src/unix/fs.c

/* Copyright Joyent, Inc. and other Node contributors. All rights reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a copy
 * of this software and associated documentation files (the "Software"), to
 * deal in the Software without restriction, including without limitation the
 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
 * sell copies of the Software, and to permit persons to whom the Software is
 * furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 */

/* Caveat emptor: this file deviates from the libuv convention of returning
 * negated errno codes. Most uv_fs_*() functions map directly to the system
 * call of the same name. For more complex wrappers, it's easier to just
 * return -1 with errno set. The dispatcher in uv__fs_work() takes care of
 * getting the errno to the right place (req->result or as the return value.)
 */

#include "uv.h"
#include "internal.h"

#include <errno.h>
#include <dlfcn.h>
#include <stdatomic.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <limits.h> /* PATH_MAX */

#include <sys/types.h>
#include <sys/socket.h>
#include <sys/stat.h>
#include <sys/time.h>
#include <sys/uio.h>
#include <unistd.h>
#include <fcntl.h>
#include <poll.h>

#if defined(__linux__)
# include <sys/sendfile.h>
#endif

#if defined(__sun)
# include <sys/sendfile.h>
# include <sys/sysmacros.h>
#endif

#if defined(__APPLE__)
# include <sys/sysctl.h>
#elif defined(__linux__) && !defined(FICLONE)
# include <sys/ioctl.h>
# define FICLONE _IOW(0x94, 9, int)
#endif

#if defined(_AIX) && !defined(_AIX71)
# include <utime.h>
#endif

#if defined(__APPLE__)            ||                                      \
    defined(__DragonFly__)        ||                                      \
    defined(__FreeBSD__)          ||                                      \
    defined(__OpenBSD__)          ||                                      \
    defined(__NetBSD__)
# include <sys/param.h>
# include <sys/mount.h>
#elif defined(__sun)      || \
      defined(__MVS__)    || \
      defined(__NetBSD__) || \
      defined(__HAIKU__)  || \
      defined(__QNX__)
# include <sys/statvfs.h>
#else
# include <sys/statfs.h>
#endif

#if defined(_AIX) && _XOPEN_SOURCE <= 600
extern char *mkdtemp(char *template); /* See issue #740 on AIX < 7 */
#endif

#define INIT(subtype)                                                         \
  do {                                                                        \
    if (req == NULL)                                                          \
      return UV_EINVAL;                                                       \
    UV_REQ_INIT(req, UV_FS);                                                  \
    req->fs_type = UV_FS_ ## subtype;                                         \
    req->result = 0;                                                          \
    req->ptr = NULL;                                                          \
    req->loop = loop;                                                         \
    req->path = NULL;                                                         \
    req->new_path = NULL;                                                     \
    req->bufs = NULL;                                                         \
    req->cb = cb;                                                             \
  }                                                                           \
  while (0)

#define PATH                                                                  \
  do {                                                                        \
    assert(path != NULL);                                                     \
    if (cb == NULL) {                                                         \
      req->path = path;                                                       \
    } else {                                                                  \
      req->path = uv__strdup(path);                                           \
      if (req->path == NULL)                                                  \
        return UV_ENOMEM;                                                     \
    }                                                                         \
  }                                                                           \
  while (0)

#define PATH2                                                                 \
  do {                                                                        \
    if (cb == NULL) {                                                         \
      req->path = path;                                                       \
      req->new_path = new_path;                                               \
    } else {                                                                  \
      size_t path_len;                                                        \
      size_t new_path_len;                                                    \
      path_len = strlen(path) + 1;                                            \
      new_path_len = strlen(new_path) + 1;                                    \
      req->path = uv__malloc(path_len + new_path_len);                        \
      if (req->path == NULL)                                                  \
        return UV_ENOMEM;                                                     \
      req->new_path = req->path + path_len;                                   \
      memcpy((void*) req->path, path, path_len);                              \
      memcpy((void*) req->new_path, new_path, new_path_len);                  \
    }                                                                         \
  }                                                                           \
  while (0)

#define POST                                                                  \
  do {                                                                        \
    if (cb != NULL) {                                                         \
      uv__req_register(loop);                                                 \
      uv__work_submit(loop,                                                   \
                      &req->work_req,                                         \
                      UV__WORK_FAST_IO,                                       \
                      uv__fs_work,                                            \
                      uv__fs_done);                                           \
      return 0;                                                               \
    }                                                                         \
    else {                                                                    \
      uv__fs_work(&req->work_req);                                            \
      return req->result;                                                     \
    }                                                                         \
  }                                                                           \
  while (0)


static int uv__fs_close(int fd) {
  int rc;

  rc = uv__close_nocancel(fd);
  if (rc == -1)
    if (errno == EINTR || errno == EINPROGRESS)
      rc = 0;  /* The close is in progress, not an error. */

  return rc;
}


static ssize_t uv__fs_fsync(uv_fs_t* req) {
#if defined(__APPLE__)
  /* Apple's fdatasync and fsync explicitly do NOT flush the drive write cache
   * to the drive platters. This is in contrast to Linux's fdatasync and fsync
   * which do, according to recent man pages. F_FULLFSYNC is Apple's equivalent
   * for flushing buffered data to permanent storage. If F_FULLFSYNC is not
   * supported by the file system we fall back to F_BARRIERFSYNC or fsync().
   * This is the same approach taken by sqlite, except sqlite does not issue
   * an F_BARRIERFSYNC call.
   */
  int r;

  r = fcntl(req->file, F_FULLFSYNC);
  if (r != 0)
    r = fcntl(req->file, 85 /* F_BARRIERFSYNC */);  /* fsync + barrier */
  if (r != 0)
    r = fsync(req->file);
  return r;
#else
  return fsync(req->file);
#endif
}


static ssize_t uv__fs_fdatasync(uv_fs_t* req) {
#if defined(__linux__) || defined(__sun) || defined(__NetBSD__)
  return fdatasync(req->file);
#elif defined(__APPLE__)
  /* See the comment in uv__fs_fsync. */
  return uv__fs_fsync(req);
#else
  return fsync(req->file);
#endif
}


UV_UNUSED(static struct timespec uv__fs_to_timespec(double time)) {
  struct timespec ts;
  ts.tv_sec  = time;
  ts.tv_nsec = (time - ts.tv_sec) * 1e9;

 /* TODO(bnoordhuis) Remove this. utimesat() has nanosecond resolution but we
  * stick to microsecond resolution for the sake of consistency with other
  * platforms. I'm the original author of this compatibility hack but I'm
  * less convinced it's useful nowadays.
  */
  ts.tv_nsec -= ts.tv_nsec % 1000;

  if (ts.tv_nsec < 0) {
    ts.tv_nsec += 1e9;
    ts.tv_sec -= 1;
  }
  return ts;
}

UV_UNUSED(static struct timeval uv__fs_to_timeval(double time)) {
  struct timeval tv;
  tv.tv_sec  = time;
  tv.tv_usec = (time - tv.tv_sec) * 1e6;
  if (tv.tv_usec < 0) {
    tv.tv_usec += 1e6;
    tv.tv_sec -= 1;
  }
  return tv;
}

static ssize_t uv__fs_futime(uv_fs_t* req) {
#if defined(__linux__)                                                        \
    || defined(_AIX71)                                                        \
    || defined(__HAIKU__)                                                     \
    || defined(__GNU__)
  struct timespec ts[2];
  ts[0] = uv__fs_to_timespec(req->atime);
  ts[1] = uv__fs_to_timespec(req->mtime);
  return futimens(req->file, ts);
#elif defined(__APPLE__)                                                      \
    || defined(__DragonFly__)                                                 \
    || defined(__FreeBSD__)                                                   \
    || defined(__NetBSD__)                                                    \
    || defined(__OpenBSD__)                                                   \
    || defined(__sun)
  struct timeval tv[2];
  tv[0] = uv__fs_to_timeval(req->atime);
  tv[1] = uv__fs_to_timeval(req->mtime);
# if defined(__sun)
  return futimesat(req->file, NULL, tv);
# else
  return futimes(req->file, tv);
# endif
#elif defined(__MVS__)
  attrib_t atr;
  memset(&atr, 0, sizeof(atr));
  atr.att_mtimechg = 1;
  atr.att_atimechg = 1;
  atr.att_mtime = req->mtime;
  atr.att_atime = req->atime;
  return __fchattr(req->file, &atr, sizeof(atr));
#else
  errno = ENOSYS;
  return -1;
#endif
}


static ssize_t uv__fs_mkdtemp(uv_fs_t* req) {
  return mkdtemp((char*) req->path) ? 0 : -1;
}


static int (*uv__mkostemp)(char*, int);


static void uv__mkostemp_initonce(void) {
  /* z/os doesn't have RTLD_DEFAULT but that's okay
   * because it doesn't have mkostemp(O_CLOEXEC) either.
   */
#ifdef RTLD_DEFAULT
  uv__mkostemp = (int (*)(char*, int)) dlsym(RTLD_DEFAULT, "mkostemp");

  /* We don't care about errors, but we do want to clean them up.
   * If there has been no error, then dlerror() will just return
   * NULL.
   */
  dlerror();
#endif  /* RTLD_DEFAULT */
}


static int uv__fs_mkstemp(uv_fs_t* req) {
  static uv_once_t once = UV_ONCE_INIT;
  int r;
#ifdef O_CLOEXEC
  static _Atomic int no_cloexec_support;
#endif
  static const char pattern[] = "XXXXXX";
  static const size_t pattern_size = sizeof(pattern) - 1;
  char* path;
  size_t path_length;

  path = (char*) req->path;
  path_length = strlen(path);

  /* EINVAL can be returned for 2 reasons:
      1. The template's last 6 characters were not XXXXXX
      2. open() didn't support O_CLOEXEC
     We want to avoid going to the fallback path in case
     of 1, so it's manually checked before. */
  if (path_length < pattern_size ||
      strcmp(path + path_length - pattern_size, pattern)) {
    errno = EINVAL;
    r = -1;
    goto clobber;
  }

  uv_once(&once, uv__mkostemp_initonce);

#ifdef O_CLOEXEC
  if (atomic_load_explicit(&no_cloexec_support, memory_order_relaxed) == 0 &&
      uv__mkostemp != NULL) {
    r = uv__mkostemp(path, O_CLOEXEC);

    if (r >= 0)
      return r;

    /* If mkostemp() returns EINVAL, it means the kernel doesn't
       support O_CLOEXEC, so we just fallback to mkstemp() below. */
    if (errno != EINVAL)
      goto clobber;

    /* We set the static variable so that next calls don't even
       try to use mkostemp. */
    atomic_store_explicit(&no_cloexec_support, 1, memory_order_relaxed);
  }
#endif  /* O_CLOEXEC */

  if (req->cb != NULL)
    uv_rwlock_rdlock(&req->loop->cloexec_lock);

  r = mkstemp(path);

  /* In case of failure `uv__cloexec` will leave error in `errno`,
   * so it is enough to just set `r` to `-1`.
   */
  if (r >= 0 && uv__cloexec(r, 1) != 0) {
    r = uv__close(r);
    if (r != 0)
      abort();
    r = -1;
  }

  if (req->cb != NULL)
    uv_rwlock_rdunlock(&req->loop->cloexec_lock);

clobber:
  if (r < 0)
    path[0] = '\0';
  return r;
}


static ssize_t uv__fs_open(uv_fs_t* req) {
#ifdef O_CLOEXEC
  return open(req->path, req->flags | O_CLOEXEC, req->mode);
#else  /* O_CLOEXEC */
  int r;

  if (req->cb != NULL)
    uv_rwlock_rdlock(&req->loop->cloexec_lock);

  r = open(req->path, req->flags, req->mode);

  /* In case of failure `uv__cloexec` will leave error in `errno`,
   * so it is enough to just set `r` to `-1`.
   */
  if (r >= 0 && uv__cloexec(r, 1) != 0) {
    r = uv__close(r);
    if (r != 0)
      abort();
    r = -1;
  }

  if (req->cb != NULL)
    uv_rwlock_rdunlock(&req->loop->cloexec_lock);

  return r;
#endif  /* O_CLOEXEC */
}


static ssize_t uv__preadv_or_pwritev_emul(int fd,
                                          const struct iovec* bufs,
                                          size_t nbufs,
                                          off_t off,
                                          int is_pread) {
  ssize_t total;
  ssize_t r;
  size_t i;
  size_t n;
  void* p;

  total = 0;
  for (i = 0; i < (size_t) nbufs; i++) {
    p = bufs[i].iov_base;
    n = bufs[i].iov_len;

    do
      if (is_pread)
        r = pread(fd, p, n, off);
      else
        r = pwrite(fd, p, n, off);
    while (r == -1 && errno == EINTR);

    if (r == -1) {
      if (total > 0)
        return total;
      return -1;
    }

    off += r;
    total += r;

    if ((size_t) r < n)
      return total;
  }

  return total;
}


#ifdef __linux__
typedef int uv__iovcnt;
#else
typedef size_t uv__iovcnt;
#endif


static ssize_t uv__preadv_emul(int fd,
                               const struct iovec* bufs,
                               uv__iovcnt nbufs,
                               off_t off) {
  return uv__preadv_or_pwritev_emul(fd, bufs, nbufs, off, /*is_pread*/1);
}


static ssize_t uv__pwritev_emul(int fd,
                                const struct iovec* bufs,
                                uv__iovcnt nbufs,
                                off_t off) {
  return uv__preadv_or_pwritev_emul(fd, bufs, nbufs, off, /*is_pread*/0);
}


/* The function pointer cache is an uintptr_t because _Atomic void*
 * doesn't work on macos/ios/etc...
 */
static ssize_t uv__preadv_or_pwritev(int fd,
                                     const struct iovec* bufs,
                                     size_t nbufs,
                                     off_t off,
                                     _Atomic uintptr_t* cache,
                                     int is_pread) {
  ssize_t (*f)(int, const struct iovec*, uv__iovcnt, off_t);
  void* p;

  p = (void*) atomic_load_explicit(cache, memory_order_relaxed);
  if (p == NULL) {
#ifdef RTLD_DEFAULT
    p = dlsym(RTLD_DEFAULT, is_pread ? "preadv" : "pwritev");
    dlerror();  /* Clear errors. */
#endif  /* RTLD_DEFAULT */
    if (p == NULL)
      p = is_pread ? uv__preadv_emul : uv__pwritev_emul;
    atomic_store_explicit(cache, (uintptr_t) p, memory_order_relaxed);
  }

  /* Use memcpy instead of `f = p` to work around a compiler bug,
   * see https://github.com/libuv/libuv/issues/4532
   */
  memcpy(&f, &p, sizeof(p));
  return f(fd, bufs, nbufs, off);
}


static ssize_t uv__preadv(int fd,
                          const struct iovec* bufs,
                          size_t nbufs,
                          off_t off) {
  static _Atomic uintptr_t cache;
  return uv__preadv_or_pwritev(fd, bufs, nbufs, off, &cache, /*is_pread*/1);
}


static ssize_t uv__pwritev(int fd,
                           const struct iovec* bufs,
                           size_t nbufs,
                           off_t off) {
  static _Atomic uintptr_t cache;
  return uv__preadv_or_pwritev(fd, bufs, nbufs, off, &cache, /*is_pread*/0);
}


static ssize_t uv__fs_read(uv_fs_t* req) {
  const struct iovec* bufs;
  unsigned int iovmax;
  size_t nbufs;
  ssize_t r;
  off_t off;
  int fd;

  fd = req->file;
  off = req->off;
  bufs = (const struct iovec*) req->bufs;
  nbufs = req->nbufs;

  iovmax = uv__getiovmax();
  if (nbufs > iovmax)
    nbufs = iovmax;

  r = 0;
  if (off < 0) {
    if (nbufs == 1)
      r = read(fd, bufs->iov_base, bufs->iov_len);
    else if (nbufs > 1)
      r = readv(fd, bufs, nbufs);
  } else {
    if (nbufs == 1)
      r = pread(fd, bufs->iov_base, bufs->iov_len, off);
    else if (nbufs > 1)
      r = uv__preadv(fd, bufs, nbufs, off);
  }

#ifdef __PASE__
  /* PASE returns EOPNOTSUPP when reading a directory, convert to EISDIR */
  if (r == -1 && errno == EOPNOTSUPP) {
    struct stat buf;
    ssize_t rc;
    rc = uv__fstat(fd, &buf);
    if (rc == 0 && S_ISDIR(buf.st_mode)) {
      errno = EISDIR;
    }
  }
#endif

  /* We don't own the buffer list in the synchronous case. */
  if (req->cb != NULL)
    if (req->bufs != req->bufsml)
      uv__free(req->bufs);

  req->bufs = NULL;
  req->nbufs = 0;

  return r;
}


static int uv__fs_scandir_filter(const uv__dirent_t* dent) {
  return strcmp(dent->d_name, ".") != 0 && strcmp(dent->d_name, "..") != 0;
}


static int uv__fs_scandir_sort(const uv__dirent_t** a, const uv__dirent_t** b) {
  return strcmp((*a)->d_name, (*b)->d_name);
}


static ssize_t uv__fs_scandir(uv_fs_t* req) {
  uv__dirent_t** dents;
  int n;

  dents = NULL;
  n = scandir(req->path, &dents, uv__fs_scandir_filter, uv__fs_scandir_sort);

  /* NOTE: We will use nbufs as an index field */
  req->nbufs = 0;

  if (n == 0) {
    /* OS X still needs to deallocate some memory.
     * Memory was allocated using the system allocator, so use free() here.
     */
    free(dents);
    dents = NULL;
  } else if (n == -1) {
    return n;
  }

  req->ptr = dents;

  return n;
}

static int uv__fs_opendir(uv_fs_t* req) {
  uv_dir_t* dir;

  dir = uv__malloc(sizeof(*dir));
  if (dir == NULL)
    goto error;

  dir->dir = opendir(req->path);
  if (dir->dir == NULL)
    goto error;

  req->ptr = dir;
  return 0;

error:
  uv__free(dir);
  req->ptr = NULL;
  return -1;
}

static int uv__fs_readdir(uv_fs_t* req) {
  uv_dir_t* dir;
  uv_dirent_t* dirent;
  struct dirent* res;
  unsigned int dirent_idx;
  unsigned int i;

  dir = req->ptr;
  dirent_idx = 0;

  while (dirent_idx < dir->nentries) {
    /* readdir() returns NULL on end of directory, as well as on error. errno
       is used to differentiate between the two conditions. */
    errno = 0;
    res = readdir(dir->dir);

    if (res == NULL) {
      if (errno != 0)
        goto error;
      break;
    }

    if (strcmp(res->d_name, ".") == 0 || strcmp(res->d_name, "..") == 0)
      continue;

    dirent = &dir->dirents[dirent_idx];
    dirent->name = uv__strdup(res->d_name);

    if (dirent->name == NULL)
      goto error;

    dirent->type = uv__fs_get_dirent_type(res);
    ++dirent_idx;
  }

  return dirent_idx;

error:
  for (i = 0; i < dirent_idx; ++i) {
    uv__free((char*) dir->dirents[i].name);
    dir->dirents[i].name = NULL;
  }

  return -1;
}

static int uv__fs_closedir(uv_fs_t* req) {
  uv_dir_t* dir;

  dir = req->ptr;

  if (dir->dir != NULL) {
    closedir(dir->dir);
    dir->dir = NULL;
  }

  uv__free(req->ptr);
  req->ptr = NULL;
  return 0;
}

static int uv__fs_statfs(uv_fs_t* req) {
  uv_statfs_t* stat_fs;
#if defined(__sun)      || \
    defined(__MVS__)    || \
    defined(__NetBSD__) || \
    defined(__HAIKU__)  || \
    defined(__QNX__)
  struct statvfs buf;

  if (0 != statvfs(req->path, &buf))
#else
  struct statfs buf;

  if (0 != statfs(req->path, &buf))
#endif /* defined(__sun) */
    return -1;

  stat_fs = uv__malloc(sizeof(*stat_fs));
  if (stat_fs == NULL) {
    errno = ENOMEM;
    return -1;
  }

#if defined(__sun)        || \
    defined(__MVS__)      || \
    defined(__OpenBSD__)  || \
    defined(__NetBSD__)   || \
    defined(__HAIKU__)    || \
    defined(__QNX__)
  stat_fs->f_type = 0;  /* f_type is not supported. */
#else
  stat_fs->f_type = buf.f_type;
#endif
  stat_fs->f_bsize = buf.f_bsize;
  stat_fs->f_blocks = buf.f_blocks;
  stat_fs->f_bfree = buf.f_bfree;
  stat_fs->f_bavail = buf.f_bavail;
  stat_fs->f_files = buf.f_files;
  stat_fs->f_ffree = buf.f_ffree;
  req->ptr = stat_fs;
  return 0;
}

static ssize_t uv__fs_pathmax_size(const char* path) {
  ssize_t pathmax;

  pathmax = pathconf(path, _PC_PATH_MAX);

  if (pathmax == -1)
    pathmax = UV__PATH_MAX;

  return pathmax;
}

static ssize_t uv__fs_readlink(uv_fs_t* req) {
  ssize_t maxlen;
  ssize_t len;
  char* buf;

#if defined(_POSIX_PATH_MAX) || defined(PATH_MAX)
  maxlen = uv__fs_pathmax_size(req->path);
#else
  /* We may not have a real PATH_MAX.  Read size of link.  */
  struct stat st;
  int ret;
  ret = uv__lstat(req->path, &st);
  if (ret != 0)
    return -1;
  if (!S_ISLNK(st.st_mode)) {
    errno = EINVAL;
    return -1;
  }

  maxlen = st.st_size;

  /* According to readlink(2) lstat can report st_size == 0
     for some symlinks, such as those in /proc or /sys.  */
  if (maxlen == 0)
    maxlen = uv__fs_pathmax_size(req->path);
#endif

  buf = uv__malloc(maxlen);

  if (buf == NULL) {
    errno = ENOMEM;
    return -1;
  }

#if defined(__MVS__)
  len = os390_readlink(req->path, buf, maxlen);
#else
  len = readlink(req->path, buf, maxlen);
#endif

  if (len == -1) {
    uv__free(buf);
    return -1;
  }

  /* Uncommon case: resize to make room for the trailing nul byte. */
  if (len == maxlen) {
    buf = uv__reallocf(buf, len + 1);

    if (buf == NULL)
      return -1;
  }

  buf[len] = '\0';
  req->ptr = buf;

  return 0;
}

static ssize_t uv__fs_realpath(uv_fs_t* req) {
  char* buf;
  char* tmp;

#if defined(_POSIX_VERSION) && _POSIX_VERSION >= 200809L
  tmp = realpath(req->path, NULL);
  if (tmp == NULL)
    return -1;
  buf = uv__strdup(tmp);
  free(tmp); /* _Not_ uv__free. */
  if (buf == NULL) {
    errno = ENOMEM;
    return -1;
  }
#else
  ssize_t len;

  (void)tmp;

  len = uv__fs_pathmax_size(req->path);
  buf = uv__malloc(len + 1);

  if (buf == NULL) {
    errno = ENOMEM;
    return -1;
  }

  if (realpath(req->path, buf) == NULL) {
    uv__free(buf);
    return -1;
  }
#endif

  req->ptr = buf;

  return 0;
}

static ssize_t uv__fs_sendfile_emul(uv_fs_t* req) {
  struct pollfd pfd;
  int use_pread;
  off_t offset;
  ssize_t nsent;
  ssize_t nread;
  ssize_t nwritten;
  size_t buflen;
  size_t len;
  ssize_t n;
  int in_fd;
  int out_fd;
  char buf[8192];

  len = req->bufsml[0].len;
  in_fd = req->flags;
  out_fd = req->file;
  offset = req->off;
  use_pread = 1;

  /* Here are the rules regarding errors:
   *
   * 1. Read errors are reported only if nsent==0, otherwise we return nsent.
   *    The user needs to know that some data has already been sent, to stop
   *    them from sending it twice.
   *
   * 2. Write errors are always reported. Write errors are bad because they
   *    mean data loss: we've read data but now we can't write it out.
   *
   * We try to use pread() and fall back to regular read() if the source fd
   * doesn't support positional reads, for example when it's a pipe fd.
   *
   * If we get EAGAIN when writing to the target fd, we poll() on it until
   * it becomes writable again.
   *
   * FIXME: If we get a write error when use_pread==1, it should be safe to
   *        return the number of sent bytes instead of an error because pread()
   *        is, in theory, idempotent. However, special files in /dev or /proc
   *        may support pread() but not necessarily return the same data on
   *        successive reads.
   *
   * FIXME: There is no way now to signal that we managed to send *some* data
   *        before a write error.
   */
  for (nsent = 0; (size_t) nsent < len; ) {
    buflen = len - nsent;

    if (buflen > sizeof(buf))
      buflen = sizeof(buf);

    do
      if (use_pread)
        nread = pread(in_fd, buf, buflen, offset);
      else
        nread = read(in_fd, buf, buflen);
    while (nread == -1 && errno == EINTR);

    if (nread == 0)
      goto out;

    if (nread == -1) {
      if (use_pread && nsent == 0 && (errno == EIO || errno == ESPIPE)) {
        use_pread = 0;
        continue;
      }

      if (nsent == 0)
        nsent = -1;

      goto out;
    }

    for (nwritten = 0; nwritten < nread; ) {
      do
        n = write(out_fd, buf + nwritten, nread - nwritten);
      while (n == -1 && errno == EINTR);

      if (n != -1) {
        nwritten += n;
        continue;
      }

      if (errno != EAGAIN && errno != EWOULDBLOCK) {
        nsent = -1;
        goto out;
      }

      pfd.fd = out_fd;
      pfd.events = POLLOUT;
      pfd.revents = 0;

      do
        n = poll(&pfd, 1, -1);
      while (n == -1 && errno == EINTR);

      if (n == -1 || (pfd.revents & ~POLLOUT) != 0) {
        errno = EIO;
        nsent = -1;
        goto out;
      }
    }

    offset += nread;
    nsent += nread;
  }

out:
  if (nsent != -1)
    req->off = offset;

  return nsent;
}


#ifdef __linux__
/* Pre-4.20 kernels have a bug where CephFS uses the RADOS copy-from command
 * in copy_file_range() when it shouldn't. There is no workaround except to
 * fall back to a regular copy.
 */
static int uv__is_buggy_cephfs(int fd) {
  struct statfs s;

  if (-1 == fstatfs(fd, &s))
    return 0;

  if (s.f_type != /* CephFS */ 0xC36400)
    return 0;

  return uv__kernel_version() < /* 4.20.0 */ 0x041400;
}


static int uv__is_cifs_or_smb(int fd) {
  struct statfs s;

  if (-1 == fstatfs(fd, &s))
    return 0;

  switch ((unsigned) s.f_type) {
  case 0x0000517Bu:  /* SMB */
  case 0xFE534D42u:  /* SMB2 */
  case 0xFF534D42u:  /* CIFS */
    return 1;
  }

  return 0;
}


static ssize_t uv__fs_try_copy_file_range(int in_fd, off_t* off,
                                          int out_fd, size_t len) {
  static _Atomic int no_copy_file_range_support;
  ssize_t r;

  if (atomic_load_explicit(&no_copy_file_range_support, memory_order_relaxed)) {
    errno = ENOSYS;
    return -1;
  }

  r = uv__fs_copy_file_range(in_fd, off, out_fd, NULL, len, 0);

  if (r != -1)
    return r;

  switch (errno) {
  case EACCES:
    /* Pre-4.20 kernels have a bug where CephFS uses the RADOS
     * copy-from command when it shouldn't.
     */
    if (uv__is_buggy_cephfs(in_fd))
      errno = ENOSYS;  /* Use fallback. */
    break;
  case ENOSYS:
    atomic_store_explicit(&no_copy_file_range_support, 1, memory_order_relaxed);
    break;
  case EPERM:
    /* It's been reported that CIFS spuriously fails.
     * Consider it a transient error.
     */
    if (uv__is_cifs_or_smb(out_fd))
      errno = ENOSYS;  /* Use fallback. */
    break;
  case ENOTSUP:
  case EXDEV:
    /* ENOTSUP - it could work on another file system type.
     * EXDEV - it will not work when in_fd and out_fd are not on the same
     *         mounted filesystem (pre Linux 5.3)
     */
    errno = ENOSYS;  /* Use fallback. */
    break;
  }

  return -1;
}

#endif  /* __linux__ */


static ssize_t uv__fs_sendfile(uv_fs_t* req) {
  int in_fd;
  int out_fd;

  in_fd = req->flags;
  out_fd = req->file;

#if defined(__linux__) || defined(__sun)
  {
    off_t off;
    ssize_t r;
    size_t len;
    int try_sendfile;

    off = req->off;
    len = req->bufsml[0].len;
    try_sendfile = 1;

#ifdef __linux__
    r = uv__fs_try_copy_file_range(in_fd, &off, out_fd, len);
    try_sendfile = (r == -1 && errno == ENOSYS);
#endif

    if (try_sendfile)
      r = sendfile(out_fd, in_fd, &off, len);

    /* sendfile() on SunOS returns EINVAL if the target fd is not a socket but
     * it still writes out data. Fortunately, we can detect it by checking if
     * the offset has been updated.
     */
    if (r != -1 || off > req->off) {
      r = off - req->off;
      req->off = off;
      return r;
    }

    if (errno == EINVAL ||
        errno == EIO ||
        errno == ENOTSOCK ||
        errno == EXDEV) {
      errno = 0;
      return uv__fs_sendfile_emul(req);
    }

    return -1;
  }
/* sendfile() on iOS(arm64) will throw SIGSYS signal cause crash. */
#elif (defined(__APPLE__) && !TARGET_OS_IPHONE)                               \
    || defined(__DragonFly__)                                                 \
    || defined(__FreeBSD__)
  {
    off_t len;
    ssize_t r;

    /* sendfile() on FreeBSD and Darwin returns EAGAIN if the target fd is in
     * non-blocking mode and not all data could be written. If a non-zero
     * number of bytes have been sent, we don't consider it an error.
     */

#if defined(__FreeBSD__) || defined(__DragonFly__)
#if defined(__FreeBSD__)
    off_t off;

    off = req->off;
    r = uv__fs_copy_file_range(in_fd, &off, out_fd, NULL, req->bufsml[0].len, 0);
    if (r >= 0) {
        r = off - req->off;
        req->off = off;
        return r;
    }
#endif
    len = 0;
    r = sendfile(in_fd, out_fd, req->off, req->bufsml[0].len, NULL, &len, 0);
#else
    /* The darwin sendfile takes len as an input for the length to send,
     * so make sure to initialize it with the caller's value. */
    len = req->bufsml[0].len;
    r = sendfile(in_fd, out_fd, req->off, &len, NULL, 0);
#endif

     /*
     * The man page for sendfile(2) on DragonFly states that `len` contains
     * a meaningful value ONLY in case of EAGAIN and EINTR.
     * Nothing is said about it's value in case of other errors, so better
     * not depend on the potential wrong assumption that is was not modified
     * by the syscall.
     */
    if (r == 0 || ((errno == EAGAIN || errno == EINTR) && len != 0)) {
      req->off += len;
      return (ssize_t) len;
    }

    if (errno == EINVAL ||
        errno == EIO ||
        errno == ENOTSOCK ||
        errno == EXDEV) {
      errno = 0;
      return uv__fs_sendfile_emul(req);
    }

    return -1;
  }
#else
  /* Squelch compiler warnings. */
  (void) &in_fd;
  (void) &out_fd;

  return uv__fs_sendfile_emul(req);
#endif
}


static ssize_t uv__fs_utime(uv_fs_t* req) {
#if defined(__linux__)                                                         \
    || defined(_AIX71)                                                         \
    || defined(__sun)                                                          \
    || defined(__HAIKU__)
  struct timespec ts[2];
  ts[0] = uv__fs_to_timespec(req->atime);
  ts[1] = uv__fs_to_timespec(req->mtime);
  return utimensat(AT_FDCWD, req->path, ts, 0);
#elif defined(__APPLE__)                                                      \
    || defined(__DragonFly__)                                                 \
    || defined(__FreeBSD__)                                                   \
    || defined(__NetBSD__)                                                    \
    || defined(__OpenBSD__)
  struct timeval tv[2];
  tv[0] = uv__fs_to_timeval(req->atime);
  tv[1] = uv__fs_to_timeval(req->mtime);
  return utimes(req->path, tv);
#elif defined(_AIX)                                                           \
    && !defined(_AIX71)
  struct utimbuf buf;
  buf.actime = req->atime;
  buf.modtime = req->mtime;
  return utime(req->path, &buf);
#elif defined(__MVS__)
  attrib_t atr;
  memset(&atr, 0, sizeof(atr));
  atr.att_mtimechg = 1;
  atr.att_atimechg = 1;
  atr.att_mtime = req->mtime;
  atr.att_atime = req->atime;
  return __lchattr((char*) req->path, &atr, sizeof(atr));
#else
  errno = ENOSYS;
  return -1;
#endif
}


static ssize_t uv__fs_lutime(uv_fs_t* req) {
#if defined(__linux__)            ||                                           \
    defined(_AIX71)               ||                                           \
    defined(__sun)                ||                                           \
    defined(__HAIKU__)            ||                                           \
    defined(__GNU__)              ||                                           \
    defined(__OpenBSD__)
  struct timespec ts[2];
  ts[0] = uv__fs_to_timespec(req->atime);
  ts[1] = uv__fs_to_timespec(req->mtime);
  return utimensat(AT_FDCWD, req->path, ts, AT_SYMLINK_NOFOLLOW);
#elif defined(__APPLE__)          ||                                          \
      defined(__DragonFly__)      ||                                          \
      defined(__FreeBSD__)        ||                                          \
      defined(__NetBSD__)
  struct timeval tv[2];
  tv[0] = uv__fs_to_timeval(req->atime);
  tv[1] = uv__fs_to_timeval(req->mtime);
  return lutimes(req->path, tv);
#else
  errno = ENOSYS;
  return -1;
#endif
}


static ssize_t uv__fs_write(uv_fs_t* req) {
  const struct iovec* bufs;
  size_t nbufs;
  ssize_t r;
  off_t off;
  int fd;

  fd = req->file;
  off = req->off;
  bufs = (const struct iovec*) req->bufs;
  nbufs = req->nbufs;

  r = 0;
  if (off < 0) {
    if (nbufs == 1)
      r = write(fd, bufs->iov_base, bufs->iov_len);
    else if (nbufs > 1)
      r = writev(fd, bufs, nbufs);
  } else {
    if (nbufs == 1)
      r = pwrite(fd, bufs->iov_base, bufs->iov_len, off);
    else if (nbufs > 1)
      r = uv__pwritev(fd, bufs, nbufs, off);
  }

  return r;
}


static ssize_t uv__fs_copyfile(uv_fs_t* req) {
  uv_fs_t fs_req;
  uv_file srcfd;
  uv_file dstfd;
  struct stat src_statsbuf;
  struct stat dst_statsbuf;
  struct timespec times[2];
  int dst_flags;
  int result;
  int err;
  off_t bytes_to_send;
  off_t in_offset;
  off_t bytes_written;
  size_t bytes_chunk;

  dstfd = -1;
  err = 0;

  /* Open the source file. */
  srcfd = uv_fs_open(NULL, &fs_req, req->path, O_RDONLY, 0, NULL);
  uv_fs_req_cleanup(&fs_req);

  if (srcfd < 0)
    return srcfd;

  /* Get the source file's mode. */
  if (uv__fstat(srcfd, &src_statsbuf)) {
    err = UV__ERR(errno);
    goto out;
  }

  dst_flags = O_WRONLY | O_CREAT;

  if (req->flags & UV_FS_COPYFILE_EXCL)
    dst_flags |= O_EXCL;

  /* Open the destination file. */
  dstfd = uv_fs_open(NULL,
                     &fs_req,
                     req->new_path,
                     dst_flags,
                     src_statsbuf.st_mode,
                     NULL);
  uv_fs_req_cleanup(&fs_req);

  if (dstfd < 0) {
    err = dstfd;
    goto out;
  }

  /* If the file is not being opened exclusively, verify that the source and
     destination are not the same file. If they are the same, bail out early. */
  if ((req->flags & UV_FS_COPYFILE_EXCL) == 0) {
    /* Get the destination file's mode. */
    if (uv__fstat(dstfd, &dst_statsbuf)) {
      err = UV__ERR(errno);
      goto out;
    }

    /* Check if srcfd and dstfd refer to the same file */
    if (src_statsbuf.st_dev == dst_statsbuf.st_dev &&
        src_statsbuf.st_ino == dst_statsbuf.st_ino) {
      goto out;
    }

    /* Truncate the file in case the destination already existed. */
    if (ftruncate(dstfd, 0) != 0) {
      err = UV__ERR(errno);

      /* ftruncate() on ceph-fuse fails with EACCES when the file is created
       * with read only permissions. Since ftruncate() on a newly created
       * file is a meaningless operation anyway, detect that condition
       * and squelch the error.
       */
      if (err != UV_EACCES)
        goto out;

      if (dst_statsbuf.st_size > 0)
        goto out;

      err = 0;
    }
  }

  /**
   * Change the timestamps of the destination file to match the source file.
   */
#if defined(__APPLE__)
  times[0] = src_statsbuf.st_atimespec;
  times[1] = src_statsbuf.st_mtimespec;
#elif defined(_AIX)
  times[0].tv_sec = src_statsbuf.st_atime;
  times[0].tv_nsec = src_statsbuf.st_atime_n;
  times[1].tv_sec = src_statsbuf.st_mtime;
  times[1].tv_nsec = src_statsbuf.st_mtime_n;
#else
  times[0] = src_statsbuf.st_atim;
  times[1] = src_statsbuf.st_mtim;
#endif

  if (futimens(dstfd, times) == -1) {
    err = UV__ERR(errno);
    goto out;
  }

  /*
   * Change the ownership and permissions of the destination file to match the
   * source file.
   * `cp -p` does not care about errors here, so we don't either. Reuse the
   * `result` variable to silence a -Wunused-result warning.
   */
  result = fchown(dstfd, src_statsbuf.st_uid, src_statsbuf.st_gid);

  if (fchmod(dstfd, src_statsbuf.st_mode) == -1) {
    err = UV__ERR(errno);
#ifdef __linux__
    /* fchmod() on CIFS shares always fails with EPERM unless the share is
     * mounted with "noperm". As fchmod() is a meaningless operation on such
     * shares anyway, detect that condition and squelch the error.
     */
    if (err != UV_EPERM)
      goto out;

    if (!uv__is_cifs_or_smb(dstfd))
      goto out;

    err = 0;
#else  /* !__linux__ */
    goto out;
#endif  /* !__linux__ */
  }

#ifdef FICLONE
  if (req->flags & UV_FS_COPYFILE_FICLONE ||
      req->flags & UV_FS_COPYFILE_FICLONE_FORCE) {
    if (ioctl(dstfd, FICLONE, srcfd) == 0) {
      /* ioctl() with FICLONE succeeded. */
      goto out;
    }
    /* If an error occurred and force was set, return the error to the caller;
     * fall back to sendfile() when force was not set. */
    if (req->flags & UV_FS_COPYFILE_FICLONE_FORCE) {
      err = UV__ERR(errno);
      goto out;
    }
  }
#else
  if (req->flags & UV_FS_COPYFILE_FICLONE_FORCE) {
    err = UV_ENOSYS;
    goto out;
  }
#endif

  bytes_to_send = src_statsbuf.st_size;
  in_offset = 0;
  while (bytes_to_send != 0) {
    bytes_chunk = SSIZE_MAX;
    if (bytes_to_send < (off_t) bytes_chunk)
      bytes_chunk = bytes_to_send;
    uv_fs_sendfile(NULL, &fs_req, dstfd, srcfd, in_offset, bytes_chunk, NULL);
    bytes_written = fs_req.result;
    uv_fs_req_cleanup(&fs_req);

    if (bytes_written < 0) {
      err = bytes_written;
      break;
    }

    bytes_to_send -= bytes_written;
    in_offset += bytes_written;
  }

out:
  if (err < 0)
    result = err;
  else
    result = 0;

  /* Close the source file. */
  err = uv__close_nocheckstdio(srcfd);

  /* Don't overwrite any existing errors. */
  if (err != 0 && result == 0)
    result = err;

  /* Close the destination file if it is open. */
  if (dstfd >= 0) {
    err = uv__close_nocheckstdio(dstfd);

    /* Don't overwrite any existing errors. */
    if (err != 0 && result == 0)
      result = err;

    /* Remove the destination file if something went wrong. */
    if (result != 0) {
      uv_fs_unlink(NULL, &fs_req, req->new_path, NULL);
      /* Ignore the unlink return value, as an error already happened. */
      uv_fs_req_cleanup(&fs_req);
    }
  }

  if (result == 0)
    return 0;

  errno = UV__ERR(result);
  return -1;
}

static void uv__to_stat(struct stat* src, uv_stat_t* dst) {
  dst->st_dev = src->st_dev;
  dst->st_mode = src->st_mode;
  dst->st_nlink = src->st_nlink;
  dst->st_uid = src->st_uid;
  dst->st_gid = src->st_gid;
  dst->st_rdev = src->st_rdev;
  dst->st_ino = src->st_ino;
  dst->st_size = src->st_size;
  dst->st_blksize = src->st_blksize;
  dst->st_blocks = src->st_blocks;

#if defined(__APPLE__)
  dst->st_atim.tv_sec = src->st_atimespec.tv_sec;
  dst->st_atim.tv_nsec = src->st_atimespec.tv_nsec;
  dst->st_mtim.tv_sec = src->st_mtimespec.tv_sec;
  dst->st_mtim.tv_nsec = src->st_mtimespec.tv_nsec;
  dst->st_ctim.tv_sec = src->st_ctimespec.tv_sec;
  dst->st_ctim.tv_nsec = src->st_ctimespec.tv_nsec;
  dst->st_birthtim.tv_sec = src->st_birthtimespec.tv_sec;
  dst->st_birthtim.tv_nsec = src->st_birthtimespec.tv_nsec;
  dst->st_flags = src->st_flags;
  dst->st_gen = src->st_gen;
#elif defined(__ANDROID__)
  dst->st_atim.tv_sec = src->st_atime;
  dst->st_atim.tv_nsec = src->st_atimensec;
  dst->st_mtim.tv_sec = src->st_mtime;
  dst->st_mtim.tv_nsec = src->st_mtimensec;
  dst->st_ctim.tv_sec = src->st_ctime;
  dst->st_ctim.tv_nsec = src->st_ctimensec;
  dst->st_birthtim.tv_sec = src->st_ctime;
  dst->st_birthtim.tv_nsec = src->st_ctimensec;
  dst->st_flags = 0;
  dst->st_gen = 0;
#elif !defined(_AIX) &&         \
    !defined(__MVS__) && (      \
    defined(__DragonFly__)   || \
    defined(__FreeBSD__)     || \
    defined(__OpenBSD__)     || \
    defined(__NetBSD__)      || \
    defined(_GNU_SOURCE)     || \
    defined(_BSD_SOURCE)     || \
    defined(_SVID_SOURCE)    || \
    defined(_XOPEN_SOURCE)   || \
    defined(_DEFAULT_SOURCE))
  dst->st_atim.tv_sec = src->st_atim.tv_sec;
  dst->st_atim.tv_nsec = src->st_atim.tv_nsec;
  dst->st_mtim.tv_sec = src->st_mtim.tv_sec;
  dst->st_mtim.tv_nsec = src->st_mtim.tv_nsec;
  dst->st_ctim.tv_sec = src->st_ctim.tv_sec;
  dst->st_ctim.tv_nsec = src->st_ctim.tv_nsec;
# if defined(__FreeBSD__)    || \
     defined(__NetBSD__)
  dst->st_birthtim.tv_sec = src->st_birthtim.tv_sec;
  dst->st_birthtim.tv_nsec = src->st_birthtim.tv_nsec;
  dst->st_flags = src->st_flags;
  dst->st_gen = src->st_gen;
# else
  dst->st_birthtim.tv_sec = src->st_ctim.tv_sec;
  dst->st_birthtim.tv_nsec = src->st_ctim.tv_nsec;
  dst->st_flags = 0;
  dst->st_gen = 0;
# endif
#else
  dst->st_atim.tv_sec = src->st_atime;
  dst->st_atim.tv_nsec = 0;
  dst->st_mtim.tv_sec = src->st_mtime;
  dst->st_mtim.tv_nsec = 0;
  dst->st_ctim.tv_sec = src->st_ctime;
  dst->st_ctim.tv_nsec = 0;
  dst->st_birthtim.tv_sec = src->st_ctime;
  dst->st_birthtim.tv_nsec = 0;
  dst->st_flags = 0;
  dst->st_gen = 0;
#endif
}


static int uv__fs_statx(int fd,
                        const char* path,
                        int is_fstat,
                        int is_lstat,
                        uv_stat_t* buf) {
  STATIC_ASSERT(UV_ENOSYS != -1);
#ifdef __linux__
  static _Atomic int no_statx;
  struct uv__statx statxbuf;
  int dirfd;
  int flags;
  int mode;
  int rc;

  if (atomic_load_explicit(&no_statx, memory_order_relaxed))
    return UV_ENOSYS;

  dirfd = AT_FDCWD;
  flags = 0; /* AT_STATX_SYNC_AS_STAT */
  mode = 0xFFF; /* STATX_BASIC_STATS + STATX_BTIME */

  if (is_fstat) {
    dirfd = fd;
    flags |= 0x1000; /* AT_EMPTY_PATH */
  }

  if (is_lstat)
    flags |= AT_SYMLINK_NOFOLLOW;

  rc = uv__statx(dirfd, path, flags, mode, &statxbuf);

  switch (rc) {
  case 0:
    break;
  case -1:
    /* EPERM happens when a seccomp filter rejects the system call.
     * Has been observed with libseccomp < 2.3.3 and docker < 18.04.
     * EOPNOTSUPP is used on DVS exported filesystems
     */
    if (errno != EINVAL && errno != EPERM && errno != ENOSYS && errno != EOPNOTSUPP)
      return -1;
    /* Fall through. */
  default:
    /* Normally on success, zero is returned and On error, -1 is returned.
     * Observed on S390 RHEL running in a docker container with statx not
     * implemented, rc might return 1 with 0 set as the error code in which
     * case we return ENOSYS.
     */
    atomic_store_explicit(&no_statx, 1, memory_order_relaxed);
    return UV_ENOSYS;
  }

  uv__statx_to_stat(&statxbuf, buf);

  return 0;
#else
  return UV_ENOSYS;
#endif /* __linux__ */
}


static int uv__fs_stat(const char *path, uv_stat_t *buf) {
  struct stat pbuf;
  int ret;

  ret = uv__fs_statx(-1, path, /* is_fstat */ 0, /* is_lstat */ 0, buf);
  if (ret != UV_ENOSYS)
    return ret;

  ret = uv__stat(path, &pbuf);
  if (ret == 0)
    uv__to_stat(&pbuf, buf);

  return ret;
}


static int uv__fs_lstat(const char *path, uv_stat_t *buf) {
  struct stat pbuf;
  int ret;

  ret = uv__fs_statx(-1, path, /* is_fstat */ 0, /* is_lstat */ 1, buf);
  if (ret != UV_ENOSYS)
    return ret;

  ret = uv__lstat(path, &pbuf);
  if (ret == 0)
    uv__to_stat(&pbuf, buf);

  return ret;
}


static int uv__fs_fstat(int fd, uv_stat_t *buf) {
  struct stat pbuf;
  int ret;

  ret = uv__fs_statx(fd, "", /* is_fstat */ 1, /* is_lstat */ 0, buf);
  if (ret != UV_ENOSYS)
    return ret;

  ret = uv__fstat(fd, &pbuf);
  if (ret == 0)
    uv__to_stat(&pbuf, buf);

  return ret;
}

static size_t uv__fs_buf_offset(uv_buf_t* bufs, size_t size) {
  size_t offset;
  /* Figure out which bufs are done */
  for (offset = 0; size > 0 && bufs[offset].len <= size; ++offset)
    size -= bufs[offset].len;

  /* Fix a partial read/write */
  if (size > 0) {
    bufs[offset].base += size;
    bufs[offset].len -= size;
  }
  return offset;
}

static ssize_t uv__fs_write_all(uv_fs_t* req) {
  unsigned int iovmax;
  unsigned int nbufs;
  uv_buf_t* bufs;
  ssize_t total;
  ssize_t result;

  iovmax = uv__getiovmax();
  nbufs = req->nbufs;
  bufs = req->bufs;
  total = 0;

  while (nbufs > 0) {
    req->nbufs = nbufs;
    if (req->nbufs > iovmax)
      req->nbufs = iovmax;

    do
      result = uv__fs_write(req);
    while (result < 0 && errno == EINTR);

    if (result <= 0) {
      if (total == 0)
        total = result;
      break;
    }

    if (req->off >= 0)
      req->off += result;

    req->nbufs = uv__fs_buf_offset(req->bufs, result);
    req->bufs += req->nbufs;
    nbufs -= req->nbufs;
    total += result;
  }

  if (bufs != req->bufsml)
    uv__free(bufs);

  req->bufs = NULL;
  req->nbufs = 0;

  return total;
}


static void uv__fs_work(struct uv__work* w) {
  int retry_on_eintr;
  uv_fs_t* req;
  ssize_t r;

  req = container_of(w, uv_fs_t, work_req);
  retry_on_eintr = !(req->fs_type == UV_FS_CLOSE ||
                     req->fs_type == UV_FS_READ);

  do {
    errno = 0;

#define X(type, action)                                                       \
  case UV_FS_ ## type:                                                        \
    r = action;                                                               \
    break;

    switch (req->fs_type) {
    X(ACCESS, access(req->path, req->flags));
    X(CHMOD, chmod(req->path, req->mode));
    X(CHOWN, chown(req->path, req->uid, req->gid));
    X(CLOSE, uv__fs_close(req->file));
    X(COPYFILE, uv__fs_copyfile(req));
    X(FCHMOD, fchmod(req->file, req->mode));
    X(FCHOWN, fchown(req->file, req->uid, req->gid));
    X(LCHOWN, lchown(req->path, req->uid, req->gid));
    X(FDATASYNC, uv__fs_fdatasync(req));
    X(FSTAT, uv__fs_fstat(req->file, &req->statbuf));
    X(FSYNC, uv__fs_fsync(req));
    X(FTRUNCATE, ftruncate(req->file, req->off));
    X(FUTIME, uv__fs_futime(req));
    X(LUTIME, uv__fs_lutime(req));
    X(LSTAT, uv__fs_lstat(req->path, &req->statbuf));
    X(LINK, link(req->path, req->new_path));
    X(MKDIR, mkdir(req->path, req->mode));
    X(MKDTEMP, uv__fs_mkdtemp(req));
    X(MKSTEMP, uv__fs_mkstemp(req));
    X(OPEN, uv__fs_open(req));
    X(READ, uv__fs_read(req));
    X(SCANDIR, uv__fs_scandir(req));
    X(OPENDIR, uv__fs_opendir(req));
    X(READDIR, uv__fs_readdir(req));
    X(CLOSEDIR, uv__fs_closedir(req));
    X(READLINK, uv__fs_readlink(req));
    X(REALPATH, uv__fs_realpath(req));
    X(RENAME, rename(req->path, req->new_path));
    X(RMDIR, rmdir(req->path));
    X(SENDFILE, uv__fs_sendfile(req));
    X(STAT, uv__fs_stat(req->path, &req->statbuf));
    X(STATFS, uv__fs_statfs(req));
    X(SYMLINK, symlink(req->path, req->new_path));
    X(UNLINK, unlink(req->path));
    X(UTIME, uv__fs_utime(req));
    X(WRITE, uv__fs_write_all(req));
    default: abort();
    }
#undef X
  } while (r == -1 && errno == EINTR && retry_on_eintr);

  if (r == -1)
    req->result = UV__ERR(errno);
  else
    req->result = r;

  if (r == 0 && (req->fs_type == UV_FS_STAT ||
                 req->fs_type == UV_FS_FSTAT ||
                 req->fs_type == UV_FS_LSTAT)) {
    req->ptr = &req->statbuf;
  }
}


static void uv__fs_done(struct uv__work* w, int status) {
  uv_fs_t* req;

  req = container_of(w, uv_fs_t, work_req);
  uv__req_unregister(req->loop);

  if (status == UV_ECANCELED) {
    assert(req->result == 0);
    req->result = UV_ECANCELED;
  }

  req->cb(req);
}


void uv__fs_post(uv_loop_t* loop, uv_fs_t* req) {
  uv__req_register(loop);
  uv__work_submit(loop,
                  &req->work_req,
                  UV__WORK_FAST_IO,
                  uv__fs_work,
                  uv__fs_done);
}


int uv_fs_access(uv_loop_t* loop,
                 uv_fs_t* req,
                 const char* path,
                 int flags,
                 uv_fs_cb cb) {
  INIT(ACCESS);
  PATH;
  req->flags = flags;
  POST;
}


int uv_fs_chmod(uv_loop_t* loop,
                uv_fs_t* req,
                const char* path,
                int mode,
                uv_fs_cb cb) {
  INIT(CHMOD);
  PATH;
  req->mode = mode;
  POST;
}


int uv_fs_chown(uv_loop_t* loop,
                uv_fs_t* req,
                const char* path,
                uv_uid_t uid,
                uv_gid_t gid,
                uv_fs_cb cb) {
  INIT(CHOWN);
  PATH;
  req->uid = uid;
  req->gid = gid;
  POST;
}


int uv_fs_close(uv_loop_t* loop, uv_fs_t* req, uv_file file, uv_fs_cb cb) {
  INIT(CLOSE);
  req->file = file;
  if (cb != NULL)
    if (uv__iou_fs_close(loop, req))
      return 0;
  POST;
}


int uv_fs_fchmod(uv_loop_t* loop,
                 uv_fs_t* req,
                 uv_file file,
                 int mode,
                 uv_fs_cb cb) {
  INIT(FCHMOD);
  req->file = file;
  req->mode = mode;
  POST;
}


int uv_fs_fchown(uv_loop_t* loop,
                 uv_fs_t* req,
                 uv_file file,
                 uv_uid_t uid,
                 uv_gid_t gid,
                 uv_fs_cb cb) {
  INIT(FCHOWN);
  req->file = file;
  req->uid = uid;
  req->gid = gid;
  POST;
}


int uv_fs_lchown(uv_loop_t* loop,
                 uv_fs_t* req,
                 const char* path,
                 uv_uid_t uid,
                 uv_gid_t gid,
                 uv_fs_cb cb) {
  INIT(LCHOWN);
  PATH;
  req->uid = uid;
  req->gid = gid;
  POST;
}


int uv_fs_fdatasync(uv_loop_t* loop, uv_fs_t* req, uv_file file, uv_fs_cb cb) {
  INIT(FDATASYNC);
  req->file = file;
  if (cb != NULL)
    if (uv__iou_fs_fsync_or_fdatasync(loop, req, /* IORING_FSYNC_DATASYNC */ 1))
      return 0;
  POST;
}


int uv_fs_fstat(uv_loop_t* loop, uv_fs_t* req, uv_file file, uv_fs_cb cb) {
  INIT(FSTAT);
  req->file = file;
  if (cb != NULL)
    if (uv__iou_fs_statx(loop, req, /* is_fstat */ 1, /* is_lstat */ 0))
      return 0;
  POST;
}


int uv_fs_fsync(uv_loop_t* loop, uv_fs_t* req, uv_file file, uv_fs_cb cb) {
  INIT(FSYNC);
  req->file = file;
  if (cb != NULL)
    if (uv__iou_fs_fsync_or_fdatasync(loop, req, /* no flags */ 0))
      return 0;
  POST;
}


int uv_fs_ftruncate(uv_loop_t* loop,
                    uv_fs_t* req,
                    uv_file file,
                    int64_t off,
                    uv_fs_cb cb) {
  INIT(FTRUNCATE);
  req->file = file;
  req->off = off;
  POST;
}


int uv_fs_futime(uv_loop_t* loop,
                 uv_fs_t* req,
                 uv_file file,
                 double atime,
                 double mtime,
                 uv_fs_cb cb) {
  INIT(FUTIME);
  req->file = file;
  req->atime = atime;
  req->mtime = mtime;
  POST;
}

int uv_fs_lutime(uv_loop_t* loop,
                 uv_fs_t* req,
                 const char* path,
                 double atime,
                 double mtime,
                 uv_fs_cb cb) {
  INIT(LUTIME);
  PATH;
  req->atime = atime;
  req->mtime = mtime;
  POST;
}


int uv_fs_lstat(uv_loop_t* loop, uv_fs_t* req, const char* path, uv_fs_cb cb) {
  INIT(LSTAT);
  PATH;
  if (cb != NULL)
    if (uv__iou_fs_statx(loop, req, /* is_fstat */ 0, /* is_lstat */ 1))
      return 0;
  POST;
}


int uv_fs_link(uv_loop_t* loop,
               uv_fs_t* req,
               const char* path,
               const char* new_path,
               uv_fs_cb cb) {
  INIT(LINK);
  PATH2;
  if (cb != NULL)
    if (uv__iou_fs_link(loop, req))
      return 0;
  POST;
}


int uv_fs_mkdir(uv_loop_t* loop,
                uv_fs_t* req,
                const char* path,
                int mode,
                uv_fs_cb cb) {
  INIT(MKDIR);
  PATH;
  req->mode = mode;
  if (cb != NULL)
    if (uv__iou_fs_mkdir(loop, req))
      return 0;
  POST;
}


int uv_fs_mkdtemp(uv_loop_t* loop,
                  uv_fs_t* req,
                  const char* tpl,
                  uv_fs_cb cb) {
  INIT(MKDTEMP);
  req->path = uv__strdup(tpl);
  if (req->path == NULL)
    return UV_ENOMEM;
  POST;
}


int uv_fs_mkstemp(uv_loop_t* loop,
                  uv_fs_t* req,
                  const char* tpl,
                  uv_fs_cb cb) {
  INIT(MKSTEMP);
  req->path = uv__strdup(tpl);
  if (req->path == NULL)
    return UV_ENOMEM;
  POST;
}


int uv_fs_open(uv_loop_t* loop,
               uv_fs_t* req,
               const char* path,
               int flags,
               int mode,
               uv_fs_cb cb) {
  INIT(OPEN);
  PATH;
  req->flags = flags;
  req->mode = mode;
  if (cb != NULL)
    if (uv__iou_fs_open(loop, req))
      return 0;
  POST;
}


int uv_fs_read(uv_loop_t* loop, uv_fs_t* req,
               uv_file file,
               const uv_buf_t bufs[],
               unsigned int nbufs,
               int64_t off,
               uv_fs_cb cb) {
  INIT(READ);

  if (bufs == NULL || nbufs == 0)
    return UV_EINVAL;

  req->off = off;
  req->file = file;
  req->bufs = (uv_buf_t*) bufs;  /* Safe, doesn't mutate |bufs| */
  req->nbufs = nbufs;

  if (cb == NULL)
    goto post;

  req->bufs = req->bufsml;
  if (nbufs > ARRAY_SIZE(req->bufsml))
    req->bufs = uv__malloc(nbufs * sizeof(*bufs));

  if (req->bufs == NULL)
    return UV_ENOMEM;

  memcpy(req->bufs, bufs, nbufs * sizeof(*bufs));

  if (uv__iou_fs_read_or_write(loop, req, /* is_read */ 1))
    return 0;

post:
  POST;
}


int uv_fs_scandir(uv_loop_t* loop,
                  uv_fs_t* req,
                  const char* path,
                  int flags,
                  uv_fs_cb cb) {
  INIT(SCANDIR);
  PATH;
  req->flags = flags;
  POST;
}

int uv_fs_opendir(uv_loop_t* loop,
                  uv_fs_t* req,
                  const char* path,
                  uv_fs_cb cb) {
  INIT(OPENDIR);
  PATH;
  POST;
}

int uv_fs_readdir(uv_loop_t* loop,
                  uv_fs_t* req,
                  uv_dir_t* dir,
                  uv_fs_cb cb) {
  INIT(READDIR);

  if (dir == NULL || dir->dir == NULL || dir->dirents == NULL)
    return UV_EINVAL;

  req->ptr = dir;
  POST;
}

int uv_fs_closedir(uv_loop_t* loop,
                   uv_fs_t* req,
                   uv_dir_t* dir,
                   uv_fs_cb cb) {
  INIT(CLOSEDIR);

  if (dir == NULL)
    return UV_EINVAL;

  req->ptr = dir;
  POST;
}

int uv_fs_readlink(uv_loop_t* loop,
                   uv_fs_t* req,
                   const char* path,
                   uv_fs_cb cb) {
  INIT(READLINK);
  PATH;
  POST;
}


int uv_fs_realpath(uv_loop_t* loop,
                  uv_fs_t* req,
                  const char * path,
                  uv_fs_cb cb) {
  INIT(REALPATH);
  PATH;
  POST;
}


int uv_fs_rename(uv_loop_t* loop,
                 uv_fs_t* req,
                 const char* path,
                 const char* new_path,
                 uv_fs_cb cb) {
  INIT(RENAME);
  PATH2;
  if (cb != NULL)
    if (uv__iou_fs_rename(loop, req))
      return 0;
  POST;
}


int uv_fs_rmdir(uv_loop_t* loop, uv_fs_t* req, const char* path, uv_fs_cb cb) {
  INIT(RMDIR);
  PATH;
  POST;
}


int uv_fs_sendfile(uv_loop_t* loop,
                   uv_fs_t* req,
                   uv_file out_fd,
                   uv_file in_fd,
                   int64_t off,
                   size_t len,
                   uv_fs_cb cb) {
  INIT(SENDFILE);
  req->flags = in_fd; /* hack */
  req->file = out_fd;
  req->off = off;
  req->bufsml[0].len = len;
  POST;
}


int uv_fs_stat(uv_loop_t* loop, uv_fs_t* req, const char* path, uv_fs_cb cb) {
  INIT(STAT);
  PATH;
  if (cb != NULL)
    if (uv__iou_fs_statx(loop, req, /* is_fstat */ 0, /* is_lstat */ 0))
      return 0;
  POST;
}


int uv_fs_symlink(uv_loop_t* loop,
                  uv_fs_t* req,
                  const char* path,
                  const char* new_path,
                  int flags,
                  uv_fs_cb cb) {
  INIT(SYMLINK);
  PATH2;
  req->flags = flags;
  if (cb != NULL)
    if (uv__iou_fs_symlink(loop, req))
      return 0;
  POST;
}


int uv_fs_unlink(uv_loop_t* loop, uv_fs_t* req, const char* path, uv_fs_cb cb) {
  INIT(UNLINK);
  PATH;
  if (cb != NULL)
    if (uv__iou_fs_unlink(loop, req))
      return 0;
  POST;
}


int uv_fs_utime(uv_loop_t* loop,
                uv_fs_t* req,
                const char* path,
                double atime,
                double mtime,
                uv_fs_cb cb) {
  INIT(UTIME);
  PATH;
  req->atime = atime;
  req->mtime = mtime;
  POST;
}


int uv_fs_write(uv_loop_t* loop,
                uv_fs_t* req,
                uv_file file,
                const uv_buf_t bufs[],
                unsigned int nbufs,
                int64_t off,
                uv_fs_cb cb) {
  INIT(WRITE);

  if (bufs == NULL || nbufs == 0)
    return UV_EINVAL;

  req->file = file;

  req->nbufs = nbufs;
  req->bufs = req->bufsml;
  if (nbufs > ARRAY_SIZE(req->bufsml))
    req->bufs = uv__malloc(nbufs * sizeof(*bufs));

  if (req->bufs == NULL)
    return UV_ENOMEM;

  memcpy(req->bufs, bufs, nbufs * sizeof(*bufs));

  req->off = off;

  if (cb != NULL)
    if (uv__iou_fs_read_or_write(loop, req, /* is_read */ 0))
      return 0;

  POST;
}


void uv_fs_req_cleanup(uv_fs_t* req) {
  if (req == NULL)
    return;

  /* Only necessary for asynchronous requests, i.e., requests with a callback.
   * Synchronous ones don't copy their arguments and have req->path and
   * req->new_path pointing to user-owned memory.  UV_FS_MKDTEMP and
   * UV_FS_MKSTEMP are the exception to the rule, they always allocate memory.
   */
  if (req->path != NULL &&
      (req->cb != NULL ||
        req->fs_type == UV_FS_MKDTEMP || req->fs_type == UV_FS_MKSTEMP))
    uv__free((void*) req->path);  /* Memory is shared with req->new_path. */

  req->path = NULL;
  req->new_path = NULL;

  if (req->fs_type == UV_FS_READDIR && req->ptr != NULL)
    uv__fs_readdir_cleanup(req);

  if (req->fs_type == UV_FS_SCANDIR && req->ptr != NULL)
    uv__fs_scandir_cleanup(req);

  if (req->bufs != req->bufsml)
    uv__free(req->bufs);
  req->bufs = NULL;

  if (req->fs_type != UV_FS_OPENDIR && req->ptr != &req->statbuf)
    uv__free(req->ptr);
  req->ptr = NULL;
}


int uv_fs_copyfile(uv_loop_t* loop,
                   uv_fs_t* req,
                   const char* path,
                   const char* new_path,
                   int flags,
                   uv_fs_cb cb) {
  INIT(COPYFILE);

  if (flags & ~(UV_FS_COPYFILE_EXCL |
                UV_FS_COPYFILE_FICLONE |
                UV_FS_COPYFILE_FICLONE_FORCE)) {
    return UV_EINVAL;
  }

  PATH2;
  req->flags = flags;
  POST;
}


int uv_fs_statfs(uv_loop_t* loop,
                 uv_fs_t* req,
                 const char* path,
                 uv_fs_cb cb) {
  INIT(STATFS);
  PATH;
  POST;
}

int uv_fs_get_system_error(const uv_fs_t* req) {
  return -req->result;
}