src/basic/copy.c

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   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 "fileio.h"
  19 #include "fs-util.h"
  20 #include "io-util.h"
  21 #include "macro.h"
  22 #include "missing_syscall.h"
  23 #include "mountpoint-util.h"
  24 #include "nulstr-util.h"
  25 #include "rm-rf.h"
  26 #include "selinux-util.h"
  27 #include "stat-util.h"
  28 #include "stdio-util.h"
  29 #include "string-util.h"
  30 #include "strv.h"
  31 #include "time-util.h"
  32 #include "tmpfile-util.h"
  33 #include "umask-util.h"
  34 #include "user-util.h"
  35 #include "xattr-util.h"
  36
  37 #define COPY_BUFFER_SIZE (16U*1024U)
  38
  39 /* A safety net for descending recursively into file system trees to copy. On Linux PATH_MAX is 4096, which means the
  40  * deepest valid path one can build is around 2048, which we hence use as a safety net here, to not spin endlessly in
  41  * case of bind mount cycles and suchlike. */
  42 #define COPY_DEPTH_MAX 2048U
  43
  44 static ssize_t try_copy_file_range(
  45                 int fd_in, loff_t *off_in,
  46                 int fd_out, loff_t *off_out,
  47                 size_t len,
  48                 unsigned flags) {
  49
  50         static int have = -1;
  51         ssize_t r;
  52
  53         if (have == 0)
  54                 return -ENOSYS;
  55
  56         r = copy_file_range(fd_in, off_in, fd_out, off_out, len, flags);
  57         if (have < 0)
  58                 have = r >= 0 || errno != ENOSYS;
  59         if (r < 0)
  60                 return -errno;
  61
  62         return r;
  63 }
  64
  65 enum {
  66         FD_IS_NO_PIPE,
  67         FD_IS_BLOCKING_PIPE,
  68         FD_IS_NONBLOCKING_PIPE,
  69 };
  70
  71 static int fd_is_nonblock_pipe(int fd) {
  72         struct stat st;
  73         int flags;
  74
  75         /* Checks whether the specified file descriptor refers to a pipe, and if so if O_NONBLOCK is set. */
  76
  77         if (fstat(fd, &st) < 0)
  78                 return -errno;
  79
  80         if (!S_ISFIFO(st.st_mode))
  81                 return FD_IS_NO_PIPE;
  82
  83         flags = fcntl(fd, F_GETFL);
  84         if (flags < 0)
  85                 return -errno;
  86
  87         return FLAGS_SET(flags, O_NONBLOCK) ? FD_IS_NONBLOCKING_PIPE : FD_IS_BLOCKING_PIPE;
  88 }
  89
  90 static int sigint_pending(void) {
  91         sigset_t ss;
  92
  93         assert_se(sigemptyset(&ss) >= 0);
  94         assert_se(sigaddset(&ss, SIGINT) >= 0);
  95
  96         if (sigtimedwait(&ss, NULL, &(struct timespec) { 0, 0 }) < 0) {
  97                 if (errno == EAGAIN)
  98                         return false;
  99
 100                 return -errno;
 101         }
 102
 103         return true;
 104 }
 105
 106 int copy_bytes_full(
 107                 int fdf, int fdt,
 108                 uint64_t max_bytes,
 109                 CopyFlags copy_flags,
 110                 void **ret_remains,
 111                 size_t *ret_remains_size,
 112                 copy_progress_bytes_t progress,
 113                 void *userdata) {
 114
 115         bool try_cfr = true, try_sendfile = true, try_splice = true;
 116         int r, nonblock_pipe = -1;
 117         size_t m = SSIZE_MAX; /* that is the maximum that sendfile and c_f_r accept */
 118
 119         assert(fdf >= 0);
 120         assert(fdt >= 0);
 121
 122         /* Tries to copy bytes from the file descriptor 'fdf' to 'fdt' in the smartest possible way. Copies a maximum
 123          * of 'max_bytes', which may be specified as UINT64_MAX, in which no maximum is applied. Returns negative on
 124          * error, zero if EOF is hit before the bytes limit is hit and positive otherwise. If the copy fails for some
 125          * reason but we read but didn't yet write some data an ret_remains/ret_remains_size is not NULL, then it will
 126          * be initialized with an allocated buffer containing this "remaining" data. Note that these two parameters are
 127          * initialized with a valid buffer only on failure and only if there's actually data already read. Otherwise
 128          * these parameters if non-NULL are set to NULL. */
 129
 130         if (ret_remains)
 131                 *ret_remains = NULL;
 132         if (ret_remains_size)
 133                 *ret_remains_size = 0;
 134
 135         /* Try btrfs reflinks first. This only works on regular, seekable files, hence let's check the file offsets of
 136          * source and destination first. */
 137         if ((copy_flags & COPY_REFLINK)) {
 138                 off_t foffset;
 139
 140                 foffset = lseek(fdf, 0, SEEK_CUR);
 141                 if (foffset >= 0) {
 142                         off_t toffset;
 143
 144                         toffset = lseek(fdt, 0, SEEK_CUR);
 145                         if (toffset >= 0) {
 146
 147                                 if (foffset == 0 && toffset == 0 && max_bytes == UINT64_MAX)
 148                                         r = btrfs_reflink(fdf, fdt); /* full file reflink */
 149                                 else
 150                                         r = btrfs_clone_range(fdf, foffset, fdt, toffset, max_bytes == UINT64_MAX ? 0 : max_bytes); /* partial reflink */
 151                                 if (r >= 0) {
 152                                         off_t t;
 153
 154                                         /* This worked, yay! Now — to be fully correct — let's adjust the file pointers */
 155                                         if (max_bytes == UINT64_MAX) {
 156
 157                                                 /* We cloned to the end of the source file, let's position the read
 158                                                  * pointer there, and query it at the same time. */
 159                                                 t = lseek(fdf, 0, SEEK_END);
 160                                                 if (t < 0)
 161                                                         return -errno;
 162                                                 if (t < foffset)
 163                                                         return -ESPIPE;
 164
 165                                                 /* Let's adjust the destination file write pointer by the same number
 166                                                  * of bytes. */
 167                                                 t = lseek(fdt, toffset + (t - foffset), SEEK_SET);
 168                                                 if (t < 0)
 169                                                         return -errno;
 170
 171                                                 return 0; /* we copied the whole thing, hence hit EOF, return 0 */
 172                                         } else {
 173                                                 t = lseek(fdf, foffset + max_bytes, SEEK_SET);
 174                                                 if (t < 0)
 175                                                         return -errno;
 176
 177                                                 t = lseek(fdt, toffset + max_bytes, SEEK_SET);
 178                                                 if (t < 0)
 179                                                         return -errno;
 180
 181                                                 return 1; /* we copied only some number of bytes, which worked, but this means we didn't hit EOF, return 1 */
 182                                         }
 183                                 }
 184                         }
 185                 }
 186         }
 187
 188         for (;;) {
 189                 ssize_t n;
 190
 191                 if (max_bytes <= 0)
 192                         return 1; /* return > 0 if we hit the max_bytes limit */
 193
 194                 if (FLAGS_SET(copy_flags, COPY_SIGINT)) {
 195                         r = sigint_pending();
 196                         if (r < 0)
 197                                 return r;
 198                         if (r > 0)
 199                                 return -EINTR;
 200                 }
 201
 202                 if (max_bytes != UINT64_MAX && m > max_bytes)
 203                         m = max_bytes;
 204
 205                 /* First try copy_file_range(), unless we already tried */
 206                 if (try_cfr) {
 207                         n = try_copy_file_range(fdf, NULL, fdt, NULL, m, 0u);
 208                         if (n < 0) {
 209                                 if (!IN_SET(n, -EINVAL, -ENOSYS, -EXDEV, -EBADF))
 210                                         return n;
 211
 212                                 try_cfr = false;
 213                                 /* use fallback below */
 214                         } else if (n == 0) /* EOF */
 215                                 break;
 216                         else
 217                                 /* Success! */
 218                                 goto next;
 219                 }
 220
 221                 /* First try sendfile(), unless we already tried */
 222                 if (try_sendfile) {
 223                         n = sendfile(fdt, fdf, NULL, m);
 224                         if (n < 0) {
 225                                 if (!IN_SET(errno, EINVAL, ENOSYS))
 226                                         return -errno;
 227
 228                                 try_sendfile = false;
 229                                 /* use fallback below */
 230                         } else if (n == 0) /* EOF */
 231                                 break;
 232                         else
 233                                 /* Success! */
 234                                 goto next;
 235                 }
 236
 237                 /* Then try splice, unless we already tried. */
 238                 if (try_splice) {
 239
 240                         /* splice()'s asynchronous I/O support is a bit weird. When it encounters a pipe file
 241                          * descriptor, then it will ignore its O_NONBLOCK flag and instead only honour the
 242                          * SPLICE_F_NONBLOCK flag specified in its flag parameter. Let's hide this behaviour here, and
 243                          * check if either of the specified fds are a pipe, and if so, let's pass the flag
 244                          * automatically, depending on O_NONBLOCK being set.
 245                          *
 246                          * Here's a twist though: when we use it to move data between two pipes of which one has
 247                          * O_NONBLOCK set and the other has not, then we have no individual control over O_NONBLOCK
 248                          * behaviour. Hence in that case we can't use splice() and still guarantee systematic
 249                          * O_NONBLOCK behaviour, hence don't. */
 250
 251                         if (nonblock_pipe < 0) {
 252                                 int a, b;
 253
 254                                 /* Check if either of these fds is a pipe, and if so non-blocking or not */
 255                                 a = fd_is_nonblock_pipe(fdf);
 256                                 if (a < 0)
 257                                         return a;
 258
 259                                 b = fd_is_nonblock_pipe(fdt);
 260                                 if (b < 0)
 261                                         return b;
 262
 263                                 if ((a == FD_IS_NO_PIPE && b == FD_IS_NO_PIPE) ||
 264                                     (a == FD_IS_BLOCKING_PIPE && b == FD_IS_NONBLOCKING_PIPE) ||
 265                                     (a == FD_IS_NONBLOCKING_PIPE && b == FD_IS_BLOCKING_PIPE))
 266
 267                                         /* splice() only works if one of the fds is a pipe. If neither is, let's skip
 268                                          * this step right-away. As mentioned above, if one of the two fds refers to a
 269                                          * blocking pipe and the other to a non-blocking pipe, we can't use splice()
 270                                          * either, hence don't try either. This hence means we can only use splice() if
 271                                          * either only one of the two fds is a pipe, or if both are pipes with the same
 272                                          * nonblocking flag setting. */
 273
 274                                         try_splice = false;
 275                                 else
 276                                         nonblock_pipe = a == FD_IS_NONBLOCKING_PIPE || b == FD_IS_NONBLOCKING_PIPE;
 277                         }
 278                 }
 279
 280                 if (try_splice) {
 281                         n = splice(fdf, NULL, fdt, NULL, m, nonblock_pipe ? SPLICE_F_NONBLOCK : 0);
 282                         if (n < 0) {
 283                                 if (!IN_SET(errno, EINVAL, ENOSYS))
 284                                         return -errno;
 285
 286                                 try_splice = false;
 287                                 /* use fallback below */
 288                         } else if (n == 0) /* EOF */
 289                                 break;
 290                         else
 291                                 /* Success! */
 292                                 goto next;
 293                 }
 294
 295                 /* As a fallback just copy bits by hand */
 296                 {
 297                         uint8_t buf[MIN(m, COPY_BUFFER_SIZE)], *p = buf;
 298                         ssize_t z;
 299
 300                         n = read(fdf, buf, sizeof buf);
 301                         if (n < 0)
 302                                 return -errno;
 303                         if (n == 0) /* EOF */
 304                                 break;
 305
 306                         z = (size_t) n;
 307                         do {
 308                                 ssize_t k;
 309
 310                                 k = write(fdt, p, z);
 311                                 if (k < 0) {
 312                                         r = -errno;
 313
 314                                         if (ret_remains) {
 315                                                 void *copy;
 316
 317                                                 copy = memdup(p, z);
 318                                                 if (!copy)
 319                                                         return -ENOMEM;
 320
 321                                                 *ret_remains = copy;
 322                                         }
 323
 324                                         if (ret_remains_size)
 325                                                 *ret_remains_size = z;
 326
 327                                         return r;
 328                                 }
 329
 330                                 assert(k <= z);
 331                                 z -= k;
 332                                 p += k;
 333                         } while (z > 0);
 334                 }
 335
 336         next:
 337                 if (progress) {
 338                         r = progress(n, userdata);
 339                         if (r < 0)
 340                                 return r;
 341                 }
 342
 343                 if (max_bytes != (uint64_t) -1) {
 344                         assert(max_bytes >= (uint64_t) n);
 345                         max_bytes -= n;
 346                 }
 347
 348                 /* sendfile accepts at most SSIZE_MAX-offset bytes to copy,
 349                  * so reduce our maximum by the amount we already copied,
 350                  * but don't go below our copy buffer size, unless we are
 351                  * close the limit of bytes we are allowed to copy. */
 352                 m = MAX(MIN(COPY_BUFFER_SIZE, max_bytes), m - n);
 353         }
 354
 355         return 0; /* return 0 if we hit EOF earlier than the size limit */
 356 }
 357
 358 static int fd_copy_symlink(
 359                 int df,
 360                 const char *from,
 361                 const struct stat *st,
 362                 int dt,
 363                 const char *to,
 364                 uid_t override_uid,
 365                 gid_t override_gid,
 366                 CopyFlags copy_flags) {
 367
 368         _cleanup_free_ char *target = NULL;
 369         int r;
 370
 371         assert(from);
 372         assert(st);
 373         assert(to);
 374
 375         r = readlinkat_malloc(df, from, &target);
 376         if (r < 0)
 377                 return r;
 378
 379         if (copy_flags & COPY_MAC_CREATE) {
 380                 r = mac_selinux_create_file_prepare_at(dt, to, S_IFLNK);
 381                 if (r < 0)
 382                         return r;
 383         }
 384         r = symlinkat(target, dt, to);
 385         if (copy_flags & COPY_MAC_CREATE)
 386                 mac_selinux_create_file_clear();
 387         if (r < 0)
 388                 return -errno;
 389
 390         if (fchownat(dt, to,
 391                      uid_is_valid(override_uid) ? override_uid : st->st_uid,
 392                      gid_is_valid(override_gid) ? override_gid : st->st_gid,
 393                      AT_SYMLINK_NOFOLLOW) < 0)
 394                 return -errno;
 395
 396         return 0;
 397 }
 398
 399 /* Encapsulates the database we store potential hardlink targets in */
 400 typedef struct HardlinkContext {
 401         int dir_fd;    /* An fd to the directory we use as lookup table. Never AT_FDCWD. Lazily created, when
 402                         * we add the first entry. */
 403
 404         /* These two fields are used to create the hardlink repository directory above — via
 405          * mkdirat(parent_fd, subdir) — and are kept so that we can automatically remove the directory again
 406          * when we are done. */
 407         int parent_fd; /* Possibly AT_FDCWD */
 408         char *subdir;
 409 } HardlinkContext;
 410
 411 static int hardlink_context_setup(
 412                 HardlinkContext *c,
 413                 int dt,
 414                 const char *to,
 415                 CopyFlags copy_flags) {
 416
 417         _cleanup_close_ int dt_copy = -1;
 418         int r;
 419
 420         assert(c);
 421         assert(c->dir_fd < 0 && c->dir_fd != AT_FDCWD);
 422         assert(c->parent_fd < 0);
 423         assert(!c->subdir);
 424
 425         /* If hardlink recreation is requested we have to maintain a database of inodes that are potential
 426          * hardlink sources. Given that generally disk sizes have to be assumed to be larger than what fits
 427          * into physical RAM we cannot maintain that database in dynamic memory alone. Here we opt to
 428          * maintain it on disk, to simplify things: inside the destination directory we'll maintain a
 429          * temporary directory consisting of hardlinks of every inode we copied that might be subject of
 430          * hardlinks. We can then use that as hardlink source later on. Yes, this means additional disk IO
 431          * but thankfully Linux is optimized for this kind of thing. If this ever becomes a performance
 432          * bottleneck we can certainly place an in-memory hash table in front of this, but for the beginning,
 433          * let's keep things simple, and just use the disk as lookup table for inodes.
 434          *
 435          * Note that this should have zero performance impact as long as .n_link of all files copied remains
 436          * <= 0, because in that case we will not actually allocate the hardlink inode lookup table directory
 437          * on disk (we do so lazily, when the first candidate with .n_link > 1 is seen). This means, in the
 438          * common case where hardlinks are not used at all or only for few files the fact that we store the
 439          * table on disk shouldn't matter perfomance-wise. */
 440
 441         if (!FLAGS_SET(copy_flags, COPY_HARDLINKS))
 442                 return 0;
 443
 444         if (dt == AT_FDCWD)
 445                 dt_copy = AT_FDCWD;
 446         else if (dt < 0)
 447                 return -EBADF;
 448         else {
 449                 dt_copy = fcntl(dt, F_DUPFD_CLOEXEC, 3);
 450                 if (dt_copy < 0)
 451                         return -errno;
 452         }
 453
 454         r = tempfn_random_child(to, "hardlink", &c->subdir);
 455         if (r < 0)
 456                 return r;
 457
 458         c->parent_fd = TAKE_FD(dt_copy);
 459
 460         /* We don't actually create the directory we keep the table in here, that's done on-demand when the
 461          * first entry is added, using hardlink_context_realize() below. */
 462         return 1;
 463 }
 464
 465 static int hardlink_context_realize(HardlinkContext *c) {
 466         int r;
 467
 468         if (!c)
 469                 return 0;
 470
 471         if (c->dir_fd >= 0) /* Already realized */
 472                 return 1;
 473
 474         if (c->parent_fd < 0 && c->parent_fd != AT_FDCWD) /* Not configured */
 475                 return 0;
 476
 477         assert(c->subdir);
 478
 479         if (mkdirat(c->parent_fd, c->subdir, 0700) < 0)
 480                 return -errno;
 481
 482         c->dir_fd = openat(c->parent_fd, c->subdir, O_RDONLY|O_DIRECTORY|O_CLOEXEC);
 483         if (c->dir_fd < 0) {
 484                 r = -errno;
 485                 (void) unlinkat(c->parent_fd, c->subdir, AT_REMOVEDIR);
 486                 return r;
 487         }
 488
 489         return 1;
 490 }
 491
 492 static void hardlink_context_destroy(HardlinkContext *c) {
 493         int r;
 494
 495         assert(c);
 496
 497         /* Automatically remove the hardlink lookup table directory again after we are done. This is used via
 498          * _cleanup_() so that we really delete this, even on failure. */
 499
 500         if (c->dir_fd >= 0) {
 501                 r = rm_rf_children(TAKE_FD(c->dir_fd), REMOVE_PHYSICAL, NULL); /* consumes dir_fd in all cases, even on failure */
 502                 if (r < 0)
 503                         log_debug_errno(r, "Failed to remove hardlink store (%s) contents, ignoring: %m", c->subdir);
 504
 505                 assert(c->parent_fd >= 0 || c->parent_fd == AT_FDCWD);
 506                 assert(c->subdir);
 507
 508                 if (unlinkat(c->parent_fd, c->subdir, AT_REMOVEDIR) < 0)
 509                         log_debug_errno(errno, "Failed to remove hardlink store (%s) directory, ignoring: %m", c->subdir);
 510         }
 511
 512         assert_cc(AT_FDCWD < 0);
 513         c->parent_fd = safe_close(c->parent_fd);
 514
 515         c->subdir = mfree(c->subdir);
 516 }
 517
 518 static int try_hardlink(
 519                 HardlinkContext *c,
 520                 const struct stat *st,
 521                 int dt,
 522                 const char *to) {
 523
 524         char dev_ino[DECIMAL_STR_MAX(dev_t)*2 + DECIMAL_STR_MAX(uint64_t) + 4];
 525
 526         assert(st);
 527         assert(dt >= 0 || dt == AT_FDCWD);
 528         assert(to);
 529
 530         if (!c) /* No temporary hardlink directory, don't bother */
 531                 return 0;
 532
 533         if (st->st_nlink <= 1) /* Source not hardlinked, don't bother */
 534                 return 0;
 535
 536         if (c->dir_fd < 0) /* not yet realized, hence empty */
 537                 return 0;
 538
 539         xsprintf(dev_ino, "%u:%u:%" PRIu64, major(st->st_dev), minor(st->st_dev), (uint64_t) st->st_ino);
 540         if (linkat(c->dir_fd, dev_ino, dt, to, 0) < 0)  {
 541                 if (errno != ENOENT) /* doesn't exist in store yet */
 542                         log_debug_errno(errno, "Failed to hardlink %s to %s, ignoring: %m", dev_ino, to);
 543                 return 0;
 544         }
 545
 546         return 1;
 547 }
 548
 549 static int memorize_hardlink(
 550                 HardlinkContext *c,
 551                 const struct stat *st,
 552                 int dt,
 553                 const char *to) {
 554
 555         char dev_ino[DECIMAL_STR_MAX(dev_t)*2 + DECIMAL_STR_MAX(uint64_t) + 4];
 556         int r;
 557
 558         assert(st);
 559         assert(dt >= 0 || dt == AT_FDCWD);
 560         assert(to);
 561
 562         if (!c) /* No temporary hardlink directory, don't bother */
 563                 return 0;
 564
 565         if (st->st_nlink <= 1) /* Source not hardlinked, don't bother */
 566                 return 0;
 567
 568         r = hardlink_context_realize(c); /* Create the hardlink store lazily */
 569         if (r < 0)
 570                 return r;
 571
 572         xsprintf(dev_ino, "%u:%u:%" PRIu64, major(st->st_dev), minor(st->st_dev), (uint64_t) st->st_ino);
 573         if (linkat(dt, to, c->dir_fd, dev_ino, 0) < 0) {
 574                 log_debug_errno(errno, "Failed to hardlink %s to %s, ignoring: %m", to, dev_ino);
 575                 return 0;
 576         }
 577
 578         return 1;
 579 }
 580
 581 static int fd_copy_regular(
 582                 int df,
 583                 const char *from,
 584                 const struct stat *st,
 585                 int dt,
 586                 const char *to,
 587                 uid_t override_uid,
 588                 gid_t override_gid,
 589                 CopyFlags copy_flags,
 590                 HardlinkContext *hardlink_context,
 591                 copy_progress_bytes_t progress,
 592                 void *userdata) {
 593
 594         _cleanup_close_ int fdf = -1, fdt = -1;
 595         struct timespec ts[2];
 596         int r, q;
 597
 598         assert(from);
 599         assert(st);
 600         assert(to);
 601
 602         r = try_hardlink(hardlink_context, st, dt, to);
 603         if (r < 0)
 604                 return r;
 605         if (r > 0) /* worked! */
 606                 return 0;
 607
 608         fdf = openat(df, from, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
 609         if (fdf < 0)
 610                 return -errno;
 611
 612         if (copy_flags & COPY_MAC_CREATE) {
 613                 r = mac_selinux_create_file_prepare_at(dt, to, S_IFREG);
 614                 if (r < 0)
 615                         return r;
 616         }
 617         fdt = openat(dt, to, O_WRONLY|O_CREAT|O_EXCL|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW, st->st_mode & 07777);
 618         if (copy_flags & COPY_MAC_CREATE)
 619                 mac_selinux_create_file_clear();
 620         if (fdt < 0)
 621                 return -errno;
 622
 623         r = copy_bytes_full(fdf, fdt, (uint64_t) -1, copy_flags, NULL, NULL, progress, userdata);
 624         if (r < 0) {
 625                 (void) unlinkat(dt, to, 0);
 626                 return r;
 627         }
 628
 629         if (fchown(fdt,
 630                    uid_is_valid(override_uid) ? override_uid : st->st_uid,
 631                    gid_is_valid(override_gid) ? override_gid : st->st_gid) < 0)
 632                 r = -errno;
 633
 634         if (fchmod(fdt, st->st_mode & 07777) < 0)
 635                 r = -errno;
 636
 637         ts[0] = st->st_atim;
 638         ts[1] = st->st_mtim;
 639         (void) futimens(fdt, ts);
 640         (void) copy_xattr(fdf, fdt);
 641
 642         q = close(fdt);
 643         fdt = -1;
 644
 645         if (q < 0) {
 646                 r = -errno;
 647                 (void) unlinkat(dt, to, 0);
 648         }
 649
 650         (void) memorize_hardlink(hardlink_context, st, dt, to);
 651         return r;
 652 }
 653
 654 static int fd_copy_fifo(
 655                 int df,
 656                 const char *from,
 657                 const struct stat *st,
 658                 int dt,
 659                 const char *to,
 660                 uid_t override_uid,
 661                 gid_t override_gid,
 662                 CopyFlags copy_flags,
 663                 HardlinkContext *hardlink_context) {
 664         int r;
 665
 666         assert(from);
 667         assert(st);
 668         assert(to);
 669
 670         r = try_hardlink(hardlink_context, st, dt, to);
 671         if (r < 0)
 672                 return r;
 673         if (r > 0) /* worked! */
 674                 return 0;
 675
 676         if (copy_flags & COPY_MAC_CREATE) {
 677                 r = mac_selinux_create_file_prepare_at(dt, to, S_IFIFO);
 678                 if (r < 0)
 679                         return r;
 680         }
 681         r = mkfifoat(dt, to, st->st_mode & 07777);
 682         if (copy_flags & COPY_MAC_CREATE)
 683                 mac_selinux_create_file_clear();
 684         if (r < 0)
 685                 return -errno;
 686
 687         if (fchownat(dt, to,
 688                      uid_is_valid(override_uid) ? override_uid : st->st_uid,
 689                      gid_is_valid(override_gid) ? override_gid : st->st_gid,
 690                      AT_SYMLINK_NOFOLLOW) < 0)
 691                 r = -errno;
 692
 693         if (fchmodat(dt, to, st->st_mode & 07777, 0) < 0)
 694                 r = -errno;
 695
 696         (void) memorize_hardlink(hardlink_context, st, dt, to);
 697         return r;
 698 }
 699
 700 static int fd_copy_node(
 701                 int df,
 702                 const char *from,
 703                 const struct stat *st,
 704                 int dt,
 705                 const char *to,
 706                 uid_t override_uid,
 707                 gid_t override_gid,
 708                 CopyFlags copy_flags,
 709                 HardlinkContext *hardlink_context) {
 710         int r;
 711
 712         assert(from);
 713         assert(st);
 714         assert(to);
 715
 716         r = try_hardlink(hardlink_context, st, dt, to);
 717         if (r < 0)
 718                 return r;
 719         if (r > 0) /* worked! */
 720                 return 0;
 721
 722         if (copy_flags & COPY_MAC_CREATE) {
 723                 r = mac_selinux_create_file_prepare_at(dt, to, st->st_mode & S_IFMT);
 724                 if (r < 0)
 725                         return r;
 726         }
 727         r = mknodat(dt, to, st->st_mode, st->st_rdev);
 728         if (copy_flags & COPY_MAC_CREATE)
 729                 mac_selinux_create_file_clear();
 730         if (r < 0)
 731                 return -errno;
 732
 733         if (fchownat(dt, to,
 734                      uid_is_valid(override_uid) ? override_uid : st->st_uid,
 735                      gid_is_valid(override_gid) ? override_gid : st->st_gid,
 736                      AT_SYMLINK_NOFOLLOW) < 0)
 737                 r = -errno;
 738
 739         if (fchmodat(dt, to, st->st_mode & 07777, 0) < 0)
 740                 r = -errno;
 741
 742         (void) memorize_hardlink(hardlink_context, st, dt, to);
 743         return r;
 744 }
 745
 746 static int fd_copy_directory(
 747                 int df,
 748                 const char *from,
 749                 const struct stat *st,
 750                 int dt,
 751                 const char *to,
 752                 dev_t original_device,
 753                 unsigned depth_left,
 754                 uid_t override_uid,
 755                 gid_t override_gid,
 756                 CopyFlags copy_flags,
 757                 HardlinkContext *hardlink_context,
 758                 const char *display_path,
 759                 copy_progress_path_t progress_path,
 760                 copy_progress_bytes_t progress_bytes,
 761                 void *userdata) {
 762
 763         _cleanup_(hardlink_context_destroy) HardlinkContext our_hardlink_context = {
 764                 .dir_fd = -1,
 765                 .parent_fd = -1,
 766         };
 767
 768         _cleanup_close_ int fdf = -1, fdt = -1;
 769         _cleanup_closedir_ DIR *d = NULL;
 770         struct dirent *de;
 771         bool exists, created;
 772         int r;
 773
 774         assert(st);
 775         assert(to);
 776
 777         if (depth_left == 0)
 778                 return -ENAMETOOLONG;
 779
 780         if (from)
 781                 fdf = openat(df, from, O_RDONLY|O_DIRECTORY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
 782         else
 783                 fdf = fcntl(df, F_DUPFD_CLOEXEC, 3);
 784         if (fdf < 0)
 785                 return -errno;
 786
 787         if (!hardlink_context) {
 788                 /* If recreating hardlinks is requested let's set up a context for that now. */
 789                 r = hardlink_context_setup(&our_hardlink_context, dt, to, copy_flags);
 790                 if (r < 0)
 791                         return r;
 792                 if (r > 0) /* It's enabled and allocated, let's now use the same context for all recursive
 793                             * invocations from here down */
 794                         hardlink_context = &our_hardlink_context;
 795         }
 796
 797         d = take_fdopendir(&fdf);
 798         if (!d)
 799                 return -errno;
 800
 801         exists = false;
 802         if (copy_flags & COPY_MERGE_EMPTY) {
 803                 r = dir_is_empty_at(dt, to);
 804                 if (r < 0 && r != -ENOENT)
 805                         return r;
 806                 else if (r == 1)
 807                         exists = true;
 808         }
 809
 810         if (exists)
 811                 created = false;
 812         else {
 813                 if (copy_flags & COPY_MAC_CREATE)
 814                         r = mkdirat_label(dt, to, st->st_mode & 07777);
 815                 else
 816                         r = mkdirat(dt, to, st->st_mode & 07777);
 817                 if (r >= 0)
 818                         created = true;
 819                 else if (errno == EEXIST && (copy_flags & COPY_MERGE))
 820                         created = false;
 821                 else
 822                         return -errno;
 823         }
 824
 825         fdt = openat(dt, to, O_RDONLY|O_DIRECTORY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
 826         if (fdt < 0)
 827                 return -errno;
 828
 829         r = 0;
 830
 831         FOREACH_DIRENT_ALL(de, d, return -errno) {
 832                 const char *child_display_path = NULL;
 833                 _cleanup_free_ char *dp = NULL;
 834                 struct stat buf;
 835                 int q;
 836
 837                 if (dot_or_dot_dot(de->d_name))
 838                         continue;
 839
 840                 if (FLAGS_SET(copy_flags, COPY_SIGINT)) {
 841                         r = sigint_pending();
 842                         if (r < 0)
 843                                 return r;
 844                         if (r > 0)
 845                                 return -EINTR;
 846                 }
 847
 848                 if (fstatat(dirfd(d), de->d_name, &buf, AT_SYMLINK_NOFOLLOW) < 0) {
 849                         r = -errno;
 850                         continue;
 851                 }
 852
 853                 if (progress_path) {
 854                         if (display_path)
 855                                 child_display_path = dp = path_join(display_path, de->d_name);
 856                         else
 857                                 child_display_path = de->d_name;
 858
 859                         r = progress_path(child_display_path, &buf, userdata);
 860                         if (r < 0)
 861                                 return r;
 862                 }
 863
 864                 if (S_ISDIR(buf.st_mode)) {
 865                         /*
 866                          * Don't descend into directories on other file systems, if this is requested. We do a simple
 867                          * .st_dev check here, which basically comes for free. Note that we do this check only on
 868                          * directories, not other kind of file system objects, for two reason:
 869                          *
 870                          * • The kernel's overlayfs pseudo file system that overlays multiple real file systems
 871                          *   propagates the .st_dev field of the file system a file originates from all the way up
 872                          *   through the stack to stat(). It doesn't do that for directories however. This means that
 873                          *   comparing .st_dev on non-directories suggests that they all are mount points. To avoid
 874                          *   confusion we hence avoid relying on this check for regular files.
 875                          *
 876                          * • The main reason we do this check at all is to protect ourselves from bind mount cycles,
 877                          *   where we really want to avoid descending down in all eternity. However the .st_dev check
 878                          *   is usually not sufficient for this protection anyway, as bind mount cycles from the same
 879                          *   file system onto itself can't be detected that way. (Note we also do a recursion depth
 880                          *   check, which is probably the better protection in this regard, which is why
 881                          *   COPY_SAME_MOUNT is optional).
 882                          */
 883
 884                         if (FLAGS_SET(copy_flags, COPY_SAME_MOUNT)) {
 885                                 if (buf.st_dev != original_device)
 886                                         continue;
 887
 888                                 r = fd_is_mount_point(dirfd(d), de->d_name, 0);
 889                                 if (r < 0)
 890                                         return r;
 891                                 if (r > 0)
 892                                         continue;
 893                         }
 894
 895                         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, hardlink_context, child_display_path, progress_path, progress_bytes, userdata);
 896                 } else if (S_ISREG(buf.st_mode))
 897                         q = fd_copy_regular(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags, hardlink_context, progress_bytes, userdata);
 898                 else if (S_ISLNK(buf.st_mode))
 899                         q = fd_copy_symlink(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags);
 900                 else if (S_ISFIFO(buf.st_mode))
 901                         q = fd_copy_fifo(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags, hardlink_context);
 902                 else if (S_ISBLK(buf.st_mode) || S_ISCHR(buf.st_mode) || S_ISSOCK(buf.st_mode))
 903                         q = fd_copy_node(dirfd(d), de->d_name, &buf, fdt, de->d_name, override_uid, override_gid, copy_flags, hardlink_context);
 904                 else
 905                         q = -EOPNOTSUPP;
 906
 907                 if (q == -EINTR) /* Propagate SIGINT up instantly */
 908                         return q;
 909                 if (q == -EEXIST && (copy_flags & COPY_MERGE))
 910                         q = 0;
 911                 if (q < 0)
 912                         r = q;
 913         }
 914
 915         if (created) {
 916                 struct timespec ut[2] = {
 917                         st->st_atim,
 918                         st->st_mtim
 919                 };
 920
 921                 if (fchown(fdt,
 922                            uid_is_valid(override_uid) ? override_uid : st->st_uid,
 923                            gid_is_valid(override_gid) ? override_gid : st->st_gid) < 0)
 924                         r = -errno;
 925
 926                 if (fchmod(fdt, st->st_mode & 07777) < 0)
 927                         r = -errno;
 928
 929                 (void) copy_xattr(dirfd(d), fdt);
 930                 (void) futimens(fdt, ut);
 931         }
 932
 933         return r;
 934 }
 935
 936 int copy_tree_at_full(
 937                 int fdf,
 938                 const char *from,
 939                 int fdt,
 940                 const char *to,
 941                 uid_t override_uid,
 942                 gid_t override_gid,
 943                 CopyFlags copy_flags,
 944                 copy_progress_path_t progress_path,
 945                 copy_progress_bytes_t progress_bytes,
 946                 void *userdata) {
 947
 948         struct stat st;
 949
 950         assert(from);
 951         assert(to);
 952
 953         if (fstatat(fdf, from, &st, AT_SYMLINK_NOFOLLOW) < 0)
 954                 return -errno;
 955
 956         if (S_ISREG(st.st_mode))
 957                 return fd_copy_regular(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags, NULL, progress_bytes, userdata);
 958         else if (S_ISDIR(st.st_mode))
 959                 return fd_copy_directory(fdf, from, &st, fdt, to, st.st_dev, COPY_DEPTH_MAX, override_uid, override_gid, copy_flags, NULL, NULL, progress_path, progress_bytes, userdata);
 960         else if (S_ISLNK(st.st_mode))
 961                 return fd_copy_symlink(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags);
 962         else if (S_ISFIFO(st.st_mode))
 963                 return fd_copy_fifo(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags, NULL);
 964         else if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode) || S_ISSOCK(st.st_mode))
 965                 return fd_copy_node(fdf, from, &st, fdt, to, override_uid, override_gid, copy_flags, NULL);
 966         else
 967                 return -EOPNOTSUPP;
 968 }
 969
 970 int copy_directory_fd_full(
 971                 int dirfd,
 972                 const char *to,
 973                 CopyFlags copy_flags,
 974                 copy_progress_path_t progress_path,
 975                 copy_progress_bytes_t progress_bytes,
 976                 void *userdata) {
 977
 978         struct stat st;
 979
 980         assert(dirfd >= 0);
 981         assert(to);
 982
 983         if (fstat(dirfd, &st) < 0)
 984                 return -errno;
 985
 986         if (!S_ISDIR(st.st_mode))
 987                 return -ENOTDIR;
 988
 989         return fd_copy_directory(dirfd, NULL, &st, AT_FDCWD, to, st.st_dev, COPY_DEPTH_MAX, UID_INVALID, GID_INVALID, copy_flags, NULL, NULL, progress_path, progress_bytes, userdata);
 990 }
 991
 992 int copy_directory_full(
 993                 const char *from,
 994                 const char *to,
 995                 CopyFlags copy_flags,
 996                 copy_progress_path_t progress_path,
 997                 copy_progress_bytes_t progress_bytes,
 998                 void *userdata) {
 999
1000         struct stat st;
1001
1002         assert(from);
1003         assert(to);
1004
1005         if (lstat(from, &st) < 0)
1006                 return -errno;
1007
1008         if (!S_ISDIR(st.st_mode))
1009                 return -ENOTDIR;
1010
1011         return fd_copy_directory(AT_FDCWD, from, &st, AT_FDCWD, to, st.st_dev, COPY_DEPTH_MAX, UID_INVALID, GID_INVALID, copy_flags, NULL, NULL, progress_path, progress_bytes, userdata);
1012 }
1013
1014 int copy_file_fd_full(
1015                 const char *from,
1016                 int fdt,
1017                 CopyFlags copy_flags,
1018                 copy_progress_bytes_t progress_bytes,
1019                 void *userdata) {
1020
1021         _cleanup_close_ int fdf = -1;
1022         int r;
1023
1024         assert(from);
1025         assert(fdt >= 0);
1026
1027         fdf = open(from, O_RDONLY|O_CLOEXEC|O_NOCTTY);
1028         if (fdf < 0)
1029                 return -errno;
1030
1031         r = copy_bytes_full(fdf, fdt, (uint64_t) -1, copy_flags, NULL, NULL, progress_bytes, userdata);
1032
1033         (void) copy_times(fdf, fdt, copy_flags);
1034         (void) copy_xattr(fdf, fdt);
1035
1036         return r;
1037 }
1038
1039 int copy_file_full(
1040                 const char *from,
1041                 const char *to,
1042                 int flags,
1043                 mode_t mode,
1044                 unsigned chattr_flags,
1045                 unsigned chattr_mask,
1046                 CopyFlags copy_flags,
1047                 copy_progress_bytes_t progress_bytes,
1048                 void *userdata) {
1049
1050         int fdt = -1, r;
1051
1052         assert(from);
1053         assert(to);
1054
1055         RUN_WITH_UMASK(0000) {
1056                 if (copy_flags & COPY_MAC_CREATE) {
1057                         r = mac_selinux_create_file_prepare(to, S_IFREG);
1058                         if (r < 0)
1059                                 return r;
1060                 }
1061                 fdt = open(to, flags|O_WRONLY|O_CREAT|O_CLOEXEC|O_NOCTTY, mode);
1062                 if (copy_flags & COPY_MAC_CREATE)
1063                         mac_selinux_create_file_clear();
1064                 if (fdt < 0)
1065                         return -errno;
1066         }
1067
1068         if (chattr_mask != 0)
1069                 (void) chattr_fd(fdt, chattr_flags, chattr_mask & CHATTR_EARLY_FL, NULL);
1070
1071         r = copy_file_fd_full(from, fdt, copy_flags, progress_bytes, userdata);
1072         if (r < 0) {
1073                 close(fdt);
1074                 (void) unlink(to);
1075                 return r;
1076         }
1077
1078         if (chattr_mask != 0)
1079                 (void) chattr_fd(fdt, chattr_flags, chattr_mask & ~CHATTR_EARLY_FL, NULL);
1080
1081         if (close(fdt) < 0) {
1082                 unlink_noerrno(to);
1083                 return -errno;
1084         }
1085
1086         return 0;
1087 }
1088
1089 int copy_file_atomic_full(
1090                 const char *from,
1091                 const char *to,
1092                 mode_t mode,
1093                 unsigned chattr_flags,
1094                 unsigned chattr_mask,
1095                 CopyFlags copy_flags,
1096                 copy_progress_bytes_t progress_bytes,
1097                 void *userdata) {
1098
1099         _cleanup_(unlink_and_freep) char *t = NULL;
1100         _cleanup_close_ int fdt = -1;
1101         int r;
1102
1103         assert(from);
1104         assert(to);
1105
1106         /* We try to use O_TMPFILE here to create the file if we can. Note that that only works if COPY_REPLACE is not
1107          * set though as we need to use linkat() for linking the O_TMPFILE file into the file system but that system
1108          * call can't replace existing files. Hence, if COPY_REPLACE is set we create a temporary name in the file
1109          * system right-away and unconditionally which we then can renameat() to the right name after we completed
1110          * writing it. */
1111
1112         if (copy_flags & COPY_REPLACE) {
1113                 r = tempfn_random(to, NULL, &t);
1114                 if (r < 0)
1115                         return r;
1116
1117                 if (copy_flags & COPY_MAC_CREATE) {
1118                         r = mac_selinux_create_file_prepare(to, S_IFREG);
1119                         if (r < 0) {
1120                                 t = mfree(t);
1121                                 return r;
1122                         }
1123                 }
1124                 fdt = open(t, O_CREAT|O_EXCL|O_NOFOLLOW|O_NOCTTY|O_WRONLY|O_CLOEXEC, 0600);
1125                 if (copy_flags & COPY_MAC_CREATE)
1126                         mac_selinux_create_file_clear();
1127                 if (fdt < 0) {
1128                         t = mfree(t);
1129                         return -errno;
1130                 }
1131         } else {
1132                 if (copy_flags & COPY_MAC_CREATE) {
1133                         r = mac_selinux_create_file_prepare(to, S_IFREG);
1134                         if (r < 0)
1135                                 return r;
1136                 }
1137                 fdt = open_tmpfile_linkable(to, O_WRONLY|O_CLOEXEC, &t);
1138                 if (copy_flags & COPY_MAC_CREATE)
1139                         mac_selinux_create_file_clear();
1140                 if (fdt < 0)
1141                         return fdt;
1142         }
1143
1144         if (chattr_mask != 0)
1145                 (void) chattr_fd(fdt, chattr_flags, chattr_mask & CHATTR_EARLY_FL, NULL);
1146
1147         r = copy_file_fd_full(from, fdt, copy_flags, progress_bytes, userdata);
1148         if (r < 0)
1149                 return r;
1150
1151         if (fchmod(fdt, mode) < 0)
1152                 return -errno;
1153
1154         if (copy_flags & COPY_REPLACE) {
1155                 if (renameat(AT_FDCWD, t, AT_FDCWD, to) < 0)
1156                         return -errno;
1157         } else {
1158                 r = link_tmpfile(fdt, t, to);
1159                 if (r < 0)
1160                         return r;
1161         }
1162
1163         if (chattr_mask != 0)
1164                 (void) chattr_fd(fdt, chattr_flags, chattr_mask & ~CHATTR_EARLY_FL, NULL);
1165
1166         t = mfree(t);
1167         return 0;
1168 }
1169
1170 int copy_times(int fdf, int fdt, CopyFlags flags) {
1171         struct timespec ut[2];
1172         struct stat st;
1173
1174         assert(fdf >= 0);
1175         assert(fdt >= 0);
1176
1177         if (fstat(fdf, &st) < 0)
1178                 return -errno;
1179
1180         ut[0] = st.st_atim;
1181         ut[1] = st.st_mtim;
1182
1183         if (futimens(fdt, ut) < 0)
1184                 return -errno;
1185
1186         if (FLAGS_SET(flags, COPY_CRTIME)) {
1187                 usec_t crtime;
1188
1189                 if (fd_getcrtime(fdf, &crtime) >= 0)
1190                         (void) fd_setcrtime(fdt, crtime);
1191         }
1192
1193         return 0;
1194 }
1195
1196 int copy_access(int fdf, int fdt) {
1197         struct stat st;
1198
1199         assert(fdf >= 0);
1200         assert(fdt >= 0);
1201
1202         if (fstat(fdf, &st) < 0)
1203                 return -errno;
1204
1205         if (fchmod(fdt, st.st_mode & 07777) < 0)
1206                 return -errno;
1207
1208         return 0;
1209 }
1210
1211 int copy_xattr(int fdf, int fdt) {
1212         _cleanup_free_ char *names = NULL;
1213         int ret = 0, r;
1214         const char *p;
1215
1216         r = flistxattr_malloc(fdf, &names);
1217         if (r < 0)
1218                 return r;
1219
1220         NULSTR_FOREACH(p, names) {
1221                 _cleanup_free_ char *value = NULL;
1222
1223                 if (!startswith(p, "user."))
1224                         continue;
1225
1226                 r = fgetxattr_malloc(fdf, p, &value);
1227                 if (r == -ENODATA)
1228                         continue; /* gone by now */
1229                 if (r < 0)
1230                         return r;
1231
1232                 if (fsetxattr(fdt, p, value, r, 0) < 0)
1233                         ret = -errno;
1234         }
1235
1236         return ret;
1237 }