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