]> 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
dissect: actually enforce policy
[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"
5228c58e 28#include "btrfs-util.h"
f461a28d 29#include "chase.h"
c2fa92e7 30#include "conf-files.h"
28db6fbf 31#include "constants.h"
3b925504 32#include "copy.h"
1e2f3230 33#include "cryptsetup-util.h"
553e15f2 34#include "device-nodes.h"
8437c059 35#include "device-util.h"
f7725647 36#include "devnum-util.h"
7718ac97 37#include "discover-image.h"
8c1be37e 38#include "dissect-image.h"
a709a315 39#include "dm-util.h"
686d13b9 40#include "env-file.h"
88b3300f 41#include "env-util.h"
d51f8eb3 42#include "extension-util.h"
18b5886e 43#include "fd-util.h"
78ebe980 44#include "fileio.h"
2eedfd2d 45#include "fs-util.h"
cf32c486 46#include "fsck-util.h"
8c1be37e 47#include "gpt.h"
78ebe980 48#include "hexdecoct.h"
e2054217 49#include "hostname-setup.h"
3b925504 50#include "id128-util.h"
593fe6c0 51#include "import-util.h"
a4e0d617 52#include "io-util.h"
254e392e 53#include "missing_mount.h"
35cd0ba5 54#include "mkdir-label.h"
8c1be37e 55#include "mount-util.h"
e4de7287 56#include "mountpoint-util.h"
6aa05ebd 57#include "namespace-util.h"
d8b4d14d 58#include "nulstr-util.h"
c2fa92e7 59#include "openssl-util.h"
d58ad743 60#include "os-util.h"
8c1be37e 61#include "path-util.h"
3b925504
LP		 62#include "process-util.h"
63#include "raw-clone.h"
81939d9d 64#include "resize-fs.h"
3b925504 65#include "signal-util.h"
05c4c59f 66#include "sparse-endian.h"
8c1be37e 67#include "stat-util.h"
18b5886e 68#include "stdio-util.h"
8c1be37e
LP		 69#include "string-table.h"
70#include "string-util.h"
2eedfd2d 71#include "strv.h"
e4de7287 72#include "tmpfile-util.h"
a8040b6d 73#include "udev-util.h"
2d3a5a73 74#include "user-util.h"
41488e1f 75#include "xattr-util.h"
8c1be37e 76
28e2641a
FF		 77/* how many times to wait for the device nodes to appear */
78#define N_DEVICE_NODE_LIST_ATTEMPTS 10
79
80ce8580
LP		 80int dissect_fstype_ok(const char *fstype) {
81 const char *e;
82 bool b;
83
84 /* When we automatically mount file systems, be a bit conservative by default what we are willing to
85 * mount, just as an extra safety net to not mount with badly maintained legacy file system
86 * drivers. */
87
88 e = secure_getenv("SYSTEMD_DISSECT_FILE_SYSTEMS");
89 if (e) {
90 _cleanup_strv_free_ char **l = NULL;
91
92 l = strv_split(e, ":");
93 if (!l)
94 return -ENOMEM;
95
96 b = strv_contains(l, fstype);
97 } else
98 b = STR_IN_SET(fstype,
99 "btrfs",
100 "erofs",
101 "ext4",
102 "squashfs",
103 "vfat",
104 "xfs");
105 if (b)
106 return true;
107
108 log_debug("File system type '%s' is not allowed to be mounted as result of automatic dissection.", fstype);
109 return false;
110}
111
05c4c59f
LP		 112int probe_sector_size(int fd, uint32_t *ret) {
113
114 struct gpt_header {
115 char signature[8];
116 le32_t revision;
117 le32_t header_size;
118 le32_t crc32;
119 le32_t reserved;
120 le64_t my_lba;
121 le64_t alternate_lba;
122 le64_t first_usable_lba;
123 le64_t last_usable_lba;
124 sd_id128_t disk_guid;
125 le64_t partition_entry_lba;
126 le32_t number_of_partition_entries;
127 le32_t size_of_partition_entry;
128 le32_t partition_entry_array_crc32;
129 } _packed_;
130
131 /* Disk images might be for 512B or for 4096 sector sizes, let's try to auto-detect that by searching
132 * for the GPT headers at the relevant byte offsets */
133
134 assert_cc(sizeof(struct gpt_header) == 92);
135
136 /* We expect a sector size in the range 512…4096. The GPT header is located in the second
137 * sector. Hence it could be at byte 512 at the earliest, and at byte 4096 at the latest. And we must
138 * read with granularity of the largest sector size we care about. Which means 8K. */
139 uint8_t sectors[2 * 4096];
140 uint32_t found = 0;
141 ssize_t n;
142
143 assert(fd >= 0);
144 assert(ret);
145
146 n = pread(fd, sectors, sizeof(sectors), 0);
147 if (n < 0)
148 return -errno;
149 if (n != sizeof(sectors)) /* too short? */
150 goto not_found;
151
152 /* Let's see if we find the GPT partition header with various expected sector sizes */
153 for (uint32_t sz = 512; sz <= 4096; sz <<= 1) {
154 struct gpt_header *p;
155
156 assert(sizeof(sectors) >= sz * 2);
157 p = (struct gpt_header*) (sectors + sz);
158
159 if (memcmp(p->signature, (const char[8]) { 'E', 'F', 'I', ' ', 'P', 'A', 'R', 'T' }, 8) != 0)
160 continue;
161
162 if (le32toh(p->revision) != UINT32_C(0x00010000)) /* the only known revision of the spec: 1.0 */
163 continue;
164
165 if (le32toh(p->header_size) < sizeof(struct gpt_header))
166 continue;
167
168 if (le32toh(p->header_size) > 4096) /* larger than a sector? something is off… */
169 continue;
170
171 if (le64toh(p->my_lba) != 1) /* this sector must claim to be at sector offset 1 */
172 continue;
173
174 if (found != 0)
175 return log_debug_errno(SYNTHETIC_ERRNO(ENOTUNIQ),
176 "Detected valid partition table at offsets matching multiple sector sizes, refusing.");
177
178 found = sz;
179 }
180
181 if (found != 0) {
182 log_debug("Determined sector size %" PRIu32 " based on discovered partition table.", found);
183 *ret = found;
184 return 1; /* indicate we *did* find it */
185 }
186
187not_found:
188 log_debug("Couldn't find any partition table to derive sector size of.");
189 *ret = 512; /* pick the traditional default */
190 return 0; /* indicate we didn't find it */
191}
192
81dde3d8
LP		 193int probe_sector_size_prefer_ioctl(int fd, uint32_t *ret) {
194 struct stat st;
195
196 assert(fd >= 0);
197 assert(ret);
198
199 /* Just like probe_sector_size(), but if we are looking at a block device, will use the already
200 * configured sector size rather than probing by contents */
201
202 if (fstat(fd, &st) < 0)
203 return -errno;
204
205 if (S_ISBLK(st.st_mode))
206 return blockdev_get_sector_size(fd, ret);
207
208 return probe_sector_size(fd, ret);
209}
210
c80c9079
LP		 211int probe_filesystem_full(
212 int fd,
213 const char *path,
214 uint64_t offset,
215 uint64_t size,
216 char **ret_fstype) {
217
7cc84b2c 218 /* Try to find device content type and return it in *ret_fstype. If nothing is found,
c80c9079 219 * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and a
7cc84b2c
ZJS		 220 * different error otherwise. */
221
349cc4a5 222#if HAVE_BLKID
8e766630 223 _cleanup_(blkid_free_probep) blkid_probe b = NULL;
d2c6e79d 224 _cleanup_free_ char *path_by_fd = NULL;
254d1313 225 _cleanup_close_ int fd_close = -EBADF;
18b5886e
LP		 226 const char *fstype;
227 int r;
228
d2c6e79d
YW		 229 assert(fd >= 0 || path);
230 assert(ret_fstype);
231
232 if (fd < 0) {
233 fd_close = open(path, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
234 if (fd_close < 0)
235 return -errno;
236
237 fd = fd_close;
238 }
239
240 if (!path) {
241 r = fd_get_path(fd, &path_by_fd);
242 if (r < 0)
243 return r;
244
245 path = path_by_fd;
246 }
247
c80c9079
LP		 248 if (size == 0) /* empty size? nothing found! */
249 goto not_found;
250
d2c6e79d 251 b = blkid_new_probe();
18b5886e 252 if (!b)
d2c6e79d
YW		 253 return -ENOMEM;
254
255 errno = 0;
c80c9079
LP		 256 r = blkid_probe_set_device(
257 b,
258 fd,
259 offset,
260 size == UINT64_MAX ? 0 : size); /* when blkid sees size=0 it understands "everything". We prefer using UINT64_MAX for that */
d2c6e79d 261 if (r != 0)
66855de7 262 return errno_or_else(ENOMEM);
18b5886e
LP		 263
264 blkid_probe_enable_superblocks(b, 1);
265 blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE);
266
267 errno = 0;
268 r = blkid_do_safeprobe(b);
2e3944b8 269 if (r == _BLKID_SAFEPROBE_NOT_FOUND)
18b5886e 270 goto not_found;
2e3944b8 271 if (r == _BLKID_SAFEPROBE_AMBIGUOUS)
58dfbfbd 272 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN),
d2c6e79d 273 "Results ambiguous for partition %s", path);
2e3944b8 274 if (r == _BLKID_SAFEPROBE_ERROR)
d2c6e79d 275 return log_debug_errno(errno_or_else(EIO), "Failed to probe partition %s: %m", path);
18b5886e 276
2e3944b8
LP		 277 assert(r == _BLKID_SAFEPROBE_FOUND);
278
18b5886e
LP		 279 (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
280
281 if (fstype) {
282 char *t;
283
d2c6e79d 284 log_debug("Probed fstype '%s' on partition %s.", fstype, path);
1cefb9a6 285
18b5886e
LP		 286 t = strdup(fstype);
287 if (!t)
288 return -ENOMEM;
289
290 *ret_fstype = t;
291 return 1;
292 }
293
294not_found:
d2c6e79d 295 log_debug("No type detected on partition %s", path);
18b5886e
LP		 296 *ret_fstype = NULL;
297 return 0;
d1c536f5
ZJS		 298#else
299 return -EOPNOTSUPP;
a75e27eb 300#endif
d1c536f5 301}
18b5886e 302
40c10d3f 303#if HAVE_BLKID
cd22d856
LP		 304static int image_policy_may_use(
305 const ImagePolicy *policy,
306 PartitionDesignator designator) {
307
308 PartitionPolicyFlags f;
309
310 /* For each partition we find in the partition table do a first check if it may exist at all given
311 * the policy, or if it shall be ignored. */
312
313 f = image_policy_get_exhaustively(policy, designator);
314 if (f < 0)
315 return f;
316
317 if ((f & _PARTITION_POLICY_USE_MASK) == PARTITION_POLICY_ABSENT)
318 /* only flag set in policy is "absent"? then this partition may not exist at all */
319 return log_debug_errno(
320 SYNTHETIC_ERRNO(ERFKILL),
321 "Partition of designator '%s' exists, but not allowed by policy, refusing.",
322 partition_designator_to_string(designator));
323 if ((f & _PARTITION_POLICY_USE_MASK & ~PARTITION_POLICY_ABSENT) == PARTITION_POLICY_UNUSED) {
324 /* only "unused" or "unused" + "absent" are set? then don't use it */
325 log_debug("Partition of designator '%s' exists, and policy dictates to ignore it, doing so.",
326 partition_designator_to_string(designator));
327 return false; /* ignore! */
328 }
329
330 return true; /* use! */
331}
332
333static int image_policy_check_protection(
334 const ImagePolicy *policy,
335 PartitionDesignator designator,
336 PartitionPolicyFlags found_flags) {
337
338 PartitionPolicyFlags policy_flags;
339
340 /* Checks if the flags in the policy for the designated partition overlap the flags of what we found */
341
342 if (found_flags < 0)
343 return found_flags;
344
345 policy_flags = image_policy_get_exhaustively(policy, designator);
346 if (policy_flags < 0)
347 return policy_flags;
348
349 if ((found_flags & policy_flags) == 0) {
350 _cleanup_free_ char *found_flags_string = NULL, *policy_flags_string = NULL;
351
352 (void) partition_policy_flags_to_string(found_flags, /* simplify= */ true, &found_flags_string);
353 (void) partition_policy_flags_to_string(policy_flags, /* simplify= */ true, &policy_flags_string);
354
355 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL), "Partition %s discovered with policy '%s' but '%s' was required, refusing.",
356 partition_designator_to_string(designator),
357 strnull(found_flags_string), strnull(policy_flags_string));
358 }
359
360 return 0;
361}
362
363static int image_policy_check_partition_flags(
364 const ImagePolicy *policy,
365 PartitionDesignator designator,
366 uint64_t gpt_flags) {
367
368 PartitionPolicyFlags policy_flags;
369 bool b;
370
371 /* Checks if the partition flags in the policy match reality */
372
373 policy_flags = image_policy_get_exhaustively(policy, designator);
374 if (policy_flags < 0)
375 return policy_flags;
376
377 b = FLAGS_SET(gpt_flags, SD_GPT_FLAG_READ_ONLY);
378 if ((policy_flags & _PARTITION_POLICY_READ_ONLY_MASK) == (b ? PARTITION_POLICY_READ_ONLY_OFF : PARTITION_POLICY_READ_ONLY_ON))
379 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL), "Partition %s has 'read-only' flag incorrectly set (must be %s, is %s), refusing.",
380 partition_designator_to_string(designator),
381 one_zero(!b), one_zero(b));
382
383 b = FLAGS_SET(gpt_flags, SD_GPT_FLAG_GROWFS);
384 if ((policy_flags & _PARTITION_POLICY_GROWFS_MASK) == (b ? PARTITION_POLICY_GROWFS_OFF : PARTITION_POLICY_GROWFS_ON))
385 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL), "Partition %s has 'growfs' flag incorrectly set (must be %s, is %s), refusing.",
386 partition_designator_to_string(designator),
387 one_zero(!b), one_zero(b));
388
389 return 0;
390}
391
392static int dissected_image_probe_filesystems(
393 DissectedImage *m,
394 int fd,
395 const ImagePolicy *policy) {
396
698bb074
YW		 397 int r;
398
399 assert(m);
400
401 /* Fill in file system types if we don't know them yet. */
402
403 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
404 DissectedPartition *p = m->partitions + i;
cd22d856 405 PartitionPolicyFlags found_flags;
698bb074
YW		 406
407 if (!p->found)
408 continue;
409
c80c9079
LP		 410 if (!p->fstype) {
411 /* If we have an fd referring to the partition block device, use that. Otherwise go
412 * via the whole block device or backing regular file, and read via offset. */
413 if (p->mount_node_fd >= 0)
414 r = probe_filesystem_full(p->mount_node_fd, p->node, 0, UINT64_MAX, &p->fstype);
415 else
416 r = probe_filesystem_full(fd, p->node, p->offset, p->size, &p->fstype);
417 if (r < 0)
698bb074
YW		 418 return r;
419 }
420
cd22d856 421 if (streq_ptr(p->fstype, "crypto_LUKS")) {
698bb074 422 m->encrypted = true;
cd22d856
LP		 423 found_flags = PARTITION_POLICY_ENCRYPTED; /* found this one, and its definitely encrypted */
424 } else
425 /* found it, but it's definitely not encrypted, hence mask the encrypted flag, but
426 * set all other ways that indicate "present". */
427 found_flags = PARTITION_POLICY_UNPROTECTED|PARTITION_POLICY_VERITY|PARTITION_POLICY_SIGNED;
698bb074
YW		 428
429 if (p->fstype && fstype_is_ro(p->fstype))
430 p->rw = false;
431
432 if (!p->rw)
433 p->growfs = false;
cd22d856
LP		 434
435 /* We might have learnt more about the file system now (i.e. whether it is encrypted or not),
436 * hence we need to validate this against policy again, to see if the policy still matches
437 * with this new information. Note that image_policy_check_protection() will check for
438 * overlap between what's allowed in the policy and what we pass as 'found_policy' here. In
439 * the unencrypted case we thus might pass an overly unspecific mask here (i.e. unprotected
440 * OR verity OR signed), but that's fine since the earlier policy check already checked more
441 * specific which of those three cases where OK. Keep in mind that this function here only
442 * looks at specific partitions (and thus can only deduce encryption or not) but not the
443 * overall partition table (and thus cannot deduce verity or not). The earlier dissection
444 * checks already did the relevant checks that look at the whole partition table, and
445 * enforced policy there as needed. */
446 r = image_policy_check_protection(policy, i, found_flags);
447 if (r < 0)
448 return r;
698bb074
YW		 449 }
450
451 return 0;
452}
453
0f7c9a3d
LP		 454static void check_partition_flags(
455 const char *node,
456 unsigned long long pflags,
457 unsigned long long supported) {
458
459 assert(node);
460
461 /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
92e72028
ZJS		 462 pflags &= ~(supported |
463 SD_GPT_FLAG_REQUIRED_PARTITION |
464 SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL |
465 SD_GPT_FLAG_LEGACY_BIOS_BOOTABLE);
0f7c9a3d
LP		 466
467 if (pflags == 0)
468 return;
469
470 /* If there are other bits set, then log about it, to make things discoverable */
471 for (unsigned i = 0; i < sizeof(pflags) * 8; i++) {
472 unsigned long long bit = 1ULL << i;
473 if (!FLAGS_SET(pflags, bit))
474 continue;
475
476 log_debug("Unexpected partition flag %llu set on %s!", bit, node);
477 }
478}
479
00e29505
YW		 480static int dissected_image_new(const char *path, DissectedImage **ret) {
481 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
482 _cleanup_free_ char *name = NULL;
483 int r;
484
485 assert(ret);
486
487 if (path) {
488 _cleanup_free_ char *filename = NULL;
489
490 r = path_extract_filename(path, &filename);
491 if (r < 0)
492 return r;
493
494 r = raw_strip_suffixes(filename, &name);
495 if (r < 0)
496 return r;
497
498 if (!image_name_is_valid(name)) {
499 log_debug("Image name %s is not valid, ignoring.", strna(name));
500 name = mfree(name);
501 }
502 }
503
504 m = new(DissectedImage, 1);
505 if (!m)
506 return -ENOMEM;
507
508 *m = (DissectedImage) {
509 .has_init_system = -1,
510 .image_name = TAKE_PTR(name),
511 };
512
babd5b08
YW		 513 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++)
514 m->partitions[i] = DISSECTED_PARTITION_NULL;
515
00e29505
YW		 516 *ret = TAKE_PTR(m);
517 return 0;
518}
519#endif
520
234c2e16 521static void dissected_partition_done(DissectedPartition *p) {
1b010ae7 522 assert(p);
786e3a52 523
1b010ae7
LP		 524 free(p->fstype);
525 free(p->node);
526 free(p->label);
527 free(p->decrypted_fstype);
528 free(p->decrypted_node);
529 free(p->mount_options);
f7725647 530 safe_close(p->mount_node_fd);
786e3a52 531
babd5b08 532 *p = DISSECTED_PARTITION_NULL;
1b010ae7 533}
786e3a52 534
1b010ae7 535#if HAVE_BLKID
1b010ae7
LP		 536static int make_partition_devname(
537 const char *whole_devname,
1a81ddef 538 uint64_t diskseq,
1b010ae7 539 int nr,
1a81ddef 540 DissectImageFlags flags,
1b010ae7 541 char **ret) {
786e3a52 542
1a81ddef
LP		 543 _cleanup_free_ char *s = NULL;
544 int r;
786e3a52 545
1b010ae7 546 assert(whole_devname);
1a81ddef
LP		 547 assert(nr != 0); /* zero is not a valid partition nr */
548 assert(ret);
aae22eb3 549
c12a0d6d 550 if (!FLAGS_SET(flags, DISSECT_IMAGE_DISKSEQ_DEVNODE) || diskseq == 0) {
4ba86848 551
1a81ddef
LP		 552 /* Given a whole block device node name (e.g. /dev/sda or /dev/loop7) generate a partition
553 * device name (e.g. /dev/sda7 or /dev/loop7p5). The rule the kernel uses is simple: if whole
554 * block device node name ends in a digit, then suffix a 'p', followed by the partition
555 * number. Otherwise, just suffix the partition number without any 'p'. */
08fe0a53 556
1a81ddef
LP		 557 if (nr < 0) { /* whole disk? */
558 s = strdup(whole_devname);
559 if (!s)
560 return -ENOMEM;
561 } else {
562 size_t l = strlen(whole_devname);
563 if (l < 1) /* underflow check for the subtraction below */
564 return -EINVAL;
08fe0a53 565
1a81ddef
LP		 566 bool need_p = ascii_isdigit(whole_devname[l-1]); /* Last char a digit? */
567
568 if (asprintf(&s, "%s%s%i", whole_devname, need_p ? "p" : "", nr) < 0)
569 return -ENOMEM;
570 }
571 } else {
572 if (nr < 0) /* whole disk? */
573 r = asprintf(&s, "/dev/disk/by-diskseq/%" PRIu64, diskseq);
574 else
575 r = asprintf(&s, "/dev/disk/by-diskseq/%" PRIu64 "-part%i", diskseq, nr);
576 if (r < 0)
577 return -ENOMEM;
578 }
579
580 *ret = TAKE_PTR(s);
581 return 0;
08fe0a53
LP		 582}
583
1a81ddef
LP		 584static int open_partition(
585 const char *node,
586 bool is_partition,
587 const LoopDevice *loop) {
588
f7725647 589 _cleanup_(sd_device_unrefp) sd_device *dev = NULL;
254d1313 590 _cleanup_close_ int fd = -EBADF;
f7725647
YW		 591 dev_t devnum;
592 int r;
593
594 assert(node);
595 assert(loop);
596
597 fd = open(node, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
598 if (fd < 0)
599 return -errno;
600
601 /* Check if the block device is a child of (or equivalent to) the originally provided one. */
602 r = block_device_new_from_fd(fd, is_partition ? BLOCK_DEVICE_LOOKUP_WHOLE_DISK : 0, &dev);
603 if (r < 0)
604 return r;
605
606 r = sd_device_get_devnum(dev, &devnum);
607 if (r < 0)
608 return r;
609
610 if (loop->devno != devnum)
611 return -ENXIO;
612
613 /* Also check diskseq. */
c12a0d6d 614 if (loop->diskseq != 0) {
f7725647
YW		 615 uint64_t diskseq;
616
617 r = fd_get_diskseq(fd, &diskseq);
618 if (r < 0)
619 return r;
620
621 if (loop->diskseq != diskseq)
622 return -ENXIO;
623 }
624
625 log_debug("Opened %s (fd=%i, whole_block_devnum=" DEVNUM_FORMAT_STR ", diskseq=%" PRIu64 ").",
626 node, fd, DEVNUM_FORMAT_VAL(loop->devno), loop->diskseq);
627 return TAKE_FD(fd);
628}
629
22e932f4
DDM		 630static int compare_arch(Architecture a, Architecture b) {
631 if (a == b)
632 return 0;
633
634 if (a == native_architecture())
635 return 1;
636
637 if (b == native_architecture())
638 return -1;
639
640#ifdef ARCHITECTURE_SECONDARY
641 if (a == ARCHITECTURE_SECONDARY)
642 return 1;
643
644 if (b == ARCHITECTURE_SECONDARY)
645 return -1;
646#endif
647
648 return 0;
649}
650
08f14be4
YW		 651static int dissect_image(
652 DissectedImage *m,
4526113f 653 int fd,
0b214aa0 654 const char *devname,
89e62e0b 655 const VeritySettings *verity,
18d73705 656 const MountOptions *mount_options,
84be0c71 657 const ImagePolicy *policy,
08f14be4 658 DissectImageFlags flags) {
8c1be37e 659
62ea0ed0 660 sd_id128_t root_uuid = SD_ID128_NULL, root_verity_uuid = SD_ID128_NULL;
62ea0ed0 661 sd_id128_t usr_uuid = SD_ID128_NULL, usr_verity_uuid = SD_ID128_NULL;
1f8fb21c 662 bool is_gpt, is_mbr, multiple_generic = false,
de98f631
LP		 663 generic_rw = false, /* initialize to appease gcc */
664 generic_growfs = false;
8e766630 665 _cleanup_(blkid_free_probep) blkid_probe b = NULL;
8c1be37e 666 _cleanup_free_ char *generic_node = NULL;
be30ad41 667 sd_id128_t generic_uuid = SD_ID128_NULL;
b387778c 668 const char *pttype = NULL, *sptuuid = NULL;
8c1be37e 669 blkid_partlist pl;
1f8fb21c 670 int r, generic_nr = -1, n_partitions;
8c1be37e 671
08f14be4 672 assert(m);
8c1be37e 673 assert(fd >= 0);
0b214aa0 674 assert(devname);
a0bff7ea 675 assert(!verity || verity->designator < 0 || IN_SET(verity->designator, PARTITION_ROOT, PARTITION_USR));
89e62e0b 676 assert(!verity || verity->root_hash || verity->root_hash_size == 0);
a0bff7ea
LP		 677 assert(!verity || verity->root_hash_sig || verity->root_hash_sig_size == 0);
678 assert(!verity || (verity->root_hash || !verity->root_hash_sig));
e7cbe5cb 679 assert(!((flags & DISSECT_IMAGE_GPT_ONLY) && (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)));
22ee78a8 680 assert(m->sector_size > 0);
8c1be37e
LP		 681
682 /* Probes a disk image, and returns information about what it found in *ret.
683 *
4623e8e6 684 * Returns -ENOPKG if no suitable partition table or file system could be found.
2679f407
LP		 685 * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found.
686 * Returns -ENXIO if we couldn't find any partition suitable as root or /usr partition
687 * Returns -ENOTUNIQ if we only found multiple generic partitions and thus don't know what to do with that */
4623e8e6 688
c12a0d6d 689 uint64_t diskseq = m->loop ? m->loop->diskseq : 0;
1a81ddef 690
89e62e0b 691 if (verity && verity->root_hash) {
aee36b4e
LP		 692 sd_id128_t fsuuid, vuuid;
693
694 /* If a root hash is supplied, then we use the root partition that has a UUID that match the
695 * first 128bit of the root hash. And we use the verity partition that has a UUID that match
696 * the final 128bit. */
4623e8e6 697
89e62e0b 698 if (verity->root_hash_size < sizeof(sd_id128_t))
4623e8e6
LP		 699 return -EINVAL;
700
aee36b4e
LP		 701 memcpy(&fsuuid, verity->root_hash, sizeof(sd_id128_t));
702 memcpy(&vuuid, (const uint8_t*) verity->root_hash + verity->root_hash_size - sizeof(sd_id128_t), sizeof(sd_id128_t));
4623e8e6 703
aee36b4e 704 if (sd_id128_is_null(fsuuid))
4623e8e6 705 return -EINVAL;
aee36b4e 706 if (sd_id128_is_null(vuuid))
4623e8e6 707 return -EINVAL;
aee36b4e
LP		 708
709 /* If the verity data declares it's for the /usr partition, then search for that, in all
710 * other cases assume it's for the root partition. */
711 if (verity->designator == PARTITION_USR) {
712 usr_uuid = fsuuid;
713 usr_verity_uuid = vuuid;
714 } else {
715 root_uuid = fsuuid;
716 root_verity_uuid = vuuid;
717 }
4623e8e6 718 }
8c1be37e 719
8c1be37e
LP		 720 b = blkid_new_probe();
721 if (!b)
722 return -ENOMEM;
723
724 errno = 0;
725 r = blkid_probe_set_device(b, fd, 0, 0);
b382db9f 726 if (r != 0)
66855de7 727 return errno_or_else(ENOMEM);
8c1be37e 728
22ee78a8
LP		 729 errno = 0;
730 r = blkid_probe_set_sectorsize(b, m->sector_size);
731 if (r != 0)
732 return errno_or_else(EIO);
733
9b6deb03
LP		 734 if ((flags & DISSECT_IMAGE_GPT_ONLY) == 0) {
735 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
736 blkid_probe_enable_superblocks(b, 1);
b387778c 737 blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE|BLKID_SUBLKS_USAGE|BLKID_SUBLKS_UUID);
9b6deb03
LP		 738 }
739
8c1be37e
LP		 740 blkid_probe_enable_partitions(b, 1);
741 blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS);
742
743 errno = 0;
744 r = blkid_do_safeprobe(b);
2e3944b8 745 if (r == _BLKID_SAFEPROBE_ERROR)
66855de7 746 return errno_or_else(EIO);
2e3944b8
LP		 747 if (IN_SET(r, _BLKID_SAFEPROBE_AMBIGUOUS, _BLKID_SAFEPROBE_NOT_FOUND))
748 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG), "Failed to identify any partition table.");
749
750 assert(r == _BLKID_SAFEPROBE_FOUND);
8c1be37e 751
e7cbe5cb 752 if ((!(flags & DISSECT_IMAGE_GPT_ONLY) &&
4b5de5dd 753 (flags & DISSECT_IMAGE_GENERIC_ROOT)) ||
e7cbe5cb 754 (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)) {
9b6deb03 755 const char *usage = NULL;
8c1be37e 756
aee36b4e
LP		 757 /* If flags permit this, also allow using non-partitioned single-filesystem images */
758
9b6deb03
LP		 759 (void) blkid_probe_lookup_value(b, "USAGE", &usage, NULL);
760 if (STRPTR_IN_SET(usage, "filesystem", "crypto")) {
18d73705 761 _cleanup_free_ char *t = NULL, *n = NULL, *o = NULL;
b387778c 762 const char *fstype = NULL, *options = NULL, *suuid = NULL;
254d1313 763 _cleanup_close_ int mount_node_fd = -EBADF;
b387778c 764 sd_id128_t uuid = SD_ID128_NULL;
cd22d856
LP		 765 PartitionPolicyFlags found_flags;
766 bool encrypted;
767
768 /* OK, we have found a file system, that's our root partition then. */
769
770 r = image_policy_may_use(policy, PARTITION_ROOT);
771 if (r < 0)
772 return r;
773 if (r == 0) /* policy says ignore this, so we ignore it */
774 return -ENOPKG;
775
776 (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
777 (void) blkid_probe_lookup_value(b, "UUID", &suuid, NULL);
778
779 encrypted = streq_ptr(fstype, "crypto_LUKS");
780
781 if (verity_settings_data_covers(verity, PARTITION_ROOT))
782 found_flags = verity->root_hash_sig ? PARTITION_POLICY_SIGNED : PARTITION_POLICY_VERITY;
783 else
784 found_flags = encrypted ? PARTITION_POLICY_ENCRYPTED : PARTITION_POLICY_UNPROTECTED;
785
786 r = image_policy_check_protection(policy, PARTITION_ROOT, found_flags);
787 if (r < 0)
788 return r;
789
790 r = image_policy_check_partition_flags(policy, PARTITION_ROOT, 0); /* we have no gpt partition flags, hence check against all bits off */
791 if (r < 0)
792 return r;
f7725647 793
73d88b80 794 if (FLAGS_SET(flags, DISSECT_IMAGE_PIN_PARTITION_DEVICES)) {
f7725647
YW		 795 mount_node_fd = open_partition(devname, /* is_partition = */ false, m->loop);
796 if (mount_node_fd < 0)
797 return mount_node_fd;
798 }
8c1be37e 799
9b6deb03
LP		 800 if (fstype) {
801 t = strdup(fstype);
802 if (!t)
803 return -ENOMEM;
804 }
805
b387778c
LP		 806 if (suuid) {
807 /* blkid will return FAT's serial number as UUID, hence it is quite possible
808 * that parsing this will fail. We'll ignore the ID, since it's just too
809 * short to be useful as tru identifier. */
810 r = sd_id128_from_string(suuid, &uuid);
811 if (r < 0)
812 log_debug_errno(r, "Failed to parse file system UUID '%s', ignoring: %m", suuid);
813 }
814
1a81ddef
LP		 815 r = make_partition_devname(devname, diskseq, -1, flags, &n);
816 if (r < 0)
817 return r;
6c544d14 818
e7cbe5cb 819 m->single_file_system = true;
cd22d856 820 m->encrypted = encrypted;
c3c88d67
LP		 821
822 m->has_verity = verity && verity->data_path;
c2534821 823 m->verity_ready = verity_settings_data_covers(verity, PARTITION_ROOT);
e7cbe5cb 824
8ee9615e 825 m->has_verity_sig = false; /* signature not embedded, must be specified */
c2534821 826 m->verity_sig_ready = m->verity_ready && verity->root_hash_sig;
8ee9615e 827
b387778c
LP		 828 m->image_uuid = uuid;
829
f5215bc8 830 options = mount_options_from_designator(mount_options, PARTITION_ROOT);
18d73705
LB		 831 if (options) {
832 o = strdup(options);
833 if (!o)
834 return -ENOMEM;
835 }
836
9b6deb03
LP		 837 m->partitions[PARTITION_ROOT] = (DissectedPartition) {
838 .found = true,
e0d53d52 839 .rw = !m->verity_ready && !fstype_is_ro(fstype),
9b6deb03
LP		 840 .partno = -1,
841 .architecture = _ARCHITECTURE_INVALID,
1cc6c93a
YW		 842 .fstype = TAKE_PTR(t),
843 .node = TAKE_PTR(n),
18d73705 844 .mount_options = TAKE_PTR(o),
f7725647 845 .mount_node_fd = TAKE_FD(mount_node_fd),
88b3300f
LP		 846 .offset = 0,
847 .size = UINT64_MAX,
9b6deb03 848 };
8c1be37e 849
9b6deb03
LP		 850 return 0;
851 }
8c1be37e
LP		 852 }
853
854 (void) blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL);
855 if (!pttype)
856 return -ENOPKG;
857
858 is_gpt = streq_ptr(pttype, "gpt");
859 is_mbr = streq_ptr(pttype, "dos");
860
9b6deb03 861 if (!is_gpt && ((flags & DISSECT_IMAGE_GPT_ONLY) || !is_mbr))
8c1be37e
LP		 862 return -ENOPKG;
863
0903fd26
LP		 864 /* We support external verity data partitions only if the image has no partition table */
865 if (verity && verity->data_path)
866 return -EBADR;
867
73d88b80 868 if (FLAGS_SET(flags, DISSECT_IMAGE_ADD_PARTITION_DEVICES)) {
08f14be4
YW		 869 /* Safety check: refuse block devices that carry a partition table but for which the kernel doesn't
870 * do partition scanning. */
871 r = blockdev_partscan_enabled(fd);
872 if (r < 0)
873 return r;
874 if (r == 0)
875 return -EPROTONOSUPPORT;
876 }
4ba86848 877
b387778c
LP		 878 (void) blkid_probe_lookup_value(b, "PTUUID", &sptuuid, NULL);
879 if (sptuuid) {
880 r = sd_id128_from_string(sptuuid, &m->image_uuid);
881 if (r < 0)
882 log_debug_errno(r, "Failed to parse partition table UUID '%s', ignoring: %m", sptuuid);
883 }
884
8c1be37e
LP		 885 errno = 0;
886 pl = blkid_probe_get_partitions(b);
b382db9f 887 if (!pl)
66855de7 888 return errno_or_else(ENOMEM);
8c1be37e 889
4ba86848
LP		 890 errno = 0;
891 n_partitions = blkid_partlist_numof_partitions(pl);
892 if (n_partitions < 0)
893 return errno_or_else(EIO);
8c1be37e 894
4ba86848 895 for (int i = 0; i < n_partitions; i++) {
1b010ae7 896 _cleanup_free_ char *node = NULL;
9b6deb03 897 unsigned long long pflags;
88b3300f 898 blkid_loff_t start, size;
8c1be37e 899 blkid_partition pp;
8c1be37e
LP		 900 int nr;
901
4ba86848
LP		 902 errno = 0;
903 pp = blkid_partlist_get_partition(pl, i);
904 if (!pp)
905 return errno_or_else(EIO);
aae22eb3 906
9b6deb03 907 pflags = blkid_partition_get_flags(pp);
8c1be37e 908
4ba86848 909 errno = 0;
8c1be37e
LP		 910 nr = blkid_partition_get_partno(pp);
911 if (nr < 0)
4ba86848 912 return errno_or_else(EIO);
8c1be37e 913
88b3300f
LP		 914 errno = 0;
915 start = blkid_partition_get_start(pp);
916 if (start < 0)
917 return errno_or_else(EIO);
918
919 assert((uint64_t) start < UINT64_MAX/512);
920
921 errno = 0;
922 size = blkid_partition_get_size(pp);
923 if (size < 0)
924 return errno_or_else(EIO);
925
926 assert((uint64_t) size < UINT64_MAX/512);
927
1a81ddef
LP		 928 /* While probing we need the non-diskseq device node name to access the thing, hence mask off
929 * DISSECT_IMAGE_DISKSEQ_DEVNODE. */
930 r = make_partition_devname(devname, diskseq, nr, flags & ~DISSECT_IMAGE_DISKSEQ_DEVNODE, &node);
1b010ae7
LP		 931 if (r < 0)
932 return r;
933
934 /* So here's the thing: after the main ("whole") block device popped up it might take a while
935 * before the kernel fully probed the partition table. Waiting for that to finish is icky in
936 * userspace. So here's what we do instead. We issue the BLKPG_ADD_PARTITION ioctl to add the
937 * partition ourselves, racing against the kernel. Good thing is: if this call fails with
938 * EBUSY then the kernel was quicker than us, and that's totally OK, the outcome is good for
939 * us: the device node will exist. If OTOH our call was successful we won the race. Which is
940 * also good as the outcome is the same: the partition block device exists, and we can use
941 * it.
942 *
943 * Kernel returns EBUSY if there's already a partition by that number or an overlapping
944 * partition already existent. */
945
73d88b80 946 if (FLAGS_SET(flags, DISSECT_IMAGE_ADD_PARTITION_DEVICES)) {
08f14be4
YW		 947 r = block_device_add_partition(fd, node, nr, (uint64_t) start * 512, (uint64_t) size * 512);
948 if (r < 0) {
949 if (r != -EBUSY)
950 return log_debug_errno(r, "BLKPG_ADD_PARTITION failed: %m");
1b010ae7 951
08f14be4
YW		 952 log_debug_errno(r, "Kernel was quicker than us in adding partition %i.", nr);
953 } else
954 log_debug("We were quicker than kernel in adding partition %i.", nr);
955 }
1b010ae7 956
8c1be37e 957 if (is_gpt) {
e3b9a5ff 958 const char *fstype = NULL, *label;
4623e8e6 959 sd_id128_t type_id, id;
22e932f4 960 GptPartitionType type;
de98f631 961 bool rw = true, growfs = false;
8c1be37e 962
e3b9a5ff
LP		 963 r = blkid_partition_get_uuid_id128(pp, &id);
964 if (r < 0) {
965 log_debug_errno(r, "Failed to read partition UUID, ignoring: %m");
4623e8e6 966 continue;
e3b9a5ff 967 }
4623e8e6 968
e3b9a5ff
LP		 969 r = blkid_partition_get_type_id128(pp, &type_id);
970 if (r < 0) {
971 log_debug_errno(r, "Failed to read partition type UUID, ignoring: %m");
8c1be37e 972 continue;
e3b9a5ff 973 }
8c1be37e 974
22e932f4
DDM		 975 type = gpt_partition_type_from_uuid(type_id);
976
08fe0a53
LP		 977 label = blkid_partition_get_name(pp); /* libblkid returns NULL here if empty */
978
fb3921b8
LP		 979 if (IN_SET(type.designator,
980 PARTITION_HOME,
981 PARTITION_SRV,
982 PARTITION_XBOOTLDR,
983 PARTITION_TMP)) {
a48dd347 984
92e72028
ZJS		 985 check_partition_flags(node, pflags,
986 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS);
0f7c9a3d 987
92e72028 988 if (pflags & SD_GPT_FLAG_NO_AUTO)
a48dd347
LP		 989 continue;
990
92e72028
ZJS		 991 rw = !(pflags & SD_GPT_FLAG_READ_ONLY);
992 growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS);
aee36b4e 993
22e932f4 994 } else if (type.designator == PARTITION_ESP) {
a48dd347 995
92e72028
ZJS		 996 /* Note that we don't check the SD_GPT_FLAG_NO_AUTO flag for the ESP, as it is
997 * not defined there. We instead check the SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as
aee36b4e
LP		 998 * recommended by the UEFI spec (See "12.3.3 Number and Location of System
999 * Partitions"). */
a48dd347 1000
92e72028 1001 if (pflags & SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL)
a48dd347
LP		 1002 continue;
1003
8c1be37e 1004 fstype = "vfat";
a8c47660 1005
22e932f4 1006 } else if (type.designator == PARTITION_ROOT) {
4623e8e6 1007
92e72028
ZJS		 1008 check_partition_flags(node, pflags,
1009 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS);
0f7c9a3d 1010
92e72028 1011 if (pflags & SD_GPT_FLAG_NO_AUTO)
a48dd347
LP		 1012 continue;
1013
4623e8e6
LP		 1014 /* If a root ID is specified, ignore everything but the root id */
1015 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
1016 continue;
1017
92e72028
ZJS		 1018 rw = !(pflags & SD_GPT_FLAG_READ_ONLY);
1019 growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS);
aee36b4e 1020
22e932f4 1021 } else if (type.designator == PARTITION_ROOT_VERITY) {
4623e8e6 1022
92e72028
ZJS		 1023 check_partition_flags(node, pflags,
1024 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY);
0f7c9a3d 1025
92e72028 1026 if (pflags & SD_GPT_FLAG_NO_AUTO)
a48dd347
LP		 1027 continue;
1028
c3c88d67 1029 m->has_verity = true;
4623e8e6 1030
8ee9615e
LP		 1031 /* If no verity configuration is specified, then don't do verity */
1032 if (!verity)
1033 continue;
1034 if (verity->designator >= 0 && verity->designator != PARTITION_ROOT)
1035 continue;
1036
1037 /* If root hash is specified, then ignore everything but the root id */
1038 if (!sd_id128_is_null(root_verity_uuid) && !sd_id128_equal(root_verity_uuid, id))
4623e8e6
LP		 1039 continue;
1040
4623e8e6 1041 fstype = "DM_verity_hash";
4623e8e6 1042 rw = false;
8ee9615e 1043
22e932f4 1044 } else if (type.designator == PARTITION_ROOT_VERITY_SIG) {
8ee9615e 1045
92e72028
ZJS		 1046 check_partition_flags(node, pflags,
1047 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY);
8ee9615e 1048
92e72028 1049 if (pflags & SD_GPT_FLAG_NO_AUTO)
8ee9615e
LP		 1050 continue;
1051
1052 m->has_verity_sig = true;
1053
8ee9615e
LP		 1054 if (!verity)
1055 continue;
1056 if (verity->designator >= 0 && verity->designator != PARTITION_ROOT)
1057 continue;
8ee9615e 1058
8ee9615e 1059 fstype = "verity_hash_signature";
4623e8e6 1060 rw = false;
8ee9615e 1061
22e932f4 1062 } else if (type.designator == PARTITION_USR) {
aee36b4e 1063
92e72028
ZJS		 1064 check_partition_flags(node, pflags,
1065 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS);
aee36b4e 1066
92e72028 1067 if (pflags & SD_GPT_FLAG_NO_AUTO)
aee36b4e
LP		 1068 continue;
1069
1070 /* If a usr ID is specified, ignore everything but the usr id */
1071 if (!sd_id128_is_null(usr_uuid) && !sd_id128_equal(usr_uuid, id))
1072 continue;
1073
92e72028
ZJS		 1074 rw = !(pflags & SD_GPT_FLAG_READ_ONLY);
1075 growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS);
aee36b4e 1076
22e932f4 1077 } else if (type.designator == PARTITION_USR_VERITY) {
aee36b4e 1078
92e72028
ZJS		 1079 check_partition_flags(node, pflags,
1080 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY);
aee36b4e 1081
92e72028 1082 if (pflags & SD_GPT_FLAG_NO_AUTO)
aee36b4e
LP		 1083 continue;
1084
c3c88d67 1085 m->has_verity = true;
aee36b4e 1086
8ee9615e
LP		 1087 if (!verity)
1088 continue;
1089 if (verity->designator >= 0 && verity->designator != PARTITION_USR)
1090 continue;
1091
1092 /* If usr hash is specified, then ignore everything but the usr id */
1093 if (!sd_id128_is_null(usr_verity_uuid) && !sd_id128_equal(usr_verity_uuid, id))
aee36b4e
LP		 1094 continue;
1095
aee36b4e 1096 fstype = "DM_verity_hash";
aee36b4e 1097 rw = false;
8ee9615e 1098
22e932f4 1099 } else if (type.designator == PARTITION_USR_VERITY_SIG) {
8ee9615e 1100
92e72028
ZJS		 1101 check_partition_flags(node, pflags,
1102 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY);
8ee9615e 1103
92e72028 1104 if (pflags & SD_GPT_FLAG_NO_AUTO)
8ee9615e
LP		 1105 continue;
1106
1107 m->has_verity_sig = true;
1108
8ee9615e
LP		 1109 if (!verity)
1110 continue;
1111 if (verity->designator >= 0 && verity->designator != PARTITION_USR)
1112 continue;
8ee9615e 1113
8ee9615e 1114 fstype = "verity_hash_signature";
aee36b4e 1115 rw = false;
8ee9615e 1116
22e932f4 1117 } else if (type.designator == PARTITION_SWAP) {
a48dd347 1118
92e72028 1119 check_partition_flags(node, pflags, SD_GPT_FLAG_NO_AUTO);
0f7c9a3d 1120
92e72028 1121 if (pflags & SD_GPT_FLAG_NO_AUTO)
a48dd347
LP		 1122 continue;
1123
41aca66b
LP		 1124 /* Note: we don't set fstype = "swap" here, because we still need to probe if
1125 * it might be encrypted (i.e. fstype "crypt_LUKS") or unencrypted
1126 * (i.e. fstype "swap"), and the only way to figure that out is via fstype
1127 * probing. */
df4524cb 1128
df655bf3
DDM		 1129 /* We don't have a designator for SD_GPT_LINUX_GENERIC so check the UUID instead. */
1130 } else if (sd_id128_equal(type.uuid, SD_GPT_LINUX_GENERIC)) {
8c1be37e 1131
92e72028
ZJS		 1132 check_partition_flags(node, pflags,
1133 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS);
0f7c9a3d 1134
92e72028 1135 if (pflags & SD_GPT_FLAG_NO_AUTO)
a48dd347
LP		 1136 continue;
1137
8c1be37e
LP		 1138 if (generic_node)
1139 multiple_generic = true;
1140 else {
1141 generic_nr = nr;
92e72028
ZJS		 1142 generic_rw = !(pflags & SD_GPT_FLAG_READ_ONLY);
1143 generic_growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS);
be30ad41 1144 generic_uuid = id;
a5590886 1145 generic_node = TAKE_PTR(node);
8c1be37e 1146 }
d4dffb85 1147
22e932f4 1148 } else if (type.designator == PARTITION_VAR) {
d4dffb85 1149
92e72028
ZJS		 1150 check_partition_flags(node, pflags,
1151 SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS);
0f7c9a3d 1152
92e72028 1153 if (pflags & SD_GPT_FLAG_NO_AUTO)
d4dffb85
LP		 1154 continue;
1155
1156 if (!FLAGS_SET(flags, DISSECT_IMAGE_RELAX_VAR_CHECK)) {
1157 sd_id128_t var_uuid;
1158
1159 /* For /var we insist that the uuid of the partition matches the
1160 * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
1161 * ID. Why? Unlike the other partitions /var is inherently
1162 * installation specific, hence we need to be careful not to mount it
1163 * in the wrong installation. By hashing the partition UUID from
1164 * /etc/machine-id we can securely bind the partition to the
1165 * installation. */
1166
92e72028 1167 r = sd_id128_get_machine_app_specific(SD_GPT_VAR, &var_uuid);
d4dffb85
LP		 1168 if (r < 0)
1169 return r;
1170
1171 if (!sd_id128_equal(var_uuid, id)) {
a52efa81
YW		 1172 log_debug("Found a /var/ partition, but its UUID didn't match our expectations "
1173 "(found: " SD_ID128_UUID_FORMAT_STR ", expected: " SD_ID128_UUID_FORMAT_STR "), ignoring.",
1174 SD_ID128_FORMAT_VAL(id), SD_ID128_FORMAT_VAL(var_uuid));
d4dffb85
LP		 1175 continue;
1176 }
1177 }
1178
92e72028
ZJS		 1179 rw = !(pflags & SD_GPT_FLAG_READ_ONLY);
1180 growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS);
8c1be37e
LP		 1181 }
1182
22e932f4 1183 if (type.designator != _PARTITION_DESIGNATOR_INVALID) {
1a81ddef 1184 _cleanup_free_ char *t = NULL, *o = NULL, *l = NULL, *n = NULL;
254d1313 1185 _cleanup_close_ int mount_node_fd = -EBADF;
18d73705 1186 const char *options = NULL;
8c1be37e 1187
cd22d856
LP		 1188 r = image_policy_may_use(policy, type.designator);
1189 if (r < 0)
1190 return r;
1191 if (r == 0) {
1192 /* Policy says: ignore; Remember this fact, so that we later can distinguish between "found but ignored" and "not found at all" */
1193
1194 if (!m->partitions[type.designator].found)
1195 m->partitions[type.designator].ignored = true;
1196
1197 continue;
1198 }
1199
22e932f4 1200 if (m->partitions[type.designator].found) {
08fe0a53
LP		 1201 /* For most partition types the first one we see wins. Except for the
1202 * rootfs and /usr, where we do a version compare of the label, and
1203 * let the newest version win. This permits a simple A/B versioning
1204 * scheme in OS images. */
1205
22e932f4
DDM		 1206 if (compare_arch(type.arch, m->partitions[type.designator].architecture) <= 0)
1207 continue;
1208
1209 if (!partition_designator_is_versioned(type.designator) ||
1210 strverscmp_improved(m->partitions[type.designator].label, label) >= 0)
08fe0a53
LP		 1211 continue;
1212
22e932f4 1213 dissected_partition_done(m->partitions + type.designator);
08fe0a53 1214 }
8c1be37e 1215
73d88b80 1216 if (FLAGS_SET(flags, DISSECT_IMAGE_PIN_PARTITION_DEVICES) &&
f52261a0 1217 type.designator != PARTITION_SWAP) {
f7725647
YW		 1218 mount_node_fd = open_partition(node, /* is_partition = */ true, m->loop);
1219 if (mount_node_fd < 0)
1220 return mount_node_fd;
1221 }
1222
1a81ddef
LP		 1223 r = make_partition_devname(devname, diskseq, nr, flags, &n);
1224 if (r < 0)
1225 return r;
1226
8c1be37e
LP		 1227 if (fstype) {
1228 t = strdup(fstype);
1229 if (!t)
1230 return -ENOMEM;
1231 }
1232
08fe0a53
LP		 1233 if (label) {
1234 l = strdup(label);
1235 if (!l)
1236 return -ENOMEM;
1237 }
1238
22e932f4 1239 options = mount_options_from_designator(mount_options, type.designator);
18d73705
LB		 1240 if (options) {
1241 o = strdup(options);
1242 if (!o)
1243 return -ENOMEM;
1244 }
1245
22e932f4 1246 m->partitions[type.designator] = (DissectedPartition) {
8c1be37e
LP		 1247 .found = true,
1248 .partno = nr,
1249 .rw = rw,
de98f631 1250 .growfs = growfs,
22e932f4 1251 .architecture = type.arch,
1a81ddef 1252 .node = TAKE_PTR(n),
1cc6c93a 1253 .fstype = TAKE_PTR(t),
08fe0a53 1254 .label = TAKE_PTR(l),
be30ad41 1255 .uuid = id,
18d73705 1256 .mount_options = TAKE_PTR(o),
f7725647 1257 .mount_node_fd = TAKE_FD(mount_node_fd),
88b3300f
LP		 1258 .offset = (uint64_t) start * 512,
1259 .size = (uint64_t) size * 512,
d90b03f8 1260 .gpt_flags = pflags,
8c1be37e 1261 };
8c1be37e
LP		 1262 }
1263
1264 } else if (is_mbr) {
1265
a8c47660 1266 switch (blkid_partition_get_type(pp)) {
8c1be37e 1267
a8c47660
LP		 1268 case 0x83: /* Linux partition */
1269
1270 if (pflags != 0x80) /* Bootable flag */
1271 continue;
8c1be37e 1272
a8c47660
LP		 1273 if (generic_node)
1274 multiple_generic = true;
1275 else {
1276 generic_nr = nr;
1277 generic_rw = true;
de98f631 1278 generic_growfs = false;
a5590886 1279 generic_node = TAKE_PTR(node);
a8c47660
LP		 1280 }
1281
1282 break;
1283
1284 case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
254d1313 1285 _cleanup_close_ int mount_node_fd = -EBADF;
1a81ddef 1286 _cleanup_free_ char *o = NULL, *n = NULL;
a8c47660 1287 sd_id128_t id = SD_ID128_NULL;
e3b9a5ff 1288 const char *options = NULL;
a8c47660 1289
cd22d856
LP		 1290 r = image_policy_may_use(policy, PARTITION_XBOOTLDR);
1291 if (r < 0)
1292 return r;
1293 if (r == 0) { /* policy says: ignore */
1294 if (!m->partitions[PARTITION_XBOOTLDR].found)
1295 m->partitions[PARTITION_XBOOTLDR].ignored = true;
1296
1297 continue;
1298 }
1299
a8c47660 1300 /* First one wins */
234c2e16 1301 if (m->partitions[PARTITION_XBOOTLDR].found)
a8c47660
LP		 1302 continue;
1303
73d88b80 1304 if (FLAGS_SET(flags, DISSECT_IMAGE_PIN_PARTITION_DEVICES)) {
f7725647
YW		 1305 mount_node_fd = open_partition(node, /* is_partition = */ true, m->loop);
1306 if (mount_node_fd < 0)
1307 return mount_node_fd;
1308 }
1309
e3b9a5ff 1310 (void) blkid_partition_get_uuid_id128(pp, &id);
a8c47660 1311
1a81ddef
LP		 1312 r = make_partition_devname(devname, diskseq, nr, flags, &n);
1313 if (r < 0)
1314 return r;
1315
f5215bc8 1316 options = mount_options_from_designator(mount_options, PARTITION_XBOOTLDR);
18d73705
LB		 1317 if (options) {
1318 o = strdup(options);
1319 if (!o)
1320 return -ENOMEM;
1321 }
1322
a8c47660
LP		 1323 m->partitions[PARTITION_XBOOTLDR] = (DissectedPartition) {
1324 .found = true,
1325 .partno = nr,
1326 .rw = true,
de98f631 1327 .growfs = false,
a8c47660 1328 .architecture = _ARCHITECTURE_INVALID,
1a81ddef 1329 .node = TAKE_PTR(n),
a8c47660 1330 .uuid = id,
18d73705 1331 .mount_options = TAKE_PTR(o),
f7725647 1332 .mount_node_fd = TAKE_FD(mount_node_fd),
88b3300f
LP		 1333 .offset = (uint64_t) start * 512,
1334 .size = (uint64_t) size * 512,
a8c47660
LP		 1335 };
1336
1337 break;
1338 }}
8c1be37e
LP		 1339 }
1340 }
1341
22e932f4
DDM		 1342 if (!m->partitions[PARTITION_ROOT].found &&
1343 (m->partitions[PARTITION_ROOT_VERITY].found ||
1344 m->partitions[PARTITION_ROOT_VERITY_SIG].found))
1345 return -EADDRNOTAVAIL; /* Verity found but no matching rootfs? Something is off, refuse. */
49ae9d91 1346
8ee9615e
LP		 1347 /* Hmm, we found a signature partition but no Verity data? Something is off. */
1348 if (m->partitions[PARTITION_ROOT_VERITY_SIG].found && !m->partitions[PARTITION_ROOT_VERITY].found)
1349 return -EADDRNOTAVAIL;
7cf66030 1350
22e932f4
DDM		 1351 if (!m->partitions[PARTITION_USR].found &&
1352 (m->partitions[PARTITION_USR_VERITY].found ||
1353 m->partitions[PARTITION_USR_VERITY_SIG].found))
1354 return -EADDRNOTAVAIL; /* as above */
7cf66030 1355
22e932f4 1356 /* as above */
8ee9615e
LP		 1357 if (m->partitions[PARTITION_USR_VERITY_SIG].found && !m->partitions[PARTITION_USR_VERITY].found)
1358 return -EADDRNOTAVAIL;
1359
cb241a69
LP		 1360 /* If root and /usr are combined then insist that the architecture matches */
1361 if (m->partitions[PARTITION_ROOT].found &&
1362 m->partitions[PARTITION_USR].found &&
1363 (m->partitions[PARTITION_ROOT].architecture >= 0 &&
1364 m->partitions[PARTITION_USR].architecture >= 0 &&
1365 m->partitions[PARTITION_ROOT].architecture != m->partitions[PARTITION_USR].architecture))
1366 return -EADDRNOTAVAIL;
1367
4ab51780
LP		 1368 if (!m->partitions[PARTITION_ROOT].found &&
1369 !m->partitions[PARTITION_USR].found &&
1370 (flags & DISSECT_IMAGE_GENERIC_ROOT) &&
00a8b34f 1371 (!verity || !verity->root_hash || verity->designator != PARTITION_USR)) {
7cf66030 1372
1b010ae7 1373 /* OK, we found nothing usable, then check if there's a single generic partition, and use
4b5de5dd
LP		 1374 * that. If the root hash was set however, then we won't fall back to a generic node, because
1375 * the root hash decides. */
7cf66030
LP		 1376
1377 /* If we didn't find a properly marked root partition, but we did find a single suitable
1378 * generic Linux partition, then use this as root partition, if the caller asked for it. */
1379 if (multiple_generic)
1380 return -ENOTUNIQ;
1381
4b5de5dd
LP		 1382 /* If we didn't find a generic node, then we can't fix this up either */
1383 if (generic_node) {
cd22d856 1384 r = image_policy_may_use(policy, PARTITION_ROOT);
1a81ddef
LP		 1385 if (r < 0)
1386 return r;
cd22d856
LP		 1387 if (r == 0)
1388 /* Policy says: ignore; remember that we did */
1389 m->partitions[PARTITION_ROOT].ignored = true;
1390 else {
1391 _cleanup_close_ int mount_node_fd = -EBADF;
1392 _cleanup_free_ char *o = NULL, *n = NULL;
1393 const char *options;
1a81ddef 1394
cd22d856
LP		 1395 if (FLAGS_SET(flags, DISSECT_IMAGE_PIN_PARTITION_DEVICES)) {
1396 mount_node_fd = open_partition(generic_node, /* is_partition = */ true, m->loop);
1397 if (mount_node_fd < 0)
1398 return mount_node_fd;
1399 }
18d73705 1400
cd22d856
LP		 1401 r = make_partition_devname(devname, diskseq, generic_nr, flags, &n);
1402 if (r < 0)
1403 return r;
1404
1405 options = mount_options_from_designator(mount_options, PARTITION_ROOT);
1406 if (options) {
1407 o = strdup(options);
1408 if (!o)
1409 return -ENOMEM;
1410 }
1411
1412 assert(generic_nr >= 0);
1413 m->partitions[PARTITION_ROOT] = (DissectedPartition) {
1414 .found = true,
1415 .rw = generic_rw,
1416 .growfs = generic_growfs,
1417 .partno = generic_nr,
1418 .architecture = _ARCHITECTURE_INVALID,
1419 .node = TAKE_PTR(n),
1420 .uuid = generic_uuid,
1421 .mount_options = TAKE_PTR(o),
1422 .mount_node_fd = TAKE_FD(mount_node_fd),
1423 .offset = UINT64_MAX,
1424 .size = UINT64_MAX,
1425 };
1426 }
e0f9e7bd 1427 }
8c1be37e
LP		 1428 }
1429
4b5de5dd
LP		 1430 /* 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 */
1431 if (FLAGS_SET(flags, DISSECT_IMAGE_REQUIRE_ROOT) &&
1432 !(m->partitions[PARTITION_ROOT].found || (m->partitions[PARTITION_USR].found && FLAGS_SET(flags, DISSECT_IMAGE_USR_NO_ROOT))))
1433 return -ENXIO;
1434
7b32164f
LP		 1435 if (m->partitions[PARTITION_ROOT_VERITY].found) {
1436 /* We only support one verity partition per image, i.e. can't do for both /usr and root fs */
1437 if (m->partitions[PARTITION_USR_VERITY].found)
1438 return -ENOTUNIQ;
1439
1440 /* We don't support verity enabled root with a split out /usr. Neither with nor without
1441 * verity there. (Note that we do support verity-less root with verity-full /usr, though.) */
1442 if (m->partitions[PARTITION_USR].found)
1443 return -EADDRNOTAVAIL;
1444 }
aee36b4e 1445
1903defc
LP		 1446 if (verity) {
1447 /* If a verity designator is specified, then insist that the matching partition exists */
1448 if (verity->designator >= 0 && !m->partitions[verity->designator].found)
1449 return -EADDRNOTAVAIL;
aee36b4e 1450
3dd73ea7
DDM		 1451 bool have_verity_sig_partition =
1452 m->partitions[verity->designator == PARTITION_USR ? PARTITION_USR_VERITY_SIG : PARTITION_ROOT_VERITY_SIG].found;
1453
1903defc 1454 if (verity->root_hash) {
8ee9615e
LP		 1455 /* If we have an explicit root hash and found the partitions for it, then we are ready to use
1456 * Verity, set things up for it */
1457
1903defc
LP		 1458 if (verity->designator < 0 || verity->designator == PARTITION_ROOT) {
1459 if (!m->partitions[PARTITION_ROOT_VERITY].found || !m->partitions[PARTITION_ROOT].found)
1460 return -EADDRNOTAVAIL;
4623e8e6 1461
1903defc
LP		 1462 /* If we found a verity setup, then the root partition is necessarily read-only. */
1463 m->partitions[PARTITION_ROOT].rw = false;
1464 m->verity_ready = true;
1903defc 1465
f9e0bb61
LP		 1466 } else {
1467 assert(verity->designator == PARTITION_USR);
1468
1903defc
LP		 1469 if (!m->partitions[PARTITION_USR_VERITY].found || !m->partitions[PARTITION_USR].found)
1470 return -EADDRNOTAVAIL;
4623e8e6 1471
1903defc
LP		 1472 m->partitions[PARTITION_USR].rw = false;
1473 m->verity_ready = true;
1474 }
8ee9615e
LP		 1475
1476 if (m->verity_ready)
3dd73ea7 1477 m->verity_sig_ready = verity->root_hash_sig || have_verity_sig_partition;
8ee9615e 1478
3dd73ea7 1479 } else if (have_verity_sig_partition) {
8ee9615e
LP		 1480
1481 /* If we found an embedded signature partition, we are ready, too. */
1482
1483 m->verity_ready = m->verity_sig_ready = true;
1484 m->partitions[verity->designator == PARTITION_USR ? PARTITION_USR : PARTITION_ROOT].rw = false;
aee36b4e 1485 }
4623e8e6
LP		 1486 }
1487
cd22d856
LP		 1488 /* After we discovered all partitions let's see if the verity requirements match the policy. (Note:
1489 * we don't check encryption requirements here, because we haven't probed the file system yet, hence
1490 * don't know if this is encrypted or not) */
1491 for (PartitionDesignator di = 0; di < _PARTITION_DESIGNATOR_MAX; di++) {
1492 PartitionDesignator vi, si;
1493 PartitionPolicyFlags found_flags;
1494
1495 vi = partition_verity_of(di);
1496 si = partition_verity_sig_of(di);
1497
1498 /* Determine the verity protection level for this partition. */
1499 found_flags = m->partitions[di].found ?
1500 (vi >= 0 && m->partitions[vi].found ?
1501 (si >= 0 && m->partitions[si].found ? PARTITION_POLICY_SIGNED : PARTITION_POLICY_VERITY) :
1502 PARTITION_POLICY_ENCRYPTED|PARTITION_POLICY_UNPROTECTED) :
1503 (m->partitions[di].ignored ? PARTITION_POLICY_UNUSED : PARTITION_POLICY_ABSENT);
1504
1505 r = image_policy_check_protection(policy, di, found_flags);
1506 if (r < 0)
1507 return r;
1508
1509 if (m->partitions[di].found) {
1510 r = image_policy_check_partition_flags(policy, di, m->partitions[di].gpt_flags);
1511 if (r < 0)
1512 return r;
1513 }
1514 }
1515
1516 r = dissected_image_probe_filesystems(m, fd, policy);
c80c9079
LP		 1517 if (r < 0)
1518 return r;
1519
08f14be4
YW		 1520 return 0;
1521}
1522#endif
18b5886e 1523
08f14be4
YW		 1524int dissect_image_file(
1525 const char *path,
1526 const VeritySettings *verity,
1527 const MountOptions *mount_options,
84be0c71 1528 const ImagePolicy *image_policy,
08f14be4
YW		 1529 DissectImageFlags flags,
1530 DissectedImage **ret) {
8c1be37e 1531
08f14be4
YW		 1532#if HAVE_BLKID
1533 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
254d1313 1534 _cleanup_close_ int fd = -EBADF;
08f14be4 1535 int r;
8c1be37e 1536
08f14be4 1537 assert(path);
08f14be4 1538 assert(ret);
8c1be37e 1539
08f14be4
YW		 1540 fd = open(path, O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY);
1541 if (fd < 0)
1542 return -errno;
896f937f 1543
08f14be4
YW		 1544 r = fd_verify_regular(fd);
1545 if (r < 0)
1546 return r;
de98f631 1547
08f14be4
YW		 1548 r = dissected_image_new(path, &m);
1549 if (r < 0)
1550 return r;
1551
22ee78a8
LP		 1552 r = probe_sector_size(fd, &m->sector_size);
1553 if (r < 0)
1554 return r;
1555
84be0c71 1556 r = dissect_image(m, fd, path, verity, mount_options, image_policy, flags);
698bb074
YW		 1557 if (r < 0)
1558 return r;
8c1be37e 1559
1cc6c93a 1560 *ret = TAKE_PTR(m);
8c1be37e
LP		 1561 return 0;
1562#else
1563 return -EOPNOTSUPP;
1564#endif
1565}
1566
1567DissectedImage* dissected_image_unref(DissectedImage *m) {
8c1be37e
LP		 1568 if (!m)
1569 return NULL;
1570
ac1e1b5f 1571 /* First, clear dissected partitions. */
08fe0a53 1572 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++)
234c2e16 1573 dissected_partition_done(m->partitions + i);
8c1be37e 1574
ac1e1b5f
YW		 1575 /* Second, free decrypted images. This must be after dissected_partition_done(), as freeing
1576 * DecryptedImage may try to deactivate partitions. */
1577 decrypted_image_unref(m->decrypted_image);
1578
1e63dc4f
YW		 1579 /* Third, unref LoopDevice. This must be called after the above two, as freeing LoopDevice may try to
1580 * remove existing partitions on the loopback block device. */
1581 loop_device_unref(m->loop);
1582
593fe6c0 1583 free(m->image_name);
3b925504
LP		 1584 free(m->hostname);
1585 strv_free(m->machine_info);
1586 strv_free(m->os_release);
fab22946 1587 strv_free(m->initrd_release);
7718ac97 1588 strv_free(m->extension_release);
3b925504 1589
5fecf46d 1590 return mfree(m);
8c1be37e
LP		 1591}
1592
18b5886e 1593static int is_loop_device(const char *path) {
553e15f2 1594 char s[SYS_BLOCK_PATH_MAX("/../loop/")];
18b5886e
LP		 1595 struct stat st;
1596
1597 assert(path);
1598
1599 if (stat(path, &st) < 0)
1600 return -errno;
1601
1602 if (!S_ISBLK(st.st_mode))
1603 return -ENOTBLK;
1604
553e15f2 1605 xsprintf_sys_block_path(s, "/loop/", st.st_dev);
18b5886e
LP		 1606 if (access(s, F_OK) < 0) {
1607 if (errno != ENOENT)
1608 return -errno;
1609
1610 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
553e15f2 1611 xsprintf_sys_block_path(s, "/../loop/", st.st_dev);
18b5886e
LP		 1612 if (access(s, F_OK) < 0)
1613 return errno == ENOENT ? false : -errno;
1614 }
1615
1616 return true;
1617}
1618
f8ab7812 1619static int run_fsck(int node_fd, const char *fstype) {
cf32c486
LP		 1620 int r, exit_status;
1621 pid_t pid;
1622
f8ab7812 1623 assert(node_fd >= 0);
cf32c486
LP		 1624 assert(fstype);
1625
13556724 1626 r = fsck_exists_for_fstype(fstype);
cf32c486
LP		 1627 if (r < 0) {
1628 log_debug_errno(r, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype);
1629 return 0;
1630 }
1631 if (r == 0) {
f8ab7812 1632 log_debug("Not checking partition %s, as fsck for %s does not exist.", FORMAT_PROC_FD_PATH(node_fd), fstype);
cf32c486
LP		 1633 return 0;
1634 }
1635
f8ab7812
LP		 1636 r = safe_fork_full(
1637 "(fsck)",
911f8f01 1638 NULL,
f8ab7812 1639 &node_fd, 1, /* Leave the node fd open */
911f8f01 1640 FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG|FORK_REARRANGE_STDIO|FORK_CLOEXEC_OFF,
f8ab7812 1641 &pid);
cf32c486
LP		 1642 if (r < 0)
1643 return log_debug_errno(r, "Failed to fork off fsck: %m");
1644 if (r == 0) {
1645 /* Child */
f8ab7812 1646 execl("/sbin/fsck", "/sbin/fsck", "-aT", FORMAT_PROC_FD_PATH(node_fd), NULL);
7e0ed2e9 1647 log_open();
cf32c486
LP		 1648 log_debug_errno(errno, "Failed to execl() fsck: %m");
1649 _exit(FSCK_OPERATIONAL_ERROR);
1650 }
1651
1652 exit_status = wait_for_terminate_and_check("fsck", pid, 0);
1653 if (exit_status < 0)
1654 return log_debug_errno(exit_status, "Failed to fork off /sbin/fsck: %m");
1655
1656 if ((exit_status & ~FSCK_ERROR_CORRECTED) != FSCK_SUCCESS) {
1657 log_debug("fsck failed with exit status %i.", exit_status);
1658
1659 if ((exit_status & (FSCK_SYSTEM_SHOULD_REBOOT|FSCK_ERRORS_LEFT_UNCORRECTED)) != 0)
1660 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN), "File system is corrupted, refusing.");
1661
1662 log_debug("Ignoring fsck error.");
1663 }
1664
1665 return 0;
1666}
1667
81939d9d 1668static int fs_grow(const char *node_path, const char *mount_path) {
254d1313 1669 _cleanup_close_ int mount_fd = -EBADF, node_fd = -EBADF;
81939d9d
LP		 1670 uint64_t size, newsize;
1671 int r;
1672
1673 node_fd = open(node_path, O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY);
1674 if (node_fd < 0)
1675 return log_debug_errno(errno, "Failed to open node device %s: %m", node_path);
1676
1677 if (ioctl(node_fd, BLKGETSIZE64, &size) != 0)
1678 return log_debug_errno(errno, "Failed to get block device size of %s: %m", node_path);
1679
1680 mount_fd = open(mount_path, O_RDONLY|O_DIRECTORY|O_CLOEXEC);
1681 if (mount_fd < 0)
1682 return log_debug_errno(errno, "Failed to open mountd file system %s: %m", mount_path);
1683
1684 log_debug("Resizing \"%s\" to %"PRIu64" bytes...", mount_path, size);
1685 r = resize_fs(mount_fd, size, &newsize);
1686 if (r < 0)
1687 return log_debug_errno(r, "Failed to resize \"%s\" to %"PRIu64" bytes: %m", mount_path, size);
1688
1689 if (newsize == size)
1690 log_debug("Successfully resized \"%s\" to %s bytes.",
2b59bf51 1691 mount_path, FORMAT_BYTES(newsize));
81939d9d
LP		 1692 else {
1693 assert(newsize < size);
1694 log_debug("Successfully resized \"%s\" to %s bytes (%"PRIu64" bytes lost due to blocksize).",
2b59bf51 1695 mount_path, FORMAT_BYTES(newsize), size - newsize);
81939d9d
LP		 1696 }
1697
1698 return 0;
1699}
1700
254e392e
LP		 1701int partition_pick_mount_options(
1702 PartitionDesignator d,
1703 const char *fstype,
1704 bool rw,
1705 bool discard,
1706 char **ret_options,
1707 unsigned long *ret_ms_flags) {
1708
1709 _cleanup_free_ char *options = NULL;
1710
1711 assert(ret_options);
1712
1713 /* Selects a baseline of bind mount flags, that should always apply.
1714 *
1715 * Firstly, we set MS_NODEV universally on all mounts, since we don't want to allow device nodes outside of /dev/.
1716 *
1717 * On /var/tmp/ we'll also set MS_NOSUID, same as we set for /tmp/ on the host.
1718 *
1719 * On the ESP and XBOOTLDR partitions we'll also disable symlinks, and execution. These file systems
1720 * are generally untrusted (i.e. not encrypted or authenticated), and typically VFAT hence we should
1721 * be as restrictive as possible, and this shouldn't hurt, since the functionality is not available
1722 * there anyway. */
1723
1724 unsigned long flags = MS_NODEV;
1725
1726 if (!rw)
1727 flags |= MS_RDONLY;
1728
1729 switch (d) {
1730
1731 case PARTITION_ESP:
1732 case PARTITION_XBOOTLDR:
1733 flags |= MS_NOSUID|MS_NOEXEC|ms_nosymfollow_supported();
1734
6eda6f7e
LP		 1735 /* The ESP might contain a pre-boot random seed. Let's make this unaccessible to regular
1736 * userspace. ESP/XBOOTLDR is almost certainly VFAT, hence if we don't know assume it is. */
1737 if (!fstype || fstype_can_umask(fstype))
254e392e
LP		 1738 if (!strextend_with_separator(&options, ",", "umask=0077"))
1739 return -ENOMEM;
1740 break;
1741
1742 case PARTITION_TMP:
1743 flags |= MS_NOSUID;
1744 break;
1745
1746 default:
1747 break;
1748 }
1749
1750 /* So, when you request MS_RDONLY from ext4, then this means nothing. It happily still writes to the
1751 * backing storage. What's worse, the BLKRO[GS]ET flag and (in case of loopback devices)
1752 * LO_FLAGS_READ_ONLY don't mean anything, they affect userspace accesses only, and write accesses
1753 * from the upper file system still get propagated through to the underlying file system,
1754 * unrestricted. To actually get ext4/xfs/btrfs to stop writing to the device we need to specify
1755 * "norecovery" as mount option, in addition to MS_RDONLY. Yes, this sucks, since it means we need to
1756 * carry a per file system table here.
1757 *
1758 * Note that this means that we might not be able to mount corrupted file systems as read-only
1759 * anymore (since in some cases the kernel implementations will refuse mounting when corrupted,
1760 * read-only and "norecovery" is specified). But I think for the case of automatically determined
1761 * mount options for loopback devices this is the right choice, since otherwise using the same
1762 * loopback file twice even in read-only mode, is going to fail badly sooner or later. The usecase of
1763 * making reuse of the immutable images "just work" is more relevant to us than having read-only
1764 * access that actually modifies stuff work on such image files. Or to say this differently: if
1765 * people want their file systems to be fixed up they should just open them in writable mode, where
1766 * all these problems don't exist. */
034ebc47 1767 if (!rw && fstype && fstype_can_norecovery(fstype))
254e392e
LP		 1768 if (!strextend_with_separator(&options, ",", "norecovery"))
1769 return -ENOMEM;
1770
1771 if (discard && fstype && fstype_can_discard(fstype))
1772 if (!strextend_with_separator(&options, ",", "discard"))
1773 return -ENOMEM;
1774
1775 if (!ret_ms_flags) /* Fold flags into option string if ret_flags specified as NULL */
1776 if (!strextend_with_separator(&options, ",",
1777 FLAGS_SET(flags, MS_RDONLY) ? "ro" : "rw",
1778 FLAGS_SET(flags, MS_NODEV) ? "nodev" : "dev",
1779 FLAGS_SET(flags, MS_NOSUID) ? "nosuid" : "suid",
1780 FLAGS_SET(flags, MS_NOEXEC) ? "noexec" : "exec",
1781 FLAGS_SET(flags, MS_NOSYMFOLLOW) ? "nosymfollow" : NULL))
1782 /* NB: we suppress 'symfollow' here, since it's the default, and old /bin/mount might not know it */
1783 return -ENOMEM;
1784
1785 if (ret_ms_flags)
1786 *ret_ms_flags = flags;
1787
1788 *ret_options = TAKE_PTR(options);
1789 return 0;
1790}
1791
18b5886e 1792static int mount_partition(
254e392e 1793 PartitionDesignator d,
18b5886e
LP		 1794 DissectedPartition *m,
1795 const char *where,
1796 const char *directory,
2d3a5a73 1797 uid_t uid_shift,
21b61b1d 1798 uid_t uid_range,
18b5886e
LP		 1799 DissectImageFlags flags) {
1800
2d3a5a73 1801 _cleanup_free_ char *chased = NULL, *options = NULL;
254e392e 1802 bool rw, discard, remap_uid_gid = false;
2d3a5a73 1803 const char *p, *node, *fstype;
254e392e 1804 unsigned long ms_flags;
2eedfd2d 1805 int r;
8c1be37e
LP		 1806
1807 assert(m);
1808 assert(where);
1809
f7725647
YW		 1810 if (m->mount_node_fd < 0)
1811 return 0;
1812
4dc28665 1813 /* Use decrypted node and matching fstype if available, otherwise use the original device */
f7725647 1814 node = FORMAT_PROC_FD_PATH(m->mount_node_fd);
c571dee7 1815 fstype = dissected_partition_fstype(m);
18b5886e 1816
4dc28665
LP		 1817 if (!fstype)
1818 return -EAFNOSUPPORT;
80ce8580
LP		 1819 r = dissect_fstype_ok(fstype);
1820 if (r < 0)
1821 return r;
1822 if (!r)
1823 return -EIDRM; /* Recognizable error */
8c1be37e 1824
68ac5118
ZJS		 1825 /* We are looking at an encrypted partition? This either means stacked encryption, or the caller
1826 * didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error for this
1827 * case. */
4dc28665 1828 if (streq(fstype, "crypto_LUKS"))
fa45d12c 1829 return -EUNATCH;
18b5886e 1830
ef9c184d 1831 rw = m->rw && !(flags & DISSECT_IMAGE_MOUNT_READ_ONLY);
8c1be37e 1832
254e392e
LP		 1833 discard = ((flags & DISSECT_IMAGE_DISCARD) ||
1834 ((flags & DISSECT_IMAGE_DISCARD_ON_LOOP) && is_loop_device(m->node) > 0));
1835
cf32c486 1836 if (FLAGS_SET(flags, DISSECT_IMAGE_FSCK) && rw) {
f8ab7812 1837 r = run_fsck(m->mount_node_fd, fstype);
cf32c486
LP		 1838 if (r < 0)
1839 return r;
1840 }
1841
2eedfd2d 1842 if (directory) {
334eb5b0
LP		 1843 /* Automatically create missing mount points inside the image, if necessary. */
1844 r = mkdir_p_root(where, directory, uid_shift, (gid_t) uid_shift, 0755);
1845 if (r < 0 && r != -EROFS)
1846 return r;
1f0f82f1 1847
f461a28d 1848 r = chase(directory, where, CHASE_PREFIX_ROOT, &chased, NULL);
2eedfd2d
LP		 1849 if (r < 0)
1850 return r;
1851
1852 p = chased;
9842905e
LP		 1853 } else {
1854 /* Create top-level mount if missing – but only if this is asked for. This won't modify the
1855 * image (as the branch above does) but the host hierarchy, and the created directory might
1856 * survive our mount in the host hierarchy hence. */
1857 if (FLAGS_SET(flags, DISSECT_IMAGE_MKDIR)) {
1858 r = mkdir_p(where, 0755);
1859 if (r < 0)
1860 return r;
1861 }
1862
8c1be37e 1863 p = where;
9842905e 1864 }
8c1be37e 1865
254e392e
LP		 1866 r = partition_pick_mount_options(d, dissected_partition_fstype(m), rw, discard, &options, &ms_flags);
1867 if (r < 0)
1868 return r;
2d3a5a73 1869
21b61b1d 1870 if (uid_is_valid(uid_shift) && uid_shift != 0) {
2d3a5a73 1871
21b61b1d
LP		 1872 if (fstype_can_uid_gid(fstype)) {
1873 _cleanup_free_ char *uid_option = NULL;
2d3a5a73 1874
21b61b1d
LP		 1875 if (asprintf(&uid_option, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0)
1876 return -ENOMEM;
1877
1878 if (!strextend_with_separator(&options, ",", uid_option))
1879 return -ENOMEM;
1880 } else if (FLAGS_SET(flags, DISSECT_IMAGE_MOUNT_IDMAPPED))
1881 remap_uid_gid = true;
2d3a5a73 1882 }
8c1be37e 1883
18d73705 1884 if (!isempty(m->mount_options))
c2bc710b 1885 if (!strextend_with_separator(&options, ",", m->mount_options))
18d73705
LB		 1886 return -ENOMEM;
1887
254e392e 1888 r = mount_nofollow_verbose(LOG_DEBUG, node, p, fstype, ms_flags, options);
d9223c07
LP		 1889 if (r < 0)
1890 return r;
1891
81939d9d
LP		 1892 if (rw && m->growfs && FLAGS_SET(flags, DISSECT_IMAGE_GROWFS))
1893 (void) fs_grow(node, p);
1894
21b61b1d 1895 if (remap_uid_gid) {
2b2777ed 1896 r = remount_idmap(p, uid_shift, uid_range, UID_INVALID, REMOUNT_IDMAPPING_HOST_ROOT);
21b61b1d
LP		 1897 if (r < 0)
1898 return r;
1899 }
1900
d9223c07 1901 return 1;
8c1be37e
LP		 1902}
1903
7cf66030
LP		 1904static int mount_root_tmpfs(const char *where, uid_t uid_shift, DissectImageFlags flags) {
1905 _cleanup_free_ char *options = NULL;
1906 int r;
1907
1908 assert(where);
1909
1910 /* For images that contain /usr/ but no rootfs, let's mount rootfs as tmpfs */
1911
1912 if (FLAGS_SET(flags, DISSECT_IMAGE_MKDIR)) {
1913 r = mkdir_p(where, 0755);
1914 if (r < 0)
1915 return r;
1916 }
1917
1918 if (uid_is_valid(uid_shift)) {
1919 if (asprintf(&options, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0)
1920 return -ENOMEM;
1921 }
1922
1923 r = mount_nofollow_verbose(LOG_DEBUG, "rootfs", where, "tmpfs", MS_NODEV, options);
1924 if (r < 0)
1925 return r;
1926
1927 return 1;
1928}
1929
21b61b1d
LP		 1930int dissected_image_mount(
1931 DissectedImage *m,
1932 const char *where,
1933 uid_t uid_shift,
1934 uid_t uid_range,
1935 DissectImageFlags flags) {
1936
1f0f82f1 1937 int r, xbootldr_mounted;
8c1be37e
LP		 1938
1939 assert(m);
1940 assert(where);
1941
fa45d12c
LP		 1942 /* Returns:
1943 *
1944 * -ENXIO → No root partition found
7718ac97 1945 * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release/extension-release file found
fa45d12c
LP		 1946 * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet
1947 * -EUCLEAN → fsck for file system failed
1948 * -EBUSY → File system already mounted/used elsewhere (kernel)
4dc28665 1949 * -EAFNOSUPPORT → File system type not supported or not known
80ce8580 1950 * -EIDRM → File system is not among allowlisted "common" file systems
fa45d12c
LP		 1951 */
1952
7cf66030
LP		 1953 if (!(m->partitions[PARTITION_ROOT].found ||
1954 (m->partitions[PARTITION_USR].found && FLAGS_SET(flags, DISSECT_IMAGE_USR_NO_ROOT))))
1955 return -ENXIO; /* Require a root fs or at least a /usr/ fs (the latter is subject to a flag of its own) */
8c1be37e 1956
2d3a5a73 1957 if ((flags & DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY) == 0) {
7cf66030
LP		 1958
1959 /* First mount the root fs. If there's none we use a tmpfs. */
1960 if (m->partitions[PARTITION_ROOT].found)
254e392e 1961 r = mount_partition(PARTITION_ROOT, m->partitions + PARTITION_ROOT, where, NULL, uid_shift, uid_range, flags);
7cf66030
LP		 1962 else
1963 r = mount_root_tmpfs(where, uid_shift, flags);
2d3a5a73
LP		 1964 if (r < 0)
1965 return r;
aee36b4e 1966
aee36b4e 1967 /* For us mounting root always means mounting /usr as well */
254e392e 1968 r = mount_partition(PARTITION_USR, m->partitions + PARTITION_USR, where, "/usr", uid_shift, uid_range, flags);
aee36b4e
LP		 1969 if (r < 0)
1970 return r;
03bcb6d4 1971
9ccb531a
LB		 1972 if ((flags & (DISSECT_IMAGE_VALIDATE_OS|DISSECT_IMAGE_VALIDATE_OS_EXT)) != 0) {
1973 /* If either one of the validation flags are set, ensure that the image qualifies
1974 * as one or the other (or both). */
1975 bool ok = false;
1976
1977 if (FLAGS_SET(flags, DISSECT_IMAGE_VALIDATE_OS)) {
1978 r = path_is_os_tree(where);
1979 if (r < 0)
1980 return r;
1981 if (r > 0)
1982 ok = true;
1983 }
1984 if (!ok && FLAGS_SET(flags, DISSECT_IMAGE_VALIDATE_OS_EXT)) {
6afa5d86 1985 r = extension_has_forbidden_content(where);
7718ac97
LB		 1986 if (r < 0)
1987 return r;
6afa5d86
LB		 1988 if (r == 0) {
1989 r = path_is_extension_tree(where, m->image_name, FLAGS_SET(flags, DISSECT_IMAGE_RELAX_SYSEXT_CHECK));
1990 if (r < 0)
1991 return r;
1992 if (r > 0)
1993 ok = true;
1994 }
7718ac97 1995 }
9ccb531a
LB		 1996
1997 if (!ok)
1998 return -ENOMEDIUM;
03bcb6d4 1999 }
2d3a5a73
LP		 2000 }
2001
705727fd 2002 if (flags & DISSECT_IMAGE_MOUNT_ROOT_ONLY)
2d3a5a73 2003 return 0;
8c1be37e 2004
254e392e 2005 r = mount_partition(PARTITION_HOME, m->partitions + PARTITION_HOME, where, "/home", uid_shift, uid_range, flags);
8c1be37e
LP		 2006 if (r < 0)
2007 return r;
2008
254e392e 2009 r = mount_partition(PARTITION_SRV, m->partitions + PARTITION_SRV, where, "/srv", uid_shift, uid_range, flags);
8c1be37e
LP		 2010 if (r < 0)
2011 return r;
2012
254e392e 2013 r = mount_partition(PARTITION_VAR, m->partitions + PARTITION_VAR, where, "/var", uid_shift, uid_range, flags);
d4dffb85
LP		 2014 if (r < 0)
2015 return r;
2016
254e392e 2017 r = mount_partition(PARTITION_TMP, m->partitions + PARTITION_TMP, where, "/var/tmp", uid_shift, uid_range, flags);
d4dffb85
LP		 2018 if (r < 0)
2019 return r;
2020
254e392e 2021 xbootldr_mounted = mount_partition(PARTITION_XBOOTLDR, m->partitions + PARTITION_XBOOTLDR, where, "/boot", uid_shift, uid_range, flags);
1f0f82f1
LP		 2022 if (xbootldr_mounted < 0)
2023 return xbootldr_mounted;
d9223c07 2024
8c1be37e 2025 if (m->partitions[PARTITION_ESP].found) {
1f0f82f1
LP		 2026 int esp_done = false;
2027
d9223c07
LP		 2028 /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it
2029 * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */
8c1be37e 2030
f461a28d 2031 r = chase("/efi", where, CHASE_PREFIX_ROOT, NULL, NULL);
1f0f82f1
LP		 2032 if (r < 0) {
2033 if (r != -ENOENT)
d9223c07 2034 return r;
8c1be37e 2035
1f0f82f1
LP		 2036 /* /efi doesn't exist. Let's see if /boot is suitable then */
2037
2038 if (!xbootldr_mounted) {
2039 _cleanup_free_ char *p = NULL;
2eedfd2d 2040
f461a28d 2041 r = chase("/boot", where, CHASE_PREFIX_ROOT, &p, NULL);
1f0f82f1
LP		 2042 if (r < 0) {
2043 if (r != -ENOENT)
2044 return r;
db55bbf2 2045 } else if (dir_is_empty(p, /* ignore_hidden_or_backup= */ false) > 0) {
1f0f82f1 2046 /* It exists and is an empty directory. Let's mount the ESP there. */
254e392e 2047 r = mount_partition(PARTITION_ESP, m->partitions + PARTITION_ESP, where, "/boot", uid_shift, uid_range, flags);
1f0f82f1
LP		 2048 if (r < 0)
2049 return r;
2050
2051 esp_done = true;
2052 }
2eedfd2d 2053 }
8c1be37e 2054 }
1f0f82f1
LP		 2055
2056 if (!esp_done) {
2057 /* OK, let's mount the ESP now to /efi (possibly creating the dir if missing) */
2058
254e392e 2059 r = mount_partition(PARTITION_ESP, m->partitions + PARTITION_ESP, where, "/efi", uid_shift, uid_range, flags);
1f0f82f1
LP		 2060 if (r < 0)
2061 return r;
2062 }
8c1be37e
LP		 2063 }
2064
2065 return 0;
2066}
2067
21b61b1d
LP		 2068int dissected_image_mount_and_warn(
2069 DissectedImage *m,
2070 const char *where,
2071 uid_t uid_shift,
2072 uid_t uid_range,
2073 DissectImageFlags flags) {
2074
af187ab2
LP		 2075 int r;
2076
2077 assert(m);
2078 assert(where);
2079
21b61b1d 2080 r = dissected_image_mount(m, where, uid_shift, uid_range, flags);
af187ab2
LP		 2081 if (r == -ENXIO)
2082 return log_error_errno(r, "Not root file system found in image.");
2083 if (r == -EMEDIUMTYPE)
7718ac97 2084 return log_error_errno(r, "No suitable os-release/extension-release file in image found.");
af187ab2
LP		 2085 if (r == -EUNATCH)
2086 return log_error_errno(r, "Encrypted file system discovered, but decryption not requested.");
2087 if (r == -EUCLEAN)
2088 return log_error_errno(r, "File system check on image failed.");
2089 if (r == -EBUSY)
2090 return log_error_errno(r, "File system already mounted elsewhere.");
4dc28665
LP		 2091 if (r == -EAFNOSUPPORT)
2092 return log_error_errno(r, "File system type not supported or not known.");
80ce8580
LP		 2093 if (r == -EIDRM)
2094 return log_error_errno(r, "File system is too uncommon, refused.");
af187ab2
LP		 2095 if (r < 0)
2096 return log_error_errno(r, "Failed to mount image: %m");
2097
2098 return r;
2099}
2100
349cc4a5 2101#if HAVE_LIBCRYPTSETUP
9321ad51 2102struct DecryptedPartition {
18b5886e
LP		 2103 struct crypt_device *device;
2104 char *name;
2105 bool relinquished;
9321ad51
YW		 2106};
2107#endif
2108
2109typedef struct DecryptedPartition DecryptedPartition;
18b5886e
LP		 2110
2111struct DecryptedImage {
9321ad51 2112 unsigned n_ref;
18b5886e
LP		 2113 DecryptedPartition *decrypted;
2114 size_t n_decrypted;
18b5886e 2115};
18b5886e 2116
9321ad51 2117static DecryptedImage* decrypted_image_free(DecryptedImage *d) {
349cc4a5 2118#if HAVE_LIBCRYPTSETUP
18b5886e
LP		 2119 int r;
2120
2121 if (!d)
2122 return NULL;
2123
67f63ee5 2124 for (size_t i = 0; i < d->n_decrypted; i++) {
18b5886e
LP		 2125 DecryptedPartition *p = d->decrypted + i;
2126
2127 if (p->device && p->name && !p->relinquished) {
5228c58e
DDM		 2128 _cleanup_free_ char *node = NULL;
2129
2130 node = path_join("/dev/mapper", p->name);
2131 if (node) {
2132 r = btrfs_forget_device(node);
2133 if (r < 0 && r != -ENOENT)
2134 log_debug_errno(r, "Failed to forget btrfs device %s, ignoring: %m", node);
2135 } else
2136 log_oom_debug();
2137
ea16d7f4
YW		 2138 /* Let's deactivate lazily, as the dm volume may be already/still used by other processes. */
2139 r = sym_crypt_deactivate_by_name(p->device, p->name, CRYPT_DEACTIVATE_DEFERRED);
18b5886e
LP		 2140 if (r < 0)
2141 log_debug_errno(r, "Failed to deactivate encrypted partition %s", p->name);
2142 }
2143
2144 if (p->device)
0d12936d 2145 sym_crypt_free(p->device);
18b5886e
LP		 2146 free(p->name);
2147 }
2148
f91861e4 2149 free(d->decrypted);
18b5886e
LP		 2150 free(d);
2151#endif
2152 return NULL;
2153}
2154
9321ad51
YW		 2155DEFINE_TRIVIAL_REF_UNREF_FUNC(DecryptedImage, decrypted_image, decrypted_image_free);
2156
349cc4a5 2157#if HAVE_LIBCRYPTSETUP
9321ad51
YW		 2158static int decrypted_image_new(DecryptedImage **ret) {
2159 _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL;
2160
2161 assert(ret);
2162
2163 d = new(DecryptedImage, 1);
2164 if (!d)
2165 return -ENOMEM;
2166
2167 *d = (DecryptedImage) {
2168 .n_ref = 1,
2169 };
2170
2171 *ret = TAKE_PTR(d);
2172 return 0;
2173}
4623e8e6
LP		 2174
2175static int make_dm_name_and_node(const void *original_node, const char *suffix, char **ret_name, char **ret_node) {
2176 _cleanup_free_ char *name = NULL, *node = NULL;
2177 const char *base;
2178
2179 assert(original_node);
2180 assert(suffix);
2181 assert(ret_name);
2182 assert(ret_node);
2183
2184 base = strrchr(original_node, '/');
2185 if (!base)
ac1f3ad0
LB		 2186 base = original_node;
2187 else
2188 base++;
4623e8e6
LP		 2189 if (isempty(base))
2190 return -EINVAL;
2191
2192 name = strjoin(base, suffix);
2193 if (!name)
2194 return -ENOMEM;
2195 if (!filename_is_valid(name))
2196 return -EINVAL;
2197
0d12936d 2198 node = path_join(sym_crypt_get_dir(), name);
4623e8e6
LP		 2199 if (!node)
2200 return -ENOMEM;
2201
1cc6c93a
YW		 2202 *ret_name = TAKE_PTR(name);
2203 *ret_node = TAKE_PTR(node);
4623e8e6 2204
4623e8e6
LP		 2205 return 0;
2206}
2207
18b5886e
LP		 2208static int decrypt_partition(
2209 DissectedPartition *m,
2210 const char *passphrase,
2211 DissectImageFlags flags,
2212 DecryptedImage *d) {
2213
2214 _cleanup_free_ char *node = NULL, *name = NULL;
0d12936d 2215 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
254d1313 2216 _cleanup_close_ int fd = -EBADF;
18b5886e
LP		 2217 int r;
2218
2219 assert(m);
2220 assert(d);
2221
2222 if (!m->found || !m->node || !m->fstype)
2223 return 0;
2224
2225 if (!streq(m->fstype, "crypto_LUKS"))
2226 return 0;
2227
bdd73ac5
ZJS		 2228 if (!passphrase)
2229 return -ENOKEY;
2230
0d12936d
LP		 2231 r = dlopen_cryptsetup();
2232 if (r < 0)
2233 return r;
2234
4623e8e6
LP		 2235 r = make_dm_name_and_node(m->node, "-decrypted", &name, &node);
2236 if (r < 0)
2237 return r;
18b5886e 2238
319a4f4b 2239 if (!GREEDY_REALLOC0(d->decrypted, d->n_decrypted + 1))
18b5886e
LP		 2240 return -ENOMEM;
2241
0d12936d 2242 r = sym_crypt_init(&cd, m->node);
18b5886e 2243 if (r < 0)
715cbb81 2244 return log_debug_errno(r, "Failed to initialize dm-crypt: %m");
18b5886e 2245
efc3b12f 2246 cryptsetup_enable_logging(cd);
1887032f 2247
0d12936d 2248 r = sym_crypt_load(cd, CRYPT_LUKS, NULL);
294bd454
ZJS		 2249 if (r < 0)
2250 return log_debug_errno(r, "Failed to load LUKS metadata: %m");
18b5886e 2251
0d12936d 2252 r = sym_crypt_activate_by_passphrase(cd, name, CRYPT_ANY_SLOT, passphrase, strlen(passphrase),
ef9c184d 2253 ((flags & DISSECT_IMAGE_DEVICE_READ_ONLY) ? CRYPT_ACTIVATE_READONLY : 0) |
0d12936d 2254 ((flags & DISSECT_IMAGE_DISCARD_ON_CRYPTO) ? CRYPT_ACTIVATE_ALLOW_DISCARDS : 0));
294bd454 2255 if (r < 0) {
715cbb81 2256 log_debug_errno(r, "Failed to activate LUKS device: %m");
294bd454 2257 return r == -EPERM ? -EKEYREJECTED : r;
18b5886e 2258 }
18b5886e 2259
f7725647
YW		 2260 fd = open(node, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
2261 if (fd < 0)
2262 return log_debug_errno(errno, "Failed to open %s: %m", node);
2263
94344385
LP		 2264 d->decrypted[d->n_decrypted++] = (DecryptedPartition) {
2265 .name = TAKE_PTR(name),
2266 .device = TAKE_PTR(cd),
2267 };
18b5886e 2268
1cc6c93a 2269 m->decrypted_node = TAKE_PTR(node);
f7725647 2270 close_and_replace(m->mount_node_fd, fd);
18b5886e
LP		 2271
2272 return 0;
4623e8e6
LP		 2273}
2274
89e62e0b
LP		 2275static int verity_can_reuse(
2276 const VeritySettings *verity,
2277 const char *name,
2278 struct crypt_device **ret_cd) {
2279
ac1f3ad0
LB		 2280 /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
2281 _cleanup_free_ char *root_hash_existing = NULL;
0d12936d 2282 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
ac1f3ad0 2283 struct crypt_params_verity crypt_params = {};
89e62e0b 2284 size_t root_hash_existing_size;
ac1f3ad0
LB		 2285 int r;
2286
89e62e0b
LP		 2287 assert(verity);
2288 assert(name);
ac1f3ad0
LB		 2289 assert(ret_cd);
2290
0d12936d 2291 r = sym_crypt_init_by_name(&cd, name);
ac1f3ad0
LB		 2292 if (r < 0)
2293 return log_debug_errno(r, "Error opening verity device, crypt_init_by_name failed: %m");
0d12936d 2294
c719805e
LP		 2295 cryptsetup_enable_logging(cd);
2296
0d12936d 2297 r = sym_crypt_get_verity_info(cd, &crypt_params);
ac1f3ad0
LB		 2298 if (r < 0)
2299 return log_debug_errno(r, "Error opening verity device, crypt_get_verity_info failed: %m");
0d12936d 2300
89e62e0b
LP		 2301 root_hash_existing_size = verity->root_hash_size;
2302 root_hash_existing = malloc0(root_hash_existing_size);
ac1f3ad0
LB		 2303 if (!root_hash_existing)
2304 return -ENOMEM;
0d12936d
LP		 2305
2306 r = sym_crypt_volume_key_get(cd, CRYPT_ANY_SLOT, root_hash_existing, &root_hash_existing_size, NULL, 0);
ac1f3ad0
LB		 2307 if (r < 0)
2308 return log_debug_errno(r, "Error opening verity device, crypt_volume_key_get failed: %m");
89e62e0b
LP		 2309 if (verity->root_hash_size != root_hash_existing_size ||
2310 memcmp(root_hash_existing, verity->root_hash, verity->root_hash_size) != 0)
ac1f3ad0 2311 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but root hashes are different.");
89e62e0b 2312
ac1f3ad0 2313#if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
89e62e0b
LP		 2314 /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the
2315 * same settings, so that a previous unsigned mount will not be reused if the user asks to use
28423d9a 2316 * signing for the new one, and vice versa. */
89e62e0b 2317 if (!!verity->root_hash_sig != !!(crypt_params.flags & CRYPT_VERITY_ROOT_HASH_SIGNATURE))
ac1f3ad0
LB		 2318 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but signature settings are not the same.");
2319#endif
2320
2321 *ret_cd = TAKE_PTR(cd);
2322 return 0;
2323}
2324
75db809a 2325static inline char* dm_deferred_remove_clean(char *name) {
ac1f3ad0 2326 if (!name)
75db809a 2327 return NULL;
0d12936d
LP		 2328
2329 (void) sym_crypt_deactivate_by_name(NULL, name, CRYPT_DEACTIVATE_DEFERRED);
75db809a 2330 return mfree(name);
ac1f3ad0
LB		 2331}
2332DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean);
2333
c2fa92e7
LP		 2334static int validate_signature_userspace(const VeritySettings *verity) {
2335#if HAVE_OPENSSL
2336 _cleanup_(sk_X509_free_allp) STACK_OF(X509) *sk = NULL;
2337 _cleanup_strv_free_ char **certs = NULL;
2338 _cleanup_(PKCS7_freep) PKCS7 *p7 = NULL;
2339 _cleanup_free_ char *s = NULL;
2340 _cleanup_(BIO_freep) BIO *bio = NULL; /* 'bio' must be freed first, 's' second, hence keep this order
2341 * of declaration in place, please */
2342 const unsigned char *d;
c2fa92e7
LP		 2343 int r;
2344
2345 assert(verity);
2346 assert(verity->root_hash);
2347 assert(verity->root_hash_sig);
2348
2349 /* Because installing a signature certificate into the kernel chain is so messy, let's optionally do
2350 * userspace validation. */
2351
2352 r = conf_files_list_nulstr(&certs, ".crt", NULL, CONF_FILES_REGULAR|CONF_FILES_FILTER_MASKED, CONF_PATHS_NULSTR("verity.d"));
2353 if (r < 0)
2354 return log_debug_errno(r, "Failed to enumerate certificates: %m");
2355 if (strv_isempty(certs)) {
2356 log_debug("No userspace dm-verity certificates found.");
2357 return 0;
2358 }
2359
2360 d = verity->root_hash_sig;
2361 p7 = d2i_PKCS7(NULL, &d, (long) verity->root_hash_sig_size);
2362 if (!p7)
2363 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to parse PKCS7 DER signature data.");
2364
2365 s = hexmem(verity->root_hash, verity->root_hash_size);
2366 if (!s)
2367 return log_oom_debug();
2368
2369 bio = BIO_new_mem_buf(s, strlen(s));
2370 if (!bio)
2371 return log_oom_debug();
2372
2373 sk = sk_X509_new_null();
2374 if (!sk)
2375 return log_oom_debug();
2376
2377 STRV_FOREACH(i, certs) {
2378 _cleanup_(X509_freep) X509 *c = NULL;
2379 _cleanup_fclose_ FILE *f = NULL;
2380
2381 f = fopen(*i, "re");
2382 if (!f) {
2383 log_debug_errno(errno, "Failed to open '%s', ignoring: %m", *i);
2384 continue;
2385 }
2386
2387 c = PEM_read_X509(f, NULL, NULL, NULL);
2388 if (!c) {
2389 log_debug("Failed to load X509 certificate '%s', ignoring.", *i);
2390 continue;
2391 }
2392
2393 if (sk_X509_push(sk, c) == 0)
2394 return log_oom_debug();
2395
2396 TAKE_PTR(c);
2397 }
2398
2399 r = PKCS7_verify(p7, sk, NULL, bio, NULL, PKCS7_NOINTERN|PKCS7_NOVERIFY);
2400 if (r)
2401 log_debug("Userspace PKCS#7 validation succeeded.");
2402 else
2403 log_debug("Userspace PKCS#7 validation failed: %s", ERR_error_string(ERR_get_error(), NULL));
2404
2405 return r;
2406#else
2407 log_debug("Not doing client-side validation of dm-verity root hash signatures, OpenSSL support disabled.");
2408 return 0;
2409#endif
2410}
2411
2412static int do_crypt_activate_verity(
2413 struct crypt_device *cd,
2414 const char *name,
2415 const VeritySettings *verity) {
2416
2417 bool check_signature;
2418 int r;
2419
2420 assert(cd);
2421 assert(name);
2422 assert(verity);
2423
2424 if (verity->root_hash_sig) {
2425 r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIGNATURE");
2426 if (r < 0 && r != -ENXIO)
2427 log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIGNATURE");
2428
2429 check_signature = r != 0;
2430 } else
2431 check_signature = false;
2432
2433 if (check_signature) {
2434
2435#if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
2436 /* First, if we have support for signed keys in the kernel, then try that first. */
2437 r = sym_crypt_activate_by_signed_key(
2438 cd,
2439 name,
2440 verity->root_hash,
2441 verity->root_hash_size,
2442 verity->root_hash_sig,
2443 verity->root_hash_sig_size,
2444 CRYPT_ACTIVATE_READONLY);
2445 if (r >= 0)
2446 return r;
2447
2448 log_debug("Validation of dm-verity signature failed via the kernel, trying userspace validation instead.");
2449#else
2450 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.",
2451 program_invocation_short_name);
2452#endif
2453
2454 /* So this didn't work via the kernel, then let's try userspace validation instead. If that
2455 * works we'll try to activate without telling the kernel the signature. */
2456
2457 r = validate_signature_userspace(verity);
2458 if (r < 0)
2459 return r;
2460 if (r == 0)
2461 return log_debug_errno(SYNTHETIC_ERRNO(ENOKEY),
2462 "Activation of signed Verity volume worked neither via the kernel nor in userspace, can't activate.");
2463 }
2464
2465 return sym_crypt_activate_by_volume_key(
2466 cd,
2467 name,
2468 verity->root_hash,
2469 verity->root_hash_size,
2470 CRYPT_ACTIVATE_READONLY);
2471}
2472
ad361a50
YW		 2473static usec_t verity_timeout(void) {
2474 usec_t t = 100 * USEC_PER_MSEC;
2475 const char *e;
2476 int r;
2477
2478 /* On slower machines, like non-KVM vm, setting up device may take a long time.
2479 * Let's make the timeout configurable. */
2480
2481 e = getenv("SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC");
2482 if (!e)
2483 return t;
2484
2485 r = parse_sec(e, &t);
2486 if (r < 0)
2487 log_debug_errno(r,
2488 "Failed to parse timeout specified in $SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC, "
2489 "using the default timeout (%s).",
2490 FORMAT_TIMESPAN(t, USEC_PER_MSEC));
2491
2492 return t;
2493}
2494
4623e8e6 2495static int verity_partition(
aee36b4e 2496 PartitionDesignator designator,
4623e8e6
LP		 2497 DissectedPartition *m,
2498 DissectedPartition *v,
89e62e0b 2499 const VeritySettings *verity,
4623e8e6
LP		 2500 DissectImageFlags flags,
2501 DecryptedImage *d) {
2502
0d12936d 2503 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
ac1f3ad0 2504 _cleanup_(dm_deferred_remove_cleanp) char *restore_deferred_remove = NULL;
89e62e0b 2505 _cleanup_free_ char *node = NULL, *name = NULL;
254d1313 2506 _cleanup_close_ int mount_node_fd = -EBADF;
4623e8e6
LP		 2507 int r;
2508
2509 assert(m);
89e62e0b 2510 assert(v || (verity && verity->data_path));
4623e8e6 2511
89e62e0b 2512 if (!verity || !verity->root_hash)
4623e8e6 2513 return 0;
aee36b4e
LP		 2514 if (!((verity->designator < 0 && designator == PARTITION_ROOT) ||
2515 (verity->designator == designator)))
2516 return 0;
4623e8e6
LP		 2517
2518 if (!m->found || !m->node || !m->fstype)
2519 return 0;
89e62e0b 2520 if (!verity->data_path) {
e7cbe5cb
LB		 2521 if (!v->found || !v->node || !v->fstype)
2522 return 0;
4623e8e6 2523
e7cbe5cb
LB		 2524 if (!streq(v->fstype, "DM_verity_hash"))
2525 return 0;
2526 }
4623e8e6 2527
0d12936d
LP		 2528 r = dlopen_cryptsetup();
2529 if (r < 0)
2530 return r;
2531
ac1f3ad0
LB		 2532 if (FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) {
2533 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
2534 _cleanup_free_ char *root_hash_encoded = NULL;
0d12936d 2535
89e62e0b 2536 root_hash_encoded = hexmem(verity->root_hash, verity->root_hash_size);
ac1f3ad0
LB		 2537 if (!root_hash_encoded)
2538 return -ENOMEM;
aee36b4e 2539
ac1f3ad0
LB		 2540 r = make_dm_name_and_node(root_hash_encoded, "-verity", &name, &node);
2541 } else
2542 r = make_dm_name_and_node(m->node, "-verity", &name, &node);
4623e8e6
LP		 2543 if (r < 0)
2544 return r;
2545
89e62e0b 2546 r = sym_crypt_init(&cd, verity->data_path ?: v->node);
4623e8e6
LP		 2547 if (r < 0)
2548 return r;
2549
efc3b12f 2550 cryptsetup_enable_logging(cd);
1887032f 2551
0d12936d 2552 r = sym_crypt_load(cd, CRYPT_VERITY, NULL);
4623e8e6 2553 if (r < 0)
294bd454 2554 return r;
4623e8e6 2555
0d12936d 2556 r = sym_crypt_set_data_device(cd, m->node);
4623e8e6 2557 if (r < 0)
294bd454 2558 return r;
4623e8e6 2559
319a4f4b 2560 if (!GREEDY_REALLOC0(d->decrypted, d->n_decrypted + 1))
ac1f3ad0
LB		 2561 return -ENOMEM;
2562
2563 /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
2564 * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
2565 * retry a few times before giving up. */
2566 for (unsigned i = 0; i < N_DEVICE_NODE_LIST_ATTEMPTS; i++) {
1d369d78 2567 _cleanup_(sym_crypt_freep) struct crypt_device *existing_cd = NULL;
254d1313 2568 _cleanup_close_ int fd = -EBADF;
c2fa92e7 2569
f80015ff
YW		 2570 /* First, check if the device already exists. */
2571 fd = open(node, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
2572 if (fd < 0 && !ERRNO_IS_DEVICE_ABSENT(errno))
2573 return log_debug_errno(errno, "Failed to open verity device %s: %m", node);
2574 if (fd >= 0)
2575 goto check; /* The device already exists. Let's check it. */
2576
2577 /* The symlink to the device node does not exist yet. Assume not activated, and let's activate it. */
c2fa92e7 2578 r = do_crypt_activate_verity(cd, name, verity);
1d369d78 2579 if (r >= 0)
f7725647 2580 goto try_open; /* The device is activated. Let's open it. */
ac1f3ad0
LB		 2581 /* libdevmapper can return EINVAL when the device is already in the activation stage.
2582 * There's no way to distinguish this situation from a genuine error due to invalid
2aed63f4 2583 * parameters, so immediately fall back to activating the device with a unique name.
89e62e0b
LP		 2584 * Improvements in libcrypsetup can ensure this never happens:
2585 * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
ac1f3ad0 2586 if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
0f75b0c5 2587 break;
1d369d78
YW		 2588 if (r == -ENODEV) /* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */
2589 goto try_again;
2590 if (!IN_SET(r,
2591 -EEXIST, /* Volume has already been opened and ready to be used. */
2592 -EBUSY /* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */))
2593 return log_debug_errno(r, "Failed to activate verity device %s: %m", node);
2594
f80015ff 2595 check:
1d369d78
YW		 2596 if (!restore_deferred_remove){
2597 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
2598 r = dm_deferred_remove_cancel(name);
12f5fbdf
YW		 2599 /* -EBUSY and -ENXIO: the device has already been removed or being removed. We cannot
2600 * use the device, try to open again. See target_message() in drivers/md/dm-ioctl.c
2601 * and dm_cancel_deferred_remove() in drivers/md/dm.c */
2602 if (IN_SET(r, -EBUSY, -ENXIO))
2603 goto try_again;
2604 if (r < 0)
1d369d78 2605 return log_debug_errno(r, "Failed to disable automated deferred removal for verity device %s: %m", node);
12f5fbdf
YW		 2606
2607 restore_deferred_remove = strdup(name);
2608 if (!restore_deferred_remove)
2609 return log_oom_debug();
1d369d78 2610 }
c2923fdc 2611
1d369d78
YW		 2612 r = verity_can_reuse(verity, name, &existing_cd);
2613 /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
2614 if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
2615 break;
2616 if (IN_SET(r,
2617 -ENOENT, /* Removed?? */
2618 -EBUSY, /* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */
2619 -ENODEV /* Volume is being opened but not ready, crypt_init_by_name() would fail, try to open again. */ ))
2620 goto try_again;
2621 if (r < 0)
2622 return log_debug_errno(r, "Failed to check if existing verity device %s can be reused: %m", node);
2623
f80015ff 2624 if (fd < 0) {
1d369d78
YW		 2625 /* devmapper might say that the device exists, but the devlink might not yet have been
2626 * created. Check and wait for the udev event in that case. */
2627 r = device_wait_for_devlink(node, "block", verity_timeout(), NULL);
2628 /* Fallback to activation with a unique device if it's taking too long */
2629 if (r == -ETIMEDOUT && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
0f75b0c5 2630 break;
1d369d78
YW		 2631 if (r < 0)
2632 return log_debug_errno(r, "Failed to wait device node symlink %s: %m", node);
c2923fdc 2633 }
ecab4c47 2634
f7725647
YW		 2635 try_open:
2636 if (fd < 0) {
2637 /* Now, the device is activated and devlink is created. Let's open it. */
2638 fd = open(node, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY);
2639 if (fd < 0) {
2640 if (!ERRNO_IS_DEVICE_ABSENT(errno))
2641 return log_debug_errno(errno, "Failed to open verity device %s: %m", node);
2642
2643 /* The device has already been removed?? */
2644 goto try_again;
2645 }
2646 }
2647
2648 mount_node_fd = TAKE_FD(fd);
1d369d78
YW		 2649 if (existing_cd)
2650 crypt_free_and_replace(cd, existing_cd);
2651
2652 goto success;
2653
2654 try_again:
2655 /* Device is being removed by another process. Let's wait for a while. */
ecab4c47 2656 (void) usleep(2 * USEC_PER_MSEC);
ac1f3ad0
LB		 2657 }
2658
0f75b0c5
YW		 2659 /* All trials failed or a conflicting verity device exists. Let's try to activate with a unique name. */
2660 if (FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) {
2661 /* Before trying to activate with unique name, we need to free crypt_device object.
2662 * Otherwise, we get error from libcryptsetup like the following:
2663 * ------
2664 * systemd[1234]: Cannot use device /dev/loop5 which is in use (already mapped or mounted).
2665 * ------
2666 */
2667 sym_crypt_free(cd);
2668 cd = NULL;
aee36b4e 2669 return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
0f75b0c5 2670 }
ac1f3ad0 2671
9972e6d6
YW		 2672 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY), "All attempts to activate verity device %s failed.", name);
2673
2674success:
ac1f3ad0
LB		 2675 /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
2676 restore_deferred_remove = mfree(restore_deferred_remove);
4623e8e6 2677
94344385
LP		 2678 d->decrypted[d->n_decrypted++] = (DecryptedPartition) {
2679 .name = TAKE_PTR(name),
2680 .device = TAKE_PTR(cd),
2681 };
4623e8e6 2682
1cc6c93a 2683 m->decrypted_node = TAKE_PTR(node);
f7725647 2684 close_and_replace(m->mount_node_fd, mount_node_fd);
4623e8e6
LP		 2685
2686 return 0;
18b5886e
LP		 2687}
2688#endif
2689
2690int dissected_image_decrypt(
2691 DissectedImage *m,
2692 const char *passphrase,
89e62e0b 2693 const VeritySettings *verity,
e330f97a 2694 DissectImageFlags flags) {
18b5886e 2695
349cc4a5 2696#if HAVE_LIBCRYPTSETUP
49b5b3b4 2697 _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL;
18b5886e
LP		 2698 int r;
2699#endif
2700
2701 assert(m);
89e62e0b 2702 assert(!verity || verity->root_hash || verity->root_hash_size == 0);
18b5886e
LP		 2703
2704 /* Returns:
2705 *
2706 * = 0 → There was nothing to decrypt
2707 * > 0 → Decrypted successfully
d1c536f5 2708 * -ENOKEY → There's something to decrypt but no key was supplied
18b5886e
LP		 2709 * -EKEYREJECTED → Passed key was not correct
2710 */
2711
89e62e0b 2712 if (verity && verity->root_hash && verity->root_hash_size < sizeof(sd_id128_t))
4623e8e6
LP		 2713 return -EINVAL;
2714
e330f97a 2715 if (!m->encrypted && !m->verity_ready)
18b5886e 2716 return 0;
18b5886e 2717
349cc4a5 2718#if HAVE_LIBCRYPTSETUP
9321ad51
YW		 2719 r = decrypted_image_new(&d);
2720 if (r < 0)
2721 return r;
18b5886e 2722
569a0e42 2723 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
18b5886e 2724 DissectedPartition *p = m->partitions + i;
22043172 2725 PartitionDesignator k;
18b5886e
LP		 2726
2727 if (!p->found)
2728 continue;
2729
2730 r = decrypt_partition(p, passphrase, flags, d);
2731 if (r < 0)
2732 return r;
2733
dd894023 2734 k = partition_verity_of(i);
4623e8e6 2735 if (k >= 0) {
aee36b4e 2736 r = verity_partition(i, p, m->partitions + k, verity, flags | DISSECT_IMAGE_VERITY_SHARE, d);
4623e8e6
LP		 2737 if (r < 0)
2738 return r;
2739 }
2740
d2c6e79d 2741 if (!p->decrypted_fstype && p->mount_node_fd >= 0 && p->decrypted_node) {
c80c9079 2742 r = probe_filesystem_full(p->mount_node_fd, p->decrypted_node, 0, UINT64_MAX, &p->decrypted_fstype);
7cc84b2c 2743 if (r < 0 && r != -EUCLEAN)
18b5886e
LP		 2744 return r;
2745 }
2746 }
2747
ac1e1b5f 2748 m->decrypted_image = TAKE_PTR(d);
18b5886e
LP		 2749
2750 return 1;
2751#else
2752 return -EOPNOTSUPP;
2753#endif
2754}
2755
2756int dissected_image_decrypt_interactively(
2757 DissectedImage *m,
2758 const char *passphrase,
89e62e0b 2759 const VeritySettings *verity,
e330f97a 2760 DissectImageFlags flags) {
18b5886e
LP		 2761
2762 _cleanup_strv_free_erase_ char **z = NULL;
2763 int n = 3, r;
2764
2765 if (passphrase)
2766 n--;
2767
2768 for (;;) {
e330f97a 2769 r = dissected_image_decrypt(m, passphrase, verity, flags);
18b5886e
LP		 2770 if (r >= 0)
2771 return r;
2772 if (r == -EKEYREJECTED)
2773 log_error_errno(r, "Incorrect passphrase, try again!");
fc95c359
YW		 2774 else if (r != -ENOKEY)
2775 return log_error_errno(r, "Failed to decrypt image: %m");
18b5886e 2776
baaa35ad
ZJS		 2777 if (--n < 0)
2778 return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED),
2779 "Too many retries.");
18b5886e
LP		 2780
2781 z = strv_free(z);
2782
8806bb4b 2783 r = ask_password_auto("Please enter image passphrase:", NULL, "dissect", "dissect", "dissect.passphrase", USEC_INFINITY, 0, &z);
18b5886e
LP		 2784 if (r < 0)
2785 return log_error_errno(r, "Failed to query for passphrase: %m");
2786
2787 passphrase = z[0];
2788 }
2789}
2790
e330f97a 2791static int decrypted_image_relinquish(DecryptedImage *d) {
18b5886e
LP		 2792 assert(d);
2793
67f63ee5
ZJS		 2794 /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a
2795 * boolean so that we don't clean it up ourselves either anymore */
18b5886e 2796
349cc4a5 2797#if HAVE_LIBCRYPTSETUP
67f63ee5
ZJS		 2798 int r;
2799
2800 for (size_t i = 0; i < d->n_decrypted; i++) {
18b5886e
LP		 2801 DecryptedPartition *p = d->decrypted + i;
2802
2803 if (p->relinquished)
2804 continue;
2805
0d12936d 2806 r = sym_crypt_deactivate_by_name(NULL, p->name, CRYPT_DEACTIVATE_DEFERRED);
18b5886e
LP		 2807 if (r < 0)
2808 return log_debug_errno(r, "Failed to mark %s for auto-removal: %m", p->name);
2809
2810 p->relinquished = true;
2811 }
2812#endif
2813
2814 return 0;
2815}
2816
f906075a
YW		 2817int dissected_image_relinquish(DissectedImage *m) {
2818 int r;
2819
2820 assert(m);
2821
2822 if (m->decrypted_image) {
2823 r = decrypted_image_relinquish(m->decrypted_image);
2824 if (r < 0)
2825 return r;
2826 }
2827
2828 if (m->loop)
2829 loop_device_relinquish(m->loop);
2830
2831 return 0;
2832}
2833
89e62e0b
LP		 2834static char *build_auxiliary_path(const char *image, const char *suffix) {
2835 const char *e;
2836 char *n;
2837
2838 assert(image);
2839 assert(suffix);
2840
2841 e = endswith(image, ".raw");
2842 if (!e)
2843 return strjoin(e, suffix);
2844
2845 n = new(char, e - image + strlen(suffix) + 1);
2846 if (!n)
2847 return NULL;
2848
2849 strcpy(mempcpy(n, image, e - image), suffix);
2850 return n;
2851}
2852
2853void verity_settings_done(VeritySettings *v) {
2854 assert(v);
2855
2856 v->root_hash = mfree(v->root_hash);
2857 v->root_hash_size = 0;
2858
2859 v->root_hash_sig = mfree(v->root_hash_sig);
2860 v->root_hash_sig_size = 0;
2861
2862 v->data_path = mfree(v->data_path);
2863}
2864
2865int verity_settings_load(
2866 VeritySettings *verity,
f5ea63a5
LP		 2867 const char *image,
2868 const char *root_hash_path,
89e62e0b
LP		 2869 const char *root_hash_sig_path) {
2870
2871 _cleanup_free_ void *root_hash = NULL, *root_hash_sig = NULL;
2872 size_t root_hash_size = 0, root_hash_sig_size = 0;
2873 _cleanup_free_ char *verity_data_path = NULL;
aee36b4e 2874 PartitionDesignator designator;
78ebe980
LP		 2875 int r;
2876
89e62e0b 2877 assert(verity);
78ebe980 2878 assert(image);
aee36b4e 2879 assert(verity->designator < 0 || IN_SET(verity->designator, PARTITION_ROOT, PARTITION_USR));
78ebe980 2880
89e62e0b
LP		 2881 /* If we are asked to load the root hash for a device node, exit early */
2882 if (is_device_path(image))
78ebe980 2883 return 0;
78ebe980 2884
d5fcc5b0
LP		 2885 r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIDECAR");
2886 if (r < 0 && r != -ENXIO)
2887 log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIDECAR, ignoring: %m");
2888 if (r == 0)
2889 return 0;
2890
aee36b4e
LP		 2891 designator = verity->designator;
2892
89e62e0b 2893 /* We only fill in what isn't already filled in */
c2923fdc 2894
89e62e0b 2895 if (!verity->root_hash) {
e7cbe5cb 2896 _cleanup_free_ char *text = NULL;
e7cbe5cb 2897
0389f4fa 2898 if (root_hash_path) {
aee36b4e 2899 /* If explicitly specified it takes precedence */
0389f4fa
LB		 2900 r = read_one_line_file(root_hash_path, &text);
2901 if (r < 0)
e7cbe5cb 2902 return r;
aee36b4e
LP		 2903
2904 if (designator < 0)
2905 designator = PARTITION_ROOT;
0389f4fa 2906 } else {
aee36b4e
LP		 2907 /* Otherwise look for xattr and separate file, and first for the data for root and if
2908 * that doesn't exist for /usr */
0389f4fa 2909
aee36b4e 2910 if (designator < 0 || designator == PARTITION_ROOT) {
c53e07e2 2911 r = getxattr_malloc(image, "user.verity.roothash", &text);
aee36b4e
LP		 2912 if (r < 0) {
2913 _cleanup_free_ char *p = NULL;
78ebe980 2914
00675c36 2915 if (r != -ENOENT && !ERRNO_IS_XATTR_ABSENT(r))
aee36b4e 2916 return r;
e7cbe5cb 2917
aee36b4e
LP		 2918 p = build_auxiliary_path(image, ".roothash");
2919 if (!p)
2920 return -ENOMEM;
2921
2922 r = read_one_line_file(p, &text);
2923 if (r < 0 && r != -ENOENT)
2924 return r;
2925 }
2926
2927 if (text)
2928 designator = PARTITION_ROOT;
2929 }
2930
2931 if (!text && (designator < 0 || designator == PARTITION_USR)) {
2932 /* So in the "roothash" xattr/file name above the "root" of course primarily
2933 * refers to the root of the Verity Merkle tree. But coincidentally it also
2934 * is the hash for the *root* file system, i.e. the "root" neatly refers to
2935 * two distinct concepts called "root". Taking benefit of this happy
2936 * coincidence we call the file with the root hash for the /usr/ file system
2937 * `usrhash`, because `usrroothash` or `rootusrhash` would just be too
2938 * confusing. We thus drop the reference to the root of the Merkle tree, and
2939 * just indicate which file system it's about. */
c53e07e2 2940 r = getxattr_malloc(image, "user.verity.usrhash", &text);
aee36b4e
LP		 2941 if (r < 0) {
2942 _cleanup_free_ char *p = NULL;
2943
00675c36 2944 if (r != -ENOENT && !ERRNO_IS_XATTR_ABSENT(r))
aee36b4e
LP		 2945 return r;
2946
2947 p = build_auxiliary_path(image, ".usrhash");
2948 if (!p)
2949 return -ENOMEM;
2950
2951 r = read_one_line_file(p, &text);
2952 if (r < 0 && r != -ENOENT)
2953 return r;
2954 }
2955
2956 if (text)
2957 designator = PARTITION_USR;
0389f4fa 2958 }
e7cbe5cb
LB		 2959 }
2960
2961 if (text) {
89e62e0b 2962 r = unhexmem(text, strlen(text), &root_hash, &root_hash_size);
e7cbe5cb
LB		 2963 if (r < 0)
2964 return r;
89e62e0b 2965 if (root_hash_size < sizeof(sd_id128_t))
e7cbe5cb
LB		 2966 return -EINVAL;
2967 }
2968 }
2969
90f98986 2970 if ((root_hash || verity->root_hash) && !verity->root_hash_sig) {
aee36b4e 2971 if (root_hash_sig_path) {
ae9cf30b 2972 r = read_full_file(root_hash_sig_path, (char**) &root_hash_sig, &root_hash_sig_size);
aee36b4e
LP		 2973 if (r < 0 && r != -ENOENT)
2974 return r;
2975
2976 if (designator < 0)
2977 designator = PARTITION_ROOT;
2978 } else {
2979 if (designator < 0 || designator == PARTITION_ROOT) {
2980 _cleanup_free_ char *p = NULL;
2981
2982 /* Follow naming convention recommended by the relevant RFC:
2983 * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
2984 p = build_auxiliary_path(image, ".roothash.p7s");
2985 if (!p)
2986 return -ENOMEM;
89e62e0b 2987
ae9cf30b 2988 r = read_full_file(p, (char**) &root_hash_sig, &root_hash_sig_size);
aee36b4e
LP		 2989 if (r < 0 && r != -ENOENT)
2990 return r;
2991 if (r >= 0)
2992 designator = PARTITION_ROOT;
2993 }
2994
2995 if (!root_hash_sig && (designator < 0 || designator == PARTITION_USR)) {
2996 _cleanup_free_ char *p = NULL;
2997
2998 p = build_auxiliary_path(image, ".usrhash.p7s");
2999 if (!p)
3000 return -ENOMEM;
89e62e0b 3001
ae9cf30b 3002 r = read_full_file(p, (char**) &root_hash_sig, &root_hash_sig_size);
aee36b4e
LP		 3003 if (r < 0 && r != -ENOENT)
3004 return r;
3005 if (r >= 0)
3006 designator = PARTITION_USR;
3007 }
89e62e0b
LP		 3008 }
3009
aee36b4e 3010 if (root_hash_sig && root_hash_sig_size == 0) /* refuse empty size signatures */
89e62e0b
LP		 3011 return -EINVAL;
3012 }
3013
3014 if (!verity->data_path) {
3015 _cleanup_free_ char *p = NULL;
3016
3017 p = build_auxiliary_path(image, ".verity");
3018 if (!p)
3019 return -ENOMEM;
3020
3021 if (access(p, F_OK) < 0) {
3022 if (errno != ENOENT)
3023 return -errno;
3024 } else
3025 verity_data_path = TAKE_PTR(p);
3026 }
3027
3028 if (root_hash) {
3029 verity->root_hash = TAKE_PTR(root_hash);
3030 verity->root_hash_size = root_hash_size;
3031 }
3032
3033 if (root_hash_sig) {
3034 verity->root_hash_sig = TAKE_PTR(root_hash_sig);
3035 verity->root_hash_sig_size = root_hash_sig_size;
e7cbe5cb 3036 }
89e62e0b
LP		 3037
3038 if (verity_data_path)
3039 verity->data_path = TAKE_PTR(verity_data_path);
78ebe980 3040
aee36b4e
LP		 3041 if (verity->designator < 0)
3042 verity->designator = designator;
3043
78ebe980
LP		 3044 return 1;
3045}
3046
88b3300f
LP		 3047int dissected_image_load_verity_sig_partition(
3048 DissectedImage *m,
3049 int fd,
3050 VeritySettings *verity) {
3051
3052 _cleanup_free_ void *root_hash = NULL, *root_hash_sig = NULL;
3053 _cleanup_(json_variant_unrefp) JsonVariant *v = NULL;
3054 size_t root_hash_size, root_hash_sig_size;
3055 _cleanup_free_ char *buf = NULL;
3056 PartitionDesignator d;
3057 DissectedPartition *p;
3058 JsonVariant *rh, *sig;
3059 ssize_t n;
3060 char *e;
3061 int r;
3062
3063 assert(m);
3064 assert(fd >= 0);
3065 assert(verity);
3066
3067 if (verity->root_hash && verity->root_hash_sig) /* Already loaded? */
3068 return 0;
3069
3070 r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_EMBEDDED");
3071 if (r < 0 && r != -ENXIO)
3072 log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_EMBEDDED, ignoring: %m");
3073 if (r == 0)
3074 return 0;
3075
dd894023 3076 d = partition_verity_sig_of(verity->designator < 0 ? PARTITION_ROOT : verity->designator);
88b3300f
LP		 3077 assert(d >= 0);
3078
3079 p = m->partitions + d;
3080 if (!p->found)
3081 return 0;
3082 if (p->offset == UINT64_MAX || p->size == UINT64_MAX)
3083 return -EINVAL;
3084
3085 if (p->size > 4*1024*1024) /* Signature data cannot possible be larger than 4M, refuse that */
3086 return -EFBIG;
3087
3088 buf = new(char, p->size+1);
3089 if (!buf)
3090 return -ENOMEM;
3091
3092 n = pread(fd, buf, p->size, p->offset);
3093 if (n < 0)
3094 return -ENOMEM;
3095 if ((uint64_t) n != p->size)
3096 return -EIO;
3097
3098 e = memchr(buf, 0, p->size);
3099 if (e) {
3100 /* If we found a NUL byte then the rest of the data must be NUL too */
3101 if (!memeqzero(e, p->size - (e - buf)))
3102 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature data contains embedded NUL byte.");
3103 } else
3104 buf[p->size] = 0;
3105
3106 r = json_parse(buf, 0, &v, NULL, NULL);
3107 if (r < 0)
3108 return log_debug_errno(r, "Failed to parse signature JSON data: %m");
3109
3110 rh = json_variant_by_key(v, "rootHash");
3111 if (!rh)
3112 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature JSON object lacks 'rootHash' field.");
3113 if (!json_variant_is_string(rh))
3114 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "'rootHash' field of signature JSON object is not a string.");
3115
3116 r = unhexmem(json_variant_string(rh), SIZE_MAX, &root_hash, &root_hash_size);
3117 if (r < 0)
3118 return log_debug_errno(r, "Failed to parse root hash field: %m");
3119
3120 /* Check if specified root hash matches if it is specified */
3121 if (verity->root_hash &&
3122 memcmp_nn(verity->root_hash, verity->root_hash_size, root_hash, root_hash_size) != 0) {
3123 _cleanup_free_ char *a = NULL, *b = NULL;
3124
3125 a = hexmem(root_hash, root_hash_size);
3126 b = hexmem(verity->root_hash, verity->root_hash_size);
3127
3128 return log_debug_errno(r, "Root hash in signature JSON data (%s) doesn't match configured hash (%s).", strna(a), strna(b));
3129 }
3130
3131 sig = json_variant_by_key(v, "signature");
3132 if (!sig)
3133 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature JSON object lacks 'signature' field.");
3134 if (!json_variant_is_string(sig))
3135 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "'signature' field of signature JSON object is not a string.");
3136
3137 r = unbase64mem(json_variant_string(sig), SIZE_MAX, &root_hash_sig, &root_hash_sig_size);
3138 if (r < 0)
3139 return log_debug_errno(r, "Failed to parse signature field: %m");
3140
3141 free_and_replace(verity->root_hash, root_hash);
3142 verity->root_hash_size = root_hash_size;
3143
3144 free_and_replace(verity->root_hash_sig, root_hash_sig);
3145 verity->root_hash_sig_size = root_hash_sig_size;
3146
3147 return 1;
3148}
3149
22847508 3150int dissected_image_acquire_metadata(DissectedImage *m, DissectImageFlags extra_flags) {
3b925504
LP		 3151
3152 enum {
3153 META_HOSTNAME,
3154 META_MACHINE_ID,
3155 META_MACHINE_INFO,
3156 META_OS_RELEASE,
fab22946 3157 META_INITRD_RELEASE,
7718ac97 3158 META_EXTENSION_RELEASE,
a4e0d617 3159 META_HAS_INIT_SYSTEM,
3b925504
LP		 3160 _META_MAX,
3161 };
3162
9a4b883b 3163 static const char *const paths[_META_MAX] = {
7718ac97
LB		 3164 [META_HOSTNAME] = "/etc/hostname\0",
3165 [META_MACHINE_ID] = "/etc/machine-id\0",
3166 [META_MACHINE_INFO] = "/etc/machine-info\0",
9a4b883b 3167 [META_OS_RELEASE] = ("/etc/os-release\0"
22847508 3168 "/usr/lib/os-release\0"),
fab22946
LP		 3169 [META_INITRD_RELEASE] = ("/etc/initrd-release\0"
3170 "/usr/lib/initrd-release\0"),
a4e0d617
LP		 3171 [META_EXTENSION_RELEASE] = "extension-release\0", /* Used only for logging. */
3172 [META_HAS_INIT_SYSTEM] = "has-init-system\0", /* ditto */
3b925504
LP		 3173 };
3174
fab22946 3175 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL, **initrd_release = NULL, **extension_release = NULL;
19ee48a6 3176 _cleanup_close_pair_ int error_pipe[2] = PIPE_EBADF;
3b925504
LP		 3177 _cleanup_(rmdir_and_freep) char *t = NULL;
3178 _cleanup_(sigkill_waitp) pid_t child = 0;
3179 sd_id128_t machine_id = SD_ID128_NULL;
3180 _cleanup_free_ char *hostname = NULL;
67f63ee5 3181 unsigned n_meta_initialized = 0;
af8219d5 3182 int fds[2 * _META_MAX], r, v;
a4e0d617 3183 int has_init_system = -1;
af8219d5 3184 ssize_t n;
3b925504
LP		 3185
3186 BLOCK_SIGNALS(SIGCHLD);
3187
3188 assert(m);
3189
7718ac97 3190 for (; n_meta_initialized < _META_MAX; n_meta_initialized ++) {
d9119c00 3191 if (!paths[n_meta_initialized]) {
254d1313 3192 fds[2*n_meta_initialized] = fds[2*n_meta_initialized+1] = -EBADF;
7718ac97 3193 continue;
d9119c00
LP		 3194 }
3195
3b925504
LP		 3196 if (pipe2(fds + 2*n_meta_initialized, O_CLOEXEC) < 0) {
3197 r = -errno;
3198 goto finish;
3199 }
7718ac97 3200 }
3b925504
LP		 3201
3202 r = mkdtemp_malloc("/tmp/dissect-XXXXXX", &t);
3203 if (r < 0)
3204 goto finish;
3205
af8219d5
LP		 3206 if (pipe2(error_pipe, O_CLOEXEC) < 0) {
3207 r = -errno;
3208 goto finish;
3209 }
3210
e2047ba9 3211 r = safe_fork("(sd-dissect)", FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE, &child);
be39f6ee 3212 if (r < 0)
3b925504 3213 goto finish;
be39f6ee 3214 if (r == 0) {
a4e0d617 3215 /* Child in a new mount namespace */
af8219d5
LP		 3216 error_pipe[0] = safe_close(error_pipe[0]);
3217
7cf66030
LP		 3218 r = dissected_image_mount(
3219 m,
3220 t,
3221 UID_INVALID,
21b61b1d 3222 UID_INVALID,
22847508
ZJS		 3223 extra_flags |
3224 DISSECT_IMAGE_READ_ONLY |
3225 DISSECT_IMAGE_MOUNT_ROOT_ONLY |
7cf66030 3226 DISSECT_IMAGE_USR_NO_ROOT);
429d4e41
LP		 3227 if (r < 0) {
3228 log_debug_errno(r, "Failed to mount dissected image: %m");
03ae68f4 3229 goto inner_fail;
429d4e41 3230 }
3b925504 3231
67f63ee5 3232 for (unsigned k = 0; k < _META_MAX; k++) {
37e44c3f 3233 _cleanup_close_ int fd = -ENOENT;
3b925504 3234
7718ac97
LB		 3235 if (!paths[k])
3236 continue;
3237
3b925504
LP		 3238 fds[2*k] = safe_close(fds[2*k]);
3239
a4e0d617
LP		 3240 switch (k) {
3241
3242 case META_EXTENSION_RELEASE:
9a4b883b
LB		 3243 /* As per the os-release spec, if the image is an extension it will have a file
3244 * named after the image name in extension-release.d/ - we use the image name
3245 * and try to resolve it with the extension-release helpers, as sometimes
3246 * the image names are mangled on deployment and do not match anymore.
3247 * Unlike other paths this is not fixed, and the image name
3248 * can be mangled on deployment, so by calling into the helper
3249 * we allow a fallback that matches on the first extension-release
3250 * file found in the directory, if one named after the image cannot
3251 * be found first. */
06768b90 3252 r = open_extension_release(t, m->image_name, /* relax_extension_release_check= */ false, NULL, &fd);
9a4b883b
LB		 3253 if (r < 0)
3254 fd = r; /* Propagate the error. */
a4e0d617
LP		 3255 break;
3256
3257 case META_HAS_INIT_SYSTEM: {
3258 bool found = false;
a4e0d617
LP		 3259
3260 FOREACH_STRING(init,
3261 "/usr/lib/systemd/systemd", /* systemd on /usr merged system */
3262 "/lib/systemd/systemd", /* systemd on /usr non-merged systems */
3263 "/sbin/init") { /* traditional path the Linux kernel invokes */
3264
f461a28d 3265 r = chase(init, t, CHASE_PREFIX_ROOT, NULL, NULL);
a4e0d617
LP		 3266 if (r < 0) {
3267 if (r != -ENOENT)
3268 log_debug_errno(r, "Failed to resolve %s, ignoring: %m", init);
3269 } else {
3270 found = true;
3271 break;
3272 }
3273 }
3274
3275 r = loop_write(fds[2*k+1], &found, sizeof(found), false);
3276 if (r < 0)
3277 goto inner_fail;
3278
3279 continue;
3280 }
3281
3282 default:
9a4b883b 3283 NULSTR_FOREACH(p, paths[k]) {
f461a28d 3284 fd = chase_and_open(p, t, CHASE_PREFIX_ROOT, O_RDONLY|O_CLOEXEC|O_NOCTTY, NULL);
9a4b883b
LB		 3285 if (fd >= 0)
3286 break;
3287 }
a4e0d617
LP		 3288 }
3289
36952d19
LP		 3290 if (fd < 0) {
3291 log_debug_errno(fd, "Failed to read %s file of image, ignoring: %m", paths[k]);
37e44c3f 3292 fds[2*k+1] = safe_close(fds[2*k+1]);
3b925504 3293 continue;
36952d19 3294 }
3b925504 3295
f5fbe71d 3296 r = copy_bytes(fd, fds[2*k+1], UINT64_MAX, 0);
a4e0d617
LP		 3297 if (r < 0)
3298 goto inner_fail;
3b925504
LP		 3299
3300 fds[2*k+1] = safe_close(fds[2*k+1]);
3301 }
3302
3303 _exit(EXIT_SUCCESS);
a4e0d617
LP		 3304
3305 inner_fail:
03ae68f4 3306 /* Let parent know the error */
a4e0d617
LP		 3307 (void) write(error_pipe[1], &r, sizeof(r));
3308 _exit(EXIT_FAILURE);
3b925504
LP		 3309 }
3310
af8219d5
LP		 3311 error_pipe[1] = safe_close(error_pipe[1]);
3312
67f63ee5 3313 for (unsigned k = 0; k < _META_MAX; k++) {
3b925504
LP		 3314 _cleanup_fclose_ FILE *f = NULL;
3315
7718ac97
LB		 3316 if (!paths[k])
3317 continue;
3318
3b925504
LP		 3319 fds[2*k+1] = safe_close(fds[2*k+1]);
3320
4fa744a3 3321 f = take_fdopen(&fds[2*k], "r");
3b925504
LP		 3322 if (!f) {
3323 r = -errno;
3324 goto finish;
3325 }
3326
3b925504
LP		 3327 switch (k) {
3328
3329 case META_HOSTNAME:
3330 r = read_etc_hostname_stream(f, &hostname);
3331 if (r < 0)
f6048e5e 3332 log_debug_errno(r, "Failed to read /etc/hostname of image: %m");
3b925504
LP		 3333
3334 break;
3335
3336 case META_MACHINE_ID: {
3337 _cleanup_free_ char *line = NULL;
3338
3339 r = read_line(f, LONG_LINE_MAX, &line);
3340 if (r < 0)
f6048e5e 3341 log_debug_errno(r, "Failed to read /etc/machine-id of image: %m");
3b925504
LP		 3342 else if (r == 33) {
3343 r = sd_id128_from_string(line, &machine_id);
3344 if (r < 0)
3345 log_debug_errno(r, "Image contains invalid /etc/machine-id: %s", line);
3346 } else if (r == 0)
f6048e5e 3347 log_debug("/etc/machine-id file of image is empty.");
ab763cb2 3348 else if (streq(line, "uninitialized"))
f6048e5e 3349 log_debug("/etc/machine-id file of image is uninitialized (likely aborted first boot).");
3b925504 3350 else
f6048e5e 3351 log_debug("/etc/machine-id file of image has unexpected length %i.", r);
3b925504
LP		 3352
3353 break;
3354 }
3355
3356 case META_MACHINE_INFO:
aa8fbc74 3357 r = load_env_file_pairs(f, "machine-info", &machine_info);
3b925504 3358 if (r < 0)
f6048e5e 3359 log_debug_errno(r, "Failed to read /etc/machine-info of image: %m");
3b925504
LP		 3360
3361 break;
3362
3363 case META_OS_RELEASE:
aa8fbc74 3364 r = load_env_file_pairs(f, "os-release", &os_release);
3b925504 3365 if (r < 0)
f6048e5e 3366 log_debug_errno(r, "Failed to read OS release file of image: %m");
3b925504
LP		 3367
3368 break;
7718ac97 3369
fab22946
LP		 3370 case META_INITRD_RELEASE:
3371 r = load_env_file_pairs(f, "initrd-release", &initrd_release);
3372 if (r < 0)
3373 log_debug_errno(r, "Failed to read initrd release file of image: %m");
3374
3375 break;
3376
7718ac97
LB		 3377 case META_EXTENSION_RELEASE:
3378 r = load_env_file_pairs(f, "extension-release", &extension_release);
3379 if (r < 0)
f6048e5e 3380 log_debug_errno(r, "Failed to read extension release file of image: %m");
7718ac97
LB		 3381
3382 break;
a4e0d617
LP		 3383
3384 case META_HAS_INIT_SYSTEM: {
3385 bool b = false;
3386 size_t nr;
3387
3388 errno = 0;
3389 nr = fread(&b, 1, sizeof(b), f);
3390 if (nr != sizeof(b))
3391 log_debug_errno(errno_or_else(EIO), "Failed to read has-init-system boolean: %m");
3392 else
3393 has_init_system = b;
3394
3395 break;
3396 }}
3b925504
LP		 3397 }
3398
2e87a1fd 3399 r = wait_for_terminate_and_check("(sd-dissect)", child, 0);
3b925504 3400 child = 0;
2e87a1fd 3401 if (r < 0)
af8219d5
LP		 3402 return r;
3403
3404 n = read(error_pipe[0], &v, sizeof(v));
3405 if (n < 0)
3406 return -errno;
3407 if (n == sizeof(v))
3408 return v; /* propagate error sent to us from child */
3409 if (n != 0)
3410 return -EIO;
3411
2e87a1fd
LP		 3412 if (r != EXIT_SUCCESS)
3413 return -EPROTO;
3b925504
LP		 3414
3415 free_and_replace(m->hostname, hostname);
3416 m->machine_id = machine_id;
3417 strv_free_and_replace(m->machine_info, machine_info);
3418 strv_free_and_replace(m->os_release, os_release);
fab22946 3419 strv_free_and_replace(m->initrd_release, initrd_release);
7718ac97 3420 strv_free_and_replace(m->extension_release, extension_release);
a4e0d617 3421 m->has_init_system = has_init_system;
3b925504
LP		 3422
3423finish:
67f63ee5 3424 for (unsigned k = 0; k < n_meta_initialized; k++)
3b925504
LP		 3425 safe_close_pair(fds + 2*k);
3426
3427 return r;
3428}
3429
2348043f
LP		 3430Architecture dissected_image_architecture(DissectedImage *img) {
3431 assert(img);
3432
3433 if (img->partitions[PARTITION_ROOT].found &&
3434 img->partitions[PARTITION_ROOT].architecture >= 0)
3435 return img->partitions[PARTITION_ROOT].architecture;
3436
3437 if (img->partitions[PARTITION_USR].found &&
3438 img->partitions[PARTITION_USR].architecture >= 0)
3439 return img->partitions[PARTITION_USR].architecture;
3440
3441 return _ARCHITECTURE_INVALID;
3442}
3443
1e63dc4f
YW		 3444int dissect_loop_device(
3445 LoopDevice *loop,
3446 const VeritySettings *verity,
3447 const MountOptions *mount_options,
84be0c71 3448 const ImagePolicy *image_policy,
1e63dc4f
YW		 3449 DissectImageFlags flags,
3450 DissectedImage **ret) {
3451
08f14be4 3452#if HAVE_BLKID
1e63dc4f
YW		 3453 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
3454 int r;
3455
3456 assert(loop);
3457 assert(ret);
3458
08f14be4 3459 r = dissected_image_new(loop->backing_file ?: loop->node, &m);
1e63dc4f
YW		 3460 if (r < 0)
3461 return r;
3462
3463 m->loop = loop_device_ref(loop);
22ee78a8 3464 m->sector_size = m->loop->sector_size;
1e63dc4f 3465
84be0c71 3466 r = dissect_image(m, loop->fd, loop->node, verity, mount_options, image_policy, flags);
08f14be4 3467 if (r < 0)
08f14be4
YW		 3468 return r;
3469
1e63dc4f
YW		 3470 *ret = TAKE_PTR(m);
3471 return 0;
08f14be4
YW		 3472#else
3473 return -EOPNOTSUPP;
3474#endif
1e63dc4f
YW		 3475}
3476
bad31660 3477int dissect_loop_device_and_warn(
1e63dc4f 3478 LoopDevice *loop,
89e62e0b 3479 const VeritySettings *verity,
18d73705 3480 const MountOptions *mount_options,
84be0c71 3481 const ImagePolicy *image_policy,
4526113f
LP		 3482 DissectImageFlags flags,
3483 DissectedImage **ret) {
3484
64dd3a24 3485 const char *name;
4526113f
LP		 3486 int r;
3487
bad31660
YW		 3488 assert(loop);
3489 assert(loop->fd >= 0);
4526113f 3490
64dd3a24 3491 name = ASSERT_PTR(loop->backing_file ?: loop->node);
4526113f 3492
84be0c71 3493 r = dissect_loop_device(loop, verity, mount_options, image_policy, flags, ret);
4526113f
LP		 3494 switch (r) {
3495
3496 case -EOPNOTSUPP:
3497 return log_error_errno(r, "Dissecting images is not supported, compiled without blkid support.");
3498
3499 case -ENOPKG:
48084df6
ZJS		 3500 return log_error_errno(r, "%s: Couldn't identify a suitable partition table or file system.", name);
3501
3502 case -ENOMEDIUM:
3503 return log_error_errno(r, "%s: The image does not pass validation.", name);
4526113f
LP		 3504
3505 case -EADDRNOTAVAIL:
48084df6 3506 return log_error_errno(r, "%s: No root partition for specified root hash found.", name);
4526113f
LP		 3507
3508 case -ENOTUNIQ:
48084df6 3509 return log_error_errno(r, "%s: Multiple suitable root partitions found in image.", name);
4526113f
LP		 3510
3511 case -ENXIO:
48084df6 3512 return log_error_errno(r, "%s: No suitable root partition found in image.", name);
4526113f
LP		 3513
3514 case -EPROTONOSUPPORT:
3515 return log_error_errno(r, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name);
3516
48084df6
ZJS		 3517 case -ENOTBLK:
3518 return log_error_errno(r, "%s: Image is not a block device.", name);
3519
a94aa2b9
LP		 3520 case -EBADR:
3521 return log_error_errno(r,
3522 "Combining partitioned images (such as '%s') with external Verity data (such as '%s') not supported. "
3523 "(Consider setting $SYSTEMD_DISSECT_VERITY_SIDECAR=0 to disable automatic discovery of external Verity data.)",
3524 name, strna(verity ? verity->data_path : NULL));
3525
4526113f
LP		 3526 default:
3527 if (r < 0)
3528 return log_error_errno(r, "Failed to dissect image '%s': %m", name);
3529
3530 return r;
3531 }
3532}
3533
49536766
LP		 3534bool dissected_image_verity_candidate(const DissectedImage *image, PartitionDesignator partition_designator) {
3535 assert(image);
3536
3537 /* Checks if this partition could theoretically do Verity. For non-partitioned images this only works
3538 * if there's an external verity file supplied, for which we can consult .has_verity. For partitioned
3539 * images we only check the partition type.
3540 *
3541 * This call is used to decide whether to suppress or show a verity column in tabular output of the
3542 * image. */
3543
e7cbe5cb 3544 if (image->single_file_system)
c3c88d67 3545 return partition_designator == PARTITION_ROOT && image->has_verity;
e7cbe5cb 3546
dd894023 3547 return partition_verity_of(partition_designator) >= 0;
e7cbe5cb
LB		 3548}
3549
49536766
LP		 3550bool dissected_image_verity_ready(const DissectedImage *image, PartitionDesignator partition_designator) {
3551 PartitionDesignator k;
3552
3553 assert(image);
3554
3555 /* Checks if this partition has verity data available that we can activate. For non-partitioned this
3556 * works for the root partition, for others only if the associated verity partition was found. */
3557
3558 if (!image->verity_ready)
3559 return false;
e7cbe5cb
LB		 3560
3561 if (image->single_file_system)
49536766 3562 return partition_designator == PARTITION_ROOT;
e7cbe5cb 3563
dd894023 3564 k = partition_verity_of(partition_designator);
e7cbe5cb
LB		 3565 return k >= 0 && image->partitions[k].found;
3566}
3567
8ee9615e
LP		 3568bool dissected_image_verity_sig_ready(const DissectedImage *image, PartitionDesignator partition_designator) {
3569 PartitionDesignator k;
3570
3571 assert(image);
3572
3573 /* Checks if this partition has verity signature data available that we can use. */
3574
3575 if (!image->verity_sig_ready)
3576 return false;
3577
3578 if (image->single_file_system)
3579 return partition_designator == PARTITION_ROOT;
3580
dd894023 3581 k = partition_verity_sig_of(partition_designator);
8ee9615e
LP		 3582 return k >= 0 && image->partitions[k].found;
3583}
3584
18d73705
LB		 3585MountOptions* mount_options_free_all(MountOptions *options) {
3586 MountOptions *m;
3587
3588 while ((m = options)) {
3589 LIST_REMOVE(mount_options, options, m);
3590 free(m->options);
3591 free(m);
3592 }
3593
3594 return NULL;
3595}
3596
569a0e42 3597const char* mount_options_from_designator(const MountOptions *options, PartitionDesignator designator) {
f5215bc8 3598 LIST_FOREACH(mount_options, m, options)
9ece6444 3599 if (designator == m->partition_designator && !isempty(m->options))
18d73705 3600 return m->options;
6aa05ebd 3601
18d73705
LB		 3602 return NULL;
3603}
3604
6aa05ebd
LP		 3605int mount_image_privately_interactively(
3606 const char *image,
84be0c71 3607 const ImagePolicy *image_policy,
6aa05ebd
LP		 3608 DissectImageFlags flags,
3609 char **ret_directory,
a133d2c3 3610 int *ret_dir_fd,
e330f97a 3611 LoopDevice **ret_loop_device) {
6aa05ebd 3612
27ec815e 3613 _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT;
6aa05ebd 3614 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
6aa05ebd
LP		 3615 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
3616 _cleanup_(rmdir_and_freep) char *created_dir = NULL;
3617 _cleanup_free_ char *temp = NULL;
3618 int r;
3619
3620 /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
3621 * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
3622 * easily. */
3623
3624 assert(image);
3625 assert(ret_directory);
3626 assert(ret_loop_device);
6aa05ebd 3627
0b75493d 3628 /* We intend to mount this right-away, hence add the partitions if needed and pin them. */
73d88b80
LP		 3629 flags |= DISSECT_IMAGE_ADD_PARTITION_DEVICES |
3630 DISSECT_IMAGE_PIN_PARTITION_DEVICES;
3631
27ec815e
LP		 3632 r = verity_settings_load(&verity, image, NULL, NULL);
3633 if (r < 0)
3634 return log_error_errno(r, "Failed to load root hash data: %m");
3635
6aa05ebd
LP		 3636 r = tempfn_random_child(NULL, program_invocation_short_name, &temp);
3637 if (r < 0)
3638 return log_error_errno(r, "Failed to generate temporary mount directory: %m");
3639
3640 r = loop_device_make_by_path(
3641 image,
ef9c184d 3642 FLAGS_SET(flags, DISSECT_IMAGE_DEVICE_READ_ONLY) ? O_RDONLY : O_RDWR,
22ee78a8 3643 /* sector_size= */ UINT32_MAX,
6aa05ebd 3644 FLAGS_SET(flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN,
7f52206a 3645 LOCK_SH,
6aa05ebd
LP		 3646 &d);
3647 if (r < 0)
7b87fe4c 3648 return log_error_errno(r, "Failed to set up loopback device for %s: %m", image);
6aa05ebd 3649
84be0c71
LP		 3650 r = dissect_loop_device_and_warn(
3651 d,
3652 &verity,
3653 /* mount_options= */ NULL,
3654 image_policy,
3655 flags,
3656 &dissected_image);
6aa05ebd
LP		 3657 if (r < 0)
3658 return r;
3659
88b3300f
LP		 3660 r = dissected_image_load_verity_sig_partition(dissected_image, d->fd, &verity);
3661 if (r < 0)
3662 return r;
3663
e330f97a 3664 r = dissected_image_decrypt_interactively(dissected_image, NULL, &verity, flags);
6aa05ebd
LP		 3665 if (r < 0)
3666 return r;
3667
3668 r = detach_mount_namespace();
3669 if (r < 0)
3670 return log_error_errno(r, "Failed to detach mount namespace: %m");
3671
3672 r = mkdir_p(temp, 0700);
3673 if (r < 0)
3674 return log_error_errno(r, "Failed to create mount point: %m");
3675
3676 created_dir = TAKE_PTR(temp);
3677
21b61b1d 3678 r = dissected_image_mount_and_warn(dissected_image, created_dir, UID_INVALID, UID_INVALID, flags);
6aa05ebd 3679 if (r < 0)
af187ab2 3680 return r;
6aa05ebd 3681
41bc4849
LP		 3682 r = loop_device_flock(d, LOCK_UN);
3683 if (r < 0)
3684 return r;
3685
3044d343
YW		 3686 r = dissected_image_relinquish(dissected_image);
3687 if (r < 0)
3688 return log_error_errno(r, "Failed to relinquish DM and loopback block devices: %m");
6aa05ebd 3689
a133d2c3
DDM		 3690 if (ret_dir_fd) {
3691 _cleanup_close_ int dir_fd = -EBADF;
3692
3693 dir_fd = open(created_dir, O_CLOEXEC|O_DIRECTORY);
3694 if (dir_fd < 0)
3695 return log_error_errno(errno, "Failed to open mount point directory: %m");
3696
3697 *ret_dir_fd = TAKE_FD(dir_fd);
3698 }
3699
6aa05ebd
LP		 3700 *ret_directory = TAKE_PTR(created_dir);
3701 *ret_loop_device = TAKE_PTR(d);
6aa05ebd
LP		 3702
3703 return 0;
3704}
3705
06768b90
LB		 3706static bool mount_options_relax_extension_release_checks(const MountOptions *options) {
3707 if (!options)
3708 return false;
3709
3710 return string_contains_word(mount_options_from_designator(options, PARTITION_ROOT), ",", "x-systemd.relax-extension-release-check") ||
3711 string_contains_word(mount_options_from_designator(options, PARTITION_USR), ",", "x-systemd.relax-extension-release-check") ||
3712 string_contains_word(options->options, ",", "x-systemd.relax-extension-release-check");
3713}
3714
93f59701 3715int verity_dissect_and_mount(
cedf5b1a 3716 int src_fd,
93f59701
LB		 3717 const char *src,
3718 const char *dest,
3719 const MountOptions *options,
84be0c71 3720 const ImagePolicy *image_policy,
93f59701
LB		 3721 const char *required_host_os_release_id,
3722 const char *required_host_os_release_version_id,
60c5f700
LP		 3723 const char *required_host_os_release_sysext_level,
3724 const char *required_sysext_scope) {
93f59701 3725
4beda316 3726 _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL;
4beda316
LB		 3727 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
3728 _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT;
3729 DissectImageFlags dissect_image_flags;
06768b90 3730 bool relax_extension_release_check;
4beda316
LB		 3731 int r;
3732
3733 assert(src);
3734 assert(dest);
3735
06768b90
LB		 3736 relax_extension_release_check = mount_options_relax_extension_release_checks(options);
3737
cedf5b1a 3738 /* We might get an FD for the image, but we use the original path to look for the dm-verity files */
4beda316
LB		 3739 r = verity_settings_load(&verity, src, NULL, NULL);
3740 if (r < 0)
3741 return log_debug_errno(r, "Failed to load root hash: %m");
3742
06768b90 3743 dissect_image_flags = (verity.data_path ? DISSECT_IMAGE_NO_PARTITION_TABLE : 0) |
73d88b80
LP		 3744 (relax_extension_release_check ? DISSECT_IMAGE_RELAX_SYSEXT_CHECK : 0) |
3745 DISSECT_IMAGE_ADD_PARTITION_DEVICES |
3746 DISSECT_IMAGE_PIN_PARTITION_DEVICES;
4beda316 3747
cedf5b1a
LB		 3748 /* Note that we don't use loop_device_make here, as the FD is most likely O_PATH which would not be
3749 * accepted by LOOP_CONFIGURE, so just let loop_device_make_by_path reopen it as a regular FD. */
4beda316 3750 r = loop_device_make_by_path(
cedf5b1a 3751 src_fd >= 0 ? FORMAT_PROC_FD_PATH(src_fd) : src,
22ee78a8
LP		 3752 /* open_flags= */ -1,
3753 /* sector_size= */ UINT32_MAX,
4beda316 3754 verity.data_path ? 0 : LO_FLAGS_PARTSCAN,
7f52206a 3755 LOCK_SH,
4beda316
LB		 3756 &loop_device);
3757 if (r < 0)
3758 return log_debug_errno(r, "Failed to create loop device for image: %m");
3759
bad31660
YW		 3760 r = dissect_loop_device(
3761 loop_device,
4beda316
LB		 3762 &verity,
3763 options,
84be0c71 3764 image_policy,
4beda316
LB		 3765 dissect_image_flags,
3766 &dissected_image);
3767 /* No partition table? Might be a single-filesystem image, try again */
3768 if (!verity.data_path && r == -ENOPKG)
bad31660
YW		 3769 r = dissect_loop_device(
3770 loop_device,
4beda316
LB		 3771 &verity,
3772 options,
84be0c71 3773 image_policy,
75dc190d 3774 dissect_image_flags | DISSECT_IMAGE_NO_PARTITION_TABLE,
4beda316
LB		 3775 &dissected_image);
3776 if (r < 0)
3777 return log_debug_errno(r, "Failed to dissect image: %m");
3778
88b3300f
LP		 3779 r = dissected_image_load_verity_sig_partition(dissected_image, loop_device->fd, &verity);
3780 if (r < 0)
3781 return r;
3782
4beda316
LB		 3783 r = dissected_image_decrypt(
3784 dissected_image,
3785 NULL,
3786 &verity,
e330f97a 3787 dissect_image_flags);
4beda316
LB		 3788 if (r < 0)
3789 return log_debug_errno(r, "Failed to decrypt dissected image: %m");
3790
3791 r = mkdir_p_label(dest, 0755);
3792 if (r < 0)
3793 return log_debug_errno(r, "Failed to create destination directory %s: %m", dest);
3794 r = umount_recursive(dest, 0);
3795 if (r < 0)
3796 return log_debug_errno(r, "Failed to umount under destination directory %s: %m", dest);
3797
21b61b1d 3798 r = dissected_image_mount(dissected_image, dest, UID_INVALID, UID_INVALID, dissect_image_flags);
4beda316
LB		 3799 if (r < 0)
3800 return log_debug_errno(r, "Failed to mount image: %m");
3801
41bc4849
LP		 3802 r = loop_device_flock(loop_device, LOCK_UN);
3803 if (r < 0)
3804 return log_debug_errno(r, "Failed to unlock loopback device: %m");
3805
93f59701
LB		 3806 /* If we got os-release values from the caller, then we need to match them with the image's
3807 * extension-release.d/ content. Return -EINVAL if there's any mismatch.
3808 * First, check the distro ID. If that matches, then check the new SYSEXT_LEVEL value if
37361f46
LB		 3809 * available, or else fallback to VERSION_ID. If neither is present (eg: rolling release),
3810 * then a simple match on the ID will be performed. */
8b2dcbbd 3811 if (required_host_os_release_id) {
93f59701
LB		 3812 _cleanup_strv_free_ char **extension_release = NULL;
3813
d30d86b7
YW		 3814 assert(!isempty(required_host_os_release_id));
3815
06768b90 3816 r = load_extension_release_pairs(dest, dissected_image->image_name, relax_extension_release_check, &extension_release);
93f59701
LB		 3817 if (r < 0)
3818 return log_debug_errno(r, "Failed to parse image %s extension-release metadata: %m", dissected_image->image_name);
3819
3820 r = extension_release_validate(
60c5f700
LP		 3821 dissected_image->image_name,
3822 required_host_os_release_id,
3823 required_host_os_release_version_id,
3824 required_host_os_release_sysext_level,
3825 required_sysext_scope,
3826 extension_release);
93f59701
LB		 3827 if (r == 0)
3828 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE), "Image %s extension-release metadata does not match the root's", dissected_image->image_name);
3829 if (r < 0)
3830 return log_debug_errno(r, "Failed to compare image %s extension-release metadata with the root's os-release: %m", dissected_image->image_name);
3831 }
3832
3044d343
YW		 3833 r = dissected_image_relinquish(dissected_image);
3834 if (r < 0)
3835 return log_debug_errno(r, "Failed to relinquish dissected image: %m");
4beda316
LB		 3836
3837 return 0;
3838}
Unnamed repository; edit this file 'description' to name the repository.