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                 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN),
  80                                        "Results ambiguous for partition %s", node);
  81         if (r != 0)
  82                 return errno_or_else(EIO);
  83
  84         (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
  85
  86         if (fstype) {
  87                 char *t;
  88
  89                 t = strdup(fstype);
  90                 if (!t)
  91                         return -ENOMEM;
  92
  93                 *ret_fstype = t;
  94                 return 1;
  95         }
  96
  97 not_found:
  98         *ret_fstype = NULL;
  99         return 0;
 100 #else
 101         return -EOPNOTSUPP;
 102 #endif
 103 }
 104
 105 #if HAVE_BLKID
 106 /* Detect RPMB and Boot partitions, which are not listed by blkid.
 107  * See https://github.com/systemd/systemd/issues/5806. */
 108 static bool device_is_mmc_special_partition(sd_device *d) {
 109         const char *sysname;
 110
 111         assert(d);
 112
 113         if (sd_device_get_sysname(d, &sysname) < 0)
 114                 return false;
 115
 116         return startswith(sysname, "mmcblk") &&
 117                 (endswith(sysname, "rpmb") || endswith(sysname, "boot0") || endswith(sysname, "boot1"));
 118 }
 119
 120 static bool device_is_block(sd_device *d) {
 121         const char *ss;
 122
 123         assert(d);
 124
 125         if (sd_device_get_subsystem(d, &ss) < 0)
 126                 return false;
 127
 128         return streq(ss, "block");
 129 }
 130
 131 static int enumerator_for_parent(sd_device *d, sd_device_enumerator **ret) {
 132         _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
 133         int r;
 134
 135         assert(d);
 136         assert(ret);
 137
 138         r = sd_device_enumerator_new(&e);
 139         if (r < 0)
 140                 return r;
 141
 142         r = sd_device_enumerator_allow_uninitialized(e);
 143         if (r < 0)
 144                 return r;
 145
 146         r = sd_device_enumerator_add_match_parent(e, d);
 147         if (r < 0)
 148                 return r;
 149
 150         *ret = TAKE_PTR(e);
 151         return 0;
 152 }
 153
 154 /* how many times to wait for the device nodes to appear */
 155 #define N_DEVICE_NODE_LIST_ATTEMPTS 10
 156
 157 static int wait_for_partitions_to_appear(
 158                 int fd,
 159                 sd_device *d,
 160                 unsigned num_partitions,
 161                 DissectImageFlags flags,
 162                 sd_device_enumerator **ret_enumerator) {
 163
 164         _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
 165         sd_device *q;
 166         unsigned n;
 167         int r;
 168
 169         assert(fd >= 0);
 170         assert(d);
 171         assert(ret_enumerator);
 172
 173         r = enumerator_for_parent(d, &e);
 174         if (r < 0)
 175                 return r;
 176
 177         /* Count the partitions enumerated by the kernel */
 178         n = 0;
 179         FOREACH_DEVICE(e, q) {
 180                 if (sd_device_get_devnum(q, NULL) < 0)
 181                         continue;
 182                 if (!device_is_block(q))
 183                         continue;
 184                 if (device_is_mmc_special_partition(q))
 185                         continue;
 186
 187                 if (!FLAGS_SET(flags, DISSECT_IMAGE_NO_UDEV)) {
 188                         r = device_wait_for_initialization(q, "block", USEC_INFINITY, NULL);
 189                         if (r < 0)
 190                                 return r;
 191                 }
 192
 193                 n++;
 194         }
 195
 196         if (n == num_partitions + 1) {
 197                 *ret_enumerator = TAKE_PTR(e);
 198                 return 0; /* success! */
 199         }
 200         if (n > num_partitions + 1)
 201                 return log_debug_errno(SYNTHETIC_ERRNO(EIO),
 202                                        "blkid and kernel partition lists do not match.");
 203
 204         /* The kernel has probed fewer partitions than blkid? Maybe the kernel prober is still running or it
 205          * got EBUSY because udev already opened the device. Let's reprobe the device, which is a synchronous
 206          * call that waits until probing is complete. */
 207
 208         for (unsigned j = 0; ; j++) {
 209                 if (j++ > 20)
 210                         return -EBUSY;
 211
 212                 if (ioctl(fd, BLKRRPART, 0) >= 0)
 213                         break;
 214                 r = -errno;
 215                 if (r == -EINVAL) {
 216                         struct loop_info64 info;
 217
 218                         /* If we are running on a loop device that has partition scanning off, return
 219                          * an explicit recognizable error about this, so that callers can generate a
 220                          * proper message explaining the situation. */
 221
 222                         if (ioctl(fd, LOOP_GET_STATUS64, &info) >= 0) {
 223 #if HAVE_VALGRIND_MEMCHECK_H
 224                                 /* Valgrind currently doesn't know LOOP_GET_STATUS64. Remove this once it does */
 225                                 VALGRIND_MAKE_MEM_DEFINED(&info, sizeof(info));
 226 #endif
 227
 228                                 if ((info.lo_flags & LO_FLAGS_PARTSCAN) == 0)
 229                                         return log_debug_errno(EPROTONOSUPPORT,
 230                                                                "Device is a loop device and partition scanning is off!");
 231                         }
 232                 }
 233                 if (r != -EBUSY)
 234                         return r;
 235
 236                 /* If something else has the device open, such as an udev rule, the ioctl will return
 237                  * EBUSY. Since there's no way to wait until it isn't busy anymore, let's just wait a bit,
 238                  * and try again.
 239                  *
 240                  * This is really something they should fix in the kernel! */
 241                 (void) usleep(50 * USEC_PER_MSEC);
 242
 243         }
 244
 245         return -EAGAIN; /* no success yet, try again */
 246 }
 247
 248 static int loop_wait_for_partitions_to_appear(
 249                 int fd,
 250                 sd_device *d,
 251                 unsigned num_partitions,
 252                 DissectImageFlags flags,
 253                 sd_device_enumerator **ret_enumerator) {
 254         _cleanup_(sd_device_unrefp) sd_device *device = NULL;
 255         int r;
 256
 257         assert(fd >= 0);
 258         assert(d);
 259         assert(ret_enumerator);
 260
 261         log_debug("Waiting for device (parent + %d partitions) to appear...", num_partitions);
 262
 263         if (!FLAGS_SET(flags, DISSECT_IMAGE_NO_UDEV)) {
 264                 r = device_wait_for_initialization(d, "block", USEC_INFINITY, &device);
 265                 if (r < 0)
 266                         return r;
 267         } else
 268                 device = sd_device_ref(d);
 269
 270         for (unsigned i = 0; i < N_DEVICE_NODE_LIST_ATTEMPTS; i++) {
 271                 r = wait_for_partitions_to_appear(fd, device, num_partitions, flags, ret_enumerator);
 272                 if (r != -EAGAIN)
 273                         return r;
 274         }
 275
 276         return log_debug_errno(SYNTHETIC_ERRNO(ENXIO),
 277                                "Kernel partitions dit not appear within %d attempts",
 278                                N_DEVICE_NODE_LIST_ATTEMPTS);
 279 }
 280
 281 static void check_partition_flags(
 282                 const char *node,
 283                 unsigned long long pflags,
 284                 unsigned long long supported) {
 285
 286         assert(node);
 287
 288         /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
 289         pflags &= ~(supported | GPT_FLAG_REQUIRED_PARTITION | GPT_FLAG_NO_BLOCK_IO_PROTOCOL | GPT_FLAG_LEGACY_BIOS_BOOTABLE);
 290
 291         if (pflags == 0)
 292                 return;
 293
 294         /* If there are other bits set, then log about it, to make things discoverable */
 295         for (unsigned i = 0; i < sizeof(pflags) * 8; i++) {
 296                 unsigned long long bit = 1ULL << i;
 297                 if (!FLAGS_SET(pflags, bit))
 298                         continue;
 299
 300                 log_debug("Unexpected partition flag %llu set on %s!", bit, node);
 301         }
 302 }
 303
 304 #endif
 305
 306 int dissect_image(
 307                 int fd,
 308                 const void *root_hash,
 309                 size_t root_hash_size,
 310                 DissectImageFlags flags,
 311                 DissectedImage **ret) {
 312
 313 #if HAVE_BLKID
 314         sd_id128_t root_uuid = SD_ID128_NULL, verity_uuid = SD_ID128_NULL;
 315         _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
 316         bool is_gpt, is_mbr, generic_rw, multiple_generic = false;
 317         _cleanup_(sd_device_unrefp) sd_device *d = NULL;
 318         _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
 319         _cleanup_(blkid_free_probep) blkid_probe b = NULL;
 320         _cleanup_free_ char *generic_node = NULL;
 321         sd_id128_t generic_uuid = SD_ID128_NULL;
 322         const char *pttype = NULL;
 323         blkid_partlist pl;
 324         int r, generic_nr;
 325         struct stat st;
 326         sd_device *q;
 327         unsigned i;
 328
 329         assert(fd >= 0);
 330         assert(ret);
 331         assert(root_hash || root_hash_size == 0);
 332
 333         /* Probes a disk image, and returns information about what it found in *ret.
 334          *
 335          * Returns -ENOPKG if no suitable partition table or file system could be found.
 336          * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found. */
 337
 338         if (root_hash) {
 339                 /* If a root hash is supplied, then we use the root partition that has a UUID that match the first
 340                  * 128bit of the root hash. And we use the verity partition that has a UUID that match the final
 341                  * 128bit. */
 342
 343                 if (root_hash_size < sizeof(sd_id128_t))
 344                         return -EINVAL;
 345
 346                 memcpy(&root_uuid, root_hash, sizeof(sd_id128_t));
 347                 memcpy(&verity_uuid, (const uint8_t*) root_hash + root_hash_size - sizeof(sd_id128_t), sizeof(sd_id128_t));
 348
 349                 if (sd_id128_is_null(root_uuid))
 350                         return -EINVAL;
 351                 if (sd_id128_is_null(verity_uuid))
 352                         return -EINVAL;
 353         }
 354
 355         if (fstat(fd, &st) < 0)
 356                 return -errno;
 357
 358         if (!S_ISBLK(st.st_mode))
 359                 return -ENOTBLK;
 360
 361         b = blkid_new_probe();
 362         if (!b)
 363                 return -ENOMEM;
 364
 365         errno = 0;
 366         r = blkid_probe_set_device(b, fd, 0, 0);
 367         if (r != 0)
 368                 return errno_or_else(ENOMEM);
 369
 370         if ((flags & DISSECT_IMAGE_GPT_ONLY) == 0) {
 371                 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
 372                 blkid_probe_enable_superblocks(b, 1);
 373                 blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE|BLKID_SUBLKS_USAGE);
 374         }
 375
 376         blkid_probe_enable_partitions(b, 1);
 377         blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS);
 378
 379         errno = 0;
 380         r = blkid_do_safeprobe(b);
 381         if (IN_SET(r, -2, 1))
 382                 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG), "Failed to identify any partition table.");
 383         if (r != 0)
 384                 return errno_or_else(EIO);
 385
 386         m = new0(DissectedImage, 1);
 387         if (!m)
 388                 return -ENOMEM;
 389
 390         r = sd_device_new_from_devnum(&d, 'b', st.st_rdev);
 391         if (r < 0)
 392                 return r;
 393
 394         if (!(flags & DISSECT_IMAGE_GPT_ONLY) &&
 395             (flags & DISSECT_IMAGE_REQUIRE_ROOT)) {
 396                 const char *usage = NULL;
 397
 398                 (void) blkid_probe_lookup_value(b, "USAGE", &usage, NULL);
 399                 if (STRPTR_IN_SET(usage, "filesystem", "crypto")) {
 400                         _cleanup_free_ char *t = NULL, *n = NULL;
 401                         const char *fstype = NULL;
 402
 403                         /* OK, we have found a file system, that's our root partition then. */
 404                         (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
 405
 406                         if (fstype) {
 407                                 t = strdup(fstype);
 408                                 if (!t)
 409                                         return -ENOMEM;
 410                         }
 411
 412                         r = device_path_make_major_minor(st.st_mode, st.st_rdev, &n);
 413                         if (r < 0)
 414                                 return r;
 415
 416                         m->partitions[PARTITION_ROOT] = (DissectedPartition) {
 417                                 .found = true,
 418                                 .rw = true,
 419                                 .partno = -1,
 420                                 .architecture = _ARCHITECTURE_INVALID,
 421                                 .fstype = TAKE_PTR(t),
 422                                 .node = TAKE_PTR(n),
 423                         };
 424
 425                         m->encrypted = streq_ptr(fstype, "crypto_LUKS");
 426
 427                         if (!streq(usage, "filesystem")) {
 428                                 r = loop_wait_for_partitions_to_appear(fd, d, 0, flags, &e);
 429                                 if (r < 0)
 430                                         return r;
 431                         }
 432                         *ret = TAKE_PTR(m);
 433
 434                         return 0;
 435                 }
 436         }
 437
 438         (void) blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL);
 439         if (!pttype)
 440                 return -ENOPKG;
 441
 442         is_gpt = streq_ptr(pttype, "gpt");
 443         is_mbr = streq_ptr(pttype, "dos");
 444
 445         if (!is_gpt && ((flags & DISSECT_IMAGE_GPT_ONLY) || !is_mbr))
 446                 return -ENOPKG;
 447
 448         errno = 0;
 449         pl = blkid_probe_get_partitions(b);
 450         if (!pl)
 451                 return errno_or_else(ENOMEM);
 452
 453         r = loop_wait_for_partitions_to_appear(fd, d, blkid_partlist_numof_partitions(pl), flags, &e);
 454         if (r < 0)
 455                 return r;
 456
 457         FOREACH_DEVICE(e, q) {
 458                 unsigned long long pflags;
 459                 blkid_partition pp;
 460                 const char *node;
 461                 dev_t qn;
 462                 int nr;
 463
 464                 r = sd_device_get_devnum(q, &qn);
 465                 if (r < 0)
 466                         continue;
 467
 468                 if (st.st_rdev == qn)
 469                         continue;
 470
 471                 if (!device_is_block(q))
 472                         continue;
 473
 474                 if (device_is_mmc_special_partition(q))
 475                         continue;
 476
 477                 r = sd_device_get_devname(q, &node);
 478                 if (r < 0)
 479                         continue;
 480
 481                 pp = blkid_partlist_devno_to_partition(pl, qn);
 482                 if (!pp)
 483                         continue;
 484
 485                 pflags = blkid_partition_get_flags(pp);
 486
 487                 nr = blkid_partition_get_partno(pp);
 488                 if (nr < 0)
 489                         continue;
 490
 491                 if (is_gpt) {
 492                         int designator = _PARTITION_DESIGNATOR_INVALID, architecture = _ARCHITECTURE_INVALID;
 493                         const char *stype, *sid, *fstype = NULL;
 494                         sd_id128_t type_id, id;
 495                         bool rw = true;
 496
 497                         sid = blkid_partition_get_uuid(pp);
 498                         if (!sid)
 499                                 continue;
 500                         if (sd_id128_from_string(sid, &id) < 0)
 501                                 continue;
 502
 503                         stype = blkid_partition_get_type_string(pp);
 504                         if (!stype)
 505                                 continue;
 506                         if (sd_id128_from_string(stype, &type_id) < 0)
 507                                 continue;
 508
 509                         if (sd_id128_equal(type_id, GPT_HOME)) {
 510
 511                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 512
 513                                 if (pflags & GPT_FLAG_NO_AUTO)
 514                                         continue;
 515
 516                                 designator = PARTITION_HOME;
 517                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 518                         } else if (sd_id128_equal(type_id, GPT_SRV)) {
 519
 520                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 521
 522                                 if (pflags & GPT_FLAG_NO_AUTO)
 523                                         continue;
 524
 525                                 designator = PARTITION_SRV;
 526                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 527                         } else if (sd_id128_equal(type_id, GPT_ESP)) {
 528
 529                                 /* Note that we don't check the GPT_FLAG_NO_AUTO flag for the ESP, as it is not defined
 530                                  * there. We instead check the GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as recommended by the
 531                                  * UEFI spec (See "12.3.3 Number and Location of System Partitions"). */
 532
 533                                 if (pflags & GPT_FLAG_NO_BLOCK_IO_PROTOCOL)
 534                                         continue;
 535
 536                                 designator = PARTITION_ESP;
 537                                 fstype = "vfat";
 538
 539                         } else if (sd_id128_equal(type_id, GPT_XBOOTLDR)) {
 540
 541                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 542
 543                                 if (pflags & GPT_FLAG_NO_AUTO)
 544                                         continue;
 545
 546                                 designator = PARTITION_XBOOTLDR;
 547                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 548                         }
 549 #ifdef GPT_ROOT_NATIVE
 550                         else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE)) {
 551
 552                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 553
 554                                 if (pflags & GPT_FLAG_NO_AUTO)
 555                                         continue;
 556
 557                                 /* If a root ID is specified, ignore everything but the root id */
 558                                 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
 559                                         continue;
 560
 561                                 designator = PARTITION_ROOT;
 562                                 architecture = native_architecture();
 563                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 564                         } else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE_VERITY)) {
 565
 566                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 567
 568                                 if (pflags & GPT_FLAG_NO_AUTO)
 569                                         continue;
 570
 571                                 m->can_verity = true;
 572
 573                                 /* Ignore verity unless a root hash is specified */
 574                                 if (sd_id128_is_null(verity_uuid) || !sd_id128_equal(verity_uuid, id))
 575                                         continue;
 576
 577                                 designator = PARTITION_ROOT_VERITY;
 578                                 fstype = "DM_verity_hash";
 579                                 architecture = native_architecture();
 580                                 rw = false;
 581                         }
 582 #endif
 583 #ifdef GPT_ROOT_SECONDARY
 584                         else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY)) {
 585
 586                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 587
 588                                 if (pflags & GPT_FLAG_NO_AUTO)
 589                                         continue;
 590
 591                                 /* If a root ID is specified, ignore everything but the root id */
 592                                 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
 593                                         continue;
 594
 595                                 designator = PARTITION_ROOT_SECONDARY;
 596                                 architecture = SECONDARY_ARCHITECTURE;
 597                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 598                         } else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY_VERITY)) {
 599
 600                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 601
 602                                 if (pflags & GPT_FLAG_NO_AUTO)
 603                                         continue;
 604
 605                                 m->can_verity = true;
 606
 607                                 /* Ignore verity unless root has is specified */
 608                                 if (sd_id128_is_null(verity_uuid) || !sd_id128_equal(verity_uuid, id))
 609                                         continue;
 610
 611                                 designator = PARTITION_ROOT_SECONDARY_VERITY;
 612                                 fstype = "DM_verity_hash";
 613                                 architecture = SECONDARY_ARCHITECTURE;
 614                                 rw = false;
 615                         }
 616 #endif
 617                         else if (sd_id128_equal(type_id, GPT_SWAP)) {
 618
 619                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO);
 620
 621                                 if (pflags & GPT_FLAG_NO_AUTO)
 622                                         continue;
 623
 624                                 designator = PARTITION_SWAP;
 625                                 fstype = "swap";
 626                         } else if (sd_id128_equal(type_id, GPT_LINUX_GENERIC)) {
 627
 628                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 629
 630                                 if (pflags & GPT_FLAG_NO_AUTO)
 631                                         continue;
 632
 633                                 if (generic_node)
 634                                         multiple_generic = true;
 635                                 else {
 636                                         generic_nr = nr;
 637                                         generic_rw = !(pflags & GPT_FLAG_READ_ONLY);
 638                                         generic_uuid = id;
 639                                         generic_node = strdup(node);
 640                                         if (!generic_node)
 641                                                 return -ENOMEM;
 642                                 }
 643
 644                         } else if (sd_id128_equal(type_id, GPT_TMP)) {
 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                                 designator = PARTITION_TMP;
 652                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 653
 654                         } else if (sd_id128_equal(type_id, GPT_VAR)) {
 655
 656                                 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
 657
 658                                 if (pflags & GPT_FLAG_NO_AUTO)
 659                                         continue;
 660
 661                                 if (!FLAGS_SET(flags, DISSECT_IMAGE_RELAX_VAR_CHECK)) {
 662                                         sd_id128_t var_uuid;
 663
 664                                         /* For /var we insist that the uuid of the partition matches the
 665                                          * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
 666                                          * ID. Why? Unlike the other partitions /var is inherently
 667                                          * installation specific, hence we need to be careful not to mount it
 668                                          * in the wrong installation. By hashing the partition UUID from
 669                                          * /etc/machine-id we can securely bind the partition to the
 670                                          * installation. */
 671
 672                                         r = sd_id128_get_machine_app_specific(GPT_VAR, &var_uuid);
 673                                         if (r < 0)
 674                                                 return r;
 675
 676                                         if (!sd_id128_equal(var_uuid, id)) {
 677                                                 log_debug("Found a /var/ partition, but its UUID didn't match our expectations, ignoring.");
 678                                                 continue;
 679                                         }
 680                                 }
 681
 682                                 designator = PARTITION_VAR;
 683                                 rw = !(pflags & GPT_FLAG_READ_ONLY);
 684                         }
 685
 686                         if (designator != _PARTITION_DESIGNATOR_INVALID) {
 687                                 _cleanup_free_ char *t = NULL, *n = NULL;
 688
 689                                 /* First one wins */
 690                                 if (m->partitions[designator].found)
 691                                         continue;
 692
 693                                 if (fstype) {
 694                                         t = strdup(fstype);
 695                                         if (!t)
 696                                                 return -ENOMEM;
 697                                 }
 698
 699                                 n = strdup(node);
 700                                 if (!n)
 701                                         return -ENOMEM;
 702
 703                                 m->partitions[designator] = (DissectedPartition) {
 704                                         .found = true,
 705                                         .partno = nr,
 706                                         .rw = rw,
 707                                         .architecture = architecture,
 708                                         .node = TAKE_PTR(n),
 709                                         .fstype = TAKE_PTR(t),
 710                                         .uuid = id,
 711                                 };
 712                         }
 713
 714                 } else if (is_mbr) {
 715
 716                         switch (blkid_partition_get_type(pp)) {
 717
 718                         case 0x83: /* Linux partition */
 719
 720                                 if (pflags != 0x80) /* Bootable flag */
 721                                         continue;
 722
 723                                 if (generic_node)
 724                                         multiple_generic = true;
 725                                 else {
 726                                         generic_nr = nr;
 727                                         generic_rw = true;
 728                                         generic_node = strdup(node);
 729                                         if (!generic_node)
 730                                                 return -ENOMEM;
 731                                 }
 732
 733                                 break;
 734
 735                         case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
 736                                 _cleanup_free_ char *n = NULL;
 737                                 sd_id128_t id = SD_ID128_NULL;
 738                                 const char *sid;
 739
 740                                 /* First one wins */
 741                                 if (m->partitions[PARTITION_XBOOTLDR].found)
 742                                         continue;
 743
 744                                 sid = blkid_partition_get_uuid(pp);
 745                                 if (sid)
 746                                         (void) sd_id128_from_string(sid, &id);
 747
 748                                 n = strdup(node);
 749                                 if (!n)
 750                                         return -ENOMEM;
 751
 752                                 m->partitions[PARTITION_XBOOTLDR] = (DissectedPartition) {
 753                                         .found = true,
 754                                         .partno = nr,
 755                                         .rw = true,
 756                                         .architecture = _ARCHITECTURE_INVALID,
 757                                         .node = TAKE_PTR(n),
 758                                         .uuid = id,
 759                                 };
 760
 761                                 break;
 762                         }}
 763                 }
 764         }
 765
 766         if (!m->partitions[PARTITION_ROOT].found) {
 767                 /* No root partition found? Then let's see if ther's one for the secondary architecture. And if not
 768                  * either, then check if there's a single generic one, and use that. */
 769
 770                 if (m->partitions[PARTITION_ROOT_VERITY].found)
 771                         return -EADDRNOTAVAIL;
 772
 773                 if (m->partitions[PARTITION_ROOT_SECONDARY].found) {
 774                         m->partitions[PARTITION_ROOT] = m->partitions[PARTITION_ROOT_SECONDARY];
 775                         zero(m->partitions[PARTITION_ROOT_SECONDARY]);
 776
 777                         m->partitions[PARTITION_ROOT_VERITY] = m->partitions[PARTITION_ROOT_SECONDARY_VERITY];
 778                         zero(m->partitions[PARTITION_ROOT_SECONDARY_VERITY]);
 779
 780                 } else if (flags & DISSECT_IMAGE_REQUIRE_ROOT) {
 781
 782                         /* If the root has was set, then we won't fallback to a generic node, because the root hash
 783                          * decides */
 784                         if (root_hash)
 785                                 return -EADDRNOTAVAIL;
 786
 787                         /* If we didn't find a generic node, then we can't fix this up either */
 788                         if (!generic_node)
 789                                 return -ENXIO;
 790
 791                         /* If we didn't find a properly marked root partition, but we did find a single suitable
 792                          * generic Linux partition, then use this as root partition, if the caller asked for it. */
 793                         if (multiple_generic)
 794                                 return -ENOTUNIQ;
 795
 796                         m->partitions[PARTITION_ROOT] = (DissectedPartition) {
 797                                 .found = true,
 798                                 .rw = generic_rw,
 799                                 .partno = generic_nr,
 800                                 .architecture = _ARCHITECTURE_INVALID,
 801                                 .node = TAKE_PTR(generic_node),
 802                                 .uuid = generic_uuid,
 803                         };
 804                 }
 805         }
 806
 807         if (root_hash) {
 808                 if (!m->partitions[PARTITION_ROOT_VERITY].found || !m->partitions[PARTITION_ROOT].found)
 809                         return -EADDRNOTAVAIL;
 810
 811                 /* If we found the primary root with the hash, then we definitely want to suppress any secondary root
 812                  * (which would be weird, after all the root hash should only be assigned to one pair of
 813                  * partitions... */
 814                 m->partitions[PARTITION_ROOT_SECONDARY].found = false;
 815                 m->partitions[PARTITION_ROOT_SECONDARY_VERITY].found = false;
 816
 817                 /* If we found a verity setup, then the root partition is necessarily read-only. */
 818                 m->partitions[PARTITION_ROOT].rw = false;
 819
 820                 m->verity = true;
 821         }
 822
 823         blkid_free_probe(b);
 824         b = NULL;
 825
 826         /* Fill in file system types if we don't know them yet. */
 827         for (i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
 828                 DissectedPartition *p = m->partitions + i;
 829
 830                 if (!p->found)
 831                         continue;
 832
 833                 if (!p->fstype && p->node) {
 834                         r = probe_filesystem(p->node, &p->fstype);
 835                         if (r < 0 && r != -EUCLEAN)
 836                                 return r;
 837                 }
 838
 839                 if (streq_ptr(p->fstype, "crypto_LUKS"))
 840                         m->encrypted = true;
 841
 842                 if (p->fstype && fstype_is_ro(p->fstype))
 843                         p->rw = false;
 844         }
 845
 846         *ret = TAKE_PTR(m);
 847
 848         return 0;
 849 #else
 850         return -EOPNOTSUPP;
 851 #endif
 852 }
 853
 854 DissectedImage* dissected_image_unref(DissectedImage *m) {
 855         unsigned i;
 856
 857         if (!m)
 858                 return NULL;
 859
 860         for (i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
 861                 free(m->partitions[i].fstype);
 862                 free(m->partitions[i].node);
 863                 free(m->partitions[i].decrypted_fstype);
 864                 free(m->partitions[i].decrypted_node);
 865         }
 866
 867         free(m->hostname);
 868         strv_free(m->machine_info);
 869         strv_free(m->os_release);
 870
 871         return mfree(m);
 872 }
 873
 874 static int is_loop_device(const char *path) {
 875         char s[SYS_BLOCK_PATH_MAX("/../loop/")];
 876         struct stat st;
 877
 878         assert(path);
 879
 880         if (stat(path, &st) < 0)
 881                 return -errno;
 882
 883         if (!S_ISBLK(st.st_mode))
 884                 return -ENOTBLK;
 885
 886         xsprintf_sys_block_path(s, "/loop/", st.st_dev);
 887         if (access(s, F_OK) < 0) {
 888                 if (errno != ENOENT)
 889                         return -errno;
 890
 891                 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
 892                 xsprintf_sys_block_path(s, "/../loop/", st.st_dev);
 893                 if (access(s, F_OK) < 0)
 894                         return errno == ENOENT ? false : -errno;
 895         }
 896
 897         return true;
 898 }
 899
 900 static int run_fsck(const char *node, const char *fstype) {
 901         int r, exit_status;
 902         pid_t pid;
 903
 904         assert(node);
 905         assert(fstype);
 906
 907         r = fsck_exists(fstype);
 908         if (r < 0) {
 909                 log_debug_errno(r, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype);
 910                 return 0;
 911         }
 912         if (r == 0) {
 913                 log_debug("Not checking partition %s, as fsck for %s does not exist.", node, fstype);
 914                 return 0;
 915         }
 916
 917         r = safe_fork("(fsck)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG|FORK_NULL_STDIO, &pid);
 918         if (r < 0)
 919                 return log_debug_errno(r, "Failed to fork off fsck: %m");
 920         if (r == 0) {
 921                 /* Child */
 922                 execl("/sbin/fsck", "/sbin/fsck", "-aT", node, NULL);
 923                 log_debug_errno(errno, "Failed to execl() fsck: %m");
 924                 _exit(FSCK_OPERATIONAL_ERROR);
 925         }
 926
 927         exit_status = wait_for_terminate_and_check("fsck", pid, 0);
 928         if (exit_status < 0)
 929                 return log_debug_errno(exit_status, "Failed to fork off /sbin/fsck: %m");
 930
 931         if ((exit_status & ~FSCK_ERROR_CORRECTED) != FSCK_SUCCESS) {
 932                 log_debug("fsck failed with exit status %i.", exit_status);
 933
 934                 if ((exit_status & (FSCK_SYSTEM_SHOULD_REBOOT|FSCK_ERRORS_LEFT_UNCORRECTED)) != 0)
 935                         return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN), "File system is corrupted, refusing.");
 936
 937                 log_debug("Ignoring fsck error.");
 938         }
 939
 940         return 0;
 941 }
 942
 943 static int mount_partition(
 944                 DissectedPartition *m,
 945                 const char *where,
 946                 const char *directory,
 947                 uid_t uid_shift,
 948                 DissectImageFlags flags) {
 949
 950         _cleanup_free_ char *chased = NULL, *options = NULL;
 951         const char *p, *node, *fstype;
 952         bool rw;
 953         int r;
 954
 955         assert(m);
 956         assert(where);
 957
 958         node = m->decrypted_node ?: m->node;
 959         fstype = m->decrypted_fstype ?: m->fstype;
 960
 961         if (!m->found || !node || !fstype)
 962                 return 0;
 963
 964         /* Stacked encryption? Yuck */
 965         if (streq_ptr(fstype, "crypto_LUKS"))
 966                 return -ELOOP;
 967
 968         rw = m->rw && !(flags & DISSECT_IMAGE_READ_ONLY);
 969
 970         if (FLAGS_SET(flags, DISSECT_IMAGE_FSCK) && rw) {
 971                 r = run_fsck(node, fstype);
 972                 if (r < 0)
 973                         return r;
 974         }
 975
 976         if (directory) {
 977                 r = chase_symlinks(directory, where, CHASE_PREFIX_ROOT, &chased, NULL);
 978                 if (r < 0)
 979                         return r;
 980
 981                 p = chased;
 982         } else
 983                 p = where;
 984
 985         /* If requested, turn on discard support. */
 986         if (fstype_can_discard(fstype) &&
 987             ((flags & DISSECT_IMAGE_DISCARD) ||
 988              ((flags & DISSECT_IMAGE_DISCARD_ON_LOOP) && is_loop_device(m->node)))) {
 989                 options = strdup("discard");
 990                 if (!options)
 991                         return -ENOMEM;
 992         }
 993
 994         if (uid_is_valid(uid_shift) && uid_shift != 0 && fstype_can_uid_gid(fstype)) {
 995                 _cleanup_free_ char *uid_option = NULL;
 996
 997                 if (asprintf(&uid_option, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0)
 998                         return -ENOMEM;
 999
1000                 if (!strextend_with_separator(&options, ",", uid_option, NULL))
1001                         return -ENOMEM;
1002         }
1003
1004         r = mount_verbose(LOG_DEBUG, node, p, fstype, MS_NODEV|(rw ? 0 : MS_RDONLY), options);
1005         if (r < 0)
1006                 return r;
1007
1008         return 1;
1009 }
1010
1011 int dissected_image_mount(DissectedImage *m, const char *where, uid_t uid_shift, DissectImageFlags flags) {
1012         int r, boot_mounted;
1013
1014         assert(m);
1015         assert(where);
1016
1017         if (!m->partitions[PARTITION_ROOT].found)
1018                 return -ENXIO;
1019
1020         if ((flags & DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY) == 0) {
1021                 r = mount_partition(m->partitions + PARTITION_ROOT, where, NULL, uid_shift, flags);
1022                 if (r < 0)
1023                         return r;
1024
1025                 if (flags & DISSECT_IMAGE_VALIDATE_OS) {
1026                         r = path_is_os_tree(where);
1027                         if (r < 0)
1028                                 return r;
1029                         if (r == 0)
1030                                 return -EMEDIUMTYPE;
1031                 }
1032         }
1033
1034         if (flags & DISSECT_IMAGE_MOUNT_ROOT_ONLY)
1035                 return 0;
1036
1037         r = mount_partition(m->partitions + PARTITION_HOME, where, "/home", uid_shift, flags);
1038         if (r < 0)
1039                 return r;
1040
1041         r = mount_partition(m->partitions + PARTITION_SRV, where, "/srv", uid_shift, flags);
1042         if (r < 0)
1043                 return r;
1044
1045         r = mount_partition(m->partitions + PARTITION_VAR, where, "/var", uid_shift, flags);
1046         if (r < 0)
1047                 return r;
1048
1049         r = mount_partition(m->partitions + PARTITION_TMP, where, "/var/tmp", uid_shift, flags);
1050         if (r < 0)
1051                 return r;
1052
1053         boot_mounted = mount_partition(m->partitions + PARTITION_XBOOTLDR, where, "/boot", uid_shift, flags);
1054         if (boot_mounted < 0)
1055                 return boot_mounted;
1056
1057         if (m->partitions[PARTITION_ESP].found) {
1058                 /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it
1059                  * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */
1060
1061                 r = chase_symlinks("/efi", where, CHASE_PREFIX_ROOT, NULL, NULL);
1062                 if (r >= 0) {
1063                         r = mount_partition(m->partitions + PARTITION_ESP, where, "/efi", uid_shift, flags);
1064                         if (r < 0)
1065                                 return r;
1066
1067                 } else if (boot_mounted <= 0) {
1068                         _cleanup_free_ char *p = NULL;
1069
1070                         r = chase_symlinks("/boot", where, CHASE_PREFIX_ROOT, &p, NULL);
1071                         if (r >= 0 && dir_is_empty(p) > 0) {
1072                                 r = mount_partition(m->partitions + PARTITION_ESP, where, "/boot", uid_shift, flags);
1073                                 if (r < 0)
1074                                         return r;
1075                         }
1076                 }
1077         }
1078
1079         return 0;
1080 }
1081
1082 #if HAVE_LIBCRYPTSETUP
1083 typedef struct DecryptedPartition {
1084         struct crypt_device *device;
1085         char *name;
1086         bool relinquished;
1087 } DecryptedPartition;
1088
1089 struct DecryptedImage {
1090         DecryptedPartition *decrypted;
1091         size_t n_decrypted;
1092         size_t n_allocated;
1093 };
1094 #endif
1095
1096 DecryptedImage* decrypted_image_unref(DecryptedImage* d) {
1097 #if HAVE_LIBCRYPTSETUP
1098         size_t i;
1099         int r;
1100
1101         if (!d)
1102                 return NULL;
1103
1104         for (i = 0; i < d->n_decrypted; i++) {
1105                 DecryptedPartition *p = d->decrypted + i;
1106
1107                 if (p->device && p->name && !p->relinquished) {
1108                         r = crypt_deactivate(p->device, p->name);
1109                         if (r < 0)
1110                                 log_debug_errno(r, "Failed to deactivate encrypted partition %s", p->name);
1111                 }
1112
1113                 if (p->device)
1114                         crypt_free(p->device);
1115                 free(p->name);
1116         }
1117
1118         free(d);
1119 #endif
1120         return NULL;
1121 }
1122
1123 #if HAVE_LIBCRYPTSETUP
1124
1125 static int make_dm_name_and_node(const void *original_node, const char *suffix, char **ret_name, char **ret_node) {
1126         _cleanup_free_ char *name = NULL, *node = NULL;
1127         const char *base;
1128
1129         assert(original_node);
1130         assert(suffix);
1131         assert(ret_name);
1132         assert(ret_node);
1133
1134         base = strrchr(original_node, '/');
1135         if (!base)
1136                 return -EINVAL;
1137         base++;
1138         if (isempty(base))
1139                 return -EINVAL;
1140
1141         name = strjoin(base, suffix);
1142         if (!name)
1143                 return -ENOMEM;
1144         if (!filename_is_valid(name))
1145                 return -EINVAL;
1146
1147         node = path_join(crypt_get_dir(), name);
1148         if (!node)
1149                 return -ENOMEM;
1150
1151         *ret_name = TAKE_PTR(name);
1152         *ret_node = TAKE_PTR(node);
1153
1154         return 0;
1155 }
1156
1157 static int decrypt_partition(
1158                 DissectedPartition *m,
1159                 const char *passphrase,
1160                 DissectImageFlags flags,
1161                 DecryptedImage *d) {
1162
1163         _cleanup_free_ char *node = NULL, *name = NULL;
1164         _cleanup_(crypt_freep) struct crypt_device *cd = NULL;
1165         int r;
1166
1167         assert(m);
1168         assert(d);
1169
1170         if (!m->found || !m->node || !m->fstype)
1171                 return 0;
1172
1173         if (!streq(m->fstype, "crypto_LUKS"))
1174                 return 0;
1175
1176         if (!passphrase)
1177                 return -ENOKEY;
1178
1179         r = make_dm_name_and_node(m->node, "-decrypted", &name, &node);
1180         if (r < 0)
1181                 return r;
1182
1183         if (!GREEDY_REALLOC0(d->decrypted, d->n_allocated, d->n_decrypted + 1))
1184                 return -ENOMEM;
1185
1186         r = crypt_init(&cd, m->node);
1187         if (r < 0)
1188                 return log_debug_errno(r, "Failed to initialize dm-crypt: %m");
1189
1190         crypt_set_log_callback(cd, cryptsetup_log_glue, NULL);
1191
1192         r = crypt_load(cd, CRYPT_LUKS, NULL);
1193         if (r < 0)
1194                 return log_debug_errno(r, "Failed to load LUKS metadata: %m");
1195
1196         r = crypt_activate_by_passphrase(cd, name, CRYPT_ANY_SLOT, passphrase, strlen(passphrase),
1197                                          ((flags & DISSECT_IMAGE_READ_ONLY) ? CRYPT_ACTIVATE_READONLY : 0) |
1198                                          ((flags & DISSECT_IMAGE_DISCARD_ON_CRYPTO) ? CRYPT_ACTIVATE_ALLOW_DISCARDS : 0));
1199         if (r < 0) {
1200                 log_debug_errno(r, "Failed to activate LUKS device: %m");
1201                 return r == -EPERM ? -EKEYREJECTED : r;
1202         }
1203
1204         d->decrypted[d->n_decrypted].name = TAKE_PTR(name);
1205         d->decrypted[d->n_decrypted].device = TAKE_PTR(cd);
1206         d->n_decrypted++;
1207
1208         m->decrypted_node = TAKE_PTR(node);
1209
1210         return 0;
1211 }
1212
1213 static int verity_partition(
1214                 DissectedPartition *m,
1215                 DissectedPartition *v,
1216                 const void *root_hash,
1217                 size_t root_hash_size,
1218                 DissectImageFlags flags,
1219                 DecryptedImage *d) {
1220
1221         _cleanup_free_ char *node = NULL, *name = NULL;
1222         _cleanup_(crypt_freep) struct crypt_device *cd = NULL;
1223         int r;
1224
1225         assert(m);
1226         assert(v);
1227
1228         if (!root_hash)
1229                 return 0;
1230
1231         if (!m->found || !m->node || !m->fstype)
1232                 return 0;
1233         if (!v->found || !v->node || !v->fstype)
1234                 return 0;
1235
1236         if (!streq(v->fstype, "DM_verity_hash"))
1237                 return 0;
1238
1239         r = make_dm_name_and_node(m->node, "-verity", &name, &node);
1240         if (r < 0)
1241                 return r;
1242
1243         if (!GREEDY_REALLOC0(d->decrypted, d->n_allocated, d->n_decrypted + 1))
1244                 return -ENOMEM;
1245
1246         r = crypt_init(&cd, v->node);
1247         if (r < 0)
1248                 return r;
1249
1250         crypt_set_log_callback(cd, cryptsetup_log_glue, NULL);
1251
1252         r = crypt_load(cd, CRYPT_VERITY, NULL);
1253         if (r < 0)
1254                 return r;
1255
1256         r = crypt_set_data_device(cd, m->node);
1257         if (r < 0)
1258                 return r;
1259
1260         r = crypt_activate_by_volume_key(cd, name, root_hash, root_hash_size, CRYPT_ACTIVATE_READONLY);
1261         if (r < 0)
1262                 return r;
1263
1264         d->decrypted[d->n_decrypted].name = TAKE_PTR(name);
1265         d->decrypted[d->n_decrypted].device = TAKE_PTR(cd);
1266         d->n_decrypted++;
1267
1268         m->decrypted_node = TAKE_PTR(node);
1269
1270         return 0;
1271 }
1272 #endif
1273
1274 int dissected_image_decrypt(
1275                 DissectedImage *m,
1276                 const char *passphrase,
1277                 const void *root_hash,
1278                 size_t root_hash_size,
1279                 DissectImageFlags flags,
1280                 DecryptedImage **ret) {
1281
1282 #if HAVE_LIBCRYPTSETUP
1283         _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL;
1284         unsigned i;
1285         int r;
1286 #endif
1287
1288         assert(m);
1289         assert(root_hash || root_hash_size == 0);
1290
1291         /* Returns:
1292          *
1293          *      = 0           → There was nothing to decrypt
1294          *      > 0           → Decrypted successfully
1295          *      -ENOKEY       → There's something to decrypt but no key was supplied
1296          *      -EKEYREJECTED → Passed key was not correct
1297          */
1298
1299         if (root_hash && root_hash_size < sizeof(sd_id128_t))
1300                 return -EINVAL;
1301
1302         if (!m->encrypted && !m->verity) {
1303                 *ret = NULL;
1304                 return 0;
1305         }
1306
1307 #if HAVE_LIBCRYPTSETUP
1308         d = new0(DecryptedImage, 1);
1309         if (!d)
1310                 return -ENOMEM;
1311
1312         for (i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
1313                 DissectedPartition *p = m->partitions + i;
1314                 int k;
1315
1316                 if (!p->found)
1317                         continue;
1318
1319                 r = decrypt_partition(p, passphrase, flags, d);
1320                 if (r < 0)
1321                         return r;
1322
1323                 k = PARTITION_VERITY_OF(i);
1324                 if (k >= 0) {
1325                         r = verity_partition(p, m->partitions + k, root_hash, root_hash_size, flags, d);
1326                         if (r < 0)
1327                                 return r;
1328                 }
1329
1330                 if (!p->decrypted_fstype && p->decrypted_node) {
1331                         r = probe_filesystem(p->decrypted_node, &p->decrypted_fstype);
1332                         if (r < 0 && r != -EUCLEAN)
1333                                 return r;
1334                 }
1335         }
1336
1337         *ret = TAKE_PTR(d);
1338
1339         return 1;
1340 #else
1341         return -EOPNOTSUPP;
1342 #endif
1343 }
1344
1345 int dissected_image_decrypt_interactively(
1346                 DissectedImage *m,
1347                 const char *passphrase,
1348                 const void *root_hash,
1349                 size_t root_hash_size,
1350                 DissectImageFlags flags,
1351                 DecryptedImage **ret) {
1352
1353         _cleanup_strv_free_erase_ char **z = NULL;
1354         int n = 3, r;
1355
1356         if (passphrase)
1357                 n--;
1358
1359         for (;;) {
1360                 r = dissected_image_decrypt(m, passphrase, root_hash, root_hash_size, flags, ret);
1361                 if (r >= 0)
1362                         return r;
1363                 if (r == -EKEYREJECTED)
1364                         log_error_errno(r, "Incorrect passphrase, try again!");
1365                 else if (r != -ENOKEY)
1366                         return log_error_errno(r, "Failed to decrypt image: %m");
1367
1368                 if (--n < 0)
1369                         return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED),
1370                                                "Too many retries.");
1371
1372                 z = strv_free(z);
1373
1374                 r = ask_password_auto("Please enter image passphrase:", NULL, "dissect", "dissect", USEC_INFINITY, 0, &z);
1375                 if (r < 0)
1376                         return log_error_errno(r, "Failed to query for passphrase: %m");
1377
1378                 passphrase = z[0];
1379         }
1380 }
1381
1382 int decrypted_image_relinquish(DecryptedImage *d) {
1383
1384 #if HAVE_LIBCRYPTSETUP
1385         size_t i;
1386         int r;
1387 #endif
1388
1389         assert(d);
1390
1391         /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a boolean so
1392          * that we don't clean it up ourselves either anymore */
1393
1394 #if HAVE_LIBCRYPTSETUP
1395         for (i = 0; i < d->n_decrypted; i++) {
1396                 DecryptedPartition *p = d->decrypted + i;
1397
1398                 if (p->relinquished)
1399                         continue;
1400
1401                 r = dm_deferred_remove(p->name);
1402                 if (r < 0)
1403                         return log_debug_errno(r, "Failed to mark %s for auto-removal: %m", p->name);
1404
1405                 p->relinquished = true;
1406         }
1407 #endif
1408
1409         return 0;
1410 }
1411
1412 int root_hash_load(const char *image, void **ret, size_t *ret_size) {
1413         _cleanup_free_ char *text = NULL;
1414         _cleanup_free_ void *k = NULL;
1415         size_t l;
1416         int r;
1417
1418         assert(image);
1419         assert(ret);
1420         assert(ret_size);
1421
1422         if (is_device_path(image)) {
1423                 /* If we are asked to load the root hash for a device node, exit early */
1424                 *ret = NULL;
1425                 *ret_size = 0;
1426                 return 0;
1427         }
1428
1429         r = getxattr_malloc(image, "user.verity.roothash", &text, true);
1430         if (r < 0) {
1431                 char *fn, *e, *n;
1432
1433                 if (!IN_SET(r, -ENODATA, -EOPNOTSUPP, -ENOENT))
1434                         return r;
1435
1436                 fn = newa(char, strlen(image) + STRLEN(".roothash") + 1);
1437                 n = stpcpy(fn, image);
1438                 e = endswith(fn, ".raw");
1439                 if (e)
1440                         n = e;
1441
1442                 strcpy(n, ".roothash");
1443
1444                 r = read_one_line_file(fn, &text);
1445                 if (r == -ENOENT) {
1446                         *ret = NULL;
1447                         *ret_size = 0;
1448                         return 0;
1449                 }
1450                 if (r < 0)
1451                         return r;
1452         }
1453
1454         r = unhexmem(text, strlen(text), &k, &l);
1455         if (r < 0)
1456                 return r;
1457         if (l < sizeof(sd_id128_t))
1458                 return -EINVAL;
1459
1460         *ret = TAKE_PTR(k);
1461         *ret_size = l;
1462
1463         return 1;
1464 }
1465
1466 int dissected_image_acquire_metadata(DissectedImage *m) {
1467
1468         enum {
1469                 META_HOSTNAME,
1470                 META_MACHINE_ID,
1471                 META_MACHINE_INFO,
1472                 META_OS_RELEASE,
1473                 _META_MAX,
1474         };
1475
1476         static const char *const paths[_META_MAX] = {
1477                 [META_HOSTNAME]     = "/etc/hostname\0",
1478                 [META_MACHINE_ID]   = "/etc/machine-id\0",
1479                 [META_MACHINE_INFO] = "/etc/machine-info\0",
1480                 [META_OS_RELEASE]   = "/etc/os-release\0"
1481                                       "/usr/lib/os-release\0",
1482         };
1483
1484         _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL;
1485         _cleanup_(rmdir_and_freep) char *t = NULL;
1486         _cleanup_(sigkill_waitp) pid_t child = 0;
1487         sd_id128_t machine_id = SD_ID128_NULL;
1488         _cleanup_free_ char *hostname = NULL;
1489         unsigned n_meta_initialized = 0, k;
1490         int fds[2 * _META_MAX], r;
1491
1492         BLOCK_SIGNALS(SIGCHLD);
1493
1494         assert(m);
1495
1496         for (; n_meta_initialized < _META_MAX; n_meta_initialized ++)
1497                 if (pipe2(fds + 2*n_meta_initialized, O_CLOEXEC) < 0) {
1498                         r = -errno;
1499                         goto finish;
1500                 }
1501
1502         r = mkdtemp_malloc("/tmp/dissect-XXXXXX", &t);
1503         if (r < 0)
1504                 goto finish;
1505
1506         r = safe_fork("(sd-dissect)", FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE, &child);
1507         if (r < 0)
1508                 goto finish;
1509         if (r == 0) {
1510                 r = dissected_image_mount(m, t, UID_INVALID, DISSECT_IMAGE_READ_ONLY|DISSECT_IMAGE_MOUNT_ROOT_ONLY|DISSECT_IMAGE_VALIDATE_OS);
1511                 if (r < 0) {
1512                         log_debug_errno(r, "Failed to mount dissected image: %m");
1513                         _exit(EXIT_FAILURE);
1514                 }
1515
1516                 for (k = 0; k < _META_MAX; k++) {
1517                         _cleanup_close_ int fd = -1;
1518                         const char *p;
1519
1520                         fds[2*k] = safe_close(fds[2*k]);
1521
1522                         NULSTR_FOREACH(p, paths[k]) {
1523                                 fd = chase_symlinks_and_open(p, t, CHASE_PREFIX_ROOT, O_RDONLY|O_CLOEXEC|O_NOCTTY, NULL);
1524                                 if (fd >= 0)
1525                                         break;
1526                         }
1527                         if (fd < 0) {
1528                                 log_debug_errno(fd, "Failed to read %s file of image, ignoring: %m", paths[k]);
1529                                 continue;
1530                         }
1531
1532                         r = copy_bytes(fd, fds[2*k+1], (uint64_t) -1, 0);
1533                         if (r < 0)
1534                                 _exit(EXIT_FAILURE);
1535
1536                         fds[2*k+1] = safe_close(fds[2*k+1]);
1537                 }
1538
1539                 _exit(EXIT_SUCCESS);
1540         }
1541
1542         for (k = 0; k < _META_MAX; k++) {
1543                 _cleanup_fclose_ FILE *f = NULL;
1544
1545                 fds[2*k+1] = safe_close(fds[2*k+1]);
1546
1547                 f = take_fdopen(&fds[2*k], "r");
1548                 if (!f) {
1549                         r = -errno;
1550                         goto finish;
1551                 }
1552
1553                 switch (k) {
1554
1555                 case META_HOSTNAME:
1556                         r = read_etc_hostname_stream(f, &hostname);
1557                         if (r < 0)
1558                                 log_debug_errno(r, "Failed to read /etc/hostname: %m");
1559
1560                         break;
1561
1562                 case META_MACHINE_ID: {
1563                         _cleanup_free_ char *line = NULL;
1564
1565                         r = read_line(f, LONG_LINE_MAX, &line);
1566                         if (r < 0)
1567                                 log_debug_errno(r, "Failed to read /etc/machine-id: %m");
1568                         else if (r == 33) {
1569                                 r = sd_id128_from_string(line, &machine_id);
1570                                 if (r < 0)
1571                                         log_debug_errno(r, "Image contains invalid /etc/machine-id: %s", line);
1572                         } else if (r == 0)
1573                                 log_debug("/etc/machine-id file is empty.");
1574                         else
1575                                 log_debug("/etc/machine-id has unexpected length %i.", r);
1576
1577                         break;
1578                 }
1579
1580                 case META_MACHINE_INFO:
1581                         r = load_env_file_pairs(f, "machine-info", &machine_info);
1582                         if (r < 0)
1583                                 log_debug_errno(r, "Failed to read /etc/machine-info: %m");
1584
1585                         break;
1586
1587                 case META_OS_RELEASE:
1588                         r = load_env_file_pairs(f, "os-release", &os_release);
1589                         if (r < 0)
1590                                 log_debug_errno(r, "Failed to read OS release file: %m");
1591
1592                         break;
1593                 }
1594         }
1595
1596         r = wait_for_terminate_and_check("(sd-dissect)", child, 0);
1597         child = 0;
1598         if (r < 0)
1599                 goto finish;
1600         if (r != EXIT_SUCCESS)
1601                 return -EPROTO;
1602
1603         free_and_replace(m->hostname, hostname);
1604         m->machine_id = machine_id;
1605         strv_free_and_replace(m->machine_info, machine_info);
1606         strv_free_and_replace(m->os_release, os_release);
1607
1608 finish:
1609         for (k = 0; k < n_meta_initialized; k++)
1610                 safe_close_pair(fds + 2*k);
1611
1612         return r;
1613 }
1614
1615 int dissect_image_and_warn(
1616                 int fd,
1617                 const char *name,
1618                 const void *root_hash,
1619                 size_t root_hash_size,
1620                 DissectImageFlags flags,
1621                 DissectedImage **ret) {
1622
1623         _cleanup_free_ char *buffer = NULL;
1624         int r;
1625
1626         if (!name) {
1627                 r = fd_get_path(fd, &buffer);
1628                 if (r < 0)
1629                         return r;
1630
1631                 name = buffer;
1632         }
1633
1634         r = dissect_image(fd, root_hash, root_hash_size, flags, ret);
1635
1636         switch (r) {
1637
1638         case -EOPNOTSUPP:
1639                 return log_error_errno(r, "Dissecting images is not supported, compiled without blkid support.");
1640
1641         case -ENOPKG:
1642                 return log_error_errno(r, "Couldn't identify a suitable partition table or file system in '%s'.", name);
1643
1644         case -EADDRNOTAVAIL:
1645                 return log_error_errno(r, "No root partition for specified root hash found in '%s'.", name);
1646
1647         case -ENOTUNIQ:
1648                 return log_error_errno(r, "Multiple suitable root partitions found in image '%s'.", name);
1649
1650         case -ENXIO:
1651                 return log_error_errno(r, "No suitable root partition found in image '%s'.", name);
1652
1653         case -EPROTONOSUPPORT:
1654                 return log_error_errno(r, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name);
1655
1656         default:
1657                 if (r < 0)
1658                         return log_error_errno(r, "Failed to dissect image '%s': %m", name);
1659
1660                 return r;
1661         }
1662 }
1663
1664 static const char *const partition_designator_table[] = {
1665         [PARTITION_ROOT] = "root",
1666         [PARTITION_ROOT_SECONDARY] = "root-secondary",
1667         [PARTITION_HOME] = "home",
1668         [PARTITION_SRV] = "srv",
1669         [PARTITION_ESP] = "esp",
1670         [PARTITION_XBOOTLDR] = "xbootldr",
1671         [PARTITION_SWAP] = "swap",
1672         [PARTITION_ROOT_VERITY] = "root-verity",
1673         [PARTITION_ROOT_SECONDARY_VERITY] = "root-secondary-verity",
1674         [PARTITION_TMP] = "tmp",
1675         [PARTITION_VAR] = "var",
1676 };
1677
1678 DEFINE_STRING_TABLE_LOOKUP(partition_designator, int);