]> git.ipfire.org Git - thirdparty/systemd.git/blame - src/shared/dissect-image.c
projects / thirdparty / systemd.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge pull request #25570 from yuwata/dissect-rootless-image
[thirdparty/systemd.git] / src / shared / dissect-image.c
CommitLineData
db9ecf05 1/* SPDX-License-Identifier: LGPL-2.1-or-later */
8c1be37e 2
10c1b188
LP		 3#if HAVE_VALGRIND_MEMCHECK_H
4#include <valgrind/memcheck.h>
5#endif
6
01234e1f
YW		 7#include <linux/dm-ioctl.h>
8#include <linux/loop.h>
19df770f 9#include <sys/file.h>
8c1be37e 10#include <sys/mount.h>
3b925504
LP		 11#include <sys/prctl.h>
12#include <sys/wait.h>
f5ea63a5 13#include <sysexits.h>
8c1be37e 14
c2fa92e7
LP		 15#if HAVE_OPENSSL
16#include <openssl/err.h>
17#include <openssl/pem.h>
18#include <openssl/x509.h>
19#endif
20
3c1f2cee 21#include "sd-device.h"
dccca82b
LP		 22#include "sd-id128.h"
23
8c1be37e 24#include "architecture.h"
18b5886e 25#include "ask-password-api.h"
8c1be37e 26#include "blkid-util.h"
18c528e9 27#include "blockdev-util.h"
f4351959 28#include "chase-symlinks.h"
c2fa92e7 29#include "conf-files.h"
28db6fbf 30#include "constants.h"
3b925504 31#include "copy.h"
1e2f3230 32#include "cryptsetup-util.h"
553e15f2 33#include "device-nodes.h"
8437c059 34#include "device-util.h"
f7725647 35#include "devnum-util.h"
7718ac97 36#include "discover-image.h"
8c1be37e 37#include "dissect-image.h"
a709a315 38#include "dm-util.h"
686d13b9 39#include "env-file.h"
88b3300f 40#include "env-util.h"
93f59701 41#include "extension-release.h"
18b5886e 42#include "fd-util.h"
78ebe980 43#include "fileio.h"
2eedfd2d 44#include "fs-util.h"
cf32c486 45#include "fsck-util.h"
8c1be37e 46#include "gpt.h"
78ebe980 47#include "hexdecoct.h"
e2054217 48#include "hostname-setup.h"
3b925504 49#include "id128-util.h"
593fe6c0 50#include "import-util.h"
a4e0d617 51#include "io-util.h"
35cd0ba5 52#include "mkdir-label.h"
8c1be37e 53#include "mount-util.h"
e4de7287 54#include "mountpoint-util.h"
6aa05ebd 55#include "namespace-util.h"
d8b4d14d 56#include "nulstr-util.h"
c2fa92e7 57#include "openssl-util.h"
d58ad743 58#include "os-util.h"
8c1be37e 59#include "path-util.h"
3b925504
LP		 60#include "process-util.h"
61#include "raw-clone.h"
81939d9d 62#include "resize-fs.h"
3b925504 63#include "signal-util.h"
8c1be37e 64#include "stat-util.h"
18b5886e 65#include "stdio-util.h"
8c1be37e
LP		 66#include "string-table.h"
67#include "string-util.h"
2eedfd2d 68#include "strv.h"
e4de7287 69#include "tmpfile-util.h"
a8040b6d 70#include "udev-util.h"
2d3a5a73 71#include "user-util.h"
41488e1f 72#include "xattr-util.h"
8c1be37e 73
28e2641a
FF		 74/* how many times to wait for the device nodes to appear */
75#define N_DEVICE_NODE_LIST_ATTEMPTS 10
76
c80c9079
LP		 77int probe_filesystem_full(
78 int fd,
79 const char *path,
80 uint64_t offset,
81 uint64_t size,
82 char **ret_fstype) {
83
7cc84b2c 84 /* Try to find device content type and return it in *ret_fstype. If nothing is found,
c80c9079 85 * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and a
7cc84b2c
ZJS		 86 * different error otherwise. */
87
349cc4a5 88#if HAVE_BLKID
8e766630 89 _cleanup_(blkid_free_probep) blkid_probe b = NULL;
d2c6e79d
YW		 90 _cleanup_free_ char *path_by_fd = NULL;
91 _cleanup_close_ int fd_close = -1;
18b5886e
LP		 92 const char *fstype;
93 int r;
94
d2c6e79d
YW		 95 assert(fd >= 0 || path);
96 assert(ret_fstype);
97
98 if (fd < 0) {
99 fd_close = open(path, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
100 if (fd_close < 0)
101 return -errno;
102
103 fd = fd_close;
104 }
105
106 if (!path) {
107 r = fd_get_path(fd, &path_by_fd);
108 if (r < 0)
109 return r;
110
111 path = path_by_fd;
112 }
113
c80c9079
LP		 114 if (size == 0) /* empty size? nothing found! */
115 goto not_found;
116
d2c6e79d 117 b = blkid_new_probe();
18b5886e 118 if (!b)
d2c6e79d
YW		 119 return -ENOMEM;
120
121 errno = 0;
c80c9079
LP		 122 r = blkid_probe_set_device(
123 b,
124 fd,
125 offset,
126 size == UINT64_MAX ? 0 : size); /* when blkid sees size=0 it understands "everything". We prefer using UINT64_MAX for that */
d2c6e79d 127 if (r != 0)
66855de7 128 return errno_or_else(ENOMEM);
18b5886e
LP		 129
130 blkid_probe_enable_superblocks(b, 1);
131 blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE);
132
133 errno = 0;
134 r = blkid_do_safeprobe(b);
2e3944b8 135 if (r == _BLKID_SAFEPROBE_NOT_FOUND)
18b5886e 136 goto not_found;
2e3944b8 137 if (r == _BLKID_SAFEPROBE_AMBIGUOUS)
58dfbfbd 138 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN),
d2c6e79d 139 "Results ambiguous for partition %s", path);
2e3944b8 140 if (r == _BLKID_SAFEPROBE_ERROR)
d2c6e79d 141 return log_debug_errno(errno_or_else(EIO), "Failed to probe partition %s: %m", path);
18b5886e 142
2e3944b8
LP		 143 assert(r == _BLKID_SAFEPROBE_FOUND);
144
18b5886e
LP		 145 (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
146
147 if (fstype) {
148 char *t;
149
d2c6e79d 150 log_debug("Probed fstype '%s' on partition %s.", fstype, path);
1cefb9a6 151
18b5886e
LP		 152 t = strdup(fstype);
153 if (!t)
154 return -ENOMEM;
155
156 *ret_fstype = t;
157 return 1;
158 }
159
160not_found:
d2c6e79d 161 log_debug("No type detected on partition %s", path);
18b5886e
LP		 162 *ret_fstype = NULL;
163 return 0;
d1c536f5
ZJS		 164#else
165 return -EOPNOTSUPP;
a75e27eb 166#endif
d1c536f5 167}
18b5886e 168
40c10d3f 169#if HAVE_BLKID
c80c9079 170static int dissected_image_probe_filesystems(DissectedImage *m, int fd) {
698bb074
YW		 171 int r;
172
173 assert(m);
174
175 /* Fill in file system types if we don't know them yet. */
176
177 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
178 DissectedPartition *p = m->partitions + i;
179
180 if (!p->found)
181 continue;
182
c80c9079
LP		 183 if (!p->fstype) {
184 /* If we have an fd referring to the partition block device, use that. Otherwise go
185 * via the whole block device or backing regular file, and read via offset. */
186 if (p->mount_node_fd >= 0)
187 r = probe_filesystem_full(p->mount_node_fd, p->node, 0, UINT64_MAX, &p->fstype);
188 else
189 r = probe_filesystem_full(fd, p->node, p->offset, p->size, &p->fstype);
190 if (r < 0)
698bb074
YW		 191 return r;
192 }
193
194 if (streq_ptr(p->fstype, "crypto_LUKS"))
195 m->encrypted = true;
196
197 if (p->fstype && fstype_is_ro(p->fstype))
198 p->rw = false;
199
200 if (!p->rw)
201 p->growfs = false;
202 }
203
204 return 0;
205}
206
0f7c9a3d
LP		 207static void check_partition_flags(
208 const char *node,
209 unsigned long long pflags,
210 unsigned long long supported) {
211
212 assert(node);
213
214 /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
92e72028
ZJS		 215 pflags &= ~(supported |
216 SD_GPT_FLAG_REQUIRED_PARTITION |
217 SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL |
218 SD_GPT_FLAG_LEGACY_BIOS_BOOTABLE);
0f7c9a3d
LP		 219
220 if (pflags == 0)
221 return;
222
223 /* If there are other bits set, then log about it, to make things discoverable */
224 for (unsigned i = 0; i < sizeof(pflags) * 8; i++) {
225 unsigned long long bit = 1ULL << i;
226 if (!FLAGS_SET(pflags, bit))
227 continue;
228
229 log_debug("Unexpected partition flag %llu set on %s!", bit, node);
230 }
231}
1b010ae7 232#endif
0f7c9a3d 233
00e29505
YW		 234#if HAVE_BLKID
235static int dissected_image_new(const char *path, DissectedImage **ret) {
236 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
237 _cleanup_free_ char *name = NULL;
238 int r;
239
240 assert(ret);
241
242 if (path) {
243 _cleanup_free_ char *filename = NULL;
244
245 r = path_extract_filename(path, &filename);
246 if (r < 0)
247 return r;
248
249 r = raw_strip_suffixes(filename, &name);
250 if (r < 0)
251 return r;
252
253 if (!image_name_is_valid(name)) {
254 log_debug("Image name %s is not valid, ignoring.", strna(name));
255 name = mfree(name);
256 }
257 }
258
259 m = new(DissectedImage, 1);
260 if (!m)
261 return -ENOMEM;
262
263 *m = (DissectedImage) {
264 .has_init_system = -1,
265 .image_name = TAKE_PTR(name),
266 };
267
babd5b08
YW		 268 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++)
269 m->partitions[i] = DISSECTED_PARTITION_NULL;
270
00e29505
YW		 271 *ret = TAKE_PTR(m);
272 return 0;
273}
274#endif
275
234c2e16 276static void dissected_partition_done(DissectedPartition *p) {
1b010ae7 277 assert(p);
786e3a52 278
1b010ae7
LP		 279 free(p->fstype);
280 free(p->node);
281 free(p->label);
282 free(p->decrypted_fstype);
283 free(p->decrypted_node);
284 free(p->mount_options);
f7725647 285 safe_close(p->mount_node_fd);
786e3a52 286
babd5b08 287 *p = DISSECTED_PARTITION_NULL;
1b010ae7 288}
786e3a52 289
1b010ae7 290#if HAVE_BLKID
1b010ae7
LP		 291static int make_partition_devname(
292 const char *whole_devname,
293 int nr,
294 char **ret) {
786e3a52 295
1b010ae7 296 bool need_p;
786e3a52 297
1b010ae7
LP		 298 assert(whole_devname);
299 assert(nr > 0);
aae22eb3 300
1b010ae7
LP		 301 /* Given a whole block device node name (e.g. /dev/sda or /dev/loop7) generate a partition device
302 * name (e.g. /dev/sda7 or /dev/loop7p5). The rule the kernel uses is simple: if whole block device
303 * node name ends in a digit, then suffix a 'p', followed by the partition number. Otherwise, just
304 * suffix the partition number without any 'p'. */
4ba86848 305
1b010ae7
LP		 306 if (isempty(whole_devname)) /* Make sure there *is* a last char */
307 return -EINVAL;
08fe0a53 308
ff25d338 309 need_p = ascii_isdigit(whole_devname[strlen(whole_devname)-1]); /* Last char a digit? */
08fe0a53 310
1b010ae7 311 return asprintf(ret, "%s%s%i", whole_devname, need_p ? "p" : "", nr);
08fe0a53
LP		 312}
313
f7725647
YW		 314static int open_partition(const char *node, bool is_partition, const LoopDevice *loop) {
315 _cleanup_(sd_device_unrefp) sd_device *dev = NULL;
316 _cleanup_close_ int fd = -1;
317 dev_t devnum;
318 int r;
319
320 assert(node);
321 assert(loop);
322
323 fd = open(node, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
324 if (fd < 0)
325 return -errno;
326
327 /* Check if the block device is a child of (or equivalent to) the originally provided one. */
328 r = block_device_new_from_fd(fd, is_partition ? BLOCK_DEVICE_LOOKUP_WHOLE_DISK : 0, &dev);
329 if (r < 0)
330 return r;
331
332 r = sd_device_get_devnum(dev, &devnum);
333 if (r < 0)
334 return r;
335
336 if (loop->devno != devnum)
337 return -ENXIO;
338
339 /* Also check diskseq. */
340 if (loop->diskseq > 0) {
341 uint64_t diskseq;
342
343 r = fd_get_diskseq(fd, &diskseq);
344 if (r < 0)
345 return r;
346
347 if (loop->diskseq != diskseq)
348 return -ENXIO;
349 }
350
351 log_debug("Opened %s (fd=%i, whole_block_devnum=" DEVNUM_FORMAT_STR ", diskseq=%" PRIu64 ").",
352 node, fd, DEVNUM_FORMAT_VAL(loop->devno), loop->diskseq);
353 return TAKE_FD(fd);
354}
355
22e932f4
DDM		 356static int compare_arch(Architecture a, Architecture b) {
357 if (a == b)
358 return 0;
359
360 if (a == native_architecture())
361 return 1;
362
363 if (b == native_architecture())
364 return -1;
365
366#ifdef ARCHITECTURE_SECONDARY
367 if (a == ARCHITECTURE_SECONDARY)
368 return 1;
369
370 if (b == ARCHITECTURE_SECONDARY)
371 return -1;
372#endif
373
374 return 0;
375}
376
08f14be4
YW		 377static int dissect_image(
378 DissectedImage *m,
4526113f 379 int fd,
0b214aa0 380 const char *devname,
89e62e0b 381 const VeritySettings *verity,
18d73705 382 const MountOptions *mount_options,
08f14be4 383 DissectImageFlags flags) {
8c1be37e 384
62ea0ed0 385 sd_id128_t root_uuid = SD_ID128_NULL, root_verity_uuid = SD_ID128_NULL;
62ea0ed0 386 sd_id128_t usr_uuid = SD_ID128_NULL, usr_verity_uuid = SD_ID128_NULL;
1f8fb21c 387 bool is_gpt, is_mbr, multiple_generic = false,
de98f631
LP		 388 generic_rw = false, /* initialize to appease gcc */
389 generic_growfs = false;
8e766630 390 _cleanup_(blkid_free_probep) blkid_probe b = NULL;
8c1be37e 391 _cleanup_free_ char *generic_node = NULL;
be30ad41 392 sd_id128_t generic_uuid = SD_ID128_NULL;
b387778c 393 const char *pttype = NULL, *sptuuid = NULL;
8c1be37e 394 blkid_partlist pl;
1f8fb21c 395 int r, generic_nr = -1, n_partitions;
8c1be37e 396
08f14be4 397 assert(m);
8c1be37e 398 assert(fd >= 0);
0b214aa0 399 assert(devname);
a0bff7ea 400 assert(!verity || verity->designator < 0 || IN_SET(verity->designator, PARTITION_ROOT, PARTITION_USR));
89e62e0b 401 assert(!verity || verity->root_hash || verity->root_hash_size == 0);
a0bff7ea
LP		 402 assert(!verity || verity->root_hash_sig || verity->root_hash_sig_size == 0);
403 assert(!verity || (verity->root_hash || !verity->root_hash_sig));
e7cbe5cb 404 assert(!((flags & DISSECT_IMAGE_GPT_ONLY) && (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)));
8c1be37e
LP		 405
406 /* Probes a disk image, and returns information about what it found in *ret.
407 *
4623e8e6 408 * Returns -ENOPKG if no suitable partition table or file system could be found.
2679f407
LP		 409 * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found.
410 * Returns -ENXIO if we couldn't find any partition suitable as root or /usr partition
411 * Returns -ENOTUNIQ if we only found multiple generic partitions and thus don't know what to do with that */
4623e8e6 412
89e62e0b 413 if (verity && verity->root_hash) {
aee36b4e
LP		 414 sd_id128_t fsuuid, vuuid;
415
416 /* If a root hash is supplied, then we use the root partition that has a UUID that match the
417 * first 128bit of the root hash. And we use the verity partition that has a UUID that match
418 * the final 128bit. */
4623e8e6 419
89e62e0b 420 if (verity->root_hash_size < sizeof(sd_id128_t))
4623e8e6
LP		 421 return -EINVAL;
422
aee36b4e
LP		 423 memcpy(&fsuuid, verity->root_hash, sizeof(sd_id128_t));
424 memcpy(&vuuid, (const uint8_t*) verity->root_hash + verity->root_hash_size - sizeof(sd_id128_t), sizeof(sd_id128_t));
4623e8e6 425
aee36b4e 426 if (sd_id128_is_null(fsuuid))
4623e8e6 427 return -EINVAL;
aee36b4e 428 if (sd_id128_is_null(vuuid))
4623e8e6 429 return -EINVAL;
aee36b4e
LP		 430
431 /* If the verity data declares it's for the /usr partition, then search for that, in all
432 * other cases assume it's for the root partition. */
433 if (verity->designator == PARTITION_USR) {
434 usr_uuid = fsuuid;
435 usr_verity_uuid = vuuid;
436 } else {
437 root_uuid = fsuuid;
438 root_verity_uuid = vuuid;
439 }
4623e8e6 440 }
8c1be37e 441
8c1be37e
LP		 442 b = blkid_new_probe();
443 if (!b)
444 return -ENOMEM;
445
446 errno = 0;
447 r = blkid_probe_set_device(b, fd, 0, 0);
b382db9f 448 if (r != 0)
66855de7 449 return errno_or_else(ENOMEM);
8c1be37e 450
9b6deb03
LP		 451 if ((flags & DISSECT_IMAGE_GPT_ONLY) == 0) {
452 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
453 blkid_probe_enable_superblocks(b, 1);
b387778c 454 blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE|BLKID_SUBLKS_USAGE|BLKID_SUBLKS_UUID);
9b6deb03
LP		 455 }
456
8c1be37e
LP		 457 blkid_probe_enable_partitions(b, 1);
458 blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS);
459
460 errno = 0;
461 r = blkid_do_safeprobe(b);
2e3944b8 462 if (r == _BLKID_SAFEPROBE_ERROR)
66855de7 463 return errno_or_else(EIO);
2e3944b8
LP		 464 if (IN_SET(r, _BLKID_SAFEPROBE_AMBIGUOUS, _BLKID_SAFEPROBE_NOT_FOUND))
465 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG), "Failed to identify any partition table.");
466
467 assert(r == _BLKID_SAFEPROBE_FOUND);
8c1be37e 468
e7cbe5cb 469 if ((!(flags & DISSECT_IMAGE_GPT_ONLY) &&
4b5de5dd 470 (flags & DISSECT_IMAGE_GENERIC_ROOT)) ||
e7cbe5cb 471 (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)) {
9b6deb03 472 const char *usage = NULL;
8c1be37e 473
aee36b4e
LP		 474 /* If flags permit this, also allow using non-partitioned single-filesystem images */
475
9b6deb03
LP		 476 (void) blkid_probe_lookup_value(b, "USAGE", &usage, NULL);
477 if (STRPTR_IN_SET(usage, "filesystem", "crypto")) {
18d73705 478 _cleanup_free_ char *t = NULL, *n = NULL, *o = NULL;
b387778c 479 const char *fstype = NULL, *options = NULL, *suuid = NULL;
f7725647 480 _cleanup_close_ int mount_node_fd = -1;
b387778c 481 sd_id128_t uuid = SD_ID128_NULL;
f7725647 482
73d88b80 483 if (FLAGS_SET(flags, DISSECT_IMAGE_PIN_PARTITION_DEVICES)) {
f7725647
YW		 484 mount_node_fd = open_partition(devname, /* is_partition = */ false, m->loop);
485 if (mount_node_fd < 0)
486 return mount_node_fd;
487 }
8c1be37e 488
9b6deb03
LP		 489 /* OK, we have found a file system, that's our root partition then. */
490 (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
b387778c 491 (void) blkid_probe_lookup_value(b, "UUID", &suuid, NULL);
8c1be37e 492
9b6deb03
LP		 493 if (fstype) {
494 t = strdup(fstype);
495 if (!t)
496 return -ENOMEM;
497 }
498
b387778c
LP		 499 if (suuid) {
500 /* blkid will return FAT's serial number as UUID, hence it is quite possible
501 * that parsing this will fail. We'll ignore the ID, since it's just too
502 * short to be useful as tru identifier. */
503 r = sd_id128_from_string(suuid, &uuid);
504 if (r < 0)
505 log_debug_errno(r, "Failed to parse file system UUID '%s', ignoring: %m", suuid);
506 }
507
6c544d14
LP		 508 n = strdup(devname);
509 if (!n)
510 return -ENOMEM;
511
e7cbe5cb 512 m->single_file_system = true;
c3c88d67
LP		 513 m->encrypted = streq_ptr(fstype, "crypto_LUKS");
514
515 m->has_verity = verity && verity->data_path;
516 m->verity_ready = m->has_verity &&
517 verity->root_hash &&
518 (verity->designator < 0 || verity->designator == PARTITION_ROOT);
e7cbe5cb 519
8ee9615e
LP		 520 m->has_verity_sig = false; /* signature not embedded, must be specified */
521 m->verity_sig_ready = m->verity_ready &&
522 verity->root_hash_sig;
523
b387778c
LP		 524 m->image_uuid = uuid;
525
f5215bc8 526 options = mount_options_from_designator(mount_options, PARTITION_ROOT);
18d73705
LB		 527 if (options) {
528 o = strdup(options);
529 if (!o)
530 return -ENOMEM;
531 }
532
9b6deb03
LP		 533 m->partitions[PARTITION_ROOT] = (DissectedPartition) {
534 .found = true,
e0d53d52 535 .rw = !m->verity_ready && !fstype_is_ro(fstype),
9b6deb03
LP		 536 .partno = -1,
537 .architecture = _ARCHITECTURE_INVALID,
1cc6c93a
YW		 538 .fstype = TAKE_PTR(t),
539 .node = TAKE_PTR(n),
18d73705 540 .mount_options = TAKE_PTR(o),
f7725647 541 .mount_node_fd = TAKE_FD(mount_node_fd),
88b3300f
LP		 542 .offset = 0,
543 .size = UINT64_MAX,
9b6deb03 544 };
8c1be37e 545
9b6deb03
LP		 546 return 0;
547 }
8c1be37e
LP		 548 }
549
550 (void) blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL);
551 if (!pttype)
552 return -ENOPKG;
553
554 is_gpt = streq_ptr(pttype, "gpt");
555 is_mbr = streq_ptr(pttype, "dos");
556
9b6deb03 557 if (!is_gpt && ((flags & DISSECT_IMAGE_GPT_ONLY) || !is_mbr))
8c1be37e
LP		 558 return -ENOPKG;
559
0903fd26
LP		 560 /* We support external verity data partitions only if the image has no partition table */
561 if (verity && verity->data_path)
562 return -EBADR;
563
73d88b80 564 if (FLAGS_SET(flags, DISSECT_IMAGE_ADD_PARTITION_DEVICES)) {
08f14be4
YW		 565 /* Safety check: refuse block devices that carry a partition table but for which the kernel doesn't
566 * do partition scanning. */
567 r = blockdev_partscan_enabled(fd);
568 if (r < 0)
569 return r;
570 if (r == 0)
571 return -EPROTONOSUPPORT;
572 }
4ba86848 573
b387778c
LP		 574 (void) blkid_probe_lookup_value(b, "PTUUID", &sptuuid, NULL);
575 if (sptuuid) {
576 r = sd_id128_from_string(sptuuid, &m->image_uuid);
577 if (r < 0)
578 log_debug_errno(r, "Failed to parse partition table UUID '%s', ignoring: %m", sptuuid);
579 }
580
8c1be37e
LP		 581 errno = 0;
582 pl = blkid_probe_get_partitions(b);
b382db9f 583 if (!pl)
66855de7 584 return errno_or_else(ENOMEM);
8c1be37e 585
4ba86848
LP		 586 errno = 0;
587 n_partitions = blkid_partlist_numof_partitions(pl);
588 if (n_partitions < 0)
589 return errno_or_else(EIO);
8c1be37e 590
4ba86848 591 for (int i = 0; i < n_partitions; i++) {
1b010ae7 592 _cleanup_free_ char *node = NULL;
9b6deb03 593 unsigned long long pflags;
88b3300f 594 blkid_loff_t start, size;
8c1be37e 595 blkid_partition pp;
8c1be37e
LP		 596 int nr;
597
4ba86848
LP		 598 errno = 0;
599 pp = blkid_partlist_get_partition(pl, i);
600 if (!pp)
601 return errno_or_else(EIO);
aae22eb3 602
9b6deb03 603 pflags = blkid_partition_get_flags(pp);
8c1be37e 604
4ba86848 605 errno = 0;
8c1be37e
LP		 606 nr = blkid_partition_get_partno(pp);
607 if (nr < 0)
4ba86848 608 return errno_or_else(EIO);
8c1be37e 609
88b3300f
LP		 610 errno = 0;
611 start = blkid_partition_get_start(pp);
612 if (start < 0)
613 return errno_or_else(EIO);
614
615 assert((uint64_t) start < UINT64_MAX/512);
616
617 errno = 0;
618 size = blkid_partition_get_size(pp);
619 if (size < 0)
620 return errno_or_else(EIO);
621
622 assert((uint64_t) size < UINT64_MAX/512);
623
1b010ae7
LP		 624 r = make_partition_devname(devname, nr, &node);
625 if (r < 0)
626 return r;
627
628 /* So here's the thing: after the main ("whole") block device popped up it might take a while
629 * before the kernel fully probed the partition table. Waiting for that to finish is icky in
630 * userspace. So here's what we do instead. We issue the BLKPG_ADD_PARTITION ioctl to add the
631 * partition ourselves, racing against the kernel. Good thing is: if this call fails with
632 * EBUSY then the kernel was quicker than us, and that's totally OK, the outcome is good for
633 * us: the device node will exist. If OTOH our call was successful we won the race. Which is
634 * also good as the outcome is the same: the partition block device exists, and we can use
635 * it.
636 *
637 * Kernel returns EBUSY if there's already a partition by that number or an overlapping
638 * partition already existent. */
639
73d88b80 640 if (FLAGS_SET(flags, DISSECT_IMAGE_ADD_PARTITION_DEVICES)) {
08f14be4
YW		 641 r = block_device_add_partition(fd, node, nr, (uint64_t) start * 512, (uint64_t) size * 512);
642 if (r < 0) {
643 if (r != -EBUSY)
644 return log_debug_errno(r, "BLKPG_ADD_PARTITION failed: %m");
1b010ae7 645
08f14be4
YW		 646 log_debug_errno(r, "Kernel was quicker than us in adding partition %i.", nr);
647 } else
648 log_debug("We were quicker than kernel in adding partition %i.", nr);
649 }
1b010ae7 650
8c1be37e 651 if (is_gpt) {
e3b9a5ff 652 const char *fstype = NULL, *label;
4623e8e6 653 sd_id128_t type_id, id;
22e932f4 654 GptPartitionType type;
de98f631 655 bool rw = true, growfs = false;
8c1be37e 656
e3b9a5ff
LP		 657 r = blkid_partition_get_uuid_id128(pp, &id);
658 if (r < 0) {
659 log_debug_errno(r, "Failed to read partition UUID, ignoring: %m");
4623e8e6 660 continue;
e3b9a5ff 661 }
4623e8e6 662
e3b9a5ff
LP		 663 r = blkid_partition_get_type_id128(pp, &type_id);
664 if (r < 0) {
665 log_debug_errno(r, "Failed to read partition type UUID, ignoring: %m");
8c1be37e 666 continue;
e3b9a5ff 667 }
8c1be37e 668
22e932f4
DDM		 669 type = gpt_partition_type_from_uuid(type_id);
670
08fe0a53
LP		 671 label = blkid_partition_get_name(pp); /* libblkid returns NULL here if empty */
672
fb3921b8
LP		 673 if (IN_SET(type.designator,
674 PARTITION_HOME,
675 PARTITION_SRV,
676 PARTITION_XBOOTLDR,
677 PARTITION_TMP)) {
a48dd347 678
92e72028
ZJS		 679 check_partition_flags(node, pflags,
680 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS);
0f7c9a3d 681
92e72028 682 if (pflags & SD_GPT_FLAG_NO_AUTO)
a48dd347
LP		 683 continue;
684
92e72028
ZJS		 685 rw = !(pflags & SD_GPT_FLAG_READ_ONLY);
686 growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS);
aee36b4e 687
22e932f4 688 } else if (type.designator == PARTITION_ESP) {
a48dd347 689
92e72028
ZJS		 690 /* Note that we don't check the SD_GPT_FLAG_NO_AUTO flag for the ESP, as it is
691 * not defined there. We instead check the SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as
aee36b4e
LP		 692 * recommended by the UEFI spec (See "12.3.3 Number and Location of System
693 * Partitions"). */
a48dd347 694
92e72028 695 if (pflags & SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL)
a48dd347
LP		 696 continue;
697
8c1be37e 698 fstype = "vfat";
a8c47660 699
22e932f4 700 } else if (type.designator == PARTITION_ROOT) {
4623e8e6 701
92e72028
ZJS		 702 check_partition_flags(node, pflags,
703 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS);
0f7c9a3d 704
92e72028 705 if (pflags & SD_GPT_FLAG_NO_AUTO)
a48dd347
LP		 706 continue;
707
4623e8e6
LP		 708 /* If a root ID is specified, ignore everything but the root id */
709 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
710 continue;
711
92e72028
ZJS		 712 rw = !(pflags & SD_GPT_FLAG_READ_ONLY);
713 growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS);
aee36b4e 714
22e932f4 715 } else if (type.designator == PARTITION_ROOT_VERITY) {
4623e8e6 716
92e72028
ZJS		 717 check_partition_flags(node, pflags,
718 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY);
0f7c9a3d 719
92e72028 720 if (pflags & SD_GPT_FLAG_NO_AUTO)
a48dd347
LP		 721 continue;
722
c3c88d67 723 m->has_verity = true;
4623e8e6 724
8ee9615e
LP		 725 /* If no verity configuration is specified, then don't do verity */
726 if (!verity)
727 continue;
728 if (verity->designator >= 0 && verity->designator != PARTITION_ROOT)
729 continue;
730
731 /* If root hash is specified, then ignore everything but the root id */
732 if (!sd_id128_is_null(root_verity_uuid) && !sd_id128_equal(root_verity_uuid, id))
4623e8e6
LP		 733 continue;
734
4623e8e6 735 fstype = "DM_verity_hash";
4623e8e6 736 rw = false;
8ee9615e 737
22e932f4 738 } else if (type.designator == PARTITION_ROOT_VERITY_SIG) {
8ee9615e 739
92e72028
ZJS		 740 check_partition_flags(node, pflags,
741 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY);
8ee9615e 742
92e72028 743 if (pflags & SD_GPT_FLAG_NO_AUTO)
8ee9615e
LP		 744 continue;
745
746 m->has_verity_sig = true;
747
8ee9615e
LP		 748 if (!verity)
749 continue;
750 if (verity->designator >= 0 && verity->designator != PARTITION_ROOT)
751 continue;
8ee9615e 752
8ee9615e 753 fstype = "verity_hash_signature";
4623e8e6 754 rw = false;
8ee9615e 755
22e932f4 756 } else if (type.designator == PARTITION_USR) {
aee36b4e 757
92e72028
ZJS		 758 check_partition_flags(node, pflags,
759 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS);
aee36b4e 760
92e72028 761 if (pflags & SD_GPT_FLAG_NO_AUTO)
aee36b4e
LP		 762 continue;
763
764 /* If a usr ID is specified, ignore everything but the usr id */
765 if (!sd_id128_is_null(usr_uuid) && !sd_id128_equal(usr_uuid, id))
766 continue;
767
92e72028
ZJS		 768 rw = !(pflags & SD_GPT_FLAG_READ_ONLY);
769 growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS);
aee36b4e 770
22e932f4 771 } else if (type.designator == PARTITION_USR_VERITY) {
aee36b4e 772
92e72028
ZJS		 773 check_partition_flags(node, pflags,
774 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY);
aee36b4e 775
92e72028 776 if (pflags & SD_GPT_FLAG_NO_AUTO)
aee36b4e
LP		 777 continue;
778
c3c88d67 779 m->has_verity = true;
aee36b4e 780
8ee9615e
LP		 781 if (!verity)
782 continue;
783 if (verity->designator >= 0 && verity->designator != PARTITION_USR)
784 continue;
785
786 /* If usr hash is specified, then ignore everything but the usr id */
787 if (!sd_id128_is_null(usr_verity_uuid) && !sd_id128_equal(usr_verity_uuid, id))
aee36b4e
LP		 788 continue;
789
aee36b4e 790 fstype = "DM_verity_hash";
aee36b4e 791 rw = false;
8ee9615e 792
22e932f4 793 } else if (type.designator == PARTITION_USR_VERITY_SIG) {
8ee9615e 794
92e72028
ZJS		 795 check_partition_flags(node, pflags,
796 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY);
8ee9615e 797
92e72028 798 if (pflags & SD_GPT_FLAG_NO_AUTO)
8ee9615e
LP		 799 continue;
800
801 m->has_verity_sig = true;
802
8ee9615e
LP		 803 if (!verity)
804 continue;
805 if (verity->designator >= 0 && verity->designator != PARTITION_USR)
806 continue;
8ee9615e 807
8ee9615e 808 fstype = "verity_hash_signature";
aee36b4e 809 rw = false;
8ee9615e 810
22e932f4 811 } else if (type.designator == PARTITION_SWAP) {
a48dd347 812
92e72028 813 check_partition_flags(node, pflags, SD_GPT_FLAG_NO_AUTO);
0f7c9a3d 814
92e72028 815 if (pflags & SD_GPT_FLAG_NO_AUTO)
a48dd347
LP		 816 continue;
817
df4524cb
LP		 818 fstype = "swap";
819
df655bf3
DDM		 820 /* We don't have a designator for SD_GPT_LINUX_GENERIC so check the UUID instead. */
821 } else if (sd_id128_equal(type.uuid, SD_GPT_LINUX_GENERIC)) {
8c1be37e 822
92e72028
ZJS		 823 check_partition_flags(node, pflags,
824 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS);
0f7c9a3d 825
92e72028 826 if (pflags & SD_GPT_FLAG_NO_AUTO)
a48dd347
LP		 827 continue;
828
8c1be37e
LP		 829 if (generic_node)
830 multiple_generic = true;
831 else {
832 generic_nr = nr;
92e72028
ZJS		 833 generic_rw = !(pflags & SD_GPT_FLAG_READ_ONLY);
834 generic_growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS);
be30ad41 835 generic_uuid = id;
8c1be37e
LP		 836 generic_node = strdup(node);
837 if (!generic_node)
838 return -ENOMEM;
839 }
d4dffb85 840
22e932f4 841 } else if (type.designator == PARTITION_VAR) {
d4dffb85 842
92e72028
ZJS		 843 check_partition_flags(node, pflags,
844 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS);
0f7c9a3d 845
92e72028 846 if (pflags & SD_GPT_FLAG_NO_AUTO)
d4dffb85
LP		 847 continue;
848
849 if (!FLAGS_SET(flags, DISSECT_IMAGE_RELAX_VAR_CHECK)) {
850 sd_id128_t var_uuid;
851
852 /* For /var we insist that the uuid of the partition matches the
853 * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
854 * ID. Why? Unlike the other partitions /var is inherently
855 * installation specific, hence we need to be careful not to mount it
856 * in the wrong installation. By hashing the partition UUID from
857 * /etc/machine-id we can securely bind the partition to the
858 * installation. */
859
92e72028 860 r = sd_id128_get_machine_app_specific(SD_GPT_VAR, &var_uuid);
d4dffb85
LP		 861 if (r < 0)
862 return r;
863
864 if (!sd_id128_equal(var_uuid, id)) {
a52efa81
YW		 865 log_debug("Found a /var/ partition, but its UUID didn't match our expectations "
866 "(found: " SD_ID128_UUID_FORMAT_STR ", expected: " SD_ID128_UUID_FORMAT_STR "), ignoring.",
867 SD_ID128_FORMAT_VAL(id), SD_ID128_FORMAT_VAL(var_uuid));
d4dffb85
LP		 868 continue;
869 }
870 }
871
92e72028
ZJS		 872 rw = !(pflags & SD_GPT_FLAG_READ_ONLY);
873 growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS);
8c1be37e
LP		 874 }
875
22e932f4 876 if (type.designator != _PARTITION_DESIGNATOR_INVALID) {
0fb5036f 877 _cleanup_free_ char *t = NULL, *o = NULL, *l = NULL;
f7725647 878 _cleanup_close_ int mount_node_fd = -1;
18d73705 879 const char *options = NULL;
8c1be37e 880
22e932f4 881 if (m->partitions[type.designator].found) {
08fe0a53
LP		 882 /* For most partition types the first one we see wins. Except for the
883 * rootfs and /usr, where we do a version compare of the label, and
884 * let the newest version win. This permits a simple A/B versioning
885 * scheme in OS images. */
886
22e932f4
DDM		 887 if (compare_arch(type.arch, m->partitions[type.designator].architecture) <= 0)
888 continue;
889
890 if (!partition_designator_is_versioned(type.designator) ||
891 strverscmp_improved(m->partitions[type.designator].label, label) >= 0)
08fe0a53
LP		 892 continue;
893
22e932f4 894 dissected_partition_done(m->partitions + type.designator);
08fe0a53 895 }
8c1be37e 896
73d88b80 897 if (FLAGS_SET(flags, DISSECT_IMAGE_PIN_PARTITION_DEVICES) &&
f52261a0 898 type.designator != PARTITION_SWAP) {
f7725647
YW		 899 mount_node_fd = open_partition(node, /* is_partition = */ true, m->loop);
900 if (mount_node_fd < 0)
901 return mount_node_fd;
902 }
903
8c1be37e
LP		 904 if (fstype) {
905 t = strdup(fstype);
906 if (!t)
907 return -ENOMEM;
908 }
909
08fe0a53
LP		 910 if (label) {
911 l = strdup(label);
912 if (!l)
913 return -ENOMEM;
914 }
915
22e932f4 916 options = mount_options_from_designator(mount_options, type.designator);
18d73705
LB		 917 if (options) {
918 o = strdup(options);
919 if (!o)
920 return -ENOMEM;
921 }
922
22e932f4 923 m->partitions[type.designator] = (DissectedPartition) {
8c1be37e
LP		 924 .found = true,
925 .partno = nr,
926 .rw = rw,
de98f631 927 .growfs = growfs,
22e932f4 928 .architecture = type.arch,
0fb5036f 929 .node = TAKE_PTR(node),
1cc6c93a 930 .fstype = TAKE_PTR(t),
08fe0a53 931 .label = TAKE_PTR(l),
be30ad41 932 .uuid = id,
18d73705 933 .mount_options = TAKE_PTR(o),
f7725647 934 .mount_node_fd = TAKE_FD(mount_node_fd),
88b3300f
LP		 935 .offset = (uint64_t) start * 512,
936 .size = (uint64_t) size * 512,
8c1be37e 937 };
8c1be37e
LP		 938 }
939
940 } else if (is_mbr) {
941
a8c47660 942 switch (blkid_partition_get_type(pp)) {
8c1be37e 943
a8c47660
LP		 944 case 0x83: /* Linux partition */
945
946 if (pflags != 0x80) /* Bootable flag */
947 continue;
8c1be37e 948
a8c47660
LP		 949 if (generic_node)
950 multiple_generic = true;
951 else {
952 generic_nr = nr;
953 generic_rw = true;
de98f631 954 generic_growfs = false;
a8c47660
LP		 955 generic_node = strdup(node);
956 if (!generic_node)
957 return -ENOMEM;
958 }
959
960 break;
961
962 case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
f7725647 963 _cleanup_close_ int mount_node_fd = -1;
0fb5036f 964 _cleanup_free_ char *o = NULL;
a8c47660 965 sd_id128_t id = SD_ID128_NULL;
e3b9a5ff 966 const char *options = NULL;
a8c47660
LP		 967
968 /* First one wins */
234c2e16 969 if (m->partitions[PARTITION_XBOOTLDR].found)
a8c47660
LP		 970 continue;
971
73d88b80 972 if (FLAGS_SET(flags, DISSECT_IMAGE_PIN_PARTITION_DEVICES)) {
f7725647
YW		 973 mount_node_fd = open_partition(node, /* is_partition = */ true, m->loop);
974 if (mount_node_fd < 0)
975 return mount_node_fd;
976 }
977
e3b9a5ff 978 (void) blkid_partition_get_uuid_id128(pp, &id);
a8c47660 979
f5215bc8 980 options = mount_options_from_designator(mount_options, PARTITION_XBOOTLDR);
18d73705
LB		 981 if (options) {
982 o = strdup(options);
983 if (!o)
984 return -ENOMEM;
985 }
986
a8c47660
LP		 987 m->partitions[PARTITION_XBOOTLDR] = (DissectedPartition) {
988 .found = true,
989 .partno = nr,
990 .rw = true,
de98f631 991 .growfs = false,
a8c47660 992 .architecture = _ARCHITECTURE_INVALID,
0fb5036f 993 .node = TAKE_PTR(node),
a8c47660 994 .uuid = id,
18d73705 995 .mount_options = TAKE_PTR(o),
f7725647 996 .mount_node_fd = TAKE_FD(mount_node_fd),
88b3300f
LP		 997 .offset = (uint64_t) start * 512,
998 .size = (uint64_t) size * 512,
a8c47660
LP		 999 };
1000
1001 break;
1002 }}
8c1be37e
LP		 1003 }
1004 }
1005
22e932f4
DDM		 1006 if (!m->partitions[PARTITION_ROOT].found &&
1007 (m->partitions[PARTITION_ROOT_VERITY].found ||
1008 m->partitions[PARTITION_ROOT_VERITY_SIG].found))
1009 return -EADDRNOTAVAIL; /* Verity found but no matching rootfs? Something is off, refuse. */
49ae9d91 1010
8ee9615e
LP		 1011 /* Hmm, we found a signature partition but no Verity data? Something is off. */
1012 if (m->partitions[PARTITION_ROOT_VERITY_SIG].found && !m->partitions[PARTITION_ROOT_VERITY].found)
1013 return -EADDRNOTAVAIL;
7cf66030 1014
22e932f4
DDM		 1015 if (!m->partitions[PARTITION_USR].found &&
1016 (m->partitions[PARTITION_USR_VERITY].found ||
1017 m->partitions[PARTITION_USR_VERITY_SIG].found))
1018 return -EADDRNOTAVAIL; /* as above */
7cf66030 1019
22e932f4 1020 /* as above */
8ee9615e
LP		 1021 if (m->partitions[PARTITION_USR_VERITY_SIG].found && !m->partitions[PARTITION_USR_VERITY].found)
1022 return -EADDRNOTAVAIL;
1023
cb241a69
LP		 1024 /* If root and /usr are combined then insist that the architecture matches */
1025 if (m->partitions[PARTITION_ROOT].found &&
1026 m->partitions[PARTITION_USR].found &&
1027 (m->partitions[PARTITION_ROOT].architecture >= 0 &&
1028 m->partitions[PARTITION_USR].architecture >= 0 &&
1029 m->partitions[PARTITION_ROOT].architecture != m->partitions[PARTITION_USR].architecture))
1030 return -EADDRNOTAVAIL;
1031
4ab51780
LP		 1032 if (!m->partitions[PARTITION_ROOT].found &&
1033 !m->partitions[PARTITION_USR].found &&
1034 (flags & DISSECT_IMAGE_GENERIC_ROOT) &&
00a8b34f 1035 (!verity || !verity->root_hash || verity->designator != PARTITION_USR)) {
7cf66030 1036
1b010ae7 1037 /* OK, we found nothing usable, then check if there's a single generic partition, and use
4b5de5dd
LP		 1038 * that. If the root hash was set however, then we won't fall back to a generic node, because
1039 * the root hash decides. */
7cf66030
LP		 1040
1041 /* If we didn't find a properly marked root partition, but we did find a single suitable
1042 * generic Linux partition, then use this as root partition, if the caller asked for it. */
1043 if (multiple_generic)
1044 return -ENOTUNIQ;
1045
4b5de5dd
LP		 1046 /* If we didn't find a generic node, then we can't fix this up either */
1047 if (generic_node) {
f7725647 1048 _cleanup_close_ int mount_node_fd = -1;
4b5de5dd
LP		 1049 _cleanup_free_ char *o = NULL;
1050 const char *options;
8c1be37e 1051
73d88b80 1052 if (FLAGS_SET(flags, DISSECT_IMAGE_PIN_PARTITION_DEVICES)) {
f7725647
YW		 1053 mount_node_fd = open_partition(generic_node, /* is_partition = */ true, m->loop);
1054 if (mount_node_fd < 0)
1055 return mount_node_fd;
1056 }
1057
f5215bc8 1058 options = mount_options_from_designator(mount_options, PARTITION_ROOT);
18d73705
LB		 1059 if (options) {
1060 o = strdup(options);
1061 if (!o)
1062 return -ENOMEM;
1063 }
1064
1f8fb21c 1065 assert(generic_nr >= 0);
8c1be37e
LP		 1066 m->partitions[PARTITION_ROOT] = (DissectedPartition) {
1067 .found = true,
1068 .rw = generic_rw,
de98f631 1069 .growfs = generic_growfs,
8c1be37e
LP		 1070 .partno = generic_nr,
1071 .architecture = _ARCHITECTURE_INVALID,
1cc6c93a 1072 .node = TAKE_PTR(generic_node),
be30ad41 1073 .uuid = generic_uuid,
18d73705 1074 .mount_options = TAKE_PTR(o),
f7725647 1075 .mount_node_fd = TAKE_FD(mount_node_fd),
88b3300f
LP		 1076 .offset = UINT64_MAX,
1077 .size = UINT64_MAX,
8c1be37e 1078 };
e0f9e7bd 1079 }
8c1be37e
LP		 1080 }
1081
4b5de5dd
LP		 1082 /* Check if we have a root fs if we are told to do check. /usr alone is fine too, but only if appropriate flag for that is set too */
1083 if (FLAGS_SET(flags, DISSECT_IMAGE_REQUIRE_ROOT) &&
1084 !(m->partitions[PARTITION_ROOT].found || (m->partitions[PARTITION_USR].found && FLAGS_SET(flags, DISSECT_IMAGE_USR_NO_ROOT))))
1085 return -ENXIO;
1086
7b32164f
LP		 1087 if (m->partitions[PARTITION_ROOT_VERITY].found) {
1088 /* We only support one verity partition per image, i.e. can't do for both /usr and root fs */
1089 if (m->partitions[PARTITION_USR_VERITY].found)
1090 return -ENOTUNIQ;
1091
1092 /* We don't support verity enabled root with a split out /usr. Neither with nor without
1093 * verity there. (Note that we do support verity-less root with verity-full /usr, though.) */
1094 if (m->partitions[PARTITION_USR].found)
1095 return -EADDRNOTAVAIL;
1096 }
aee36b4e 1097
1903defc
LP		 1098 if (verity) {
1099 /* If a verity designator is specified, then insist that the matching partition exists */
1100 if (verity->designator >= 0 && !m->partitions[verity->designator].found)
1101 return -EADDRNOTAVAIL;
aee36b4e 1102
3dd73ea7
DDM		 1103 bool have_verity_sig_partition =
1104 m->partitions[verity->designator == PARTITION_USR ? PARTITION_USR_VERITY_SIG : PARTITION_ROOT_VERITY_SIG].found;
1105
1903defc 1106 if (verity->root_hash) {
8ee9615e
LP		 1107 /* If we have an explicit root hash and found the partitions for it, then we are ready to use
1108 * Verity, set things up for it */
1109
1903defc
LP		 1110 if (verity->designator < 0 || verity->designator == PARTITION_ROOT) {
1111 if (!m->partitions[PARTITION_ROOT_VERITY].found || !m->partitions[PARTITION_ROOT].found)
1112 return -EADDRNOTAVAIL;
4623e8e6 1113
1903defc
LP		 1114 /* If we found a verity setup, then the root partition is necessarily read-only. */
1115 m->partitions[PARTITION_ROOT].rw = false;
1116 m->verity_ready = true;
1903defc 1117
f9e0bb61
LP		 1118 } else {
1119 assert(verity->designator == PARTITION_USR);
1120
1903defc
LP		 1121 if (!m->partitions[PARTITION_USR_VERITY].found || !m->partitions[PARTITION_USR].found)
1122 return -EADDRNOTAVAIL;
4623e8e6 1123
1903defc
LP		 1124 m->partitions[PARTITION_USR].rw = false;
1125 m->verity_ready = true;
1126 }
8ee9615e
LP		 1127
1128 if (m->verity_ready)
3dd73ea7 1129 m->verity_sig_ready = verity->root_hash_sig || have_verity_sig_partition;
8ee9615e 1130
3dd73ea7 1131 } else if (have_verity_sig_partition) {
8ee9615e
LP		 1132
1133 /* If we found an embedded signature partition, we are ready, too. */
1134
1135 m->verity_ready = m->verity_sig_ready = true;
1136 m->partitions[verity->designator == PARTITION_USR ? PARTITION_USR : PARTITION_ROOT].rw = false;
aee36b4e 1137 }
4623e8e6
LP		 1138 }
1139
c80c9079
LP		 1140 r = dissected_image_probe_filesystems(m, fd);
1141 if (r < 0)
1142 return r;
1143
08f14be4
YW		 1144 return 0;
1145}
1146#endif
18b5886e 1147
08f14be4
YW		 1148int dissect_image_file(
1149 const char *path,
1150 const VeritySettings *verity,
1151 const MountOptions *mount_options,
1152 DissectImageFlags flags,
1153 DissectedImage **ret) {
8c1be37e 1154
08f14be4
YW		 1155#if HAVE_BLKID
1156 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
1157 _cleanup_close_ int fd = -1;
1158 int r;
8c1be37e 1159
08f14be4 1160 assert(path);
08f14be4 1161 assert(ret);
8c1be37e 1162
08f14be4
YW		 1163 fd = open(path, O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY);
1164 if (fd < 0)
1165 return -errno;
896f937f 1166
08f14be4
YW		 1167 r = fd_verify_regular(fd);
1168 if (r < 0)
1169 return r;
de98f631 1170
08f14be4
YW		 1171 r = dissected_image_new(path, &m);
1172 if (r < 0)
1173 return r;
1174
1175 r = dissect_image(m, fd, path, verity, mount_options, flags);
698bb074
YW		 1176 if (r < 0)
1177 return r;
8c1be37e 1178
1cc6c93a 1179 *ret = TAKE_PTR(m);
8c1be37e
LP		 1180 return 0;
1181#else
1182 return -EOPNOTSUPP;
1183#endif
1184}
1185
1186DissectedImage* dissected_image_unref(DissectedImage *m) {
8c1be37e
LP		 1187 if (!m)
1188 return NULL;
1189
ac1e1b5f 1190 /* First, clear dissected partitions. */
08fe0a53 1191 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++)
234c2e16 1192 dissected_partition_done(m->partitions + i);
8c1be37e 1193
ac1e1b5f
YW		 1194 /* Second, free decrypted images. This must be after dissected_partition_done(), as freeing
1195 * DecryptedImage may try to deactivate partitions. */
1196 decrypted_image_unref(m->decrypted_image);
1197
1e63dc4f
YW		 1198 /* Third, unref LoopDevice. This must be called after the above two, as freeing LoopDevice may try to
1199 * remove existing partitions on the loopback block device. */
1200 loop_device_unref(m->loop);
1201
593fe6c0 1202 free(m->image_name);
3b925504
LP		 1203 free(m->hostname);
1204 strv_free(m->machine_info);
1205 strv_free(m->os_release);
fab22946 1206 strv_free(m->initrd_release);
7718ac97 1207 strv_free(m->extension_release);
3b925504 1208
5fecf46d 1209 return mfree(m);
8c1be37e
LP		 1210}
1211
18b5886e 1212static int is_loop_device(const char *path) {
553e15f2 1213 char s[SYS_BLOCK_PATH_MAX("/../loop/")];
18b5886e
LP		 1214 struct stat st;
1215
1216 assert(path);
1217
1218 if (stat(path, &st) < 0)
1219 return -errno;
1220
1221 if (!S_ISBLK(st.st_mode))
1222 return -ENOTBLK;
1223
553e15f2 1224 xsprintf_sys_block_path(s, "/loop/", st.st_dev);
18b5886e
LP		 1225 if (access(s, F_OK) < 0) {
1226 if (errno != ENOENT)
1227 return -errno;
1228
1229 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
553e15f2 1230 xsprintf_sys_block_path(s, "/../loop/", st.st_dev);
18b5886e
LP		 1231 if (access(s, F_OK) < 0)
1232 return errno == ENOENT ? false : -errno;
1233 }
1234
1235 return true;
1236}
1237
f8ab7812 1238static int run_fsck(int node_fd, const char *fstype) {
cf32c486
LP		 1239 int r, exit_status;
1240 pid_t pid;
1241
f8ab7812 1242 assert(node_fd >= 0);
cf32c486
LP		 1243 assert(fstype);
1244
13556724 1245 r = fsck_exists_for_fstype(fstype);
cf32c486
LP		 1246 if (r < 0) {
1247 log_debug_errno(r, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype);
1248 return 0;
1249 }
1250 if (r == 0) {
f8ab7812 1251 log_debug("Not checking partition %s, as fsck for %s does not exist.", FORMAT_PROC_FD_PATH(node_fd), fstype);
cf32c486
LP		 1252 return 0;
1253 }
1254
f8ab7812
LP		 1255 r = safe_fork_full(
1256 "(fsck)",
1257 &node_fd, 1, /* Leave the node fd open */
1258 FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG|FORK_NULL_STDIO|FORK_CLOEXEC_OFF,
1259 &pid);
cf32c486
LP		 1260 if (r < 0)
1261 return log_debug_errno(r, "Failed to fork off fsck: %m");
1262 if (r == 0) {
1263 /* Child */
f8ab7812 1264 execl("/sbin/fsck", "/sbin/fsck", "-aT", FORMAT_PROC_FD_PATH(node_fd), NULL);
7e0ed2e9 1265 log_open();
cf32c486
LP		 1266 log_debug_errno(errno, "Failed to execl() fsck: %m");
1267 _exit(FSCK_OPERATIONAL_ERROR);
1268 }
1269
1270 exit_status = wait_for_terminate_and_check("fsck", pid, 0);
1271 if (exit_status < 0)
1272 return log_debug_errno(exit_status, "Failed to fork off /sbin/fsck: %m");
1273
1274 if ((exit_status & ~FSCK_ERROR_CORRECTED) != FSCK_SUCCESS) {
1275 log_debug("fsck failed with exit status %i.", exit_status);
1276
1277 if ((exit_status & (FSCK_SYSTEM_SHOULD_REBOOT|FSCK_ERRORS_LEFT_UNCORRECTED)) != 0)
1278 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN), "File system is corrupted, refusing.");
1279
1280 log_debug("Ignoring fsck error.");
1281 }
1282
1283 return 0;
1284}
1285
81939d9d
LP		 1286static int fs_grow(const char *node_path, const char *mount_path) {
1287 _cleanup_close_ int mount_fd = -1, node_fd = -1;
81939d9d
LP		 1288 uint64_t size, newsize;
1289 int r;
1290
1291 node_fd = open(node_path, O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY);
1292 if (node_fd < 0)
1293 return log_debug_errno(errno, "Failed to open node device %s: %m", node_path);
1294
1295 if (ioctl(node_fd, BLKGETSIZE64, &size) != 0)
1296 return log_debug_errno(errno, "Failed to get block device size of %s: %m", node_path);
1297
1298 mount_fd = open(mount_path, O_RDONLY|O_DIRECTORY|O_CLOEXEC);
1299 if (mount_fd < 0)
1300 return log_debug_errno(errno, "Failed to open mountd file system %s: %m", mount_path);
1301
1302 log_debug("Resizing \"%s\" to %"PRIu64" bytes...", mount_path, size);
1303 r = resize_fs(mount_fd, size, &newsize);
1304 if (r < 0)
1305 return log_debug_errno(r, "Failed to resize \"%s\" to %"PRIu64" bytes: %m", mount_path, size);
1306
1307 if (newsize == size)
1308 log_debug("Successfully resized \"%s\" to %s bytes.",
2b59bf51 1309 mount_path, FORMAT_BYTES(newsize));
81939d9d
LP		 1310 else {
1311 assert(newsize < size);
1312 log_debug("Successfully resized \"%s\" to %s bytes (%"PRIu64" bytes lost due to blocksize).",
2b59bf51 1313 mount_path, FORMAT_BYTES(newsize), size - newsize);
81939d9d
LP		 1314 }
1315
1316 return 0;
1317}
1318
18b5886e
LP		 1319static int mount_partition(
1320 DissectedPartition *m,
1321 const char *where,
1322 const char *directory,
2d3a5a73 1323 uid_t uid_shift,
21b61b1d 1324 uid_t uid_range,
18b5886e
LP		 1325 DissectImageFlags flags) {
1326
2d3a5a73
LP		 1327 _cleanup_free_ char *chased = NULL, *options = NULL;
1328 const char *p, *node, *fstype;
21b61b1d 1329 bool rw, remap_uid_gid = false;
2eedfd2d 1330 int r;
8c1be37e
LP		 1331
1332 assert(m);
1333 assert(where);
1334
f7725647
YW		 1335 if (m->mount_node_fd < 0)
1336 return 0;
1337
4dc28665 1338 /* Use decrypted node and matching fstype if available, otherwise use the original device */
f7725647 1339 node = FORMAT_PROC_FD_PATH(m->mount_node_fd);
4dc28665 1340 fstype = m->decrypted_node ? m->decrypted_fstype: m->fstype;
18b5886e 1341
4dc28665
LP		 1342 if (!fstype)
1343 return -EAFNOSUPPORT;
8c1be37e 1344
68ac5118
ZJS		 1345 /* We are looking at an encrypted partition? This either means stacked encryption, or the caller
1346 * didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error for this
1347 * case. */
4dc28665 1348 if (streq(fstype, "crypto_LUKS"))
fa45d12c 1349 return -EUNATCH;
18b5886e 1350
ef9c184d 1351 rw = m->rw && !(flags & DISSECT_IMAGE_MOUNT_READ_ONLY);
8c1be37e 1352
cf32c486 1353 if (FLAGS_SET(flags, DISSECT_IMAGE_FSCK) && rw) {
f8ab7812 1354 r = run_fsck(m->mount_node_fd, fstype);
cf32c486
LP		 1355 if (r < 0)
1356 return r;
1357 }
1358
2eedfd2d 1359 if (directory) {
334eb5b0
LP		 1360 /* Automatically create missing mount points inside the image, if necessary. */
1361 r = mkdir_p_root(where, directory, uid_shift, (gid_t) uid_shift, 0755);
1362 if (r < 0 && r != -EROFS)
1363 return r;
1f0f82f1 1364
a5648b80 1365 r = chase_symlinks(directory, where, CHASE_PREFIX_ROOT, &chased, NULL);
2eedfd2d
LP		 1366 if (r < 0)
1367 return r;
1368
1369 p = chased;
9842905e
LP		 1370 } else {
1371 /* Create top-level mount if missing – but only if this is asked for. This won't modify the
1372 * image (as the branch above does) but the host hierarchy, and the created directory might
1373 * survive our mount in the host hierarchy hence. */
1374 if (FLAGS_SET(flags, DISSECT_IMAGE_MKDIR)) {
1375 r = mkdir_p(where, 0755);
1376 if (r < 0)
1377 return r;
1378 }
1379
8c1be37e 1380 p = where;
9842905e 1381 }
8c1be37e 1382
18b5886e 1383 /* If requested, turn on discard support. */
154d2269 1384 if (fstype_can_discard(fstype) &&
18b5886e 1385 ((flags & DISSECT_IMAGE_DISCARD) ||
3afda7c7 1386 ((flags & DISSECT_IMAGE_DISCARD_ON_LOOP) && is_loop_device(m->node) > 0))) {
2d3a5a73
LP		 1387 options = strdup("discard");
1388 if (!options)
1389 return -ENOMEM;
1390 }
1391
21b61b1d 1392 if (uid_is_valid(uid_shift) && uid_shift != 0) {
2d3a5a73 1393
21b61b1d
LP		 1394 if (fstype_can_uid_gid(fstype)) {
1395 _cleanup_free_ char *uid_option = NULL;
2d3a5a73 1396
21b61b1d
LP		 1397 if (asprintf(&uid_option, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0)
1398 return -ENOMEM;
1399
1400 if (!strextend_with_separator(&options, ",", uid_option))
1401 return -ENOMEM;
1402 } else if (FLAGS_SET(flags, DISSECT_IMAGE_MOUNT_IDMAPPED))
1403 remap_uid_gid = true;
2d3a5a73 1404 }
8c1be37e 1405
18d73705 1406 if (!isempty(m->mount_options))
c2bc710b 1407 if (!strextend_with_separator(&options, ",", m->mount_options))
18d73705
LB		 1408 return -ENOMEM;
1409
b620bf33
LP		 1410 /* So, when you request MS_RDONLY from ext4, then this means nothing. It happily still writes to the
1411 * backing storage. What's worse, the BLKRO[GS]ET flag and (in case of loopback devices)
1412 * LO_FLAGS_READ_ONLY don't mean anything, they affect userspace accesses only, and write accesses
1413 * from the upper file system still get propagated through to the underlying file system,
1414 * unrestricted. To actually get ext4/xfs/btrfs to stop writing to the device we need to specify
1415 * "norecovery" as mount option, in addition to MS_RDONLY. Yes, this sucks, since it means we need to
1416 * carry a per file system table here.
1417 *
1418 * Note that this means that we might not be able to mount corrupted file systems as read-only
1419 * anymore (since in some cases the kernel implementations will refuse mounting when corrupted,
1420 * read-only and "norecovery" is specified). But I think for the case of automatically determined
1421 * mount options for loopback devices this is the right choice, since otherwise using the same
1422 * loopback file twice even in read-only mode, is going to fail badly sooner or later. The usecase of
1423 * making reuse of the immutable images "just work" is more relevant to us than having read-only
1424 * access that actually modifies stuff work on such image files. Or to say this differently: if
1425 * people want their file systems to be fixed up they should just open them in writable mode, where
1426 * all these problems don't exist. */
1427 if (!rw && STRPTR_IN_SET(fstype, "ext3", "ext4", "xfs", "btrfs"))
1428 if (!strextend_with_separator(&options, ",", "norecovery"))
1429 return -ENOMEM;
1430
511a8cfe 1431 r = mount_nofollow_verbose(LOG_DEBUG, node, p, fstype, MS_NODEV|(rw ? 0 : MS_RDONLY), options);
d9223c07
LP		 1432 if (r < 0)
1433 return r;
1434
81939d9d
LP		 1435 if (rw && m->growfs && FLAGS_SET(flags, DISSECT_IMAGE_GROWFS))
1436 (void) fs_grow(node, p);
1437
21b61b1d 1438 if (remap_uid_gid) {
2b2777ed 1439 r = remount_idmap(p, uid_shift, uid_range, UID_INVALID, REMOUNT_IDMAPPING_HOST_ROOT);
21b61b1d
LP		 1440 if (r < 0)
1441 return r;
1442 }
1443
d9223c07 1444 return 1;
8c1be37e
LP		 1445}
1446
7cf66030
LP		 1447static int mount_root_tmpfs(const char *where, uid_t uid_shift, DissectImageFlags flags) {
1448 _cleanup_free_ char *options = NULL;
1449 int r;
1450
1451 assert(where);
1452
1453 /* For images that contain /usr/ but no rootfs, let's mount rootfs as tmpfs */
1454
1455 if (FLAGS_SET(flags, DISSECT_IMAGE_MKDIR)) {
1456 r = mkdir_p(where, 0755);
1457 if (r < 0)
1458 return r;
1459 }
1460
1461 if (uid_is_valid(uid_shift)) {
1462 if (asprintf(&options, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0)
1463 return -ENOMEM;
1464 }
1465
1466 r = mount_nofollow_verbose(LOG_DEBUG, "rootfs", where, "tmpfs", MS_NODEV, options);
1467 if (r < 0)
1468 return r;
1469
1470 return 1;
1471}
1472
21b61b1d
LP		 1473int dissected_image_mount(
1474 DissectedImage *m,
1475 const char *where,
1476 uid_t uid_shift,
1477 uid_t uid_range,
1478 DissectImageFlags flags) {
1479
1f0f82f1 1480 int r, xbootldr_mounted;
8c1be37e
LP		 1481
1482 assert(m);
1483 assert(where);
1484
fa45d12c
LP		 1485 /* Returns:
1486 *
1487 * -ENXIO → No root partition found
7718ac97 1488 * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release/extension-release file found
fa45d12c
LP		 1489 * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet
1490 * -EUCLEAN → fsck for file system failed
1491 * -EBUSY → File system already mounted/used elsewhere (kernel)
4dc28665 1492 * -EAFNOSUPPORT → File system type not supported or not known
fa45d12c
LP		 1493 */
1494
7cf66030
LP		 1495 if (!(m->partitions[PARTITION_ROOT].found ||
1496 (m->partitions[PARTITION_USR].found && FLAGS_SET(flags, DISSECT_IMAGE_USR_NO_ROOT))))
1497 return -ENXIO; /* Require a root fs or at least a /usr/ fs (the latter is subject to a flag of its own) */
8c1be37e 1498
2d3a5a73 1499 if ((flags & DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY) == 0) {
7cf66030
LP		 1500
1501 /* First mount the root fs. If there's none we use a tmpfs. */
1502 if (m->partitions[PARTITION_ROOT].found)
21b61b1d 1503 r = mount_partition(m->partitions + PARTITION_ROOT, where, NULL, uid_shift, uid_range, flags);
7cf66030
LP		 1504 else
1505 r = mount_root_tmpfs(where, uid_shift, flags);
2d3a5a73
LP		 1506 if (r < 0)
1507 return r;
aee36b4e 1508
aee36b4e 1509 /* For us mounting root always means mounting /usr as well */
21b61b1d 1510 r = mount_partition(m->partitions + PARTITION_USR, where, "/usr", uid_shift, uid_range, flags);
aee36b4e
LP		 1511 if (r < 0)
1512 return r;
03bcb6d4 1513
9ccb531a
LB		 1514 if ((flags & (DISSECT_IMAGE_VALIDATE_OS|DISSECT_IMAGE_VALIDATE_OS_EXT)) != 0) {
1515 /* If either one of the validation flags are set, ensure that the image qualifies
1516 * as one or the other (or both). */
1517 bool ok = false;
1518
1519 if (FLAGS_SET(flags, DISSECT_IMAGE_VALIDATE_OS)) {
1520 r = path_is_os_tree(where);
1521 if (r < 0)
1522 return r;
1523 if (r > 0)
1524 ok = true;
1525 }
1526 if (!ok && FLAGS_SET(flags, DISSECT_IMAGE_VALIDATE_OS_EXT)) {
06768b90 1527 r = path_is_extension_tree(where, m->image_name, FLAGS_SET(flags, DISSECT_IMAGE_RELAX_SYSEXT_CHECK));
7718ac97
LB		 1528 if (r < 0)
1529 return r;
9ccb531a
LB		 1530 if (r > 0)
1531 ok = true;
7718ac97 1532 }
9ccb531a
LB		 1533
1534 if (!ok)
1535 return -ENOMEDIUM;
03bcb6d4 1536 }
2d3a5a73
LP		 1537 }
1538
705727fd 1539 if (flags & DISSECT_IMAGE_MOUNT_ROOT_ONLY)
2d3a5a73 1540 return 0;
8c1be37e 1541
21b61b1d 1542 r = mount_partition(m->partitions + PARTITION_HOME, where, "/home", uid_shift, uid_range, flags);
8c1be37e
LP		 1543 if (r < 0)
1544 return r;
1545
21b61b1d 1546 r = mount_partition(m->partitions + PARTITION_SRV, where, "/srv", uid_shift, uid_range, flags);
8c1be37e
LP		 1547 if (r < 0)
1548 return r;
1549
21b61b1d 1550 r = mount_partition(m->partitions + PARTITION_VAR, where, "/var", uid_shift, uid_range, flags);
d4dffb85
LP		 1551 if (r < 0)
1552 return r;
1553
21b61b1d 1554 r = mount_partition(m->partitions + PARTITION_TMP, where, "/var/tmp", uid_shift, uid_range, flags);
d4dffb85
LP		 1555 if (r < 0)
1556 return r;
1557
21b61b1d 1558 xbootldr_mounted = mount_partition(m->partitions + PARTITION_XBOOTLDR, where, "/boot", uid_shift, uid_range, flags);
1f0f82f1
LP		 1559 if (xbootldr_mounted < 0)
1560 return xbootldr_mounted;
d9223c07 1561
8c1be37e 1562 if (m->partitions[PARTITION_ESP].found) {
1f0f82f1
LP		 1563 int esp_done = false;
1564
d9223c07
LP		 1565 /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it
1566 * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */
8c1be37e 1567
a5648b80 1568 r = chase_symlinks("/efi", where, CHASE_PREFIX_ROOT, NULL, NULL);
1f0f82f1
LP		 1569 if (r < 0) {
1570 if (r != -ENOENT)
d9223c07 1571 return r;
8c1be37e 1572
1f0f82f1
LP		 1573 /* /efi doesn't exist. Let's see if /boot is suitable then */
1574
1575 if (!xbootldr_mounted) {
1576 _cleanup_free_ char *p = NULL;
2eedfd2d 1577
1f0f82f1
LP		 1578 r = chase_symlinks("/boot", where, CHASE_PREFIX_ROOT, &p, NULL);
1579 if (r < 0) {
1580 if (r != -ENOENT)
1581 return r;
db55bbf2 1582 } else if (dir_is_empty(p, /* ignore_hidden_or_backup= */ false) > 0) {
1f0f82f1 1583 /* It exists and is an empty directory. Let's mount the ESP there. */
21b61b1d 1584 r = mount_partition(m->partitions + PARTITION_ESP, where, "/boot", uid_shift, uid_range, flags);
1f0f82f1
LP		 1585 if (r < 0)
1586 return r;
1587
1588 esp_done = true;
1589 }
2eedfd2d 1590 }
8c1be37e 1591 }
1f0f82f1
LP		 1592
1593 if (!esp_done) {
1594 /* OK, let's mount the ESP now to /efi (possibly creating the dir if missing) */
1595
21b61b1d 1596 r = mount_partition(m->partitions + PARTITION_ESP, where, "/efi", uid_shift, uid_range, flags);
1f0f82f1
LP		 1597 if (r < 0)
1598 return r;
1599 }
8c1be37e
LP		 1600 }
1601
1602 return 0;
1603}
1604
21b61b1d
LP		 1605int dissected_image_mount_and_warn(
1606 DissectedImage *m,
1607 const char *where,
1608 uid_t uid_shift,
1609 uid_t uid_range,
1610 DissectImageFlags flags) {
1611
af187ab2
LP		 1612 int r;
1613
1614 assert(m);
1615 assert(where);
1616
21b61b1d 1617 r = dissected_image_mount(m, where, uid_shift, uid_range, flags);
af187ab2
LP		 1618 if (r == -ENXIO)
1619 return log_error_errno(r, "Not root file system found in image.");
1620 if (r == -EMEDIUMTYPE)
7718ac97 1621 return log_error_errno(r, "No suitable os-release/extension-release file in image found.");
af187ab2
LP		 1622 if (r == -EUNATCH)
1623 return log_error_errno(r, "Encrypted file system discovered, but decryption not requested.");
1624 if (r == -EUCLEAN)
1625 return log_error_errno(r, "File system check on image failed.");
1626 if (r == -EBUSY)
1627 return log_error_errno(r, "File system already mounted elsewhere.");
4dc28665
LP		 1628 if (r == -EAFNOSUPPORT)
1629 return log_error_errno(r, "File system type not supported or not known.");
af187ab2
LP		 1630 if (r < 0)
1631 return log_error_errno(r, "Failed to mount image: %m");
1632
1633 return r;
1634}
1635
349cc4a5 1636#if HAVE_LIBCRYPTSETUP
9321ad51 1637struct DecryptedPartition {
18b5886e
LP		 1638 struct crypt_device *device;
1639 char *name;
1640 bool relinquished;
9321ad51
YW		 1641};
1642#endif
1643
1644typedef struct DecryptedPartition DecryptedPartition;
18b5886e
LP		 1645
1646struct DecryptedImage {
9321ad51 1647 unsigned n_ref;
18b5886e
LP		 1648 DecryptedPartition *decrypted;
1649 size_t n_decrypted;
18b5886e 1650};
18b5886e 1651
9321ad51 1652static DecryptedImage* decrypted_image_free(DecryptedImage *d) {
349cc4a5 1653#if HAVE_LIBCRYPTSETUP
18b5886e
LP		 1654 int r;
1655
1656 if (!d)
1657 return NULL;
1658
67f63ee5 1659 for (size_t i = 0; i < d->n_decrypted; i++) {
18b5886e
LP		 1660 DecryptedPartition *p = d->decrypted + i;
1661
1662 if (p->device && p->name && !p->relinquished) {
ea16d7f4
YW		 1663 /* Let's deactivate lazily, as the dm volume may be already/still used by other processes. */
1664 r = sym_crypt_deactivate_by_name(p->device, p->name, CRYPT_DEACTIVATE_DEFERRED);
18b5886e
LP		 1665 if (r < 0)
1666 log_debug_errno(r, "Failed to deactivate encrypted partition %s", p->name);
1667 }
1668
1669 if (p->device)
0d12936d 1670 sym_crypt_free(p->device);
18b5886e
LP		 1671 free(p->name);
1672 }
1673
f91861e4 1674 free(d->decrypted);
18b5886e
LP		 1675 free(d);
1676#endif
1677 return NULL;
1678}
1679
9321ad51
YW		 1680DEFINE_TRIVIAL_REF_UNREF_FUNC(DecryptedImage, decrypted_image, decrypted_image_free);
1681
349cc4a5 1682#if HAVE_LIBCRYPTSETUP
9321ad51
YW		 1683static int decrypted_image_new(DecryptedImage **ret) {
1684 _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL;
1685
1686 assert(ret);
1687
1688 d = new(DecryptedImage, 1);
1689 if (!d)
1690 return -ENOMEM;
1691
1692 *d = (DecryptedImage) {
1693 .n_ref = 1,
1694 };
1695
1696 *ret = TAKE_PTR(d);
1697 return 0;
1698}
4623e8e6
LP		 1699
1700static int make_dm_name_and_node(const void *original_node, const char *suffix, char **ret_name, char **ret_node) {
1701 _cleanup_free_ char *name = NULL, *node = NULL;
1702 const char *base;
1703
1704 assert(original_node);
1705 assert(suffix);
1706 assert(ret_name);
1707 assert(ret_node);
1708
1709 base = strrchr(original_node, '/');
1710 if (!base)
ac1f3ad0
LB		 1711 base = original_node;
1712 else
1713 base++;
4623e8e6
LP		 1714 if (isempty(base))
1715 return -EINVAL;
1716
1717 name = strjoin(base, suffix);
1718 if (!name)
1719 return -ENOMEM;
1720 if (!filename_is_valid(name))
1721 return -EINVAL;
1722
0d12936d 1723 node = path_join(sym_crypt_get_dir(), name);
4623e8e6
LP		 1724 if (!node)
1725 return -ENOMEM;
1726
1cc6c93a
YW		 1727 *ret_name = TAKE_PTR(name);
1728 *ret_node = TAKE_PTR(node);
4623e8e6 1729
4623e8e6
LP		 1730 return 0;
1731}
1732
18b5886e
LP		 1733static int decrypt_partition(
1734 DissectedPartition *m,
1735 const char *passphrase,
1736 DissectImageFlags flags,
1737 DecryptedImage *d) {
1738
1739 _cleanup_free_ char *node = NULL, *name = NULL;
0d12936d 1740 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
f7725647 1741 _cleanup_close_ int fd = -1;
18b5886e
LP		 1742 int r;
1743
1744 assert(m);
1745 assert(d);
1746
1747 if (!m->found || !m->node || !m->fstype)
1748 return 0;
1749
1750 if (!streq(m->fstype, "crypto_LUKS"))
1751 return 0;
1752
bdd73ac5
ZJS		 1753 if (!passphrase)
1754 return -ENOKEY;
1755
0d12936d
LP		 1756 r = dlopen_cryptsetup();
1757 if (r < 0)
1758 return r;
1759
4623e8e6
LP		 1760 r = make_dm_name_and_node(m->node, "-decrypted", &name, &node);
1761 if (r < 0)
1762 return r;
18b5886e 1763
319a4f4b 1764 if (!GREEDY_REALLOC0(d->decrypted, d->n_decrypted + 1))
18b5886e
LP		 1765 return -ENOMEM;
1766
0d12936d 1767 r = sym_crypt_init(&cd, m->node);
18b5886e 1768 if (r < 0)
715cbb81 1769 return log_debug_errno(r, "Failed to initialize dm-crypt: %m");
18b5886e 1770
efc3b12f 1771 cryptsetup_enable_logging(cd);
1887032f 1772
0d12936d 1773 r = sym_crypt_load(cd, CRYPT_LUKS, NULL);
294bd454
ZJS		 1774 if (r < 0)
1775 return log_debug_errno(r, "Failed to load LUKS metadata: %m");
18b5886e 1776
0d12936d 1777 r = sym_crypt_activate_by_passphrase(cd, name, CRYPT_ANY_SLOT, passphrase, strlen(passphrase),
ef9c184d 1778 ((flags & DISSECT_IMAGE_DEVICE_READ_ONLY) ? CRYPT_ACTIVATE_READONLY : 0) |
0d12936d 1779 ((flags & DISSECT_IMAGE_DISCARD_ON_CRYPTO) ? CRYPT_ACTIVATE_ALLOW_DISCARDS : 0));
294bd454 1780 if (r < 0) {
715cbb81 1781 log_debug_errno(r, "Failed to activate LUKS device: %m");
294bd454 1782 return r == -EPERM ? -EKEYREJECTED : r;
18b5886e 1783 }
18b5886e 1784
f7725647
YW		 1785 fd = open(node, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
1786 if (fd < 0)
1787 return log_debug_errno(errno, "Failed to open %s: %m", node);
1788
94344385
LP		 1789 d->decrypted[d->n_decrypted++] = (DecryptedPartition) {
1790 .name = TAKE_PTR(name),
1791 .device = TAKE_PTR(cd),
1792 };
18b5886e 1793
1cc6c93a 1794 m->decrypted_node = TAKE_PTR(node);
f7725647 1795 close_and_replace(m->mount_node_fd, fd);
18b5886e
LP		 1796
1797 return 0;
4623e8e6
LP		 1798}
1799
89e62e0b
LP		 1800static int verity_can_reuse(
1801 const VeritySettings *verity,
1802 const char *name,
1803 struct crypt_device **ret_cd) {
1804
ac1f3ad0
LB		 1805 /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
1806 _cleanup_free_ char *root_hash_existing = NULL;
0d12936d 1807 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
ac1f3ad0 1808 struct crypt_params_verity crypt_params = {};
89e62e0b 1809 size_t root_hash_existing_size;
ac1f3ad0
LB		 1810 int r;
1811
89e62e0b
LP		 1812 assert(verity);
1813 assert(name);
ac1f3ad0
LB		 1814 assert(ret_cd);
1815
0d12936d 1816 r = sym_crypt_init_by_name(&cd, name);
ac1f3ad0
LB		 1817 if (r < 0)
1818 return log_debug_errno(r, "Error opening verity device, crypt_init_by_name failed: %m");
0d12936d 1819
c719805e
LP		 1820 cryptsetup_enable_logging(cd);
1821
0d12936d 1822 r = sym_crypt_get_verity_info(cd, &crypt_params);
ac1f3ad0
LB		 1823 if (r < 0)
1824 return log_debug_errno(r, "Error opening verity device, crypt_get_verity_info failed: %m");
0d12936d 1825
89e62e0b
LP		 1826 root_hash_existing_size = verity->root_hash_size;
1827 root_hash_existing = malloc0(root_hash_existing_size);
ac1f3ad0
LB		 1828 if (!root_hash_existing)
1829 return -ENOMEM;
0d12936d
LP		 1830
1831 r = sym_crypt_volume_key_get(cd, CRYPT_ANY_SLOT, root_hash_existing, &root_hash_existing_size, NULL, 0);
ac1f3ad0
LB		 1832 if (r < 0)
1833 return log_debug_errno(r, "Error opening verity device, crypt_volume_key_get failed: %m");
89e62e0b
LP		 1834 if (verity->root_hash_size != root_hash_existing_size ||
1835 memcmp(root_hash_existing, verity->root_hash, verity->root_hash_size) != 0)
ac1f3ad0 1836 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but root hashes are different.");
89e62e0b 1837
ac1f3ad0 1838#if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
89e62e0b
LP		 1839 /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the
1840 * same settings, so that a previous unsigned mount will not be reused if the user asks to use
28423d9a 1841 * signing for the new one, and vice versa. */
89e62e0b 1842 if (!!verity->root_hash_sig != !!(crypt_params.flags & CRYPT_VERITY_ROOT_HASH_SIGNATURE))
ac1f3ad0
LB		 1843 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but signature settings are not the same.");
1844#endif
1845
1846 *ret_cd = TAKE_PTR(cd);
1847 return 0;
1848}
1849
75db809a 1850static inline char* dm_deferred_remove_clean(char *name) {
ac1f3ad0 1851 if (!name)
75db809a 1852 return NULL;
0d12936d
LP		 1853
1854 (void) sym_crypt_deactivate_by_name(NULL, name, CRYPT_DEACTIVATE_DEFERRED);
75db809a 1855 return mfree(name);
ac1f3ad0
LB		 1856}
1857DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean);
1858
c2fa92e7
LP		 1859static int validate_signature_userspace(const VeritySettings *verity) {
1860#if HAVE_OPENSSL
1861 _cleanup_(sk_X509_free_allp) STACK_OF(X509) *sk = NULL;
1862 _cleanup_strv_free_ char **certs = NULL;
1863 _cleanup_(PKCS7_freep) PKCS7 *p7 = NULL;
1864 _cleanup_free_ char *s = NULL;
1865 _cleanup_(BIO_freep) BIO *bio = NULL; /* 'bio' must be freed first, 's' second, hence keep this order
1866 * of declaration in place, please */
1867 const unsigned char *d;
c2fa92e7
LP		 1868 int r;
1869
1870 assert(verity);
1871 assert(verity->root_hash);
1872 assert(verity->root_hash_sig);
1873
1874 /* Because installing a signature certificate into the kernel chain is so messy, let's optionally do
1875 * userspace validation. */
1876
1877 r = conf_files_list_nulstr(&certs, ".crt", NULL, CONF_FILES_REGULAR|CONF_FILES_FILTER_MASKED, CONF_PATHS_NULSTR("verity.d"));
1878 if (r < 0)
1879 return log_debug_errno(r, "Failed to enumerate certificates: %m");
1880 if (strv_isempty(certs)) {
1881 log_debug("No userspace dm-verity certificates found.");
1882 return 0;
1883 }
1884
1885 d = verity->root_hash_sig;
1886 p7 = d2i_PKCS7(NULL, &d, (long) verity->root_hash_sig_size);
1887 if (!p7)
1888 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to parse PKCS7 DER signature data.");
1889
1890 s = hexmem(verity->root_hash, verity->root_hash_size);
1891 if (!s)
1892 return log_oom_debug();
1893
1894 bio = BIO_new_mem_buf(s, strlen(s));
1895 if (!bio)
1896 return log_oom_debug();
1897
1898 sk = sk_X509_new_null();
1899 if (!sk)
1900 return log_oom_debug();
1901
1902 STRV_FOREACH(i, certs) {
1903 _cleanup_(X509_freep) X509 *c = NULL;
1904 _cleanup_fclose_ FILE *f = NULL;
1905
1906 f = fopen(*i, "re");
1907 if (!f) {
1908 log_debug_errno(errno, "Failed to open '%s', ignoring: %m", *i);
1909 continue;
1910 }
1911
1912 c = PEM_read_X509(f, NULL, NULL, NULL);
1913 if (!c) {
1914 log_debug("Failed to load X509 certificate '%s', ignoring.", *i);
1915 continue;
1916 }
1917
1918 if (sk_X509_push(sk, c) == 0)
1919 return log_oom_debug();
1920
1921 TAKE_PTR(c);
1922 }
1923
1924 r = PKCS7_verify(p7, sk, NULL, bio, NULL, PKCS7_NOINTERN|PKCS7_NOVERIFY);
1925 if (r)
1926 log_debug("Userspace PKCS#7 validation succeeded.");
1927 else
1928 log_debug("Userspace PKCS#7 validation failed: %s", ERR_error_string(ERR_get_error(), NULL));
1929
1930 return r;
1931#else
1932 log_debug("Not doing client-side validation of dm-verity root hash signatures, OpenSSL support disabled.");
1933 return 0;
1934#endif
1935}
1936
1937static int do_crypt_activate_verity(
1938 struct crypt_device *cd,
1939 const char *name,
1940 const VeritySettings *verity) {
1941
1942 bool check_signature;
1943 int r;
1944
1945 assert(cd);
1946 assert(name);
1947 assert(verity);
1948
1949 if (verity->root_hash_sig) {
1950 r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIGNATURE");
1951 if (r < 0 && r != -ENXIO)
1952 log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIGNATURE");
1953
1954 check_signature = r != 0;
1955 } else
1956 check_signature = false;
1957
1958 if (check_signature) {
1959
1960#if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1961 /* First, if we have support for signed keys in the kernel, then try that first. */
1962 r = sym_crypt_activate_by_signed_key(
1963 cd,
1964 name,
1965 verity->root_hash,
1966 verity->root_hash_size,
1967 verity->root_hash_sig,
1968 verity->root_hash_sig_size,
1969 CRYPT_ACTIVATE_READONLY);
1970 if (r >= 0)
1971 return r;
1972
1973 log_debug("Validation of dm-verity signature failed via the kernel, trying userspace validation instead.");
1974#else
1975 log_debug("Activation of verity device with signature requested, but not supported via the kernel by %s due to missing crypt_activate_by_signed_key(), trying userspace validation instead.",
1976 program_invocation_short_name);
1977#endif
1978
1979 /* So this didn't work via the kernel, then let's try userspace validation instead. If that
1980 * works we'll try to activate without telling the kernel the signature. */
1981
1982 r = validate_signature_userspace(verity);
1983 if (r < 0)
1984 return r;
1985 if (r == 0)
1986 return log_debug_errno(SYNTHETIC_ERRNO(ENOKEY),
1987 "Activation of signed Verity volume worked neither via the kernel nor in userspace, can't activate.");
1988 }
1989
1990 return sym_crypt_activate_by_volume_key(
1991 cd,
1992 name,
1993 verity->root_hash,
1994 verity->root_hash_size,
1995 CRYPT_ACTIVATE_READONLY);
1996}
1997
ad361a50
YW		 1998static usec_t verity_timeout(void) {
1999 usec_t t = 100 * USEC_PER_MSEC;
2000 const char *e;
2001 int r;
2002
2003 /* On slower machines, like non-KVM vm, setting up device may take a long time.
2004 * Let's make the timeout configurable. */
2005
2006 e = getenv("SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC");
2007 if (!e)
2008 return t;
2009
2010 r = parse_sec(e, &t);
2011 if (r < 0)
2012 log_debug_errno(r,
2013 "Failed to parse timeout specified in $SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC, "
2014 "using the default timeout (%s).",
2015 FORMAT_TIMESPAN(t, USEC_PER_MSEC));
2016
2017 return t;
2018}
2019
4623e8e6 2020static int verity_partition(
aee36b4e 2021 PartitionDesignator designator,
4623e8e6
LP		 2022 DissectedPartition *m,
2023 DissectedPartition *v,
89e62e0b 2024 const VeritySettings *verity,
4623e8e6
LP		 2025 DissectImageFlags flags,
2026 DecryptedImage *d) {
2027
0d12936d 2028 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
ac1f3ad0 2029 _cleanup_(dm_deferred_remove_cleanp) char *restore_deferred_remove = NULL;
89e62e0b 2030 _cleanup_free_ char *node = NULL, *name = NULL;
f7725647 2031 _cleanup_close_ int mount_node_fd = -1;
4623e8e6
LP		 2032 int r;
2033
2034 assert(m);
89e62e0b 2035 assert(v || (verity && verity->data_path));
4623e8e6 2036
89e62e0b 2037 if (!verity || !verity->root_hash)
4623e8e6 2038 return 0;
aee36b4e
LP		 2039 if (!((verity->designator < 0 && designator == PARTITION_ROOT) ||
2040 (verity->designator == designator)))
2041 return 0;
4623e8e6
LP		 2042
2043 if (!m->found || !m->node || !m->fstype)
2044 return 0;
89e62e0b 2045 if (!verity->data_path) {
e7cbe5cb
LB		 2046 if (!v->found || !v->node || !v->fstype)
2047 return 0;
4623e8e6 2048
e7cbe5cb
LB		 2049 if (!streq(v->fstype, "DM_verity_hash"))
2050 return 0;
2051 }
4623e8e6 2052
0d12936d
LP		 2053 r = dlopen_cryptsetup();
2054 if (r < 0)
2055 return r;
2056
ac1f3ad0
LB		 2057 if (FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) {
2058 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
2059 _cleanup_free_ char *root_hash_encoded = NULL;
0d12936d 2060
89e62e0b 2061 root_hash_encoded = hexmem(verity->root_hash, verity->root_hash_size);
ac1f3ad0
LB		 2062 if (!root_hash_encoded)
2063 return -ENOMEM;
aee36b4e 2064
ac1f3ad0
LB		 2065 r = make_dm_name_and_node(root_hash_encoded, "-verity", &name, &node);
2066 } else
2067 r = make_dm_name_and_node(m->node, "-verity", &name, &node);
4623e8e6
LP		 2068 if (r < 0)
2069 return r;
2070
89e62e0b 2071 r = sym_crypt_init(&cd, verity->data_path ?: v->node);
4623e8e6
LP		 2072 if (r < 0)
2073 return r;
2074
efc3b12f 2075 cryptsetup_enable_logging(cd);
1887032f 2076
0d12936d 2077 r = sym_crypt_load(cd, CRYPT_VERITY, NULL);
4623e8e6 2078 if (r < 0)
294bd454 2079 return r;
4623e8e6 2080
0d12936d 2081 r = sym_crypt_set_data_device(cd, m->node);
4623e8e6 2082 if (r < 0)
294bd454 2083 return r;
4623e8e6 2084
319a4f4b 2085 if (!GREEDY_REALLOC0(d->decrypted, d->n_decrypted + 1))
ac1f3ad0
LB		 2086 return -ENOMEM;
2087
2088 /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
2089 * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
2090 * retry a few times before giving up. */
2091 for (unsigned i = 0; i < N_DEVICE_NODE_LIST_ATTEMPTS; i++) {
1d369d78 2092 _cleanup_(sym_crypt_freep) struct crypt_device *existing_cd = NULL;
f80015ff 2093 _cleanup_close_ int fd = -1;
c2fa92e7 2094
f80015ff
YW		 2095 /* First, check if the device already exists. */
2096 fd = open(node, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
2097 if (fd < 0 && !ERRNO_IS_DEVICE_ABSENT(errno))
2098 return log_debug_errno(errno, "Failed to open verity device %s: %m", node);
2099 if (fd >= 0)
2100 goto check; /* The device already exists. Let's check it. */
2101
2102 /* The symlink to the device node does not exist yet. Assume not activated, and let's activate it. */
c2fa92e7 2103 r = do_crypt_activate_verity(cd, name, verity);
1d369d78 2104 if (r >= 0)
f7725647 2105 goto try_open; /* The device is activated. Let's open it. */
ac1f3ad0
LB		 2106 /* libdevmapper can return EINVAL when the device is already in the activation stage.
2107 * There's no way to distinguish this situation from a genuine error due to invalid
2aed63f4 2108 * parameters, so immediately fall back to activating the device with a unique name.
89e62e0b
LP		 2109 * Improvements in libcrypsetup can ensure this never happens:
2110 * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
ac1f3ad0 2111 if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
0f75b0c5 2112 break;
1d369d78
YW		 2113 if (r == -ENODEV) /* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */
2114 goto try_again;
2115 if (!IN_SET(r,
2116 -EEXIST, /* Volume has already been opened and ready to be used. */
2117 -EBUSY /* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */))
2118 return log_debug_errno(r, "Failed to activate verity device %s: %m", node);
2119
f80015ff 2120 check:
1d369d78
YW		 2121 if (!restore_deferred_remove){
2122 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
2123 r = dm_deferred_remove_cancel(name);
12f5fbdf
YW		 2124 /* -EBUSY and -ENXIO: the device has already been removed or being removed. We cannot
2125 * use the device, try to open again. See target_message() in drivers/md/dm-ioctl.c
2126 * and dm_cancel_deferred_remove() in drivers/md/dm.c */
2127 if (IN_SET(r, -EBUSY, -ENXIO))
2128 goto try_again;
2129 if (r < 0)
1d369d78 2130 return log_debug_errno(r, "Failed to disable automated deferred removal for verity device %s: %m", node);
12f5fbdf
YW		 2131
2132 restore_deferred_remove = strdup(name);
2133 if (!restore_deferred_remove)
2134 return log_oom_debug();
1d369d78 2135 }
c2923fdc 2136
1d369d78
YW		 2137 r = verity_can_reuse(verity, name, &existing_cd);
2138 /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
2139 if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
2140 break;
2141 if (IN_SET(r,
2142 -ENOENT, /* Removed?? */
2143 -EBUSY, /* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */
2144 -ENODEV /* Volume is being opened but not ready, crypt_init_by_name() would fail, try to open again. */ ))
2145 goto try_again;
2146 if (r < 0)
2147 return log_debug_errno(r, "Failed to check if existing verity device %s can be reused: %m", node);
2148
f80015ff 2149 if (fd < 0) {
1d369d78
YW		 2150 /* devmapper might say that the device exists, but the devlink might not yet have been
2151 * created. Check and wait for the udev event in that case. */
2152 r = device_wait_for_devlink(node, "block", verity_timeout(), NULL);
2153 /* Fallback to activation with a unique device if it's taking too long */
2154 if (r == -ETIMEDOUT && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
0f75b0c5 2155 break;
1d369d78
YW		 2156 if (r < 0)
2157 return log_debug_errno(r, "Failed to wait device node symlink %s: %m", node);
c2923fdc 2158 }
ecab4c47 2159
f7725647
YW		 2160 try_open:
2161 if (fd < 0) {
2162 /* Now, the device is activated and devlink is created. Let's open it. */
2163 fd = open(node, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
2164 if (fd < 0) {
2165 if (!ERRNO_IS_DEVICE_ABSENT(errno))
2166 return log_debug_errno(errno, "Failed to open verity device %s: %m", node);
2167
2168 /* The device has already been removed?? */
2169 goto try_again;
2170 }
2171 }
2172
2173 mount_node_fd = TAKE_FD(fd);
1d369d78
YW		 2174 if (existing_cd)
2175 crypt_free_and_replace(cd, existing_cd);
2176
2177 goto success;
2178
2179 try_again:
2180 /* Device is being removed by another process. Let's wait for a while. */
ecab4c47 2181 (void) usleep(2 * USEC_PER_MSEC);
ac1f3ad0
LB		 2182 }
2183
0f75b0c5
YW		 2184 /* All trials failed or a conflicting verity device exists. Let's try to activate with a unique name. */
2185 if (FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) {
2186 /* Before trying to activate with unique name, we need to free crypt_device object.
2187 * Otherwise, we get error from libcryptsetup like the following:
2188 * ------
2189 * systemd[1234]: Cannot use device /dev/loop5 which is in use (already mapped or mounted).
2190 * ------
2191 */
2192 sym_crypt_free(cd);
2193 cd = NULL;
aee36b4e 2194 return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
0f75b0c5 2195 }
ac1f3ad0 2196
9972e6d6
YW		 2197 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY), "All attempts to activate verity device %s failed.", name);
2198
2199success:
ac1f3ad0
LB		 2200 /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
2201 restore_deferred_remove = mfree(restore_deferred_remove);
4623e8e6 2202
94344385
LP		 2203 d->decrypted[d->n_decrypted++] = (DecryptedPartition) {
2204 .name = TAKE_PTR(name),
2205 .device = TAKE_PTR(cd),
2206 };
4623e8e6 2207
1cc6c93a 2208 m->decrypted_node = TAKE_PTR(node);
f7725647 2209 close_and_replace(m->mount_node_fd, mount_node_fd);
4623e8e6
LP		 2210
2211 return 0;
18b5886e
LP		 2212}
2213#endif
2214
2215int dissected_image_decrypt(
2216 DissectedImage *m,
2217 const char *passphrase,
89e62e0b 2218 const VeritySettings *verity,
e330f97a 2219 DissectImageFlags flags) {
18b5886e 2220
349cc4a5 2221#if HAVE_LIBCRYPTSETUP
49b5b3b4 2222 _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL;
18b5886e
LP		 2223 int r;
2224#endif
2225
2226 assert(m);
89e62e0b 2227 assert(!verity || verity->root_hash || verity->root_hash_size == 0);
18b5886e
LP		 2228
2229 /* Returns:
2230 *
2231 * = 0 → There was nothing to decrypt
2232 * > 0 → Decrypted successfully
d1c536f5 2233 * -ENOKEY → There's something to decrypt but no key was supplied
18b5886e
LP		 2234 * -EKEYREJECTED → Passed key was not correct
2235 */
2236
89e62e0b 2237 if (verity && verity->root_hash && verity->root_hash_size < sizeof(sd_id128_t))
4623e8e6
LP		 2238 return -EINVAL;
2239
e330f97a 2240 if (!m->encrypted && !m->verity_ready)
18b5886e 2241 return 0;
18b5886e 2242
349cc4a5 2243#if HAVE_LIBCRYPTSETUP
9321ad51
YW		 2244 r = decrypted_image_new(&d);
2245 if (r < 0)
2246 return r;
18b5886e 2247
569a0e42 2248 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
18b5886e 2249 DissectedPartition *p = m->partitions + i;
22043172 2250 PartitionDesignator k;
18b5886e
LP		 2251
2252 if (!p->found)
2253 continue;
2254
2255 r = decrypt_partition(p, passphrase, flags, d);
2256 if (r < 0)
2257 return r;
2258
dd894023 2259 k = partition_verity_of(i);
4623e8e6 2260 if (k >= 0) {
aee36b4e 2261 r = verity_partition(i, p, m->partitions + k, verity, flags | DISSECT_IMAGE_VERITY_SHARE, d);
4623e8e6
LP		 2262 if (r < 0)
2263 return r;
2264 }
2265
d2c6e79d 2266 if (!p->decrypted_fstype && p->mount_node_fd >= 0 && p->decrypted_node) {
c80c9079 2267 r = probe_filesystem_full(p->mount_node_fd, p->decrypted_node, 0, UINT64_MAX, &p->decrypted_fstype);
7cc84b2c 2268 if (r < 0 && r != -EUCLEAN)
18b5886e
LP		 2269 return r;
2270 }
2271 }
2272
ac1e1b5f 2273 m->decrypted_image = TAKE_PTR(d);
18b5886e
LP		 2274
2275 return 1;
2276#else
2277 return -EOPNOTSUPP;
2278#endif
2279}
2280
2281int dissected_image_decrypt_interactively(
2282 DissectedImage *m,
2283 const char *passphrase,
89e62e0b 2284 const VeritySettings *verity,
e330f97a 2285 DissectImageFlags flags) {
18b5886e
LP		 2286
2287 _cleanup_strv_free_erase_ char **z = NULL;
2288 int n = 3, r;
2289
2290 if (passphrase)
2291 n--;
2292
2293 for (;;) {
e330f97a 2294 r = dissected_image_decrypt(m, passphrase, verity, flags);
18b5886e
LP		 2295 if (r >= 0)
2296 return r;
2297 if (r == -EKEYREJECTED)
2298 log_error_errno(r, "Incorrect passphrase, try again!");
fc95c359
YW		 2299 else if (r != -ENOKEY)
2300 return log_error_errno(r, "Failed to decrypt image: %m");
18b5886e 2301
baaa35ad
ZJS		 2302 if (--n < 0)
2303 return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED),
2304 "Too many retries.");
18b5886e
LP		 2305
2306 z = strv_free(z);
2307
8806bb4b 2308 r = ask_password_auto("Please enter image passphrase:", NULL, "dissect", "dissect", "dissect.passphrase", USEC_INFINITY, 0, &z);
18b5886e
LP		 2309 if (r < 0)
2310 return log_error_errno(r, "Failed to query for passphrase: %m");
2311
2312 passphrase = z[0];
2313 }
2314}
2315
e330f97a 2316static int decrypted_image_relinquish(DecryptedImage *d) {
18b5886e
LP		 2317 assert(d);
2318
67f63ee5
ZJS		 2319 /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a
2320 * boolean so that we don't clean it up ourselves either anymore */
18b5886e 2321
349cc4a5 2322#if HAVE_LIBCRYPTSETUP
67f63ee5
ZJS		 2323 int r;
2324
2325 for (size_t i = 0; i < d->n_decrypted; i++) {
18b5886e
LP		 2326 DecryptedPartition *p = d->decrypted + i;
2327
2328 if (p->relinquished)
2329 continue;
2330
0d12936d 2331 r = sym_crypt_deactivate_by_name(NULL, p->name, CRYPT_DEACTIVATE_DEFERRED);
18b5886e
LP		 2332 if (r < 0)
2333 return log_debug_errno(r, "Failed to mark %s for auto-removal: %m", p->name);
2334
2335 p->relinquished = true;
2336 }
2337#endif
2338
2339 return 0;
2340}
2341
f906075a
YW		 2342int dissected_image_relinquish(DissectedImage *m) {
2343 int r;
2344
2345 assert(m);
2346
2347 if (m->decrypted_image) {
2348 r = decrypted_image_relinquish(m->decrypted_image);
2349 if (r < 0)
2350 return r;
2351 }
2352
2353 if (m->loop)
2354 loop_device_relinquish(m->loop);
2355
2356 return 0;
2357}
2358
89e62e0b
LP		 2359static char *build_auxiliary_path(const char *image, const char *suffix) {
2360 const char *e;
2361 char *n;
2362
2363 assert(image);
2364 assert(suffix);
2365
2366 e = endswith(image, ".raw");
2367 if (!e)
2368 return strjoin(e, suffix);
2369
2370 n = new(char, e - image + strlen(suffix) + 1);
2371 if (!n)
2372 return NULL;
2373
2374 strcpy(mempcpy(n, image, e - image), suffix);
2375 return n;
2376}
2377
2378void verity_settings_done(VeritySettings *v) {
2379 assert(v);
2380
2381 v->root_hash = mfree(v->root_hash);
2382 v->root_hash_size = 0;
2383
2384 v->root_hash_sig = mfree(v->root_hash_sig);
2385 v->root_hash_sig_size = 0;
2386
2387 v->data_path = mfree(v->data_path);
2388}
2389
2390int verity_settings_load(
2391 VeritySettings *verity,
f5ea63a5
LP		 2392 const char *image,
2393 const char *root_hash_path,
89e62e0b
LP		 2394 const char *root_hash_sig_path) {
2395
2396 _cleanup_free_ void *root_hash = NULL, *root_hash_sig = NULL;
2397 size_t root_hash_size = 0, root_hash_sig_size = 0;
2398 _cleanup_free_ char *verity_data_path = NULL;
aee36b4e 2399 PartitionDesignator designator;
78ebe980
LP		 2400 int r;
2401
89e62e0b 2402 assert(verity);
78ebe980 2403 assert(image);
aee36b4e 2404 assert(verity->designator < 0 || IN_SET(verity->designator, PARTITION_ROOT, PARTITION_USR));
78ebe980 2405
89e62e0b
LP		 2406 /* If we are asked to load the root hash for a device node, exit early */
2407 if (is_device_path(image))
78ebe980 2408 return 0;
78ebe980 2409
d5fcc5b0
LP		 2410 r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIDECAR");
2411 if (r < 0 && r != -ENXIO)
2412 log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIDECAR, ignoring: %m");
2413 if (r == 0)
2414 return 0;
2415
aee36b4e
LP		 2416 designator = verity->designator;
2417
89e62e0b 2418 /* We only fill in what isn't already filled in */
c2923fdc 2419
89e62e0b 2420 if (!verity->root_hash) {
e7cbe5cb 2421 _cleanup_free_ char *text = NULL;
e7cbe5cb 2422
0389f4fa 2423 if (root_hash_path) {
aee36b4e 2424 /* If explicitly specified it takes precedence */
0389f4fa
LB		 2425 r = read_one_line_file(root_hash_path, &text);
2426 if (r < 0)
e7cbe5cb 2427 return r;
aee36b4e
LP		 2428
2429 if (designator < 0)
2430 designator = PARTITION_ROOT;
0389f4fa 2431 } else {
aee36b4e
LP		 2432 /* Otherwise look for xattr and separate file, and first for the data for root and if
2433 * that doesn't exist for /usr */
0389f4fa 2434
aee36b4e 2435 if (designator < 0 || designator == PARTITION_ROOT) {
c53e07e2 2436 r = getxattr_malloc(image, "user.verity.roothash", &text);
aee36b4e
LP		 2437 if (r < 0) {
2438 _cleanup_free_ char *p = NULL;
78ebe980 2439
00675c36 2440 if (r != -ENOENT && !ERRNO_IS_XATTR_ABSENT(r))
aee36b4e 2441 return r;
e7cbe5cb 2442
aee36b4e
LP		 2443 p = build_auxiliary_path(image, ".roothash");
2444 if (!p)
2445 return -ENOMEM;
2446
2447 r = read_one_line_file(p, &text);
2448 if (r < 0 && r != -ENOENT)
2449 return r;
2450 }
2451
2452 if (text)
2453 designator = PARTITION_ROOT;
2454 }
2455
2456 if (!text && (designator < 0 || designator == PARTITION_USR)) {
2457 /* So in the "roothash" xattr/file name above the "root" of course primarily
2458 * refers to the root of the Verity Merkle tree. But coincidentally it also
2459 * is the hash for the *root* file system, i.e. the "root" neatly refers to
2460 * two distinct concepts called "root". Taking benefit of this happy
2461 * coincidence we call the file with the root hash for the /usr/ file system
2462 * `usrhash`, because `usrroothash` or `rootusrhash` would just be too
2463 * confusing. We thus drop the reference to the root of the Merkle tree, and
2464 * just indicate which file system it's about. */
c53e07e2 2465 r = getxattr_malloc(image, "user.verity.usrhash", &text);
aee36b4e
LP		 2466 if (r < 0) {
2467 _cleanup_free_ char *p = NULL;
2468
00675c36 2469 if (r != -ENOENT && !ERRNO_IS_XATTR_ABSENT(r))
aee36b4e
LP		 2470 return r;
2471
2472 p = build_auxiliary_path(image, ".usrhash");
2473 if (!p)
2474 return -ENOMEM;
2475
2476 r = read_one_line_file(p, &text);
2477 if (r < 0 && r != -ENOENT)
2478 return r;
2479 }
2480
2481 if (text)
2482 designator = PARTITION_USR;
0389f4fa 2483 }
e7cbe5cb
LB		 2484 }
2485
2486 if (text) {
89e62e0b 2487 r = unhexmem(text, strlen(text), &root_hash, &root_hash_size);
e7cbe5cb
LB		 2488 if (r < 0)
2489 return r;
89e62e0b 2490 if (root_hash_size < sizeof(sd_id128_t))
e7cbe5cb
LB		 2491 return -EINVAL;
2492 }
2493 }
2494
90f98986 2495 if ((root_hash || verity->root_hash) && !verity->root_hash_sig) {
aee36b4e 2496 if (root_hash_sig_path) {
ae9cf30b 2497 r = read_full_file(root_hash_sig_path, (char**) &root_hash_sig, &root_hash_sig_size);
aee36b4e
LP		 2498 if (r < 0 && r != -ENOENT)
2499 return r;
2500
2501 if (designator < 0)
2502 designator = PARTITION_ROOT;
2503 } else {
2504 if (designator < 0 || designator == PARTITION_ROOT) {
2505 _cleanup_free_ char *p = NULL;
2506
2507 /* Follow naming convention recommended by the relevant RFC:
2508 * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
2509 p = build_auxiliary_path(image, ".roothash.p7s");
2510 if (!p)
2511 return -ENOMEM;
89e62e0b 2512
ae9cf30b 2513 r = read_full_file(p, (char**) &root_hash_sig, &root_hash_sig_size);
aee36b4e
LP		 2514 if (r < 0 && r != -ENOENT)
2515 return r;
2516 if (r >= 0)
2517 designator = PARTITION_ROOT;
2518 }
2519
2520 if (!root_hash_sig && (designator < 0 || designator == PARTITION_USR)) {
2521 _cleanup_free_ char *p = NULL;
2522
2523 p = build_auxiliary_path(image, ".usrhash.p7s");
2524 if (!p)
2525 return -ENOMEM;
89e62e0b 2526
ae9cf30b 2527 r = read_full_file(p, (char**) &root_hash_sig, &root_hash_sig_size);
aee36b4e
LP		 2528 if (r < 0 && r != -ENOENT)
2529 return r;
2530 if (r >= 0)
2531 designator = PARTITION_USR;
2532 }
89e62e0b
LP		 2533 }
2534
aee36b4e 2535 if (root_hash_sig && root_hash_sig_size == 0) /* refuse empty size signatures */
89e62e0b
LP		 2536 return -EINVAL;
2537 }
2538
2539 if (!verity->data_path) {
2540 _cleanup_free_ char *p = NULL;
2541
2542 p = build_auxiliary_path(image, ".verity");
2543 if (!p)
2544 return -ENOMEM;
2545
2546 if (access(p, F_OK) < 0) {
2547 if (errno != ENOENT)
2548 return -errno;
2549 } else
2550 verity_data_path = TAKE_PTR(p);
2551 }
2552
2553 if (root_hash) {
2554 verity->root_hash = TAKE_PTR(root_hash);
2555 verity->root_hash_size = root_hash_size;
2556 }
2557
2558 if (root_hash_sig) {
2559 verity->root_hash_sig = TAKE_PTR(root_hash_sig);
2560 verity->root_hash_sig_size = root_hash_sig_size;
e7cbe5cb 2561 }
89e62e0b
LP		 2562
2563 if (verity_data_path)
2564 verity->data_path = TAKE_PTR(verity_data_path);
78ebe980 2565
aee36b4e
LP		 2566 if (verity->designator < 0)
2567 verity->designator = designator;
2568
78ebe980
LP		 2569 return 1;
2570}
2571
88b3300f
LP		 2572int dissected_image_load_verity_sig_partition(
2573 DissectedImage *m,
2574 int fd,
2575 VeritySettings *verity) {
2576
2577 _cleanup_free_ void *root_hash = NULL, *root_hash_sig = NULL;
2578 _cleanup_(json_variant_unrefp) JsonVariant *v = NULL;
2579 size_t root_hash_size, root_hash_sig_size;
2580 _cleanup_free_ char *buf = NULL;
2581 PartitionDesignator d;
2582 DissectedPartition *p;
2583 JsonVariant *rh, *sig;
2584 ssize_t n;
2585 char *e;
2586 int r;
2587
2588 assert(m);
2589 assert(fd >= 0);
2590 assert(verity);
2591
2592 if (verity->root_hash && verity->root_hash_sig) /* Already loaded? */
2593 return 0;
2594
2595 r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_EMBEDDED");
2596 if (r < 0 && r != -ENXIO)
2597 log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_EMBEDDED, ignoring: %m");
2598 if (r == 0)
2599 return 0;
2600
dd894023 2601 d = partition_verity_sig_of(verity->designator < 0 ? PARTITION_ROOT : verity->designator);
88b3300f
LP		 2602 assert(d >= 0);
2603
2604 p = m->partitions + d;
2605 if (!p->found)
2606 return 0;
2607 if (p->offset == UINT64_MAX || p->size == UINT64_MAX)
2608 return -EINVAL;
2609
2610 if (p->size > 4*1024*1024) /* Signature data cannot possible be larger than 4M, refuse that */
2611 return -EFBIG;
2612
2613 buf = new(char, p->size+1);
2614 if (!buf)
2615 return -ENOMEM;
2616
2617 n = pread(fd, buf, p->size, p->offset);
2618 if (n < 0)
2619 return -ENOMEM;
2620 if ((uint64_t) n != p->size)
2621 return -EIO;
2622
2623 e = memchr(buf, 0, p->size);
2624 if (e) {
2625 /* If we found a NUL byte then the rest of the data must be NUL too */
2626 if (!memeqzero(e, p->size - (e - buf)))
2627 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature data contains embedded NUL byte.");
2628 } else
2629 buf[p->size] = 0;
2630
2631 r = json_parse(buf, 0, &v, NULL, NULL);
2632 if (r < 0)
2633 return log_debug_errno(r, "Failed to parse signature JSON data: %m");
2634
2635 rh = json_variant_by_key(v, "rootHash");
2636 if (!rh)
2637 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature JSON object lacks 'rootHash' field.");
2638 if (!json_variant_is_string(rh))
2639 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "'rootHash' field of signature JSON object is not a string.");
2640
2641 r = unhexmem(json_variant_string(rh), SIZE_MAX, &root_hash, &root_hash_size);
2642 if (r < 0)
2643 return log_debug_errno(r, "Failed to parse root hash field: %m");
2644
2645 /* Check if specified root hash matches if it is specified */
2646 if (verity->root_hash &&
2647 memcmp_nn(verity->root_hash, verity->root_hash_size, root_hash, root_hash_size) != 0) {
2648 _cleanup_free_ char *a = NULL, *b = NULL;
2649
2650 a = hexmem(root_hash, root_hash_size);
2651 b = hexmem(verity->root_hash, verity->root_hash_size);
2652
2653 return log_debug_errno(r, "Root hash in signature JSON data (%s) doesn't match configured hash (%s).", strna(a), strna(b));
2654 }
2655
2656 sig = json_variant_by_key(v, "signature");
2657 if (!sig)
2658 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature JSON object lacks 'signature' field.");
2659 if (!json_variant_is_string(sig))
2660 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "'signature' field of signature JSON object is not a string.");
2661
2662 r = unbase64mem(json_variant_string(sig), SIZE_MAX, &root_hash_sig, &root_hash_sig_size);
2663 if (r < 0)
2664 return log_debug_errno(r, "Failed to parse signature field: %m");
2665
2666 free_and_replace(verity->root_hash, root_hash);
2667 verity->root_hash_size = root_hash_size;
2668
2669 free_and_replace(verity->root_hash_sig, root_hash_sig);
2670 verity->root_hash_sig_size = root_hash_sig_size;
2671
2672 return 1;
2673}
2674
22847508 2675int dissected_image_acquire_metadata(DissectedImage *m, DissectImageFlags extra_flags) {
3b925504
LP		 2676
2677 enum {
2678 META_HOSTNAME,
2679 META_MACHINE_ID,
2680 META_MACHINE_INFO,
2681 META_OS_RELEASE,
fab22946 2682 META_INITRD_RELEASE,
7718ac97 2683 META_EXTENSION_RELEASE,
a4e0d617 2684 META_HAS_INIT_SYSTEM,
3b925504
LP		 2685 _META_MAX,
2686 };
2687
9a4b883b 2688 static const char *const paths[_META_MAX] = {
7718ac97
LB		 2689 [META_HOSTNAME] = "/etc/hostname\0",
2690 [META_MACHINE_ID] = "/etc/machine-id\0",
2691 [META_MACHINE_INFO] = "/etc/machine-info\0",
9a4b883b 2692 [META_OS_RELEASE] = ("/etc/os-release\0"
22847508 2693 "/usr/lib/os-release\0"),
fab22946
LP		 2694 [META_INITRD_RELEASE] = ("/etc/initrd-release\0"
2695 "/usr/lib/initrd-release\0"),
a4e0d617
LP		 2696 [META_EXTENSION_RELEASE] = "extension-release\0", /* Used only for logging. */
2697 [META_HAS_INIT_SYSTEM] = "has-init-system\0", /* ditto */
3b925504
LP		 2698 };
2699
fab22946 2700 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL, **initrd_release = NULL, **extension_release = NULL;
af8219d5 2701 _cleanup_close_pair_ int error_pipe[2] = { -1, -1 };
3b925504
LP		 2702 _cleanup_(rmdir_and_freep) char *t = NULL;
2703 _cleanup_(sigkill_waitp) pid_t child = 0;
2704 sd_id128_t machine_id = SD_ID128_NULL;
2705 _cleanup_free_ char *hostname = NULL;
67f63ee5 2706 unsigned n_meta_initialized = 0;
af8219d5 2707 int fds[2 * _META_MAX], r, v;
a4e0d617 2708 int has_init_system = -1;
af8219d5 2709 ssize_t n;
3b925504
LP		 2710
2711 BLOCK_SIGNALS(SIGCHLD);
2712
2713 assert(m);
2714
7718ac97 2715 for (; n_meta_initialized < _META_MAX; n_meta_initialized ++) {
d9119c00
LP		 2716 if (!paths[n_meta_initialized]) {
2717 fds[2*n_meta_initialized] = fds[2*n_meta_initialized+1] = -1;
7718ac97 2718 continue;
d9119c00
LP		 2719 }
2720
3b925504
LP		 2721 if (pipe2(fds + 2*n_meta_initialized, O_CLOEXEC) < 0) {
2722 r = -errno;
2723 goto finish;
2724 }
7718ac97 2725 }
3b925504
LP		 2726
2727 r = mkdtemp_malloc("/tmp/dissect-XXXXXX", &t);
2728 if (r < 0)
2729 goto finish;
2730
af8219d5
LP		 2731 if (pipe2(error_pipe, O_CLOEXEC) < 0) {
2732 r = -errno;
2733 goto finish;
2734 }
2735
e2047ba9 2736 r = safe_fork("(sd-dissect)", FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE, &child);
be39f6ee 2737 if (r < 0)
3b925504 2738 goto finish;
be39f6ee 2739 if (r == 0) {
a4e0d617 2740 /* Child in a new mount namespace */
af8219d5
LP		 2741 error_pipe[0] = safe_close(error_pipe[0]);
2742
7cf66030
LP		 2743 r = dissected_image_mount(
2744 m,
2745 t,
2746 UID_INVALID,
21b61b1d 2747 UID_INVALID,
22847508
ZJS		 2748 extra_flags |
2749 DISSECT_IMAGE_READ_ONLY |
2750 DISSECT_IMAGE_MOUNT_ROOT_ONLY |
7cf66030 2751 DISSECT_IMAGE_USR_NO_ROOT);
429d4e41
LP		 2752 if (r < 0) {
2753 log_debug_errno(r, "Failed to mount dissected image: %m");
03ae68f4 2754 goto inner_fail;
429d4e41 2755 }
3b925504 2756
67f63ee5 2757 for (unsigned k = 0; k < _META_MAX; k++) {
37e44c3f 2758 _cleanup_close_ int fd = -ENOENT;
3b925504 2759
7718ac97
LB		 2760 if (!paths[k])
2761 continue;
2762
3b925504
LP		 2763 fds[2*k] = safe_close(fds[2*k]);
2764
a4e0d617
LP		 2765 switch (k) {
2766
2767 case META_EXTENSION_RELEASE:
9a4b883b
LB		 2768 /* As per the os-release spec, if the image is an extension it will have a file
2769 * named after the image name in extension-release.d/ - we use the image name
2770 * and try to resolve it with the extension-release helpers, as sometimes
2771 * the image names are mangled on deployment and do not match anymore.
2772 * Unlike other paths this is not fixed, and the image name
2773 * can be mangled on deployment, so by calling into the helper
2774 * we allow a fallback that matches on the first extension-release
2775 * file found in the directory, if one named after the image cannot
2776 * be found first. */
06768b90 2777 r = open_extension_release(t, m->image_name, /* relax_extension_release_check= */ false, NULL, &fd);
9a4b883b
LB		 2778 if (r < 0)
2779 fd = r; /* Propagate the error. */
a4e0d617
LP		 2780 break;
2781
2782 case META_HAS_INIT_SYSTEM: {
2783 bool found = false;
a4e0d617
LP		 2784
2785 FOREACH_STRING(init,
2786 "/usr/lib/systemd/systemd", /* systemd on /usr merged system */
2787 "/lib/systemd/systemd", /* systemd on /usr non-merged systems */
2788 "/sbin/init") { /* traditional path the Linux kernel invokes */
2789
2790 r = chase_symlinks(init, t, CHASE_PREFIX_ROOT, NULL, NULL);
2791 if (r < 0) {
2792 if (r != -ENOENT)
2793 log_debug_errno(r, "Failed to resolve %s, ignoring: %m", init);
2794 } else {
2795 found = true;
2796 break;
2797 }
2798 }
2799
2800 r = loop_write(fds[2*k+1], &found, sizeof(found), false);
2801 if (r < 0)
2802 goto inner_fail;
2803
2804 continue;
2805 }
2806
2807 default:
9a4b883b
LB		 2808 NULSTR_FOREACH(p, paths[k]) {
2809 fd = chase_symlinks_and_open(p, t, CHASE_PREFIX_ROOT, O_RDONLY|O_CLOEXEC|O_NOCTTY, NULL);
2810 if (fd >= 0)
2811 break;
2812 }
a4e0d617
LP		 2813 }
2814
36952d19
LP		 2815 if (fd < 0) {
2816 log_debug_errno(fd, "Failed to read %s file of image, ignoring: %m", paths[k]);
37e44c3f 2817 fds[2*k+1] = safe_close(fds[2*k+1]);
3b925504 2818 continue;
36952d19 2819 }
3b925504 2820
f5fbe71d 2821 r = copy_bytes(fd, fds[2*k+1], UINT64_MAX, 0);
a4e0d617
LP		 2822 if (r < 0)
2823 goto inner_fail;
3b925504
LP		 2824
2825 fds[2*k+1] = safe_close(fds[2*k+1]);
2826 }
2827
2828 _exit(EXIT_SUCCESS);
a4e0d617
LP		 2829
2830 inner_fail:
03ae68f4 2831 /* Let parent know the error */
a4e0d617
LP		 2832 (void) write(error_pipe[1], &r, sizeof(r));
2833 _exit(EXIT_FAILURE);
3b925504
LP		 2834 }
2835
af8219d5
LP		 2836 error_pipe[1] = safe_close(error_pipe[1]);
2837
67f63ee5 2838 for (unsigned k = 0; k < _META_MAX; k++) {
3b925504
LP		 2839 _cleanup_fclose_ FILE *f = NULL;
2840
7718ac97
LB		 2841 if (!paths[k])
2842 continue;
2843
3b925504
LP		 2844 fds[2*k+1] = safe_close(fds[2*k+1]);
2845
4fa744a3 2846 f = take_fdopen(&fds[2*k], "r");
3b925504
LP		 2847 if (!f) {
2848 r = -errno;
2849 goto finish;
2850 }
2851
3b925504
LP		 2852 switch (k) {
2853
2854 case META_HOSTNAME:
2855 r = read_etc_hostname_stream(f, &hostname);
2856 if (r < 0)
f6048e5e 2857 log_debug_errno(r, "Failed to read /etc/hostname of image: %m");
3b925504
LP		 2858
2859 break;
2860
2861 case META_MACHINE_ID: {
2862 _cleanup_free_ char *line = NULL;
2863
2864 r = read_line(f, LONG_LINE_MAX, &line);
2865 if (r < 0)
f6048e5e 2866 log_debug_errno(r, "Failed to read /etc/machine-id of image: %m");
3b925504
LP		 2867 else if (r == 33) {
2868 r = sd_id128_from_string(line, &machine_id);
2869 if (r < 0)
2870 log_debug_errno(r, "Image contains invalid /etc/machine-id: %s", line);
2871 } else if (r == 0)
f6048e5e 2872 log_debug("/etc/machine-id file of image is empty.");
ab763cb2 2873 else if (streq(line, "uninitialized"))
f6048e5e 2874 log_debug("/etc/machine-id file of image is uninitialized (likely aborted first boot).");
3b925504 2875 else
f6048e5e 2876 log_debug("/etc/machine-id file of image has unexpected length %i.", r);
3b925504
LP		 2877
2878 break;
2879 }
2880
2881 case META_MACHINE_INFO:
aa8fbc74 2882 r = load_env_file_pairs(f, "machine-info", &machine_info);
3b925504 2883 if (r < 0)
f6048e5e 2884 log_debug_errno(r, "Failed to read /etc/machine-info of image: %m");
3b925504
LP		 2885
2886 break;
2887
2888 case META_OS_RELEASE:
aa8fbc74 2889 r = load_env_file_pairs(f, "os-release", &os_release);
3b925504 2890 if (r < 0)
f6048e5e 2891 log_debug_errno(r, "Failed to read OS release file of image: %m");
3b925504
LP		 2892
2893 break;
7718ac97 2894
fab22946
LP		 2895 case META_INITRD_RELEASE:
2896 r = load_env_file_pairs(f, "initrd-release", &initrd_release);
2897 if (r < 0)
2898 log_debug_errno(r, "Failed to read initrd release file of image: %m");
2899
2900 break;
2901
7718ac97
LB		 2902 case META_EXTENSION_RELEASE:
2903 r = load_env_file_pairs(f, "extension-release", &extension_release);
2904 if (r < 0)
f6048e5e 2905 log_debug_errno(r, "Failed to read extension release file of image: %m");
7718ac97
LB		 2906
2907 break;
a4e0d617
LP		 2908
2909 case META_HAS_INIT_SYSTEM: {
2910 bool b = false;
2911 size_t nr;
2912
2913 errno = 0;
2914 nr = fread(&b, 1, sizeof(b), f);
2915 if (nr != sizeof(b))
2916 log_debug_errno(errno_or_else(EIO), "Failed to read has-init-system boolean: %m");
2917 else
2918 has_init_system = b;
2919
2920 break;
2921 }}
3b925504
LP		 2922 }
2923
2e87a1fd 2924 r = wait_for_terminate_and_check("(sd-dissect)", child, 0);
3b925504 2925 child = 0;
2e87a1fd 2926 if (r < 0)
af8219d5
LP		 2927 return r;
2928
2929 n = read(error_pipe[0], &v, sizeof(v));
2930 if (n < 0)
2931 return -errno;
2932 if (n == sizeof(v))
2933 return v; /* propagate error sent to us from child */
2934 if (n != 0)
2935 return -EIO;
2936
2e87a1fd
LP		 2937 if (r != EXIT_SUCCESS)
2938 return -EPROTO;
3b925504
LP		 2939
2940 free_and_replace(m->hostname, hostname);
2941 m->machine_id = machine_id;
2942 strv_free_and_replace(m->machine_info, machine_info);
2943 strv_free_and_replace(m->os_release, os_release);
fab22946 2944 strv_free_and_replace(m->initrd_release, initrd_release);
7718ac97 2945 strv_free_and_replace(m->extension_release, extension_release);
a4e0d617 2946 m->has_init_system = has_init_system;
3b925504
LP		 2947
2948finish:
67f63ee5 2949 for (unsigned k = 0; k < n_meta_initialized; k++)
3b925504
LP		 2950 safe_close_pair(fds + 2*k);
2951
2952 return r;
2953}
2954
1e63dc4f
YW		 2955int dissect_loop_device(
2956 LoopDevice *loop,
2957 const VeritySettings *verity,
2958 const MountOptions *mount_options,
2959 DissectImageFlags flags,
2960 DissectedImage **ret) {
2961
08f14be4 2962#if HAVE_BLKID
1e63dc4f
YW		 2963 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
2964 int r;
2965
2966 assert(loop);
2967 assert(ret);
2968
08f14be4 2969 r = dissected_image_new(loop->backing_file ?: loop->node, &m);
1e63dc4f
YW		 2970 if (r < 0)
2971 return r;
2972
2973 m->loop = loop_device_ref(loop);
2974
73d88b80 2975 r = dissect_image(m, loop->fd, loop->node, verity, mount_options, flags);
08f14be4 2976 if (r < 0)
08f14be4
YW		 2977 return r;
2978
1e63dc4f
YW		 2979 *ret = TAKE_PTR(m);
2980 return 0;
08f14be4
YW		 2981#else
2982 return -EOPNOTSUPP;
2983#endif
1e63dc4f
YW		 2984}
2985
bad31660 2986int dissect_loop_device_and_warn(
1e63dc4f 2987 LoopDevice *loop,
89e62e0b 2988 const VeritySettings *verity,
18d73705 2989 const MountOptions *mount_options,
4526113f
LP		 2990 DissectImageFlags flags,
2991 DissectedImage **ret) {
2992
64dd3a24 2993 const char *name;
4526113f
LP		 2994 int r;
2995
bad31660
YW		 2996 assert(loop);
2997 assert(loop->fd >= 0);
4526113f 2998
64dd3a24 2999 name = ASSERT_PTR(loop->backing_file ?: loop->node);
4526113f 3000
369de26f 3001 r = dissect_loop_device(loop, verity, mount_options, flags, ret);
4526113f
LP		 3002 switch (r) {
3003
3004 case -EOPNOTSUPP:
3005 return log_error_errno(r, "Dissecting images is not supported, compiled without blkid support.");
3006
3007 case -ENOPKG:
48084df6
ZJS		 3008 return log_error_errno(r, "%s: Couldn't identify a suitable partition table or file system.", name);
3009
3010 case -ENOMEDIUM:
3011 return log_error_errno(r, "%s: The image does not pass validation.", name);
4526113f
LP		 3012
3013 case -EADDRNOTAVAIL:
48084df6 3014 return log_error_errno(r, "%s: No root partition for specified root hash found.", name);
4526113f
LP		 3015
3016 case -ENOTUNIQ:
48084df6 3017 return log_error_errno(r, "%s: Multiple suitable root partitions found in image.", name);
4526113f
LP		 3018
3019 case -ENXIO:
48084df6 3020 return log_error_errno(r, "%s: No suitable root partition found in image.", name);
4526113f
LP		 3021
3022 case -EPROTONOSUPPORT:
3023 return log_error_errno(r, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name);
3024
48084df6
ZJS		 3025 case -ENOTBLK:
3026 return log_error_errno(r, "%s: Image is not a block device.", name);
3027
a94aa2b9
LP		 3028 case -EBADR:
3029 return log_error_errno(r,
3030 "Combining partitioned images (such as '%s') with external Verity data (such as '%s') not supported. "
3031 "(Consider setting $SYSTEMD_DISSECT_VERITY_SIDECAR=0 to disable automatic discovery of external Verity data.)",
3032 name, strna(verity ? verity->data_path : NULL));
3033
4526113f
LP		 3034 default:
3035 if (r < 0)
3036 return log_error_errno(r, "Failed to dissect image '%s': %m", name);
3037
3038 return r;
3039 }
3040}
3041
49536766
LP		 3042bool dissected_image_verity_candidate(const DissectedImage *image, PartitionDesignator partition_designator) {
3043 assert(image);
3044
3045 /* Checks if this partition could theoretically do Verity. For non-partitioned images this only works
3046 * if there's an external verity file supplied, for which we can consult .has_verity. For partitioned
3047 * images we only check the partition type.
3048 *
3049 * This call is used to decide whether to suppress or show a verity column in tabular output of the
3050 * image. */
3051
e7cbe5cb 3052 if (image->single_file_system)
c3c88d67 3053 return partition_designator == PARTITION_ROOT && image->has_verity;
e7cbe5cb 3054
dd894023 3055 return partition_verity_of(partition_designator) >= 0;
e7cbe5cb
LB		 3056}
3057
49536766
LP		 3058bool dissected_image_verity_ready(const DissectedImage *image, PartitionDesignator partition_designator) {
3059 PartitionDesignator k;
3060
3061 assert(image);
3062
3063 /* Checks if this partition has verity data available that we can activate. For non-partitioned this
3064 * works for the root partition, for others only if the associated verity partition was found. */
3065
3066 if (!image->verity_ready)
3067 return false;
e7cbe5cb
LB		 3068
3069 if (image->single_file_system)
49536766 3070 return partition_designator == PARTITION_ROOT;
e7cbe5cb 3071
dd894023 3072 k = partition_verity_of(partition_designator);
e7cbe5cb
LB		 3073 return k >= 0 && image->partitions[k].found;
3074}
3075
8ee9615e
LP		 3076bool dissected_image_verity_sig_ready(const DissectedImage *image, PartitionDesignator partition_designator) {
3077 PartitionDesignator k;
3078
3079 assert(image);
3080
3081 /* Checks if this partition has verity signature data available that we can use. */
3082
3083 if (!image->verity_sig_ready)
3084 return false;
3085
3086 if (image->single_file_system)
3087 return partition_designator == PARTITION_ROOT;
3088
dd894023 3089 k = partition_verity_sig_of(partition_designator);
8ee9615e
LP		 3090 return k >= 0 && image->partitions[k].found;
3091}
3092
18d73705
LB		 3093MountOptions* mount_options_free_all(MountOptions *options) {
3094 MountOptions *m;
3095
3096 while ((m = options)) {
3097 LIST_REMOVE(mount_options, options, m);
3098 free(m->options);
3099 free(m);
3100 }
3101
3102 return NULL;
3103}
3104
569a0e42 3105const char* mount_options_from_designator(const MountOptions *options, PartitionDesignator designator) {
f5215bc8 3106 LIST_FOREACH(mount_options, m, options)
9ece6444 3107 if (designator == m->partition_designator && !isempty(m->options))
18d73705 3108 return m->options;
6aa05ebd 3109
18d73705
LB		 3110 return NULL;
3111}
3112
6aa05ebd
LP		 3113int mount_image_privately_interactively(
3114 const char *image,
3115 DissectImageFlags flags,
3116 char **ret_directory,
e330f97a 3117 LoopDevice **ret_loop_device) {
6aa05ebd 3118
27ec815e 3119 _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT;
6aa05ebd 3120 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
6aa05ebd
LP		 3121 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
3122 _cleanup_(rmdir_and_freep) char *created_dir = NULL;
3123 _cleanup_free_ char *temp = NULL;
3124 int r;
3125
3126 /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
3127 * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
3128 * easily. */
3129
3130 assert(image);
3131 assert(ret_directory);
3132 assert(ret_loop_device);
6aa05ebd 3133
73d88b80
LP		 3134 /* We intend to mount this right-away, hence add the partitions if needed and pin them*/
3135 flags |= DISSECT_IMAGE_ADD_PARTITION_DEVICES |
3136 DISSECT_IMAGE_PIN_PARTITION_DEVICES;
3137
27ec815e
LP		 3138 r = verity_settings_load(&verity, image, NULL, NULL);
3139 if (r < 0)
3140 return log_error_errno(r, "Failed to load root hash data: %m");
3141
6aa05ebd
LP		 3142 r = tempfn_random_child(NULL, program_invocation_short_name, &temp);
3143 if (r < 0)
3144 return log_error_errno(r, "Failed to generate temporary mount directory: %m");
3145
3146 r = loop_device_make_by_path(
3147 image,
ef9c184d 3148 FLAGS_SET(flags, DISSECT_IMAGE_DEVICE_READ_ONLY) ? O_RDONLY : O_RDWR,
6aa05ebd 3149 FLAGS_SET(flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN,
7f52206a 3150 LOCK_SH,
6aa05ebd
LP		 3151 &d);
3152 if (r < 0)
7b87fe4c 3153 return log_error_errno(r, "Failed to set up loopback device for %s: %m", image);
6aa05ebd 3154
64dd3a24 3155 r = dissect_loop_device_and_warn(d, &verity, NULL, flags, &dissected_image);
6aa05ebd
LP		 3156 if (r < 0)
3157 return r;
3158
88b3300f
LP		 3159 r = dissected_image_load_verity_sig_partition(dissected_image, d->fd, &verity);
3160 if (r < 0)
3161 return r;
3162
e330f97a 3163 r = dissected_image_decrypt_interactively(dissected_image, NULL, &verity, flags);
6aa05ebd
LP		 3164 if (r < 0)
3165 return r;
3166
3167 r = detach_mount_namespace();
3168 if (r < 0)
3169 return log_error_errno(r, "Failed to detach mount namespace: %m");
3170
3171 r = mkdir_p(temp, 0700);
3172 if (r < 0)
3173 return log_error_errno(r, "Failed to create mount point: %m");
3174
3175 created_dir = TAKE_PTR(temp);
3176
21b61b1d 3177 r = dissected_image_mount_and_warn(dissected_image, created_dir, UID_INVALID, UID_INVALID, flags);
6aa05ebd 3178 if (r < 0)
af187ab2 3179 return r;
6aa05ebd 3180
41bc4849
LP		 3181 r = loop_device_flock(d, LOCK_UN);
3182 if (r < 0)
3183 return r;
3184
3044d343
YW		 3185 r = dissected_image_relinquish(dissected_image);
3186 if (r < 0)
3187 return log_error_errno(r, "Failed to relinquish DM and loopback block devices: %m");
6aa05ebd
LP		 3188
3189 *ret_directory = TAKE_PTR(created_dir);
3190 *ret_loop_device = TAKE_PTR(d);
6aa05ebd
LP		 3191
3192 return 0;
3193}
3194
06768b90
LB		 3195static bool mount_options_relax_extension_release_checks(const MountOptions *options) {
3196 if (!options)
3197 return false;
3198
3199 return string_contains_word(mount_options_from_designator(options, PARTITION_ROOT), ",", "x-systemd.relax-extension-release-check") ||
3200 string_contains_word(mount_options_from_designator(options, PARTITION_USR), ",", "x-systemd.relax-extension-release-check") ||
3201 string_contains_word(options->options, ",", "x-systemd.relax-extension-release-check");
3202}
3203
93f59701 3204int verity_dissect_and_mount(
cedf5b1a 3205 int src_fd,
93f59701
LB		 3206 const char *src,
3207 const char *dest,
3208 const MountOptions *options,
3209 const char *required_host_os_release_id,
3210 const char *required_host_os_release_version_id,
60c5f700
LP		 3211 const char *required_host_os_release_sysext_level,
3212 const char *required_sysext_scope) {
93f59701 3213
4beda316 3214 _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL;
4beda316
LB		 3215 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
3216 _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT;
3217 DissectImageFlags dissect_image_flags;
06768b90 3218 bool relax_extension_release_check;
4beda316
LB		 3219 int r;
3220
3221 assert(src);
3222 assert(dest);
3223
06768b90
LB		 3224 relax_extension_release_check = mount_options_relax_extension_release_checks(options);
3225
cedf5b1a 3226 /* We might get an FD for the image, but we use the original path to look for the dm-verity files */
4beda316
LB		 3227 r = verity_settings_load(&verity, src, NULL, NULL);
3228 if (r < 0)
3229 return log_debug_errno(r, "Failed to load root hash: %m");
3230
06768b90 3231 dissect_image_flags = (verity.data_path ? DISSECT_IMAGE_NO_PARTITION_TABLE : 0) |
73d88b80
LP		 3232 (relax_extension_release_check ? DISSECT_IMAGE_RELAX_SYSEXT_CHECK : 0) |
3233 DISSECT_IMAGE_ADD_PARTITION_DEVICES |
3234 DISSECT_IMAGE_PIN_PARTITION_DEVICES;
4beda316 3235
cedf5b1a
LB		 3236 /* Note that we don't use loop_device_make here, as the FD is most likely O_PATH which would not be
3237 * accepted by LOOP_CONFIGURE, so just let loop_device_make_by_path reopen it as a regular FD. */
4beda316 3238 r = loop_device_make_by_path(
cedf5b1a 3239 src_fd >= 0 ? FORMAT_PROC_FD_PATH(src_fd) : src,
4beda316
LB		 3240 -1,
3241 verity.data_path ? 0 : LO_FLAGS_PARTSCAN,
7f52206a 3242 LOCK_SH,
4beda316
LB		 3243 &loop_device);
3244 if (r < 0)
3245 return log_debug_errno(r, "Failed to create loop device for image: %m");
3246
bad31660
YW		 3247 r = dissect_loop_device(
3248 loop_device,
4beda316
LB		 3249 &verity,
3250 options,
3251 dissect_image_flags,
3252 &dissected_image);
3253 /* No partition table? Might be a single-filesystem image, try again */
3254 if (!verity.data_path && r == -ENOPKG)
bad31660
YW		 3255 r = dissect_loop_device(
3256 loop_device,
4beda316
LB		 3257 &verity,
3258 options,
75dc190d 3259 dissect_image_flags | DISSECT_IMAGE_NO_PARTITION_TABLE,
4beda316
LB		 3260 &dissected_image);
3261 if (r < 0)
3262 return log_debug_errno(r, "Failed to dissect image: %m");
3263
88b3300f
LP		 3264 r = dissected_image_load_verity_sig_partition(dissected_image, loop_device->fd, &verity);
3265 if (r < 0)
3266 return r;
3267
4beda316
LB		 3268 r = dissected_image_decrypt(
3269 dissected_image,
3270 NULL,
3271 &verity,
e330f97a 3272 dissect_image_flags);
4beda316
LB		 3273 if (r < 0)
3274 return log_debug_errno(r, "Failed to decrypt dissected image: %m");
3275
3276 r = mkdir_p_label(dest, 0755);
3277 if (r < 0)
3278 return log_debug_errno(r, "Failed to create destination directory %s: %m", dest);
3279 r = umount_recursive(dest, 0);
3280 if (r < 0)
3281 return log_debug_errno(r, "Failed to umount under destination directory %s: %m", dest);
3282
21b61b1d 3283 r = dissected_image_mount(dissected_image, dest, UID_INVALID, UID_INVALID, dissect_image_flags);
4beda316
LB		 3284 if (r < 0)
3285 return log_debug_errno(r, "Failed to mount image: %m");
3286
41bc4849
LP		 3287 r = loop_device_flock(loop_device, LOCK_UN);
3288 if (r < 0)
3289 return log_debug_errno(r, "Failed to unlock loopback device: %m");
3290
93f59701
LB		 3291 /* If we got os-release values from the caller, then we need to match them with the image's
3292 * extension-release.d/ content. Return -EINVAL if there's any mismatch.
3293 * First, check the distro ID. If that matches, then check the new SYSEXT_LEVEL value if
37361f46
LB		 3294 * available, or else fallback to VERSION_ID. If neither is present (eg: rolling release),
3295 * then a simple match on the ID will be performed. */
8b2dcbbd 3296 if (required_host_os_release_id) {
93f59701
LB		 3297 _cleanup_strv_free_ char **extension_release = NULL;
3298
d30d86b7
YW		 3299 assert(!isempty(required_host_os_release_id));
3300
06768b90 3301 r = load_extension_release_pairs(dest, dissected_image->image_name, relax_extension_release_check, &extension_release);
93f59701
LB		 3302 if (r < 0)
3303 return log_debug_errno(r, "Failed to parse image %s extension-release metadata: %m", dissected_image->image_name);
3304
3305 r = extension_release_validate(
60c5f700
LP		 3306 dissected_image->image_name,
3307 required_host_os_release_id,
3308 required_host_os_release_version_id,
3309 required_host_os_release_sysext_level,
3310 required_sysext_scope,
3311 extension_release);
93f59701
LB		 3312 if (r == 0)
3313 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE), "Image %s extension-release metadata does not match the root's", dissected_image->image_name);
3314 if (r < 0)
3315 return log_debug_errno(r, "Failed to compare image %s extension-release metadata with the root's os-release: %m", dissected_image->image_name);
3316 }
3317
3044d343
YW		 3318 r = dissected_image_relinquish(dissected_image);
3319 if (r < 0)
3320 return log_debug_errno(r, "Failed to relinquish dissected image: %m");
4beda316
LB		 3321
3322 return 0;
3323}
Unnamed repository; edit this file 'description' to name the repository.