src/shared/dissect-image.c

   1 /* SPDX-License-Identifier: LGPL-2.1+ */
   2
   3 #if HAVE_VALGRIND_MEMCHECK_H
   4 #include <valgrind/memcheck.h>
   5 #endif
   6
   7 #include <linux/dm-ioctl.h>
   8 #include <linux/loop.h>
   9 #include <sys/mount.h>
  10 #include <sys/prctl.h>
  11 #include <sys/wait.h>
  12
  13 #include "sd-device.h"
  14 #include "sd-id128.h"
  15
  16 #include "architecture.h"
  17 #include "ask-password-api.h"
  18 #include "blkid-util.h"
  19 #include "blockdev-util.h"
  20 #include "copy.h"
  21 #include "crypt-util.h"
  22 #include "def.h"
  23 #include "device-nodes.h"
  24 #include "device-util.h"
  25 #include "dissect-image.h"
  26 #include "dm-util.h"
  27 #include "env-file.h"
  28 #include "fd-util.h"
  29 #include "fileio.h"
  30 #include "fs-util.h"
  31 #include "fsck-util.h"
  32 #include "gpt.h"
  33 #include "hexdecoct.h"
  34 #include "hostname-util.h"
  35 #include "id128-util.h"
  36 #include "mkdir.h"
  37 #include "mount-util.h"
  38 #include "mountpoint-util.h"
  39 #include "namespace-util.h"
  40 #include "nulstr-util.h"
  41 #include "os-util.h"
  42 #include "path-util.h"
  43 #include "process-util.h"
  44 #include "raw-clone.h"
  45 #include "signal-util.h"
  46 #include "stat-util.h"
  47 #include "stdio-util.h"
  48 #include "string-table.h"
  49 #include "string-util.h"
  50 #include "strv.h"
  51 #include "tmpfile-util.h"
  52 #include "udev-util.h"
  53 #include "user-util.h"
  54 #include "xattr-util.h"
  55
  56 int probe_filesystem(const char *node, char **ret_fstype) {
  57         /* Try to find device content type and return it in *ret_fstype. If nothing is found,
  58          * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and an
  59          * different error otherwise. */
  60
  61 #if HAVE_BLKID
  62         _cleanup_(blkid_free_probep) blkid_probe b = NULL;
  63         const char *fstype;
  64         int r;
  65
  66         errno = 0;
  67         b = blkid_new_probe_from_filename(node);
  68         if (!b)
  69                 return errno_or_else(ENOMEM);
  70
  71         blkid_probe_enable_superblocks(b, 1);
  72         blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE);
  73
  74         errno = 0;
  75         r = blkid_do_safeprobe(b);
  76         if (r == 1) {
  77                 log_debug("No type detected on partition %s", node);
  78                 goto not_found;
  79         }
  80         if (r == -2)
  81                 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN),
  82                                        "Results ambiguous for partition %s", node);
  83         if (r != 0)
  84                 return errno_or_else(EIO);
  85
  86         (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
  87
  88         if (fstype) {
  89                 char *t;
  90
  91                 t = strdup(fstype);
  92                 if (!t)
  93                         return -ENOMEM;
  94
  95                 *ret_fstype = t;
  96                 return 1;
  97         }
  98
  99 not_found:
 100         *ret_fstype = NULL;
 101         return 0;
 102 #else
 103         return -EOPNOTSUPP;
 104 #endif
 105 }
 106
 107 #if HAVE_BLKID
 108 /* Detect RPMB and Boot partitions, which are not listed by blkid.
 109  * See https://github.com/systemd/systemd/issues/5806. */
 110 static bool device_is_mmc_special_partition(sd_device *d) {
 111         const char *sysname;
 112
 113         assert(d);
 114
 115         if (sd_device_get_sysname(d, &sysname) < 0)
 116                 return false;
 117
 118         return startswith(sysname, "mmcblk") &&
 119                 (endswith(sysname, "rpmb") || endswith(sysname, "boot0") || endswith(sysname, "boot1"));
 120 }
 121
 122 static bool device_is_block(sd_device *d) {
 123         const char *ss;
 124
 125         assert(d);
 126
 127         if (sd_device_get_subsystem(d, &ss) < 0)
 128                 return false;
 129
 130         return streq(ss, "block");
 131 }
 132
 133 static int enumerator_for_parent(sd_device *d, sd_device_enumerator **ret) {
 134         _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
 135         int r;
 136
 137         assert(d);
 138         assert(ret);
 139
 140         r = sd_device_enumerator_new(&e);
 141         if (r < 0)
 142                 return r;
 143
 144         r = sd_device_enumerator_allow_uninitialized(e);
 145         if (r < 0)
 146                 return r;
 147
 148         r = sd_device_enumerator_add_match_parent(e, d);
 149         if (r < 0)
 150                 return r;
 151
 152         *ret = TAKE_PTR(e);
 153         return 0;
 154 }
 155
 156 /* how many times to wait for the device nodes to appear */
 157 #define N_DEVICE_NODE_LIST_ATTEMPTS 10
 158
 159 static int wait_for_partitions_to_appear(
 160                 int fd,
 161                 sd_device *d,
 162                 unsigned num_partitions,
 163                 DissectImageFlags flags,
 164                 sd_device_enumerator **ret_enumerator) {
 165
 166         _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
 167         sd_device *q;
 168         unsigned n;
 169         int r;
 170
 171         assert(fd >= 0);
 172         assert(d);
 173         assert(ret_enumerator);
 174
 175         r = enumerator_for_parent(d, &e);
 176         if (r < 0)
 177                 return r;
 178
 179         /* Count the partitions enumerated by the kernel */
 180         n = 0;
 181         FOREACH_DEVICE(e, q) {
 182                 if (sd_device_get_devnum(q, NULL) < 0)
 183                         continue;
 184                 if (!device_is_block(q))
 185                         continue;
 186                 if (device_is_mmc_special_partition(q))
 187                         continue;
 188
 189                 if (!FLAGS_SET(flags, DISSECT_IMAGE_NO_UDEV)) {
 190                         r = device_wait_for_initialization(q, "block", USEC_INFINITY, NULL);
 191                         if (r < 0)
 192                                 return r;
 193                 }
 194
 195                 n++;
 196         }
 197
 198         if (n == num_partitions + 1) {
 199                 *ret_enumerator = TAKE_PTR(e);
 200                 return 0; /* success! */
 201         }
 202         if (n > num_partitions + 1)
 203                 return log_debug_errno(SYNTHETIC_ERRNO(EIO),
 204                                        "blkid and kernel partition lists do not match.");
 205
 206         /* The kernel has probed fewer partitions than blkid? Maybe the kernel prober is still running or it
 207          * got EBUSY because udev already opened the device. Let's reprobe the device, which is a synchronous
 208          * call that waits until probing is complete. */
 209
 210         for (unsigned j = 0; ; j++) {
 211                 if (j++ > 20)
 212                         return -EBUSY;
 213
 214                 if (ioctl(fd, BLKRRPART, 0) >= 0)
 215                         break;
 216                 r = -errno;
 217                 if (r == -EINVAL) {
 218                         /* If we are running on a block device that has partition scanning off, return an
 219                          * explicit recognizable error about this, so that callers can generate a proper
 220                          * message explaining the situation. */
 221
 222                         r = blockdev_partscan_enabled(fd);
 223                         if (r < 0)
 224                                 return r;
 225                         if (r == 0)
 226                                 return log_debug_errno(EPROTONOSUPPORT,
 227                                                        "Device is a loop device and partition scanning is off!");
 228
 229                         return -EINVAL; /* original error */
 230                 }
 231                 if (r != -EBUSY)
 232                         return r;
 233
 234                 /* If something else has the device open, such as an udev rule, the ioctl will return
 235                  * EBUSY. Since there's no way to wait until it isn't busy anymore, let's just wait a bit,
 236                  * and try again.
 237                  *
 238                  * This is really something they should fix in the kernel! */
 239                 (void) usleep(50 * USEC_PER_MSEC);
 240
 241         }
 242
 243         return -EAGAIN; /* no success yet, try again */
 244 }
 245
 246 static int loop_wait_for_partitions_to_appear(
 247                 int fd,
 248                 sd_device *d,
 249                 unsigned num_partitions,
 250                 DissectImageFlags flags,
 251                 sd_device_enumerator **ret_enumerator) {
 252         _cleanup_(sd_device_unrefp) sd_device *device = NULL;
 253         int r;
 254
 255         assert(fd >= 0);
 256         assert(d);
 257         assert(ret_enumerator);
 258
 259         log_debug("Waiting for device (parent + %d partitions) to appear...", num_partitions);
 260
 261         if (!FLAGS_SET(flags, DISSECT_IMAGE_NO_UDEV)) {
 262                 r = device_wait_for_initialization(d, "block", USEC_INFINITY, &device);
 263                 if (r < 0)
 264                         return r;
 265         } else
 266                 device = sd_device_ref(d);
 267
 268         for (unsigned i = 0; i < N_DEVICE_NODE_LIST_ATTEMPTS; i++) {
 269                 r = wait_for_partitions_to_appear(fd, device, num_partitions, flags, ret_enumerator);
 270                 if (r != -EAGAIN)
 271                         return r;
 272         }
 273
 274         return log_debug_errno(SYNTHETIC_ERRNO(ENXIO),
 275                                "Kernel partitions dit not appear within %d attempts",
 276                                N_DEVICE_NODE_LIST_ATTEMPTS);
 277 }
 278
 279 static void check_partition_flags(
 280                 const char *node,
 281                 unsigned long long pflags,
 282                 unsigned long long supported) {
 283
 284         assert(node);
 285
 286         /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
 287         pflags &= ~(supported | GPT_FLAG_REQUIRED_PARTITION | GPT_FLAG_NO_BLOCK_IO_PROTOCOL | GPT_FLAG_LEGACY_BIOS_BOOTABLE);
 288
 289         if (pflags == 0)
 290                 return;
 291
 292         /* If there are other bits set, then log about it, to make things discoverable */
 293         for (unsigned i = 0; i < sizeof(pflags) * 8; i++) {
 294                 unsigned long long bit = 1ULL << i;
 295                 if (!FLAGS_SET(pflags, bit))
 296                         continue;
 297
 298                 log_debug("Unexpected partition flag %llu set on %s!", bit, node);
 299         }
 300 }
 301
 302 #endif
 303
 304 int dissect_image(
 305                 int fd,
 306                 const void *root_hash,
 307                 size_t root_hash_size,
 308                 const char *verity_data,
 309                 const MountOptions *mount_options,
 310                 DissectImageFlags flags,
 311                 DissectedImage **ret) {
 312
 313 #if HAVE_BLKID
 314         sd_id128_t root_uuid = SD_ID128_NULL, verity_uuid = SD_ID128_NULL;
 315         _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
 316         bool is_gpt, is_mbr, generic_rw, multiple_generic = false;
 317         _cleanup_(sd_device_unrefp) sd_device *d = NULL;
 318         _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
 319         _cleanup_(blkid_free_probep) blkid_probe b = NULL;
 320         _cleanup_free_ char *generic_node = NULL;
 321         sd_id128_t generic_uuid = SD_ID128_NULL;
 322         const char *pttype = NULL;
 323         blkid_partlist pl;
 324         int r, generic_nr;
 325         struct stat st;
 326         sd_device *q;
 327         unsigned i;
 328
 329         assert(fd >= 0);
 330         assert(ret);
 331         assert(root_hash || root_hash_size == 0);
 332         assert(!((flags & DISSECT_IMAGE_GPT_ONLY) && (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)));
 333
 334         /* Probes a disk image, and returns information about what it found in *ret.
 335          *
 336          * Returns -ENOPKG if no suitable partition table or file system could be found.
 337          * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found. */
 338
 339         if (root_hash) {
 340                 /* If a root hash is supplied, then we use the root partition that has a UUID that match the first
 341                  * 128bit of the root hash. And we use the verity partition that has a UUID that match the final
 342                  * 128bit. */
 343
 344                 if (root_hash_size < sizeof(sd_id128_t))
 345                         return -EINVAL;
 346
 347                 memcpy(&root_uuid, root_hash, sizeof(sd_id128_t));
 348                 memcpy(&verity_uuid, (const uint8_t*) root_hash + root_hash_size - sizeof(sd_id128_t), sizeof(sd_id128_t));
 349
 350                 if (sd_id128_is_null(root_uuid))
 351                         return -EINVAL;
 352                 if (sd_id128_is_null(verity_uuid))
 353                         return -EINVAL;
 354         }
 355
 356         if (fstat(fd, &st) < 0)
 357                 return -errno;
 358
 359         if (!S_ISBLK(st.st_mode))
 360                 return -ENOTBLK;
 361
 362         b = blkid_new_probe();
 363         if (!b)
 364                 return -ENOMEM;
 365
 366         errno = 0;
 367         r = blkid_probe_set_device(b, fd, 0, 0);
 368         if (r != 0)
 369                 return errno_or_else(ENOMEM);
 370
 371         if ((flags & DISSECT_IMAGE_GPT_ONLY) == 0) {
 372                 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
 373                 blkid_probe_enable_superblocks(b, 1);
 374                 blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE|BLKID_SUBLKS_USAGE);
 375         }
 376
 377         blkid_probe_enable_partitions(b, 1);
 378         blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS);
 379
 380         errno = 0;
 381         r = blkid_do_safeprobe(b);
 382         if (IN_SET(r, -2, 1))
 383                 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG), "Failed to identify any partition table.");
 384         if (r != 0)
 385                 return errno_or_else(EIO);
 386
 387         m = new0(DissectedImage, 1);
 388         if (!m)
 389                 return -ENOMEM;
 390
 391         r = sd_device_new_from_devnum(&d, 'b', st.st_rdev);
 392         if (r < 0)
 393                 return r;
 394
 395         if ((!(flags & DISSECT_IMAGE_GPT_ONLY) &&
 396             (flags & DISSECT_IMAGE_REQUIRE_ROOT)) ||
 397             (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)) {
 398                 const char *usage = NULL;
 399
 400                 (void) blkid_probe_lookup_value(b, "USAGE", &usage, NULL);
 401                 if (STRPTR_IN_SET(usage, "filesystem", "crypto")) {
 402                         _cleanup_free_ char *t = NULL, *n = NULL, *o = NULL;
 403                         const char *fstype = NULL, *options = NULL;
 404
 405                         /* OK, we have found a file system, that's our root partition then. */
 406                         (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
 407
 408                         if (fstype) {
 409                                 t = strdup(fstype);
 410                                 if (!t)
 411                                         return -ENOMEM;
 412                         }
 413
 414                         r = device_path_make_major_minor(st.st_mode, st.st_rdev, &n);
 415                         if (r < 0)
 416                                 return r;
 417
 418                         m->single_file_system = true;
 419                         m->verity = root_hash && verity_data;
 420                         m->can_verity = !!verity_data;
 421
 422                         options = mount_options_from_part(mount_options, 0);
 423                         if (options) {
 424                                 o = strdup(options);
 425                                 if (!o)
 426                                         return -ENOMEM;
 427                         }
 428
 429                         m->partitions[PARTITION_ROOT] = (DissectedPartition) {
 430                                 .found = true,
 431                                 .rw = !m->verity,
 432                                 .partno = -1,
 433                                 .architecture = _ARCHITECTURE_INVALID,
 434                                 .fstype = TAKE_PTR(t),
 435                                 .node = TAKE_PTR(n),
 436                                 .mount_options = TAKE_PTR(o),
 437                         };
 438
 439                         m->encrypted = streq_ptr(fstype, "crypto_LUKS");
 440
 441                         /* Even on a single partition we need to wait for udev to create the
 442                          * /dev/block/X:Y symlink to /dev/loopZ */
 443                         r = loop_wait_for_partitions_to_appear(fd, d, 0, flags, &e);
 444                         if (r < 0)
 445                                 return r;
 446                         *ret = TAKE_PTR(m);
 447
 448                         return 0;
 449                 }
 450         }
 451
 452         (void) blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL);
 453         if (!pttype)
 454                 return -ENOPKG;
 455
 456         is_gpt = streq_ptr(pttype, "gpt");
 457         is_mbr = streq_ptr(pttype, "dos");
 458
 459         if (!is_gpt && ((flags & DISSECT_IMAGE_GPT_ONLY) || !is_mbr))
 460                 return -ENOPKG;
 461
 462         errno = 0;
 463         pl = blkid_probe_get_partitions(b);
 464         if (!pl)
 465                 return errno_or_else(ENOMEM);
 466
 467         r = loop_wait_for_partitions_to_appear(fd, d, blkid_partlist_numof_partitions(pl), flags, &e);
 468         if (r < 0)
 469                 return r;
 470
 471         FOREACH_DEVICE(e, q) {
 472                 unsigned long long pflags;
 473                 blkid_partition pp;
 474                 const char *node;
 475                 dev_t qn;
 476                 int nr;
 477
 478                 r = sd_device_get_devnum(q, &qn);
 479                 if (r < 0)
 480                         continue;
 481
 482                 if (st.st_rdev == qn)
 483                         continue;
 484
 485                 if (!device_is_block(q))
 486                         continue;
 487
 488                 if (device_is_mmc_special_partition(q))
 489                         continue;
 490
 491                 r = sd_device_get_devname(q, &node);
 492                 if (r < 0)
 493                         continue;
 494
 495                 pp = blkid_partlist_devno_to_partition(pl, qn);
 496                 if (!pp)
 497                         continue;
 498
 499                 pflags = blkid_partition_get_flags(pp);
 500
 501                 nr = blkid_partition_get_partno(pp);
 502                 if (nr < 0)
 503                         continue;
 504
 505                 if (is_gpt) {
 506                         int designator = _PARTITION_DESIGNATOR_INVALID, architecture = _ARCHITECTURE_INVALID;
 507                         const char *stype, *sid, *fstype = NULL;
 508                         sd_id128_t type_id, id;
 509                         bool rw = true;
 510
 511                         sid = blkid_partition_get_uuid(pp);
 512                         if (!sid)
 513                                 continue;
 514                         if (sd_id128_from_string(sid, &id) < 0)
 515                                 continue;
 516
 517                         stype = blkid_partition_get_type_string(pp);
 518                         if (!stype)
 519                                 continue;
 520                         if (sd_id128_from_string(stype, &type_id) < 0)
 521                                 continue;
 522
 523                         if (sd_id128_equal(type_id, GPT_HOME)) {
 524
 525                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 526
 527                                 if (pflags & GPT_FLAG_NO_AUTO)
 528                                         continue;
 529
 530                                 designator = PARTITION_HOME;
 531                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 532                         } else if (sd_id128_equal(type_id, GPT_SRV)) {
 533
 534                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 535
 536                                 if (pflags & GPT_FLAG_NO_AUTO)
 537                                         continue;
 538
 539                                 designator = PARTITION_SRV;
 540                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 541                         } else if (sd_id128_equal(type_id, GPT_ESP)) {
 542
 543                                 /* Note that we don't check the GPT_FLAG_NO_AUTO flag for the ESP, as it is not defined
 544                                  * there. We instead check the GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as recommended by the
 545                                  * UEFI spec (See "12.3.3 Number and Location of System Partitions"). */
 546
 547                                 if (pflags & GPT_FLAG_NO_BLOCK_IO_PROTOCOL)
 548                                         continue;
 549
 550                                 designator = PARTITION_ESP;
 551                                 fstype = "vfat";
 552
 553                         } else if (sd_id128_equal(type_id, GPT_XBOOTLDR)) {
 554
 555                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 556
 557                                 if (pflags & GPT_FLAG_NO_AUTO)
 558                                         continue;
 559
 560                                 designator = PARTITION_XBOOTLDR;
 561                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 562                         }
 563 #ifdef GPT_ROOT_NATIVE
 564                         else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE)) {
 565
 566                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 567
 568                                 if (pflags & GPT_FLAG_NO_AUTO)
 569                                         continue;
 570
 571                                 /* If a root ID is specified, ignore everything but the root id */
 572                                 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
 573                                         continue;
 574
 575                                 designator = PARTITION_ROOT;
 576                                 architecture = native_architecture();
 577                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 578                         } else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE_VERITY)) {
 579
 580                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 581
 582                                 if (pflags & GPT_FLAG_NO_AUTO)
 583                                         continue;
 584
 585                                 m->can_verity = true;
 586
 587                                 /* Ignore verity unless a root hash is specified */
 588                                 if (sd_id128_is_null(verity_uuid) || !sd_id128_equal(verity_uuid, id))
 589                                         continue;
 590
 591                                 designator = PARTITION_ROOT_VERITY;
 592                                 fstype = "DM_verity_hash";
 593                                 architecture = native_architecture();
 594                                 rw = false;
 595                         }
 596 #endif
 597 #ifdef GPT_ROOT_SECONDARY
 598                         else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY)) {
 599
 600                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 601
 602                                 if (pflags & GPT_FLAG_NO_AUTO)
 603                                         continue;
 604
 605                                 /* If a root ID is specified, ignore everything but the root id */
 606                                 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
 607                                         continue;
 608
 609                                 designator = PARTITION_ROOT_SECONDARY;
 610                                 architecture = SECONDARY_ARCHITECTURE;
 611                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 612                         } else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY_VERITY)) {
 613
 614                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 615
 616                                 if (pflags & GPT_FLAG_NO_AUTO)
 617                                         continue;
 618
 619                                 m->can_verity = true;
 620
 621                                 /* Ignore verity unless root has is specified */
 622                                 if (sd_id128_is_null(verity_uuid) || !sd_id128_equal(verity_uuid, id))
 623                                         continue;
 624
 625                                 designator = PARTITION_ROOT_SECONDARY_VERITY;
 626                                 fstype = "DM_verity_hash";
 627                                 architecture = SECONDARY_ARCHITECTURE;
 628                                 rw = false;
 629                         }
 630 #endif
 631                         else if (sd_id128_equal(type_id, GPT_SWAP)) {
 632
 633                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO);
 634
 635                                 if (pflags & GPT_FLAG_NO_AUTO)
 636                                         continue;
 637
 638                                 designator = PARTITION_SWAP;
 639                                 fstype = "swap";
 640                         } else if (sd_id128_equal(type_id, GPT_LINUX_GENERIC)) {
 641
 642                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 643
 644                                 if (pflags & GPT_FLAG_NO_AUTO)
 645                                         continue;
 646
 647                                 if (generic_node)
 648                                         multiple_generic = true;
 649                                 else {
 650                                         generic_nr = nr;
 651                                         generic_rw = !(pflags & GPT_FLAG_READ_ONLY);
 652                                         generic_uuid = id;
 653                                         generic_node = strdup(node);
 654                                         if (!generic_node)
 655                                                 return -ENOMEM;
 656                                 }
 657
 658                         } else if (sd_id128_equal(type_id, GPT_TMP)) {
 659
 660                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 661
 662                                 if (pflags & GPT_FLAG_NO_AUTO)
 663                                         continue;
 664
 665                                 designator = PARTITION_TMP;
 666                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 667
 668                         } else if (sd_id128_equal(type_id, GPT_VAR)) {
 669
 670                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 671
 672                                 if (pflags & GPT_FLAG_NO_AUTO)
 673                                         continue;
 674
 675                                 if (!FLAGS_SET(flags, DISSECT_IMAGE_RELAX_VAR_CHECK)) {
 676                                         sd_id128_t var_uuid;
 677
 678                                         /* For /var we insist that the uuid of the partition matches the
 679                                          * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
 680                                          * ID. Why? Unlike the other partitions /var is inherently
 681                                          * installation specific, hence we need to be careful not to mount it
 682                                          * in the wrong installation. By hashing the partition UUID from
 683                                          * /etc/machine-id we can securely bind the partition to the
 684                                          * installation. */
 685
 686                                         r = sd_id128_get_machine_app_specific(GPT_VAR, &var_uuid);
 687                                         if (r < 0)
 688                                                 return r;
 689
 690                                         if (!sd_id128_equal(var_uuid, id)) {
 691                                                 log_debug("Found a /var/ partition, but its UUID didn't match our expectations, ignoring.");
 692                                                 continue;
 693                                         }
 694                                 }
 695
 696                                 designator = PARTITION_VAR;
 697                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 698                         }
 699
 700                         if (designator != _PARTITION_DESIGNATOR_INVALID) {
 701                                 _cleanup_free_ char *t = NULL, *n = NULL, *o = NULL;
 702                                 const char *options = NULL;
 703
 704                                 /* First one wins */
 705                                 if (m->partitions[designator].found)
 706                                         continue;
 707
 708                                 if (fstype) {
 709                                         t = strdup(fstype);
 710                                         if (!t)
 711                                                 return -ENOMEM;
 712                                 }
 713
 714                                 n = strdup(node);
 715                                 if (!n)
 716                                         return -ENOMEM;
 717
 718                                 options = mount_options_from_part(mount_options, nr);
 719                                 if (options) {
 720                                         o = strdup(options);
 721                                         if (!o)
 722                                                 return -ENOMEM;
 723                                 }
 724
 725                                 m->partitions[designator] = (DissectedPartition) {
 726                                         .found = true,
 727                                         .partno = nr,
 728                                         .rw = rw,
 729                                         .architecture = architecture,
 730                                         .node = TAKE_PTR(n),
 731                                         .fstype = TAKE_PTR(t),
 732                                         .uuid = id,
 733                                         .mount_options = TAKE_PTR(o),
 734                                 };
 735                         }
 736
 737                 } else if (is_mbr) {
 738
 739                         switch (blkid_partition_get_type(pp)) {
 740
 741                         case 0x83: /* Linux partition */
 742
 743                                 if (pflags != 0x80) /* Bootable flag */
 744                                         continue;
 745
 746                                 if (generic_node)
 747                                         multiple_generic = true;
 748                                 else {
 749                                         generic_nr = nr;
 750                                         generic_rw = true;
 751                                         generic_node = strdup(node);
 752                                         if (!generic_node)
 753                                                 return -ENOMEM;
 754                                 }
 755
 756                                 break;
 757
 758                         case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
 759                                 _cleanup_free_ char *n = NULL, *o = NULL;
 760                                 sd_id128_t id = SD_ID128_NULL;
 761                                 const char *sid, *options = NULL;
 762
 763                                 /* First one wins */
 764                                 if (m->partitions[PARTITION_XBOOTLDR].found)
 765                                         continue;
 766
 767                                 sid = blkid_partition_get_uuid(pp);
 768                                 if (sid)
 769                                         (void) sd_id128_from_string(sid, &id);
 770
 771                                 n = strdup(node);
 772                                 if (!n)
 773                                         return -ENOMEM;
 774
 775                                 options = mount_options_from_part(mount_options, nr);
 776                                 if (options) {
 777                                         o = strdup(options);
 778                                         if (!o)
 779                                                 return -ENOMEM;
 780                                 }
 781
 782                                 m->partitions[PARTITION_XBOOTLDR] = (DissectedPartition) {
 783                                         .found = true,
 784                                         .partno = nr,
 785                                         .rw = true,
 786                                         .architecture = _ARCHITECTURE_INVALID,
 787                                         .node = TAKE_PTR(n),
 788                                         .uuid = id,
 789                                         .mount_options = TAKE_PTR(o),
 790                                 };
 791
 792                                 break;
 793                         }}
 794                 }
 795         }
 796
 797         if (!m->partitions[PARTITION_ROOT].found) {
 798                 /* No root partition found? Then let's see if ther's one for the secondary architecture. And if not
 799                  * either, then check if there's a single generic one, and use that. */
 800
 801                 if (m->partitions[PARTITION_ROOT_VERITY].found)
 802                         return -EADDRNOTAVAIL;
 803
 804                 if (m->partitions[PARTITION_ROOT_SECONDARY].found) {
 805                         m->partitions[PARTITION_ROOT] = m->partitions[PARTITION_ROOT_SECONDARY];
 806                         zero(m->partitions[PARTITION_ROOT_SECONDARY]);
 807
 808                         m->partitions[PARTITION_ROOT_VERITY] = m->partitions[PARTITION_ROOT_SECONDARY_VERITY];
 809                         zero(m->partitions[PARTITION_ROOT_SECONDARY_VERITY]);
 810
 811                 } else if (flags & DISSECT_IMAGE_REQUIRE_ROOT) {
 812                         _cleanup_free_ char *o = NULL;
 813                         const char *options = NULL;
 814
 815                         /* If the root has was set, then we won't fallback to a generic node, because the root hash
 816                          * decides */
 817                         if (root_hash)
 818                                 return -EADDRNOTAVAIL;
 819
 820                         /* If we didn't find a generic node, then we can't fix this up either */
 821                         if (!generic_node)
 822                                 return -ENXIO;
 823
 824                         /* If we didn't find a properly marked root partition, but we did find a single suitable
 825                          * generic Linux partition, then use this as root partition, if the caller asked for it. */
 826                         if (multiple_generic)
 827                                 return -ENOTUNIQ;
 828
 829                         options = mount_options_from_part(mount_options, generic_nr);
 830                         if (options) {
 831                                 o = strdup(options);
 832                                 if (!o)
 833                                         return -ENOMEM;
 834                         }
 835
 836                         m->partitions[PARTITION_ROOT] = (DissectedPartition) {
 837                                 .found = true,
 838                                 .rw = generic_rw,
 839                                 .partno = generic_nr,
 840                                 .architecture = _ARCHITECTURE_INVALID,
 841                                 .node = TAKE_PTR(generic_node),
 842                                 .uuid = generic_uuid,
 843                                 .mount_options = TAKE_PTR(o),
 844                         };
 845                 }
 846         }
 847
 848         if (root_hash) {
 849                 if (!m->partitions[PARTITION_ROOT_VERITY].found || !m->partitions[PARTITION_ROOT].found)
 850                         return -EADDRNOTAVAIL;
 851
 852                 /* If we found the primary root with the hash, then we definitely want to suppress any secondary root
 853                  * (which would be weird, after all the root hash should only be assigned to one pair of
 854                  * partitions... */
 855                 m->partitions[PARTITION_ROOT_SECONDARY].found = false;
 856                 m->partitions[PARTITION_ROOT_SECONDARY_VERITY].found = false;
 857
 858                 /* If we found a verity setup, then the root partition is necessarily read-only. */
 859                 m->partitions[PARTITION_ROOT].rw = false;
 860
 861                 m->verity = true;
 862         }
 863
 864         blkid_free_probe(b);
 865         b = NULL;
 866
 867         /* Fill in file system types if we don't know them yet. */
 868         for (i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
 869                 DissectedPartition *p = m->partitions + i;
 870
 871                 if (!p->found)
 872                         continue;
 873
 874                 if (!p->fstype && p->node) {
 875                         r = probe_filesystem(p->node, &p->fstype);
 876                         if (r < 0 && r != -EUCLEAN)
 877                                 return r;
 878                 }
 879
 880                 if (streq_ptr(p->fstype, "crypto_LUKS"))
 881                         m->encrypted = true;
 882
 883                 if (p->fstype && fstype_is_ro(p->fstype))
 884                         p->rw = false;
 885         }
 886
 887         *ret = TAKE_PTR(m);
 888
 889         return 0;
 890 #else
 891         return -EOPNOTSUPP;
 892 #endif
 893 }
 894
 895 DissectedImage* dissected_image_unref(DissectedImage *m) {
 896         unsigned i;
 897
 898         if (!m)
 899                 return NULL;
 900
 901         for (i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
 902                 free(m->partitions[i].fstype);
 903                 free(m->partitions[i].node);
 904                 free(m->partitions[i].decrypted_fstype);
 905                 free(m->partitions[i].decrypted_node);
 906                 free(m->partitions[i].mount_options);
 907         }
 908
 909         free(m->hostname);
 910         strv_free(m->machine_info);
 911         strv_free(m->os_release);
 912
 913         return mfree(m);
 914 }
 915
 916 static int is_loop_device(const char *path) {
 917         char s[SYS_BLOCK_PATH_MAX("/../loop/")];
 918         struct stat st;
 919
 920         assert(path);
 921
 922         if (stat(path, &st) < 0)
 923                 return -errno;
 924
 925         if (!S_ISBLK(st.st_mode))
 926                 return -ENOTBLK;
 927
 928         xsprintf_sys_block_path(s, "/loop/", st.st_dev);
 929         if (access(s, F_OK) < 0) {
 930                 if (errno != ENOENT)
 931                         return -errno;
 932
 933                 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
 934                 xsprintf_sys_block_path(s, "/../loop/", st.st_dev);
 935                 if (access(s, F_OK) < 0)
 936                         return errno == ENOENT ? false : -errno;
 937         }
 938
 939         return true;
 940 }
 941
 942 static int run_fsck(const char *node, const char *fstype) {
 943         int r, exit_status;
 944         pid_t pid;
 945
 946         assert(node);
 947         assert(fstype);
 948
 949         r = fsck_exists(fstype);
 950         if (r < 0) {
 951                 log_debug_errno(r, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype);
 952                 return 0;
 953         }
 954         if (r == 0) {
 955                 log_debug("Not checking partition %s, as fsck for %s does not exist.", node, fstype);
 956                 return 0;
 957         }
 958
 959         r = safe_fork("(fsck)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG|FORK_NULL_STDIO, &pid);
 960         if (r < 0)
 961                 return log_debug_errno(r, "Failed to fork off fsck: %m");
 962         if (r == 0) {
 963                 /* Child */
 964                 execl("/sbin/fsck", "/sbin/fsck", "-aT", node, NULL);
 965                 log_debug_errno(errno, "Failed to execl() fsck: %m");
 966                 _exit(FSCK_OPERATIONAL_ERROR);
 967         }
 968
 969         exit_status = wait_for_terminate_and_check("fsck", pid, 0);
 970         if (exit_status < 0)
 971                 return log_debug_errno(exit_status, "Failed to fork off /sbin/fsck: %m");
 972
 973         if ((exit_status & ~FSCK_ERROR_CORRECTED) != FSCK_SUCCESS) {
 974                 log_debug("fsck failed with exit status %i.", exit_status);
 975
 976                 if ((exit_status & (FSCK_SYSTEM_SHOULD_REBOOT|FSCK_ERRORS_LEFT_UNCORRECTED)) != 0)
 977                         return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN), "File system is corrupted, refusing.");
 978
 979                 log_debug("Ignoring fsck error.");
 980         }
 981
 982         return 0;
 983 }
 984
 985 static int mount_partition(
 986                 DissectedPartition *m,
 987                 const char *where,
 988                 const char *directory,
 989                 uid_t uid_shift,
 990                 DissectImageFlags flags) {
 991
 992         _cleanup_free_ char *chased = NULL, *options = NULL;
 993         const char *p, *node, *fstype;
 994         bool rw;
 995         int r;
 996
 997         assert(m);
 998         assert(where);
 999
1000         node = m->decrypted_node ?: m->node;
1001         fstype = m->decrypted_fstype ?: m->fstype;
1002
1003         if (!m->found || !node || !fstype)
1004                 return 0;
1005
1006         /* Stacked encryption? Yuck */
1007         if (streq_ptr(fstype, "crypto_LUKS"))
1008                 return -ELOOP;
1009
1010         rw = m->rw && !(flags & DISSECT_IMAGE_READ_ONLY);
1011
1012         if (FLAGS_SET(flags, DISSECT_IMAGE_FSCK) && rw) {
1013                 r = run_fsck(node, fstype);
1014                 if (r < 0)
1015                         return r;
1016         }
1017
1018         if (directory) {
1019                 r = chase_symlinks(directory, where, CHASE_PREFIX_ROOT, &chased, NULL);
1020                 if (r < 0)
1021                         return r;
1022
1023                 p = chased;
1024         } else
1025                 p = where;
1026
1027         /* If requested, turn on discard support. */
1028         if (fstype_can_discard(fstype) &&
1029             ((flags & DISSECT_IMAGE_DISCARD) ||
1030              ((flags & DISSECT_IMAGE_DISCARD_ON_LOOP) && is_loop_device(m->node)))) {
1031                 options = strdup("discard");
1032                 if (!options)
1033                         return -ENOMEM;
1034         }
1035
1036         if (uid_is_valid(uid_shift) && uid_shift != 0 && fstype_can_uid_gid(fstype)) {
1037                 _cleanup_free_ char *uid_option = NULL;
1038
1039                 if (asprintf(&uid_option, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0)
1040                         return -ENOMEM;
1041
1042                 if (!strextend_with_separator(&options, ",", uid_option, NULL))
1043                         return -ENOMEM;
1044         }
1045
1046         if (!isempty(m->mount_options))
1047                 if (!strextend_with_separator(&options, ",", m->mount_options, NULL))
1048                         return -ENOMEM;
1049
1050         r = mount_verbose(LOG_DEBUG, node, p, fstype, MS_NODEV|(rw ? 0 : MS_RDONLY), options);
1051         if (r < 0)
1052                 return r;
1053
1054         return 1;
1055 }
1056
1057 int dissected_image_mount(DissectedImage *m, const char *where, uid_t uid_shift, DissectImageFlags flags) {
1058         int r, boot_mounted;
1059
1060         assert(m);
1061         assert(where);
1062
1063         if (!m->partitions[PARTITION_ROOT].found)
1064                 return -ENXIO;
1065
1066         if ((flags & DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY) == 0) {
1067                 r = mount_partition(m->partitions + PARTITION_ROOT, where, NULL, uid_shift, flags);
1068                 if (r < 0)
1069                         return r;
1070
1071                 if (flags & DISSECT_IMAGE_VALIDATE_OS) {
1072                         r = path_is_os_tree(where);
1073                         if (r < 0)
1074                                 return r;
1075                         if (r == 0)
1076                                 return -EMEDIUMTYPE;
1077                 }
1078         }
1079
1080         if (flags & DISSECT_IMAGE_MOUNT_ROOT_ONLY)
1081                 return 0;
1082
1083         r = mount_partition(m->partitions + PARTITION_HOME, where, "/home", uid_shift, flags);
1084         if (r < 0)
1085                 return r;
1086
1087         r = mount_partition(m->partitions + PARTITION_SRV, where, "/srv", uid_shift, flags);
1088         if (r < 0)
1089                 return r;
1090
1091         r = mount_partition(m->partitions + PARTITION_VAR, where, "/var", uid_shift, flags);
1092         if (r < 0)
1093                 return r;
1094
1095         r = mount_partition(m->partitions + PARTITION_TMP, where, "/var/tmp", uid_shift, flags);
1096         if (r < 0)
1097                 return r;
1098
1099         boot_mounted = mount_partition(m->partitions + PARTITION_XBOOTLDR, where, "/boot", uid_shift, flags);
1100         if (boot_mounted < 0)
1101                 return boot_mounted;
1102
1103         if (m->partitions[PARTITION_ESP].found) {
1104                 /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it
1105                  * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */
1106
1107                 r = chase_symlinks("/efi", where, CHASE_PREFIX_ROOT, NULL, NULL);
1108                 if (r >= 0) {
1109                         r = mount_partition(m->partitions + PARTITION_ESP, where, "/efi", uid_shift, flags);
1110                         if (r < 0)
1111                                 return r;
1112
1113                 } else if (boot_mounted <= 0) {
1114                         _cleanup_free_ char *p = NULL;
1115
1116                         r = chase_symlinks("/boot", where, CHASE_PREFIX_ROOT, &p, NULL);
1117                         if (r >= 0 && dir_is_empty(p) > 0) {
1118                                 r = mount_partition(m->partitions + PARTITION_ESP, where, "/boot", uid_shift, flags);
1119                                 if (r < 0)
1120                                         return r;
1121                         }
1122                 }
1123         }
1124
1125         return 0;
1126 }
1127
1128 #if HAVE_LIBCRYPTSETUP
1129 typedef struct DecryptedPartition {
1130         struct crypt_device *device;
1131         char *name;
1132         bool relinquished;
1133 } DecryptedPartition;
1134
1135 struct DecryptedImage {
1136         DecryptedPartition *decrypted;
1137         size_t n_decrypted;
1138         size_t n_allocated;
1139 };
1140 #endif
1141
1142 DecryptedImage* decrypted_image_unref(DecryptedImage* d) {
1143 #if HAVE_LIBCRYPTSETUP
1144         size_t i;
1145         int r;
1146
1147         if (!d)
1148                 return NULL;
1149
1150         for (i = 0; i < d->n_decrypted; i++) {
1151                 DecryptedPartition *p = d->decrypted + i;
1152
1153                 if (p->device && p->name && !p->relinquished) {
1154                         r = crypt_deactivate(p->device, p->name);
1155                         if (r < 0)
1156                                 log_debug_errno(r, "Failed to deactivate encrypted partition %s", p->name);
1157                 }
1158
1159                 if (p->device)
1160                         crypt_free(p->device);
1161                 free(p->name);
1162         }
1163
1164         free(d);
1165 #endif
1166         return NULL;
1167 }
1168
1169 #if HAVE_LIBCRYPTSETUP
1170
1171 static int make_dm_name_and_node(const void *original_node, const char *suffix, char **ret_name, char **ret_node) {
1172         _cleanup_free_ char *name = NULL, *node = NULL;
1173         const char *base;
1174
1175         assert(original_node);
1176         assert(suffix);
1177         assert(ret_name);
1178         assert(ret_node);
1179
1180         base = strrchr(original_node, '/');
1181         if (!base)
1182                 base = original_node;
1183         else
1184                 base++;
1185         if (isempty(base))
1186                 return -EINVAL;
1187
1188         name = strjoin(base, suffix);
1189         if (!name)
1190                 return -ENOMEM;
1191         if (!filename_is_valid(name))
1192                 return -EINVAL;
1193
1194         node = path_join(crypt_get_dir(), name);
1195         if (!node)
1196                 return -ENOMEM;
1197
1198         *ret_name = TAKE_PTR(name);
1199         *ret_node = TAKE_PTR(node);
1200
1201         return 0;
1202 }
1203
1204 static int decrypt_partition(
1205                 DissectedPartition *m,
1206                 const char *passphrase,
1207                 DissectImageFlags flags,
1208                 DecryptedImage *d) {
1209
1210         _cleanup_free_ char *node = NULL, *name = NULL;
1211         _cleanup_(crypt_freep) struct crypt_device *cd = NULL;
1212         int r;
1213
1214         assert(m);
1215         assert(d);
1216
1217         if (!m->found || !m->node || !m->fstype)
1218                 return 0;
1219
1220         if (!streq(m->fstype, "crypto_LUKS"))
1221                 return 0;
1222
1223         if (!passphrase)
1224                 return -ENOKEY;
1225
1226         r = make_dm_name_and_node(m->node, "-decrypted", &name, &node);
1227         if (r < 0)
1228                 return r;
1229
1230         if (!GREEDY_REALLOC0(d->decrypted, d->n_allocated, d->n_decrypted + 1))
1231                 return -ENOMEM;
1232
1233         r = crypt_init(&cd, m->node);
1234         if (r < 0)
1235                 return log_debug_errno(r, "Failed to initialize dm-crypt: %m");
1236
1237         crypt_set_log_callback(cd, cryptsetup_log_glue, NULL);
1238
1239         r = crypt_load(cd, CRYPT_LUKS, NULL);
1240         if (r < 0)
1241                 return log_debug_errno(r, "Failed to load LUKS metadata: %m");
1242
1243         r = crypt_activate_by_passphrase(cd, name, CRYPT_ANY_SLOT, passphrase, strlen(passphrase),
1244                                          ((flags & DISSECT_IMAGE_READ_ONLY) ? CRYPT_ACTIVATE_READONLY : 0) |
1245                                          ((flags & DISSECT_IMAGE_DISCARD_ON_CRYPTO) ? CRYPT_ACTIVATE_ALLOW_DISCARDS : 0));
1246         if (r < 0) {
1247                 log_debug_errno(r, "Failed to activate LUKS device: %m");
1248                 return r == -EPERM ? -EKEYREJECTED : r;
1249         }
1250
1251         d->decrypted[d->n_decrypted].name = TAKE_PTR(name);
1252         d->decrypted[d->n_decrypted].device = TAKE_PTR(cd);
1253         d->n_decrypted++;
1254
1255         m->decrypted_node = TAKE_PTR(node);
1256
1257         return 0;
1258 }
1259
1260 static int verity_can_reuse(const void *root_hash, size_t root_hash_size, bool has_sig, const char *name, struct crypt_device **ret_cd) {
1261         /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
1262         _cleanup_free_ char *root_hash_existing = NULL;
1263         _cleanup_(crypt_freep) struct crypt_device *cd = NULL;
1264         struct crypt_params_verity crypt_params = {};
1265         size_t root_hash_existing_size = root_hash_size;
1266         int r;
1267
1268         assert(ret_cd);
1269
1270         r = crypt_init_by_name(&cd, name);
1271         if (r < 0)
1272                 return log_debug_errno(r, "Error opening verity device, crypt_init_by_name failed: %m");
1273         r = crypt_get_verity_info(cd, &crypt_params);
1274         if (r < 0)
1275                 return log_debug_errno(r, "Error opening verity device, crypt_get_verity_info failed: %m");
1276         root_hash_existing = malloc0(root_hash_size);
1277         if (!root_hash_existing)
1278                 return -ENOMEM;
1279         r = crypt_volume_key_get(cd, CRYPT_ANY_SLOT, root_hash_existing, &root_hash_existing_size, NULL, 0);
1280         if (r < 0)
1281                 return log_debug_errno(r, "Error opening verity device, crypt_volume_key_get failed: %m");
1282         if (root_hash_size != root_hash_existing_size || memcmp(root_hash_existing, root_hash, root_hash_size) != 0)
1283                 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but root hashes are different.");
1284 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1285         /* Ensure that, if signatures are supported, we only reuse the device if the previous mount
1286          * used the same settings, so that a previous unsigned mount will not be reused if the user
1287          * asks to use signing for the new one, and viceversa. */
1288         if (has_sig != !!(crypt_params.flags & CRYPT_VERITY_ROOT_HASH_SIGNATURE))
1289                 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but signature settings are not the same.");
1290 #endif
1291
1292         *ret_cd = TAKE_PTR(cd);
1293         return 0;
1294 }
1295
1296 static inline void dm_deferred_remove_clean(char *name) {
1297         if (!name)
1298                 return;
1299         (void) crypt_deactivate_by_name(NULL, name, CRYPT_DEACTIVATE_DEFERRED);
1300         free(name);
1301 }
1302 DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean);
1303
1304 static int verity_partition(
1305                 DissectedPartition *m,
1306                 DissectedPartition *v,
1307                 const void *root_hash,
1308                 size_t root_hash_size,
1309                 const char *verity_data,
1310                 const char *root_hash_sig_path,
1311                 const void *root_hash_sig,
1312                 size_t root_hash_sig_size,
1313                 DissectImageFlags flags,
1314                 DecryptedImage *d) {
1315
1316         _cleanup_free_ char *node = NULL, *name = NULL, *hash_sig_from_file = NULL;
1317         _cleanup_(crypt_freep) struct crypt_device *cd = NULL;
1318         _cleanup_(dm_deferred_remove_cleanp) char *restore_deferred_remove = NULL;
1319         int r;
1320
1321         assert(m);
1322         assert(v || verity_data);
1323
1324         if (!root_hash)
1325                 return 0;
1326
1327         if (!m->found || !m->node || !m->fstype)
1328                 return 0;
1329         if (!verity_data) {
1330                 if (!v->found || !v->node || !v->fstype)
1331                         return 0;
1332
1333                 if (!streq(v->fstype, "DM_verity_hash"))
1334                         return 0;
1335         }
1336
1337         if (FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) {
1338                 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
1339                 _cleanup_free_ char *root_hash_encoded = NULL;
1340                 root_hash_encoded = hexmem(root_hash, root_hash_size);
1341                 if (!root_hash_encoded)
1342                         return -ENOMEM;
1343                 r = make_dm_name_and_node(root_hash_encoded, "-verity", &name, &node);
1344         } else
1345                 r = make_dm_name_and_node(m->node, "-verity", &name, &node);
1346         if (r < 0)
1347                 return r;
1348
1349         if (!root_hash_sig && root_hash_sig_path) {
1350                 r = read_full_file_full(AT_FDCWD, root_hash_sig_path, 0, &hash_sig_from_file, &root_hash_sig_size);
1351                 if (r < 0)
1352                         return r;
1353         }
1354
1355         r = crypt_init(&cd, verity_data ?: v->node);
1356         if (r < 0)
1357                 return r;
1358
1359         crypt_set_log_callback(cd, cryptsetup_log_glue, NULL);
1360
1361         r = crypt_load(cd, CRYPT_VERITY, NULL);
1362         if (r < 0)
1363                 return r;
1364
1365         r = crypt_set_data_device(cd, m->node);
1366         if (r < 0)
1367                 return r;
1368
1369         if (!GREEDY_REALLOC0(d->decrypted, d->n_allocated, d->n_decrypted + 1))
1370                 return -ENOMEM;
1371
1372         /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
1373          * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
1374          * retry a few times before giving up. */
1375         for (unsigned i = 0; i < N_DEVICE_NODE_LIST_ATTEMPTS; i++) {
1376                 if (root_hash_sig || hash_sig_from_file) {
1377 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1378                         r = crypt_activate_by_signed_key(cd, name, root_hash, root_hash_size, root_hash_sig ?: hash_sig_from_file, root_hash_sig_size, CRYPT_ACTIVATE_READONLY);
1379 #else
1380                         r = log_debug_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "activation of verity device with signature requested, but not supported by cryptsetup due to missing crypt_activate_by_signed_key()");
1381 #endif
1382                 } else
1383                         r = crypt_activate_by_volume_key(cd, name, root_hash, root_hash_size, CRYPT_ACTIVATE_READONLY);
1384                 /* libdevmapper can return EINVAL when the device is already in the activation stage.
1385                  * There's no way to distinguish this situation from a genuine error due to invalid
1386                  * parameters, so immediately fallback to activating the device with a unique name.
1387                  * Improvements in libcrypsetup can ensure this never happens: https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
1388                 if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
1389                         return verity_partition(m, v, root_hash, root_hash_size, verity_data, NULL, root_hash_sig ?: hash_sig_from_file, root_hash_sig_size, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
1390                 if (!IN_SET(r, 0, -EEXIST, -ENODEV))
1391                         return r;
1392                 if (r == -EEXIST) {
1393                         struct crypt_device *existing_cd = NULL;
1394
1395                         if (!restore_deferred_remove){
1396                                 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
1397                                 r = dm_deferred_remove_cancel(name);
1398                                 if (r < 0)
1399                                         return log_debug_errno(r, "Disabling automated deferred removal for verity device %s failed: %m", node);
1400                                 restore_deferred_remove = strdup(name);
1401                                 if (!restore_deferred_remove)
1402                                         return -ENOMEM;
1403                         }
1404
1405                         r = verity_can_reuse(root_hash, root_hash_size, !!root_hash_sig || !!hash_sig_from_file, name, &existing_cd);
1406                         /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
1407                         if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
1408                                 return verity_partition(m, v, root_hash, root_hash_size, verity_data, NULL, root_hash_sig ?: hash_sig_from_file, root_hash_sig_size, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
1409                         if (!IN_SET(r, 0, -ENODEV, -ENOENT))
1410                                 return log_debug_errno(r, "Checking whether existing verity device %s can be reused failed: %m", node);
1411                         if (r == 0) {
1412                                 if (cd)
1413                                         crypt_free(cd);
1414                                 cd = existing_cd;
1415                         }
1416                 }
1417                 if (r == 0)
1418                         break;
1419         }
1420
1421         /* Sanity check: libdevmapper is known to report that the device already exists and is active,
1422         * but it's actually not there, so the later filesystem probe or mount would fail. */
1423         if (r == 0)
1424                 r = access(node, F_OK);
1425         /* An existing verity device was reported by libcryptsetup/libdevmapper, but we can't use it at this time.
1426          * Fall back to activating it with a unique device name. */
1427         if (r != 0 && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
1428                 return verity_partition(m, v, root_hash, root_hash_size, verity_data, NULL, root_hash_sig ?: hash_sig_from_file, root_hash_sig_size, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
1429
1430         /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
1431         restore_deferred_remove = mfree(restore_deferred_remove);
1432
1433         d->decrypted[d->n_decrypted].name = TAKE_PTR(name);
1434         d->decrypted[d->n_decrypted].device = TAKE_PTR(cd);
1435         d->n_decrypted++;
1436
1437         m->decrypted_node = TAKE_PTR(node);
1438
1439         return 0;
1440 }
1441 #endif
1442
1443 int dissected_image_decrypt(
1444                 DissectedImage *m,
1445                 const char *passphrase,
1446                 const void *root_hash,
1447                 size_t root_hash_size,
1448                 const char *verity_data,
1449                 const char *root_hash_sig_path,
1450                 const void *root_hash_sig,
1451                 size_t root_hash_sig_size,
1452                 DissectImageFlags flags,
1453                 DecryptedImage **ret) {
1454
1455 #if HAVE_LIBCRYPTSETUP
1456         _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL;
1457         unsigned i;
1458         int r;
1459 #endif
1460
1461         assert(m);
1462         assert(root_hash || root_hash_size == 0);
1463
1464         /* Returns:
1465          *
1466          *      = 0           → There was nothing to decrypt
1467          *      > 0           → Decrypted successfully
1468          *      -ENOKEY       → There's something to decrypt but no key was supplied
1469          *      -EKEYREJECTED → Passed key was not correct
1470          */
1471
1472         if (root_hash && root_hash_size < sizeof(sd_id128_t))
1473                 return -EINVAL;
1474
1475         if (!m->encrypted && !m->verity) {
1476                 *ret = NULL;
1477                 return 0;
1478         }
1479
1480 #if HAVE_LIBCRYPTSETUP
1481         d = new0(DecryptedImage, 1);
1482         if (!d)
1483                 return -ENOMEM;
1484
1485         for (i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
1486                 DissectedPartition *p = m->partitions + i;
1487                 int k;
1488
1489                 if (!p->found)
1490                         continue;
1491
1492                 r = decrypt_partition(p, passphrase, flags, d);
1493                 if (r < 0)
1494                         return r;
1495
1496                 k = PARTITION_VERITY_OF(i);
1497                 if (k >= 0) {
1498                         r = verity_partition(p, m->partitions + k, root_hash, root_hash_size, verity_data, root_hash_sig_path, root_hash_sig, root_hash_sig_size, flags | DISSECT_IMAGE_VERITY_SHARE, d);
1499                         if (r < 0)
1500                                 return r;
1501                 }
1502
1503                 if (!p->decrypted_fstype && p->decrypted_node) {
1504                         r = probe_filesystem(p->decrypted_node, &p->decrypted_fstype);
1505                         if (r < 0 && r != -EUCLEAN)
1506                                 return r;
1507                 }
1508         }
1509
1510         *ret = TAKE_PTR(d);
1511
1512         return 1;
1513 #else
1514         return -EOPNOTSUPP;
1515 #endif
1516 }
1517
1518 int dissected_image_decrypt_interactively(
1519                 DissectedImage *m,
1520                 const char *passphrase,
1521                 const void *root_hash,
1522                 size_t root_hash_size,
1523                 const char *verity_data,
1524                 const char *root_hash_sig_path,
1525                 const void *root_hash_sig,
1526                 size_t root_hash_sig_size,
1527                 DissectImageFlags flags,
1528                 DecryptedImage **ret) {
1529
1530         _cleanup_strv_free_erase_ char **z = NULL;
1531         int n = 3, r;
1532
1533         if (passphrase)
1534                 n--;
1535
1536         for (;;) {
1537                 r = dissected_image_decrypt(m, passphrase, root_hash, root_hash_size, verity_data, root_hash_sig_path, root_hash_sig, root_hash_sig_size, flags, ret);
1538                 if (r >= 0)
1539                         return r;
1540                 if (r == -EKEYREJECTED)
1541                         log_error_errno(r, "Incorrect passphrase, try again!");
1542                 else if (r != -ENOKEY)
1543                         return log_error_errno(r, "Failed to decrypt image: %m");
1544
1545                 if (--n < 0)
1546                         return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED),
1547                                                "Too many retries.");
1548
1549                 z = strv_free(z);
1550
1551                 r = ask_password_auto("Please enter image passphrase:", NULL, "dissect", "dissect", USEC_INFINITY, 0, &z);
1552                 if (r < 0)
1553                         return log_error_errno(r, "Failed to query for passphrase: %m");
1554
1555                 passphrase = z[0];
1556         }
1557 }
1558
1559 int decrypted_image_relinquish(DecryptedImage *d) {
1560
1561 #if HAVE_LIBCRYPTSETUP
1562         size_t i;
1563         int r;
1564 #endif
1565
1566         assert(d);
1567
1568         /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a boolean so
1569          * that we don't clean it up ourselves either anymore */
1570
1571 #if HAVE_LIBCRYPTSETUP
1572         for (i = 0; i < d->n_decrypted; i++) {
1573                 DecryptedPartition *p = d->decrypted + i;
1574
1575                 if (p->relinquished)
1576                         continue;
1577
1578                 r = crypt_deactivate_by_name(NULL, p->name, CRYPT_DEACTIVATE_DEFERRED);
1579                 if (r < 0)
1580                         return log_debug_errno(r, "Failed to mark %s for auto-removal: %m", p->name);
1581
1582                 p->relinquished = true;
1583         }
1584 #endif
1585
1586         return 0;
1587 }
1588
1589 int verity_metadata_load(const char *image, const char *root_hash_path, void **ret_roothash, size_t *ret_roothash_size, char **ret_verity_data, char **ret_roothashsig) {
1590         _cleanup_free_ char *verity_filename = NULL, *roothashsig_filename = NULL;
1591         _cleanup_free_ void *roothash_decoded = NULL;
1592         size_t roothash_decoded_size = 0;
1593         int r;
1594
1595         assert(image);
1596
1597         if (is_device_path(image)) {
1598                 /* If we are asked to load the root hash for a device node, exit early */
1599                 if (ret_roothash)
1600                         *ret_roothash = NULL;
1601                 if (ret_roothash_size)
1602                         *ret_roothash_size = 0;
1603                 if (ret_verity_data)
1604                         *ret_verity_data = NULL;
1605                 if (ret_roothashsig)
1606                         *ret_roothashsig = NULL;
1607                 return 0;
1608         }
1609
1610         if (ret_verity_data) {
1611                 char *e;
1612
1613                 verity_filename = new(char, strlen(image) + STRLEN(".verity") + 1);
1614                 if (!verity_filename)
1615                         return -ENOMEM;
1616                 strcpy(verity_filename, image);
1617                 e = endswith(verity_filename, ".raw");
1618                 if (e)
1619                         strcpy(e, ".verity");
1620                 else
1621                         strcat(verity_filename, ".verity");
1622
1623                 r = access(verity_filename, F_OK);
1624                 if (r < 0) {
1625                         if (errno != ENOENT)
1626                                 return -errno;
1627                         verity_filename = mfree(verity_filename);
1628                 }
1629         }
1630
1631         if (ret_roothashsig) {
1632                 char *e;
1633
1634                 /* Follow naming convention recommended by the relevant RFC:
1635                  * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
1636                 roothashsig_filename = new(char, strlen(image) + STRLEN(".roothash.p7s") + 1);
1637                 if (!roothashsig_filename)
1638                         return -ENOMEM;
1639                 strcpy(roothashsig_filename, image);
1640                 e = endswith(roothashsig_filename, ".raw");
1641                 if (e)
1642                         strcpy(e, ".roothash.p7s");
1643                 else
1644                         strcat(roothashsig_filename, ".roothash.p7s");
1645
1646                 r = access(roothashsig_filename, R_OK);
1647                 if (r < 0) {
1648                         if (errno != ENOENT)
1649                                 return -errno;
1650                         roothashsig_filename = mfree(roothashsig_filename);
1651                 }
1652         }
1653
1654         if (ret_roothash) {
1655                 _cleanup_free_ char *text = NULL;
1656                 assert(ret_roothash_size);
1657
1658                 if (root_hash_path) {
1659                         /* We have the path to a roothash to load and decode, eg: RootHash=/foo/bar.roothash */
1660                         r = read_one_line_file(root_hash_path, &text);
1661                         if (r < 0)
1662                                 return r;
1663                 } else {
1664                         r = getxattr_malloc(image, "user.verity.roothash", &text, true);
1665                         if (r < 0) {
1666                                 char *fn, *e, *n;
1667
1668                                 if (!IN_SET(r, -ENODATA, -EOPNOTSUPP, -ENOENT))
1669                                         return r;
1670
1671                                 fn = newa(char, strlen(image) + STRLEN(".roothash") + 1);
1672                                 n = stpcpy(fn, image);
1673                                 e = endswith(fn, ".raw");
1674                                 if (e)
1675                                         n = e;
1676
1677                                 strcpy(n, ".roothash");
1678
1679                                 r = read_one_line_file(fn, &text);
1680                                 if (r < 0 && r != -ENOENT)
1681                                         return r;
1682                         }
1683                 }
1684
1685                 if (text) {
1686                         r = unhexmem(text, strlen(text), &roothash_decoded, &roothash_decoded_size);
1687                         if (r < 0)
1688                                 return r;
1689                         if (roothash_decoded_size < sizeof(sd_id128_t))
1690                                 return -EINVAL;
1691                 }
1692         }
1693
1694         if (ret_roothash) {
1695                 *ret_roothash = TAKE_PTR(roothash_decoded);
1696                 *ret_roothash_size = roothash_decoded_size;
1697         }
1698         if (ret_verity_data)
1699                 *ret_verity_data = TAKE_PTR(verity_filename);
1700         if (roothashsig_filename)
1701                 *ret_roothashsig = TAKE_PTR(roothashsig_filename);
1702
1703         return 1;
1704 }
1705
1706 int dissected_image_acquire_metadata(DissectedImage *m) {
1707
1708         enum {
1709                 META_HOSTNAME,
1710                 META_MACHINE_ID,
1711                 META_MACHINE_INFO,
1712                 META_OS_RELEASE,
1713                 _META_MAX,
1714         };
1715
1716         static const char *const paths[_META_MAX] = {
1717                 [META_HOSTNAME]     = "/etc/hostname\0",
1718                 [META_MACHINE_ID]   = "/etc/machine-id\0",
1719                 [META_MACHINE_INFO] = "/etc/machine-info\0",
1720                 [META_OS_RELEASE]   = "/etc/os-release\0"
1721                                       "/usr/lib/os-release\0",
1722         };
1723
1724         _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL;
1725         _cleanup_(rmdir_and_freep) char *t = NULL;
1726         _cleanup_(sigkill_waitp) pid_t child = 0;
1727         sd_id128_t machine_id = SD_ID128_NULL;
1728         _cleanup_free_ char *hostname = NULL;
1729         unsigned n_meta_initialized = 0, k;
1730         int fds[2 * _META_MAX], r;
1731
1732         BLOCK_SIGNALS(SIGCHLD);
1733
1734         assert(m);
1735
1736         for (; n_meta_initialized < _META_MAX; n_meta_initialized ++)
1737                 if (pipe2(fds + 2*n_meta_initialized, O_CLOEXEC) < 0) {
1738                         r = -errno;
1739                         goto finish;
1740                 }
1741
1742         r = mkdtemp_malloc("/tmp/dissect-XXXXXX", &t);
1743         if (r < 0)
1744                 goto finish;
1745
1746         r = safe_fork("(sd-dissect)", FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE, &child);
1747         if (r < 0)
1748                 goto finish;
1749         if (r == 0) {
1750                 r = dissected_image_mount(m, t, UID_INVALID, DISSECT_IMAGE_READ_ONLY|DISSECT_IMAGE_MOUNT_ROOT_ONLY|DISSECT_IMAGE_VALIDATE_OS);
1751                 if (r < 0) {
1752                         log_debug_errno(r, "Failed to mount dissected image: %m");
1753                         _exit(EXIT_FAILURE);
1754                 }
1755
1756                 for (k = 0; k < _META_MAX; k++) {
1757                         _cleanup_close_ int fd = -1;
1758                         const char *p;
1759
1760                         fds[2*k] = safe_close(fds[2*k]);
1761
1762                         NULSTR_FOREACH(p, paths[k]) {
1763                                 fd = chase_symlinks_and_open(p, t, CHASE_PREFIX_ROOT, O_RDONLY|O_CLOEXEC|O_NOCTTY, NULL);
1764                                 if (fd >= 0)
1765                                         break;
1766                         }
1767                         if (fd < 0) {
1768                                 log_debug_errno(fd, "Failed to read %s file of image, ignoring: %m", paths[k]);
1769                                 continue;
1770                         }
1771
1772                         r = copy_bytes(fd, fds[2*k+1], (uint64_t) -1, 0);
1773                         if (r < 0)
1774                                 _exit(EXIT_FAILURE);
1775
1776                         fds[2*k+1] = safe_close(fds[2*k+1]);
1777                 }
1778
1779                 _exit(EXIT_SUCCESS);
1780         }
1781
1782         for (k = 0; k < _META_MAX; k++) {
1783                 _cleanup_fclose_ FILE *f = NULL;
1784
1785                 fds[2*k+1] = safe_close(fds[2*k+1]);
1786
1787                 f = take_fdopen(&fds[2*k], "r");
1788                 if (!f) {
1789                         r = -errno;
1790                         goto finish;
1791                 }
1792
1793                 switch (k) {
1794
1795                 case META_HOSTNAME:
1796                         r = read_etc_hostname_stream(f, &hostname);
1797                         if (r < 0)
1798                                 log_debug_errno(r, "Failed to read /etc/hostname: %m");
1799
1800                         break;
1801
1802                 case META_MACHINE_ID: {
1803                         _cleanup_free_ char *line = NULL;
1804
1805                         r = read_line(f, LONG_LINE_MAX, &line);
1806                         if (r < 0)
1807                                 log_debug_errno(r, "Failed to read /etc/machine-id: %m");
1808                         else if (r == 33) {
1809                                 r = sd_id128_from_string(line, &machine_id);
1810                                 if (r < 0)
1811                                         log_debug_errno(r, "Image contains invalid /etc/machine-id: %s", line);
1812                         } else if (r == 0)
1813                                 log_debug("/etc/machine-id file is empty.");
1814                         else
1815                                 log_debug("/etc/machine-id has unexpected length %i.", r);
1816
1817                         break;
1818                 }
1819
1820                 case META_MACHINE_INFO:
1821                         r = load_env_file_pairs(f, "machine-info", &machine_info);
1822                         if (r < 0)
1823                                 log_debug_errno(r, "Failed to read /etc/machine-info: %m");
1824
1825                         break;
1826
1827                 case META_OS_RELEASE:
1828                         r = load_env_file_pairs(f, "os-release", &os_release);
1829                         if (r < 0)
1830                                 log_debug_errno(r, "Failed to read OS release file: %m");
1831
1832                         break;
1833                 }
1834         }
1835
1836         r = wait_for_terminate_and_check("(sd-dissect)", child, 0);
1837         child = 0;
1838         if (r < 0)
1839                 goto finish;
1840         if (r != EXIT_SUCCESS)
1841                 return -EPROTO;
1842
1843         free_and_replace(m->hostname, hostname);
1844         m->machine_id = machine_id;
1845         strv_free_and_replace(m->machine_info, machine_info);
1846         strv_free_and_replace(m->os_release, os_release);
1847
1848 finish:
1849         for (k = 0; k < n_meta_initialized; k++)
1850                 safe_close_pair(fds + 2*k);
1851
1852         return r;
1853 }
1854
1855 int dissect_image_and_warn(
1856                 int fd,
1857                 const char *name,
1858                 const void *root_hash,
1859                 size_t root_hash_size,
1860                 const char *verity_data,
1861                 const MountOptions *mount_options,
1862                 DissectImageFlags flags,
1863                 DissectedImage **ret) {
1864
1865         _cleanup_free_ char *buffer = NULL;
1866         int r;
1867
1868         if (!name) {
1869                 r = fd_get_path(fd, &buffer);
1870                 if (r < 0)
1871                         return r;
1872
1873                 name = buffer;
1874         }
1875
1876         r = dissect_image(fd, root_hash, root_hash_size, verity_data, mount_options, flags, ret);
1877
1878         switch (r) {
1879
1880         case -EOPNOTSUPP:
1881                 return log_error_errno(r, "Dissecting images is not supported, compiled without blkid support.");
1882
1883         case -ENOPKG:
1884                 return log_error_errno(r, "Couldn't identify a suitable partition table or file system in '%s'.", name);
1885
1886         case -EADDRNOTAVAIL:
1887                 return log_error_errno(r, "No root partition for specified root hash found in '%s'.", name);
1888
1889         case -ENOTUNIQ:
1890                 return log_error_errno(r, "Multiple suitable root partitions found in image '%s'.", name);
1891
1892         case -ENXIO:
1893                 return log_error_errno(r, "No suitable root partition found in image '%s'.", name);
1894
1895         case -EPROTONOSUPPORT:
1896                 return log_error_errno(r, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name);
1897
1898         default:
1899                 if (r < 0)
1900                         return log_error_errno(r, "Failed to dissect image '%s': %m", name);
1901
1902                 return r;
1903         }
1904 }
1905
1906 bool dissected_image_can_do_verity(const DissectedImage *image, unsigned partition_designator) {
1907         if (image->single_file_system)
1908                 return partition_designator == PARTITION_ROOT && image->can_verity;
1909
1910         return PARTITION_VERITY_OF(partition_designator) >= 0;
1911 }
1912
1913 bool dissected_image_has_verity(const DissectedImage *image, unsigned partition_designator) {
1914         int k;
1915
1916         if (image->single_file_system)
1917                 return partition_designator == PARTITION_ROOT && image->verity;
1918
1919         k = PARTITION_VERITY_OF(partition_designator);
1920         return k >= 0 && image->partitions[k].found;
1921 }
1922
1923 MountOptions* mount_options_free_all(MountOptions *options) {
1924         MountOptions *m;
1925
1926         while ((m = options)) {
1927                 LIST_REMOVE(mount_options, options, m);
1928                 free(m->options);
1929                 free(m);
1930         }
1931
1932         return NULL;
1933 }
1934
1935 const char* mount_options_from_part(const MountOptions *options, unsigned int partition_number) {
1936         MountOptions *m;
1937
1938         LIST_FOREACH(mount_options, m, (MountOptions *)options)
1939                 if (partition_number == m->partition_number && !isempty(m->options))
1940                         return m->options;
1941
1942         return NULL;
1943 }
1944
1945 int mount_image_privately_interactively(
1946                 const char *image,
1947                 DissectImageFlags flags,
1948                 char **ret_directory,
1949                 LoopDevice **ret_loop_device,
1950                 DecryptedImage **ret_decrypted_image) {
1951
1952         _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
1953         _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
1954         _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
1955         _cleanup_(rmdir_and_freep) char *created_dir = NULL;
1956         _cleanup_free_ char *temp = NULL;
1957         int r;
1958
1959         /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
1960          * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
1961          * easily. */
1962
1963         assert(image);
1964         assert(ret_directory);
1965         assert(ret_loop_device);
1966         assert(ret_decrypted_image);
1967
1968         r = tempfn_random_child(NULL, program_invocation_short_name, &temp);
1969         if (r < 0)
1970                 return log_error_errno(r, "Failed to generate temporary mount directory: %m");
1971
1972         r = loop_device_make_by_path(
1973                         image,
1974                         FLAGS_SET(flags, DISSECT_IMAGE_READ_ONLY) ? O_RDONLY : O_RDWR,
1975                         FLAGS_SET(flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN,
1976                         &d);
1977         if (r < 0)
1978                 return log_error_errno(r, "Failed to set up loopback device: %m");
1979
1980         r = dissect_image_and_warn(d->fd, image, NULL, 0, NULL, NULL, flags, &dissected_image);
1981         if (r < 0)
1982                 return r;
1983
1984         r = dissected_image_decrypt_interactively(dissected_image, NULL, NULL, 0, NULL, NULL, NULL, 0, flags, &decrypted_image);
1985         if (r < 0)
1986                 return r;
1987
1988         r = detach_mount_namespace();
1989         if (r < 0)
1990                 return log_error_errno(r, "Failed to detach mount namespace: %m");
1991
1992         r = mkdir_p(temp, 0700);
1993         if (r < 0)
1994                 return log_error_errno(r, "Failed to create mount point: %m");
1995
1996         created_dir = TAKE_PTR(temp);
1997
1998         r = dissected_image_mount(dissected_image, created_dir, UID_INVALID, flags);
1999         if (r == -EUCLEAN)
2000                 return log_error_errno(r, "File system check on image failed: %m");
2001         if (r < 0)
2002                 return log_error_errno(r, "Failed to mount image: %m");
2003
2004         if (decrypted_image) {
2005                 r = decrypted_image_relinquish(decrypted_image);
2006                 if (r < 0)
2007                         return log_error_errno(r, "Failed to relinquish DM devices: %m");
2008         }
2009
2010         loop_device_relinquish(d);
2011
2012         *ret_directory = TAKE_PTR(created_dir);
2013         *ret_loop_device = TAKE_PTR(d);
2014         *ret_decrypted_image = TAKE_PTR(decrypted_image);
2015
2016         return 0;
2017 }
2018
2019 static const char *const partition_designator_table[] = {
2020         [PARTITION_ROOT] = "root",
2021         [PARTITION_ROOT_SECONDARY] = "root-secondary",
2022         [PARTITION_HOME] = "home",
2023         [PARTITION_SRV] = "srv",
2024         [PARTITION_ESP] = "esp",
2025         [PARTITION_XBOOTLDR] = "xbootldr",
2026         [PARTITION_SWAP] = "swap",
2027         [PARTITION_ROOT_VERITY] = "root-verity",
2028         [PARTITION_ROOT_SECONDARY_VERITY] = "root-secondary-verity",
2029         [PARTITION_TMP] = "tmp",
2030         [PARTITION_VAR] = "var",
2031 };
2032
2033 DEFINE_STRING_TABLE_LOOKUP(partition_designator, int);