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 "mount-util.h"
  37 #include "mountpoint-util.h"
  38 #include "nulstr-util.h"
  39 #include "os-util.h"
  40 #include "path-util.h"
  41 #include "process-util.h"
  42 #include "raw-clone.h"
  43 #include "signal-util.h"
  44 #include "stat-util.h"
  45 #include "stdio-util.h"
  46 #include "string-table.h"
  47 #include "string-util.h"
  48 #include "strv.h"
  49 #include "tmpfile-util.h"
  50 #include "udev-util.h"
  51 #include "user-util.h"
  52 #include "xattr-util.h"
  53
  54 int probe_filesystem(const char *node, char **ret_fstype) {
  55         /* Try to find device content type and return it in *ret_fstype. If nothing is found,
  56          * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and an
  57          * different error otherwise. */
  58
  59 #if HAVE_BLKID
  60         _cleanup_(blkid_free_probep) blkid_probe b = NULL;
  61         const char *fstype;
  62         int r;
  63
  64         errno = 0;
  65         b = blkid_new_probe_from_filename(node);
  66         if (!b)
  67                 return errno_or_else(ENOMEM);
  68
  69         blkid_probe_enable_superblocks(b, 1);
  70         blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE);
  71
  72         errno = 0;
  73         r = blkid_do_safeprobe(b);
  74         if (r == 1) {
  75                 log_debug("No type detected on partition %s", node);
  76                 goto not_found;
  77         }
  78         if (r == -2) {
  79                 log_debug("Results ambiguous for partition %s", node);
  80                 return -EUCLEAN;
  81         }
  82         if (r != 0)
  83                 return errno_or_else(EIO);
  84
  85         (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
  86
  87         if (fstype) {
  88                 char *t;
  89
  90                 t = strdup(fstype);
  91                 if (!t)
  92                         return -ENOMEM;
  93
  94                 *ret_fstype = t;
  95                 return 1;
  96         }
  97
  98 not_found:
  99         *ret_fstype = NULL;
 100         return 0;
 101 #else
 102         return -EOPNOTSUPP;
 103 #endif
 104 }
 105
 106 #if HAVE_BLKID
 107 /* Detect RPMB and Boot partitions, which are not listed by blkid.
 108  * See https://github.com/systemd/systemd/issues/5806. */
 109 static bool device_is_mmc_special_partition(sd_device *d) {
 110         const char *sysname;
 111
 112         assert(d);
 113
 114         if (sd_device_get_sysname(d, &sysname) < 0)
 115                 return false;
 116
 117         return startswith(sysname, "mmcblk") &&
 118                 (endswith(sysname, "rpmb") || endswith(sysname, "boot0") || endswith(sysname, "boot1"));
 119 }
 120
 121 static bool device_is_block(sd_device *d) {
 122         const char *ss;
 123
 124         assert(d);
 125
 126         if (sd_device_get_subsystem(d, &ss) < 0)
 127                 return false;
 128
 129         return streq(ss, "block");
 130 }
 131
 132 static int enumerator_for_parent(sd_device *d, sd_device_enumerator **ret) {
 133         _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
 134         int r;
 135
 136         assert(d);
 137         assert(ret);
 138
 139         r = sd_device_enumerator_new(&e);
 140         if (r < 0)
 141                 return r;
 142
 143         r = sd_device_enumerator_allow_uninitialized(e);
 144         if (r < 0)
 145                 return r;
 146
 147         r = sd_device_enumerator_add_match_parent(e, d);
 148         if (r < 0)
 149                 return r;
 150
 151         *ret = TAKE_PTR(e);
 152         return 0;
 153 }
 154
 155 /* how many times to wait for the device nodes to appear */
 156 #define N_DEVICE_NODE_LIST_ATTEMPTS 10
 157
 158 static int wait_for_partitions_to_appear(
 159                 int fd,
 160                 sd_device *d,
 161                 unsigned num_partitions,
 162                 DissectImageFlags flags,
 163                 sd_device_enumerator **ret_enumerator) {
 164
 165         _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
 166         sd_device *q;
 167         unsigned n;
 168         int r;
 169
 170         assert(fd >= 0);
 171         assert(d);
 172         assert(ret_enumerator);
 173
 174         r = enumerator_for_parent(d, &e);
 175         if (r < 0)
 176                 return r;
 177
 178         /* Count the partitions enumerated by the kernel */
 179         n = 0;
 180         FOREACH_DEVICE(e, q) {
 181                 if (sd_device_get_devnum(q, NULL) < 0)
 182                         continue;
 183                 if (!device_is_block(q))
 184                         continue;
 185                 if (device_is_mmc_special_partition(q))
 186                         continue;
 187
 188                 if (!FLAGS_SET(flags, DISSECT_IMAGE_NO_UDEV)) {
 189                         r = device_wait_for_initialization(q, "block", USEC_INFINITY, NULL);
 190                         if (r < 0)
 191                                 return r;
 192                 }
 193
 194                 n++;
 195         }
 196
 197         if (n == num_partitions + 1) {
 198                 *ret_enumerator = TAKE_PTR(e);
 199                 return 0; /* success! */
 200         }
 201         if (n > num_partitions + 1)
 202                 return log_debug_errno(SYNTHETIC_ERRNO(EIO),
 203                                        "blkid and kernel partition lists do not match.");
 204
 205         /* The kernel has probed fewer partitions than blkid? Maybe the kernel prober is still running or it
 206          * got EBUSY because udev already opened the device. Let's reprobe the device, which is a synchronous
 207          * call that waits until probing is complete. */
 208
 209         for (unsigned j = 0; ; j++) {
 210                 if (j++ > 20)
 211                         return -EBUSY;
 212
 213                 if (ioctl(fd, BLKRRPART, 0) >= 0)
 214                         break;
 215                 r = -errno;
 216                 if (r == -EINVAL) {
 217                         struct loop_info64 info;
 218
 219                         /* If we are running on a loop device that has partition scanning off, return
 220                          * an explicit recognizable error about this, so that callers can generate a
 221                          * proper message explaining the situation. */
 222
 223                         if (ioctl(fd, LOOP_GET_STATUS64, &info) >= 0) {
 224 #if HAVE_VALGRIND_MEMCHECK_H
 225                                 /* Valgrind currently doesn't know LOOP_GET_STATUS64. Remove this once it does */
 226                                 VALGRIND_MAKE_MEM_DEFINED(&info, sizeof(info));
 227 #endif
 228
 229                                 if ((info.lo_flags & LO_FLAGS_PARTSCAN) == 0)
 230                                         return log_debug_errno(EPROTONOSUPPORT,
 231                                                                "Device is a loop device and partition scanning is off!");
 232                         }
 233                 }
 234                 if (r != -EBUSY)
 235                         return r;
 236
 237                 /* If something else has the device open, such as an udev rule, the ioctl will return
 238                  * EBUSY. Since there's no way to wait until it isn't busy anymore, let's just wait a bit,
 239                  * and try again.
 240                  *
 241                  * This is really something they should fix in the kernel! */
 242                 (void) usleep(50 * USEC_PER_MSEC);
 243
 244         }
 245
 246         return -EAGAIN; /* no success yet, try again */
 247 }
 248
 249 static int loop_wait_for_partitions_to_appear(
 250                 int fd,
 251                 sd_device *d,
 252                 unsigned num_partitions,
 253                 DissectImageFlags flags,
 254                 sd_device_enumerator **ret_enumerator) {
 255         _cleanup_(sd_device_unrefp) sd_device *device = NULL;
 256         int r;
 257
 258         assert(fd >= 0);
 259         assert(d);
 260         assert(ret_enumerator);
 261
 262         log_debug("Waiting for device (parent + %d partitions) to appear...", num_partitions);
 263
 264         if (!FLAGS_SET(flags, DISSECT_IMAGE_NO_UDEV)) {
 265                 r = device_wait_for_initialization(d, "block", USEC_INFINITY, &device);
 266                 if (r < 0)
 267                         return r;
 268         } else
 269                 device = sd_device_ref(d);
 270
 271         for (unsigned i = 0; i < N_DEVICE_NODE_LIST_ATTEMPTS; i++) {
 272                 r = wait_for_partitions_to_appear(fd, device, num_partitions, flags, ret_enumerator);
 273                 if (r != -EAGAIN)
 274                         return r;
 275         }
 276
 277         return log_debug_errno(SYNTHETIC_ERRNO(ENXIO),
 278                                "Kernel partitions dit not appear within %d attempts",
 279                                N_DEVICE_NODE_LIST_ATTEMPTS);
 280 }
 281
 282 static void check_partition_flags(
 283                 const char *node,
 284                 unsigned long long pflags,
 285                 unsigned long long supported) {
 286
 287         assert(node);
 288
 289         /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
 290         pflags &= ~(supported | GPT_FLAG_REQUIRED_PARTITION | GPT_FLAG_NO_BLOCK_IO_PROTOCOL | GPT_FLAG_LEGACY_BIOS_BOOTABLE);
 291
 292         if (pflags == 0)
 293                 return;
 294
 295         /* If there are other bits set, then log about it, to make things discoverable */
 296         for (unsigned i = 0; i < sizeof(pflags) * 8; i++) {
 297                 unsigned long long bit = 1ULL << i;
 298                 if (!FLAGS_SET(pflags, bit))
 299                         continue;
 300
 301                 log_debug("Unexpected partition flag %llu set on %s!", bit, node);
 302         }
 303 }
 304
 305 #endif
 306
 307 int dissect_image(
 308                 int fd,
 309                 const void *root_hash,
 310                 size_t root_hash_size,
 311                 DissectImageFlags flags,
 312                 DissectedImage **ret) {
 313
 314 #if HAVE_BLKID
 315         sd_id128_t root_uuid = SD_ID128_NULL, verity_uuid = SD_ID128_NULL;
 316         _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
 317         bool is_gpt, is_mbr, generic_rw, multiple_generic = false;
 318         _cleanup_(sd_device_unrefp) sd_device *d = NULL;
 319         _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
 320         _cleanup_(blkid_free_probep) blkid_probe b = NULL;
 321         _cleanup_free_ char *generic_node = NULL;
 322         sd_id128_t generic_uuid = SD_ID128_NULL;
 323         const char *pttype = NULL;
 324         blkid_partlist pl;
 325         int r, generic_nr;
 326         struct stat st;
 327         sd_device *q;
 328         unsigned i;
 329
 330         assert(fd >= 0);
 331         assert(ret);
 332         assert(root_hash || root_hash_size == 0);
 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                 const char *usage = NULL;
 398
 399                 (void) blkid_probe_lookup_value(b, "USAGE", &usage, NULL);
 400                 if (STRPTR_IN_SET(usage, "filesystem", "crypto")) {
 401                         _cleanup_free_ char *t = NULL, *n = NULL;
 402                         const char *fstype = NULL;
 403
 404                         /* OK, we have found a file system, that's our root partition then. */
 405                         (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
 406
 407                         if (fstype) {
 408                                 t = strdup(fstype);
 409                                 if (!t)
 410                                         return -ENOMEM;
 411                         }
 412
 413                         r = device_path_make_major_minor(st.st_mode, st.st_rdev, &n);
 414                         if (r < 0)
 415                                 return r;
 416
 417                         m->partitions[PARTITION_ROOT] = (DissectedPartition) {
 418                                 .found = true,
 419                                 .rw = true,
 420                                 .partno = -1,
 421                                 .architecture = _ARCHITECTURE_INVALID,
 422                                 .fstype = TAKE_PTR(t),
 423                                 .node = TAKE_PTR(n),
 424                         };
 425
 426                         m->encrypted = streq_ptr(fstype, "crypto_LUKS");
 427
 428                         if (!streq(usage, "filesystem")) {
 429                                 r = loop_wait_for_partitions_to_appear(fd, d, 0, flags, &e);
 430                                 if (r < 0)
 431                                         return r;
 432                         }
 433                         *ret = TAKE_PTR(m);
 434
 435                         return 0;
 436                 }
 437         }
 438
 439         (void) blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL);
 440         if (!pttype)
 441                 return -ENOPKG;
 442
 443         is_gpt = streq_ptr(pttype, "gpt");
 444         is_mbr = streq_ptr(pttype, "dos");
 445
 446         if (!is_gpt && ((flags & DISSECT_IMAGE_GPT_ONLY) || !is_mbr))
 447                 return -ENOPKG;
 448
 449         errno = 0;
 450         pl = blkid_probe_get_partitions(b);
 451         if (!pl)
 452                 return errno_or_else(ENOMEM);
 453
 454         r = loop_wait_for_partitions_to_appear(fd, d, blkid_partlist_numof_partitions(pl), flags, &e);
 455         if (r < 0)
 456                 return r;
 457
 458         FOREACH_DEVICE(e, q) {
 459                 unsigned long long pflags;
 460                 blkid_partition pp;
 461                 const char *node;
 462                 dev_t qn;
 463                 int nr;
 464
 465                 r = sd_device_get_devnum(q, &qn);
 466                 if (r < 0)
 467                         continue;
 468
 469                 if (st.st_rdev == qn)
 470                         continue;
 471
 472                 if (!device_is_block(q))
 473                         continue;
 474
 475                 if (device_is_mmc_special_partition(q))
 476                         continue;
 477
 478                 r = sd_device_get_devname(q, &node);
 479                 if (r < 0)
 480                         continue;
 481
 482                 pp = blkid_partlist_devno_to_partition(pl, qn);
 483                 if (!pp)
 484                         continue;
 485
 486                 pflags = blkid_partition_get_flags(pp);
 487
 488                 nr = blkid_partition_get_partno(pp);
 489                 if (nr < 0)
 490                         continue;
 491
 492                 if (is_gpt) {
 493                         int designator = _PARTITION_DESIGNATOR_INVALID, architecture = _ARCHITECTURE_INVALID;
 494                         const char *stype, *sid, *fstype = NULL;
 495                         sd_id128_t type_id, id;
 496                         bool rw = true;
 497
 498                         sid = blkid_partition_get_uuid(pp);
 499                         if (!sid)
 500                                 continue;
 501                         if (sd_id128_from_string(sid, &id) < 0)
 502                                 continue;
 503
 504                         stype = blkid_partition_get_type_string(pp);
 505                         if (!stype)
 506                                 continue;
 507                         if (sd_id128_from_string(stype, &type_id) < 0)
 508                                 continue;
 509
 510                         if (sd_id128_equal(type_id, GPT_HOME)) {
 511
 512                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 513
 514                                 if (pflags & GPT_FLAG_NO_AUTO)
 515                                         continue;
 516
 517                                 designator = PARTITION_HOME;
 518                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 519                         } else if (sd_id128_equal(type_id, GPT_SRV)) {
 520
 521                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 522
 523                                 if (pflags & GPT_FLAG_NO_AUTO)
 524                                         continue;
 525
 526                                 designator = PARTITION_SRV;
 527                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 528                         } else if (sd_id128_equal(type_id, GPT_ESP)) {
 529
 530                                 /* Note that we don't check the GPT_FLAG_NO_AUTO flag for the ESP, as it is not defined
 531                                  * there. We instead check the GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as recommended by the
 532                                  * UEFI spec (See "12.3.3 Number and Location of System Partitions"). */
 533
 534                                 if (pflags & GPT_FLAG_NO_BLOCK_IO_PROTOCOL)
 535                                         continue;
 536
 537                                 designator = PARTITION_ESP;
 538                                 fstype = "vfat";
 539
 540                         } else if (sd_id128_equal(type_id, GPT_XBOOTLDR)) {
 541
 542                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 543
 544                                 if (pflags & GPT_FLAG_NO_AUTO)
 545                                         continue;
 546
 547                                 designator = PARTITION_XBOOTLDR;
 548                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 549                         }
 550 #ifdef GPT_ROOT_NATIVE
 551                         else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE)) {
 552
 553                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 554
 555                                 if (pflags & GPT_FLAG_NO_AUTO)
 556                                         continue;
 557
 558                                 /* If a root ID is specified, ignore everything but the root id */
 559                                 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
 560                                         continue;
 561
 562                                 designator = PARTITION_ROOT;
 563                                 architecture = native_architecture();
 564                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 565                         } else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE_VERITY)) {
 566
 567                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 568
 569                                 if (pflags & GPT_FLAG_NO_AUTO)
 570                                         continue;
 571
 572                                 m->can_verity = true;
 573
 574                                 /* Ignore verity unless a root hash is specified */
 575                                 if (sd_id128_is_null(verity_uuid) || !sd_id128_equal(verity_uuid, id))
 576                                         continue;
 577
 578                                 designator = PARTITION_ROOT_VERITY;
 579                                 fstype = "DM_verity_hash";
 580                                 architecture = native_architecture();
 581                                 rw = false;
 582                         }
 583 #endif
 584 #ifdef GPT_ROOT_SECONDARY
 585                         else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY)) {
 586
 587                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 588
 589                                 if (pflags & GPT_FLAG_NO_AUTO)
 590                                         continue;
 591
 592                                 /* If a root ID is specified, ignore everything but the root id */
 593                                 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
 594                                         continue;
 595
 596                                 designator = PARTITION_ROOT_SECONDARY;
 597                                 architecture = SECONDARY_ARCHITECTURE;
 598                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 599                         } else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY_VERITY)) {
 600
 601                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 602
 603                                 if (pflags & GPT_FLAG_NO_AUTO)
 604                                         continue;
 605
 606                                 m->can_verity = true;
 607
 608                                 /* Ignore verity unless root has is specified */
 609                                 if (sd_id128_is_null(verity_uuid) || !sd_id128_equal(verity_uuid, id))
 610                                         continue;
 611
 612                                 designator = PARTITION_ROOT_SECONDARY_VERITY;
 613                                 fstype = "DM_verity_hash";
 614                                 architecture = SECONDARY_ARCHITECTURE;
 615                                 rw = false;
 616                         }
 617 #endif
 618                         else if (sd_id128_equal(type_id, GPT_SWAP)) {
 619
 620                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO);
 621
 622                                 if (pflags & GPT_FLAG_NO_AUTO)
 623                                         continue;
 624
 625                                 designator = PARTITION_SWAP;
 626                                 fstype = "swap";
 627                         } else if (sd_id128_equal(type_id, GPT_LINUX_GENERIC)) {
 628
 629                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 630
 631                                 if (pflags & GPT_FLAG_NO_AUTO)
 632                                         continue;
 633
 634                                 if (generic_node)
 635                                         multiple_generic = true;
 636                                 else {
 637                                         generic_nr = nr;
 638                                         generic_rw = !(pflags & GPT_FLAG_READ_ONLY);
 639                                         generic_uuid = id;
 640                                         generic_node = strdup(node);
 641                                         if (!generic_node)
 642                                                 return -ENOMEM;
 643                                 }
 644
 645                         } else if (sd_id128_equal(type_id, GPT_TMP)) {
 646
 647                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 648
 649                                 if (pflags & GPT_FLAG_NO_AUTO)
 650                                         continue;
 651
 652                                 designator = PARTITION_TMP;
 653                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 654
 655                         } else if (sd_id128_equal(type_id, GPT_VAR)) {
 656
 657                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 658
 659                                 if (pflags & GPT_FLAG_NO_AUTO)
 660                                         continue;
 661
 662                                 if (!FLAGS_SET(flags, DISSECT_IMAGE_RELAX_VAR_CHECK)) {
 663                                         sd_id128_t var_uuid;
 664
 665                                         /* For /var we insist that the uuid of the partition matches the
 666                                          * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
 667                                          * ID. Why? Unlike the other partitions /var is inherently
 668                                          * installation specific, hence we need to be careful not to mount it
 669                                          * in the wrong installation. By hashing the partition UUID from
 670                                          * /etc/machine-id we can securely bind the partition to the
 671                                          * installation. */
 672
 673                                         r = sd_id128_get_machine_app_specific(GPT_VAR, &var_uuid);
 674                                         if (r < 0)
 675                                                 return r;
 676
 677                                         if (!sd_id128_equal(var_uuid, id)) {
 678                                                 log_debug("Found a /var/ partition, but its UUID didn't match our expectations, ignoring.");
 679                                                 continue;
 680                                         }
 681                                 }
 682
 683                                 designator = PARTITION_VAR;
 684                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 685                         }
 686
 687                         if (designator != _PARTITION_DESIGNATOR_INVALID) {
 688                                 _cleanup_free_ char *t = NULL, *n = NULL;
 689
 690                                 /* First one wins */
 691                                 if (m->partitions[designator].found)
 692                                         continue;
 693
 694                                 if (fstype) {
 695                                         t = strdup(fstype);
 696                                         if (!t)
 697                                                 return -ENOMEM;
 698                                 }
 699
 700                                 n = strdup(node);
 701                                 if (!n)
 702                                         return -ENOMEM;
 703
 704                                 m->partitions[designator] = (DissectedPartition) {
 705                                         .found = true,
 706                                         .partno = nr,
 707                                         .rw = rw,
 708                                         .architecture = architecture,
 709                                         .node = TAKE_PTR(n),
 710                                         .fstype = TAKE_PTR(t),
 711                                         .uuid = id,
 712                                 };
 713                         }
 714
 715                 } else if (is_mbr) {
 716
 717                         switch (blkid_partition_get_type(pp)) {
 718
 719                         case 0x83: /* Linux partition */
 720
 721                                 if (pflags != 0x80) /* Bootable flag */
 722                                         continue;
 723
 724                                 if (generic_node)
 725                                         multiple_generic = true;
 726                                 else {
 727                                         generic_nr = nr;
 728                                         generic_rw = true;
 729                                         generic_node = strdup(node);
 730                                         if (!generic_node)
 731                                                 return -ENOMEM;
 732                                 }
 733
 734                                 break;
 735
 736                         case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
 737                                 _cleanup_free_ char *n = NULL;
 738                                 sd_id128_t id = SD_ID128_NULL;
 739                                 const char *sid;
 740
 741                                 /* First one wins */
 742                                 if (m->partitions[PARTITION_XBOOTLDR].found)
 743                                         continue;
 744
 745                                 sid = blkid_partition_get_uuid(pp);
 746                                 if (sid)
 747                                         (void) sd_id128_from_string(sid, &id);
 748
 749                                 n = strdup(node);
 750                                 if (!n)
 751                                         return -ENOMEM;
 752
 753                                 m->partitions[PARTITION_XBOOTLDR] = (DissectedPartition) {
 754                                         .found = true,
 755                                         .partno = nr,
 756                                         .rw = true,
 757                                         .architecture = _ARCHITECTURE_INVALID,
 758                                         .node = TAKE_PTR(n),
 759                                         .uuid = id,
 760                                 };
 761
 762                                 break;
 763                         }}
 764                 }
 765         }
 766
 767         if (!m->partitions[PARTITION_ROOT].found) {
 768                 /* No root partition found? Then let's see if ther's one for the secondary architecture. And if not
 769                  * either, then check if there's a single generic one, and use that. */
 770
 771                 if (m->partitions[PARTITION_ROOT_VERITY].found)
 772                         return -EADDRNOTAVAIL;
 773
 774                 if (m->partitions[PARTITION_ROOT_SECONDARY].found) {
 775                         m->partitions[PARTITION_ROOT] = m->partitions[PARTITION_ROOT_SECONDARY];
 776                         zero(m->partitions[PARTITION_ROOT_SECONDARY]);
 777
 778                         m->partitions[PARTITION_ROOT_VERITY] = m->partitions[PARTITION_ROOT_SECONDARY_VERITY];
 779                         zero(m->partitions[PARTITION_ROOT_SECONDARY_VERITY]);
 780
 781                 } else if (flags & DISSECT_IMAGE_REQUIRE_ROOT) {
 782
 783                         /* If the root has was set, then we won't fallback to a generic node, because the root hash
 784                          * decides */
 785                         if (root_hash)
 786                                 return -EADDRNOTAVAIL;
 787
 788                         /* If we didn't find a generic node, then we can't fix this up either */
 789                         if (!generic_node)
 790                                 return -ENXIO;
 791
 792                         /* If we didn't find a properly marked root partition, but we did find a single suitable
 793                          * generic Linux partition, then use this as root partition, if the caller asked for it. */
 794                         if (multiple_generic)
 795                                 return -ENOTUNIQ;
 796
 797                         m->partitions[PARTITION_ROOT] = (DissectedPartition) {
 798                                 .found = true,
 799                                 .rw = generic_rw,
 800                                 .partno = generic_nr,
 801                                 .architecture = _ARCHITECTURE_INVALID,
 802                                 .node = TAKE_PTR(generic_node),
 803                                 .uuid = generic_uuid,
 804                         };
 805                 }
 806         }
 807
 808         if (root_hash) {
 809                 if (!m->partitions[PARTITION_ROOT_VERITY].found || !m->partitions[PARTITION_ROOT].found)
 810                         return -EADDRNOTAVAIL;
 811
 812                 /* If we found the primary root with the hash, then we definitely want to suppress any secondary root
 813                  * (which would be weird, after all the root hash should only be assigned to one pair of
 814                  * partitions... */
 815                 m->partitions[PARTITION_ROOT_SECONDARY].found = false;
 816                 m->partitions[PARTITION_ROOT_SECONDARY_VERITY].found = false;
 817
 818                 /* If we found a verity setup, then the root partition is necessarily read-only. */
 819                 m->partitions[PARTITION_ROOT].rw = false;
 820
 821                 m->verity = true;
 822         }
 823
 824         blkid_free_probe(b);
 825         b = NULL;
 826
 827         /* Fill in file system types if we don't know them yet. */
 828         for (i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
 829                 DissectedPartition *p = m->partitions + i;
 830
 831                 if (!p->found)
 832                         continue;
 833
 834                 if (!p->fstype && p->node) {
 835                         r = probe_filesystem(p->node, &p->fstype);
 836                         if (r < 0 && r != -EUCLEAN)
 837                                 return r;
 838                 }
 839
 840                 if (streq_ptr(p->fstype, "crypto_LUKS"))
 841                         m->encrypted = true;
 842
 843                 if (p->fstype && fstype_is_ro(p->fstype))
 844                         p->rw = false;
 845         }
 846
 847         *ret = TAKE_PTR(m);
 848
 849         return 0;
 850 #else
 851         return -EOPNOTSUPP;
 852 #endif
 853 }
 854
 855 DissectedImage* dissected_image_unref(DissectedImage *m) {
 856         unsigned i;
 857
 858         if (!m)
 859                 return NULL;
 860
 861         for (i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
 862                 free(m->partitions[i].fstype);
 863                 free(m->partitions[i].node);
 864                 free(m->partitions[i].decrypted_fstype);
 865                 free(m->partitions[i].decrypted_node);
 866         }
 867
 868         free(m->hostname);
 869         strv_free(m->machine_info);
 870         strv_free(m->os_release);
 871
 872         return mfree(m);
 873 }
 874
 875 static int is_loop_device(const char *path) {
 876         char s[SYS_BLOCK_PATH_MAX("/../loop/")];
 877         struct stat st;
 878
 879         assert(path);
 880
 881         if (stat(path, &st) < 0)
 882                 return -errno;
 883
 884         if (!S_ISBLK(st.st_mode))
 885                 return -ENOTBLK;
 886
 887         xsprintf_sys_block_path(s, "/loop/", st.st_dev);
 888         if (access(s, F_OK) < 0) {
 889                 if (errno != ENOENT)
 890                         return -errno;
 891
 892                 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
 893                 xsprintf_sys_block_path(s, "/../loop/", st.st_dev);
 894                 if (access(s, F_OK) < 0)
 895                         return errno == ENOENT ? false : -errno;
 896         }
 897
 898         return true;
 899 }
 900
 901 static int run_fsck(const char *node, const char *fstype) {
 902         int r, exit_status;
 903         pid_t pid;
 904
 905         assert(node);
 906         assert(fstype);
 907
 908         r = fsck_exists(fstype);
 909         if (r < 0) {
 910                 log_debug_errno(r, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype);
 911                 return 0;
 912         }
 913         if (r == 0) {
 914                 log_debug("Not checking partition %s, as fsck for %s does not exist.", node, fstype);
 915                 return 0;
 916         }
 917
 918         r = safe_fork("(fsck)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG|FORK_NULL_STDIO, &pid);
 919         if (r < 0)
 920                 return log_debug_errno(r, "Failed to fork off fsck: %m");
 921         if (r == 0) {
 922                 /* Child */
 923                 execl("/sbin/fsck", "/sbin/fsck", "-aT", node, NULL);
 924                 log_debug_errno(errno, "Failed to execl() fsck: %m");
 925                 _exit(FSCK_OPERATIONAL_ERROR);
 926         }
 927
 928         exit_status = wait_for_terminate_and_check("fsck", pid, 0);
 929         if (exit_status < 0)
 930                 return log_debug_errno(exit_status, "Failed to fork off /sbin/fsck: %m");
 931
 932         if ((exit_status & ~FSCK_ERROR_CORRECTED) != FSCK_SUCCESS) {
 933                 log_debug("fsck failed with exit status %i.", exit_status);
 934
 935                 if ((exit_status & (FSCK_SYSTEM_SHOULD_REBOOT|FSCK_ERRORS_LEFT_UNCORRECTED)) != 0)
 936                         return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN), "File system is corrupted, refusing.");
 937
 938                 log_debug("Ignoring fsck error.");
 939         }
 940
 941         return 0;
 942 }
 943
 944 static int mount_partition(
 945                 DissectedPartition *m,
 946                 const char *where,
 947                 const char *directory,
 948                 uid_t uid_shift,
 949                 DissectImageFlags flags) {
 950
 951         _cleanup_free_ char *chased = NULL, *options = NULL;
 952         const char *p, *node, *fstype;
 953         bool rw;
 954         int r;
 955
 956         assert(m);
 957         assert(where);
 958
 959         node = m->decrypted_node ?: m->node;
 960         fstype = m->decrypted_fstype ?: m->fstype;
 961
 962         if (!m->found || !node || !fstype)
 963                 return 0;
 964
 965         /* Stacked encryption? Yuck */
 966         if (streq_ptr(fstype, "crypto_LUKS"))
 967                 return -ELOOP;
 968
 969         rw = m->rw && !(flags & DISSECT_IMAGE_READ_ONLY);
 970
 971         if (FLAGS_SET(flags, DISSECT_IMAGE_FSCK) && rw) {
 972                 r = run_fsck(node, fstype);
 973                 if (r < 0)
 974                         return r;
 975         }
 976
 977         if (directory) {
 978                 r = chase_symlinks(directory, where, CHASE_PREFIX_ROOT, &chased, NULL);
 979                 if (r < 0)
 980                         return r;
 981
 982                 p = chased;
 983         } else
 984                 p = where;
 985
 986         /* If requested, turn on discard support. */
 987         if (fstype_can_discard(fstype) &&
 988             ((flags & DISSECT_IMAGE_DISCARD) ||
 989              ((flags & DISSECT_IMAGE_DISCARD_ON_LOOP) && is_loop_device(m->node)))) {
 990                 options = strdup("discard");
 991                 if (!options)
 992                         return -ENOMEM;
 993         }
 994
 995         if (uid_is_valid(uid_shift) && uid_shift != 0 && fstype_can_uid_gid(fstype)) {
 996                 _cleanup_free_ char *uid_option = NULL;
 997
 998                 if (asprintf(&uid_option, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0)
 999                         return -ENOMEM;
1000
1001                 if (!strextend_with_separator(&options, ",", uid_option, NULL))
1002                         return -ENOMEM;
1003         }
1004
1005         r = mount_verbose(LOG_DEBUG, node, p, fstype, MS_NODEV|(rw ? 0 : MS_RDONLY), options);
1006         if (r < 0)
1007                 return r;
1008
1009         return 1;
1010 }
1011
1012 int dissected_image_mount(DissectedImage *m, const char *where, uid_t uid_shift, DissectImageFlags flags) {
1013         int r, boot_mounted;
1014
1015         assert(m);
1016         assert(where);
1017
1018         if (!m->partitions[PARTITION_ROOT].found)
1019                 return -ENXIO;
1020
1021         if ((flags & DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY) == 0) {
1022                 r = mount_partition(m->partitions + PARTITION_ROOT, where, NULL, uid_shift, flags);
1023                 if (r < 0)
1024                         return r;
1025
1026                 if (flags & DISSECT_IMAGE_VALIDATE_OS) {
1027                         r = path_is_os_tree(where);
1028                         if (r < 0)
1029                                 return r;
1030                         if (r == 0)
1031                                 return -EMEDIUMTYPE;
1032                 }
1033         }
1034
1035         if (flags & DISSECT_IMAGE_MOUNT_ROOT_ONLY)
1036                 return 0;
1037
1038         r = mount_partition(m->partitions + PARTITION_HOME, where, "/home", uid_shift, flags);
1039         if (r < 0)
1040                 return r;
1041
1042         r = mount_partition(m->partitions + PARTITION_SRV, where, "/srv", uid_shift, flags);
1043         if (r < 0)
1044                 return r;
1045
1046         r = mount_partition(m->partitions + PARTITION_VAR, where, "/var", uid_shift, flags);
1047         if (r < 0)
1048                 return r;
1049
1050         r = mount_partition(m->partitions + PARTITION_TMP, where, "/var/tmp", uid_shift, flags);
1051         if (r < 0)
1052                 return r;
1053
1054         boot_mounted = mount_partition(m->partitions + PARTITION_XBOOTLDR, where, "/boot", uid_shift, flags);
1055         if (boot_mounted < 0)
1056                 return boot_mounted;
1057
1058         if (m->partitions[PARTITION_ESP].found) {
1059                 /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it
1060                  * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */
1061
1062                 r = chase_symlinks("/efi", where, CHASE_PREFIX_ROOT, NULL, NULL);
1063                 if (r >= 0) {
1064                         r = mount_partition(m->partitions + PARTITION_ESP, where, "/efi", uid_shift, flags);
1065                         if (r < 0)
1066                                 return r;
1067
1068                 } else if (boot_mounted <= 0) {
1069                         _cleanup_free_ char *p = NULL;
1070
1071                         r = chase_symlinks("/boot", where, CHASE_PREFIX_ROOT, &p, NULL);
1072                         if (r >= 0 && dir_is_empty(p) > 0) {
1073                                 r = mount_partition(m->partitions + PARTITION_ESP, where, "/boot", uid_shift, flags);
1074                                 if (r < 0)
1075                                         return r;
1076                         }
1077                 }
1078         }
1079
1080         return 0;
1081 }
1082
1083 #if HAVE_LIBCRYPTSETUP
1084 typedef struct DecryptedPartition {
1085         struct crypt_device *device;
1086         char *name;
1087         bool relinquished;
1088 } DecryptedPartition;
1089
1090 struct DecryptedImage {
1091         DecryptedPartition *decrypted;
1092         size_t n_decrypted;
1093         size_t n_allocated;
1094 };
1095 #endif
1096
1097 DecryptedImage* decrypted_image_unref(DecryptedImage* d) {
1098 #if HAVE_LIBCRYPTSETUP
1099         size_t i;
1100         int r;
1101
1102         if (!d)
1103                 return NULL;
1104
1105         for (i = 0; i < d->n_decrypted; i++) {
1106                 DecryptedPartition *p = d->decrypted + i;
1107
1108                 if (p->device && p->name && !p->relinquished) {
1109                         r = crypt_deactivate(p->device, p->name);
1110                         if (r < 0)
1111                                 log_debug_errno(r, "Failed to deactivate encrypted partition %s", p->name);
1112                 }
1113
1114                 if (p->device)
1115                         crypt_free(p->device);
1116                 free(p->name);
1117         }
1118
1119         free(d);
1120 #endif
1121         return NULL;
1122 }
1123
1124 #if HAVE_LIBCRYPTSETUP
1125
1126 static int make_dm_name_and_node(const void *original_node, const char *suffix, char **ret_name, char **ret_node) {
1127         _cleanup_free_ char *name = NULL, *node = NULL;
1128         const char *base;
1129
1130         assert(original_node);
1131         assert(suffix);
1132         assert(ret_name);
1133         assert(ret_node);
1134
1135         base = strrchr(original_node, '/');
1136         if (!base)
1137                 return -EINVAL;
1138         base++;
1139         if (isempty(base))
1140                 return -EINVAL;
1141
1142         name = strjoin(base, suffix);
1143         if (!name)
1144                 return -ENOMEM;
1145         if (!filename_is_valid(name))
1146                 return -EINVAL;
1147
1148         node = path_join(crypt_get_dir(), name);
1149         if (!node)
1150                 return -ENOMEM;
1151
1152         *ret_name = TAKE_PTR(name);
1153         *ret_node = TAKE_PTR(node);
1154
1155         return 0;
1156 }
1157
1158 static int decrypt_partition(
1159                 DissectedPartition *m,
1160                 const char *passphrase,
1161                 DissectImageFlags flags,
1162                 DecryptedImage *d) {
1163
1164         _cleanup_free_ char *node = NULL, *name = NULL;
1165         _cleanup_(crypt_freep) struct crypt_device *cd = NULL;
1166         int r;
1167
1168         assert(m);
1169         assert(d);
1170
1171         if (!m->found || !m->node || !m->fstype)
1172                 return 0;
1173
1174         if (!streq(m->fstype, "crypto_LUKS"))
1175                 return 0;
1176
1177         if (!passphrase)
1178                 return -ENOKEY;
1179
1180         r = make_dm_name_and_node(m->node, "-decrypted", &name, &node);
1181         if (r < 0)
1182                 return r;
1183
1184         if (!GREEDY_REALLOC0(d->decrypted, d->n_allocated, d->n_decrypted + 1))
1185                 return -ENOMEM;
1186
1187         r = crypt_init(&cd, m->node);
1188         if (r < 0)
1189                 return log_debug_errno(r, "Failed to initialize dm-crypt: %m");
1190
1191         r = crypt_load(cd, CRYPT_LUKS, NULL);
1192         if (r < 0)
1193                 return log_debug_errno(r, "Failed to load LUKS metadata: %m");
1194
1195         r = crypt_activate_by_passphrase(cd, name, CRYPT_ANY_SLOT, passphrase, strlen(passphrase),
1196                                          ((flags & DISSECT_IMAGE_READ_ONLY) ? CRYPT_ACTIVATE_READONLY : 0) |
1197                                          ((flags & DISSECT_IMAGE_DISCARD_ON_CRYPTO) ? CRYPT_ACTIVATE_ALLOW_DISCARDS : 0));
1198         if (r < 0) {
1199                 log_debug_errno(r, "Failed to activate LUKS device: %m");
1200                 return r == -EPERM ? -EKEYREJECTED : r;
1201         }
1202
1203         d->decrypted[d->n_decrypted].name = TAKE_PTR(name);
1204         d->decrypted[d->n_decrypted].device = TAKE_PTR(cd);
1205         d->n_decrypted++;
1206
1207         m->decrypted_node = TAKE_PTR(node);
1208
1209         return 0;
1210 }
1211
1212 static int verity_partition(
1213                 DissectedPartition *m,
1214                 DissectedPartition *v,
1215                 const void *root_hash,
1216                 size_t root_hash_size,
1217                 DissectImageFlags flags,
1218                 DecryptedImage *d) {
1219
1220         _cleanup_free_ char *node = NULL, *name = NULL;
1221         _cleanup_(crypt_freep) struct crypt_device *cd = NULL;
1222         int r;
1223
1224         assert(m);
1225         assert(v);
1226
1227         if (!root_hash)
1228                 return 0;
1229
1230         if (!m->found || !m->node || !m->fstype)
1231                 return 0;
1232         if (!v->found || !v->node || !v->fstype)
1233                 return 0;
1234
1235         if (!streq(v->fstype, "DM_verity_hash"))
1236                 return 0;
1237
1238         r = make_dm_name_and_node(m->node, "-verity", &name, &node);
1239         if (r < 0)
1240                 return r;
1241
1242         if (!GREEDY_REALLOC0(d->decrypted, d->n_allocated, d->n_decrypted + 1))
1243                 return -ENOMEM;
1244
1245         r = crypt_init(&cd, v->node);
1246         if (r < 0)
1247                 return r;
1248
1249         r = crypt_load(cd, CRYPT_VERITY, NULL);
1250         if (r < 0)
1251                 return r;
1252
1253         r = crypt_set_data_device(cd, m->node);
1254         if (r < 0)
1255                 return r;
1256
1257         r = crypt_activate_by_volume_key(cd, name, root_hash, root_hash_size, CRYPT_ACTIVATE_READONLY);
1258         if (r < 0)
1259                 return r;
1260
1261         d->decrypted[d->n_decrypted].name = TAKE_PTR(name);
1262         d->decrypted[d->n_decrypted].device = TAKE_PTR(cd);
1263         d->n_decrypted++;
1264
1265         m->decrypted_node = TAKE_PTR(node);
1266
1267         return 0;
1268 }
1269 #endif
1270
1271 int dissected_image_decrypt(
1272                 DissectedImage *m,
1273                 const char *passphrase,
1274                 const void *root_hash,
1275                 size_t root_hash_size,
1276                 DissectImageFlags flags,
1277                 DecryptedImage **ret) {
1278
1279 #if HAVE_LIBCRYPTSETUP
1280         _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL;
1281         unsigned i;
1282         int r;
1283 #endif
1284
1285         assert(m);
1286         assert(root_hash || root_hash_size == 0);
1287
1288         /* Returns:
1289          *
1290          *      = 0           → There was nothing to decrypt
1291          *      > 0           → Decrypted successfully
1292          *      -ENOKEY       → There's something to decrypt but no key was supplied
1293          *      -EKEYREJECTED → Passed key was not correct
1294          */
1295
1296         if (root_hash && root_hash_size < sizeof(sd_id128_t))
1297                 return -EINVAL;
1298
1299         if (!m->encrypted && !m->verity) {
1300                 *ret = NULL;
1301                 return 0;
1302         }
1303
1304 #if HAVE_LIBCRYPTSETUP
1305         d = new0(DecryptedImage, 1);
1306         if (!d)
1307                 return -ENOMEM;
1308
1309         for (i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
1310                 DissectedPartition *p = m->partitions + i;
1311                 int k;
1312
1313                 if (!p->found)
1314                         continue;
1315
1316                 r = decrypt_partition(p, passphrase, flags, d);
1317                 if (r < 0)
1318                         return r;
1319
1320                 k = PARTITION_VERITY_OF(i);
1321                 if (k >= 0) {
1322                         r = verity_partition(p, m->partitions + k, root_hash, root_hash_size, flags, d);
1323                         if (r < 0)
1324                                 return r;
1325                 }
1326
1327                 if (!p->decrypted_fstype && p->decrypted_node) {
1328                         r = probe_filesystem(p->decrypted_node, &p->decrypted_fstype);
1329                         if (r < 0 && r != -EUCLEAN)
1330                                 return r;
1331                 }
1332         }
1333
1334         *ret = TAKE_PTR(d);
1335
1336         return 1;
1337 #else
1338         return -EOPNOTSUPP;
1339 #endif
1340 }
1341
1342 int dissected_image_decrypt_interactively(
1343                 DissectedImage *m,
1344                 const char *passphrase,
1345                 const void *root_hash,
1346                 size_t root_hash_size,
1347                 DissectImageFlags flags,
1348                 DecryptedImage **ret) {
1349
1350         _cleanup_strv_free_erase_ char **z = NULL;
1351         int n = 3, r;
1352
1353         if (passphrase)
1354                 n--;
1355
1356         for (;;) {
1357                 r = dissected_image_decrypt(m, passphrase, root_hash, root_hash_size, flags, ret);
1358                 if (r >= 0)
1359                         return r;
1360                 if (r == -EKEYREJECTED)
1361                         log_error_errno(r, "Incorrect passphrase, try again!");
1362                 else if (r != -ENOKEY)
1363                         return log_error_errno(r, "Failed to decrypt image: %m");
1364
1365                 if (--n < 0)
1366                         return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED),
1367                                                "Too many retries.");
1368
1369                 z = strv_free(z);
1370
1371                 r = ask_password_auto("Please enter image passphrase:", NULL, "dissect", "dissect", USEC_INFINITY, 0, &z);
1372                 if (r < 0)
1373                         return log_error_errno(r, "Failed to query for passphrase: %m");
1374
1375                 passphrase = z[0];
1376         }
1377 }
1378
1379 int decrypted_image_relinquish(DecryptedImage *d) {
1380
1381 #if HAVE_LIBCRYPTSETUP
1382         size_t i;
1383         int r;
1384 #endif
1385
1386         assert(d);
1387
1388         /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a boolean so
1389          * that we don't clean it up ourselves either anymore */
1390
1391 #if HAVE_LIBCRYPTSETUP
1392         for (i = 0; i < d->n_decrypted; i++) {
1393                 DecryptedPartition *p = d->decrypted + i;
1394
1395                 if (p->relinquished)
1396                         continue;
1397
1398                 r = dm_deferred_remove(p->name);
1399                 if (r < 0)
1400                         return log_debug_errno(r, "Failed to mark %s for auto-removal: %m", p->name);
1401
1402                 p->relinquished = true;
1403         }
1404 #endif
1405
1406         return 0;
1407 }
1408
1409 int root_hash_load(const char *image, void **ret, size_t *ret_size) {
1410         _cleanup_free_ char *text = NULL;
1411         _cleanup_free_ void *k = NULL;
1412         size_t l;
1413         int r;
1414
1415         assert(image);
1416         assert(ret);
1417         assert(ret_size);
1418
1419         if (is_device_path(image)) {
1420                 /* If we are asked to load the root hash for a device node, exit early */
1421                 *ret = NULL;
1422                 *ret_size = 0;
1423                 return 0;
1424         }
1425
1426         r = getxattr_malloc(image, "user.verity.roothash", &text, true);
1427         if (r < 0) {
1428                 char *fn, *e, *n;
1429
1430                 if (!IN_SET(r, -ENODATA, -EOPNOTSUPP, -ENOENT))
1431                         return r;
1432
1433                 fn = newa(char, strlen(image) + STRLEN(".roothash") + 1);
1434                 n = stpcpy(fn, image);
1435                 e = endswith(fn, ".raw");
1436                 if (e)
1437                         n = e;
1438
1439                 strcpy(n, ".roothash");
1440
1441                 r = read_one_line_file(fn, &text);
1442                 if (r == -ENOENT) {
1443                         *ret = NULL;
1444                         *ret_size = 0;
1445                         return 0;
1446                 }
1447                 if (r < 0)
1448                         return r;
1449         }
1450
1451         r = unhexmem(text, strlen(text), &k, &l);
1452         if (r < 0)
1453                 return r;
1454         if (l < sizeof(sd_id128_t))
1455                 return -EINVAL;
1456
1457         *ret = TAKE_PTR(k);
1458         *ret_size = l;
1459
1460         return 1;
1461 }
1462
1463 int dissected_image_acquire_metadata(DissectedImage *m) {
1464
1465         enum {
1466                 META_HOSTNAME,
1467                 META_MACHINE_ID,
1468                 META_MACHINE_INFO,
1469                 META_OS_RELEASE,
1470                 _META_MAX,
1471         };
1472
1473         static const char *const paths[_META_MAX] = {
1474                 [META_HOSTNAME]     = "/etc/hostname\0",
1475                 [META_MACHINE_ID]   = "/etc/machine-id\0",
1476                 [META_MACHINE_INFO] = "/etc/machine-info\0",
1477                 [META_OS_RELEASE]   = "/etc/os-release\0"
1478                                       "/usr/lib/os-release\0",
1479         };
1480
1481         _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL;
1482         _cleanup_(rmdir_and_freep) char *t = NULL;
1483         _cleanup_(sigkill_waitp) pid_t child = 0;
1484         sd_id128_t machine_id = SD_ID128_NULL;
1485         _cleanup_free_ char *hostname = NULL;
1486         unsigned n_meta_initialized = 0, k;
1487         int fds[2 * _META_MAX], r;
1488
1489         BLOCK_SIGNALS(SIGCHLD);
1490
1491         assert(m);
1492
1493         for (; n_meta_initialized < _META_MAX; n_meta_initialized ++)
1494                 if (pipe2(fds + 2*n_meta_initialized, O_CLOEXEC) < 0) {
1495                         r = -errno;
1496                         goto finish;
1497                 }
1498
1499         r = mkdtemp_malloc("/tmp/dissect-XXXXXX", &t);
1500         if (r < 0)
1501                 goto finish;
1502
1503         r = safe_fork("(sd-dissect)", FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE, &child);
1504         if (r < 0)
1505                 goto finish;
1506         if (r == 0) {
1507                 r = dissected_image_mount(m, t, UID_INVALID, DISSECT_IMAGE_READ_ONLY|DISSECT_IMAGE_MOUNT_ROOT_ONLY|DISSECT_IMAGE_VALIDATE_OS);
1508                 if (r < 0) {
1509                         log_debug_errno(r, "Failed to mount dissected image: %m");
1510                         _exit(EXIT_FAILURE);
1511                 }
1512
1513                 for (k = 0; k < _META_MAX; k++) {
1514                         _cleanup_close_ int fd = -1;
1515                         const char *p;
1516
1517                         fds[2*k] = safe_close(fds[2*k]);
1518
1519                         NULSTR_FOREACH(p, paths[k]) {
1520                                 fd = chase_symlinks_and_open(p, t, CHASE_PREFIX_ROOT, O_RDONLY|O_CLOEXEC|O_NOCTTY, NULL);
1521                                 if (fd >= 0)
1522                                         break;
1523                         }
1524                         if (fd < 0) {
1525                                 log_debug_errno(fd, "Failed to read %s file of image, ignoring: %m", paths[k]);
1526                                 continue;
1527                         }
1528
1529                         r = copy_bytes(fd, fds[2*k+1], (uint64_t) -1, 0);
1530                         if (r < 0)
1531                                 _exit(EXIT_FAILURE);
1532
1533                         fds[2*k+1] = safe_close(fds[2*k+1]);
1534                 }
1535
1536                 _exit(EXIT_SUCCESS);
1537         }
1538
1539         for (k = 0; k < _META_MAX; k++) {
1540                 _cleanup_fclose_ FILE *f = NULL;
1541
1542                 fds[2*k+1] = safe_close(fds[2*k+1]);
1543
1544                 f = fdopen(fds[2*k], "r");
1545                 if (!f) {
1546                         r = -errno;
1547                         goto finish;
1548                 }
1549
1550                 fds[2*k] = -1;
1551
1552                 switch (k) {
1553
1554                 case META_HOSTNAME:
1555                         r = read_etc_hostname_stream(f, &hostname);
1556                         if (r < 0)
1557                                 log_debug_errno(r, "Failed to read /etc/hostname: %m");
1558
1559                         break;
1560
1561                 case META_MACHINE_ID: {
1562                         _cleanup_free_ char *line = NULL;
1563
1564                         r = read_line(f, LONG_LINE_MAX, &line);
1565                         if (r < 0)
1566                                 log_debug_errno(r, "Failed to read /etc/machine-id: %m");
1567                         else if (r == 33) {
1568                                 r = sd_id128_from_string(line, &machine_id);
1569                                 if (r < 0)
1570                                         log_debug_errno(r, "Image contains invalid /etc/machine-id: %s", line);
1571                         } else if (r == 0)
1572                                 log_debug("/etc/machine-id file is empty.");
1573                         else
1574                                 log_debug("/etc/machine-id has unexpected length %i.", r);
1575
1576                         break;
1577                 }
1578
1579                 case META_MACHINE_INFO:
1580                         r = load_env_file_pairs(f, "machine-info", &machine_info);
1581                         if (r < 0)
1582                                 log_debug_errno(r, "Failed to read /etc/machine-info: %m");
1583
1584                         break;
1585
1586                 case META_OS_RELEASE:
1587                         r = load_env_file_pairs(f, "os-release", &os_release);
1588                         if (r < 0)
1589                                 log_debug_errno(r, "Failed to read OS release file: %m");
1590
1591                         break;
1592                 }
1593         }
1594
1595         r = wait_for_terminate_and_check("(sd-dissect)", child, 0);
1596         child = 0;
1597         if (r < 0)
1598                 goto finish;
1599         if (r != EXIT_SUCCESS)
1600                 return -EPROTO;
1601
1602         free_and_replace(m->hostname, hostname);
1603         m->machine_id = machine_id;
1604         strv_free_and_replace(m->machine_info, machine_info);
1605         strv_free_and_replace(m->os_release, os_release);
1606
1607 finish:
1608         for (k = 0; k < n_meta_initialized; k++)
1609                 safe_close_pair(fds + 2*k);
1610
1611         return r;
1612 }
1613
1614 int dissect_image_and_warn(
1615                 int fd,
1616                 const char *name,
1617                 const void *root_hash,
1618                 size_t root_hash_size,
1619                 DissectImageFlags flags,
1620                 DissectedImage **ret) {
1621
1622         _cleanup_free_ char *buffer = NULL;
1623         int r;
1624
1625         if (!name) {
1626                 r = fd_get_path(fd, &buffer);
1627                 if (r < 0)
1628                         return r;
1629
1630                 name = buffer;
1631         }
1632
1633         r = dissect_image(fd, root_hash, root_hash_size, flags, ret);
1634
1635         switch (r) {
1636
1637         case -EOPNOTSUPP:
1638                 return log_error_errno(r, "Dissecting images is not supported, compiled without blkid support.");
1639
1640         case -ENOPKG:
1641                 return log_error_errno(r, "Couldn't identify a suitable partition table or file system in '%s'.", name);
1642
1643         case -EADDRNOTAVAIL:
1644                 return log_error_errno(r, "No root partition for specified root hash found in '%s'.", name);
1645
1646         case -ENOTUNIQ:
1647                 return log_error_errno(r, "Multiple suitable root partitions found in image '%s'.", name);
1648
1649         case -ENXIO:
1650                 return log_error_errno(r, "No suitable root partition found in image '%s'.", name);
1651
1652         case -EPROTONOSUPPORT:
1653                 return log_error_errno(r, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name);
1654
1655         default:
1656                 if (r < 0)
1657                         return log_error_errno(r, "Failed to dissect image '%s': %m", name);
1658
1659                 return r;
1660         }
1661 }
1662
1663 static const char *const partition_designator_table[] = {
1664         [PARTITION_ROOT] = "root",
1665         [PARTITION_ROOT_SECONDARY] = "root-secondary",
1666         [PARTITION_HOME] = "home",
1667         [PARTITION_SRV] = "srv",
1668         [PARTITION_ESP] = "esp",
1669         [PARTITION_XBOOTLDR] = "xbootldr",
1670         [PARTITION_SWAP] = "swap",
1671         [PARTITION_ROOT_VERITY] = "root-verity",
1672         [PARTITION_ROOT_SECONDARY_VERITY] = "root-secondary-verity",
1673         [PARTITION_TMP] = "tmp",
1674         [PARTITION_VAR] = "var",
1675 };
1676
1677 DEFINE_STRING_TABLE_LOOKUP(partition_designator, int);