buffer/size of filename excluding zero - termination in bytes*/ -{ - HANDLE dir; /* Handle to directory. */ - WIN32_FIND_DATAW findData; /* Data found by FindFirstFile() or FindNext(). */ - /* Alignment is not honored, this works on x86 because of alignment fixup by processor. - Not perfect, but faster than alinging by hand (really) */ - WCHAR *wbuffer = (WCHAR *) buffer; - - /* - * First time we must setup everything. - */ - - if (*dir_handle == NULL) { - WCHAR *wname = (WCHAR *) name; - WCHAR* wildcard; - int length; - WCHAR* s; - - ensure_wpath_max(state, &wname, MAX_PATH-2); - length = wcslen(wname); - - wildcard = wpath_tmp_alloc(state, length+3); - - wcscpy(wildcard, wname); - s = wildcard+length-1; - if (*s != L'/' && *s != L'\\') - *++s = L'\\'; - *++s = L'*'; - *++s = L'\0'; - DEBUGF(("Reading %ws\n", wildcard)); - dir = FindFirstFileW(wildcard, &findData); - if (dir == INVALID_HANDLE_VALUE) { - set_error(state->errInfo); - return 0; - } - *dir_handle = (EFILE_DIR_HANDLE) dir; - - if (!IS_DOT_OR_DOTDOT(findData.cFileName)) { - wcscpy(wbuffer, findData.cFileName); - *size = wcslen(wbuffer)*2; - return 1; - } - } - - /* - * Retrieve the name of the next file using the directory handle. - */ - - dir = (HANDLE) *dir_handle; - - for (;;) { - if (FindNextFileW(dir, &findData)) { - if (IS_DOT_OR_DOTDOT(findData.cFileName)) - continue; - wcscpy(wbuffer, findData.cFileName); - *size = wcslen(wbuffer)*2; - return 1; - } - - if (GetLastError() == ERROR_NO_MORE_FILES) { - state->errInfo->posix_errno = state->errInfo->os_errno = 0; - } - else { - set_error(state->errInfo); - } - FindClose(dir); - return 0; - } -} - -int -efile_openfile(Efile_error* errInfo, char* name, int flags, int* pfd, Sint64* pSize) -{ - Efile_call_state state; - int ret; - DBG_TRACE1(1, "openfile(%s)", name); - call_state_init(&state, errInfo); - ret = do_openfile(&state, name, flags, pfd, pSize); - call_state_free(&state); - return ret; -} - -static -int do_openfile(Efile_call_state* state, /* Where to return error codes. */ - char* name, /* Name of directory to open. */ - int flags, /* Flags to use for opening. */ - int* pfd, /* Where to store the file descriptor. */ - Sint64* pSize) /* Where to store the size of the file. */ -{ - Efile_error* errInfo = state->errInfo; - BY_HANDLE_FILE_INFORMATION fileInfo; /* File information from a handle. */ - HANDLE fd; /* Handle to open file. */ - DWORD access; /* Access mode: GENERIC_READ, GENERIC_WRITE. */ - DWORD crFlags; - DWORD flagsAndAttrs = FILE_ATTRIBUTE_NORMAL; - WCHAR *wname = (WCHAR *) name; - - switch (flags & (EFILE_MODE_READ|EFILE_MODE_WRITE)) { - case EFILE_MODE_READ: - access = GENERIC_READ; - crFlags = OPEN_EXISTING; - break; - case EFILE_MODE_WRITE: - access = GENERIC_WRITE; - crFlags = CREATE_ALWAYS; - break; - case EFILE_MODE_READ_WRITE: - access = GENERIC_READ|GENERIC_WRITE; - crFlags = OPEN_ALWAYS; - break; - default: - errno = EINVAL; - check_error(-1, errInfo); - return 0; - } - - if (flags & EFILE_MODE_SYNC) { - flagsAndAttrs = FILE_FLAG_WRITE_THROUGH; - } - - if (flags & EFILE_MODE_APPEND) { - crFlags = OPEN_ALWAYS; - } - if (flags & EFILE_MODE_EXCL) { - crFlags = CREATE_NEW; - } - ensure_wpath(state, &wname); - fd = CreateFileW(wname, access, - FILE_SHARE_FLAGS, - NULL, crFlags, flagsAndAttrs, NULL); - - /* - * Check for errors. - */ - - if (fd == INVALID_HANDLE_VALUE) { - DWORD attr; - - set_error(errInfo); - - /* - * If the error is EACESS, the reason could be that we tried to - * open a directory. In that case, we'll change the error code - * to EISDIR. - */ - if (errInfo->posix_errno && - (attr = GetFileAttributesW(wname)) != INVALID_FILE_ATTRIBUTES && - (attr & FILE_ATTRIBUTE_DIRECTORY)) { - errInfo->posix_errno = EISDIR; - } - return 0; - } - - /* - * Get and return the length of the open file. - */ - - if (!GetFileInformationByHandle(fd, &fileInfo)) - return set_error(errInfo); - *pfd = (int) fd; - if (pSize) { - *pSize = (Sint64) - (((Uint64)fileInfo.nFileSizeHigh << 32) | - (Uint64)fileInfo.nFileSizeLow); - } - return 1; -} - -int -efile_may_openfile(Efile_error* errInfo, char *name) -{ - Efile_call_state state; - WCHAR *wname = (WCHAR *) name; - DWORD attr; - int ret; - - DBG_TRACE(1, name); - call_state_init(&state, errInfo); - ensure_wpath(&state, &wname); - if ((attr = GetFileAttributesW(wname)) == INVALID_FILE_ATTRIBUTES) { - errno = ENOENT; - ret = check_error(-1, errInfo); - } - else if (attr & FILE_ATTRIBUTE_DIRECTORY) { - errno = EISDIR; - ret = check_error(-1, errInfo); - } - else ret = 1; - - call_state_free(&state); - return ret; -} - -void -efile_closefile(fd) -int fd; /* File descriptor for file to close. */ -{ - DBG_TRACE(2, L""); - CloseHandle((HANDLE) fd); -} - -FILE* efile_wfopen(const WCHAR* name, const WCHAR* mode) -{ - Efile_call_state state; - Efile_error dummy; - FILE* f; - call_state_init(&state, &dummy); - ensure_wpath(&state, (WCHAR**)&name); - f = _wfopen(name, mode); - call_state_free(&state); - return f; -} - -int -efile_fdatasync(errInfo, fd) -Efile_error* errInfo; /* Where to return error codes. */ -int fd; /* File descriptor for file to sync. */ -{ - DBG_TRACE(2, L""); - /* Not available in Windows, just call regular fsync */ - return efile_fsync(errInfo, fd); -} - -int -efile_fsync(errInfo, fd) -Efile_error* errInfo; /* Where to return error codes. */ -int fd; /* File descriptor for file to sync. */ -{ - DBG_TRACE(2, L""); - if (!FlushFileBuffers((HANDLE) fd)) { - return check_error(-1, errInfo); - } - return 1; -} - -int -efile_fileinfo(Efile_error* errInfo, Efile_info* pInfo, - char* orig_name, int info_for_link) -{ - Efile_call_state state; - int ret; - DBG_TRACE(1, L""); - call_state_init(&state, errInfo); - ret = do_fileinfo(&state, pInfo, orig_name, info_for_link); - call_state_free(&state); - return ret; -} - -static int -do_fileinfo(Efile_call_state* state, Efile_info* pInfo, - char* orig_name, int info_for_link) -{ - Efile_error* errInfo = state->errInfo; - HANDLE findhandle; /* Handle returned by FindFirstFile(). */ - WIN32_FIND_DATAW findbuf; /* Data return by FindFirstFile(). */ - WCHAR* name = NULL; - WCHAR* win_path; - int name_len; - int drive; /* Drive for filename (1 = A:, 2 = B: etc). */ - WCHAR *worig_name = (WCHAR *) orig_name; - - ensure_wpath(state, &worig_name); - /* Don't allow wildcards to be interpreted by system */ - - - /* - * Move the name to a buffer and make sure to remove a trailing - * slash, because it causes FindFirstFile() to fail on Win95. - */ - - name_len = wcslen(worig_name); - - name = wpath_tmp_alloc(state, name_len+1); - wcscpy(name, worig_name); - if (name_len > 2 && ISSLASH(name[name_len-1]) && - name[name_len-2] != L':') { - name[name_len-1] = L'\0'; - } - - win_path = name; - if (wcsncmp(name, L"\\\\?\\", 4) == 0) { - win_path += 4; - } - - if (wcspbrk(win_path, L"?*")) { - enoent: - errInfo->posix_errno = ENOENT; - errInfo->os_errno = ERROR_FILE_NOT_FOUND; - return 0; - } - - /* Try to get disk from name. If none, get current disk. */ - - if (win_path[1] != L':') { - WCHAR* cwd_path = get_cwd_wpath_tmp(state); - drive = 0; - if (cwd_path[1] == L':') { - drive = towlower(cwd_path[0]) - L'a' + 1; - } - } else if (*win_path && win_path[2] == L'\0') { - /* - * X: and nothing more is an error. - */ - errInfo->posix_errno = ENOENT; - errInfo->os_errno = ERROR_FILE_NOT_FOUND; - return 0; - } else { - drive = towlower(*win_path) - L'a' + 1; - } - - findhandle = FindFirstFileW(name, &findbuf); - if (findhandle == INVALID_HANDLE_VALUE) { - WCHAR* path = NULL; - - if (!(wcspbrk(name, L"./\\") && - (path = get_full_wpath_tmp(state, name, NULL, 0)) && - /* root dir. ('C:\') or UNC root dir. ('\\server\share\') */ - ((wcslen(path) == 3) || is_root_unc_name(path)) && - (GetDriveTypeW(path) > 1) ) ) { - - errInfo->posix_errno = ENOENT; - errInfo->os_errno = ERROR_FILE_NOT_FOUND; - return 0; - } - - /* - * Root directories (such as C:\ or \\server\share\ are fabricated. - */ - - findbuf.dwFileAttributes = FILE_ATTRIBUTE_DIRECTORY; - findbuf.nFileSizeHigh = 0; - findbuf.nFileSizeLow = 0; - findbuf.cFileName[0] = L'\0'; - - pInfo->links = 1; - pInfo->cTime = pInfo->accessTime = pInfo->modifyTime = 0; - } else { - SYSTEMTIME SystemTime; - FILETIME LocalFTime; - - /*first check if we are a symlink */ - if (!info_for_link && (findbuf.dwFileAttributes & - FILE_ATTRIBUTE_REPARSE_POINT)){ - /* - * given that we know this is a symlink, - we should be able to find its target */ - WCHAR* target_name = (WCHAR*) do_readlink(state, (char *) name, NULL, 0); - if (target_name) { - FindClose(findhandle); - return do_fileinfo(state, pInfo, - (char *) target_name, info_for_link); - } - } - - /* number of links: */ - { - HANDLE handle; /* Handle returned by CreateFile() */ - BY_HANDLE_FILE_INFORMATION fileInfo; /* from CreateFile() */ - - /* We initialise nNumberOfLinks as GetFileInformationByHandle - does not always initialise this field */ - fileInfo.nNumberOfLinks = 1; - if (handle = CreateFileW(name, GENERIC_READ, FILE_SHARE_FLAGS, NULL, - OPEN_EXISTING, 0, NULL)) { - GetFileInformationByHandle(handle, &fileInfo); - pInfo->links = fileInfo.nNumberOfLinks; - CloseHandle(handle); - } else { - pInfo->links = 1; - } - } - - FILETIME_TO_EPOCH(pInfo->modifyTime, findbuf.ftLastWriteTime); - - if (findbuf.ftLastAccessTime.dwLowDateTime == 0 && - findbuf.ftLastAccessTime.dwHighDateTime == 0) { - pInfo->accessTime = pInfo->modifyTime; - } else { - FILETIME_TO_EPOCH(pInfo->accessTime, findbuf.ftLastAccessTime); - } - - if (findbuf.ftCreationTime.dwLowDateTime == 0 && - findbuf.ftCreationTime.dwHighDateTime == 0) { - pInfo->cTime = pInfo->modifyTime; - } else { - FILETIME_TO_EPOCH(pInfo->cTime ,findbuf.ftCreationTime); - } - FindClose(findhandle); - } - - pInfo->size_low = findbuf.nFileSizeLow; - pInfo->size_high = findbuf.nFileSizeHigh; - - if (info_for_link && (findbuf.dwFileAttributes & - FILE_ATTRIBUTE_REPARSE_POINT)) - pInfo->type = FT_SYMLINK; - else if (findbuf.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY) - pInfo->type = FT_DIRECTORY; - else - pInfo->type = FT_REGULAR; - - if (findbuf.dwFileAttributes & FILE_ATTRIBUTE_READONLY) - pInfo->access = FA_READ; - else - pInfo->access = FA_READ|FA_WRITE; - - pInfo->mode = dos_to_posix_mode(findbuf.dwFileAttributes, name); - pInfo->major_device = drive; - pInfo->minor_device = 0; - pInfo->inode = 0; - pInfo->uid = 0; - pInfo->gid = 0; - - return 1; -} - -int -efile_write_info(Efile_error* errInfo, - Efile_info* pInfo, - char* name) -{ - Efile_call_state state; - int ret; - call_state_init(&state, errInfo); - ret = do_write_info(&state, pInfo, name); - call_state_free(&state); - return ret; -} - -static int -do_write_info(Efile_call_state* state, - Efile_info* pInfo, - char* name) -{ - Efile_error* errInfo = state->errInfo; - SYSTEMTIME timebuf; - FILETIME ModifyFileTime; - FILETIME AccessFileTime; - FILETIME CreationFileTime; - HANDLE fd; - DWORD attr; - DWORD tempAttr; - WCHAR *wname = (WCHAR *) name; - - DBG_TRACE(1, name); - - ensure_wpath(state, &wname); - - /* - * Get the attributes for the file. - */ - - tempAttr = attr = GetFileAttributesW(wname); - if (attr == 0xffffffff) { - return set_error(errInfo); - } - if (pInfo->mode != -1) { - if (pInfo->mode & _S_IWRITE) { - /* clear read only bit */ - attr &= ~FILE_ATTRIBUTE_READONLY; - } else { - /* set read only bit */ - attr |= FILE_ATTRIBUTE_READONLY; - } - } - - /* - * Construct all file times. - */ - - EPOCH_TO_FILETIME(ModifyFileTime, pInfo->modifyTime); - EPOCH_TO_FILETIME(AccessFileTime, pInfo->accessTime); - EPOCH_TO_FILETIME(CreationFileTime, pInfo->cTime); - - /* - * If necessary, set the file times. - */ - - /* - * If the has read only access, we must temporarily turn on - * write access (this is necessary for native filesystems, - * but not for NFS filesystems). - */ - - if (tempAttr & FILE_ATTRIBUTE_READONLY) { - tempAttr &= ~FILE_ATTRIBUTE_READONLY; - if (!SetFileAttributesW(wname, tempAttr)) { - return set_error(errInfo); - } - } - - fd = CreateFileW(wname, GENERIC_READ|GENERIC_WRITE, - FILE_SHARE_FLAGS, - NULL, OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, NULL); - if (fd != INVALID_HANDLE_VALUE) { - BOOL result = SetFileTime(fd, &CreationFileTime, &AccessFileTime, &ModifyFileTime); - if (!result) { - return set_error(errInfo); - } - CloseHandle(fd); - } - - /* - * If the file doesn't have the correct attributes, set them now. - * (It could have been done before setting the file times, above). - */ - - if (tempAttr != attr) { - if (!SetFileAttributesW(wname, attr)) { - return set_error(errInfo); - } - } - return 1; -} - - -int -efile_pwrite(errInfo, fd, buf, count, offset) -Efile_error* errInfo; /* Where to return error codes. */ -int fd; /* File descriptor to write to. */ -char* buf; /* Buffer to write. */ -size_t count; /* Number of bytes to write. */ -Sint64 offset; /* where to write it */ -{ - int res; - DBG_TRACE(2, L""); - res = efile_seek(errInfo, fd, offset, EFILE_SEEK_SET, NULL); - if (res) { - return efile_write(errInfo, EFILE_MODE_WRITE, fd, buf, count); - } else { - return res; - } -} - -/* position and read/write as a single atomic op */ -int -efile_pread(errInfo, fd, offset, buf, count, pBytesRead) -Efile_error* errInfo; /* Where to return error codes. */ -int fd; /* File descriptor to read from. */ -Sint64 offset; /* Offset in bytes from BOF. */ -char* buf; /* Buffer to read into. */ -size_t count; /* Number of bytes to read. */ -size_t* pBytesRead; /* Where to return number of bytes read. */ -{ - int res; - DBG_TRACE(2, L""); - res = efile_seek(errInfo, fd, offset, EFILE_SEEK_SET, NULL); - if (res) { - return efile_read(errInfo, EFILE_MODE_READ, fd, buf, count, pBytesRead); - } else { - return res; - } -} - - - -int -efile_write(errInfo, flags, fd, buf, count) -Efile_error* errInfo; /* Where to return error codes. */ -int flags; /* Flags given when file was opened. */ -int fd; /* File descriptor to write to. */ -char* buf; /* Buffer to write. */ -size_t count; /* Number of bytes to write. */ -{ - DWORD written; /* Bytes written in last operation. */ - OVERLAPPED overlapped; - OVERLAPPED* pOverlapped = NULL; - - DBG_TRACE(2, L""); - if (flags & EFILE_MODE_APPEND) { - memset(&overlapped, 0, sizeof(overlapped)); - overlapped.Offset = 0xffffffff; - overlapped.OffsetHigh = 0xffffffff; - pOverlapped = &overlapped; - } - while (count > 0) { - if (!WriteFile((HANDLE) fd, buf, count, &written, pOverlapped)) - return set_error(errInfo); - buf += written; - count -= written; - } - return 1; -} - -int -efile_writev(Efile_error* errInfo, /* Where to return error codes */ - int flags, /* Flags given when file was - * opened */ - int fd, /* File descriptor to write to */ - SysIOVec* iov, /* Vector of buffer structs. - * The structs are unchanged - * after the call */ - int iovcnt) /* Number of structs in vector */ -{ - int cnt; /* Buffers so far written */ - OVERLAPPED overlapped; - OVERLAPPED* pOverlapped = NULL; - - DBG_TRACE(2, L""); - ASSERT(iovcnt >= 0); - - if (flags & EFILE_MODE_APPEND) { - memset(&overlapped, 0, sizeof(overlapped)); - overlapped.Offset = 0xffffffff; - overlapped.OffsetHigh = 0xffffffff; - pOverlapped = &overlapped; - } - for (cnt = 0; cnt < iovcnt; cnt++) { - if (iov[cnt].iov_base && iov[cnt].iov_len > 0) { - /* Non-empty buffer */ - int p; /* Position in buffer */ - int w = iov[cnt].iov_len;/* Bytes written in this call */ - for (p = 0; p < iov[cnt].iov_len; p += w) { - if (!WriteFile((HANDLE) fd, - iov[cnt].iov_base + p, - iov[cnt].iov_len - p, - &w, - pOverlapped)) - return set_error(errInfo); - } - } - } - return 1; -} - -int -efile_read(errInfo, flags, fd, buf, count, pBytesRead) -Efile_error* errInfo; /* Where to return error codes. */ -int flags; /* Flags given when file was opened. */ -int fd; /* File descriptor to read from. */ -char* buf; /* Buffer to read into. */ -size_t count; /* Number of bytes to read. */ -size_t* pBytesRead; /* Where to return number of bytes read. */ -{ - DWORD nbytes = 0; - - DBG_TRACE(2, L""); - if (!ReadFile((HANDLE) fd, buf, count, &nbytes, NULL)) - return set_error(errInfo); - - *pBytesRead = nbytes; - return 1; -} - -int -efile_seek(errInfo, fd, offset, origin, new_location) -Efile_error* errInfo; /* Where to return error codes. */ -int fd; /* File descriptor to do the seek on. */ -Sint64 offset; /* Offset in bytes from the given origin. */ -int origin; /* Origin of seek (SEEK_SET, SEEK_CUR, - * SEEK_END). - */ -Sint64* new_location; /* Resulting new location in file. */ -{ - LARGE_INTEGER off, new_loc; - - DBG_TRACE(2, L""); - switch (origin) { - case EFILE_SEEK_SET: origin = FILE_BEGIN; break; - case EFILE_SEEK_CUR: origin = FILE_CURRENT; break; - case EFILE_SEEK_END: origin = FILE_END; break; - default: - errno = EINVAL; - check_error(-1, errInfo); - break; - } - - off.QuadPart = offset; - if (! SetFilePointerEx((HANDLE) fd, off, - new_location ? &new_loc : NULL, origin)) { - return set_error(errInfo); - } - if (new_location) { - *new_location = new_loc.QuadPart; - DEBUGF(("efile_seek(offset=%ld, origin=%d) -> %ld\n", - (long) offset, origin, (long) *new_location)); - } else { - DEBUGF(("efile_seek(offset=%ld, origin=%d)\n", (long) offset, origin)); - } - return 1; -} - -int -efile_truncate_file(errInfo, fd, flags) -Efile_error* errInfo; /* Where to return error codes. */ -int *fd; /* File descriptor for file to truncate. */ -int flags; -{ - DBG_TRACE(2, L""); - if (!SetEndOfFile((HANDLE) (*fd))) - return set_error(errInfo); - return 1; -} - - -/* - * is_root_unc_name - returns TRUE if the argument is a UNC name specifying - * a root share. That is, if it is of the form \\server\share\. - * This routine will also return true if the argument is of the - * form \\server\share (no trailing slash) but Win32 currently - * does not like that form. - * - * Forward slashes ('/') may be used instead of backslashes ('\'). - */ - -static int -is_root_unc_name(const WCHAR *path) -{ - /* - * If a root UNC name, path will start with 2 (but not 3) slashes - */ - - if ((wcslen(path) >= 5) /* minimum string is "//x/y" */ - && ISSLASH(path[0]) && ISSLASH(path[1])) - { - const WCHAR *p = path + 2; - - /* - * find the slash between the server name and share name - */ - while ( * ++ p ) - if ( ISSLASH(*p) ) - break ; - - if ( *p && p[1] ) - { - /* - * is there a further slash? - */ - while ( * ++ p ) - if ( ISSLASH(*p) ) - break ; - - /* - * just final slash (or no final slash) - */ - if ( !*p || !p[1]) - return 1; - } - } - - return 0 ; -} - -/* - * Extracts the root part of an absolute filename (by modifying the string - * pointed to by the name argument). The name can start - * with either a driver letter (for example, C:\), or a UNC name - * (for example, \\guinness\bjorn). - * - * If the name is invalid, the buffer will be modified to point to - * an empty string. - * - * Returns: 1 if the name consists of just the root part, 0 if - * the name was longer. - */ - -static int -extract_root(WCHAR* name) -{ - int len = wcslen(name); - - if (iswalpha(name[0]) && name[1] == L':' && ISSLASH(name[2])) { - WCHAR c = name[3]; - name[3] = L'\0'; - return c == L'\0'; - } else if (len < 5 || !ISSLASH(name[0]) || !ISSLASH(name[1])) { - goto error; - } else { /* Try to find the end of the UNC name. */ - WCHAR* p; - WCHAR c; - - /* - * Find the slash between the server name and share name. - */ - - for (p = name + 2; *p; p++) - if (ISSLASH(*p)) - break; - if (*p == L'\0') - goto error; - - /* - * Find the slash after the share name. - */ - - for (p++; *p; p++) - if (ISSLASH(*p)) - break; - c = *p; - *p = L'\0'; - return c == L'\0' || p[1] == L'\0'; - } - - error: - *name = L'\0'; - return 1; -} - -static unsigned short -dos_to_posix_mode(int attr, const WCHAR *name) -{ - register unsigned short uxmode; - unsigned dosmode; - register const WCHAR *p; - - dosmode = attr & 0xff; - if ((p = name)[1] == L':') - p += 2; - - /* check to see if this is a directory - note we must make a special - * check for the root, which DOS thinks is not a directory - */ - - uxmode = (unsigned short) - (((ISSLASH(*p) && !p[1]) || (dosmode & FILE_ATTRIBUTE_DIRECTORY) || - *p == L'\0') ? _S_IFDIR|_S_IEXEC : _S_IFREG); - - /* If attribute byte does not have read-only bit, it is read-write */ - - uxmode |= (dosmode & FILE_ATTRIBUTE_READONLY) ? - _S_IREAD : (_S_IREAD|_S_IWRITE); - - /* see if file appears to be executable - check extension of name */ - - if (p = wcsrchr(name, L'.')) { - if (!_wcsicmp(p, L".exe") || - !_wcsicmp(p, L".cmd") || - !_wcsicmp(p, L".bat") || - !_wcsicmp(p, L".com")) - uxmode |= _S_IEXEC; - } - - /* propagate user read/write/execute bits to group/other fields */ - - uxmode |= (uxmode & 0700) >> 3; - uxmode |= (uxmode & 0700) >> 6; - - return uxmode; -} - - -int -efile_readlink(Efile_error* errInfo, char* name, char* buffer, size_t size) -{ - Efile_call_state state; - int ret; - DBG_TRACE(1, name); - call_state_init(&state, errInfo); - ret = !!do_readlink(&state, name, buffer, size); - call_state_free(&state); - return ret; -} - -/* If buffer==0, return buffer allocated by wpath_tmp_allocate -*/ -static char* -do_readlink(Efile_call_state* state, char* name, char* buffer, size_t size) -{ - /* - * load dll and see if we have CreateSymbolicLink at runtime: - * (Vista only) - */ - HINSTANCE hModule = NULL; - WCHAR *wname = (WCHAR *) name; - WCHAR *wbuffer = (WCHAR *) buffer; - DWORD wsize = size / sizeof(WCHAR); - char* ret = NULL; - - if ((hModule = LoadLibrary("kernel32.dll")) != NULL) { - typedef DWORD (WINAPI * GETFINALPATHNAMEBYHANDLEPTR)( - HANDLE hFile, - LPCWSTR lpFilePath, - DWORD cchFilePath, - DWORD dwFlags); - - GETFINALPATHNAMEBYHANDLEPTR pGetFinalPathNameByHandle = - (GETFINALPATHNAMEBYHANDLEPTR)GetProcAddress(hModule, "GetFinalPathNameByHandleW"); - - if (pGetFinalPathNameByHandle != NULL) { - DWORD fileAttributes; - ensure_wpath(state, &wname); - /* first check if file is a symlink; {error, einval} otherwise */ - fileAttributes = GetFileAttributesW(wname); - if ((fileAttributes & FILE_ATTRIBUTE_REPARSE_POINT)) { - DWORD success = 0; - HANDLE h = CreateFileW(wname, GENERIC_READ, FILE_SHARE_FLAGS, NULL, OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL); - int len; - if(h != INVALID_HANDLE_VALUE) { - if (!wbuffer) { /* dynamic allocation */ - WCHAR dummy; - wsize = pGetFinalPathNameByHandle(h, &dummy, 0, 0); - if (wsize) { - wbuffer = wpath_tmp_alloc(state, wsize); - } - } - if (wbuffer - && (success = pGetFinalPathNameByHandle(h, wbuffer, wsize, 0)) - && success < wsize) { - WCHAR* wp; - - /* GetFinalPathNameByHandle prepends path with "\\?\": */ - len = wcslen(wbuffer); - wmemmove(wbuffer,wbuffer+4,len-3); - if (len - 4 >= 2 && wbuffer[1] == L':' && wbuffer[0] >= L'A' && - wbuffer[0] <= L'Z') { - wbuffer[0] = wbuffer[0] + L'a' - L'A'; - } - - for (wp=wbuffer ; *wp; wp++) - if (*wp == L'\\') - *wp = L'/'; - } - CloseHandle(h); - } - if (success) { - ret = (char*) wbuffer; - } else { - set_error(state->errInfo); - } - } else { - errno = EINVAL; - save_last_error(state->errInfo); - } - goto done; - } - } - errno = ENOTSUP; - save_last_error(state->errInfo); - -done: - if (hModule) - FreeLibrary(hModule); - return ret; -} - - -int -efile_altname(Efile_error* errInfo, char* orig_name, char* buffer, size_t size) -{ - Efile_call_state state; - int ret; - DBG_TRACE(1, orig_name); - call_state_init(&state, errInfo); - ret = do_altname(&state, orig_name, buffer, size); - call_state_free(&state); - return ret; -} - -static int -do_altname(Efile_call_state* state, char* orig_name, char* buffer, size_t size) -{ - WIN32_FIND_DATAW wfd; - HANDLE fh; - WCHAR* name; - int name_len; - WCHAR* full_path = NULL; - WCHAR *worig_name = (WCHAR *) orig_name; - WCHAR *wbuffer = (WCHAR *) buffer; - int drive; /* Drive for filename (1 = A:, 2 = B: etc). */ - - /* Don't allow wildcards to be interpreted by system */ - - if (wcspbrk(worig_name, L"?*")) { - enoent: - state->errInfo->posix_errno = ENOENT; - state->errInfo->os_errno = ERROR_FILE_NOT_FOUND; - return 0; - } - - /* - * Move the name to a buffer and make sure to remove a trailing - * slash, because it causes FindFirstFile() to fail on Win95. - */ - ensure_wpath(state, &worig_name); - name_len = wcslen(worig_name); - - name = wpath_tmp_alloc(state, name_len + 1); - wcscpy(name, worig_name); - if (name_len > 2 && ISSLASH(name[name_len-1]) && - name[name_len-2] != L':') { - name[name_len-1] = L'\0'; - } - - /* Try to get disk from name. If none, get current disk. */ - - if (name[1] != L':') { - WCHAR* cwd_path = get_cwd_wpath_tmp(state); - drive = 0; - if (cwd_path[1] == L':') { - drive = towlower(cwd_path[0]) - L'a' + 1; - } - } else if (*name && name[2] == L'\0') { - /* - * X: and nothing more is an error. - */ - goto enoent; - } else { - drive = towlower(*name) - L'a' + 1; - } - fh = FindFirstFileW(name,&wfd); - if (fh == INVALID_HANDLE_VALUE) { - DWORD fff_error = GetLastError(); - if (!(wcspbrk(name, L"./\\") && - (full_path = get_full_wpath_tmp(state, name, NULL, 0)) && - /* root dir. ('C:\') or UNC root dir. ('\\server\share\') */ - ((wcslen(full_path) == 3) || is_root_unc_name(full_path)) && - (GetDriveTypeW(full_path) > 1) ) ) { - - set_os_errno(state->errInfo, fff_error); - return 0; - } - /* - * Root directories (such as C:\ or \\server\share\ are fabricated. - */ - wcscpy(wbuffer,name); - return 1; - } - - wcscpy(wbuffer,wfd.cAlternateFileName); - if (!*wbuffer) { - wcscpy(wbuffer,wfd.cFileName); - } - FindClose(fh); - return 1; -} - - -int -efile_link(Efile_error* errInfo, char* old, char* new) -{ - Efile_call_state state; - WCHAR *wold = (WCHAR *) old; - WCHAR *wnew = (WCHAR *) new; - int ret; - DBG_TRACE(1, old); - call_state_init(&state, errInfo); - ensure_wpath(&state, &wold); - ensure_wpath(&state, &wnew); - if(!CreateHardLinkW(wnew, wold, NULL)) { - ret = set_error(errInfo); - } - else ret =1; - call_state_free(&state); - return ret; -} - -int -efile_symlink(Efile_error* errInfo, char* old, char* new) -{ - Efile_call_state state; - int ret; - DBG_TRACE2(1, "symlink(%s <- %s)", old, new); - call_state_init(&state, errInfo); - ret = do_symlink(&state, old, new); - call_state_free(&state); - return ret; -} - -static int -do_symlink(Efile_call_state* state, char* old, char* new) -{ - /* - * Load dll and see if we have CreateSymbolicLink at runtime: - * (Vista only) - */ - HINSTANCE hModule = NULL; - WCHAR *wold = (WCHAR *) old; - WCHAR *wnew = (WCHAR *) new; - - DBG_TRACE(1, old); - if ((hModule = LoadLibrary("kernel32.dll")) != NULL) { - typedef BOOLEAN (WINAPI * CREATESYMBOLICLINKFUNCPTR) ( - LPCWSTR lpSymlinkFileName, - LPCWSTR lpTargetFileName, - DWORD dwFlags); - - CREATESYMBOLICLINKFUNCPTR pCreateSymbolicLink = - (CREATESYMBOLICLINKFUNCPTR) GetProcAddress(hModule, - "CreateSymbolicLinkW"); - /* A for MBCS, W for UNICODE... char* above implies 'W'! */ - if (pCreateSymbolicLink != NULL) { - ensure_wpath(state, &wold); - ensure_wpath(state, &wnew); - { - DWORD attr = GetFileAttributesW(wold); - int flag = (attr != INVALID_FILE_ATTRIBUTES && - attr & FILE_ATTRIBUTE_DIRECTORY) ? 1 : 0; - /* SYMBOLIC_LINK_FLAG_DIRECTORY = 1 */ - BOOLEAN success = pCreateSymbolicLink(wnew, wold, flag); - FreeLibrary(hModule); - - if (success) { - return 1; - } else { - return set_error(state->errInfo); - } - } - } else - FreeLibrary(hModule); - } - errno = ENOTSUP; - return check_error(-1, state->errInfo); -} - -int -efile_fadvise(Efile_error* errInfo, int fd, Sint64 offset, - Sint64 length, int advise) -{ - DBG_TRACE(2, L""); - /* posix_fadvise is not available on Windows, do nothing */ - errno = ERROR_SUCCESS; - return check_error(0, errInfo); -} - -int -efile_fallocate(Efile_error* errInfo, int fd, Sint64 offset, Sint64 length) -{ - DBG_TRACE(2, L""); - /* No file preallocation method available in Windows. */ - errno = errno_map(ERROR_NOT_SUPPORTED); - SetLastError(ERROR_NOT_SUPPORTED); - - return check_error(-1, errInfo); -} |