src/basic/copy.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #include <errno.h>
   4 #include <fcntl.h>
   5 #include <stddef.h>
   6 #include <stdio.h>
   7 #include <stdlib.h>
   8 #include <sys/sendfile.h>
   9 #include <sys/xattr.h>
  10 #include <unistd.h>
  11
  12 #include "alloc-util.h"
  13 #include "btrfs-util.h"
  14 #include "chattr-util.h"
  15 #include "copy.h"
  16 #include "dirent-util.h"
  17 #include "fd-util.h"
  18 #include "fs-util.h"
  19 #include "io-util.h"
  20 #include "macro.h"
  21 #include "missing_syscall.h"
  22 #include "mountpoint-util.h"
  23 #include "stat-util.h"
  24 #include "string-util.h"
  25 #include "strv.h"
  26 #include "time-util.h"
  27 #include "tmpfile-util.h"
  28 #include "umask-util.h"
  29 #include "user-util.h"
  30 #include "xattr-util.h"
  31
  32 #define COPY_BUFFER_SIZE (16U*1024U)
  33
  34 /* A safety net for descending recursively into file system trees to copy. On Linux PATH_MAX is 4096, which means the
  35  * deepest valid path one can build is around 2048, which we hence use as a safety net here, to not spin endlessly in
  36  * case of bind mount cycles and suchlike. */
  37 #define COPY_DEPTH_MAX 2048U
  38
  39 static ssize_t try_copy_file_range(
  40                 int fd_in, loff_t *off_in,
  41                 int fd_out, loff_t *off_out,
  42                 size_t len,
  43                 unsigned flags) {
  44
  45         static int have = -1;
  46         ssize_t r;
  47
  48         if (have == 0)
  49                 return -ENOSYS;
  50
  51         r = copy_file_range(fd_in, off_in, fd_out, off_out, len, flags);
  52         if (have < 0)
  53                 have = r >= 0 || errno != ENOSYS;
  54         if (r < 0)
  55                 return -errno;
  56
  57         return r;
  58 }
  59
  60 enum {
  61         FD_IS_NO_PIPE,
  62         FD_IS_BLOCKING_PIPE,
  63         FD_IS_NONBLOCKING_PIPE,
  64 };
  65
  66 static int fd_is_nonblock_pipe(int fd) {
  67         struct stat st;
  68         int flags;
  69
  70         /* Checks whether the specified file descriptor refers to a pipe, and if so if O_NONBLOCK is set. */
  71
  72         if (fstat(fd, &st) < 0)
  73                 return -errno;
  74
  75         if (!S_ISFIFO(st.st_mode))
  76                 return FD_IS_NO_PIPE;
  77
  78         flags = fcntl(fd, F_GETFL);
  79         if (flags < 0)
  80                 return -errno;
  81
  82         return FLAGS_SET(flags, O_NONBLOCK) ? FD_IS_NONBLOCKING_PIPE : FD_IS_BLOCKING_PIPE;
  83 }
  84
  85 static int sigint_pending(void) {
  86         sigset_t ss;
  87
  88         assert_se(sigemptyset(&ss) >= 0);
  89         assert_se(sigaddset(&ss, SIGINT) >= 0);
  90
  91         if (sigtimedwait(&ss, NULL, &(struct timespec) { 0, 0 }) < 0) {
  92                 if (errno == EAGAIN)
  93                         return false;
  94
  95                 return -errno;
  96         }
  97
  98         return true;
  99 }
 100
 101 int copy_bytes_full(
 102                 int fdf, int fdt,
 103                 uint64_t max_bytes,
 104                 CopyFlags copy_flags,
 105                 void **ret_remains,
 106                 size_t *ret_remains_size,
 107                 copy_progress_bytes_t progress,
 108                 void *userdata) {
 109
 110         bool try_cfr = true, try_sendfile = true, try_splice = true;
 111         int r, nonblock_pipe = -1;
 112         size_t m = SSIZE_MAX; /* that is the maximum that sendfile and c_f_r accept */
 113
 114         assert(fdf >= 0);
 115         assert(fdt >= 0);
 116
 117         /* Tries to copy bytes from the file descriptor 'fdf' to 'fdt' in the smartest possible way. Copies a maximum
 118          * of 'max_bytes', which may be specified as UINT64_MAX, in which no maximum is applied. Returns negative on
 119          * error, zero if EOF is hit before the bytes limit is hit and positive otherwise. If the copy fails for some
 120          * reason but we read but didn't yet write some data an ret_remains/ret_remains_size is not NULL, then it will
 121          * be initialized with an allocated buffer containing this "remaining" data. Note that these two parameters are
 122          * initialized with a valid buffer only on failure and only if there's actually data already read. Otherwise
 123          * these parameters if non-NULL are set to NULL. */
 124
 125         if (ret_remains)
 126                 *ret_remains = NULL;
 127         if (ret_remains_size)
 128                 *ret_remains_size = 0;
 129
 130         /* Try btrfs reflinks first. This only works on regular, seekable files, hence let's check the file offsets of
 131          * source and destination first. */
 132         if ((copy_flags & COPY_REFLINK)) {
 133                 off_t foffset;
 134
 135                 foffset = lseek(fdf, 0, SEEK_CUR);
 136                 if (foffset >= 0) {
 137                         off_t toffset;
 138
 139                         toffset = lseek(fdt, 0, SEEK_CUR);
 140                         if (toffset >= 0) {
 141
 142                                 if (foffset == 0 && toffset == 0 && max_bytes == UINT64_MAX)
 143                                         r = btrfs_reflink(fdf, fdt); /* full file reflink */
 144                                 else
 145                                         r = btrfs_clone_range(fdf, foffset, fdt, toffset, max_bytes == UINT64_MAX ? 0 : max_bytes); /* partial reflink */
 146                                 if (r >= 0) {
 147                                         off_t t;
 148
 149                                         /* This worked, yay! Now — to be fully correct — let's adjust the file pointers */
 150                                         if (max_bytes == UINT64_MAX) {
 151
 152                                                 /* We cloned to the end of the source file, let's position the read
 153                                                  * pointer there, and query it at the same time. */
 154                                                 t = lseek(fdf, 0, SEEK_END);
 155                                                 if (t < 0)
 156                                                         return -errno;
 157                                                 if (t < foffset)
 158                                                         return -ESPIPE;
 159
 160                                                 /* Let's adjust the destination file write pointer by the same number
 161                                                  * of bytes. */
 162                                                 t = lseek(fdt, toffset + (t - foffset), SEEK_SET);
 163                                                 if (t < 0)
 164                                                         return -errno;
 165
 166                                                 return 0; /* we copied the whole thing, hence hit EOF, return 0 */
 167                                         } else {
 168                                                 t = lseek(fdf, foffset + max_bytes, SEEK_SET);
 169                                                 if (t < 0)
 170                                                         return -errno;
 171
 172                                                 t = lseek(fdt, toffset + max_bytes, SEEK_SET);
 173                                                 if (t < 0)
 174                                                         return -errno;
 175
 176                                                 return 1; /* we copied only some number of bytes, which worked, but this means we didn't hit EOF, return 1 */
 177                                         }
 178                                 }
 179                         }
 180                 }
 181         }
 182
 183         for (;;) {
 184                 ssize_t n;
 185
 186                 if (max_bytes <= 0)
 187                         return 1; /* return > 0 if we hit the max_bytes limit */
 188
 189                 if (FLAGS_SET(copy_flags, COPY_SIGINT)) {
 190                         r = sigint_pending();
 191                         if (r < 0)
 192                                 return r;
 193                         if (r > 0)
 194                                 return -EINTR;
 195                 }
 196
 197                 if (max_bytes != UINT64_MAX && m > max_bytes)
 198                         m = max_bytes;
 199
 200                 /* First try copy_file_range(), unless we already tried */
 201                 if (try_cfr) {
 202                         n = try_copy_file_range(fdf, NULL, fdt, NULL, m, 0u);
 203                         if (n < 0) {
 204                                 if (!IN_SET(n, -EINVAL, -ENOSYS, -EXDEV, -EBADF))
 205                                         return n;
 206
 207                                 try_cfr = false;
 208                                 /* use fallback below */
 209                         } else if (n == 0) /* EOF */
 210                                 break;
 211                         else
 212                                 /* Success! */
 213                                 goto next;
 214                 }
 215
 216                 /* First try sendfile(), unless we already tried */
 217                 if (try_sendfile) {
 218                         n = sendfile(fdt, fdf, NULL, m);
 219                         if (n < 0) {
 220                                 if (!IN_SET(errno, EINVAL, ENOSYS))
 221                                         return -errno;
 222
 223                                 try_sendfile = false;
 224                                 /* use fallback below */
 225                         } else if (n == 0) /* EOF */
 226                                 break;
 227                         else
 228                                 /* Success! */
 229                                 goto next;
 230                 }
 231
 232                 /* Then try splice, unless we already tried. */
 233                 if (try_splice) {
 234
 235                         /* splice()'s asynchronous I/O support is a bit weird. When it encounters a pipe file
 236                          * descriptor, then it will ignore its O_NONBLOCK flag and instead only honour the
 237                          * SPLICE_F_NONBLOCK flag specified in its flag parameter. Let's hide this behaviour here, and
 238                          * check if either of the specified fds are a pipe, and if so, let's pass the flag
 239                          * automatically, depending on O_NONBLOCK being set.
 240                          *
 241                          * Here's a twist though: when we use it to move data between two pipes of which one has
 242                          * O_NONBLOCK set and the other has not, then we have no individual control over O_NONBLOCK
 243                          * behaviour. Hence in that case we can't use splice() and still guarantee systematic
 244                          * O_NONBLOCK behaviour, hence don't. */
 245
 246                         if (nonblock_pipe < 0) {
 247                                 int a, b;
 248
 249                                 /* Check if either of these fds is a pipe, and if so non-blocking or not */
 250                                 a = fd_is_nonblock_pipe(fdf);
 251                                 if (a < 0)
 252                                         return a;
 253
 254                                 b = fd_is_nonblock_pipe(fdt);
 255                                 if (b < 0)
 256                                         return b;
 257
 258                                 if ((a == FD_IS_NO_PIPE && b == FD_IS_NO_PIPE) ||
 259                                     (a == FD_IS_BLOCKING_PIPE && b == FD_IS_NONBLOCKING_PIPE) ||
 260                                     (a == FD_IS_NONBLOCKING_PIPE && b == FD_IS_BLOCKING_PIPE))
 261
 262                                         /* splice() only works if one of the fds is a pipe. If neither is, let's skip
 263                                          * this step right-away. As mentioned above, if one of the two fds refers to a
 264                                          * blocking pipe and the other to a non-blocking pipe, we can't use splice()
 265                                          * either, hence don't try either. This hence means we can only use splice() if
 266                                          * either only one of the two fds is a pipe, or if both are pipes with the same
 267                                          * nonblocking flag setting. */
 268
 269                                         try_splice = false;
 270                                 else
 271                                         nonblock_pipe = a == FD_IS_NONBLOCKING_PIPE || b == FD_IS_NONBLOCKING_PIPE;
 272                         }
 273                 }
 274
 275                 if (try_splice) {
 276                         n = splice(fdf, NULL, fdt, NULL, m, nonblock_pipe ? SPLICE_F_NONBLOCK : 0);
 277                         if (n < 0) {
 278                                 if (!IN_SET(errno, EINVAL, ENOSYS))
 279                                         return -errno;
 280
 281                                 try_splice = false;
 282                                 /* use fallback below */
 283                         } else if (n == 0) /* EOF */
 284                                 break;
 285                         else
 286                                 /* Success! */
 287                                 goto next;
 288                 }
 289
 290                 /* As a fallback just copy bits by hand */
 291                 {
 292                         uint8_t buf[MIN(m, COPY_BUFFER_SIZE)], *p = buf;
 293                         ssize_t z;
 294
 295                         n = read(fdf, buf, sizeof buf);
 296                         if (n < 0)
 297                                 return -errno;
 298                         if (n == 0) /* EOF */
 299                                 break;
 300
 301                         z = (size_t) n;
 302                         do {
 303                                 ssize_t k;
 304
 305                                 k = write(fdt, p, z);
 306                                 if (k < 0) {
 307                                         r = -errno;
 308
 309                                         if (ret_remains) {
 310                                                 void *copy;
 311
 312                                                 copy = memdup(p, z);
 313                                                 if (!copy)
 314                                                         return -ENOMEM;
 315
 316                                                 *ret_remains = copy;
 317                                         }
 318
 319                                         if (ret_remains_size)
 320                                                 *ret_remains_size = z;
 321
 322                                         return r;
 323                                 }
 324
 325                                 assert(k <= z);
 326                                 z -= k;
 327                                 p += k;
 328                         } while (z > 0);
 329                 }
 330
 331         next:
 332                 if (progress) {
 333                         r = progress(n, userdata);
 334                         if (r < 0)
 335                                 return r;
 336                 }
 337
 338                 if (max_bytes != (uint64_t) -1) {
 339                         assert(max_bytes >= (uint64_t) n);
 340                         max_bytes -= n;
 341                 }
 342
 343                 /* sendfile accepts at most SSIZE_MAX-offset bytes to copy,
 344                  * so reduce our maximum by the amount we already copied,
 345                  * but don't go below our copy buffer size, unless we are
 346                  * close the limit of bytes we are allowed to copy. */
 347                 m = MAX(MIN(COPY_BUFFER_SIZE, max_bytes), m - n);
 348         }
 349
 350         return 0; /* return 0 if we hit EOF earlier than the size limit */
 351 }
 352
 353 static int fd_copy_symlink(
 354                 int df,
 355                 const char *from,
 356                 const struct stat *st,
 357                 int dt,
 358                 const char *to,
 359                 uid_t override_uid,
 360                 gid_t override_gid,
 361                 CopyFlags copy_flags) {
 362
 363         _cleanup_free_ char *target = NULL;
 364         int r;
 365
 366         assert(from);
 367         assert(st);
 368         assert(to);
 369
 370         r = readlinkat_malloc(df, from, &target);
 371         if (r < 0)
 372                 return r;
 373
 374         if (symlinkat(target, dt, to) < 0)
 375                 return -errno;
 376
 377         if (fchownat(dt, to,
 378                      uid_is_valid(override_uid) ? override_uid : st->st_uid,
 379                      gid_is_valid(override_gid) ? override_gid : st->st_gid,
 380                      AT_SYMLINK_NOFOLLOW) < 0)
 381                 return -errno;
 382
 383         return 0;
 384 }
 385
 386 static int fd_copy_regular(
 387                 int df,
 388                 const char *from,
 389                 const struct stat *st,
 390                 int dt,
 391                 const char *to,
 392                 uid_t override_uid,
 393                 gid_t override_gid,
 394                 CopyFlags copy_flags,
 395                 copy_progress_bytes_t progress,
 396                 void *userdata) {
 397
 398         _cleanup_close_ int fdf = -1, fdt = -1;
 399         struct timespec ts[2];
 400         int r, q;
 401
 402         assert(from);
 403         assert(st);
 404         assert(to);
 405
 406         fdf = openat(df, from, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
 407         if (fdf < 0)
 408                 return -errno;
 409
 410         fdt = openat(dt, to, O_WRONLY|O_CREAT|O_EXCL|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW, st->st_mode & 07777);
 411         if (fdt < 0)
 412                 return -errno;
 413
 414         r = copy_bytes_full(fdf, fdt, (uint64_t) -1, copy_flags, NULL, NULL, progress, userdata);
 415         if (r < 0) {
 416                 (void) unlinkat(dt, to, 0);
 417                 return r;
 418         }
 419
 420         if (fchown(fdt,
 421                    uid_is_valid(override_uid) ? override_uid : st->st_uid,
 422                    gid_is_valid(override_gid) ? override_gid : st->st_gid) < 0)
 423                 r = -errno;
 424
 425         if (fchmod(fdt, st->st_mode & 07777) < 0)
 426                 r = -errno;
 427
 428         ts[0] = st->st_atim;
 429         ts[1] = st->st_mtim;
 430         (void) futimens(fdt, ts);
 431         (void) copy_xattr(fdf, fdt);
 432
 433         q = close(fdt);
 434         fdt = -1;
 435
 436         if (q < 0) {
 437                 r = -errno;
 438                 (void) unlinkat(dt, to, 0);
 439         }
 440
 441         return r;
 442 }
 443
 444 static int fd_copy_fifo(
 445                 int df,
 446                 const char *from,
 447                 const struct stat *st,
 448                 int dt,
 449                 const char *to,
 450                 uid_t override_uid,
 451                 gid_t override_gid,
 452                 CopyFlags copy_flags) {
 453         int r;
 454
 455         assert(from);
 456         assert(st);
 457         assert(to);
 458
 459         r = mkfifoat(dt, to, st->st_mode & 07777);
 460         if (r < 0)
 461                 return -errno;
 462
 463         if (fchownat(dt, to,
 464                      uid_is_valid(override_uid) ? override_uid : st->st_uid,
 465                      gid_is_valid(override_gid) ? override_gid : st->st_gid,
 466                      AT_SYMLINK_NOFOLLOW) < 0)
 467                 r = -errno;
 468
 469         if (fchmodat(dt, to, st->st_mode & 07777, 0) < 0)
 470                 r = -errno;
 471
 472         return r;
 473 }
 474
 475 static int fd_copy_node(
 476                 int df,
 477                 const char *from,
 478                 const struct stat *st,
 479                 int dt,
 480                 const char *to,
 481                 uid_t override_uid,
 482                 gid_t override_gid,
 483                 CopyFlags copy_flags) {
 484         int r;
 485
 486         assert(from);
 487         assert(st);
 488         assert(to);
 489
 490         r = mknodat(dt, to, st->st_mode, st->st_rdev);
 491         if (r < 0)
 492                 return -errno;
 493
 494         if (fchownat(dt, to,
 495                      uid_is_valid(override_uid) ? override_uid : st->st_uid,
 496                      gid_is_valid(override_gid) ? override_gid : st->st_gid,
 497                      AT_SYMLINK_NOFOLLOW) < 0)
 498                 r = -errno;
 499
 500         if (fchmodat(dt, to, st->st_mode & 07777, 0) < 0)
 501                 r = -errno;
 502
 503         return r;
 504 }
 505
 506 static int fd_copy_directory(
 507                 int df,
 508                 const char *from,
 509                 const struct stat *st,
 510                 int dt,
 511                 const char *to,
 512                 dev_t original_device,
 513                 unsigned depth_left,
 514                 uid_t override_uid,
 515                 gid_t override_gid,
 516                 CopyFlags copy_flags,
 517                 const char *display_path,
 518                 copy_progress_path_t progress_path,
 519                 copy_progress_bytes_t progress_bytes,
 520                 void *userdata) {
 521
 522         _cleanup_close_ int fdf = -1, fdt = -1;
 523         _cleanup_closedir_ DIR *d = NULL;
 524         struct dirent *de;
 525         bool exists, created;
 526         int r;
 527
 528         assert(st);
 529         assert(to);
 530
 531         if (depth_left == 0)
 532                 return -ENAMETOOLONG;
 533
 534         if (from)
 535                 fdf = openat(df, from, O_RDONLY|O_DIRECTORY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
 536         else
 537                 fdf = fcntl(df, F_DUPFD_CLOEXEC, 3);
 538         if (fdf < 0)
 539                 return -errno;
 540
 541         d = fdopendir(fdf);
 542         if (!d)
 543                 return -errno;
 544         fdf = -1;
 545
 546         exists = false;
 547         if (copy_flags & COPY_MERGE_EMPTY) {
 548                 r = dir_is_empty_at(dt, to);
 549                 if (r < 0 && r != -ENOENT)
 550                         return r;
 551                 else if (r == 1)
 552                         exists = true;
 553         }
 554
 555         if (exists)
 556                 created = false;
 557         else {
 558                 r = mkdirat(dt, to, st->st_mode & 07777);
 559                 if (r >= 0)
 560                         created = true;
 561                 else if (errno == EEXIST && (copy_flags & COPY_MERGE))
 562                         created = false;
 563                 else
 564                         return -errno;
 565         }
 566
 567         fdt = openat(dt, to, O_RDONLY|O_DIRECTORY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
 568         if (fdt < 0)
 569                 return -errno;
 570
 571         r = 0;
 572
 573         FOREACH_DIRENT_ALL(de, d, return -errno) {
 574                 const char *child_display_path = NULL;
 575                 _cleanup_free_ char *dp = NULL;
 576                 struct stat buf;
 577                 int q;
 578
 579                 if (dot_or_dot_dot(de->d_name))
 580                         continue;
 581
 582                 if (FLAGS_SET(copy_flags, COPY_SIGINT)) {
 583                         r = sigint_pending();
 584                         if (r < 0)
 585                                 return r;
 586                         if (r > 0)
 587                                 return -EINTR;
 588                 }
 589
 590                 if (fstatat(dirfd(d), de->d_name, &buf, AT_SYMLINK_NOFOLLOW) < 0) {
 591                         r = -errno;
 592                         continue;
 593                 }
 594
 595                 if (progress_path) {
 596                         if (display_path)
 597                                 child_display_path = dp = path_join(display_path, de->d_name);
 598                         else
 599                                 child_display_path = de->d_name;
 600
 601                         r = progress_path(child_display_path, &buf, userdata);
 602                         if (r < 0)
 603                                 return r;
 604                 }
 605
 606                 if (S_ISDIR(buf.st_mode)) {
 607                         /*
 608                          * Don't descend into directories on other file systems, if this is requested. We do a simple
 609                          * .st_dev check here, which basically comes for free. Note that we do this check only on
 610                          * directories, not other kind of file system objects, for two reason:
 611                          *
 612                          * • The kernel's overlayfs pseudo file system that overlays multiple real file systems
 613                          *   propagates the .st_dev field of the file system a file originates from all the way up
 614                          *   through the stack to stat(). It doesn't do that for directories however. This means that
 615                          *   comparing .st_dev on non-directories suggests that they all are mount points. To avoid
 616                          *   confusion we hence avoid relying on this check for regular files.
 617                          *
 618                          * • The main reason we do this check at all is to protect ourselves from bind mount cycles,
 619                          *   where we really want to avoid descending down in all eternity. However the .st_dev check
 620                          *   is usually not sufficient for this protection anyway, as bind mount cycles from the same
 621                          *   file system onto itself can't be detected that way. (Note we also do a recursion depth
 622                          *   check, which is probably the better protection in this regard, which is why
 623                          *   COPY_SAME_MOUNT is optional).
 624                          */
 625
 626                         if (FLAGS_SET(copy_flags, COPY_SAME_MOUNT)) {
 627                                 if (buf.st_dev != original_device)
 628                                         continue;
 629
 630                                 r = fd_is_mount_point(dirfd(d), de->d_name, 0);
 631                                 if (r < 0)
 632                                         return r;
 633                                 if (r > 0)
 634                                         continue;
 635                         }
 636
 637                         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);
 638                 } else if (S_ISREG(buf.st_mode))
 639                         q = fd_copy_regular(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags, progress_bytes, userdata);
 640                 else if (S_ISLNK(buf.st_mode))
 641                         q = fd_copy_symlink(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
 642                 else if (S_ISFIFO(buf.st_mode))
 643                         q = fd_copy_fifo(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
 644                 else if (S_ISBLK(buf.st_mode) || S_ISCHR(buf.st_mode) || S_ISSOCK(buf.st_mode))
 645                         q = fd_copy_node(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
 646                 else
 647                         q = -EOPNOTSUPP;
 648
 649                 if (q == -EINTR) /* Propagate SIGINT up instantly */
 650                         return q;
 651                 if (q == -EEXIST && (copy_flags & COPY_MERGE))
 652                         q = 0;
 653                 if (q < 0)
 654                         r = q;
 655         }
 656
 657         if (created) {
 658                 struct timespec ut[2] = {
 659                         st->st_atim,
 660                         st->st_mtim
 661                 };
 662
 663                 if (fchown(fdt,
 664                            uid_is_valid(override_uid) ? override_uid : st->st_uid,
 665                            gid_is_valid(override_gid) ? override_gid : st->st_gid) < 0)
 666                         r = -errno;
 667
 668                 if (fchmod(fdt, st->st_mode & 07777) < 0)
 669                         r = -errno;
 670
 671                 (void) copy_xattr(dirfd(d), fdt);
 672                 (void) futimens(fdt, ut);
 673         }
 674
 675         return r;
 676 }
 677
 678 int copy_tree_at_full(
 679                 int fdf,
 680                 const char *from,
 681                 int fdt,
 682                 const char *to,
 683                 uid_t override_uid,
 684                 gid_t override_gid,
 685                 CopyFlags copy_flags,
 686                 copy_progress_path_t progress_path,
 687                 copy_progress_bytes_t progress_bytes,
 688                 void *userdata) {
 689
 690         struct stat st;
 691
 692         assert(from);
 693         assert(to);
 694
 695         if (fstatat(fdf, from, &st, AT_SYMLINK_NOFOLLOW) < 0)
 696                 return -errno;
 697
 698         if (S_ISREG(st.st_mode))
 699                 return fd_copy_regular(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags, progress_bytes, userdata);
 700         else if (S_ISDIR(st.st_mode))
 701                 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);
 702         else if (S_ISLNK(st.st_mode))
 703                 return fd_copy_symlink(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
 704         else if (S_ISFIFO(st.st_mode))
 705                 return fd_copy_fifo(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
 706         else if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode) || S_ISSOCK(st.st_mode))
 707                 return fd_copy_node(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
 708         else
 709                 return -EOPNOTSUPP;
 710 }
 711
 712 int copy_directory_fd_full(
 713                 int dirfd,
 714                 const char *to,
 715                 CopyFlags copy_flags,
 716                 copy_progress_path_t progress_path,
 717                 copy_progress_bytes_t progress_bytes,
 718                 void *userdata) {
 719
 720         struct stat st;
 721
 722         assert(dirfd >= 0);
 723         assert(to);
 724
 725         if (fstat(dirfd, &st) < 0)
 726                 return -errno;
 727
 728         if (!S_ISDIR(st.st_mode))
 729                 return -ENOTDIR;
 730
 731         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);
 732 }
 733
 734 int copy_directory_full(
 735                 const char *from,
 736                 const char *to,
 737                 CopyFlags copy_flags,
 738                 copy_progress_path_t progress_path,
 739                 copy_progress_bytes_t progress_bytes,
 740                 void *userdata) {
 741
 742         struct stat st;
 743
 744         assert(from);
 745         assert(to);
 746
 747         if (lstat(from, &st) < 0)
 748                 return -errno;
 749
 750         if (!S_ISDIR(st.st_mode))
 751                 return -ENOTDIR;
 752
 753         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);
 754 }
 755
 756 int copy_file_fd_full(
 757                 const char *from,
 758                 int fdt,
 759                 CopyFlags copy_flags,
 760                 copy_progress_bytes_t progress_bytes,
 761                 void *userdata) {
 762
 763         _cleanup_close_ int fdf = -1;
 764         int r;
 765
 766         assert(from);
 767         assert(fdt >= 0);
 768
 769         fdf = open(from, O_RDONLY|O_CLOEXEC|O_NOCTTY);
 770         if (fdf < 0)
 771                 return -errno;
 772
 773         r = copy_bytes_full(fdf, fdt, (uint64_t) -1, copy_flags, NULL, NULL, progress_bytes, userdata);
 774
 775         (void) copy_times(fdf, fdt, copy_flags);
 776         (void) copy_xattr(fdf, fdt);
 777
 778         return r;
 779 }
 780
 781 int copy_file_full(
 782                 const char *from,
 783                 const char *to,
 784                 int flags,
 785                 mode_t mode,
 786                 unsigned chattr_flags,
 787                 unsigned chattr_mask,
 788                 CopyFlags copy_flags,
 789                 copy_progress_bytes_t progress_bytes,
 790                 void *userdata) {
 791
 792         int fdt = -1, r;
 793
 794         assert(from);
 795         assert(to);
 796
 797         RUN_WITH_UMASK(0000) {
 798                 fdt = open(to, flags|O_WRONLY|O_CREAT|O_CLOEXEC|O_NOCTTY, mode);
 799                 if (fdt < 0)
 800                         return -errno;
 801         }
 802
 803         if (chattr_mask != 0)
 804                 (void) chattr_fd(fdt, chattr_flags, chattr_mask & CHATTR_EARLY_FL, NULL);
 805
 806         r = copy_file_fd_full(from, fdt, copy_flags, progress_bytes, userdata);
 807         if (r < 0) {
 808                 close(fdt);
 809                 (void) unlink(to);
 810                 return r;
 811         }
 812
 813         if (chattr_mask != 0)
 814                 (void) chattr_fd(fdt, chattr_flags, chattr_mask & ~CHATTR_EARLY_FL, NULL);
 815
 816         if (close(fdt) < 0) {
 817                 unlink_noerrno(to);
 818                 return -errno;
 819         }
 820
 821         return 0;
 822 }
 823
 824 int copy_file_atomic_full(
 825                 const char *from,
 826                 const char *to,
 827                 mode_t mode,
 828                 unsigned chattr_flags,
 829                 unsigned chattr_mask,
 830                 CopyFlags copy_flags,
 831                 copy_progress_bytes_t progress_bytes,
 832                 void *userdata) {
 833
 834         _cleanup_(unlink_and_freep) char *t = NULL;
 835         _cleanup_close_ int fdt = -1;
 836         int r;
 837
 838         assert(from);
 839         assert(to);
 840
 841         /* We try to use O_TMPFILE here to create the file if we can. Note that that only works if COPY_REPLACE is not
 842          * set though as we need to use linkat() for linking the O_TMPFILE file into the file system but that system
 843          * call can't replace existing files. Hence, if COPY_REPLACE is set we create a temporary name in the file
 844          * system right-away and unconditionally which we then can renameat() to the right name after we completed
 845          * writing it. */
 846
 847         if (copy_flags & COPY_REPLACE) {
 848                 r = tempfn_random(to, NULL, &t);
 849                 if (r < 0)
 850                         return r;
 851
 852                 fdt = open(t, O_CREAT|O_EXCL|O_NOFOLLOW|O_NOCTTY|O_WRONLY|O_CLOEXEC, 0600);
 853                 if (fdt < 0) {
 854                         t = mfree(t);
 855                         return -errno;
 856                 }
 857         } else {
 858                 fdt = open_tmpfile_linkable(to, O_WRONLY|O_CLOEXEC, &t);
 859                 if (fdt < 0)
 860                         return fdt;
 861         }
 862
 863         if (chattr_mask != 0)
 864                 (void) chattr_fd(fdt, chattr_flags, chattr_mask & CHATTR_EARLY_FL, NULL);
 865
 866         r = copy_file_fd_full(from, fdt, copy_flags, progress_bytes, userdata);
 867         if (r < 0)
 868                 return r;
 869
 870         if (fchmod(fdt, mode) < 0)
 871                 return -errno;
 872
 873         if (copy_flags & COPY_REPLACE) {
 874                 if (renameat(AT_FDCWD, t, AT_FDCWD, to) < 0)
 875                         return -errno;
 876         } else {
 877                 r = link_tmpfile(fdt, t, to);
 878                 if (r < 0)
 879                         return r;
 880         }
 881
 882         if (chattr_mask != 0)
 883                 (void) chattr_fd(fdt, chattr_flags, chattr_mask & ~CHATTR_EARLY_FL, NULL);
 884
 885         t = mfree(t);
 886         return 0;
 887 }
 888
 889 int copy_times(int fdf, int fdt, CopyFlags flags) {
 890         struct timespec ut[2];
 891         struct stat st;
 892
 893         assert(fdf >= 0);
 894         assert(fdt >= 0);
 895
 896         if (fstat(fdf, &st) < 0)
 897                 return -errno;
 898
 899         ut[0] = st.st_atim;
 900         ut[1] = st.st_mtim;
 901
 902         if (futimens(fdt, ut) < 0)
 903                 return -errno;
 904
 905         if (FLAGS_SET(flags, COPY_CRTIME)) {
 906                 usec_t crtime;
 907
 908                 if (fd_getcrtime(fdf, &crtime) >= 0)
 909                         (void) fd_setcrtime(fdt, crtime);
 910         }
 911
 912         return 0;
 913 }
 914
 915 int copy_xattr(int fdf, int fdt) {
 916         _cleanup_free_ char *bufa = NULL, *bufb = NULL;
 917         size_t sza = 100, szb = 100;
 918         ssize_t n;
 919         int ret = 0;
 920         const char *p;
 921
 922         for (;;) {
 923                 bufa = malloc(sza);
 924                 if (!bufa)
 925                         return -ENOMEM;
 926
 927                 n = flistxattr(fdf, bufa, sza);
 928                 if (n == 0)
 929                         return 0;
 930                 if (n > 0)
 931                         break;
 932                 if (errno != ERANGE)
 933                         return -errno;
 934
 935                 sza *= 2;
 936
 937                 bufa = mfree(bufa);
 938         }
 939
 940         p = bufa;
 941         while (n > 0) {
 942                 size_t l;
 943
 944                 l = strlen(p);
 945                 assert(l < (size_t) n);
 946
 947                 if (startswith(p, "user.")) {
 948                         ssize_t m;
 949
 950                         if (!bufb) {
 951                                 bufb = malloc(szb);
 952                                 if (!bufb)
 953                                         return -ENOMEM;
 954                         }
 955
 956                         m = fgetxattr(fdf, p, bufb, szb);
 957                         if (m < 0) {
 958                                 if (errno == ERANGE) {
 959                                         szb *= 2;
 960                                         bufb = mfree(bufb);
 961                                         continue;
 962                                 }
 963
 964                                 return -errno;
 965                         }
 966
 967                         if (fsetxattr(fdt, p, bufb, m, 0) < 0)
 968                                 ret = -errno;
 969                 }
 970
 971                 p += l + 1;
 972                 n -= l + 1;
 973         }
 974
 975         return ret;
 976 }