xref: /third_party/libuv/src/unix/fs.c (revision e66f31c5)
1/* Copyright Joyent, Inc. and other Node contributors. All rights reserved.
2 *
3 * Permission is hereby granted, free of charge, to any person obtaining a copy
4 * of this software and associated documentation files (the "Software"), to
5 * deal in the Software without restriction, including without limitation the
6 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
7 * sell copies of the Software, and to permit persons to whom the Software is
8 * furnished to do so, subject to the following conditions:
9 *
10 * The above copyright notice and this permission notice shall be included in
11 * all copies or substantial portions of the Software.
12 *
13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
16 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
17 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
18 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
19 * IN THE SOFTWARE.
20 */
21
22/* Caveat emptor: this file deviates from the libuv convention of returning
23 * negated errno codes. Most uv_fs_*() functions map directly to the system
24 * call of the same name. For more complex wrappers, it's easier to just
25 * return -1 with errno set. The dispatcher in uv__fs_work() takes care of
26 * getting the errno to the right place (req->result or as the return value.)
27 */
28
29#include "uv.h"
30#include "internal.h"
31
32#include <errno.h>
33#include <dlfcn.h>
34#include <stdio.h>
35#include <stdlib.h>
36#include <string.h>
37#include <limits.h> /* PATH_MAX */
38
39#include <sys/types.h>
40#include <sys/socket.h>
41#include <sys/stat.h>
42#include <sys/time.h>
43#include <sys/uio.h>
44#include <unistd.h>
45#include <fcntl.h>
46#include <poll.h>
47
48#if defined(__linux__)
49# include <sys/sendfile.h>
50#endif
51
52#if defined(__sun)
53# include <sys/sendfile.h>
54# include <sys/sysmacros.h>
55#endif
56
57#if defined(__APPLE__)
58# include <sys/sysctl.h>
59#elif defined(__linux__) && !defined(FICLONE)
60# include <sys/ioctl.h>
61# define FICLONE _IOW(0x94, 9, int)
62#endif
63
64#if defined(_AIX) && !defined(_AIX71)
65# include <utime.h>
66#endif
67
68#if defined(__APPLE__)            ||                                      \
69    defined(__DragonFly__)        ||                                      \
70    defined(__FreeBSD__)          ||                                      \
71    defined(__OpenBSD__)          ||                                      \
72    defined(__NetBSD__)
73# include <sys/param.h>
74# include <sys/mount.h>
75#elif defined(__sun)      || \
76      defined(__MVS__)    || \
77      defined(__NetBSD__) || \
78      defined(__HAIKU__)  || \
79      defined(__QNX__)
80# include <sys/statvfs.h>
81#else
82# include <sys/statfs.h>
83#endif
84
85#if defined(__CYGWIN__) ||                                                    \
86    (defined(__HAIKU__) && B_HAIKU_VERSION < B_HAIKU_VERSION_1_PRE_BETA_5) || \
87    (defined(__sun) && !defined(__illumos__)) ||                              \
88    (defined(__APPLE__) && !TARGET_OS_IPHONE &&                               \
89     MAC_OS_X_VERSION_MIN_REQUIRED < 110000)
90#define preadv(fd, bufs, nbufs, off)                                          \
91  pread(fd, (bufs)->iov_base, (bufs)->iov_len, off)
92#define pwritev(fd, bufs, nbufs, off)                                         \
93  pwrite(fd, (bufs)->iov_base, (bufs)->iov_len, off)
94#endif
95
96#if defined(_AIX) && _XOPEN_SOURCE <= 600
97extern char *mkdtemp(char *template); /* See issue #740 on AIX < 7 */
98#endif
99
100#define INIT(subtype)                                                         \
101  do {                                                                        \
102    if (req == NULL)                                                          \
103      return UV_EINVAL;                                                       \
104    UV_REQ_INIT(req, UV_FS);                                                  \
105    req->fs_type = UV_FS_ ## subtype;                                         \
106    req->result = 0;                                                          \
107    req->ptr = NULL;                                                          \
108    req->loop = loop;                                                         \
109    req->path = NULL;                                                         \
110    req->new_path = NULL;                                                     \
111    req->bufs = NULL;                                                         \
112    req->cb = cb;                                                             \
113  }                                                                           \
114  while (0)
115
116#define PATH                                                                  \
117  do {                                                                        \
118    assert(path != NULL);                                                     \
119    if (cb == NULL) {                                                         \
120      req->path = path;                                                       \
121    } else {                                                                  \
122      req->path = uv__strdup(path);                                           \
123      if (req->path == NULL)                                                  \
124        return UV_ENOMEM;                                                     \
125    }                                                                         \
126  }                                                                           \
127  while (0)
128
129#define PATH2                                                                 \
130  do {                                                                        \
131    if (cb == NULL) {                                                         \
132      req->path = path;                                                       \
133      req->new_path = new_path;                                               \
134    } else {                                                                  \
135      size_t path_len;                                                        \
136      size_t new_path_len;                                                    \
137      path_len = strlen(path) + 1;                                            \
138      new_path_len = strlen(new_path) + 1;                                    \
139      req->path = uv__malloc(path_len + new_path_len);                        \
140      if (req->path == NULL)                                                  \
141        return UV_ENOMEM;                                                     \
142      req->new_path = req->path + path_len;                                   \
143      memcpy((void*) req->path, path, path_len);                              \
144      memcpy((void*) req->new_path, new_path, new_path_len);                  \
145    }                                                                         \
146  }                                                                           \
147  while (0)
148
149#ifdef USE_FFRT
150#define POST                                                                  \
151  do {                                                                        \
152    if (cb != NULL) {                                                         \
153      uv__req_register(loop, req);                                            \
154      uv__work_submit(loop,                                                   \
155                      (uv_req_t*)req,                                         \
156                      &req->work_req,                                         \
157                      UV__WORK_FAST_IO,                                       \
158                      uv__fs_work,                                            \
159                      uv__fs_done);                                           \
160      return 0;                                                               \
161    }                                                                         \
162    else {                                                                    \
163      uv__fs_work(&req->work_req);                                            \
164      return req->result;                                                     \
165    }                                                                         \
166  }                                                                           \
167  while (0)
168#else
169#define POST                                                                  \
170  do {                                                                        \
171    if (cb != NULL) {                                                         \
172      uv__req_register(loop, req);                                            \
173      uv__work_submit(loop,                                                   \
174                      &req->work_req,                                         \
175                      UV__WORK_FAST_IO,                                       \
176                      uv__fs_work,                                            \
177                      uv__fs_done);                                           \
178      return 0;                                                               \
179    }                                                                         \
180    else {                                                                    \
181      uv__fs_work(&req->work_req);                                            \
182      return req->result;                                                     \
183    }                                                                         \
184  }                                                                           \
185  while (0)
186#endif
187
188
189static int uv__fs_close(int fd) {
190  int rc;
191
192  rc = uv__close_nocancel(fd);
193  if (rc == -1)
194    if (errno == EINTR || errno == EINPROGRESS)
195      rc = 0;  /* The close is in progress, not an error. */
196
197  return rc;
198}
199
200
201static ssize_t uv__fs_fsync(uv_fs_t* req) {
202#if defined(__APPLE__)
203  /* Apple's fdatasync and fsync explicitly do NOT flush the drive write cache
204   * to the drive platters. This is in contrast to Linux's fdatasync and fsync
205   * which do, according to recent man pages. F_FULLFSYNC is Apple's equivalent
206   * for flushing buffered data to permanent storage. If F_FULLFSYNC is not
207   * supported by the file system we fall back to F_BARRIERFSYNC or fsync().
208   * This is the same approach taken by sqlite, except sqlite does not issue
209   * an F_BARRIERFSYNC call.
210   */
211  int r;
212
213  r = fcntl(req->file, F_FULLFSYNC);
214  if (r != 0)
215    r = fcntl(req->file, 85 /* F_BARRIERFSYNC */);  /* fsync + barrier */
216  if (r != 0)
217    r = fsync(req->file);
218  return r;
219#else
220  return fsync(req->file);
221#endif
222}
223
224
225static ssize_t uv__fs_fdatasync(uv_fs_t* req) {
226#if defined(__linux__) || defined(__sun) || defined(__NetBSD__)
227  return fdatasync(req->file);
228#elif defined(__APPLE__)
229  /* See the comment in uv__fs_fsync. */
230  return uv__fs_fsync(req);
231#else
232  return fsync(req->file);
233#endif
234}
235
236
237UV_UNUSED(static struct timespec uv__fs_to_timespec(double time)) {
238  struct timespec ts;
239  ts.tv_sec  = time;
240  ts.tv_nsec = (time - ts.tv_sec) * 1e9;
241
242 /* TODO(bnoordhuis) Remove this. utimesat() has nanosecond resolution but we
243  * stick to microsecond resolution for the sake of consistency with other
244  * platforms. I'm the original author of this compatibility hack but I'm
245  * less convinced it's useful nowadays.
246  */
247  ts.tv_nsec -= ts.tv_nsec % 1000;
248
249  if (ts.tv_nsec < 0) {
250    ts.tv_nsec += 1e9;
251    ts.tv_sec -= 1;
252  }
253  return ts;
254}
255
256UV_UNUSED(static struct timeval uv__fs_to_timeval(double time)) {
257  struct timeval tv;
258  tv.tv_sec  = time;
259  tv.tv_usec = (time - tv.tv_sec) * 1e6;
260  if (tv.tv_usec < 0) {
261    tv.tv_usec += 1e6;
262    tv.tv_sec -= 1;
263  }
264  return tv;
265}
266
267static ssize_t uv__fs_futime(uv_fs_t* req) {
268#if defined(__linux__)                                                        \
269    || defined(_AIX71)                                                        \
270    || defined(__HAIKU__)                                                     \
271    || defined(__GNU__)
272  struct timespec ts[2];
273  ts[0] = uv__fs_to_timespec(req->atime);
274  ts[1] = uv__fs_to_timespec(req->mtime);
275  return futimens(req->file, ts);
276#elif defined(__APPLE__)                                                      \
277    || defined(__DragonFly__)                                                 \
278    || defined(__FreeBSD__)                                                   \
279    || defined(__NetBSD__)                                                    \
280    || defined(__OpenBSD__)                                                   \
281    || defined(__sun)
282  struct timeval tv[2];
283  tv[0] = uv__fs_to_timeval(req->atime);
284  tv[1] = uv__fs_to_timeval(req->mtime);
285# if defined(__sun)
286  return futimesat(req->file, NULL, tv);
287# else
288  return futimes(req->file, tv);
289# endif
290#elif defined(__MVS__)
291  attrib_t atr;
292  memset(&atr, 0, sizeof(atr));
293  atr.att_mtimechg = 1;
294  atr.att_atimechg = 1;
295  atr.att_mtime = req->mtime;
296  atr.att_atime = req->atime;
297  return __fchattr(req->file, &atr, sizeof(atr));
298#else
299  errno = ENOSYS;
300  return -1;
301#endif
302}
303
304
305static ssize_t uv__fs_mkdtemp(uv_fs_t* req) {
306  return mkdtemp((char*) req->path) ? 0 : -1;
307}
308
309
310static int (*uv__mkostemp)(char*, int);
311
312
313static void uv__mkostemp_initonce(void) {
314  /* z/os doesn't have RTLD_DEFAULT but that's okay
315   * because it doesn't have mkostemp(O_CLOEXEC) either.
316   */
317#ifdef RTLD_DEFAULT
318  uv__mkostemp = (int (*)(char*, int)) dlsym(RTLD_DEFAULT, "mkostemp");
319
320  /* We don't care about errors, but we do want to clean them up.
321   * If there has been no error, then dlerror() will just return
322   * NULL.
323   */
324  dlerror();
325#endif  /* RTLD_DEFAULT */
326}
327
328
329static int uv__fs_mkstemp(uv_fs_t* req) {
330  static uv_once_t once = UV_ONCE_INIT;
331  int r;
332#ifdef O_CLOEXEC
333  static _Atomic int no_cloexec_support;
334#endif
335  static const char pattern[] = "XXXXXX";
336  static const size_t pattern_size = sizeof(pattern) - 1;
337  char* path;
338  size_t path_length;
339
340  path = (char*) req->path;
341  path_length = strlen(path);
342
343  /* EINVAL can be returned for 2 reasons:
344      1. The template's last 6 characters were not XXXXXX
345      2. open() didn't support O_CLOEXEC
346     We want to avoid going to the fallback path in case
347     of 1, so it's manually checked before. */
348  if (path_length < pattern_size ||
349      strcmp(path + path_length - pattern_size, pattern)) {
350    errno = EINVAL;
351    r = -1;
352    goto clobber;
353  }
354
355  uv_once(&once, uv__mkostemp_initonce);
356
357#ifdef O_CLOEXEC
358  if (atomic_load_explicit(&no_cloexec_support, memory_order_relaxed) == 0 &&
359      uv__mkostemp != NULL) {
360    r = uv__mkostemp(path, O_CLOEXEC);
361
362    if (r >= 0)
363      return r;
364
365    /* If mkostemp() returns EINVAL, it means the kernel doesn't
366       support O_CLOEXEC, so we just fallback to mkstemp() below. */
367    if (errno != EINVAL)
368      goto clobber;
369
370    /* We set the static variable so that next calls don't even
371       try to use mkostemp. */
372    atomic_store_explicit(&no_cloexec_support, 1, memory_order_relaxed);
373  }
374#endif  /* O_CLOEXEC */
375
376  if (req->cb != NULL)
377    uv_rwlock_rdlock(&req->loop->cloexec_lock);
378
379  r = mkstemp(path);
380
381  /* In case of failure `uv__cloexec` will leave error in `errno`,
382   * so it is enough to just set `r` to `-1`.
383   */
384  if (r >= 0 && uv__cloexec(r, 1) != 0) {
385    r = uv__close(r);
386    if (r != 0)
387      abort();
388    r = -1;
389  }
390
391  if (req->cb != NULL)
392    uv_rwlock_rdunlock(&req->loop->cloexec_lock);
393
394clobber:
395  if (r < 0)
396    path[0] = '\0';
397  return r;
398}
399
400
401static ssize_t uv__fs_open(uv_fs_t* req) {
402#ifdef O_CLOEXEC
403  return open(req->path, req->flags | O_CLOEXEC, req->mode);
404#else  /* O_CLOEXEC */
405  int r;
406
407  if (req->cb != NULL)
408    uv_rwlock_rdlock(&req->loop->cloexec_lock);
409
410  r = open(req->path, req->flags, req->mode);
411
412  /* In case of failure `uv__cloexec` will leave error in `errno`,
413   * so it is enough to just set `r` to `-1`.
414   */
415  if (r >= 0 && uv__cloexec(r, 1) != 0) {
416    r = uv__close(r);
417    if (r != 0)
418      abort();
419    r = -1;
420  }
421
422  if (req->cb != NULL)
423    uv_rwlock_rdunlock(&req->loop->cloexec_lock);
424
425  return r;
426#endif  /* O_CLOEXEC */
427}
428
429
430static ssize_t uv__fs_read(uv_fs_t* req) {
431  const struct iovec* bufs;
432  unsigned int iovmax;
433  size_t nbufs;
434  ssize_t r;
435  off_t off;
436  int fd;
437
438  fd = req->file;
439  off = req->off;
440  bufs = (const struct iovec*) req->bufs;
441  nbufs = req->nbufs;
442
443  iovmax = uv__getiovmax();
444  if (nbufs > iovmax)
445    nbufs = iovmax;
446
447  r = 0;
448  if (off < 0) {
449    if (nbufs == 1)
450      r = read(fd, bufs->iov_base, bufs->iov_len);
451    else if (nbufs > 1)
452      r = readv(fd, bufs, nbufs);
453  } else {
454    if (nbufs == 1)
455      r = pread(fd, bufs->iov_base, bufs->iov_len, off);
456    else if (nbufs > 1)
457      r = preadv(fd, bufs, nbufs, off);
458  }
459
460#ifdef __PASE__
461  /* PASE returns EOPNOTSUPP when reading a directory, convert to EISDIR */
462  if (r == -1 && errno == EOPNOTSUPP) {
463    struct stat buf;
464    ssize_t rc;
465    rc = uv__fstat(fd, &buf);
466    if (rc == 0 && S_ISDIR(buf.st_mode)) {
467      errno = EISDIR;
468    }
469  }
470#endif
471
472  /* We don't own the buffer list in the synchronous case. */
473  if (req->cb != NULL)
474    if (req->bufs != req->bufsml)
475      uv__free(req->bufs);
476
477  req->bufs = NULL;
478  req->nbufs = 0;
479
480  return r;
481}
482
483
484static int uv__fs_scandir_filter(const uv__dirent_t* dent) {
485  return strcmp(dent->d_name, ".") != 0 && strcmp(dent->d_name, "..") != 0;
486}
487
488
489static int uv__fs_scandir_sort(const uv__dirent_t** a, const uv__dirent_t** b) {
490  return strcmp((*a)->d_name, (*b)->d_name);
491}
492
493
494static ssize_t uv__fs_scandir(uv_fs_t* req) {
495  uv__dirent_t** dents;
496  int n;
497
498  dents = NULL;
499  n = scandir(req->path, &dents, uv__fs_scandir_filter, uv__fs_scandir_sort);
500
501  /* NOTE: We will use nbufs as an index field */
502  req->nbufs = 0;
503
504  if (n == 0) {
505    /* OS X still needs to deallocate some memory.
506     * Memory was allocated using the system allocator, so use free() here.
507     */
508    free(dents);
509    dents = NULL;
510  } else if (n == -1) {
511    return n;
512  }
513
514  req->ptr = dents;
515
516  return n;
517}
518
519static int uv__fs_opendir(uv_fs_t* req) {
520  uv_dir_t* dir;
521
522  dir = uv__malloc(sizeof(*dir));
523  if (dir == NULL)
524    goto error;
525
526  dir->dir = opendir(req->path);
527  if (dir->dir == NULL)
528    goto error;
529
530  req->ptr = dir;
531  return 0;
532
533error:
534  uv__free(dir);
535  req->ptr = NULL;
536  return -1;
537}
538
539static int uv__fs_readdir(uv_fs_t* req) {
540  uv_dir_t* dir;
541  uv_dirent_t* dirent;
542  struct dirent* res;
543  unsigned int dirent_idx;
544  unsigned int i;
545
546  dir = req->ptr;
547  dirent_idx = 0;
548
549  while (dirent_idx < dir->nentries) {
550    /* readdir() returns NULL on end of directory, as well as on error. errno
551       is used to differentiate between the two conditions. */
552    errno = 0;
553    res = readdir(dir->dir);
554
555    if (res == NULL) {
556      if (errno != 0)
557        goto error;
558      break;
559    }
560
561    if (strcmp(res->d_name, ".") == 0 || strcmp(res->d_name, "..") == 0)
562      continue;
563
564    dirent = &dir->dirents[dirent_idx];
565    dirent->name = uv__strdup(res->d_name);
566
567    if (dirent->name == NULL)
568      goto error;
569
570    dirent->type = uv__fs_get_dirent_type(res);
571    ++dirent_idx;
572  }
573
574  return dirent_idx;
575
576error:
577  for (i = 0; i < dirent_idx; ++i) {
578    uv__free((char*) dir->dirents[i].name);
579    dir->dirents[i].name = NULL;
580  }
581
582  return -1;
583}
584
585static int uv__fs_closedir(uv_fs_t* req) {
586  uv_dir_t* dir;
587
588  dir = req->ptr;
589
590  if (dir->dir != NULL) {
591    closedir(dir->dir);
592    dir->dir = NULL;
593  }
594
595  uv__free(req->ptr);
596  req->ptr = NULL;
597  return 0;
598}
599
600static int uv__fs_statfs(uv_fs_t* req) {
601  uv_statfs_t* stat_fs;
602#if defined(__sun)      || \
603    defined(__MVS__)    || \
604    defined(__NetBSD__) || \
605    defined(__HAIKU__)  || \
606    defined(__QNX__)
607  struct statvfs buf;
608
609  if (0 != statvfs(req->path, &buf))
610#else
611  struct statfs buf;
612
613  if (0 != statfs(req->path, &buf))
614#endif /* defined(__sun) */
615    return -1;
616
617  stat_fs = uv__malloc(sizeof(*stat_fs));
618  if (stat_fs == NULL) {
619    errno = ENOMEM;
620    return -1;
621  }
622
623#if defined(__sun)        || \
624    defined(__MVS__)      || \
625    defined(__OpenBSD__)  || \
626    defined(__NetBSD__)   || \
627    defined(__HAIKU__)    || \
628    defined(__QNX__)
629  stat_fs->f_type = 0;  /* f_type is not supported. */
630#else
631  stat_fs->f_type = buf.f_type;
632#endif
633  stat_fs->f_bsize = buf.f_bsize;
634  stat_fs->f_blocks = buf.f_blocks;
635  stat_fs->f_bfree = buf.f_bfree;
636  stat_fs->f_bavail = buf.f_bavail;
637  stat_fs->f_files = buf.f_files;
638  stat_fs->f_ffree = buf.f_ffree;
639  req->ptr = stat_fs;
640  return 0;
641}
642
643static ssize_t uv__fs_pathmax_size(const char* path) {
644  ssize_t pathmax;
645
646  pathmax = pathconf(path, _PC_PATH_MAX);
647
648  if (pathmax == -1)
649    pathmax = UV__PATH_MAX;
650
651  return pathmax;
652}
653
654static ssize_t uv__fs_readlink(uv_fs_t* req) {
655  ssize_t maxlen;
656  ssize_t len;
657  char* buf;
658
659#if defined(_POSIX_PATH_MAX) || defined(PATH_MAX)
660  maxlen = uv__fs_pathmax_size(req->path);
661#else
662  /* We may not have a real PATH_MAX.  Read size of link.  */
663  struct stat st;
664  int ret;
665  ret = uv__lstat(req->path, &st);
666  if (ret != 0)
667    return -1;
668  if (!S_ISLNK(st.st_mode)) {
669    errno = EINVAL;
670    return -1;
671  }
672
673  maxlen = st.st_size;
674
675  /* According to readlink(2) lstat can report st_size == 0
676     for some symlinks, such as those in /proc or /sys.  */
677  if (maxlen == 0)
678    maxlen = uv__fs_pathmax_size(req->path);
679#endif
680
681  buf = uv__malloc(maxlen);
682
683  if (buf == NULL) {
684    errno = ENOMEM;
685    return -1;
686  }
687
688#if defined(__MVS__)
689  len = os390_readlink(req->path, buf, maxlen);
690#else
691  len = readlink(req->path, buf, maxlen);
692#endif
693
694  if (len == -1) {
695    uv__free(buf);
696    return -1;
697  }
698
699  /* Uncommon case: resize to make room for the trailing nul byte. */
700  if (len == maxlen) {
701    buf = uv__reallocf(buf, len + 1);
702
703    if (buf == NULL)
704      return -1;
705  }
706
707  buf[len] = '\0';
708  req->ptr = buf;
709
710  return 0;
711}
712
713static ssize_t uv__fs_realpath(uv_fs_t* req) {
714  char* buf;
715
716#if defined(_POSIX_VERSION) && _POSIX_VERSION >= 200809L
717  buf = realpath(req->path, NULL);
718  if (buf == NULL)
719    return -1;
720#else
721  ssize_t len;
722
723  len = uv__fs_pathmax_size(req->path);
724  buf = uv__malloc(len + 1);
725
726  if (buf == NULL) {
727    errno = ENOMEM;
728    return -1;
729  }
730
731  if (realpath(req->path, buf) == NULL) {
732    uv__free(buf);
733    return -1;
734  }
735#endif
736
737  req->ptr = buf;
738
739  return 0;
740}
741
742static ssize_t uv__fs_sendfile_emul(uv_fs_t* req) {
743  struct pollfd pfd;
744  int use_pread;
745  off_t offset;
746  ssize_t nsent;
747  ssize_t nread;
748  ssize_t nwritten;
749  size_t buflen;
750  size_t len;
751  ssize_t n;
752  int in_fd;
753  int out_fd;
754  char buf[8192];
755
756  len = req->bufsml[0].len;
757  in_fd = req->flags;
758  out_fd = req->file;
759  offset = req->off;
760  use_pread = 1;
761
762  /* Here are the rules regarding errors:
763   *
764   * 1. Read errors are reported only if nsent==0, otherwise we return nsent.
765   *    The user needs to know that some data has already been sent, to stop
766   *    them from sending it twice.
767   *
768   * 2. Write errors are always reported. Write errors are bad because they
769   *    mean data loss: we've read data but now we can't write it out.
770   *
771   * We try to use pread() and fall back to regular read() if the source fd
772   * doesn't support positional reads, for example when it's a pipe fd.
773   *
774   * If we get EAGAIN when writing to the target fd, we poll() on it until
775   * it becomes writable again.
776   *
777   * FIXME: If we get a write error when use_pread==1, it should be safe to
778   *        return the number of sent bytes instead of an error because pread()
779   *        is, in theory, idempotent. However, special files in /dev or /proc
780   *        may support pread() but not necessarily return the same data on
781   *        successive reads.
782   *
783   * FIXME: There is no way now to signal that we managed to send *some* data
784   *        before a write error.
785   */
786  for (nsent = 0; (size_t) nsent < len; ) {
787    buflen = len - nsent;
788
789    if (buflen > sizeof(buf))
790      buflen = sizeof(buf);
791
792    do
793      if (use_pread)
794        nread = pread(in_fd, buf, buflen, offset);
795      else
796        nread = read(in_fd, buf, buflen);
797    while (nread == -1 && errno == EINTR);
798
799    if (nread == 0)
800      goto out;
801
802    if (nread == -1) {
803      if (use_pread && nsent == 0 && (errno == EIO || errno == ESPIPE)) {
804        use_pread = 0;
805        continue;
806      }
807
808      if (nsent == 0)
809        nsent = -1;
810
811      goto out;
812    }
813
814    for (nwritten = 0; nwritten < nread; ) {
815      do
816        n = write(out_fd, buf + nwritten, nread - nwritten);
817      while (n == -1 && errno == EINTR);
818
819      if (n != -1) {
820        nwritten += n;
821        continue;
822      }
823
824      if (errno != EAGAIN && errno != EWOULDBLOCK) {
825        nsent = -1;
826        goto out;
827      }
828
829      pfd.fd = out_fd;
830      pfd.events = POLLOUT;
831      pfd.revents = 0;
832
833      do
834        n = poll(&pfd, 1, -1);
835      while (n == -1 && errno == EINTR);
836
837      if (n == -1 || (pfd.revents & ~POLLOUT) != 0) {
838        errno = EIO;
839        nsent = -1;
840        goto out;
841      }
842    }
843
844    offset += nread;
845    nsent += nread;
846  }
847
848out:
849  if (nsent != -1)
850    req->off = offset;
851
852  return nsent;
853}
854
855
856#ifdef __linux__
857/* Pre-4.20 kernels have a bug where CephFS uses the RADOS copy-from command
858 * in copy_file_range() when it shouldn't. There is no workaround except to
859 * fall back to a regular copy.
860 */
861static int uv__is_buggy_cephfs(int fd) {
862  struct statfs s;
863
864  if (-1 == fstatfs(fd, &s))
865    return 0;
866
867  if (s.f_type != /* CephFS */ 0xC36400)
868    return 0;
869
870  return uv__kernel_version() < /* 4.20.0 */ 0x041400;
871}
872
873
874static int uv__is_cifs_or_smb(int fd) {
875  struct statfs s;
876
877  if (-1 == fstatfs(fd, &s))
878    return 0;
879
880  switch ((unsigned) s.f_type) {
881  case 0x0000517Bu:  /* SMB */
882  case 0xFE534D42u:  /* SMB2 */
883  case 0xFF534D42u:  /* CIFS */
884    return 1;
885  }
886
887  return 0;
888}
889
890
891static ssize_t uv__fs_try_copy_file_range(int in_fd, off_t* off,
892                                          int out_fd, size_t len) {
893  static _Atomic int no_copy_file_range_support;
894  ssize_t r;
895
896  if (atomic_load_explicit(&no_copy_file_range_support, memory_order_relaxed)) {
897    errno = ENOSYS;
898    return -1;
899  }
900
901  r = uv__fs_copy_file_range(in_fd, off, out_fd, NULL, len, 0);
902
903  if (r != -1)
904    return r;
905
906  switch (errno) {
907  case EACCES:
908    /* Pre-4.20 kernels have a bug where CephFS uses the RADOS
909     * copy-from command when it shouldn't.
910     */
911    if (uv__is_buggy_cephfs(in_fd))
912      errno = ENOSYS;  /* Use fallback. */
913    break;
914  case ENOSYS:
915    atomic_store_explicit(&no_copy_file_range_support, 1, memory_order_relaxed);
916    break;
917  case EPERM:
918    /* It's been reported that CIFS spuriously fails.
919     * Consider it a transient error.
920     */
921    if (uv__is_cifs_or_smb(out_fd))
922      errno = ENOSYS;  /* Use fallback. */
923    break;
924  case ENOTSUP:
925  case EXDEV:
926    /* ENOTSUP - it could work on another file system type.
927     * EXDEV - it will not work when in_fd and out_fd are not on the same
928     *         mounted filesystem (pre Linux 5.3)
929     */
930    errno = ENOSYS;  /* Use fallback. */
931    break;
932  }
933
934  return -1;
935}
936
937#endif  /* __linux__ */
938
939
940static ssize_t uv__fs_sendfile(uv_fs_t* req) {
941  int in_fd;
942  int out_fd;
943
944  in_fd = req->flags;
945  out_fd = req->file;
946
947#if defined(__linux__) || defined(__sun)
948  {
949    off_t off;
950    ssize_t r;
951    size_t len;
952    int try_sendfile;
953
954    off = req->off;
955    len = req->bufsml[0].len;
956    try_sendfile = 1;
957
958#ifdef __linux__
959    r = uv__fs_try_copy_file_range(in_fd, &off, out_fd, len);
960    try_sendfile = (r == -1 && errno == ENOSYS);
961#endif
962
963    if (try_sendfile)
964      r = sendfile(out_fd, in_fd, &off, len);
965
966    /* sendfile() on SunOS returns EINVAL if the target fd is not a socket but
967     * it still writes out data. Fortunately, we can detect it by checking if
968     * the offset has been updated.
969     */
970    if (r != -1 || off > req->off) {
971      r = off - req->off;
972      req->off = off;
973      return r;
974    }
975
976    if (errno == EINVAL ||
977        errno == EIO ||
978        errno == ENOTSOCK ||
979        errno == EXDEV) {
980      errno = 0;
981      return uv__fs_sendfile_emul(req);
982    }
983
984    return -1;
985  }
986#elif defined(__APPLE__) || defined(__DragonFly__) || defined(__FreeBSD__)
987  {
988    off_t len;
989    ssize_t r;
990
991    /* sendfile() on FreeBSD and Darwin returns EAGAIN if the target fd is in
992     * non-blocking mode and not all data could be written. If a non-zero
993     * number of bytes have been sent, we don't consider it an error.
994     */
995
996#if defined(__FreeBSD__) || defined(__DragonFly__)
997#if defined(__FreeBSD__)
998    off_t off;
999
1000    off = req->off;
1001    r = uv__fs_copy_file_range(in_fd, &off, out_fd, NULL, req->bufsml[0].len, 0);
1002    if (r >= 0) {
1003        r = off - req->off;
1004        req->off = off;
1005        return r;
1006    }
1007#endif
1008    len = 0;
1009    r = sendfile(in_fd, out_fd, req->off, req->bufsml[0].len, NULL, &len, 0);
1010#else
1011    /* The darwin sendfile takes len as an input for the length to send,
1012     * so make sure to initialize it with the caller's value. */
1013    len = req->bufsml[0].len;
1014    r = sendfile(in_fd, out_fd, req->off, &len, NULL, 0);
1015#endif
1016
1017     /*
1018     * The man page for sendfile(2) on DragonFly states that `len` contains
1019     * a meaningful value ONLY in case of EAGAIN and EINTR.
1020     * Nothing is said about it's value in case of other errors, so better
1021     * not depend on the potential wrong assumption that is was not modified
1022     * by the syscall.
1023     */
1024    if (r == 0 || ((errno == EAGAIN || errno == EINTR) && len != 0)) {
1025      req->off += len;
1026      return (ssize_t) len;
1027    }
1028
1029    if (errno == EINVAL ||
1030        errno == EIO ||
1031        errno == ENOTSOCK ||
1032        errno == EXDEV) {
1033      errno = 0;
1034      return uv__fs_sendfile_emul(req);
1035    }
1036
1037    return -1;
1038  }
1039#else
1040  /* Squelch compiler warnings. */
1041  (void) &in_fd;
1042  (void) &out_fd;
1043
1044  return uv__fs_sendfile_emul(req);
1045#endif
1046}
1047
1048
1049static ssize_t uv__fs_utime(uv_fs_t* req) {
1050#if defined(__linux__)                                                         \
1051    || defined(_AIX71)                                                         \
1052    || defined(__sun)                                                          \
1053    || defined(__HAIKU__)
1054  struct timespec ts[2];
1055  ts[0] = uv__fs_to_timespec(req->atime);
1056  ts[1] = uv__fs_to_timespec(req->mtime);
1057  return utimensat(AT_FDCWD, req->path, ts, 0);
1058#elif defined(__APPLE__)                                                      \
1059    || defined(__DragonFly__)                                                 \
1060    || defined(__FreeBSD__)                                                   \
1061    || defined(__NetBSD__)                                                    \
1062    || defined(__OpenBSD__)
1063  struct timeval tv[2];
1064  tv[0] = uv__fs_to_timeval(req->atime);
1065  tv[1] = uv__fs_to_timeval(req->mtime);
1066  return utimes(req->path, tv);
1067#elif defined(_AIX)                                                           \
1068    && !defined(_AIX71)
1069  struct utimbuf buf;
1070  buf.actime = req->atime;
1071  buf.modtime = req->mtime;
1072  return utime(req->path, &buf);
1073#elif defined(__MVS__)
1074  attrib_t atr;
1075  memset(&atr, 0, sizeof(atr));
1076  atr.att_mtimechg = 1;
1077  atr.att_atimechg = 1;
1078  atr.att_mtime = req->mtime;
1079  atr.att_atime = req->atime;
1080  return __lchattr((char*) req->path, &atr, sizeof(atr));
1081#else
1082  errno = ENOSYS;
1083  return -1;
1084#endif
1085}
1086
1087
1088static ssize_t uv__fs_lutime(uv_fs_t* req) {
1089#if defined(__linux__)            ||                                           \
1090    defined(_AIX71)               ||                                           \
1091    defined(__sun)                ||                                           \
1092    defined(__HAIKU__)            ||                                           \
1093    defined(__GNU__)              ||                                           \
1094    defined(__OpenBSD__)
1095  struct timespec ts[2];
1096  ts[0] = uv__fs_to_timespec(req->atime);
1097  ts[1] = uv__fs_to_timespec(req->mtime);
1098  return utimensat(AT_FDCWD, req->path, ts, AT_SYMLINK_NOFOLLOW);
1099#elif defined(__APPLE__)          ||                                          \
1100      defined(__DragonFly__)      ||                                          \
1101      defined(__FreeBSD__)        ||                                          \
1102      defined(__NetBSD__)
1103  struct timeval tv[2];
1104  tv[0] = uv__fs_to_timeval(req->atime);
1105  tv[1] = uv__fs_to_timeval(req->mtime);
1106  return lutimes(req->path, tv);
1107#else
1108  errno = ENOSYS;
1109  return -1;
1110#endif
1111}
1112
1113
1114static ssize_t uv__fs_write(uv_fs_t* req) {
1115  const struct iovec* bufs;
1116  size_t nbufs;
1117  ssize_t r;
1118  off_t off;
1119  int fd;
1120
1121  fd = req->file;
1122  off = req->off;
1123  bufs = (const struct iovec*) req->bufs;
1124  nbufs = req->nbufs;
1125
1126  r = 0;
1127  if (off < 0) {
1128    if (nbufs == 1)
1129      r = write(fd, bufs->iov_base, bufs->iov_len);
1130    else if (nbufs > 1)
1131      r = writev(fd, bufs, nbufs);
1132  } else {
1133    if (nbufs == 1)
1134      r = pwrite(fd, bufs->iov_base, bufs->iov_len, off);
1135    else if (nbufs > 1)
1136      r = pwritev(fd, bufs, nbufs, off);
1137  }
1138
1139  return r;
1140}
1141
1142
1143static ssize_t uv__fs_copyfile(uv_fs_t* req) {
1144  uv_fs_t fs_req;
1145  uv_file srcfd;
1146  uv_file dstfd;
1147  struct stat src_statsbuf;
1148  struct stat dst_statsbuf;
1149  int dst_flags;
1150  int result;
1151  int err;
1152  off_t bytes_to_send;
1153  off_t in_offset;
1154  off_t bytes_written;
1155  size_t bytes_chunk;
1156
1157  dstfd = -1;
1158  err = 0;
1159
1160  /* Open the source file. */
1161  srcfd = uv_fs_open(NULL, &fs_req, req->path, O_RDONLY, 0, NULL);
1162  uv_fs_req_cleanup(&fs_req);
1163
1164  if (srcfd < 0)
1165    return srcfd;
1166
1167  /* Get the source file's mode. */
1168  if (uv__fstat(srcfd, &src_statsbuf)) {
1169    err = UV__ERR(errno);
1170    goto out;
1171  }
1172
1173  dst_flags = O_WRONLY | O_CREAT;
1174
1175  if (req->flags & UV_FS_COPYFILE_EXCL)
1176    dst_flags |= O_EXCL;
1177
1178  /* Open the destination file. */
1179  dstfd = uv_fs_open(NULL,
1180                     &fs_req,
1181                     req->new_path,
1182                     dst_flags,
1183                     src_statsbuf.st_mode,
1184                     NULL);
1185  uv_fs_req_cleanup(&fs_req);
1186
1187  if (dstfd < 0) {
1188    err = dstfd;
1189    goto out;
1190  }
1191
1192  /* If the file is not being opened exclusively, verify that the source and
1193     destination are not the same file. If they are the same, bail out early. */
1194  if ((req->flags & UV_FS_COPYFILE_EXCL) == 0) {
1195    /* Get the destination file's mode. */
1196    if (uv__fstat(dstfd, &dst_statsbuf)) {
1197      err = UV__ERR(errno);
1198      goto out;
1199    }
1200
1201    /* Check if srcfd and dstfd refer to the same file */
1202    if (src_statsbuf.st_dev == dst_statsbuf.st_dev &&
1203        src_statsbuf.st_ino == dst_statsbuf.st_ino) {
1204      goto out;
1205    }
1206
1207    /* Truncate the file in case the destination already existed. */
1208    if (ftruncate(dstfd, 0) != 0) {
1209      err = UV__ERR(errno);
1210
1211      /* ftruncate() on ceph-fuse fails with EACCES when the file is created
1212       * with read only permissions. Since ftruncate() on a newly created
1213       * file is a meaningless operation anyway, detect that condition
1214       * and squelch the error.
1215       */
1216      if (err != UV_EACCES)
1217        goto out;
1218
1219      if (dst_statsbuf.st_size > 0)
1220        goto out;
1221
1222      err = 0;
1223    }
1224  }
1225
1226  if (fchmod(dstfd, src_statsbuf.st_mode) == -1) {
1227    err = UV__ERR(errno);
1228#ifdef __linux__
1229    /* fchmod() on CIFS shares always fails with EPERM unless the share is
1230     * mounted with "noperm". As fchmod() is a meaningless operation on such
1231     * shares anyway, detect that condition and squelch the error.
1232     */
1233    if (err != UV_EPERM)
1234      goto out;
1235
1236    if (!uv__is_cifs_or_smb(dstfd))
1237      goto out;
1238
1239    err = 0;
1240#else  /* !__linux__ */
1241    goto out;
1242#endif  /* !__linux__ */
1243  }
1244
1245#ifdef FICLONE
1246  if (req->flags & UV_FS_COPYFILE_FICLONE ||
1247      req->flags & UV_FS_COPYFILE_FICLONE_FORCE) {
1248    if (ioctl(dstfd, FICLONE, srcfd) == 0) {
1249      /* ioctl() with FICLONE succeeded. */
1250      goto out;
1251    }
1252    /* If an error occurred and force was set, return the error to the caller;
1253     * fall back to sendfile() when force was not set. */
1254    if (req->flags & UV_FS_COPYFILE_FICLONE_FORCE) {
1255      err = UV__ERR(errno);
1256      goto out;
1257    }
1258  }
1259#else
1260  if (req->flags & UV_FS_COPYFILE_FICLONE_FORCE) {
1261    err = UV_ENOSYS;
1262    goto out;
1263  }
1264#endif
1265
1266  bytes_to_send = src_statsbuf.st_size;
1267  in_offset = 0;
1268  while (bytes_to_send != 0) {
1269    bytes_chunk = SSIZE_MAX;
1270    if (bytes_to_send < (off_t) bytes_chunk)
1271      bytes_chunk = bytes_to_send;
1272    uv_fs_sendfile(NULL, &fs_req, dstfd, srcfd, in_offset, bytes_chunk, NULL);
1273    bytes_written = fs_req.result;
1274    uv_fs_req_cleanup(&fs_req);
1275
1276    if (bytes_written < 0) {
1277      err = bytes_written;
1278      break;
1279    }
1280
1281    bytes_to_send -= bytes_written;
1282    in_offset += bytes_written;
1283  }
1284
1285out:
1286  if (err < 0)
1287    result = err;
1288  else
1289    result = 0;
1290
1291  /* Close the source file. */
1292  err = uv__close_nocheckstdio(srcfd);
1293
1294  /* Don't overwrite any existing errors. */
1295  if (err != 0 && result == 0)
1296    result = err;
1297
1298  /* Close the destination file if it is open. */
1299  if (dstfd >= 0) {
1300    err = uv__close_nocheckstdio(dstfd);
1301
1302    /* Don't overwrite any existing errors. */
1303    if (err != 0 && result == 0)
1304      result = err;
1305
1306    /* Remove the destination file if something went wrong. */
1307    if (result != 0) {
1308      uv_fs_unlink(NULL, &fs_req, req->new_path, NULL);
1309      /* Ignore the unlink return value, as an error already happened. */
1310      uv_fs_req_cleanup(&fs_req);
1311    }
1312  }
1313
1314  if (result == 0)
1315    return 0;
1316
1317  errno = UV__ERR(result);
1318  return -1;
1319}
1320
1321static void uv__to_stat(struct stat* src, uv_stat_t* dst) {
1322  dst->st_dev = src->st_dev;
1323  dst->st_mode = src->st_mode;
1324  dst->st_nlink = src->st_nlink;
1325  dst->st_uid = src->st_uid;
1326  dst->st_gid = src->st_gid;
1327  dst->st_rdev = src->st_rdev;
1328  dst->st_ino = src->st_ino;
1329  dst->st_size = src->st_size;
1330  dst->st_blksize = src->st_blksize;
1331  dst->st_blocks = src->st_blocks;
1332
1333#if defined(__APPLE__)
1334  dst->st_atim.tv_sec = src->st_atimespec.tv_sec;
1335  dst->st_atim.tv_nsec = src->st_atimespec.tv_nsec;
1336  dst->st_mtim.tv_sec = src->st_mtimespec.tv_sec;
1337  dst->st_mtim.tv_nsec = src->st_mtimespec.tv_nsec;
1338  dst->st_ctim.tv_sec = src->st_ctimespec.tv_sec;
1339  dst->st_ctim.tv_nsec = src->st_ctimespec.tv_nsec;
1340  dst->st_birthtim.tv_sec = src->st_birthtimespec.tv_sec;
1341  dst->st_birthtim.tv_nsec = src->st_birthtimespec.tv_nsec;
1342  dst->st_flags = src->st_flags;
1343  dst->st_gen = src->st_gen;
1344#elif defined(__ANDROID__)
1345  dst->st_atim.tv_sec = src->st_atime;
1346  dst->st_atim.tv_nsec = src->st_atimensec;
1347  dst->st_mtim.tv_sec = src->st_mtime;
1348  dst->st_mtim.tv_nsec = src->st_mtimensec;
1349  dst->st_ctim.tv_sec = src->st_ctime;
1350  dst->st_ctim.tv_nsec = src->st_ctimensec;
1351  dst->st_birthtim.tv_sec = src->st_ctime;
1352  dst->st_birthtim.tv_nsec = src->st_ctimensec;
1353  dst->st_flags = 0;
1354  dst->st_gen = 0;
1355#elif !defined(_AIX) &&         \
1356    !defined(__MVS__) && (      \
1357    defined(__DragonFly__)   || \
1358    defined(__FreeBSD__)     || \
1359    defined(__OpenBSD__)     || \
1360    defined(__NetBSD__)      || \
1361    defined(_GNU_SOURCE)     || \
1362    defined(_BSD_SOURCE)     || \
1363    defined(_SVID_SOURCE)    || \
1364    defined(_XOPEN_SOURCE)   || \
1365    defined(_DEFAULT_SOURCE))
1366  dst->st_atim.tv_sec = src->st_atim.tv_sec;
1367  dst->st_atim.tv_nsec = src->st_atim.tv_nsec;
1368  dst->st_mtim.tv_sec = src->st_mtim.tv_sec;
1369  dst->st_mtim.tv_nsec = src->st_mtim.tv_nsec;
1370  dst->st_ctim.tv_sec = src->st_ctim.tv_sec;
1371  dst->st_ctim.tv_nsec = src->st_ctim.tv_nsec;
1372# if defined(__FreeBSD__)    || \
1373     defined(__NetBSD__)
1374  dst->st_birthtim.tv_sec = src->st_birthtim.tv_sec;
1375  dst->st_birthtim.tv_nsec = src->st_birthtim.tv_nsec;
1376  dst->st_flags = src->st_flags;
1377  dst->st_gen = src->st_gen;
1378# else
1379  dst->st_birthtim.tv_sec = src->st_ctim.tv_sec;
1380  dst->st_birthtim.tv_nsec = src->st_ctim.tv_nsec;
1381  dst->st_flags = 0;
1382  dst->st_gen = 0;
1383# endif
1384#else
1385  dst->st_atim.tv_sec = src->st_atime;
1386  dst->st_atim.tv_nsec = 0;
1387  dst->st_mtim.tv_sec = src->st_mtime;
1388  dst->st_mtim.tv_nsec = 0;
1389  dst->st_ctim.tv_sec = src->st_ctime;
1390  dst->st_ctim.tv_nsec = 0;
1391  dst->st_birthtim.tv_sec = src->st_ctime;
1392  dst->st_birthtim.tv_nsec = 0;
1393  dst->st_flags = 0;
1394  dst->st_gen = 0;
1395#endif
1396}
1397
1398
1399static int uv__fs_statx(int fd,
1400                        const char* path,
1401                        int is_fstat,
1402                        int is_lstat,
1403                        uv_stat_t* buf) {
1404  STATIC_ASSERT(UV_ENOSYS != -1);
1405#ifdef __linux__
1406  static _Atomic int no_statx;
1407  struct uv__statx statxbuf;
1408  int dirfd;
1409  int flags;
1410  int mode;
1411  int rc;
1412
1413  if (atomic_load_explicit(&no_statx, memory_order_relaxed))
1414    return UV_ENOSYS;
1415
1416  dirfd = AT_FDCWD;
1417  flags = 0; /* AT_STATX_SYNC_AS_STAT */
1418  mode = 0xFFF; /* STATX_BASIC_STATS + STATX_BTIME */
1419
1420  if (is_fstat) {
1421    dirfd = fd;
1422    flags |= 0x1000; /* AT_EMPTY_PATH */
1423  }
1424
1425  if (is_lstat)
1426    flags |= AT_SYMLINK_NOFOLLOW;
1427
1428  rc = uv__statx(dirfd, path, flags, mode, &statxbuf);
1429
1430  switch (rc) {
1431  case 0:
1432    break;
1433  case -1:
1434    /* EPERM happens when a seccomp filter rejects the system call.
1435     * Has been observed with libseccomp < 2.3.3 and docker < 18.04.
1436     * EOPNOTSUPP is used on DVS exported filesystems
1437     */
1438    if (errno != EINVAL && errno != EPERM && errno != ENOSYS && errno != EOPNOTSUPP)
1439      return -1;
1440    /* Fall through. */
1441  default:
1442    /* Normally on success, zero is returned and On error, -1 is returned.
1443     * Observed on S390 RHEL running in a docker container with statx not
1444     * implemented, rc might return 1 with 0 set as the error code in which
1445     * case we return ENOSYS.
1446     */
1447    atomic_store_explicit(&no_statx, 1, memory_order_relaxed);
1448    return UV_ENOSYS;
1449  }
1450
1451  uv__statx_to_stat(&statxbuf, buf);
1452
1453  return 0;
1454#else
1455  return UV_ENOSYS;
1456#endif /* __linux__ */
1457}
1458
1459
1460static int uv__fs_stat(const char *path, uv_stat_t *buf) {
1461  struct stat pbuf;
1462  int ret;
1463
1464  ret = uv__fs_statx(-1, path, /* is_fstat */ 0, /* is_lstat */ 0, buf);
1465  if (ret != UV_ENOSYS)
1466    return ret;
1467
1468  ret = uv__stat(path, &pbuf);
1469  if (ret == 0)
1470    uv__to_stat(&pbuf, buf);
1471
1472  return ret;
1473}
1474
1475
1476static int uv__fs_lstat(const char *path, uv_stat_t *buf) {
1477  struct stat pbuf;
1478  int ret;
1479
1480  ret = uv__fs_statx(-1, path, /* is_fstat */ 0, /* is_lstat */ 1, buf);
1481  if (ret != UV_ENOSYS)
1482    return ret;
1483
1484  ret = uv__lstat(path, &pbuf);
1485  if (ret == 0)
1486    uv__to_stat(&pbuf, buf);
1487
1488  return ret;
1489}
1490
1491
1492static int uv__fs_fstat(int fd, uv_stat_t *buf) {
1493  struct stat pbuf;
1494  int ret;
1495
1496  ret = uv__fs_statx(fd, "", /* is_fstat */ 1, /* is_lstat */ 0, buf);
1497  if (ret != UV_ENOSYS)
1498    return ret;
1499
1500  ret = uv__fstat(fd, &pbuf);
1501  if (ret == 0)
1502    uv__to_stat(&pbuf, buf);
1503
1504  return ret;
1505}
1506
1507static size_t uv__fs_buf_offset(uv_buf_t* bufs, size_t size) {
1508  size_t offset;
1509  /* Figure out which bufs are done */
1510  for (offset = 0; size > 0 && bufs[offset].len <= size; ++offset)
1511    size -= bufs[offset].len;
1512
1513  /* Fix a partial read/write */
1514  if (size > 0) {
1515    bufs[offset].base += size;
1516    bufs[offset].len -= size;
1517  }
1518  return offset;
1519}
1520
1521static ssize_t uv__fs_write_all(uv_fs_t* req) {
1522  unsigned int iovmax;
1523  unsigned int nbufs;
1524  uv_buf_t* bufs;
1525  ssize_t total;
1526  ssize_t result;
1527
1528  iovmax = uv__getiovmax();
1529  nbufs = req->nbufs;
1530  bufs = req->bufs;
1531  total = 0;
1532
1533  while (nbufs > 0) {
1534    req->nbufs = nbufs;
1535    if (req->nbufs > iovmax)
1536      req->nbufs = iovmax;
1537
1538    do
1539      result = uv__fs_write(req);
1540    while (result < 0 && errno == EINTR);
1541
1542    if (result <= 0) {
1543      if (total == 0)
1544        total = result;
1545      break;
1546    }
1547
1548    if (req->off >= 0)
1549      req->off += result;
1550
1551    req->nbufs = uv__fs_buf_offset(req->bufs, result);
1552    req->bufs += req->nbufs;
1553    nbufs -= req->nbufs;
1554    total += result;
1555  }
1556
1557  if (bufs != req->bufsml)
1558    uv__free(bufs);
1559
1560  req->bufs = NULL;
1561  req->nbufs = 0;
1562
1563  return total;
1564}
1565
1566
1567static void uv__fs_work(struct uv__work* w) {
1568  int retry_on_eintr;
1569  uv_fs_t* req;
1570  ssize_t r;
1571
1572  req = container_of(w, uv_fs_t, work_req);
1573  retry_on_eintr = !(req->fs_type == UV_FS_CLOSE ||
1574                     req->fs_type == UV_FS_READ);
1575
1576  do {
1577    errno = 0;
1578
1579#define X(type, action)                                                       \
1580  case UV_FS_ ## type:                                                        \
1581    r = action;                                                               \
1582    break;
1583
1584    switch (req->fs_type) {
1585    X(ACCESS, access(req->path, req->flags));
1586    X(CHMOD, chmod(req->path, req->mode));
1587    X(CHOWN, chown(req->path, req->uid, req->gid));
1588    X(CLOSE, uv__fs_close(req->file));
1589    X(COPYFILE, uv__fs_copyfile(req));
1590    X(FCHMOD, fchmod(req->file, req->mode));
1591    X(FCHOWN, fchown(req->file, req->uid, req->gid));
1592    X(LCHOWN, lchown(req->path, req->uid, req->gid));
1593    X(FDATASYNC, uv__fs_fdatasync(req));
1594    X(FSTAT, uv__fs_fstat(req->file, &req->statbuf));
1595    X(FSYNC, uv__fs_fsync(req));
1596    X(FTRUNCATE, ftruncate(req->file, req->off));
1597    X(FUTIME, uv__fs_futime(req));
1598    X(LUTIME, uv__fs_lutime(req));
1599    X(LSTAT, uv__fs_lstat(req->path, &req->statbuf));
1600    X(LINK, link(req->path, req->new_path));
1601    X(MKDIR, mkdir(req->path, req->mode));
1602    X(MKDTEMP, uv__fs_mkdtemp(req));
1603    X(MKSTEMP, uv__fs_mkstemp(req));
1604    X(OPEN, uv__fs_open(req));
1605    X(READ, uv__fs_read(req));
1606    X(SCANDIR, uv__fs_scandir(req));
1607    X(OPENDIR, uv__fs_opendir(req));
1608    X(READDIR, uv__fs_readdir(req));
1609    X(CLOSEDIR, uv__fs_closedir(req));
1610    X(READLINK, uv__fs_readlink(req));
1611    X(REALPATH, uv__fs_realpath(req));
1612    X(RENAME, rename(req->path, req->new_path));
1613    X(RMDIR, rmdir(req->path));
1614    X(SENDFILE, uv__fs_sendfile(req));
1615    X(STAT, uv__fs_stat(req->path, &req->statbuf));
1616    X(STATFS, uv__fs_statfs(req));
1617    X(SYMLINK, symlink(req->path, req->new_path));
1618    X(UNLINK, unlink(req->path));
1619    X(UTIME, uv__fs_utime(req));
1620    X(WRITE, uv__fs_write_all(req));
1621    default: abort();
1622    }
1623#undef X
1624  } while (r == -1 && errno == EINTR && retry_on_eintr);
1625
1626  if (r == -1)
1627    req->result = UV__ERR(errno);
1628  else
1629    req->result = r;
1630
1631  if (r == 0 && (req->fs_type == UV_FS_STAT ||
1632                 req->fs_type == UV_FS_FSTAT ||
1633                 req->fs_type == UV_FS_LSTAT)) {
1634    req->ptr = &req->statbuf;
1635  }
1636}
1637
1638
1639static void uv__fs_done(struct uv__work* w, int status) {
1640  uv_fs_t* req;
1641
1642  req = container_of(w, uv_fs_t, work_req);
1643  uv__req_unregister(req->loop, req);
1644
1645  if (status == UV_ECANCELED) {
1646    assert(req->result == 0);
1647    req->result = UV_ECANCELED;
1648  }
1649
1650  req->cb(req);
1651}
1652
1653
1654void uv__fs_post(uv_loop_t* loop, uv_fs_t* req) {
1655  uv__req_register(loop, req);
1656  uv__work_submit(loop,
1657#ifdef USE_FFRT
1658                  NULL,
1659#endif
1660                  &req->work_req,
1661                  UV__WORK_FAST_IO,
1662                  uv__fs_work,
1663                  uv__fs_done);
1664}
1665
1666
1667int uv_fs_access(uv_loop_t* loop,
1668                 uv_fs_t* req,
1669                 const char* path,
1670                 int flags,
1671                 uv_fs_cb cb) {
1672  INIT(ACCESS);
1673  PATH;
1674  req->flags = flags;
1675  POST;
1676}
1677
1678
1679int uv_fs_chmod(uv_loop_t* loop,
1680                uv_fs_t* req,
1681                const char* path,
1682                int mode,
1683                uv_fs_cb cb) {
1684  INIT(CHMOD);
1685  PATH;
1686  req->mode = mode;
1687  POST;
1688}
1689
1690
1691int uv_fs_chown(uv_loop_t* loop,
1692                uv_fs_t* req,
1693                const char* path,
1694                uv_uid_t uid,
1695                uv_gid_t gid,
1696                uv_fs_cb cb) {
1697  INIT(CHOWN);
1698  PATH;
1699  req->uid = uid;
1700  req->gid = gid;
1701  POST;
1702}
1703
1704
1705int uv_fs_close(uv_loop_t* loop, uv_fs_t* req, uv_file file, uv_fs_cb cb) {
1706  INIT(CLOSE);
1707  req->file = file;
1708  if (cb != NULL)
1709    if (uv__iou_fs_close(loop, req))
1710      return 0;
1711  POST;
1712}
1713
1714
1715int uv_fs_fchmod(uv_loop_t* loop,
1716                 uv_fs_t* req,
1717                 uv_file file,
1718                 int mode,
1719                 uv_fs_cb cb) {
1720  INIT(FCHMOD);
1721  req->file = file;
1722  req->mode = mode;
1723  POST;
1724}
1725
1726
1727int uv_fs_fchown(uv_loop_t* loop,
1728                 uv_fs_t* req,
1729                 uv_file file,
1730                 uv_uid_t uid,
1731                 uv_gid_t gid,
1732                 uv_fs_cb cb) {
1733  INIT(FCHOWN);
1734  req->file = file;
1735  req->uid = uid;
1736  req->gid = gid;
1737  POST;
1738}
1739
1740
1741int uv_fs_lchown(uv_loop_t* loop,
1742                 uv_fs_t* req,
1743                 const char* path,
1744                 uv_uid_t uid,
1745                 uv_gid_t gid,
1746                 uv_fs_cb cb) {
1747  INIT(LCHOWN);
1748  PATH;
1749  req->uid = uid;
1750  req->gid = gid;
1751  POST;
1752}
1753
1754
1755int uv_fs_fdatasync(uv_loop_t* loop, uv_fs_t* req, uv_file file, uv_fs_cb cb) {
1756  INIT(FDATASYNC);
1757  req->file = file;
1758  if (cb != NULL)
1759    if (uv__iou_fs_fsync_or_fdatasync(loop, req, /* IORING_FSYNC_DATASYNC */ 1))
1760      return 0;
1761  POST;
1762}
1763
1764
1765int uv_fs_fstat(uv_loop_t* loop, uv_fs_t* req, uv_file file, uv_fs_cb cb) {
1766  INIT(FSTAT);
1767  req->file = file;
1768  if (cb != NULL)
1769    if (uv__iou_fs_statx(loop, req, /* is_fstat */ 1, /* is_lstat */ 0))
1770      return 0;
1771  POST;
1772}
1773
1774
1775int uv_fs_fsync(uv_loop_t* loop, uv_fs_t* req, uv_file file, uv_fs_cb cb) {
1776  INIT(FSYNC);
1777  req->file = file;
1778  if (cb != NULL)
1779    if (uv__iou_fs_fsync_or_fdatasync(loop, req, /* no flags */ 0))
1780      return 0;
1781  POST;
1782}
1783
1784
1785int uv_fs_ftruncate(uv_loop_t* loop,
1786                    uv_fs_t* req,
1787                    uv_file file,
1788                    int64_t off,
1789                    uv_fs_cb cb) {
1790  INIT(FTRUNCATE);
1791  req->file = file;
1792  req->off = off;
1793  POST;
1794}
1795
1796
1797int uv_fs_futime(uv_loop_t* loop,
1798                 uv_fs_t* req,
1799                 uv_file file,
1800                 double atime,
1801                 double mtime,
1802                 uv_fs_cb cb) {
1803  INIT(FUTIME);
1804  req->file = file;
1805  req->atime = atime;
1806  req->mtime = mtime;
1807  POST;
1808}
1809
1810int uv_fs_lutime(uv_loop_t* loop,
1811                 uv_fs_t* req,
1812                 const char* path,
1813                 double atime,
1814                 double mtime,
1815                 uv_fs_cb cb) {
1816  INIT(LUTIME);
1817  PATH;
1818  req->atime = atime;
1819  req->mtime = mtime;
1820  POST;
1821}
1822
1823
1824int uv_fs_lstat(uv_loop_t* loop, uv_fs_t* req, const char* path, uv_fs_cb cb) {
1825  INIT(LSTAT);
1826  PATH;
1827  if (cb != NULL)
1828    if (uv__iou_fs_statx(loop, req, /* is_fstat */ 0, /* is_lstat */ 1))
1829      return 0;
1830  POST;
1831}
1832
1833
1834int uv_fs_link(uv_loop_t* loop,
1835               uv_fs_t* req,
1836               const char* path,
1837               const char* new_path,
1838               uv_fs_cb cb) {
1839  INIT(LINK);
1840  PATH2;
1841  if (cb != NULL)
1842    if (uv__iou_fs_link(loop, req))
1843      return 0;
1844  POST;
1845}
1846
1847
1848int uv_fs_mkdir(uv_loop_t* loop,
1849                uv_fs_t* req,
1850                const char* path,
1851                int mode,
1852                uv_fs_cb cb) {
1853  INIT(MKDIR);
1854  PATH;
1855  req->mode = mode;
1856  if (cb != NULL)
1857    if (uv__iou_fs_mkdir(loop, req))
1858      return 0;
1859  POST;
1860}
1861
1862
1863int uv_fs_mkdtemp(uv_loop_t* loop,
1864                  uv_fs_t* req,
1865                  const char* tpl,
1866                  uv_fs_cb cb) {
1867  INIT(MKDTEMP);
1868  req->path = uv__strdup(tpl);
1869  if (req->path == NULL)
1870    return UV_ENOMEM;
1871  POST;
1872}
1873
1874
1875int uv_fs_mkstemp(uv_loop_t* loop,
1876                  uv_fs_t* req,
1877                  const char* tpl,
1878                  uv_fs_cb cb) {
1879  INIT(MKSTEMP);
1880  req->path = uv__strdup(tpl);
1881  if (req->path == NULL)
1882    return UV_ENOMEM;
1883  POST;
1884}
1885
1886
1887int uv_fs_open(uv_loop_t* loop,
1888               uv_fs_t* req,
1889               const char* path,
1890               int flags,
1891               int mode,
1892               uv_fs_cb cb) {
1893  INIT(OPEN);
1894  PATH;
1895  req->flags = flags;
1896  req->mode = mode;
1897  if (cb != NULL)
1898    if (uv__iou_fs_open(loop, req))
1899      return 0;
1900  POST;
1901}
1902
1903
1904int uv_fs_read(uv_loop_t* loop, uv_fs_t* req,
1905               uv_file file,
1906               const uv_buf_t bufs[],
1907               unsigned int nbufs,
1908               int64_t off,
1909               uv_fs_cb cb) {
1910  INIT(READ);
1911
1912  if (bufs == NULL || nbufs == 0)
1913    return UV_EINVAL;
1914
1915  req->off = off;
1916  req->file = file;
1917  req->bufs = (uv_buf_t*) bufs;  /* Safe, doesn't mutate |bufs| */
1918  req->nbufs = nbufs;
1919
1920  if (cb == NULL)
1921    goto post;
1922
1923  req->bufs = req->bufsml;
1924  if (nbufs > ARRAY_SIZE(req->bufsml))
1925    req->bufs = uv__malloc(nbufs * sizeof(*bufs));
1926
1927  if (req->bufs == NULL)
1928    return UV_ENOMEM;
1929
1930  memcpy(req->bufs, bufs, nbufs * sizeof(*bufs));
1931
1932  if (uv__iou_fs_read_or_write(loop, req, /* is_read */ 1))
1933    return 0;
1934
1935post:
1936  POST;
1937}
1938
1939
1940int uv_fs_scandir(uv_loop_t* loop,
1941                  uv_fs_t* req,
1942                  const char* path,
1943                  int flags,
1944                  uv_fs_cb cb) {
1945  INIT(SCANDIR);
1946  PATH;
1947  req->flags = flags;
1948  POST;
1949}
1950
1951int uv_fs_opendir(uv_loop_t* loop,
1952                  uv_fs_t* req,
1953                  const char* path,
1954                  uv_fs_cb cb) {
1955  INIT(OPENDIR);
1956  PATH;
1957  POST;
1958}
1959
1960int uv_fs_readdir(uv_loop_t* loop,
1961                  uv_fs_t* req,
1962                  uv_dir_t* dir,
1963                  uv_fs_cb cb) {
1964  INIT(READDIR);
1965
1966  if (dir == NULL || dir->dir == NULL || dir->dirents == NULL)
1967    return UV_EINVAL;
1968
1969  req->ptr = dir;
1970  POST;
1971}
1972
1973int uv_fs_closedir(uv_loop_t* loop,
1974                   uv_fs_t* req,
1975                   uv_dir_t* dir,
1976                   uv_fs_cb cb) {
1977  INIT(CLOSEDIR);
1978
1979  if (dir == NULL)
1980    return UV_EINVAL;
1981
1982  req->ptr = dir;
1983  POST;
1984}
1985
1986int uv_fs_readlink(uv_loop_t* loop,
1987                   uv_fs_t* req,
1988                   const char* path,
1989                   uv_fs_cb cb) {
1990  INIT(READLINK);
1991  PATH;
1992  POST;
1993}
1994
1995
1996int uv_fs_realpath(uv_loop_t* loop,
1997                  uv_fs_t* req,
1998                  const char * path,
1999                  uv_fs_cb cb) {
2000  INIT(REALPATH);
2001  PATH;
2002  POST;
2003}
2004
2005
2006int uv_fs_rename(uv_loop_t* loop,
2007                 uv_fs_t* req,
2008                 const char* path,
2009                 const char* new_path,
2010                 uv_fs_cb cb) {
2011  INIT(RENAME);
2012  PATH2;
2013  if (cb != NULL)
2014    if (uv__iou_fs_rename(loop, req))
2015      return 0;
2016  POST;
2017}
2018
2019
2020int uv_fs_rmdir(uv_loop_t* loop, uv_fs_t* req, const char* path, uv_fs_cb cb) {
2021  INIT(RMDIR);
2022  PATH;
2023  POST;
2024}
2025
2026
2027int uv_fs_sendfile(uv_loop_t* loop,
2028                   uv_fs_t* req,
2029                   uv_file out_fd,
2030                   uv_file in_fd,
2031                   int64_t off,
2032                   size_t len,
2033                   uv_fs_cb cb) {
2034  INIT(SENDFILE);
2035  req->flags = in_fd; /* hack */
2036  req->file = out_fd;
2037  req->off = off;
2038  req->bufsml[0].len = len;
2039  POST;
2040}
2041
2042
2043int uv_fs_stat(uv_loop_t* loop, uv_fs_t* req, const char* path, uv_fs_cb cb) {
2044  INIT(STAT);
2045  PATH;
2046  if (cb != NULL)
2047    if (uv__iou_fs_statx(loop, req, /* is_fstat */ 0, /* is_lstat */ 0))
2048      return 0;
2049  POST;
2050}
2051
2052
2053int uv_fs_symlink(uv_loop_t* loop,
2054                  uv_fs_t* req,
2055                  const char* path,
2056                  const char* new_path,
2057                  int flags,
2058                  uv_fs_cb cb) {
2059  INIT(SYMLINK);
2060  PATH2;
2061  req->flags = flags;
2062  if (cb != NULL)
2063    if (uv__iou_fs_symlink(loop, req))
2064      return 0;
2065  POST;
2066}
2067
2068
2069int uv_fs_unlink(uv_loop_t* loop, uv_fs_t* req, const char* path, uv_fs_cb cb) {
2070  INIT(UNLINK);
2071  PATH;
2072  if (cb != NULL)
2073    if (uv__iou_fs_unlink(loop, req))
2074      return 0;
2075  POST;
2076}
2077
2078
2079int uv_fs_utime(uv_loop_t* loop,
2080                uv_fs_t* req,
2081                const char* path,
2082                double atime,
2083                double mtime,
2084                uv_fs_cb cb) {
2085  INIT(UTIME);
2086  PATH;
2087  req->atime = atime;
2088  req->mtime = mtime;
2089  POST;
2090}
2091
2092
2093int uv_fs_write(uv_loop_t* loop,
2094                uv_fs_t* req,
2095                uv_file file,
2096                const uv_buf_t bufs[],
2097                unsigned int nbufs,
2098                int64_t off,
2099                uv_fs_cb cb) {
2100  INIT(WRITE);
2101
2102  if (bufs == NULL || nbufs == 0)
2103    return UV_EINVAL;
2104
2105  req->file = file;
2106
2107  req->nbufs = nbufs;
2108  req->bufs = req->bufsml;
2109  if (nbufs > ARRAY_SIZE(req->bufsml))
2110    req->bufs = uv__malloc(nbufs * sizeof(*bufs));
2111
2112  if (req->bufs == NULL)
2113    return UV_ENOMEM;
2114
2115  memcpy(req->bufs, bufs, nbufs * sizeof(*bufs));
2116
2117  req->off = off;
2118
2119  if (cb != NULL)
2120    if (uv__iou_fs_read_or_write(loop, req, /* is_read */ 0))
2121      return 0;
2122
2123  POST;
2124}
2125
2126
2127void uv_fs_req_cleanup(uv_fs_t* req) {
2128  if (req == NULL)
2129    return;
2130
2131  /* Only necessary for asynchronous requests, i.e., requests with a callback.
2132   * Synchronous ones don't copy their arguments and have req->path and
2133   * req->new_path pointing to user-owned memory.  UV_FS_MKDTEMP and
2134   * UV_FS_MKSTEMP are the exception to the rule, they always allocate memory.
2135   */
2136  if (req->path != NULL &&
2137      (req->cb != NULL ||
2138        req->fs_type == UV_FS_MKDTEMP || req->fs_type == UV_FS_MKSTEMP))
2139    uv__free((void*) req->path);  /* Memory is shared with req->new_path. */
2140
2141  req->path = NULL;
2142  req->new_path = NULL;
2143
2144  if (req->fs_type == UV_FS_READDIR && req->ptr != NULL)
2145    uv__fs_readdir_cleanup(req);
2146
2147  if (req->fs_type == UV_FS_SCANDIR && req->ptr != NULL)
2148    uv__fs_scandir_cleanup(req);
2149
2150  if (req->bufs != req->bufsml)
2151    uv__free(req->bufs);
2152  req->bufs = NULL;
2153
2154  if (req->fs_type != UV_FS_OPENDIR && req->ptr != &req->statbuf)
2155    uv__free(req->ptr);
2156  req->ptr = NULL;
2157}
2158
2159
2160int uv_fs_copyfile(uv_loop_t* loop,
2161                   uv_fs_t* req,
2162                   const char* path,
2163                   const char* new_path,
2164                   int flags,
2165                   uv_fs_cb cb) {
2166  INIT(COPYFILE);
2167
2168  if (flags & ~(UV_FS_COPYFILE_EXCL |
2169                UV_FS_COPYFILE_FICLONE |
2170                UV_FS_COPYFILE_FICLONE_FORCE)) {
2171    return UV_EINVAL;
2172  }
2173
2174  PATH2;
2175  req->flags = flags;
2176  POST;
2177}
2178
2179
2180int uv_fs_statfs(uv_loop_t* loop,
2181                 uv_fs_t* req,
2182                 const char* path,
2183                 uv_fs_cb cb) {
2184  INIT(STATFS);
2185  PATH;
2186  POST;
2187}
2188
2189int uv_fs_get_system_error(const uv_fs_t* req) {
2190  return -req->result;
2191}
2192