src/basic/copy.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <dirent.h>
   4 #include <errno.h>
   5 #include <fcntl.h>
   6 #include <stddef.h>
   7 #include <stdio.h>
   8 #include <stdlib.h>
   9 #include <string.h>
  10 #include <sys/sendfile.h>
  11 #include <sys/stat.h>
  12 #include <sys/xattr.h>
  13 #include <time.h>
  14 #include <unistd.h>
  15
  16 #include "alloc-util.h"
  17 #include "btrfs-util.h"
  18 #include "chattr-util.h"
  19 #include "copy.h"
  20 #include "dirent-util.h"
  21 #include "fd-util.h"
  22 #include "fs-util.h"
  23 #include "io-util.h"
  24 #include "macro.h"
  25 #include "missing.h"
  26 #include "mountpoint-util.h"
  27 #include "stat-util.h"
  28 #include "string-util.h"
  29 #include "strv.h"
  30 #include "time-util.h"
  31 #include "tmpfile-util.h"
  32 #include "umask-util.h"
  33 #include "user-util.h"
  34 #include "xattr-util.h"
  35
  36 #define COPY_BUFFER_SIZE (16U*1024U)
  37
  38 /* A safety net for descending recursively into file system trees to copy. On Linux PATH_MAX is 4096, which means the
  39  * deepest valid path one can build is around 2048, which we hence use as a safety net here, to not spin endlessly in
  40  * case of bind mount cycles and suchlike. */
  41 #define COPY_DEPTH_MAX 2048U
  42
  43 static ssize_t try_copy_file_range(
  44                 int fd_in, loff_t *off_in,
  45                 int fd_out, loff_t *off_out,
  46                 size_t len,
  47                 unsigned flags) {
  48
  49         static int have = -1;
  50         ssize_t r;
  51
  52         if (have == 0)
  53                 return -ENOSYS;
  54
  55         r = copy_file_range(fd_in, off_in, fd_out, off_out, len, flags);
  56         if (have < 0)
  57                 have = r >= 0 || errno != ENOSYS;
  58         if (r < 0)
  59                 return -errno;
  60
  61         return r;
  62 }
  63
  64 enum {
  65         FD_IS_NO_PIPE,
  66         FD_IS_BLOCKING_PIPE,
  67         FD_IS_NONBLOCKING_PIPE,
  68 };
  69
  70 static int fd_is_nonblock_pipe(int fd) {
  71         struct stat st;
  72         int flags;
  73
  74         /* Checks whether the specified file descriptor refers to a pipe, and if so if O_NONBLOCK is set. */
  75
  76         if (fstat(fd, &st) < 0)
  77                 return -errno;
  78
  79         if (!S_ISFIFO(st.st_mode))
  80                 return FD_IS_NO_PIPE;
  81
  82         flags = fcntl(fd, F_GETFL);
  83         if (flags < 0)
  84                 return -errno;
  85
  86         return FLAGS_SET(flags, O_NONBLOCK) ? FD_IS_NONBLOCKING_PIPE : FD_IS_BLOCKING_PIPE;
  87 }
  88
  89 int copy_bytes_full(
  90                 int fdf, int fdt,
  91                 uint64_t max_bytes,
  92                 CopyFlags copy_flags,
  93                 void **ret_remains,
  94                 size_t *ret_remains_size,
  95                 copy_progress_bytes_t progress,
  96                 void *userdata) {
  97
  98         bool try_cfr = true, try_sendfile = true, try_splice = true;
  99         int r, nonblock_pipe = -1;
 100         size_t m = SSIZE_MAX; /* that is the maximum that sendfile and c_f_r accept */
 101
 102         assert(fdf >= 0);
 103         assert(fdt >= 0);
 104
 105         /* Tries to copy bytes from the file descriptor 'fdf' to 'fdt' in the smartest possible way. Copies a maximum
 106          * of 'max_bytes', which may be specified as UINT64_MAX, in which no maximum is applied. Returns negative on
 107          * error, zero if EOF is hit before the bytes limit is hit and positive otherwise. If the copy fails for some
 108          * reason but we read but didn't yet write some data an ret_remains/ret_remains_size is not NULL, then it will
 109          * be initialized with an allocated buffer containing this "remaining" data. Note that these two parameters are
 110          * initialized with a valid buffer only on failure and only if there's actually data already read. Otherwise
 111          * these parameters if non-NULL are set to NULL. */
 112
 113         if (ret_remains)
 114                 *ret_remains = NULL;
 115         if (ret_remains_size)
 116                 *ret_remains_size = 0;
 117
 118         /* Try btrfs reflinks first. This only works on regular, seekable files, hence let's check the file offsets of
 119          * source and destination first. */
 120         if ((copy_flags & COPY_REFLINK)) {
 121                 off_t foffset;
 122
 123                 foffset = lseek(fdf, 0, SEEK_CUR);
 124                 if (foffset >= 0) {
 125                         off_t toffset;
 126
 127                         toffset = lseek(fdt, 0, SEEK_CUR);
 128                         if (toffset >= 0) {
 129
 130                                 if (foffset == 0 && toffset == 0 && max_bytes == UINT64_MAX)
 131                                         r = btrfs_reflink(fdf, fdt); /* full file reflink */
 132                                 else
 133                                         r = btrfs_clone_range(fdf, foffset, fdt, toffset, max_bytes == UINT64_MAX ? 0 : max_bytes); /* partial reflink */
 134                                 if (r >= 0) {
 135                                         off_t t;
 136
 137                                         /* This worked, yay! Now — to be fully correct — let's adjust the file pointers */
 138                                         if (max_bytes == UINT64_MAX) {
 139
 140                                                 /* We cloned to the end of the source file, let's position the read
 141                                                  * pointer there, and query it at the same time. */
 142                                                 t = lseek(fdf, 0, SEEK_END);
 143                                                 if (t < 0)
 144                                                         return -errno;
 145                                                 if (t < foffset)
 146                                                         return -ESPIPE;
 147
 148                                                 /* Let's adjust the destination file write pointer by the same number
 149                                                  * of bytes. */
 150                                                 t = lseek(fdt, toffset + (t - foffset), SEEK_SET);
 151                                                 if (t < 0)
 152                                                         return -errno;
 153
 154                                                 return 0; /* we copied the whole thing, hence hit EOF, return 0 */
 155                                         } else {
 156                                                 t = lseek(fdf, foffset + max_bytes, SEEK_SET);
 157                                                 if (t < 0)
 158                                                         return -errno;
 159
 160                                                 t = lseek(fdt, toffset + max_bytes, SEEK_SET);
 161                                                 if (t < 0)
 162                                                         return -errno;
 163
 164                                                 return 1; /* we copied only some number of bytes, which worked, but this means we didn't hit EOF, return 1 */
 165                                         }
 166                                 }
 167                         }
 168                 }
 169         }
 170
 171         for (;;) {
 172                 ssize_t n;
 173
 174                 if (max_bytes <= 0)
 175                         return 1; /* return > 0 if we hit the max_bytes limit */
 176
 177                 if (max_bytes != UINT64_MAX && m > max_bytes)
 178                         m = max_bytes;
 179
 180                 /* First try copy_file_range(), unless we already tried */
 181                 if (try_cfr) {
 182                         n = try_copy_file_range(fdf, NULL, fdt, NULL, m, 0u);
 183                         if (n < 0) {
 184                                 if (!IN_SET(n, -EINVAL, -ENOSYS, -EXDEV, -EBADF))
 185                                         return n;
 186
 187                                 try_cfr = false;
 188                                 /* use fallback below */
 189                         } else if (n == 0) /* EOF */
 190                                 break;
 191                         else
 192                                 /* Success! */
 193                                 goto next;
 194                 }
 195
 196                 /* First try sendfile(), unless we already tried */
 197                 if (try_sendfile) {
 198                         n = sendfile(fdt, fdf, NULL, m);
 199                         if (n < 0) {
 200                                 if (!IN_SET(errno, EINVAL, ENOSYS))
 201                                         return -errno;
 202
 203                                 try_sendfile = false;
 204                                 /* use fallback below */
 205                         } else if (n == 0) /* EOF */
 206                                 break;
 207                         else
 208                                 /* Success! */
 209                                 goto next;
 210                 }
 211
 212                 /* Then try splice, unless we already tried. */
 213                 if (try_splice) {
 214
 215                         /* splice()'s asynchronous I/O support is a bit weird. When it encounters a pipe file
 216                          * descriptor, then it will ignore its O_NONBLOCK flag and instead only honour the
 217                          * SPLICE_F_NONBLOCK flag specified in its flag parameter. Let's hide this behaviour here, and
 218                          * check if either of the specified fds are a pipe, and if so, let's pass the flag
 219                          * automatically, depending on O_NONBLOCK being set.
 220                          *
 221                          * Here's a twist though: when we use it to move data between two pipes of which one has
 222                          * O_NONBLOCK set and the other has not, then we have no individual control over O_NONBLOCK
 223                          * behaviour. Hence in that case we can't use splice() and still guarantee systematic
 224                          * O_NONBLOCK behaviour, hence don't. */
 225
 226                         if (nonblock_pipe < 0) {
 227                                 int a, b;
 228
 229                                 /* Check if either of these fds is a pipe, and if so non-blocking or not */
 230                                 a = fd_is_nonblock_pipe(fdf);
 231                                 if (a < 0)
 232                                         return a;
 233
 234                                 b = fd_is_nonblock_pipe(fdt);
 235                                 if (b < 0)
 236                                         return b;
 237
 238                                 if ((a == FD_IS_NO_PIPE && b == FD_IS_NO_PIPE) ||
 239                                     (a == FD_IS_BLOCKING_PIPE && b == FD_IS_NONBLOCKING_PIPE) ||
 240                                     (a == FD_IS_NONBLOCKING_PIPE && b == FD_IS_BLOCKING_PIPE))
 241
 242                                         /* splice() only works if one of the fds is a pipe. If neither is, let's skip
 243                                          * this step right-away. As mentioned above, if one of the two fds refers to a
 244                                          * blocking pipe and the other to a non-blocking pipe, we can't use splice()
 245                                          * either, hence don't try either. This hence means we can only use splice() if
 246                                          * either only one of the two fds is a pipe, or if both are pipes with the same
 247                                          * nonblocking flag setting. */
 248
 249                                         try_splice = false;
 250                                 else
 251                                         nonblock_pipe = a == FD_IS_NONBLOCKING_PIPE || b == FD_IS_NONBLOCKING_PIPE;
 252                         }
 253                 }
 254
 255                 if (try_splice) {
 256                         n = splice(fdf, NULL, fdt, NULL, m, nonblock_pipe ? SPLICE_F_NONBLOCK : 0);
 257                         if (n < 0) {
 258                                 if (!IN_SET(errno, EINVAL, ENOSYS))
 259                                         return -errno;
 260
 261                                 try_splice = false;
 262                                 /* use fallback below */
 263                         } else if (n == 0) /* EOF */
 264                                 break;
 265                         else
 266                                 /* Success! */
 267                                 goto next;
 268                 }
 269
 270                 /* As a fallback just copy bits by hand */
 271                 {
 272                         uint8_t buf[MIN(m, COPY_BUFFER_SIZE)], *p = buf;
 273                         ssize_t z;
 274
 275                         n = read(fdf, buf, sizeof buf);
 276                         if (n < 0)
 277                                 return -errno;
 278                         if (n == 0) /* EOF */
 279                                 break;
 280
 281                         z = (size_t) n;
 282                         do {
 283                                 ssize_t k;
 284
 285                                 k = write(fdt, p, z);
 286                                 if (k < 0) {
 287                                         r = -errno;
 288
 289                                         if (ret_remains) {
 290                                                 void *copy;
 291
 292                                                 copy = memdup(p, z);
 293                                                 if (!copy)
 294                                                         return -ENOMEM;
 295
 296                                                 *ret_remains = copy;
 297                                         }
 298
 299                                         if (ret_remains_size)
 300                                                 *ret_remains_size = z;
 301
 302                                         return r;
 303                                 }
 304
 305                                 assert(k <= z);
 306                                 z -= k;
 307                                 p += k;
 308                         } while (z > 0);
 309                 }
 310
 311         next:
 312                 if (progress) {
 313                         r = progress(n, userdata);
 314                         if (r < 0)
 315                                 return r;
 316                 }
 317
 318                 if (max_bytes != (uint64_t) -1) {
 319                         assert(max_bytes >= (uint64_t) n);
 320                         max_bytes -= n;
 321                 }
 322
 323                 /* sendfile accepts at most SSIZE_MAX-offset bytes to copy,
 324                  * so reduce our maximum by the amount we already copied,
 325                  * but don't go below our copy buffer size, unless we are
 326                  * close the limit of bytes we are allowed to copy. */
 327                 m = MAX(MIN(COPY_BUFFER_SIZE, max_bytes), m - n);
 328         }
 329
 330         return 0; /* return 0 if we hit EOF earlier than the size limit */
 331 }
 332
 333 static int fd_copy_symlink(
 334                 int df,
 335                 const char *from,
 336                 const struct stat *st,
 337                 int dt,
 338                 const char *to,
 339                 uid_t override_uid,
 340                 gid_t override_gid,
 341                 CopyFlags copy_flags) {
 342
 343         _cleanup_free_ char *target = NULL;
 344         int r;
 345
 346         assert(from);
 347         assert(st);
 348         assert(to);
 349
 350         r = readlinkat_malloc(df, from, &target);
 351         if (r < 0)
 352                 return r;
 353
 354         if (symlinkat(target, dt, to) < 0)
 355                 return -errno;
 356
 357         if (fchownat(dt, to,
 358                      uid_is_valid(override_uid) ? override_uid : st->st_uid,
 359                      gid_is_valid(override_gid) ? override_gid : st->st_gid,
 360                      AT_SYMLINK_NOFOLLOW) < 0)
 361                 return -errno;
 362
 363         return 0;
 364 }
 365
 366 static int fd_copy_regular(
 367                 int df,
 368                 const char *from,
 369                 const struct stat *st,
 370                 int dt,
 371                 const char *to,
 372                 uid_t override_uid,
 373                 gid_t override_gid,
 374                 CopyFlags copy_flags,
 375                 copy_progress_bytes_t progress,
 376                 void *userdata) {
 377
 378         _cleanup_close_ int fdf = -1, fdt = -1;
 379         struct timespec ts[2];
 380         int r, q;
 381
 382         assert(from);
 383         assert(st);
 384         assert(to);
 385
 386         fdf = openat(df, from, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
 387         if (fdf < 0)
 388                 return -errno;
 389
 390         fdt = openat(dt, to, O_WRONLY|O_CREAT|O_EXCL|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW, st->st_mode & 07777);
 391         if (fdt < 0)
 392                 return -errno;
 393
 394         r = copy_bytes_full(fdf, fdt, (uint64_t) -1, copy_flags, NULL, NULL, progress, userdata);
 395         if (r < 0) {
 396                 (void) unlinkat(dt, to, 0);
 397                 return r;
 398         }
 399
 400         if (fchown(fdt,
 401                    uid_is_valid(override_uid) ? override_uid : st->st_uid,
 402                    gid_is_valid(override_gid) ? override_gid : st->st_gid) < 0)
 403                 r = -errno;
 404
 405         if (fchmod(fdt, st->st_mode & 07777) < 0)
 406                 r = -errno;
 407
 408         ts[0] = st->st_atim;
 409         ts[1] = st->st_mtim;
 410         (void) futimens(fdt, ts);
 411         (void) copy_xattr(fdf, fdt);
 412
 413         q = close(fdt);
 414         fdt = -1;
 415
 416         if (q < 0) {
 417                 r = -errno;
 418                 (void) unlinkat(dt, to, 0);
 419         }
 420
 421         return r;
 422 }
 423
 424 static int fd_copy_fifo(
 425                 int df,
 426                 const char *from,
 427                 const struct stat *st,
 428                 int dt,
 429                 const char *to,
 430                 uid_t override_uid,
 431                 gid_t override_gid,
 432                 CopyFlags copy_flags) {
 433         int r;
 434
 435         assert(from);
 436         assert(st);
 437         assert(to);
 438
 439         r = mkfifoat(dt, to, st->st_mode & 07777);
 440         if (r < 0)
 441                 return -errno;
 442
 443         if (fchownat(dt, to,
 444                      uid_is_valid(override_uid) ? override_uid : st->st_uid,
 445                      gid_is_valid(override_gid) ? override_gid : st->st_gid,
 446                      AT_SYMLINK_NOFOLLOW) < 0)
 447                 r = -errno;
 448
 449         if (fchmodat(dt, to, st->st_mode & 07777, 0) < 0)
 450                 r = -errno;
 451
 452         return r;
 453 }
 454
 455 static int fd_copy_node(
 456                 int df,
 457                 const char *from,
 458                 const struct stat *st,
 459                 int dt,
 460                 const char *to,
 461                 uid_t override_uid,
 462                 gid_t override_gid,
 463                 CopyFlags copy_flags) {
 464         int r;
 465
 466         assert(from);
 467         assert(st);
 468         assert(to);
 469
 470         r = mknodat(dt, to, st->st_mode, st->st_rdev);
 471         if (r < 0)
 472                 return -errno;
 473
 474         if (fchownat(dt, to,
 475                      uid_is_valid(override_uid) ? override_uid : st->st_uid,
 476                      gid_is_valid(override_gid) ? override_gid : st->st_gid,
 477                      AT_SYMLINK_NOFOLLOW) < 0)
 478                 r = -errno;
 479
 480         if (fchmodat(dt, to, st->st_mode & 07777, 0) < 0)
 481                 r = -errno;
 482
 483         return r;
 484 }
 485
 486 static int fd_copy_directory(
 487                 int df,
 488                 const char *from,
 489                 const struct stat *st,
 490                 int dt,
 491                 const char *to,
 492                 dev_t original_device,
 493                 unsigned depth_left,
 494                 uid_t override_uid,
 495                 gid_t override_gid,
 496                 CopyFlags copy_flags,
 497                 const char *display_path,
 498                 copy_progress_path_t progress_path,
 499                 copy_progress_bytes_t progress_bytes,
 500                 void *userdata) {
 501
 502         _cleanup_close_ int fdf = -1, fdt = -1;
 503         _cleanup_closedir_ DIR *d = NULL;
 504         struct dirent *de;
 505         bool exists, created;
 506         int r;
 507
 508         assert(st);
 509         assert(to);
 510
 511         if (depth_left == 0)
 512                 return -ENAMETOOLONG;
 513
 514         if (from)
 515                 fdf = openat(df, from, O_RDONLY|O_DIRECTORY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
 516         else
 517                 fdf = fcntl(df, F_DUPFD_CLOEXEC, 3);
 518         if (fdf < 0)
 519                 return -errno;
 520
 521         d = fdopendir(fdf);
 522         if (!d)
 523                 return -errno;
 524         fdf = -1;
 525
 526         exists = false;
 527         if (copy_flags & COPY_MERGE_EMPTY) {
 528                 r = dir_is_empty_at(dt, to);
 529                 if (r < 0 && r != -ENOENT)
 530                         return r;
 531                 else if (r == 1)
 532                         exists = true;
 533         }
 534
 535         if (exists)
 536                 created = false;
 537         else {
 538                 r = mkdirat(dt, to, st->st_mode & 07777);
 539                 if (r >= 0)
 540                         created = true;
 541                 else if (errno == EEXIST && (copy_flags & COPY_MERGE))
 542                         created = false;
 543                 else
 544                         return -errno;
 545         }
 546
 547         fdt = openat(dt, to, O_RDONLY|O_DIRECTORY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
 548         if (fdt < 0)
 549                 return -errno;
 550
 551         r = 0;
 552
 553         FOREACH_DIRENT_ALL(de, d, return -errno) {
 554                 const char *child_display_path = NULL;
 555                 _cleanup_free_ char *dp = NULL;
 556                 struct stat buf;
 557                 int q;
 558
 559                 if (dot_or_dot_dot(de->d_name))
 560                         continue;
 561
 562                 if (fstatat(dirfd(d), de->d_name, &buf, AT_SYMLINK_NOFOLLOW) < 0) {
 563                         r = -errno;
 564                         continue;
 565                 }
 566
 567                 if (progress_path) {
 568                         if (display_path)
 569                                 child_display_path = dp = strjoin(display_path, "/", de->d_name);
 570                         else
 571                                 child_display_path = de->d_name;
 572
 573                         r = progress_path(child_display_path, &buf, userdata);
 574                         if (r < 0)
 575                                 return r;
 576                 }
 577
 578                 if (S_ISDIR(buf.st_mode)) {
 579                         /*
 580                          * Don't descend into directories on other file systems, if this is requested. We do a simple
 581                          * .st_dev check here, which basically comes for free. Note that we do this check only on
 582                          * directories, not other kind of file system objects, for two reason:
 583                          *
 584                          * • The kernel's overlayfs pseudo file system that overlays multiple real file systems
 585                          *   propagates the .st_dev field of the file system a file originates from all the way up
 586                          *   through the stack to stat(). It doesn't do that for directories however. This means that
 587                          *   comparing .st_dev on non-directories suggests that they all are mount points. To avoid
 588                          *   confusion we hence avoid relying on this check for regular files.
 589                          *
 590                          * • The main reason we do this check at all is to protect ourselves from bind mount cycles,
 591                          *   where we really want to avoid descending down in all eternity. However the .st_dev check
 592                          *   is usually not sufficient for this protection anyway, as bind mount cycles from the same
 593                          *   file system onto itself can't be detected that way. (Note we also do a recursion depth
 594                          *   check, which is probably the better protection in this regard, which is why
 595                          *   COPY_SAME_MOUNT is optional).
 596                          */
 597
 598                         if (FLAGS_SET(copy_flags, COPY_SAME_MOUNT)) {
 599                                 if (buf.st_dev != original_device)
 600                                         continue;
 601
 602                                 r = fd_is_mount_point(dirfd(d), de->d_name, 0);
 603                                 if (r < 0)
 604                                         return r;
 605                                 if (r > 0)
 606                                         continue;
 607                         }
 608
 609                         q = fd_copy_directory(dirfd(d), de->d_name, &buf, fdt, de->d_name, original_device, depth_left-1, override_uid, override_gid, copy_flags, child_display_path, progress_path, progress_bytes, userdata);
 610                 } else if (S_ISREG(buf.st_mode))
 611                         q = fd_copy_regular(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags, progress_bytes, userdata);
 612                 else if (S_ISLNK(buf.st_mode))
 613                         q = fd_copy_symlink(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
 614                 else if (S_ISFIFO(buf.st_mode))
 615                         q = fd_copy_fifo(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
 616                 else if (S_ISBLK(buf.st_mode) || S_ISCHR(buf.st_mode) || S_ISSOCK(buf.st_mode))
 617                         q = fd_copy_node(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
 618                 else
 619                         q = -EOPNOTSUPP;
 620
 621                 if (q == -EEXIST && (copy_flags & COPY_MERGE))
 622                         q = 0;
 623
 624                 if (q < 0)
 625                         r = q;
 626         }
 627
 628         if (created) {
 629                 struct timespec ut[2] = {
 630                         st->st_atim,
 631                         st->st_mtim
 632                 };
 633
 634                 if (fchown(fdt,
 635                            uid_is_valid(override_uid) ? override_uid : st->st_uid,
 636                            gid_is_valid(override_gid) ? override_gid : st->st_gid) < 0)
 637                         r = -errno;
 638
 639                 if (fchmod(fdt, st->st_mode & 07777) < 0)
 640                         r = -errno;
 641
 642                 (void) copy_xattr(dirfd(d), fdt);
 643                 (void) futimens(fdt, ut);
 644         }
 645
 646         return r;
 647 }
 648
 649 int copy_tree_at_full(
 650                 int fdf,
 651                 const char *from,
 652                 int fdt,
 653                 const char *to,
 654                 uid_t override_uid,
 655                 gid_t override_gid,
 656                 CopyFlags copy_flags,
 657                 copy_progress_path_t progress_path,
 658                 copy_progress_bytes_t progress_bytes,
 659                 void *userdata) {
 660
 661         struct stat st;
 662
 663         assert(from);
 664         assert(to);
 665
 666         if (fstatat(fdf, from, &st, AT_SYMLINK_NOFOLLOW) < 0)
 667                 return -errno;
 668
 669         if (S_ISREG(st.st_mode))
 670                 return fd_copy_regular(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags, progress_bytes, userdata);
 671         else if (S_ISDIR(st.st_mode))
 672                 return fd_copy_directory(fdf, from, &st, fdt, to, st.st_dev, COPY_DEPTH_MAX, override_uid, override_gid, copy_flags, NULL, progress_path, progress_bytes, userdata);
 673         else if (S_ISLNK(st.st_mode))
 674                 return fd_copy_symlink(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
 675         else if (S_ISFIFO(st.st_mode))
 676                 return fd_copy_fifo(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
 677         else if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode) || S_ISSOCK(st.st_mode))
 678                 return fd_copy_node(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
 679         else
 680                 return -EOPNOTSUPP;
 681 }
 682
 683 int copy_directory_fd_full(
 684                 int dirfd,
 685                 const char *to,
 686                 CopyFlags copy_flags,
 687                 copy_progress_path_t progress_path,
 688                 copy_progress_bytes_t progress_bytes,
 689                 void *userdata) {
 690
 691         struct stat st;
 692
 693         assert(dirfd >= 0);
 694         assert(to);
 695
 696         if (fstat(dirfd, &st) < 0)
 697                 return -errno;
 698
 699         if (!S_ISDIR(st.st_mode))
 700                 return -ENOTDIR;
 701
 702         return fd_copy_directory(dirfd, NULL, &st, AT_FDCWD, to, st.st_dev, COPY_DEPTH_MAX, UID_INVALID, GID_INVALID, copy_flags, NULL, progress_path, progress_bytes, userdata);
 703 }
 704
 705 int copy_directory_full(
 706                 const char *from,
 707                 const char *to,
 708                 CopyFlags copy_flags,
 709                 copy_progress_path_t progress_path,
 710                 copy_progress_bytes_t progress_bytes,
 711                 void *userdata) {
 712
 713         struct stat st;
 714
 715         assert(from);
 716         assert(to);
 717
 718         if (lstat(from, &st) < 0)
 719                 return -errno;
 720
 721         if (!S_ISDIR(st.st_mode))
 722                 return -ENOTDIR;
 723
 724         return fd_copy_directory(AT_FDCWD, from, &st, AT_FDCWD, to, st.st_dev, COPY_DEPTH_MAX, UID_INVALID, GID_INVALID, copy_flags, NULL, progress_path, progress_bytes, userdata);
 725 }
 726
 727 int copy_file_fd_full(
 728                 const char *from,
 729                 int fdt,
 730                 CopyFlags copy_flags,
 731                 copy_progress_bytes_t progress_bytes,
 732                 void *userdata) {
 733
 734         _cleanup_close_ int fdf = -1;
 735         int r;
 736
 737         assert(from);
 738         assert(fdt >= 0);
 739
 740         fdf = open(from, O_RDONLY|O_CLOEXEC|O_NOCTTY);
 741         if (fdf < 0)
 742                 return -errno;
 743
 744         r = copy_bytes_full(fdf, fdt, (uint64_t) -1, copy_flags, NULL, NULL, progress_bytes, userdata);
 745
 746         (void) copy_times(fdf, fdt);
 747         (void) copy_xattr(fdf, fdt);
 748
 749         return r;
 750 }
 751
 752 int copy_file_full(
 753                 const char *from,
 754                 const char *to,
 755                 int flags,
 756                 mode_t mode,
 757                 unsigned chattr_flags,
 758                 CopyFlags copy_flags,
 759                 copy_progress_bytes_t progress_bytes,
 760                 void *userdata) {
 761
 762         int fdt = -1, r;
 763
 764         assert(from);
 765         assert(to);
 766
 767         RUN_WITH_UMASK(0000) {
 768                 fdt = open(to, flags|O_WRONLY|O_CREAT|O_CLOEXEC|O_NOCTTY, mode);
 769                 if (fdt < 0)
 770                         return -errno;
 771         }
 772
 773         if (chattr_flags != 0)
 774                 (void) chattr_fd(fdt, chattr_flags, (unsigned) -1, NULL);
 775
 776         r = copy_file_fd_full(from, fdt, copy_flags, progress_bytes, userdata);
 777         if (r < 0) {
 778                 close(fdt);
 779                 (void) unlink(to);
 780                 return r;
 781         }
 782
 783         if (close(fdt) < 0) {
 784                 unlink_noerrno(to);
 785                 return -errno;
 786         }
 787
 788         return 0;
 789 }
 790
 791 int copy_file_atomic_full(
 792                 const char *from,
 793                 const char *to,
 794                 mode_t mode,
 795                 unsigned chattr_flags,
 796                 CopyFlags copy_flags,
 797                 copy_progress_bytes_t progress_bytes,
 798                 void *userdata) {
 799
 800         _cleanup_(unlink_and_freep) char *t = NULL;
 801         _cleanup_close_ int fdt = -1;
 802         int r;
 803
 804         assert(from);
 805         assert(to);
 806
 807         /* We try to use O_TMPFILE here to create the file if we can. Note that that only works if COPY_REPLACE is not
 808          * set though as we need to use linkat() for linking the O_TMPFILE file into the file system but that system
 809          * call can't replace existing files. Hence, if COPY_REPLACE is set we create a temporary name in the file
 810          * system right-away and unconditionally which we then can renameat() to the right name after we completed
 811          * writing it. */
 812
 813         if (copy_flags & COPY_REPLACE) {
 814                 r = tempfn_random(to, NULL, &t);
 815                 if (r < 0)
 816                         return r;
 817
 818                 fdt = open(t, O_CREAT|O_EXCL|O_NOFOLLOW|O_NOCTTY|O_WRONLY|O_CLOEXEC, 0600);
 819                 if (fdt < 0) {
 820                         t = mfree(t);
 821                         return -errno;
 822                 }
 823         } else {
 824                 fdt = open_tmpfile_linkable(to, O_WRONLY|O_CLOEXEC, &t);
 825                 if (fdt < 0)
 826                         return fdt;
 827         }
 828
 829         if (chattr_flags != 0)
 830                 (void) chattr_fd(fdt, chattr_flags, (unsigned) -1, NULL);
 831
 832         r = copy_file_fd_full(from, fdt, copy_flags, progress_bytes, userdata);
 833         if (r < 0)
 834                 return r;
 835
 836         if (fchmod(fdt, mode) < 0)
 837                 return -errno;
 838
 839         if (copy_flags & COPY_REPLACE) {
 840                 if (renameat(AT_FDCWD, t, AT_FDCWD, to) < 0)
 841                         return -errno;
 842         } else {
 843                 r = link_tmpfile(fdt, t, to);
 844                 if (r < 0)
 845                         return r;
 846         }
 847
 848         t = mfree(t);
 849         return 0;
 850 }
 851
 852 int copy_times(int fdf, int fdt) {
 853         struct timespec ut[2];
 854         struct stat st;
 855         usec_t crtime = 0;
 856
 857         assert(fdf >= 0);
 858         assert(fdt >= 0);
 859
 860         if (fstat(fdf, &st) < 0)
 861                 return -errno;
 862
 863         ut[0] = st.st_atim;
 864         ut[1] = st.st_mtim;
 865
 866         if (futimens(fdt, ut) < 0)
 867                 return -errno;
 868
 869         if (fd_getcrtime(fdf, &crtime) >= 0)
 870                 (void) fd_setcrtime(fdt, crtime);
 871
 872         return 0;
 873 }
 874
 875 int copy_xattr(int fdf, int fdt) {
 876         _cleanup_free_ char *bufa = NULL, *bufb = NULL;
 877         size_t sza = 100, szb = 100;
 878         ssize_t n;
 879         int ret = 0;
 880         const char *p;
 881
 882         for (;;) {
 883                 bufa = malloc(sza);
 884                 if (!bufa)
 885                         return -ENOMEM;
 886
 887                 n = flistxattr(fdf, bufa, sza);
 888                 if (n == 0)
 889                         return 0;
 890                 if (n > 0)
 891                         break;
 892                 if (errno != ERANGE)
 893                         return -errno;
 894
 895                 sza *= 2;
 896
 897                 bufa = mfree(bufa);
 898         }
 899
 900         p = bufa;
 901         while (n > 0) {
 902                 size_t l;
 903
 904                 l = strlen(p);
 905                 assert(l < (size_t) n);
 906
 907                 if (startswith(p, "user.")) {
 908                         ssize_t m;
 909
 910                         if (!bufb) {
 911                                 bufb = malloc(szb);
 912                                 if (!bufb)
 913                                         return -ENOMEM;
 914                         }
 915
 916                         m = fgetxattr(fdf, p, bufb, szb);
 917                         if (m < 0) {
 918                                 if (errno == ERANGE) {
 919                                         szb *= 2;
 920                                         bufb = mfree(bufb);
 921                                         continue;
 922                                 }
 923
 924                                 return -errno;
 925                         }
 926
 927                         if (fsetxattr(fdt, p, bufb, m, 0) < 0)
 928                                 ret = -errno;
 929                 }
 930
 931                 p += l + 1;
 932                 n -= l + 1;
 933         }
 934
 935         return ret;
 936 }