src/basic/copy.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2 /***
   3   Copyright 2014 Lennart Poettering
   4 ***/
   5
   6 #include <dirent.h>
   7 #include <errno.h>
   8 #include <fcntl.h>
   9 #include <stddef.h>
  10 #include <stdio.h>
  11 #include <stdlib.h>
  12 #include <string.h>
  13 #include <sys/sendfile.h>
  14 #include <sys/stat.h>
  15 #include <sys/xattr.h>
  16 #include <time.h>
  17 #include <unistd.h>
  18
  19 #include "alloc-util.h"
  20 #include "btrfs-util.h"
  21 #include "chattr-util.h"
  22 #include "copy.h"
  23 #include "dirent-util.h"
  24 #include "fd-util.h"
  25 #include "fileio.h"
  26 #include "fs-util.h"
  27 #include "io-util.h"
  28 #include "macro.h"
  29 #include "missing.h"
  30 #include "mount-util.h"
  31 #include "string-util.h"
  32 #include "strv.h"
  33 #include "time-util.h"
  34 #include "umask-util.h"
  35 #include "user-util.h"
  36 #include "xattr-util.h"
  37
  38 #define COPY_BUFFER_SIZE (16U*1024U)
  39
  40 /* A safety net for descending recursively into file system trees to copy. On Linux PATH_MAX is 4096, which means the
  41  * deepest valid path one can build is around 2048, which we hence use as a safety net here, to not spin endlessly in
  42  * case of bind mount cycles and suchlike. */
  43 #define COPY_DEPTH_MAX 2048U
  44
  45 static ssize_t try_copy_file_range(
  46                 int fd_in, loff_t *off_in,
  47                 int fd_out, loff_t *off_out,
  48                 size_t len,
  49                 unsigned int flags) {
  50
  51         static int have = -1;
  52         ssize_t r;
  53
  54         if (have == 0)
  55                 return -ENOSYS;
  56
  57         r = copy_file_range(fd_in, off_in, fd_out, off_out, len, flags);
  58         if (have < 0)
  59                 have = r >= 0 || errno != ENOSYS;
  60         if (r < 0)
  61                 return -errno;
  62
  63         return r;
  64 }
  65
  66 enum {
  67         FD_IS_NO_PIPE,
  68         FD_IS_BLOCKING_PIPE,
  69         FD_IS_NONBLOCKING_PIPE,
  70 };
  71
  72 static int fd_is_nonblock_pipe(int fd) {
  73         struct stat st;
  74         int flags;
  75
  76         /* Checks whether the specified file descriptor refers to a pipe, and if so if O_NONBLOCK is set. */
  77
  78         if (fstat(fd, &st) < 0)
  79                 return -errno;
  80
  81         if (!S_ISFIFO(st.st_mode))
  82                 return FD_IS_NO_PIPE;
  83
  84         flags = fcntl(fd, F_GETFL);
  85         if (flags < 0)
  86                 return -errno;
  87
  88         return FLAGS_SET(flags, O_NONBLOCK) ? FD_IS_NONBLOCKING_PIPE : FD_IS_BLOCKING_PIPE;
  89 }
  90
  91 int copy_bytes_full(
  92                 int fdf, int fdt,
  93                 uint64_t max_bytes,
  94                 CopyFlags copy_flags,
  95                 void **ret_remains,
  96                 size_t *ret_remains_size) {
  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                                 log_debug_errno(r, "Reflinking didn't work, falling back to non-reflink copying: %m");
 169                         }
 170                 }
 171         }
 172
 173         for (;;) {
 174                 ssize_t n;
 175
 176                 if (max_bytes <= 0)
 177                         return 1; /* return > 0 if we hit the max_bytes limit */
 178
 179                 if (max_bytes != UINT64_MAX && m > max_bytes)
 180                         m = max_bytes;
 181
 182                 /* First try copy_file_range(), unless we already tried */
 183                 if (try_cfr) {
 184                         n = try_copy_file_range(fdf, NULL, fdt, NULL, m, 0u);
 185                         if (n < 0) {
 186                                 if (!IN_SET(n, -EINVAL, -ENOSYS, -EXDEV, -EBADF))
 187                                         return n;
 188
 189                                 try_cfr = false;
 190                                 /* use fallback below */
 191                         } else if (n == 0) /* EOF */
 192                                 break;
 193                         else
 194                                 /* Success! */
 195                                 goto next;
 196                 }
 197
 198                 /* First try sendfile(), unless we already tried */
 199                 if (try_sendfile) {
 200                         n = sendfile(fdt, fdf, NULL, m);
 201                         if (n < 0) {
 202                                 if (!IN_SET(errno, EINVAL, ENOSYS))
 203                                         return -errno;
 204
 205                                 try_sendfile = false;
 206                                 /* use fallback below */
 207                         } else if (n == 0) /* EOF */
 208                                 break;
 209                         else
 210                                 /* Success! */
 211                                 goto next;
 212                 }
 213
 214                 /* Then try splice, unless we already tried. */
 215                 if (try_splice) {
 216
 217                         /* splice()'s asynchronous I/O support is a bit weird. When it encounters a pipe file
 218                          * descriptor, then it will ignore its O_NONBLOCK flag and instead only honour the
 219                          * SPLICE_F_NONBLOCK flag specified in its flag parameter. Let's hide this behaviour here, and
 220                          * check if either of the specified fds are a pipe, and if so, let's pass the flag
 221                          * automatically, depending on O_NONBLOCK being set.
 222                          *
 223                          * Here's a twist though: when we use it to move data between two pipes of which one has
 224                          * O_NONBLOCK set and the other has not, then we have no individual control over O_NONBLOCK
 225                          * behaviour. Hence in that case we can't use splice() and still guarantee systematic
 226                          * O_NONBLOCK behaviour, hence don't. */
 227
 228                         if (nonblock_pipe < 0) {
 229                                 int a, b;
 230
 231                                 /* Check if either of these fds is a pipe, and if so non-blocking or not */
 232                                 a = fd_is_nonblock_pipe(fdf);
 233                                 if (a < 0)
 234                                         return a;
 235
 236                                 b = fd_is_nonblock_pipe(fdt);
 237                                 if (b < 0)
 238                                         return b;
 239
 240                                 if ((a == FD_IS_NO_PIPE && b == FD_IS_NO_PIPE) ||
 241                                     (a == FD_IS_BLOCKING_PIPE && b == FD_IS_NONBLOCKING_PIPE) ||
 242                                     (a == FD_IS_NONBLOCKING_PIPE && b == FD_IS_BLOCKING_PIPE))
 243
 244                                         /* splice() only works if one of the fds is a pipe. If neither is, let's skip
 245                                          * this step right-away. As mentioned above, if one of the two fds refers to a
 246                                          * blocking pipe and the other to a non-blocking pipe, we can't use splice()
 247                                          * either, hence don't try either. This hence means we can only use splice() if
 248                                          * either only one of the two fds is a pipe, or if both are pipes with the same
 249                                          * nonblocking flag setting. */
 250
 251                                         try_splice = false;
 252                                 else
 253                                         nonblock_pipe = a == FD_IS_NONBLOCKING_PIPE || b == FD_IS_NONBLOCKING_PIPE;
 254                         }
 255                 }
 256
 257                 if (try_splice) {
 258                         n = splice(fdf, NULL, fdt, NULL, m, nonblock_pipe ? SPLICE_F_NONBLOCK : 0);
 259                         if (n < 0) {
 260                                 if (!IN_SET(errno, EINVAL, ENOSYS))
 261                                         return -errno;
 262
 263                                 try_splice = false;
 264                                 /* use fallback below */
 265                         } else if (n == 0) /* EOF */
 266                                 break;
 267                         else
 268                                 /* Success! */
 269                                 goto next;
 270                 }
 271
 272                 /* As a fallback just copy bits by hand */
 273                 {
 274                         uint8_t buf[MIN(m, COPY_BUFFER_SIZE)], *p = buf;
 275                         ssize_t z;
 276
 277                         n = read(fdf, buf, sizeof buf);
 278                         if (n < 0)
 279                                 return -errno;
 280                         if (n == 0) /* EOF */
 281                                 break;
 282
 283                         z = (size_t) n;
 284                         do {
 285                                 ssize_t k;
 286
 287                                 k = write(fdt, p, z);
 288                                 if (k < 0) {
 289                                         r = -errno;
 290
 291                                         if (ret_remains) {
 292                                                 void *copy;
 293
 294                                                 copy = memdup(p, z);
 295                                                 if (!copy)
 296                                                         return -ENOMEM;
 297
 298                                                 *ret_remains = copy;
 299                                         }
 300
 301                                         if (ret_remains_size)
 302                                                 *ret_remains_size = z;
 303
 304                                         return r;
 305                                 }
 306
 307                                 assert(k <= z);
 308                                 z -= k;
 309                                 p += k;
 310                         } while (z > 0);
 311                 }
 312
 313         next:
 314                 if (max_bytes != (uint64_t) -1) {
 315                         assert(max_bytes >= (uint64_t) n);
 316                         max_bytes -= n;
 317                 }
 318                 /* sendfile accepts at most SSIZE_MAX-offset bytes to copy,
 319                  * so reduce our maximum by the amount we already copied,
 320                  * but don't go below our copy buffer size, unless we are
 321                  * close the limit of bytes we are allowed to copy. */
 322                 m = MAX(MIN(COPY_BUFFER_SIZE, max_bytes), m - n);
 323         }
 324
 325         return 0; /* return 0 if we hit EOF earlier than the size limit */
 326 }
 327
 328 static int fd_copy_symlink(
 329                 int df,
 330                 const char *from,
 331                 const struct stat *st,
 332                 int dt,
 333                 const char *to,
 334                 uid_t override_uid,
 335                 gid_t override_gid,
 336                 CopyFlags copy_flags) {
 337
 338         _cleanup_free_ char *target = NULL;
 339         int r;
 340
 341         assert(from);
 342         assert(st);
 343         assert(to);
 344
 345         r = readlinkat_malloc(df, from, &target);
 346         if (r < 0)
 347                 return r;
 348
 349         if (symlinkat(target, dt, to) < 0)
 350                 return -errno;
 351
 352         if (fchownat(dt, to,
 353                      uid_is_valid(override_uid) ? override_uid : st->st_uid,
 354                      gid_is_valid(override_gid) ? override_gid : st->st_gid,
 355                      AT_SYMLINK_NOFOLLOW) < 0)
 356                 return -errno;
 357
 358         return 0;
 359 }
 360
 361 static int fd_copy_regular(
 362                 int df,
 363                 const char *from,
 364                 const struct stat *st,
 365                 int dt,
 366                 const char *to,
 367                 uid_t override_uid,
 368                 gid_t override_gid,
 369                 CopyFlags copy_flags) {
 370
 371         _cleanup_close_ int fdf = -1, fdt = -1;
 372         struct timespec ts[2];
 373         int r, q;
 374
 375         assert(from);
 376         assert(st);
 377         assert(to);
 378
 379         fdf = openat(df, from, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
 380         if (fdf < 0)
 381                 return -errno;
 382
 383         fdt = openat(dt, to, O_WRONLY|O_CREAT|O_EXCL|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW, st->st_mode & 07777);
 384         if (fdt < 0)
 385                 return -errno;
 386
 387         r = copy_bytes(fdf, fdt, (uint64_t) -1, copy_flags);
 388         if (r < 0) {
 389                 (void) unlinkat(dt, to, 0);
 390                 return r;
 391         }
 392
 393         if (fchown(fdt,
 394                    uid_is_valid(override_uid) ? override_uid : st->st_uid,
 395                    gid_is_valid(override_gid) ? override_gid : st->st_gid) < 0)
 396                 r = -errno;
 397
 398         if (fchmod(fdt, st->st_mode & 07777) < 0)
 399                 r = -errno;
 400
 401         ts[0] = st->st_atim;
 402         ts[1] = st->st_mtim;
 403         (void) futimens(fdt, ts);
 404         (void) copy_xattr(fdf, fdt);
 405
 406         q = close(fdt);
 407         fdt = -1;
 408
 409         if (q < 0) {
 410                 r = -errno;
 411                 (void) unlinkat(dt, to, 0);
 412         }
 413
 414         return r;
 415 }
 416
 417 static int fd_copy_fifo(
 418                 int df,
 419                 const char *from,
 420                 const struct stat *st,
 421                 int dt,
 422                 const char *to,
 423                 uid_t override_uid,
 424                 gid_t override_gid,
 425                 CopyFlags copy_flags) {
 426         int r;
 427
 428         assert(from);
 429         assert(st);
 430         assert(to);
 431
 432         r = mkfifoat(dt, to, st->st_mode & 07777);
 433         if (r < 0)
 434                 return -errno;
 435
 436         if (fchownat(dt, to,
 437                      uid_is_valid(override_uid) ? override_uid : st->st_uid,
 438                      gid_is_valid(override_gid) ? override_gid : st->st_gid,
 439                      AT_SYMLINK_NOFOLLOW) < 0)
 440                 r = -errno;
 441
 442         if (fchmodat(dt, to, st->st_mode & 07777, 0) < 0)
 443                 r = -errno;
 444
 445         return r;
 446 }
 447
 448 static int fd_copy_node(
 449                 int df,
 450                 const char *from,
 451                 const struct stat *st,
 452                 int dt,
 453                 const char *to,
 454                 uid_t override_uid,
 455                 gid_t override_gid,
 456                 CopyFlags copy_flags) {
 457         int r;
 458
 459         assert(from);
 460         assert(st);
 461         assert(to);
 462
 463         r = mknodat(dt, to, st->st_mode, st->st_rdev);
 464         if (r < 0)
 465                 return -errno;
 466
 467         if (fchownat(dt, to,
 468                      uid_is_valid(override_uid) ? override_uid : st->st_uid,
 469                      gid_is_valid(override_gid) ? override_gid : st->st_gid,
 470                      AT_SYMLINK_NOFOLLOW) < 0)
 471                 r = -errno;
 472
 473         if (fchmodat(dt, to, st->st_mode & 07777, 0) < 0)
 474                 r = -errno;
 475
 476         return r;
 477 }
 478
 479 static int fd_copy_directory(
 480                 int df,
 481                 const char *from,
 482                 const struct stat *st,
 483                 int dt,
 484                 const char *to,
 485                 dev_t original_device,
 486                 unsigned depth_left,
 487                 uid_t override_uid,
 488                 gid_t override_gid,
 489                 CopyFlags copy_flags) {
 490
 491         _cleanup_close_ int fdf = -1, fdt = -1;
 492         _cleanup_closedir_ DIR *d = NULL;
 493         struct dirent *de;
 494         bool created;
 495         int r;
 496
 497         assert(st);
 498         assert(to);
 499
 500         if (depth_left == 0)
 501                 return -ENAMETOOLONG;
 502
 503         if (from)
 504                 fdf = openat(df, from, O_RDONLY|O_DIRECTORY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
 505         else
 506                 fdf = fcntl(df, F_DUPFD_CLOEXEC, 3);
 507         if (fdf < 0)
 508                 return -errno;
 509
 510         d = fdopendir(fdf);
 511         if (!d)
 512                 return -errno;
 513         fdf = -1;
 514
 515         r = mkdirat(dt, to, st->st_mode & 07777);
 516         if (r >= 0)
 517                 created = true;
 518         else if (errno == EEXIST && (copy_flags & COPY_MERGE))
 519                 created = false;
 520         else
 521                 return -errno;
 522
 523         fdt = openat(dt, to, O_RDONLY|O_DIRECTORY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
 524         if (fdt < 0)
 525                 return -errno;
 526
 527         r = 0;
 528
 529         FOREACH_DIRENT_ALL(de, d, return -errno) {
 530                 struct stat buf;
 531                 int q;
 532
 533                 if (dot_or_dot_dot(de->d_name))
 534                         continue;
 535
 536                 if (fstatat(dirfd(d), de->d_name, &buf, AT_SYMLINK_NOFOLLOW) < 0) {
 537                         r = -errno;
 538                         continue;
 539                 }
 540
 541                 if (S_ISDIR(buf.st_mode)) {
 542                         /*
 543                          * Don't descend into directories on other file systems, if this is requested. We do a simple
 544                          * .st_dev check here, which basically comes for free. Note that we do this check only on
 545                          * directories, not other kind of file system objects, for two reason:
 546                          *
 547                          * • The kernel's overlayfs pseudo file system that overlays multiple real file systems
 548                          *   propagates the .st_dev field of the file system a file originates from all the way up
 549                          *   through the stack to stat(). It doesn't do that for directories however. This means that
 550                          *   comparing .st_dev on non-directories suggests that they all are mount points. To avoid
 551                          *   confusion we hence avoid relying on this check for regular files.
 552                          *
 553                          * • The main reason we do this check at all is to protect ourselves from bind mount cycles,
 554                          *   where we really want to avoid descending down in all eternity. However the .st_dev check
 555                          *   is usually not sufficient for this protection anyway, as bind mount cycles from the same
 556                          *   file system onto itself can't be detected that way. (Note we also do a recursion depth
 557                          *   check, which is probably the better protection in this regard, which is why
 558                          *   COPY_SAME_MOUNT is optional).
 559                          */
 560
 561                         if (FLAGS_SET(copy_flags, COPY_SAME_MOUNT)) {
 562                                 if (buf.st_dev != original_device)
 563                                         continue;
 564
 565                                 r = fd_is_mount_point(dirfd(d), de->d_name, 0);
 566                                 if (r < 0)
 567                                         return r;
 568                                 if (r > 0)
 569                                         continue;
 570                         }
 571
 572                         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);
 573                 } else if (S_ISREG(buf.st_mode))
 574                         q = fd_copy_regular(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
 575                 else if (S_ISLNK(buf.st_mode))
 576                         q = fd_copy_symlink(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
 577                 else if (S_ISFIFO(buf.st_mode))
 578                         q = fd_copy_fifo(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
 579                 else if (S_ISBLK(buf.st_mode) || S_ISCHR(buf.st_mode) || S_ISSOCK(buf.st_mode))
 580                         q = fd_copy_node(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
 581                 else
 582                         q = -EOPNOTSUPP;
 583
 584                 if (q == -EEXIST && (copy_flags & COPY_MERGE))
 585                         q = 0;
 586
 587                 if (q < 0)
 588                         r = q;
 589         }
 590
 591         if (created) {
 592                 struct timespec ut[2] = {
 593                         st->st_atim,
 594                         st->st_mtim
 595                 };
 596
 597                 if (fchown(fdt,
 598                            uid_is_valid(override_uid) ? override_uid : st->st_uid,
 599                            gid_is_valid(override_gid) ? override_gid : st->st_gid) < 0)
 600                         r = -errno;
 601
 602                 if (fchmod(fdt, st->st_mode & 07777) < 0)
 603                         r = -errno;
 604
 605                 (void) copy_xattr(dirfd(d), fdt);
 606                 (void) futimens(fdt, ut);
 607         }
 608
 609         return r;
 610 }
 611
 612 int copy_tree_at(int fdf, const char *from, int fdt, const char *to, uid_t override_uid, gid_t override_gid, CopyFlags copy_flags) {
 613         struct stat st;
 614
 615         assert(from);
 616         assert(to);
 617
 618         if (fstatat(fdf, from, &st, AT_SYMLINK_NOFOLLOW) < 0)
 619                 return -errno;
 620
 621         if (S_ISREG(st.st_mode))
 622                 return fd_copy_regular(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
 623         else if (S_ISDIR(st.st_mode))
 624                 return fd_copy_directory(fdf, from, &st, fdt, to, st.st_dev, COPY_DEPTH_MAX, override_uid, override_gid, copy_flags);
 625         else if (S_ISLNK(st.st_mode))
 626                 return fd_copy_symlink(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
 627         else if (S_ISFIFO(st.st_mode))
 628                 return fd_copy_fifo(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
 629         else if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode) || S_ISSOCK(st.st_mode))
 630                 return fd_copy_node(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
 631         else
 632                 return -EOPNOTSUPP;
 633 }
 634
 635 int copy_tree(const char *from, const char *to, uid_t override_uid, gid_t override_gid, CopyFlags copy_flags) {
 636         return copy_tree_at(AT_FDCWD, from, AT_FDCWD, to, override_uid, override_gid, copy_flags);
 637 }
 638
 639 int copy_directory_fd(int dirfd, const char *to, CopyFlags copy_flags) {
 640         struct stat st;
 641
 642         assert(dirfd >= 0);
 643         assert(to);
 644
 645         if (fstat(dirfd, &st) < 0)
 646                 return -errno;
 647
 648         if (!S_ISDIR(st.st_mode))
 649                 return -ENOTDIR;
 650
 651         return fd_copy_directory(dirfd, NULL, &st, AT_FDCWD, to, st.st_dev, COPY_DEPTH_MAX, UID_INVALID, GID_INVALID, copy_flags);
 652 }
 653
 654 int copy_directory(const char *from, const char *to, CopyFlags copy_flags) {
 655         struct stat st;
 656
 657         assert(from);
 658         assert(to);
 659
 660         if (lstat(from, &st) < 0)
 661                 return -errno;
 662
 663         if (!S_ISDIR(st.st_mode))
 664                 return -ENOTDIR;
 665
 666         return fd_copy_directory(AT_FDCWD, from, &st, AT_FDCWD, to, st.st_dev, COPY_DEPTH_MAX, UID_INVALID, GID_INVALID, copy_flags);
 667 }
 668
 669 int copy_file_fd(const char *from, int fdt, CopyFlags copy_flags) {
 670         _cleanup_close_ int fdf = -1;
 671         int r;
 672
 673         assert(from);
 674         assert(fdt >= 0);
 675
 676         fdf = open(from, O_RDONLY|O_CLOEXEC|O_NOCTTY);
 677         if (fdf < 0)
 678                 return -errno;
 679
 680         r = copy_bytes(fdf, fdt, (uint64_t) -1, copy_flags);
 681
 682         (void) copy_times(fdf, fdt);
 683         (void) copy_xattr(fdf, fdt);
 684
 685         return r;
 686 }
 687
 688 int copy_file(const char *from, const char *to, int flags, mode_t mode, unsigned chattr_flags, CopyFlags copy_flags) {
 689         int fdt = -1, r;
 690
 691         assert(from);
 692         assert(to);
 693
 694         RUN_WITH_UMASK(0000) {
 695                 fdt = open(to, flags|O_WRONLY|O_CREAT|O_CLOEXEC|O_NOCTTY, mode);
 696                 if (fdt < 0)
 697                         return -errno;
 698         }
 699
 700         if (chattr_flags != 0)
 701                 (void) chattr_fd(fdt, chattr_flags, (unsigned) -1);
 702
 703         r = copy_file_fd(from, fdt, copy_flags);
 704         if (r < 0) {
 705                 close(fdt);
 706                 (void) unlink(to);
 707                 return r;
 708         }
 709
 710         if (close(fdt) < 0) {
 711                 unlink_noerrno(to);
 712                 return -errno;
 713         }
 714
 715         return 0;
 716 }
 717
 718 int copy_file_atomic(const char *from, const char *to, mode_t mode, unsigned chattr_flags, CopyFlags copy_flags) {
 719         _cleanup_free_ char *t = NULL;
 720         int r;
 721
 722         assert(from);
 723         assert(to);
 724
 725         r = tempfn_random(to, NULL, &t);
 726         if (r < 0)
 727                 return r;
 728
 729         r = copy_file(from, t, O_NOFOLLOW|O_EXCL, mode, chattr_flags, copy_flags);
 730         if (r < 0)
 731                 return r;
 732
 733         if (copy_flags & COPY_REPLACE) {
 734                 r = renameat(AT_FDCWD, t, AT_FDCWD, to);
 735                 if (r < 0)
 736                         r = -errno;
 737         } else
 738                 r = rename_noreplace(AT_FDCWD, t, AT_FDCWD, to);
 739         if (r < 0) {
 740                 (void) unlink(t);
 741                 return r;
 742         }
 743
 744         return 0;
 745 }
 746
 747 int copy_times(int fdf, int fdt) {
 748         struct timespec ut[2];
 749         struct stat st;
 750         usec_t crtime = 0;
 751
 752         assert(fdf >= 0);
 753         assert(fdt >= 0);
 754
 755         if (fstat(fdf, &st) < 0)
 756                 return -errno;
 757
 758         ut[0] = st.st_atim;
 759         ut[1] = st.st_mtim;
 760
 761         if (futimens(fdt, ut) < 0)
 762                 return -errno;
 763
 764         if (fd_getcrtime(fdf, &crtime) >= 0)
 765                 (void) fd_setcrtime(fdt, crtime);
 766
 767         return 0;
 768 }
 769
 770 int copy_xattr(int fdf, int fdt) {
 771         _cleanup_free_ char *bufa = NULL, *bufb = NULL;
 772         size_t sza = 100, szb = 100;
 773         ssize_t n;
 774         int ret = 0;
 775         const char *p;
 776
 777         for (;;) {
 778                 bufa = malloc(sza);
 779                 if (!bufa)
 780                         return -ENOMEM;
 781
 782                 n = flistxattr(fdf, bufa, sza);
 783                 if (n == 0)
 784                         return 0;
 785                 if (n > 0)
 786                         break;
 787                 if (errno != ERANGE)
 788                         return -errno;
 789
 790                 sza *= 2;
 791
 792                 bufa = mfree(bufa);
 793         }
 794
 795         p = bufa;
 796         while (n > 0) {
 797                 size_t l;
 798
 799                 l = strlen(p);
 800                 assert(l < (size_t) n);
 801
 802                 if (startswith(p, "user.")) {
 803                         ssize_t m;
 804
 805                         if (!bufb) {
 806                                 bufb = malloc(szb);
 807                                 if (!bufb)
 808                                         return -ENOMEM;
 809                         }
 810
 811                         m = fgetxattr(fdf, p, bufb, szb);
 812                         if (m < 0) {
 813                                 if (errno == ERANGE) {
 814                                         szb *= 2;
 815                                         bufb = mfree(bufb);
 816                                         continue;
 817                                 }
 818
 819                                 return -errno;
 820                         }
 821
 822                         if (fsetxattr(fdt, p, bufb, m, 0) < 0)
 823                                 ret = -errno;
 824                 }
 825
 826                 p += l + 1;
 827                 n -= l + 1;
 828         }
 829
 830         return ret;
 831 }