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                         struct loop_info64 info;
 219
 220                         /* If we are running on a loop device that has partition scanning off, return
 221                          * an explicit recognizable error about this, so that callers can generate a
 222                          * proper message explaining the situation. */
 223
 224                         if (ioctl(fd, LOOP_GET_STATUS64, &info) >= 0) {
 225 #if HAVE_VALGRIND_MEMCHECK_H
 226                                 /* Valgrind currently doesn't know LOOP_GET_STATUS64. Remove this once it does */
 227                                 VALGRIND_MAKE_MEM_DEFINED(&info, sizeof(info));
 228 #endif
 229
 230                                 if ((info.lo_flags & LO_FLAGS_PARTSCAN) == 0)
 231                                         return log_debug_errno(EPROTONOSUPPORT,
 232                                                                "Device is a loop device and partition scanning is off!");
 233                         }
 234                 }
 235                 if (r != -EBUSY)
 236                         return r;
 237
 238                 /* If something else has the device open, such as an udev rule, the ioctl will return
 239                  * EBUSY. Since there's no way to wait until it isn't busy anymore, let's just wait a bit,
 240                  * and try again.
 241                  *
 242                  * This is really something they should fix in the kernel! */
 243                 (void) usleep(50 * USEC_PER_MSEC);
 244
 245         }
 246
 247         return -EAGAIN; /* no success yet, try again */
 248 }
 249
 250 static int loop_wait_for_partitions_to_appear(
 251                 int fd,
 252                 sd_device *d,
 253                 unsigned num_partitions,
 254                 DissectImageFlags flags,
 255                 sd_device_enumerator **ret_enumerator) {
 256         _cleanup_(sd_device_unrefp) sd_device *device = NULL;
 257         int r;
 258
 259         assert(fd >= 0);
 260         assert(d);
 261         assert(ret_enumerator);
 262
 263         log_debug("Waiting for device (parent + %d partitions) to appear...", num_partitions);
 264
 265         if (!FLAGS_SET(flags, DISSECT_IMAGE_NO_UDEV)) {
 266                 r = device_wait_for_initialization(d, "block", USEC_INFINITY, &device);
 267                 if (r < 0)
 268                         return r;
 269         } else
 270                 device = sd_device_ref(d);
 271
 272         for (unsigned i = 0; i < N_DEVICE_NODE_LIST_ATTEMPTS; i++) {
 273                 r = wait_for_partitions_to_appear(fd, device, num_partitions, flags, ret_enumerator);
 274                 if (r != -EAGAIN)
 275                         return r;
 276         }
 277
 278         return log_debug_errno(SYNTHETIC_ERRNO(ENXIO),
 279                                "Kernel partitions dit not appear within %d attempts",
 280                                N_DEVICE_NODE_LIST_ATTEMPTS);
 281 }
 282
 283 static void check_partition_flags(
 284                 const char *node,
 285                 unsigned long long pflags,
 286                 unsigned long long supported) {
 287
 288         assert(node);
 289
 290         /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
 291         pflags &= ~(supported | GPT_FLAG_REQUIRED_PARTITION | GPT_FLAG_NO_BLOCK_IO_PROTOCOL | GPT_FLAG_LEGACY_BIOS_BOOTABLE);
 292
 293         if (pflags == 0)
 294                 return;
 295
 296         /* If there are other bits set, then log about it, to make things discoverable */
 297         for (unsigned i = 0; i < sizeof(pflags) * 8; i++) {
 298                 unsigned long long bit = 1ULL << i;
 299                 if (!FLAGS_SET(pflags, bit))
 300                         continue;
 301
 302                 log_debug("Unexpected partition flag %llu set on %s!", bit, node);
 303         }
 304 }
 305
 306 #endif
 307
 308 int dissect_image(
 309                 int fd,
 310                 const void *root_hash,
 311                 size_t root_hash_size,
 312                 const char *verity_data,
 313                 const MountOptions *mount_options,
 314                 DissectImageFlags flags,
 315                 DissectedImage **ret) {
 316
 317 #if HAVE_BLKID
 318         sd_id128_t root_uuid = SD_ID128_NULL, verity_uuid = SD_ID128_NULL;
 319         _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
 320         bool is_gpt, is_mbr, generic_rw, multiple_generic = false;
 321         _cleanup_(sd_device_unrefp) sd_device *d = NULL;
 322         _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
 323         _cleanup_(blkid_free_probep) blkid_probe b = NULL;
 324         _cleanup_free_ char *generic_node = NULL;
 325         sd_id128_t generic_uuid = SD_ID128_NULL;
 326         const char *pttype = NULL;
 327         blkid_partlist pl;
 328         int r, generic_nr;
 329         struct stat st;
 330         sd_device *q;
 331         unsigned i;
 332
 333         assert(fd >= 0);
 334         assert(ret);
 335         assert(root_hash || root_hash_size == 0);
 336         assert(!((flags & DISSECT_IMAGE_GPT_ONLY) && (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)));
 337
 338         /* Probes a disk image, and returns information about what it found in *ret.
 339          *
 340          * Returns -ENOPKG if no suitable partition table or file system could be found.
 341          * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found. */
 342
 343         if (root_hash) {
 344                 /* If a root hash is supplied, then we use the root partition that has a UUID that match the first
 345                  * 128bit of the root hash. And we use the verity partition that has a UUID that match the final
 346                  * 128bit. */
 347
 348                 if (root_hash_size < sizeof(sd_id128_t))
 349                         return -EINVAL;
 350
 351                 memcpy(&root_uuid, root_hash, sizeof(sd_id128_t));
 352                 memcpy(&verity_uuid, (const uint8_t*) root_hash + root_hash_size - sizeof(sd_id128_t), sizeof(sd_id128_t));
 353
 354                 if (sd_id128_is_null(root_uuid))
 355                         return -EINVAL;
 356                 if (sd_id128_is_null(verity_uuid))
 357                         return -EINVAL;
 358         }
 359
 360         if (fstat(fd, &st) < 0)
 361                 return -errno;
 362
 363         if (!S_ISBLK(st.st_mode))
 364                 return -ENOTBLK;
 365
 366         b = blkid_new_probe();
 367         if (!b)
 368                 return -ENOMEM;
 369
 370         errno = 0;
 371         r = blkid_probe_set_device(b, fd, 0, 0);
 372         if (r != 0)
 373                 return errno_or_else(ENOMEM);
 374
 375         if ((flags & DISSECT_IMAGE_GPT_ONLY) == 0) {
 376                 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
 377                 blkid_probe_enable_superblocks(b, 1);
 378                 blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE|BLKID_SUBLKS_USAGE);
 379         }
 380
 381         blkid_probe_enable_partitions(b, 1);
 382         blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS);
 383
 384         errno = 0;
 385         r = blkid_do_safeprobe(b);
 386         if (IN_SET(r, -2, 1))
 387                 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG), "Failed to identify any partition table.");
 388         if (r != 0)
 389                 return errno_or_else(EIO);
 390
 391         m = new0(DissectedImage, 1);
 392         if (!m)
 393                 return -ENOMEM;
 394
 395         r = sd_device_new_from_devnum(&d, 'b', st.st_rdev);
 396         if (r < 0)
 397                 return r;
 398
 399         if ((!(flags & DISSECT_IMAGE_GPT_ONLY) &&
 400             (flags & DISSECT_IMAGE_REQUIRE_ROOT)) ||
 401             (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)) {
 402                 const char *usage = NULL;
 403
 404                 (void) blkid_probe_lookup_value(b, "USAGE", &usage, NULL);
 405                 if (STRPTR_IN_SET(usage, "filesystem", "crypto")) {
 406                         _cleanup_free_ char *t = NULL, *n = NULL, *o = NULL;
 407                         const char *fstype = NULL, *options = NULL;
 408
 409                         /* OK, we have found a file system, that's our root partition then. */
 410                         (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
 411
 412                         if (fstype) {
 413                                 t = strdup(fstype);
 414                                 if (!t)
 415                                         return -ENOMEM;
 416                         }
 417
 418                         r = device_path_make_major_minor(st.st_mode, st.st_rdev, &n);
 419                         if (r < 0)
 420                                 return r;
 421
 422                         m->single_file_system = true;
 423                         m->verity = root_hash && verity_data;
 424                         m->can_verity = !!verity_data;
 425
 426                         options = mount_options_from_part(mount_options, 0);
 427                         if (options) {
 428                                 o = strdup(options);
 429                                 if (!o)
 430                                         return -ENOMEM;
 431                         }
 432
 433                         m->partitions[PARTITION_ROOT] = (DissectedPartition) {
 434                                 .found = true,
 435                                 .rw = !m->verity,
 436                                 .partno = -1,
 437                                 .architecture = _ARCHITECTURE_INVALID,
 438                                 .fstype = TAKE_PTR(t),
 439                                 .node = TAKE_PTR(n),
 440                                 .mount_options = TAKE_PTR(o),
 441                         };
 442
 443                         m->encrypted = streq_ptr(fstype, "crypto_LUKS");
 444
 445                         /* Even on a single partition we need to wait for udev to create the
 446                          * /dev/block/X:Y symlink to /dev/loopZ */
 447                         r = loop_wait_for_partitions_to_appear(fd, d, 0, flags, &e);
 448                         if (r < 0)
 449                                 return r;
 450                         *ret = TAKE_PTR(m);
 451
 452                         return 0;
 453                 }
 454         }
 455
 456         (void) blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL);
 457         if (!pttype)
 458                 return -ENOPKG;
 459
 460         is_gpt = streq_ptr(pttype, "gpt");
 461         is_mbr = streq_ptr(pttype, "dos");
 462
 463         if (!is_gpt && ((flags & DISSECT_IMAGE_GPT_ONLY) || !is_mbr))
 464                 return -ENOPKG;
 465
 466         errno = 0;
 467         pl = blkid_probe_get_partitions(b);
 468         if (!pl)
 469                 return errno_or_else(ENOMEM);
 470
 471         r = loop_wait_for_partitions_to_appear(fd, d, blkid_partlist_numof_partitions(pl), flags, &e);
 472         if (r < 0)
 473                 return r;
 474
 475         FOREACH_DEVICE(e, q) {
 476                 unsigned long long pflags;
 477                 blkid_partition pp;
 478                 const char *node;
 479                 dev_t qn;
 480                 int nr;
 481
 482                 r = sd_device_get_devnum(q, &qn);
 483                 if (r < 0)
 484                         continue;
 485
 486                 if (st.st_rdev == qn)
 487                         continue;
 488
 489                 if (!device_is_block(q))
 490                         continue;
 491
 492                 if (device_is_mmc_special_partition(q))
 493                         continue;
 494
 495                 r = sd_device_get_devname(q, &node);
 496                 if (r < 0)
 497                         continue;
 498
 499                 pp = blkid_partlist_devno_to_partition(pl, qn);
 500                 if (!pp)
 501                         continue;
 502
 503                 pflags = blkid_partition_get_flags(pp);
 504
 505                 nr = blkid_partition_get_partno(pp);
 506                 if (nr < 0)
 507                         continue;
 508
 509                 if (is_gpt) {
 510                         int designator = _PARTITION_DESIGNATOR_INVALID, architecture = _ARCHITECTURE_INVALID;
 511                         const char *stype, *sid, *fstype = NULL;
 512                         sd_id128_t type_id, id;
 513                         bool rw = true;
 514
 515                         sid = blkid_partition_get_uuid(pp);
 516                         if (!sid)
 517                                 continue;
 518                         if (sd_id128_from_string(sid, &id) < 0)
 519                                 continue;
 520
 521                         stype = blkid_partition_get_type_string(pp);
 522                         if (!stype)
 523                                 continue;
 524                         if (sd_id128_from_string(stype, &type_id) < 0)
 525                                 continue;
 526
 527                         if (sd_id128_equal(type_id, GPT_HOME)) {
 528
 529                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 530
 531                                 if (pflags & GPT_FLAG_NO_AUTO)
 532                                         continue;
 533
 534                                 designator = PARTITION_HOME;
 535                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 536                         } else if (sd_id128_equal(type_id, GPT_SRV)) {
 537
 538                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 539
 540                                 if (pflags & GPT_FLAG_NO_AUTO)
 541                                         continue;
 542
 543                                 designator = PARTITION_SRV;
 544                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 545                         } else if (sd_id128_equal(type_id, GPT_ESP)) {
 546
 547                                 /* Note that we don't check the GPT_FLAG_NO_AUTO flag for the ESP, as it is not defined
 548                                  * there. We instead check the GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as recommended by the
 549                                  * UEFI spec (See "12.3.3 Number and Location of System Partitions"). */
 550
 551                                 if (pflags & GPT_FLAG_NO_BLOCK_IO_PROTOCOL)
 552                                         continue;
 553
 554                                 designator = PARTITION_ESP;
 555                                 fstype = "vfat";
 556
 557                         } else if (sd_id128_equal(type_id, GPT_XBOOTLDR)) {
 558
 559                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 560
 561                                 if (pflags & GPT_FLAG_NO_AUTO)
 562                                         continue;
 563
 564                                 designator = PARTITION_XBOOTLDR;
 565                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 566                         }
 567 #ifdef GPT_ROOT_NATIVE
 568                         else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE)) {
 569
 570                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 571
 572                                 if (pflags & GPT_FLAG_NO_AUTO)
 573                                         continue;
 574
 575                                 /* If a root ID is specified, ignore everything but the root id */
 576                                 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
 577                                         continue;
 578
 579                                 designator = PARTITION_ROOT;
 580                                 architecture = native_architecture();
 581                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 582                         } else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE_VERITY)) {
 583
 584                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 585
 586                                 if (pflags & GPT_FLAG_NO_AUTO)
 587                                         continue;
 588
 589                                 m->can_verity = true;
 590
 591                                 /* Ignore verity unless a root hash is specified */
 592                                 if (sd_id128_is_null(verity_uuid) || !sd_id128_equal(verity_uuid, id))
 593                                         continue;
 594
 595                                 designator = PARTITION_ROOT_VERITY;
 596                                 fstype = "DM_verity_hash";
 597                                 architecture = native_architecture();
 598                                 rw = false;
 599                         }
 600 #endif
 601 #ifdef GPT_ROOT_SECONDARY
 602                         else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY)) {
 603
 604                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 605
 606                                 if (pflags & GPT_FLAG_NO_AUTO)
 607                                         continue;
 608
 609                                 /* If a root ID is specified, ignore everything but the root id */
 610                                 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
 611                                         continue;
 612
 613                                 designator = PARTITION_ROOT_SECONDARY;
 614                                 architecture = SECONDARY_ARCHITECTURE;
 615                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 616                         } else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY_VERITY)) {
 617
 618                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 619
 620                                 if (pflags & GPT_FLAG_NO_AUTO)
 621                                         continue;
 622
 623                                 m->can_verity = true;
 624
 625                                 /* Ignore verity unless root has is specified */
 626                                 if (sd_id128_is_null(verity_uuid) || !sd_id128_equal(verity_uuid, id))
 627                                         continue;
 628
 629                                 designator = PARTITION_ROOT_SECONDARY_VERITY;
 630                                 fstype = "DM_verity_hash";
 631                                 architecture = SECONDARY_ARCHITECTURE;
 632                                 rw = false;
 633                         }
 634 #endif
 635                         else if (sd_id128_equal(type_id, GPT_SWAP)) {
 636
 637                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO);
 638
 639                                 if (pflags & GPT_FLAG_NO_AUTO)
 640                                         continue;
 641
 642                                 designator = PARTITION_SWAP;
 643                                 fstype = "swap";
 644                         } else if (sd_id128_equal(type_id, GPT_LINUX_GENERIC)) {
 645
 646                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 647
 648                                 if (pflags & GPT_FLAG_NO_AUTO)
 649                                         continue;
 650
 651                                 if (generic_node)
 652                                         multiple_generic = true;
 653                                 else {
 654                                         generic_nr = nr;
 655                                         generic_rw = !(pflags & GPT_FLAG_READ_ONLY);
 656                                         generic_uuid = id;
 657                                         generic_node = strdup(node);
 658                                         if (!generic_node)
 659                                                 return -ENOMEM;
 660                                 }
 661
 662                         } else if (sd_id128_equal(type_id, GPT_TMP)) {
 663
 664                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 665
 666                                 if (pflags & GPT_FLAG_NO_AUTO)
 667                                         continue;
 668
 669                                 designator = PARTITION_TMP;
 670                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 671
 672                         } else if (sd_id128_equal(type_id, GPT_VAR)) {
 673
 674                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 675
 676                                 if (pflags & GPT_FLAG_NO_AUTO)
 677                                         continue;
 678
 679                                 if (!FLAGS_SET(flags, DISSECT_IMAGE_RELAX_VAR_CHECK)) {
 680                                         sd_id128_t var_uuid;
 681
 682                                         /* For /var we insist that the uuid of the partition matches the
 683                                          * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
 684                                          * ID. Why? Unlike the other partitions /var is inherently
 685                                          * installation specific, hence we need to be careful not to mount it
 686                                          * in the wrong installation. By hashing the partition UUID from
 687                                          * /etc/machine-id we can securely bind the partition to the
 688                                          * installation. */
 689
 690                                         r = sd_id128_get_machine_app_specific(GPT_VAR, &var_uuid);
 691                                         if (r < 0)
 692                                                 return r;
 693
 694                                         if (!sd_id128_equal(var_uuid, id)) {
 695                                                 log_debug("Found a /var/ partition, but its UUID didn't match our expectations, ignoring.");
 696                                                 continue;
 697                                         }
 698                                 }
 699
 700                                 designator = PARTITION_VAR;
 701                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 702                         }
 703
 704                         if (designator != _PARTITION_DESIGNATOR_INVALID) {
 705                                 _cleanup_free_ char *t = NULL, *n = NULL, *o = NULL;
 706                                 const char *options = NULL;
 707
 708                                 /* First one wins */
 709                                 if (m->partitions[designator].found)
 710                                         continue;
 711
 712                                 if (fstype) {
 713                                         t = strdup(fstype);
 714                                         if (!t)
 715                                                 return -ENOMEM;
 716                                 }
 717
 718                                 n = strdup(node);
 719                                 if (!n)
 720                                         return -ENOMEM;
 721
 722                                 options = mount_options_from_part(mount_options, nr);
 723                                 if (options) {
 724                                         o = strdup(options);
 725                                         if (!o)
 726                                                 return -ENOMEM;
 727                                 }
 728
 729                                 m->partitions[designator] = (DissectedPartition) {
 730                                         .found = true,
 731                                         .partno = nr,
 732                                         .rw = rw,
 733                                         .architecture = architecture,
 734                                         .node = TAKE_PTR(n),
 735                                         .fstype = TAKE_PTR(t),
 736                                         .uuid = id,
 737                                         .mount_options = TAKE_PTR(o),
 738                                 };
 739                         }
 740
 741                 } else if (is_mbr) {
 742
 743                         switch (blkid_partition_get_type(pp)) {
 744
 745                         case 0x83: /* Linux partition */
 746
 747                                 if (pflags != 0x80) /* Bootable flag */
 748                                         continue;
 749
 750                                 if (generic_node)
 751                                         multiple_generic = true;
 752                                 else {
 753                                         generic_nr = nr;
 754                                         generic_rw = true;
 755                                         generic_node = strdup(node);
 756                                         if (!generic_node)
 757                                                 return -ENOMEM;
 758                                 }
 759
 760                                 break;
 761
 762                         case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
 763                                 _cleanup_free_ char *n = NULL, *o = NULL;
 764                                 sd_id128_t id = SD_ID128_NULL;
 765                                 const char *sid, *options = NULL;
 766
 767                                 /* First one wins */
 768                                 if (m->partitions[PARTITION_XBOOTLDR].found)
 769                                         continue;
 770
 771                                 sid = blkid_partition_get_uuid(pp);
 772                                 if (sid)
 773                                         (void) sd_id128_from_string(sid, &id);
 774
 775                                 n = strdup(node);
 776                                 if (!n)
 777                                         return -ENOMEM;
 778
 779                                 options = mount_options_from_part(mount_options, nr);
 780                                 if (options) {
 781                                         o = strdup(options);
 782                                         if (!o)
 783                                                 return -ENOMEM;
 784                                 }
 785
 786                                 m->partitions[PARTITION_XBOOTLDR] = (DissectedPartition) {
 787                                         .found = true,
 788                                         .partno = nr,
 789                                         .rw = true,
 790                                         .architecture = _ARCHITECTURE_INVALID,
 791                                         .node = TAKE_PTR(n),
 792                                         .uuid = id,
 793                                         .mount_options = TAKE_PTR(o),
 794                                 };
 795
 796                                 break;
 797                         }}
 798                 }
 799         }
 800
 801         if (!m->partitions[PARTITION_ROOT].found) {
 802                 /* No root partition found? Then let's see if ther's one for the secondary architecture. And if not
 803                  * either, then check if there's a single generic one, and use that. */
 804
 805                 if (m->partitions[PARTITION_ROOT_VERITY].found)
 806                         return -EADDRNOTAVAIL;
 807
 808                 if (m->partitions[PARTITION_ROOT_SECONDARY].found) {
 809                         m->partitions[PARTITION_ROOT] = m->partitions[PARTITION_ROOT_SECONDARY];
 810                         zero(m->partitions[PARTITION_ROOT_SECONDARY]);
 811
 812                         m->partitions[PARTITION_ROOT_VERITY] = m->partitions[PARTITION_ROOT_SECONDARY_VERITY];
 813                         zero(m->partitions[PARTITION_ROOT_SECONDARY_VERITY]);
 814
 815                 } else if (flags & DISSECT_IMAGE_REQUIRE_ROOT) {
 816                         _cleanup_free_ char *o = NULL;
 817                         const char *options = NULL;
 818
 819                         /* If the root has was set, then we won't fallback to a generic node, because the root hash
 820                          * decides */
 821                         if (root_hash)
 822                                 return -EADDRNOTAVAIL;
 823
 824                         /* If we didn't find a generic node, then we can't fix this up either */
 825                         if (!generic_node)
 826                                 return -ENXIO;
 827
 828                         /* If we didn't find a properly marked root partition, but we did find a single suitable
 829                          * generic Linux partition, then use this as root partition, if the caller asked for it. */
 830                         if (multiple_generic)
 831                                 return -ENOTUNIQ;
 832
 833                         options = mount_options_from_part(mount_options, generic_nr);
 834                         if (options) {
 835                                 o = strdup(options);
 836                                 if (!o)
 837                                         return -ENOMEM;
 838                         }
 839
 840                         m->partitions[PARTITION_ROOT] = (DissectedPartition) {
 841                                 .found = true,
 842                                 .rw = generic_rw,
 843                                 .partno = generic_nr,
 844                                 .architecture = _ARCHITECTURE_INVALID,
 845                                 .node = TAKE_PTR(generic_node),
 846                                 .uuid = generic_uuid,
 847                                 .mount_options = TAKE_PTR(o),
 848                         };
 849                 }
 850         }
 851
 852         if (root_hash) {
 853                 if (!m->partitions[PARTITION_ROOT_VERITY].found || !m->partitions[PARTITION_ROOT].found)
 854                         return -EADDRNOTAVAIL;
 855
 856                 /* If we found the primary root with the hash, then we definitely want to suppress any secondary root
 857                  * (which would be weird, after all the root hash should only be assigned to one pair of
 858                  * partitions... */
 859                 m->partitions[PARTITION_ROOT_SECONDARY].found = false;
 860                 m->partitions[PARTITION_ROOT_SECONDARY_VERITY].found = false;
 861
 862                 /* If we found a verity setup, then the root partition is necessarily read-only. */
 863                 m->partitions[PARTITION_ROOT].rw = false;
 864
 865                 m->verity = true;
 866         }
 867
 868         blkid_free_probe(b);
 869         b = NULL;
 870
 871         /* Fill in file system types if we don't know them yet. */
 872         for (i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
 873                 DissectedPartition *p = m->partitions + i;
 874
 875                 if (!p->found)
 876                         continue;
 877
 878                 if (!p->fstype && p->node) {
 879                         r = probe_filesystem(p->node, &p->fstype);
 880                         if (r < 0 && r != -EUCLEAN)
 881                                 return r;
 882                 }
 883
 884                 if (streq_ptr(p->fstype, "crypto_LUKS"))
 885                         m->encrypted = true;
 886
 887                 if (p->fstype && fstype_is_ro(p->fstype))
 888                         p->rw = false;
 889         }
 890
 891         *ret = TAKE_PTR(m);
 892
 893         return 0;
 894 #else
 895         return -EOPNOTSUPP;
 896 #endif
 897 }
 898
 899 DissectedImage* dissected_image_unref(DissectedImage *m) {
 900         unsigned i;
 901
 902         if (!m)
 903                 return NULL;
 904
 905         for (i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
 906                 free(m->partitions[i].fstype);
 907                 free(m->partitions[i].node);
 908                 free(m->partitions[i].decrypted_fstype);
 909                 free(m->partitions[i].decrypted_node);
 910                 free(m->partitions[i].mount_options);
 911         }
 912
 913         free(m->hostname);
 914         strv_free(m->machine_info);
 915         strv_free(m->os_release);
 916
 917         return mfree(m);
 918 }
 919
 920 static int is_loop_device(const char *path) {
 921         char s[SYS_BLOCK_PATH_MAX("/../loop/")];
 922         struct stat st;
 923
 924         assert(path);
 925
 926         if (stat(path, &st) < 0)
 927                 return -errno;
 928
 929         if (!S_ISBLK(st.st_mode))
 930                 return -ENOTBLK;
 931
 932         xsprintf_sys_block_path(s, "/loop/", st.st_dev);
 933         if (access(s, F_OK) < 0) {
 934                 if (errno != ENOENT)
 935                         return -errno;
 936
 937                 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
 938                 xsprintf_sys_block_path(s, "/../loop/", st.st_dev);
 939                 if (access(s, F_OK) < 0)
 940                         return errno == ENOENT ? false : -errno;
 941         }
 942
 943         return true;
 944 }
 945
 946 static int run_fsck(const char *node, const char *fstype) {
 947         int r, exit_status;
 948         pid_t pid;
 949
 950         assert(node);
 951         assert(fstype);
 952
 953         r = fsck_exists(fstype);
 954         if (r < 0) {
 955                 log_debug_errno(r, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype);
 956                 return 0;
 957         }
 958         if (r == 0) {
 959                 log_debug("Not checking partition %s, as fsck for %s does not exist.", node, fstype);
 960                 return 0;
 961         }
 962
 963         r = safe_fork("(fsck)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG|FORK_NULL_STDIO, &pid);
 964         if (r < 0)
 965                 return log_debug_errno(r, "Failed to fork off fsck: %m");
 966         if (r == 0) {
 967                 /* Child */
 968                 execl("/sbin/fsck", "/sbin/fsck", "-aT", node, NULL);
 969                 log_debug_errno(errno, "Failed to execl() fsck: %m");
 970                 _exit(FSCK_OPERATIONAL_ERROR);
 971         }
 972
 973         exit_status = wait_for_terminate_and_check("fsck", pid, 0);
 974         if (exit_status < 0)
 975                 return log_debug_errno(exit_status, "Failed to fork off /sbin/fsck: %m");
 976
 977         if ((exit_status & ~FSCK_ERROR_CORRECTED) != FSCK_SUCCESS) {
 978                 log_debug("fsck failed with exit status %i.", exit_status);
 979
 980                 if ((exit_status & (FSCK_SYSTEM_SHOULD_REBOOT|FSCK_ERRORS_LEFT_UNCORRECTED)) != 0)
 981                         return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN), "File system is corrupted, refusing.");
 982
 983                 log_debug("Ignoring fsck error.");
 984         }
 985
 986         return 0;
 987 }
 988
 989 static int mount_partition(
 990                 DissectedPartition *m,
 991                 const char *where,
 992                 const char *directory,
 993                 uid_t uid_shift,
 994                 DissectImageFlags flags) {
 995
 996         _cleanup_free_ char *chased = NULL, *options = NULL;
 997         const char *p, *node, *fstype;
 998         bool rw;
 999         int r;
1000
1001         assert(m);
1002         assert(where);
1003
1004         node = m->decrypted_node ?: m->node;
1005         fstype = m->decrypted_fstype ?: m->fstype;
1006
1007         if (!m->found || !node || !fstype)
1008                 return 0;
1009
1010         /* Stacked encryption? Yuck */
1011         if (streq_ptr(fstype, "crypto_LUKS"))
1012                 return -ELOOP;
1013
1014         rw = m->rw && !(flags & DISSECT_IMAGE_READ_ONLY);
1015
1016         if (FLAGS_SET(flags, DISSECT_IMAGE_FSCK) && rw) {
1017                 r = run_fsck(node, fstype);
1018                 if (r < 0)
1019                         return r;
1020         }
1021
1022         if (directory) {
1023                 r = chase_symlinks(directory, where, CHASE_PREFIX_ROOT, &chased, NULL);
1024                 if (r < 0)
1025                         return r;
1026
1027                 p = chased;
1028         } else
1029                 p = where;
1030
1031         /* If requested, turn on discard support. */
1032         if (fstype_can_discard(fstype) &&
1033             ((flags & DISSECT_IMAGE_DISCARD) ||
1034              ((flags & DISSECT_IMAGE_DISCARD_ON_LOOP) && is_loop_device(m->node)))) {
1035                 options = strdup("discard");
1036                 if (!options)
1037                         return -ENOMEM;
1038         }
1039
1040         if (uid_is_valid(uid_shift) && uid_shift != 0 && fstype_can_uid_gid(fstype)) {
1041                 _cleanup_free_ char *uid_option = NULL;
1042
1043                 if (asprintf(&uid_option, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0)
1044                         return -ENOMEM;
1045
1046                 if (!strextend_with_separator(&options, ",", uid_option, NULL))
1047                         return -ENOMEM;
1048         }
1049
1050         if (!isempty(m->mount_options))
1051                 if (!strextend_with_separator(&options, ",", m->mount_options, NULL))
1052                         return -ENOMEM;
1053
1054         r = mount_verbose(LOG_DEBUG, node, p, fstype, MS_NODEV|(rw ? 0 : MS_RDONLY), options);
1055         if (r < 0)
1056                 return r;
1057
1058         return 1;
1059 }
1060
1061 int dissected_image_mount(DissectedImage *m, const char *where, uid_t uid_shift, DissectImageFlags flags) {
1062         int r, boot_mounted;
1063
1064         assert(m);
1065         assert(where);
1066
1067         if (!m->partitions[PARTITION_ROOT].found)
1068                 return -ENXIO;
1069
1070         if ((flags & DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY) == 0) {
1071                 r = mount_partition(m->partitions + PARTITION_ROOT, where, NULL, uid_shift, flags);
1072                 if (r < 0)
1073                         return r;
1074
1075                 if (flags & DISSECT_IMAGE_VALIDATE_OS) {
1076                         r = path_is_os_tree(where);
1077                         if (r < 0)
1078                                 return r;
1079                         if (r == 0)
1080                                 return -EMEDIUMTYPE;
1081                 }
1082         }
1083
1084         if (flags & DISSECT_IMAGE_MOUNT_ROOT_ONLY)
1085                 return 0;
1086
1087         r = mount_partition(m->partitions + PARTITION_HOME, where, "/home", uid_shift, flags);
1088         if (r < 0)
1089                 return r;
1090
1091         r = mount_partition(m->partitions + PARTITION_SRV, where, "/srv", uid_shift, flags);
1092         if (r < 0)
1093                 return r;
1094
1095         r = mount_partition(m->partitions + PARTITION_VAR, where, "/var", uid_shift, flags);
1096         if (r < 0)
1097                 return r;
1098
1099         r = mount_partition(m->partitions + PARTITION_TMP, where, "/var/tmp", uid_shift, flags);
1100         if (r < 0)
1101                 return r;
1102
1103         boot_mounted = mount_partition(m->partitions + PARTITION_XBOOTLDR, where, "/boot", uid_shift, flags);
1104         if (boot_mounted < 0)
1105                 return boot_mounted;
1106
1107         if (m->partitions[PARTITION_ESP].found) {
1108                 /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it
1109                  * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */
1110
1111                 r = chase_symlinks("/efi", where, CHASE_PREFIX_ROOT, NULL, NULL);
1112                 if (r >= 0) {
1113                         r = mount_partition(m->partitions + PARTITION_ESP, where, "/efi", uid_shift, flags);
1114                         if (r < 0)
1115                                 return r;
1116
1117                 } else if (boot_mounted <= 0) {
1118                         _cleanup_free_ char *p = NULL;
1119
1120                         r = chase_symlinks("/boot", where, CHASE_PREFIX_ROOT, &p, NULL);
1121                         if (r >= 0 && dir_is_empty(p) > 0) {
1122                                 r = mount_partition(m->partitions + PARTITION_ESP, where, "/boot", uid_shift, flags);
1123                                 if (r < 0)
1124                                         return r;
1125                         }
1126                 }
1127         }
1128
1129         return 0;
1130 }
1131
1132 #if HAVE_LIBCRYPTSETUP
1133 typedef struct DecryptedPartition {
1134         struct crypt_device *device;
1135         char *name;
1136         bool relinquished;
1137 } DecryptedPartition;
1138
1139 struct DecryptedImage {
1140         DecryptedPartition *decrypted;
1141         size_t n_decrypted;
1142         size_t n_allocated;
1143 };
1144 #endif
1145
1146 DecryptedImage* decrypted_image_unref(DecryptedImage* d) {
1147 #if HAVE_LIBCRYPTSETUP
1148         size_t i;
1149         int r;
1150
1151         if (!d)
1152                 return NULL;
1153
1154         for (i = 0; i < d->n_decrypted; i++) {
1155                 DecryptedPartition *p = d->decrypted + i;
1156
1157                 if (p->device && p->name && !p->relinquished) {
1158                         r = crypt_deactivate(p->device, p->name);
1159                         if (r < 0)
1160                                 log_debug_errno(r, "Failed to deactivate encrypted partition %s", p->name);
1161                 }
1162
1163                 if (p->device)
1164                         crypt_free(p->device);
1165                 free(p->name);
1166         }
1167
1168         free(d);
1169 #endif
1170         return NULL;
1171 }
1172
1173 #if HAVE_LIBCRYPTSETUP
1174
1175 static int make_dm_name_and_node(const void *original_node, const char *suffix, char **ret_name, char **ret_node) {
1176         _cleanup_free_ char *name = NULL, *node = NULL;
1177         const char *base;
1178
1179         assert(original_node);
1180         assert(suffix);
1181         assert(ret_name);
1182         assert(ret_node);
1183
1184         base = strrchr(original_node, '/');
1185         if (!base)
1186                 base = original_node;
1187         else
1188                 base++;
1189         if (isempty(base))
1190                 return -EINVAL;
1191
1192         name = strjoin(base, suffix);
1193         if (!name)
1194                 return -ENOMEM;
1195         if (!filename_is_valid(name))
1196                 return -EINVAL;
1197
1198         node = path_join(crypt_get_dir(), name);
1199         if (!node)
1200                 return -ENOMEM;
1201
1202         *ret_name = TAKE_PTR(name);
1203         *ret_node = TAKE_PTR(node);
1204
1205         return 0;
1206 }
1207
1208 static int decrypt_partition(
1209                 DissectedPartition *m,
1210                 const char *passphrase,
1211                 DissectImageFlags flags,
1212                 DecryptedImage *d) {
1213
1214         _cleanup_free_ char *node = NULL, *name = NULL;
1215         _cleanup_(crypt_freep) struct crypt_device *cd = NULL;
1216         int r;
1217
1218         assert(m);
1219         assert(d);
1220
1221         if (!m->found || !m->node || !m->fstype)
1222                 return 0;
1223
1224         if (!streq(m->fstype, "crypto_LUKS"))
1225                 return 0;
1226
1227         if (!passphrase)
1228                 return -ENOKEY;
1229
1230         r = make_dm_name_and_node(m->node, "-decrypted", &name, &node);
1231         if (r < 0)
1232                 return r;
1233
1234         if (!GREEDY_REALLOC0(d->decrypted, d->n_allocated, d->n_decrypted + 1))
1235                 return -ENOMEM;
1236
1237         r = crypt_init(&cd, m->node);
1238         if (r < 0)
1239                 return log_debug_errno(r, "Failed to initialize dm-crypt: %m");
1240
1241         crypt_set_log_callback(cd, cryptsetup_log_glue, NULL);
1242
1243         r = crypt_load(cd, CRYPT_LUKS, NULL);
1244         if (r < 0)
1245                 return log_debug_errno(r, "Failed to load LUKS metadata: %m");
1246
1247         r = crypt_activate_by_passphrase(cd, name, CRYPT_ANY_SLOT, passphrase, strlen(passphrase),
1248                                          ((flags & DISSECT_IMAGE_READ_ONLY) ? CRYPT_ACTIVATE_READONLY : 0) |
1249                                          ((flags & DISSECT_IMAGE_DISCARD_ON_CRYPTO) ? CRYPT_ACTIVATE_ALLOW_DISCARDS : 0));
1250         if (r < 0) {
1251                 log_debug_errno(r, "Failed to activate LUKS device: %m");
1252                 return r == -EPERM ? -EKEYREJECTED : r;
1253         }
1254
1255         d->decrypted[d->n_decrypted].name = TAKE_PTR(name);
1256         d->decrypted[d->n_decrypted].device = TAKE_PTR(cd);
1257         d->n_decrypted++;
1258
1259         m->decrypted_node = TAKE_PTR(node);
1260
1261         return 0;
1262 }
1263
1264 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) {
1265         /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
1266         _cleanup_free_ char *root_hash_existing = NULL;
1267         _cleanup_(crypt_freep) struct crypt_device *cd = NULL;
1268         struct crypt_params_verity crypt_params = {};
1269         size_t root_hash_existing_size = root_hash_size;
1270         int r;
1271
1272         assert(ret_cd);
1273
1274         r = crypt_init_by_name(&cd, name);
1275         if (r < 0)
1276                 return log_debug_errno(r, "Error opening verity device, crypt_init_by_name failed: %m");
1277         r = crypt_get_verity_info(cd, &crypt_params);
1278         if (r < 0)
1279                 return log_debug_errno(r, "Error opening verity device, crypt_get_verity_info failed: %m");
1280         root_hash_existing = malloc0(root_hash_size);
1281         if (!root_hash_existing)
1282                 return -ENOMEM;
1283         r = crypt_volume_key_get(cd, CRYPT_ANY_SLOT, root_hash_existing, &root_hash_existing_size, NULL, 0);
1284         if (r < 0)
1285                 return log_debug_errno(r, "Error opening verity device, crypt_volume_key_get failed: %m");
1286         if (root_hash_size != root_hash_existing_size || memcmp(root_hash_existing, root_hash, root_hash_size) != 0)
1287                 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but root hashes are different.");
1288 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1289         /* Ensure that, if signatures are supported, we only reuse the device if the previous mount
1290          * used the same settings, so that a previous unsigned mount will not be reused if the user
1291          * asks to use signing for the new one, and viceversa. */
1292         if (has_sig != !!(crypt_params.flags & CRYPT_VERITY_ROOT_HASH_SIGNATURE))
1293                 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but signature settings are not the same.");
1294 #endif
1295
1296         *ret_cd = TAKE_PTR(cd);
1297         return 0;
1298 }
1299
1300 static inline void dm_deferred_remove_clean(char *name) {
1301         if (!name)
1302                 return;
1303         (void) crypt_deactivate_by_name(NULL, name, CRYPT_DEACTIVATE_DEFERRED);
1304         free(name);
1305 }
1306 DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean);
1307
1308 static int verity_partition(
1309                 DissectedPartition *m,
1310                 DissectedPartition *v,
1311                 const void *root_hash,
1312                 size_t root_hash_size,
1313                 const char *verity_data,
1314                 const char *root_hash_sig_path,
1315                 const void *root_hash_sig,
1316                 size_t root_hash_sig_size,
1317                 DissectImageFlags flags,
1318                 DecryptedImage *d) {
1319
1320         _cleanup_free_ char *node = NULL, *name = NULL, *hash_sig_from_file = NULL;
1321         _cleanup_(crypt_freep) struct crypt_device *cd = NULL;
1322         _cleanup_(dm_deferred_remove_cleanp) char *restore_deferred_remove = NULL;
1323         int r;
1324
1325         assert(m);
1326         assert(v || verity_data);
1327
1328         if (!root_hash)
1329                 return 0;
1330
1331         if (!m->found || !m->node || !m->fstype)
1332                 return 0;
1333         if (!verity_data) {
1334                 if (!v->found || !v->node || !v->fstype)
1335                         return 0;
1336
1337                 if (!streq(v->fstype, "DM_verity_hash"))
1338                         return 0;
1339         }
1340
1341         if (FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) {
1342                 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
1343                 _cleanup_free_ char *root_hash_encoded = NULL;
1344                 root_hash_encoded = hexmem(root_hash, root_hash_size);
1345                 if (!root_hash_encoded)
1346                         return -ENOMEM;
1347                 r = make_dm_name_and_node(root_hash_encoded, "-verity", &name, &node);
1348         } else
1349                 r = make_dm_name_and_node(m->node, "-verity", &name, &node);
1350         if (r < 0)
1351                 return r;
1352
1353         if (!root_hash_sig && root_hash_sig_path) {
1354                 r = read_full_file_full(AT_FDCWD, root_hash_sig_path, 0, &hash_sig_from_file, &root_hash_sig_size);
1355                 if (r < 0)
1356                         return r;
1357         }
1358
1359         r = crypt_init(&cd, verity_data ?: v->node);
1360         if (r < 0)
1361                 return r;
1362
1363         crypt_set_log_callback(cd, cryptsetup_log_glue, NULL);
1364
1365         r = crypt_load(cd, CRYPT_VERITY, NULL);
1366         if (r < 0)
1367                 return r;
1368
1369         r = crypt_set_data_device(cd, m->node);
1370         if (r < 0)
1371                 return r;
1372
1373         if (!GREEDY_REALLOC0(d->decrypted, d->n_allocated, d->n_decrypted + 1))
1374                 return -ENOMEM;
1375
1376         /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
1377          * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
1378          * retry a few times before giving up. */
1379         for (unsigned i = 0; i < N_DEVICE_NODE_LIST_ATTEMPTS; i++) {
1380                 if (root_hash_sig || hash_sig_from_file) {
1381 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1382                         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);
1383 #else
1384                         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()");
1385 #endif
1386                 } else
1387                         r = crypt_activate_by_volume_key(cd, name, root_hash, root_hash_size, CRYPT_ACTIVATE_READONLY);
1388                 /* libdevmapper can return EINVAL when the device is already in the activation stage.
1389                  * There's no way to distinguish this situation from a genuine error due to invalid
1390                  * parameters, so immediately fallback to activating the device with a unique name.
1391                  * Improvements in libcrypsetup can ensure this never happens: https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
1392                 if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
1393                         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);
1394                 if (!IN_SET(r, 0, -EEXIST, -ENODEV))
1395                         return r;
1396                 if (r == -EEXIST) {
1397                         struct crypt_device *existing_cd = NULL;
1398
1399                         if (!restore_deferred_remove){
1400                                 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
1401                                 r = dm_deferred_remove_cancel(name);
1402                                 if (r < 0)
1403                                         return log_debug_errno(r, "Disabling automated deferred removal for verity device %s failed: %m", node);
1404                                 restore_deferred_remove = strdup(name);
1405                                 if (!restore_deferred_remove)
1406                                         return -ENOMEM;
1407                         }
1408
1409                         r = verity_can_reuse(root_hash, root_hash_size, !!root_hash_sig || !!hash_sig_from_file, name, &existing_cd);
1410                         /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
1411                         if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
1412                                 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);
1413                         if (!IN_SET(r, 0, -ENODEV, -ENOENT))
1414                                 return log_debug_errno(r, "Checking whether existing verity device %s can be reused failed: %m", node);
1415                         if (r == 0) {
1416                                 if (cd)
1417                                         crypt_free(cd);
1418                                 cd = existing_cd;
1419                         }
1420                 }
1421                 if (r == 0)
1422                         break;
1423         }
1424
1425         /* Sanity check: libdevmapper is known to report that the device already exists and is active,
1426         * but it's actually not there, so the later filesystem probe or mount would fail. */
1427         if (r == 0)
1428                 r = access(node, F_OK);
1429         /* An existing verity device was reported by libcryptsetup/libdevmapper, but we can't use it at this time.
1430          * Fall back to activating it with a unique device name. */
1431         if (r != 0 && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
1432                 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);
1433
1434         /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
1435         restore_deferred_remove = mfree(restore_deferred_remove);
1436
1437         d->decrypted[d->n_decrypted].name = TAKE_PTR(name);
1438         d->decrypted[d->n_decrypted].device = TAKE_PTR(cd);
1439         d->n_decrypted++;
1440
1441         m->decrypted_node = TAKE_PTR(node);
1442
1443         return 0;
1444 }
1445 #endif
1446
1447 int dissected_image_decrypt(
1448                 DissectedImage *m,
1449                 const char *passphrase,
1450                 const void *root_hash,
1451                 size_t root_hash_size,
1452                 const char *verity_data,
1453                 const char *root_hash_sig_path,
1454                 const void *root_hash_sig,
1455                 size_t root_hash_sig_size,
1456                 DissectImageFlags flags,
1457                 DecryptedImage **ret) {
1458
1459 #if HAVE_LIBCRYPTSETUP
1460         _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL;
1461         unsigned i;
1462         int r;
1463 #endif
1464
1465         assert(m);
1466         assert(root_hash || root_hash_size == 0);
1467
1468         /* Returns:
1469          *
1470          *      = 0           → There was nothing to decrypt
1471          *      > 0           → Decrypted successfully
1472          *      -ENOKEY       → There's something to decrypt but no key was supplied
1473          *      -EKEYREJECTED → Passed key was not correct
1474          */
1475
1476         if (root_hash && root_hash_size < sizeof(sd_id128_t))
1477                 return -EINVAL;
1478
1479         if (!m->encrypted && !m->verity) {
1480                 *ret = NULL;
1481                 return 0;
1482         }
1483
1484 #if HAVE_LIBCRYPTSETUP
1485         d = new0(DecryptedImage, 1);
1486         if (!d)
1487                 return -ENOMEM;
1488
1489         for (i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
1490                 DissectedPartition *p = m->partitions + i;
1491                 int k;
1492
1493                 if (!p->found)
1494                         continue;
1495
1496                 r = decrypt_partition(p, passphrase, flags, d);
1497                 if (r < 0)
1498                         return r;
1499
1500                 k = PARTITION_VERITY_OF(i);
1501                 if (k >= 0) {
1502                         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);
1503                         if (r < 0)
1504                                 return r;
1505                 }
1506
1507                 if (!p->decrypted_fstype && p->decrypted_node) {
1508                         r = probe_filesystem(p->decrypted_node, &p->decrypted_fstype);
1509                         if (r < 0 && r != -EUCLEAN)
1510                                 return r;
1511                 }
1512         }
1513
1514         *ret = TAKE_PTR(d);
1515
1516         return 1;
1517 #else
1518         return -EOPNOTSUPP;
1519 #endif
1520 }
1521
1522 int dissected_image_decrypt_interactively(
1523                 DissectedImage *m,
1524                 const char *passphrase,
1525                 const void *root_hash,
1526                 size_t root_hash_size,
1527                 const char *verity_data,
1528                 const char *root_hash_sig_path,
1529                 const void *root_hash_sig,
1530                 size_t root_hash_sig_size,
1531                 DissectImageFlags flags,
1532                 DecryptedImage **ret) {
1533
1534         _cleanup_strv_free_erase_ char **z = NULL;
1535         int n = 3, r;
1536
1537         if (passphrase)
1538                 n--;
1539
1540         for (;;) {
1541                 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);
1542                 if (r >= 0)
1543                         return r;
1544                 if (r == -EKEYREJECTED)
1545                         log_error_errno(r, "Incorrect passphrase, try again!");
1546                 else if (r != -ENOKEY)
1547                         return log_error_errno(r, "Failed to decrypt image: %m");
1548
1549                 if (--n < 0)
1550                         return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED),
1551                                                "Too many retries.");
1552
1553                 z = strv_free(z);
1554
1555                 r = ask_password_auto("Please enter image passphrase:", NULL, "dissect", "dissect", USEC_INFINITY, 0, &z);
1556                 if (r < 0)
1557                         return log_error_errno(r, "Failed to query for passphrase: %m");
1558
1559                 passphrase = z[0];
1560         }
1561 }
1562
1563 int decrypted_image_relinquish(DecryptedImage *d) {
1564
1565 #if HAVE_LIBCRYPTSETUP
1566         size_t i;
1567         int r;
1568 #endif
1569
1570         assert(d);
1571
1572         /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a boolean so
1573          * that we don't clean it up ourselves either anymore */
1574
1575 #if HAVE_LIBCRYPTSETUP
1576         for (i = 0; i < d->n_decrypted; i++) {
1577                 DecryptedPartition *p = d->decrypted + i;
1578
1579                 if (p->relinquished)
1580                         continue;
1581
1582                 r = crypt_deactivate_by_name(NULL, p->name, CRYPT_DEACTIVATE_DEFERRED);
1583                 if (r < 0)
1584                         return log_debug_errno(r, "Failed to mark %s for auto-removal: %m", p->name);
1585
1586                 p->relinquished = true;
1587         }
1588 #endif
1589
1590         return 0;
1591 }
1592
1593 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) {
1594         _cleanup_free_ char *verity_filename = NULL, *roothashsig_filename = NULL;
1595         _cleanup_free_ void *roothash_decoded = NULL;
1596         size_t roothash_decoded_size = 0;
1597         int r;
1598
1599         assert(image);
1600
1601         if (is_device_path(image)) {
1602                 /* If we are asked to load the root hash for a device node, exit early */
1603                 if (ret_roothash)
1604                         *ret_roothash = NULL;
1605                 if (ret_roothash_size)
1606                         *ret_roothash_size = 0;
1607                 if (ret_verity_data)
1608                         *ret_verity_data = NULL;
1609                 if (ret_roothashsig)
1610                         *ret_roothashsig = NULL;
1611                 return 0;
1612         }
1613
1614         if (ret_verity_data) {
1615                 char *e;
1616
1617                 verity_filename = new(char, strlen(image) + STRLEN(".verity") + 1);
1618                 if (!verity_filename)
1619                         return -ENOMEM;
1620                 strcpy(verity_filename, image);
1621                 e = endswith(verity_filename, ".raw");
1622                 if (e)
1623                         strcpy(e, ".verity");
1624                 else
1625                         strcat(verity_filename, ".verity");
1626
1627                 r = access(verity_filename, F_OK);
1628                 if (r < 0) {
1629                         if (errno != ENOENT)
1630                                 return -errno;
1631                         verity_filename = mfree(verity_filename);
1632                 }
1633         }
1634
1635         if (ret_roothashsig) {
1636                 char *e;
1637
1638                 /* Follow naming convention recommended by the relevant RFC:
1639                  * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
1640                 roothashsig_filename = new(char, strlen(image) + STRLEN(".roothash.p7s") + 1);
1641                 if (!roothashsig_filename)
1642                         return -ENOMEM;
1643                 strcpy(roothashsig_filename, image);
1644                 e = endswith(roothashsig_filename, ".raw");
1645                 if (e)
1646                         strcpy(e, ".roothash.p7s");
1647                 else
1648                         strcat(roothashsig_filename, ".roothash.p7s");
1649
1650                 r = access(roothashsig_filename, R_OK);
1651                 if (r < 0) {
1652                         if (errno != ENOENT)
1653                                 return -errno;
1654                         roothashsig_filename = mfree(roothashsig_filename);
1655                 }
1656         }
1657
1658         if (ret_roothash) {
1659                 _cleanup_free_ char *text = NULL;
1660                 assert(ret_roothash_size);
1661
1662                 if (root_hash_path) {
1663                         /* We have the path to a roothash to load and decode, eg: RootHash=/foo/bar.roothash */
1664                         r = read_one_line_file(root_hash_path, &text);
1665                         if (r < 0)
1666                                 return r;
1667                 } else {
1668                         r = getxattr_malloc(image, "user.verity.roothash", &text, true);
1669                         if (r < 0) {
1670                                 char *fn, *e, *n;
1671
1672                                 if (!IN_SET(r, -ENODATA, -EOPNOTSUPP, -ENOENT))
1673                                         return r;
1674
1675                                 fn = newa(char, strlen(image) + STRLEN(".roothash") + 1);
1676                                 n = stpcpy(fn, image);
1677                                 e = endswith(fn, ".raw");
1678                                 if (e)
1679                                         n = e;
1680
1681                                 strcpy(n, ".roothash");
1682
1683                                 r = read_one_line_file(fn, &text);
1684                                 if (r < 0 && r != -ENOENT)
1685                                         return r;
1686                         }
1687                 }
1688
1689                 if (text) {
1690                         r = unhexmem(text, strlen(text), &roothash_decoded, &roothash_decoded_size);
1691                         if (r < 0)
1692                                 return r;
1693                         if (roothash_decoded_size < sizeof(sd_id128_t))
1694                                 return -EINVAL;
1695                 }
1696         }
1697
1698         if (ret_roothash) {
1699                 *ret_roothash = TAKE_PTR(roothash_decoded);
1700                 *ret_roothash_size = roothash_decoded_size;
1701         }
1702         if (ret_verity_data)
1703                 *ret_verity_data = TAKE_PTR(verity_filename);
1704         if (roothashsig_filename)
1705                 *ret_roothashsig = TAKE_PTR(roothashsig_filename);
1706
1707         return 1;
1708 }
1709
1710 int dissected_image_acquire_metadata(DissectedImage *m) {
1711
1712         enum {
1713                 META_HOSTNAME,
1714                 META_MACHINE_ID,
1715                 META_MACHINE_INFO,
1716                 META_OS_RELEASE,
1717                 _META_MAX,
1718         };
1719
1720         static const char *const paths[_META_MAX] = {
1721                 [META_HOSTNAME]     = "/etc/hostname\0",
1722                 [META_MACHINE_ID]   = "/etc/machine-id\0",
1723                 [META_MACHINE_INFO] = "/etc/machine-info\0",
1724                 [META_OS_RELEASE]   = "/etc/os-release\0"
1725                                       "/usr/lib/os-release\0",
1726         };
1727
1728         _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL;
1729         _cleanup_(rmdir_and_freep) char *t = NULL;
1730         _cleanup_(sigkill_waitp) pid_t child = 0;
1731         sd_id128_t machine_id = SD_ID128_NULL;
1732         _cleanup_free_ char *hostname = NULL;
1733         unsigned n_meta_initialized = 0, k;
1734         int fds[2 * _META_MAX], r;
1735
1736         BLOCK_SIGNALS(SIGCHLD);
1737
1738         assert(m);
1739
1740         for (; n_meta_initialized < _META_MAX; n_meta_initialized ++)
1741                 if (pipe2(fds + 2*n_meta_initialized, O_CLOEXEC) < 0) {
1742                         r = -errno;
1743                         goto finish;
1744                 }
1745
1746         r = mkdtemp_malloc("/tmp/dissect-XXXXXX", &t);
1747         if (r < 0)
1748                 goto finish;
1749
1750         r = safe_fork("(sd-dissect)", FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE, &child);
1751         if (r < 0)
1752                 goto finish;
1753         if (r == 0) {
1754                 r = dissected_image_mount(m, t, UID_INVALID, DISSECT_IMAGE_READ_ONLY|DISSECT_IMAGE_MOUNT_ROOT_ONLY|DISSECT_IMAGE_VALIDATE_OS);
1755                 if (r < 0) {
1756                         log_debug_errno(r, "Failed to mount dissected image: %m");
1757                         _exit(EXIT_FAILURE);
1758                 }
1759
1760                 for (k = 0; k < _META_MAX; k++) {
1761                         _cleanup_close_ int fd = -1;
1762                         const char *p;
1763
1764                         fds[2*k] = safe_close(fds[2*k]);
1765
1766                         NULSTR_FOREACH(p, paths[k]) {
1767                                 fd = chase_symlinks_and_open(p, t, CHASE_PREFIX_ROOT, O_RDONLY|O_CLOEXEC|O_NOCTTY, NULL);
1768                                 if (fd >= 0)
1769                                         break;
1770                         }
1771                         if (fd < 0) {
1772                                 log_debug_errno(fd, "Failed to read %s file of image, ignoring: %m", paths[k]);
1773                                 continue;
1774                         }
1775
1776                         r = copy_bytes(fd, fds[2*k+1], (uint64_t) -1, 0);
1777                         if (r < 0)
1778                                 _exit(EXIT_FAILURE);
1779
1780                         fds[2*k+1] = safe_close(fds[2*k+1]);
1781                 }
1782
1783                 _exit(EXIT_SUCCESS);
1784         }
1785
1786         for (k = 0; k < _META_MAX; k++) {
1787                 _cleanup_fclose_ FILE *f = NULL;
1788
1789                 fds[2*k+1] = safe_close(fds[2*k+1]);
1790
1791                 f = take_fdopen(&fds[2*k], "r");
1792                 if (!f) {
1793                         r = -errno;
1794                         goto finish;
1795                 }
1796
1797                 switch (k) {
1798
1799                 case META_HOSTNAME:
1800                         r = read_etc_hostname_stream(f, &hostname);
1801                         if (r < 0)
1802                                 log_debug_errno(r, "Failed to read /etc/hostname: %m");
1803
1804                         break;
1805
1806                 case META_MACHINE_ID: {
1807                         _cleanup_free_ char *line = NULL;
1808
1809                         r = read_line(f, LONG_LINE_MAX, &line);
1810                         if (r < 0)
1811                                 log_debug_errno(r, "Failed to read /etc/machine-id: %m");
1812                         else if (r == 33) {
1813                                 r = sd_id128_from_string(line, &machine_id);
1814                                 if (r < 0)
1815                                         log_debug_errno(r, "Image contains invalid /etc/machine-id: %s", line);
1816                         } else if (r == 0)
1817                                 log_debug("/etc/machine-id file is empty.");
1818                         else
1819                                 log_debug("/etc/machine-id has unexpected length %i.", r);
1820
1821                         break;
1822                 }
1823
1824                 case META_MACHINE_INFO:
1825                         r = load_env_file_pairs(f, "machine-info", &machine_info);
1826                         if (r < 0)
1827                                 log_debug_errno(r, "Failed to read /etc/machine-info: %m");
1828
1829                         break;
1830
1831                 case META_OS_RELEASE:
1832                         r = load_env_file_pairs(f, "os-release", &os_release);
1833                         if (r < 0)
1834                                 log_debug_errno(r, "Failed to read OS release file: %m");
1835
1836                         break;
1837                 }
1838         }
1839
1840         r = wait_for_terminate_and_check("(sd-dissect)", child, 0);
1841         child = 0;
1842         if (r < 0)
1843                 goto finish;
1844         if (r != EXIT_SUCCESS)
1845                 return -EPROTO;
1846
1847         free_and_replace(m->hostname, hostname);
1848         m->machine_id = machine_id;
1849         strv_free_and_replace(m->machine_info, machine_info);
1850         strv_free_and_replace(m->os_release, os_release);
1851
1852 finish:
1853         for (k = 0; k < n_meta_initialized; k++)
1854                 safe_close_pair(fds + 2*k);
1855
1856         return r;
1857 }
1858
1859 int dissect_image_and_warn(
1860                 int fd,
1861                 const char *name,
1862                 const void *root_hash,
1863                 size_t root_hash_size,
1864                 const char *verity_data,
1865                 const MountOptions *mount_options,
1866                 DissectImageFlags flags,
1867                 DissectedImage **ret) {
1868
1869         _cleanup_free_ char *buffer = NULL;
1870         int r;
1871
1872         if (!name) {
1873                 r = fd_get_path(fd, &buffer);
1874                 if (r < 0)
1875                         return r;
1876
1877                 name = buffer;
1878         }
1879
1880         r = dissect_image(fd, root_hash, root_hash_size, verity_data, mount_options, flags, ret);
1881
1882         switch (r) {
1883
1884         case -EOPNOTSUPP:
1885                 return log_error_errno(r, "Dissecting images is not supported, compiled without blkid support.");
1886
1887         case -ENOPKG:
1888                 return log_error_errno(r, "Couldn't identify a suitable partition table or file system in '%s'.", name);
1889
1890         case -EADDRNOTAVAIL:
1891                 return log_error_errno(r, "No root partition for specified root hash found in '%s'.", name);
1892
1893         case -ENOTUNIQ:
1894                 return log_error_errno(r, "Multiple suitable root partitions found in image '%s'.", name);
1895
1896         case -ENXIO:
1897                 return log_error_errno(r, "No suitable root partition found in image '%s'.", name);
1898
1899         case -EPROTONOSUPPORT:
1900                 return log_error_errno(r, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name);
1901
1902         default:
1903                 if (r < 0)
1904                         return log_error_errno(r, "Failed to dissect image '%s': %m", name);
1905
1906                 return r;
1907         }
1908 }
1909
1910 bool dissected_image_can_do_verity(const DissectedImage *image, unsigned partition_designator) {
1911         if (image->single_file_system)
1912                 return partition_designator == PARTITION_ROOT && image->can_verity;
1913
1914         return PARTITION_VERITY_OF(partition_designator) >= 0;
1915 }
1916
1917 bool dissected_image_has_verity(const DissectedImage *image, unsigned partition_designator) {
1918         int k;
1919
1920         if (image->single_file_system)
1921                 return partition_designator == PARTITION_ROOT && image->verity;
1922
1923         k = PARTITION_VERITY_OF(partition_designator);
1924         return k >= 0 && image->partitions[k].found;
1925 }
1926
1927 MountOptions* mount_options_free_all(MountOptions *options) {
1928         MountOptions *m;
1929
1930         while ((m = options)) {
1931                 LIST_REMOVE(mount_options, options, m);
1932                 free(m->options);
1933                 free(m);
1934         }
1935
1936         return NULL;
1937 }
1938
1939 const char* mount_options_from_part(const MountOptions *options, unsigned int partition_number) {
1940         MountOptions *m;
1941
1942         LIST_FOREACH(mount_options, m, (MountOptions *)options)
1943                 if (partition_number == m->partition_number && !isempty(m->options))
1944                         return m->options;
1945
1946         return NULL;
1947 }
1948
1949 int mount_image_privately_interactively(
1950                 const char *image,
1951                 DissectImageFlags flags,
1952                 char **ret_directory,
1953                 LoopDevice **ret_loop_device,
1954                 DecryptedImage **ret_decrypted_image) {
1955
1956         _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
1957         _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
1958         _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
1959         _cleanup_(rmdir_and_freep) char *created_dir = NULL;
1960         _cleanup_free_ char *temp = NULL;
1961         int r;
1962
1963         /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
1964          * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
1965          * easily. */
1966
1967         assert(image);
1968         assert(ret_directory);
1969         assert(ret_loop_device);
1970         assert(ret_decrypted_image);
1971
1972         r = tempfn_random_child(NULL, program_invocation_short_name, &temp);
1973         if (r < 0)
1974                 return log_error_errno(r, "Failed to generate temporary mount directory: %m");
1975
1976         r = loop_device_make_by_path(
1977                         image,
1978                         FLAGS_SET(flags, DISSECT_IMAGE_READ_ONLY) ? O_RDONLY : O_RDWR,
1979                         FLAGS_SET(flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN,
1980                         &d);
1981         if (r < 0)
1982                 return log_error_errno(r, "Failed to set up loopback device: %m");
1983
1984         r = dissect_image_and_warn(d->fd, image, NULL, 0, NULL, NULL, flags, &dissected_image);
1985         if (r < 0)
1986                 return r;
1987
1988         r = dissected_image_decrypt_interactively(dissected_image, NULL, NULL, 0, NULL, NULL, NULL, 0, flags, &decrypted_image);
1989         if (r < 0)
1990                 return r;
1991
1992         r = detach_mount_namespace();
1993         if (r < 0)
1994                 return log_error_errno(r, "Failed to detach mount namespace: %m");
1995
1996         r = mkdir_p(temp, 0700);
1997         if (r < 0)
1998                 return log_error_errno(r, "Failed to create mount point: %m");
1999
2000         created_dir = TAKE_PTR(temp);
2001
2002         r = dissected_image_mount(dissected_image, created_dir, UID_INVALID, flags);
2003         if (r == -EUCLEAN)
2004                 return log_error_errno(r, "File system check on image failed: %m");
2005         if (r < 0)
2006                 return log_error_errno(r, "Failed to mount image: %m");
2007
2008         if (decrypted_image) {
2009                 r = decrypted_image_relinquish(decrypted_image);
2010                 if (r < 0)
2011                         return log_error_errno(r, "Failed to relinquish DM devices: %m");
2012         }
2013
2014         loop_device_relinquish(d);
2015
2016         *ret_directory = TAKE_PTR(created_dir);
2017         *ret_loop_device = TAKE_PTR(d);
2018         *ret_decrypted_image = TAKE_PTR(decrypted_image);
2019
2020         return 0;
2021 }
2022
2023 static const char *const partition_designator_table[] = {
2024         [PARTITION_ROOT] = "root",
2025         [PARTITION_ROOT_SECONDARY] = "root-secondary",
2026         [PARTITION_HOME] = "home",
2027         [PARTITION_SRV] = "srv",
2028         [PARTITION_ESP] = "esp",
2029         [PARTITION_XBOOTLDR] = "xbootldr",
2030         [PARTITION_SWAP] = "swap",
2031         [PARTITION_ROOT_VERITY] = "root-verity",
2032         [PARTITION_ROOT_SECONDARY_VERITY] = "root-secondary-verity",
2033         [PARTITION_TMP] = "tmp",
2034         [PARTITION_VAR] = "var",
2035 };
2036
2037 DEFINE_STRING_TABLE_LOOKUP(partition_designator, int);