]> 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
Merge pull request #18470 from mrc0mmand/ci-clang-12
[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/mount.h>
10 #include <sys/prctl.h>
11 #include <sys/wait.h>
12 #include <sysexits.h>
13
14 #include "sd-device.h"
15 #include "sd-id128.h"
16
17 #include "architecture.h"
18 #include "ask-password-api.h"
19 #include "blkid-util.h"
20 #include "blockdev-util.h"
21 #include "copy.h"
22 #include "cryptsetup-util.h"
23 #include "def.h"
24 #include "device-nodes.h"
25 #include "device-util.h"
26 #include "dissect-image.h"
27 #include "dm-util.h"
28 #include "env-file.h"
29 #include "fd-util.h"
30 #include "fileio.h"
31 #include "fs-util.h"
32 #include "fsck-util.h"
33 #include "gpt.h"
34 #include "hexdecoct.h"
35 #include "hostname-setup.h"
36 #include "id128-util.h"
37 #include "mkdir.h"
38 #include "mount-util.h"
39 #include "mountpoint-util.h"
40 #include "namespace-util.h"
41 #include "nulstr-util.h"
42 #include "os-util.h"
43 #include "path-util.h"
44 #include "process-util.h"
45 #include "raw-clone.h"
46 #include "signal-util.h"
47 #include "stat-util.h"
48 #include "stdio-util.h"
49 #include "string-table.h"
50 #include "string-util.h"
51 #include "strv.h"
52 #include "tmpfile-util.h"
53 #include "udev-util.h"
54 #include "user-util.h"
55 #include "xattr-util.h"
56
57 /* how many times to wait for the device nodes to appear */
58 #define N_DEVICE_NODE_LIST_ATTEMPTS 10
59
60 int probe_filesystem(const char *node, char **ret_fstype) {
61 /* Try to find device content type and return it in *ret_fstype. If nothing is found,
62 * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and an
63 * different error otherwise. */
64
65 #if HAVE_BLKID
66 _cleanup_(blkid_free_probep) blkid_probe b = NULL;
67 const char *fstype;
68 int r;
69
70 errno = 0;
71 b = blkid_new_probe_from_filename(node);
72 if (!b)
73 return errno_or_else(ENOMEM);
74
75 blkid_probe_enable_superblocks(b, 1);
76 blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE);
77
78 errno = 0;
79 r = blkid_do_safeprobe(b);
80 if (r == 1) {
81 log_debug("No type detected on partition %s", node);
82 goto not_found;
83 }
84 if (r == -2)
85 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN),
86 "Results ambiguous for partition %s", node);
87 if (r != 0)
88 return errno_or_else(EIO);
89
90 (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
91
92 if (fstype) {
93 char *t;
94
95 t = strdup(fstype);
96 if (!t)
97 return -ENOMEM;
98
99 *ret_fstype = t;
100 return 1;
101 }
102
103 not_found:
104 *ret_fstype = NULL;
105 return 0;
106 #else
107 return -EOPNOTSUPP;
108 #endif
109 }
110
111 #if HAVE_BLKID
112 static int enumerator_for_parent(sd_device *d, sd_device_enumerator **ret) {
113 _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
114 int r;
115
116 assert(d);
117 assert(ret);
118
119 r = sd_device_enumerator_new(&e);
120 if (r < 0)
121 return r;
122
123 r = sd_device_enumerator_allow_uninitialized(e);
124 if (r < 0)
125 return r;
126
127 r = sd_device_enumerator_add_match_parent(e, d);
128 if (r < 0)
129 return r;
130
131 *ret = TAKE_PTR(e);
132 return 0;
133 }
134
135 static int device_is_partition(sd_device *d, blkid_partition pp) {
136 blkid_loff_t bsize, bstart;
137 uint64_t size, start;
138 int partno, bpartno, r;
139 const char *ss, *v;
140
141 assert(d);
142 assert(pp);
143
144 r = sd_device_get_subsystem(d, &ss);
145 if (r < 0)
146 return r;
147 if (!streq(ss, "block"))
148 return false;
149
150 r = sd_device_get_sysattr_value(d, "partition", &v);
151 if (r == -ENOENT) /* Not a partition device */
152 return false;
153 if (r < 0)
154 return r;
155 r = safe_atoi(v, &partno);
156 if (r < 0)
157 return r;
158
159 errno = 0;
160 bpartno = blkid_partition_get_partno(pp);
161 if (bpartno < 0)
162 return errno_or_else(EIO);
163
164 if (partno != bpartno)
165 return false;
166
167 r = sd_device_get_sysattr_value(d, "start", &v);
168 if (r < 0)
169 return r;
170 r = safe_atou64(v, &start);
171 if (r < 0)
172 return r;
173
174 errno = 0;
175 bstart = blkid_partition_get_start(pp);
176 if (bstart < 0)
177 return errno_or_else(EIO);
178
179 if (start != (uint64_t) bstart)
180 return false;
181
182 r = sd_device_get_sysattr_value(d, "size", &v);
183 if (r < 0)
184 return r;
185 r = safe_atou64(v, &size);
186 if (r < 0)
187 return r;
188
189 errno = 0;
190 bsize = blkid_partition_get_size(pp);
191 if (bsize < 0)
192 return errno_or_else(EIO);
193
194 if (size != (uint64_t) bsize)
195 return false;
196
197 return true;
198 }
199
200 static int find_partition(
201 sd_device *parent,
202 blkid_partition pp,
203 sd_device **ret) {
204
205 _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
206 sd_device *q;
207 int r;
208
209 assert(parent);
210 assert(pp);
211 assert(ret);
212
213 r = enumerator_for_parent(parent, &e);
214 if (r < 0)
215 return r;
216
217 FOREACH_DEVICE(e, q) {
218 r = device_is_partition(q, pp);
219 if (r < 0)
220 return r;
221 if (r > 0) {
222 *ret = sd_device_ref(q);
223 return 0;
224 }
225 }
226
227 return -ENXIO;
228 }
229
230 struct wait_data {
231 sd_device *parent_device;
232 blkid_partition blkidp;
233 sd_device *found;
234 };
235
236 static inline void wait_data_done(struct wait_data *d) {
237 sd_device_unref(d->found);
238 }
239
240 static int device_monitor_handler(sd_device_monitor *monitor, sd_device *device, void *userdata) {
241 const char *parent1_path, *parent2_path;
242 struct wait_data *w = userdata;
243 sd_device *pp;
244 int r;
245
246 assert(w);
247
248 if (device_for_action(device, DEVICE_ACTION_REMOVE))
249 return 0;
250
251 r = sd_device_get_parent(device, &pp);
252 if (r < 0)
253 return 0; /* Doesn't have a parent? No relevant to us */
254
255 r = sd_device_get_syspath(pp, &parent1_path); /* Check parent of device of this action */
256 if (r < 0)
257 goto finish;
258
259 r = sd_device_get_syspath(w->parent_device, &parent2_path); /* Check parent of device we are looking for */
260 if (r < 0)
261 goto finish;
262
263 if (!path_equal(parent1_path, parent2_path))
264 return 0; /* Has a different parent than what we need, not interesting to us */
265
266 r = device_is_partition(device, w->blkidp);
267 if (r < 0)
268 goto finish;
269 if (r == 0) /* Not the one we need */
270 return 0;
271
272 /* It's the one we need! Yay! */
273 assert(!w->found);
274 w->found = sd_device_ref(device);
275 r = 0;
276
277 finish:
278 return sd_event_exit(sd_device_monitor_get_event(monitor), r);
279 }
280
281 static int wait_for_partition_device(
282 sd_device *parent,
283 blkid_partition pp,
284 usec_t deadline,
285 sd_device **ret) {
286
287 _cleanup_(sd_event_source_unrefp) sd_event_source *timeout_source = NULL;
288 _cleanup_(sd_device_monitor_unrefp) sd_device_monitor *monitor = NULL;
289 _cleanup_(sd_event_unrefp) sd_event *event = NULL;
290 int r;
291
292 assert(parent);
293 assert(pp);
294 assert(ret);
295
296 r = find_partition(parent, pp, ret);
297 if (r != -ENXIO)
298 return r;
299
300 r = sd_event_new(&event);
301 if (r < 0)
302 return r;
303
304 r = sd_device_monitor_new(&monitor);
305 if (r < 0)
306 return r;
307
308 r = sd_device_monitor_filter_add_match_subsystem_devtype(monitor, "block", "partition");
309 if (r < 0)
310 return r;
311
312 r = sd_device_monitor_attach_event(monitor, event);
313 if (r < 0)
314 return r;
315
316 _cleanup_(wait_data_done) struct wait_data w = {
317 .parent_device = parent,
318 .blkidp = pp,
319 };
320
321 r = sd_device_monitor_start(monitor, device_monitor_handler, &w);
322 if (r < 0)
323 return r;
324
325 /* Check again, the partition might have appeared in the meantime */
326 r = find_partition(parent, pp, ret);
327 if (r != -ENXIO)
328 return r;
329
330 if (deadline != USEC_INFINITY) {
331 r = sd_event_add_time(
332 event, &timeout_source,
333 CLOCK_MONOTONIC, deadline, 0,
334 NULL, INT_TO_PTR(-ETIMEDOUT));
335 if (r < 0)
336 return r;
337 }
338
339 r = sd_event_loop(event);
340 if (r < 0)
341 return r;
342
343 assert(w.found);
344 *ret = TAKE_PTR(w.found);
345 return 0;
346 }
347
348 static void check_partition_flags(
349 const char *node,
350 unsigned long long pflags,
351 unsigned long long supported) {
352
353 assert(node);
354
355 /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
356 pflags &= ~(supported | GPT_FLAG_REQUIRED_PARTITION | GPT_FLAG_NO_BLOCK_IO_PROTOCOL | GPT_FLAG_LEGACY_BIOS_BOOTABLE);
357
358 if (pflags == 0)
359 return;
360
361 /* If there are other bits set, then log about it, to make things discoverable */
362 for (unsigned i = 0; i < sizeof(pflags) * 8; i++) {
363 unsigned long long bit = 1ULL << i;
364 if (!FLAGS_SET(pflags, bit))
365 continue;
366
367 log_debug("Unexpected partition flag %llu set on %s!", bit, node);
368 }
369 }
370
371 static int device_wait_for_initialization_harder(
372 sd_device *device,
373 const char *subsystem,
374 usec_t deadline,
375 sd_device **ret) {
376
377 _cleanup_free_ char *uevent = NULL;
378 usec_t start, left, retrigger_timeout;
379 int r;
380
381 start = now(CLOCK_MONOTONIC);
382 left = usec_sub_unsigned(deadline, start);
383
384 if (DEBUG_LOGGING) {
385 char buf[FORMAT_TIMESPAN_MAX];
386 const char *sn = NULL;
387
388 (void) sd_device_get_sysname(device, &sn);
389 log_debug("Waiting for device '%s' to initialize for %s.", strna(sn), format_timespan(buf, sizeof(buf), left, 0));
390 }
391
392 if (left != USEC_INFINITY)
393 retrigger_timeout = CLAMP(left / 4, 1 * USEC_PER_SEC, 5 * USEC_PER_SEC); /* A fourth of the total timeout, but let's clamp to 1s…5s range */
394 else
395 retrigger_timeout = 2 * USEC_PER_SEC;
396
397 for (;;) {
398 usec_t local_deadline, n;
399 bool last_try;
400
401 n = now(CLOCK_MONOTONIC);
402 assert(n >= start);
403
404 /* Find next deadline, when we'll retrigger */
405 local_deadline = start +
406 DIV_ROUND_UP(n - start, retrigger_timeout) * retrigger_timeout;
407
408 if (deadline != USEC_INFINITY && deadline <= local_deadline) {
409 local_deadline = deadline;
410 last_try = true;
411 } else
412 last_try = false;
413
414 r = device_wait_for_initialization(device, subsystem, local_deadline, ret);
415 if (r >= 0 && DEBUG_LOGGING) {
416 char buf[FORMAT_TIMESPAN_MAX];
417 const char *sn = NULL;
418
419 (void) sd_device_get_sysname(device, &sn);
420 log_debug("Successfully waited for device '%s' to initialize for %s.", strna(sn), format_timespan(buf, sizeof(buf), usec_sub_unsigned(now(CLOCK_MONOTONIC), start), 0));
421
422 }
423 if (r != -ETIMEDOUT || last_try)
424 return r;
425
426 if (!uevent) {
427 const char *syspath;
428
429 r = sd_device_get_syspath(device, &syspath);
430 if (r < 0)
431 return r;
432
433 uevent = path_join(syspath, "uevent");
434 if (!uevent)
435 return -ENOMEM;
436 }
437
438 if (DEBUG_LOGGING) {
439 char buf[FORMAT_TIMESPAN_MAX];
440
441 log_debug("Device didn't initialize within %s, assuming lost event. Retriggering device through %s.",
442 format_timespan(buf, sizeof(buf), usec_sub_unsigned(now(CLOCK_MONOTONIC), start), 0),
443 uevent);
444 }
445
446 r = write_string_file(uevent, "change", WRITE_STRING_FILE_DISABLE_BUFFER);
447 if (r < 0)
448 return r;
449 }
450 }
451 #endif
452
453 #define DEVICE_TIMEOUT_USEC (45 * USEC_PER_SEC)
454
455 int dissect_image(
456 int fd,
457 const VeritySettings *verity,
458 const MountOptions *mount_options,
459 DissectImageFlags flags,
460 DissectedImage **ret) {
461
462 #if HAVE_BLKID
463 #ifdef GPT_ROOT_NATIVE
464 sd_id128_t root_uuid = SD_ID128_NULL, root_verity_uuid = SD_ID128_NULL;
465 #endif
466 #ifdef GPT_USR_NATIVE
467 sd_id128_t usr_uuid = SD_ID128_NULL, usr_verity_uuid = SD_ID128_NULL;
468 #endif
469 bool is_gpt, is_mbr, generic_rw, multiple_generic = false;
470 _cleanup_(sd_device_unrefp) sd_device *d = NULL;
471 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
472 _cleanup_(blkid_free_probep) blkid_probe b = NULL;
473 _cleanup_free_ char *generic_node = NULL;
474 sd_id128_t generic_uuid = SD_ID128_NULL;
475 const char *pttype = NULL;
476 blkid_partlist pl;
477 int r, generic_nr, n_partitions;
478 struct stat st;
479 usec_t deadline;
480
481 assert(fd >= 0);
482 assert(ret);
483 assert(!verity || verity->root_hash || verity->root_hash_size == 0);
484 assert(!((flags & DISSECT_IMAGE_GPT_ONLY) && (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)));
485
486 /* Probes a disk image, and returns information about what it found in *ret.
487 *
488 * Returns -ENOPKG if no suitable partition table or file system could be found.
489 * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found. */
490
491 if (verity && verity->root_hash) {
492 sd_id128_t fsuuid, vuuid;
493
494 /* If a root hash is supplied, then we use the root partition that has a UUID that match the
495 * first 128bit of the root hash. And we use the verity partition that has a UUID that match
496 * the final 128bit. */
497
498 if (verity->root_hash_size < sizeof(sd_id128_t))
499 return -EINVAL;
500
501 memcpy(&fsuuid, verity->root_hash, sizeof(sd_id128_t));
502 memcpy(&vuuid, (const uint8_t*) verity->root_hash + verity->root_hash_size - sizeof(sd_id128_t), sizeof(sd_id128_t));
503
504 if (sd_id128_is_null(fsuuid))
505 return -EINVAL;
506 if (sd_id128_is_null(vuuid))
507 return -EINVAL;
508
509 /* If the verity data declares it's for the /usr partition, then search for that, in all
510 * other cases assume it's for the root partition. */
511 #ifdef GPT_USR_NATIVE
512 if (verity->designator == PARTITION_USR) {
513 usr_uuid = fsuuid;
514 usr_verity_uuid = vuuid;
515 } else {
516 #endif
517 #ifdef GPT_ROOT_NATIVE
518 root_uuid = fsuuid;
519 root_verity_uuid = vuuid;
520 #endif
521 #ifdef GPT_USR_NATIVE
522 }
523 #endif
524 }
525
526 if (fstat(fd, &st) < 0)
527 return -errno;
528
529 if (!S_ISBLK(st.st_mode))
530 return -ENOTBLK;
531
532 r = sd_device_new_from_devnum(&d, 'b', st.st_rdev);
533 if (r < 0)
534 return r;
535
536 if (!FLAGS_SET(flags, DISSECT_IMAGE_NO_UDEV)) {
537 _cleanup_(sd_device_unrefp) sd_device *initialized = NULL;
538
539 /* If udev support is enabled, then let's wait for the device to be initialized before we doing anything. */
540
541 r = device_wait_for_initialization_harder(
542 d,
543 "block",
544 usec_add(now(CLOCK_MONOTONIC), DEVICE_TIMEOUT_USEC),
545 &initialized);
546 if (r < 0)
547 return r;
548
549 sd_device_unref(d);
550 d = TAKE_PTR(initialized);
551 }
552
553 b = blkid_new_probe();
554 if (!b)
555 return -ENOMEM;
556
557 errno = 0;
558 r = blkid_probe_set_device(b, fd, 0, 0);
559 if (r != 0)
560 return errno_or_else(ENOMEM);
561
562 if ((flags & DISSECT_IMAGE_GPT_ONLY) == 0) {
563 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
564 blkid_probe_enable_superblocks(b, 1);
565 blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE|BLKID_SUBLKS_USAGE);
566 }
567
568 blkid_probe_enable_partitions(b, 1);
569 blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS);
570
571 errno = 0;
572 r = blkid_do_safeprobe(b);
573 if (IN_SET(r, -2, 1))
574 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG), "Failed to identify any partition table.");
575 if (r != 0)
576 return errno_or_else(EIO);
577
578 m = new0(DissectedImage, 1);
579 if (!m)
580 return -ENOMEM;
581
582 if ((!(flags & DISSECT_IMAGE_GPT_ONLY) &&
583 (flags & DISSECT_IMAGE_REQUIRE_ROOT)) ||
584 (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)) {
585 const char *usage = NULL;
586
587 /* If flags permit this, also allow using non-partitioned single-filesystem images */
588
589 (void) blkid_probe_lookup_value(b, "USAGE", &usage, NULL);
590 if (STRPTR_IN_SET(usage, "filesystem", "crypto")) {
591 const char *fstype = NULL, *options = NULL, *devname = NULL;
592 _cleanup_free_ char *t = NULL, *n = NULL, *o = NULL;
593
594 /* OK, we have found a file system, that's our root partition then. */
595 (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
596
597 if (fstype) {
598 t = strdup(fstype);
599 if (!t)
600 return -ENOMEM;
601 }
602
603 r = sd_device_get_devname(d, &devname);
604 if (r < 0)
605 return r;
606
607 n = strdup(devname);
608 if (!n)
609 return -ENOMEM;
610
611 m->single_file_system = true;
612 m->verity = verity && verity->root_hash && verity->data_path && (verity->designator < 0 || verity->designator == PARTITION_ROOT);
613 m->can_verity = verity && verity->data_path;
614
615 options = mount_options_from_designator(mount_options, PARTITION_ROOT);
616 if (options) {
617 o = strdup(options);
618 if (!o)
619 return -ENOMEM;
620 }
621
622 m->partitions[PARTITION_ROOT] = (DissectedPartition) {
623 .found = true,
624 .rw = !m->verity,
625 .partno = -1,
626 .architecture = _ARCHITECTURE_INVALID,
627 .fstype = TAKE_PTR(t),
628 .node = TAKE_PTR(n),
629 .mount_options = TAKE_PTR(o),
630 };
631
632 m->encrypted = streq_ptr(fstype, "crypto_LUKS");
633
634 *ret = TAKE_PTR(m);
635 return 0;
636 }
637 }
638
639 (void) blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL);
640 if (!pttype)
641 return -ENOPKG;
642
643 is_gpt = streq_ptr(pttype, "gpt");
644 is_mbr = streq_ptr(pttype, "dos");
645
646 if (!is_gpt && ((flags & DISSECT_IMAGE_GPT_ONLY) || !is_mbr))
647 return -ENOPKG;
648
649 /* Safety check: refuse block devices that carry a partition table but for which the kernel doesn't
650 * do partition scanning. */
651 r = blockdev_partscan_enabled(fd);
652 if (r < 0)
653 return r;
654 if (r == 0)
655 return -EPROTONOSUPPORT;
656
657 errno = 0;
658 pl = blkid_probe_get_partitions(b);
659 if (!pl)
660 return errno_or_else(ENOMEM);
661
662 errno = 0;
663 n_partitions = blkid_partlist_numof_partitions(pl);
664 if (n_partitions < 0)
665 return errno_or_else(EIO);
666
667 deadline = usec_add(now(CLOCK_MONOTONIC), DEVICE_TIMEOUT_USEC);
668 for (int i = 0; i < n_partitions; i++) {
669 _cleanup_(sd_device_unrefp) sd_device *q = NULL;
670 unsigned long long pflags;
671 blkid_partition pp;
672 const char *node;
673 int nr;
674
675 errno = 0;
676 pp = blkid_partlist_get_partition(pl, i);
677 if (!pp)
678 return errno_or_else(EIO);
679
680 r = wait_for_partition_device(d, pp, deadline, &q);
681 if (r < 0)
682 return r;
683
684 r = sd_device_get_devname(q, &node);
685 if (r < 0)
686 return r;
687
688 pflags = blkid_partition_get_flags(pp);
689
690 errno = 0;
691 nr = blkid_partition_get_partno(pp);
692 if (nr < 0)
693 return errno_or_else(EIO);
694
695 if (is_gpt) {
696 PartitionDesignator designator = _PARTITION_DESIGNATOR_INVALID;
697 int architecture = _ARCHITECTURE_INVALID;
698 const char *stype, *sid, *fstype = NULL;
699 sd_id128_t type_id, id;
700 bool rw = true;
701
702 sid = blkid_partition_get_uuid(pp);
703 if (!sid)
704 continue;
705 if (sd_id128_from_string(sid, &id) < 0)
706 continue;
707
708 stype = blkid_partition_get_type_string(pp);
709 if (!stype)
710 continue;
711 if (sd_id128_from_string(stype, &type_id) < 0)
712 continue;
713
714 if (sd_id128_equal(type_id, GPT_HOME)) {
715
716 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
717
718 if (pflags & GPT_FLAG_NO_AUTO)
719 continue;
720
721 designator = PARTITION_HOME;
722 rw = !(pflags & GPT_FLAG_READ_ONLY);
723
724 } else if (sd_id128_equal(type_id, GPT_SRV)) {
725
726 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
727
728 if (pflags & GPT_FLAG_NO_AUTO)
729 continue;
730
731 designator = PARTITION_SRV;
732 rw = !(pflags & GPT_FLAG_READ_ONLY);
733
734 } else if (sd_id128_equal(type_id, GPT_ESP)) {
735
736 /* Note that we don't check the GPT_FLAG_NO_AUTO flag for the ESP, as it is
737 * not defined there. We instead check the GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as
738 * recommended by the UEFI spec (See "12.3.3 Number and Location of System
739 * Partitions"). */
740
741 if (pflags & GPT_FLAG_NO_BLOCK_IO_PROTOCOL)
742 continue;
743
744 designator = PARTITION_ESP;
745 fstype = "vfat";
746
747 } else if (sd_id128_equal(type_id, GPT_XBOOTLDR)) {
748
749 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
750
751 if (pflags & GPT_FLAG_NO_AUTO)
752 continue;
753
754 designator = PARTITION_XBOOTLDR;
755 rw = !(pflags & GPT_FLAG_READ_ONLY);
756 }
757 #ifdef GPT_ROOT_NATIVE
758 else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE)) {
759
760 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
761
762 if (pflags & GPT_FLAG_NO_AUTO)
763 continue;
764
765 /* If a root ID is specified, ignore everything but the root id */
766 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
767 continue;
768
769 designator = PARTITION_ROOT;
770 architecture = native_architecture();
771 rw = !(pflags & GPT_FLAG_READ_ONLY);
772
773 } else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE_VERITY)) {
774
775 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
776
777 if (pflags & GPT_FLAG_NO_AUTO)
778 continue;
779
780 m->can_verity = true;
781
782 /* Ignore verity unless a root hash is specified */
783 if (sd_id128_is_null(root_verity_uuid) || !sd_id128_equal(root_verity_uuid, id))
784 continue;
785
786 designator = PARTITION_ROOT_VERITY;
787 fstype = "DM_verity_hash";
788 architecture = native_architecture();
789 rw = false;
790 }
791 #endif
792 #ifdef GPT_ROOT_SECONDARY
793 else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY)) {
794
795 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
796
797 if (pflags & GPT_FLAG_NO_AUTO)
798 continue;
799
800 /* If a root ID is specified, ignore everything but the root id */
801 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
802 continue;
803
804 designator = PARTITION_ROOT_SECONDARY;
805 architecture = SECONDARY_ARCHITECTURE;
806 rw = !(pflags & GPT_FLAG_READ_ONLY);
807
808 } else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY_VERITY)) {
809
810 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
811
812 if (pflags & GPT_FLAG_NO_AUTO)
813 continue;
814
815 m->can_verity = true;
816
817 /* Ignore verity unless root has is specified */
818 if (sd_id128_is_null(root_verity_uuid) || !sd_id128_equal(root_verity_uuid, id))
819 continue;
820
821 designator = PARTITION_ROOT_SECONDARY_VERITY;
822 fstype = "DM_verity_hash";
823 architecture = SECONDARY_ARCHITECTURE;
824 rw = false;
825 }
826 #endif
827 #ifdef GPT_USR_NATIVE
828 else if (sd_id128_equal(type_id, GPT_USR_NATIVE)) {
829
830 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
831
832 if (pflags & GPT_FLAG_NO_AUTO)
833 continue;
834
835 /* If a usr ID is specified, ignore everything but the usr id */
836 if (!sd_id128_is_null(usr_uuid) && !sd_id128_equal(usr_uuid, id))
837 continue;
838
839 designator = PARTITION_USR;
840 architecture = native_architecture();
841 rw = !(pflags & GPT_FLAG_READ_ONLY);
842
843 } else if (sd_id128_equal(type_id, GPT_USR_NATIVE_VERITY)) {
844
845 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
846
847 if (pflags & GPT_FLAG_NO_AUTO)
848 continue;
849
850 m->can_verity = true;
851
852 /* Ignore verity unless a usr hash is specified */
853 if (sd_id128_is_null(usr_verity_uuid) || !sd_id128_equal(usr_verity_uuid, id))
854 continue;
855
856 designator = PARTITION_USR_VERITY;
857 fstype = "DM_verity_hash";
858 architecture = native_architecture();
859 rw = false;
860 }
861 #endif
862 #ifdef GPT_USR_SECONDARY
863 else if (sd_id128_equal(type_id, GPT_USR_SECONDARY)) {
864
865 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
866
867 if (pflags & GPT_FLAG_NO_AUTO)
868 continue;
869
870 /* If a usr ID is specified, ignore everything but the usr id */
871 if (!sd_id128_is_null(usr_uuid) && !sd_id128_equal(usr_uuid, id))
872 continue;
873
874 designator = PARTITION_USR_SECONDARY;
875 architecture = SECONDARY_ARCHITECTURE;
876 rw = !(pflags & GPT_FLAG_READ_ONLY);
877
878 } else if (sd_id128_equal(type_id, GPT_USR_SECONDARY_VERITY)) {
879
880 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
881
882 if (pflags & GPT_FLAG_NO_AUTO)
883 continue;
884
885 m->can_verity = true;
886
887 /* Ignore verity unless usr has is specified */
888 if (sd_id128_is_null(usr_verity_uuid) || !sd_id128_equal(usr_verity_uuid, id))
889 continue;
890
891 designator = PARTITION_USR_SECONDARY_VERITY;
892 fstype = "DM_verity_hash";
893 architecture = SECONDARY_ARCHITECTURE;
894 rw = false;
895 }
896 #endif
897 else if (sd_id128_equal(type_id, GPT_SWAP)) {
898
899 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO);
900
901 if (pflags & GPT_FLAG_NO_AUTO)
902 continue;
903
904 designator = PARTITION_SWAP;
905 fstype = "swap";
906
907 } else if (sd_id128_equal(type_id, GPT_LINUX_GENERIC)) {
908
909 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
910
911 if (pflags & GPT_FLAG_NO_AUTO)
912 continue;
913
914 if (generic_node)
915 multiple_generic = true;
916 else {
917 generic_nr = nr;
918 generic_rw = !(pflags & GPT_FLAG_READ_ONLY);
919 generic_uuid = id;
920 generic_node = strdup(node);
921 if (!generic_node)
922 return -ENOMEM;
923 }
924
925 } else if (sd_id128_equal(type_id, GPT_TMP)) {
926
927 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
928
929 if (pflags & GPT_FLAG_NO_AUTO)
930 continue;
931
932 designator = PARTITION_TMP;
933 rw = !(pflags & GPT_FLAG_READ_ONLY);
934
935 } else if (sd_id128_equal(type_id, GPT_VAR)) {
936
937 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
938
939 if (pflags & GPT_FLAG_NO_AUTO)
940 continue;
941
942 if (!FLAGS_SET(flags, DISSECT_IMAGE_RELAX_VAR_CHECK)) {
943 sd_id128_t var_uuid;
944
945 /* For /var we insist that the uuid of the partition matches the
946 * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
947 * ID. Why? Unlike the other partitions /var is inherently
948 * installation specific, hence we need to be careful not to mount it
949 * in the wrong installation. By hashing the partition UUID from
950 * /etc/machine-id we can securely bind the partition to the
951 * installation. */
952
953 r = sd_id128_get_machine_app_specific(GPT_VAR, &var_uuid);
954 if (r < 0)
955 return r;
956
957 if (!sd_id128_equal(var_uuid, id)) {
958 log_debug("Found a /var/ partition, but its UUID didn't match our expectations, ignoring.");
959 continue;
960 }
961 }
962
963 designator = PARTITION_VAR;
964 rw = !(pflags & GPT_FLAG_READ_ONLY);
965 }
966
967 if (designator != _PARTITION_DESIGNATOR_INVALID) {
968 _cleanup_free_ char *t = NULL, *n = NULL, *o = NULL;
969 const char *options = NULL;
970
971 /* First one wins */
972 if (m->partitions[designator].found)
973 continue;
974
975 if (fstype) {
976 t = strdup(fstype);
977 if (!t)
978 return -ENOMEM;
979 }
980
981 n = strdup(node);
982 if (!n)
983 return -ENOMEM;
984
985 options = mount_options_from_designator(mount_options, designator);
986 if (options) {
987 o = strdup(options);
988 if (!o)
989 return -ENOMEM;
990 }
991
992 m->partitions[designator] = (DissectedPartition) {
993 .found = true,
994 .partno = nr,
995 .rw = rw,
996 .architecture = architecture,
997 .node = TAKE_PTR(n),
998 .fstype = TAKE_PTR(t),
999 .uuid = id,
1000 .mount_options = TAKE_PTR(o),
1001 };
1002 }
1003
1004 } else if (is_mbr) {
1005
1006 switch (blkid_partition_get_type(pp)) {
1007
1008 case 0x83: /* Linux partition */
1009
1010 if (pflags != 0x80) /* Bootable flag */
1011 continue;
1012
1013 if (generic_node)
1014 multiple_generic = true;
1015 else {
1016 generic_nr = nr;
1017 generic_rw = true;
1018 generic_node = strdup(node);
1019 if (!generic_node)
1020 return -ENOMEM;
1021 }
1022
1023 break;
1024
1025 case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
1026 _cleanup_free_ char *n = NULL, *o = NULL;
1027 sd_id128_t id = SD_ID128_NULL;
1028 const char *sid, *options = NULL;
1029
1030 /* First one wins */
1031 if (m->partitions[PARTITION_XBOOTLDR].found)
1032 continue;
1033
1034 sid = blkid_partition_get_uuid(pp);
1035 if (sid)
1036 (void) sd_id128_from_string(sid, &id);
1037
1038 n = strdup(node);
1039 if (!n)
1040 return -ENOMEM;
1041
1042 options = mount_options_from_designator(mount_options, PARTITION_XBOOTLDR);
1043 if (options) {
1044 o = strdup(options);
1045 if (!o)
1046 return -ENOMEM;
1047 }
1048
1049 m->partitions[PARTITION_XBOOTLDR] = (DissectedPartition) {
1050 .found = true,
1051 .partno = nr,
1052 .rw = true,
1053 .architecture = _ARCHITECTURE_INVALID,
1054 .node = TAKE_PTR(n),
1055 .uuid = id,
1056 .mount_options = TAKE_PTR(o),
1057 };
1058
1059 break;
1060 }}
1061 }
1062 }
1063
1064 if (m->partitions[PARTITION_ROOT].found) {
1065 /* If we found the primary arch, then invalidate the secondary arch to avoid any ambiguities,
1066 * since we never want to mount the secondary arch in this case. */
1067 m->partitions[PARTITION_ROOT_SECONDARY].found = false;
1068 m->partitions[PARTITION_ROOT_SECONDARY_VERITY].found = false;
1069 m->partitions[PARTITION_USR_SECONDARY].found = false;
1070 m->partitions[PARTITION_USR_SECONDARY_VERITY].found = false;
1071 } else {
1072 /* No root partition found? Then let's see if ther's one for the secondary architecture. And if not
1073 * either, then check if there's a single generic one, and use that. */
1074
1075 if (m->partitions[PARTITION_ROOT_VERITY].found)
1076 return -EADDRNOTAVAIL;
1077
1078 /* We didn't find a primary architecture root, but we found a primary architecture /usr? Refuse that for now. */
1079 if (m->partitions[PARTITION_USR].found || m->partitions[PARTITION_USR_VERITY].found)
1080 return -EADDRNOTAVAIL;
1081
1082 if (m->partitions[PARTITION_ROOT_SECONDARY].found) {
1083 /* Upgrade secondary arch to first */
1084 m->partitions[PARTITION_ROOT] = m->partitions[PARTITION_ROOT_SECONDARY];
1085 zero(m->partitions[PARTITION_ROOT_SECONDARY]);
1086 m->partitions[PARTITION_ROOT_VERITY] = m->partitions[PARTITION_ROOT_SECONDARY_VERITY];
1087 zero(m->partitions[PARTITION_ROOT_SECONDARY_VERITY]);
1088
1089 m->partitions[PARTITION_USR] = m->partitions[PARTITION_USR_SECONDARY];
1090 zero(m->partitions[PARTITION_USR_SECONDARY]);
1091 m->partitions[PARTITION_USR_VERITY] = m->partitions[PARTITION_USR_SECONDARY_VERITY];
1092 zero(m->partitions[PARTITION_USR_SECONDARY_VERITY]);
1093
1094 } else if (flags & DISSECT_IMAGE_REQUIRE_ROOT) {
1095 _cleanup_free_ char *o = NULL;
1096 const char *options = NULL;
1097
1098 /* If the root hash was set, then we won't fall back to a generic node, because the
1099 * root hash decides. */
1100 if (verity && verity->root_hash)
1101 return -EADDRNOTAVAIL;
1102
1103 /* If we didn't find a generic node, then we can't fix this up either */
1104 if (!generic_node)
1105 return -ENXIO;
1106
1107 /* If we didn't find a properly marked root partition, but we did find a single suitable
1108 * generic Linux partition, then use this as root partition, if the caller asked for it. */
1109 if (multiple_generic)
1110 return -ENOTUNIQ;
1111
1112 options = mount_options_from_designator(mount_options, PARTITION_ROOT);
1113 if (options) {
1114 o = strdup(options);
1115 if (!o)
1116 return -ENOMEM;
1117 }
1118
1119 m->partitions[PARTITION_ROOT] = (DissectedPartition) {
1120 .found = true,
1121 .rw = generic_rw,
1122 .partno = generic_nr,
1123 .architecture = _ARCHITECTURE_INVALID,
1124 .node = TAKE_PTR(generic_node),
1125 .uuid = generic_uuid,
1126 .mount_options = TAKE_PTR(o),
1127 };
1128 }
1129 }
1130
1131 /* Refuse if we found a verity partition for /usr but no matching file system partition */
1132 if (!m->partitions[PARTITION_USR].found && m->partitions[PARTITION_USR_VERITY].found)
1133 return -EADDRNOTAVAIL;
1134
1135 /* Combinations of verity /usr with verity-less root is OK, but the reverse is not */
1136 if (m->partitions[PARTITION_ROOT_VERITY].found && m->partitions[PARTITION_USR].found && !m->partitions[PARTITION_USR_VERITY].found)
1137 return -EADDRNOTAVAIL;
1138
1139 if (verity && verity->root_hash) {
1140 if (verity->designator < 0 || verity->designator == PARTITION_ROOT) {
1141 if (!m->partitions[PARTITION_ROOT_VERITY].found || !m->partitions[PARTITION_ROOT].found)
1142 return -EADDRNOTAVAIL;
1143
1144 /* If we found a verity setup, then the root partition is necessarily read-only. */
1145 m->partitions[PARTITION_ROOT].rw = false;
1146 m->verity = true;
1147 }
1148
1149 if (verity->designator == PARTITION_USR) {
1150 if (!m->partitions[PARTITION_USR_VERITY].found || !m->partitions[PARTITION_USR].found)
1151 return -EADDRNOTAVAIL;
1152
1153 m->partitions[PARTITION_USR].rw = false;
1154 m->verity = true;
1155 }
1156 }
1157
1158 blkid_free_probe(b);
1159 b = NULL;
1160
1161 /* Fill in file system types if we don't know them yet. */
1162 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
1163 DissectedPartition *p = m->partitions + i;
1164
1165 if (!p->found)
1166 continue;
1167
1168 if (!p->fstype && p->node) {
1169 r = probe_filesystem(p->node, &p->fstype);
1170 if (r < 0 && r != -EUCLEAN)
1171 return r;
1172 }
1173
1174 if (streq_ptr(p->fstype, "crypto_LUKS"))
1175 m->encrypted = true;
1176
1177 if (p->fstype && fstype_is_ro(p->fstype))
1178 p->rw = false;
1179 }
1180
1181 *ret = TAKE_PTR(m);
1182 return 0;
1183 #else
1184 return -EOPNOTSUPP;
1185 #endif
1186 }
1187
1188 DissectedImage* dissected_image_unref(DissectedImage *m) {
1189 if (!m)
1190 return NULL;
1191
1192 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
1193 free(m->partitions[i].fstype);
1194 free(m->partitions[i].node);
1195 free(m->partitions[i].decrypted_fstype);
1196 free(m->partitions[i].decrypted_node);
1197 free(m->partitions[i].mount_options);
1198 }
1199
1200 free(m->hostname);
1201 strv_free(m->machine_info);
1202 strv_free(m->os_release);
1203
1204 return mfree(m);
1205 }
1206
1207 static int is_loop_device(const char *path) {
1208 char s[SYS_BLOCK_PATH_MAX("/../loop/")];
1209 struct stat st;
1210
1211 assert(path);
1212
1213 if (stat(path, &st) < 0)
1214 return -errno;
1215
1216 if (!S_ISBLK(st.st_mode))
1217 return -ENOTBLK;
1218
1219 xsprintf_sys_block_path(s, "/loop/", st.st_dev);
1220 if (access(s, F_OK) < 0) {
1221 if (errno != ENOENT)
1222 return -errno;
1223
1224 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
1225 xsprintf_sys_block_path(s, "/../loop/", st.st_dev);
1226 if (access(s, F_OK) < 0)
1227 return errno == ENOENT ? false : -errno;
1228 }
1229
1230 return true;
1231 }
1232
1233 static int run_fsck(const char *node, const char *fstype) {
1234 int r, exit_status;
1235 pid_t pid;
1236
1237 assert(node);
1238 assert(fstype);
1239
1240 r = fsck_exists(fstype);
1241 if (r < 0) {
1242 log_debug_errno(r, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype);
1243 return 0;
1244 }
1245 if (r == 0) {
1246 log_debug("Not checking partition %s, as fsck for %s does not exist.", node, fstype);
1247 return 0;
1248 }
1249
1250 r = safe_fork("(fsck)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG|FORK_NULL_STDIO, &pid);
1251 if (r < 0)
1252 return log_debug_errno(r, "Failed to fork off fsck: %m");
1253 if (r == 0) {
1254 /* Child */
1255 execl("/sbin/fsck", "/sbin/fsck", "-aT", node, NULL);
1256 log_debug_errno(errno, "Failed to execl() fsck: %m");
1257 _exit(FSCK_OPERATIONAL_ERROR);
1258 }
1259
1260 exit_status = wait_for_terminate_and_check("fsck", pid, 0);
1261 if (exit_status < 0)
1262 return log_debug_errno(exit_status, "Failed to fork off /sbin/fsck: %m");
1263
1264 if ((exit_status & ~FSCK_ERROR_CORRECTED) != FSCK_SUCCESS) {
1265 log_debug("fsck failed with exit status %i.", exit_status);
1266
1267 if ((exit_status & (FSCK_SYSTEM_SHOULD_REBOOT|FSCK_ERRORS_LEFT_UNCORRECTED)) != 0)
1268 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN), "File system is corrupted, refusing.");
1269
1270 log_debug("Ignoring fsck error.");
1271 }
1272
1273 return 0;
1274 }
1275
1276 static int mount_partition(
1277 DissectedPartition *m,
1278 const char *where,
1279 const char *directory,
1280 uid_t uid_shift,
1281 DissectImageFlags flags) {
1282
1283 _cleanup_free_ char *chased = NULL, *options = NULL;
1284 const char *p, *node, *fstype;
1285 bool rw;
1286 int r;
1287
1288 assert(m);
1289 assert(where);
1290
1291 /* Use decrypted node and matching fstype if available, otherwise use the original device */
1292 node = m->decrypted_node ?: m->node;
1293 fstype = m->decrypted_node ? m->decrypted_fstype: m->fstype;
1294
1295 if (!m->found || !node)
1296 return 0;
1297 if (!fstype)
1298 return -EAFNOSUPPORT;
1299
1300 /* We are looking at an encrypted partition? This either means stacked encryption, or the caller didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error for this case. */
1301 if (streq(fstype, "crypto_LUKS"))
1302 return -EUNATCH;
1303
1304 rw = m->rw && !(flags & DISSECT_IMAGE_READ_ONLY);
1305
1306 if (FLAGS_SET(flags, DISSECT_IMAGE_FSCK) && rw) {
1307 r = run_fsck(node, fstype);
1308 if (r < 0)
1309 return r;
1310 }
1311
1312 if (directory) {
1313 if (!FLAGS_SET(flags, DISSECT_IMAGE_READ_ONLY)) {
1314 /* Automatically create missing mount points, if necessary. */
1315 r = mkdir_p_root(where, directory, uid_shift, (gid_t) uid_shift, 0755);
1316 if (r < 0)
1317 return r;
1318 }
1319
1320 r = chase_symlinks(directory, where, CHASE_PREFIX_ROOT, &chased, NULL);
1321 if (r < 0)
1322 return r;
1323
1324 p = chased;
1325 } else
1326 p = where;
1327
1328 /* If requested, turn on discard support. */
1329 if (fstype_can_discard(fstype) &&
1330 ((flags & DISSECT_IMAGE_DISCARD) ||
1331 ((flags & DISSECT_IMAGE_DISCARD_ON_LOOP) && is_loop_device(m->node) > 0))) {
1332 options = strdup("discard");
1333 if (!options)
1334 return -ENOMEM;
1335 }
1336
1337 if (uid_is_valid(uid_shift) && uid_shift != 0 && fstype_can_uid_gid(fstype)) {
1338 _cleanup_free_ char *uid_option = NULL;
1339
1340 if (asprintf(&uid_option, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0)
1341 return -ENOMEM;
1342
1343 if (!strextend_with_separator(&options, ",", uid_option))
1344 return -ENOMEM;
1345 }
1346
1347 if (!isempty(m->mount_options))
1348 if (!strextend_with_separator(&options, ",", m->mount_options))
1349 return -ENOMEM;
1350
1351 if (FLAGS_SET(flags, DISSECT_IMAGE_MKDIR)) {
1352 r = mkdir_p(p, 0755);
1353 if (r < 0)
1354 return r;
1355 }
1356
1357 r = mount_nofollow_verbose(LOG_DEBUG, node, p, fstype, MS_NODEV|(rw ? 0 : MS_RDONLY), options);
1358 if (r < 0)
1359 return r;
1360
1361 return 1;
1362 }
1363
1364 int dissected_image_mount(DissectedImage *m, const char *where, uid_t uid_shift, DissectImageFlags flags) {
1365 int r, xbootldr_mounted;
1366
1367 assert(m);
1368 assert(where);
1369
1370 /* Returns:
1371 *
1372 * -ENXIO → No root partition found
1373 * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release file found
1374 * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet
1375 * -EUCLEAN → fsck for file system failed
1376 * -EBUSY → File system already mounted/used elsewhere (kernel)
1377 * -EAFNOSUPPORT → File system type not supported or not known
1378 */
1379
1380 if (!m->partitions[PARTITION_ROOT].found)
1381 return -ENXIO;
1382
1383 if ((flags & DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY) == 0) {
1384 r = mount_partition(m->partitions + PARTITION_ROOT, where, NULL, uid_shift, flags);
1385 if (r < 0)
1386 return r;
1387 }
1388
1389 /* Mask DISSECT_IMAGE_MKDIR for all subdirs: the idea is that only the top-level mount point is
1390 * created if needed, but the image itself not modified. */
1391 flags &= ~DISSECT_IMAGE_MKDIR;
1392
1393 if ((flags & DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY) == 0) {
1394 /* For us mounting root always means mounting /usr as well */
1395 r = mount_partition(m->partitions + PARTITION_USR, where, "/usr", uid_shift, flags);
1396 if (r < 0)
1397 return r;
1398
1399 if (flags & DISSECT_IMAGE_VALIDATE_OS) {
1400 r = path_is_os_tree(where);
1401 if (r < 0)
1402 return r;
1403 if (r == 0)
1404 return -EMEDIUMTYPE;
1405 }
1406 }
1407
1408 if (flags & DISSECT_IMAGE_MOUNT_ROOT_ONLY)
1409 return 0;
1410
1411 r = mount_partition(m->partitions + PARTITION_HOME, where, "/home", uid_shift, flags);
1412 if (r < 0)
1413 return r;
1414
1415 r = mount_partition(m->partitions + PARTITION_SRV, where, "/srv", uid_shift, flags);
1416 if (r < 0)
1417 return r;
1418
1419 r = mount_partition(m->partitions + PARTITION_VAR, where, "/var", uid_shift, flags);
1420 if (r < 0)
1421 return r;
1422
1423 r = mount_partition(m->partitions + PARTITION_TMP, where, "/var/tmp", uid_shift, flags);
1424 if (r < 0)
1425 return r;
1426
1427 xbootldr_mounted = mount_partition(m->partitions + PARTITION_XBOOTLDR, where, "/boot", uid_shift, flags);
1428 if (xbootldr_mounted < 0)
1429 return xbootldr_mounted;
1430
1431 if (m->partitions[PARTITION_ESP].found) {
1432 int esp_done = false;
1433
1434 /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it
1435 * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */
1436
1437 r = chase_symlinks("/efi", where, CHASE_PREFIX_ROOT, NULL, NULL);
1438 if (r < 0) {
1439 if (r != -ENOENT)
1440 return r;
1441
1442 /* /efi doesn't exist. Let's see if /boot is suitable then */
1443
1444 if (!xbootldr_mounted) {
1445 _cleanup_free_ char *p = NULL;
1446
1447 r = chase_symlinks("/boot", where, CHASE_PREFIX_ROOT, &p, NULL);
1448 if (r < 0) {
1449 if (r != -ENOENT)
1450 return r;
1451 } else if (dir_is_empty(p) > 0) {
1452 /* It exists and is an empty directory. Let's mount the ESP there. */
1453 r = mount_partition(m->partitions + PARTITION_ESP, where, "/boot", uid_shift, flags);
1454 if (r < 0)
1455 return r;
1456
1457 esp_done = true;
1458 }
1459 }
1460 }
1461
1462 if (!esp_done) {
1463 /* OK, let's mount the ESP now to /efi (possibly creating the dir if missing) */
1464
1465 r = mount_partition(m->partitions + PARTITION_ESP, where, "/efi", uid_shift, flags);
1466 if (r < 0)
1467 return r;
1468 }
1469 }
1470
1471 return 0;
1472 }
1473
1474 int dissected_image_mount_and_warn(DissectedImage *m, const char *where, uid_t uid_shift, DissectImageFlags flags) {
1475 int r;
1476
1477 assert(m);
1478 assert(where);
1479
1480 r = dissected_image_mount(m, where, uid_shift, flags);
1481 if (r == -ENXIO)
1482 return log_error_errno(r, "Not root file system found in image.");
1483 if (r == -EMEDIUMTYPE)
1484 return log_error_errno(r, "No suitable os-release file in image found.");
1485 if (r == -EUNATCH)
1486 return log_error_errno(r, "Encrypted file system discovered, but decryption not requested.");
1487 if (r == -EUCLEAN)
1488 return log_error_errno(r, "File system check on image failed.");
1489 if (r == -EBUSY)
1490 return log_error_errno(r, "File system already mounted elsewhere.");
1491 if (r == -EAFNOSUPPORT)
1492 return log_error_errno(r, "File system type not supported or not known.");
1493 if (r < 0)
1494 return log_error_errno(r, "Failed to mount image: %m");
1495
1496 return r;
1497 }
1498
1499 #if HAVE_LIBCRYPTSETUP
1500 typedef struct DecryptedPartition {
1501 struct crypt_device *device;
1502 char *name;
1503 bool relinquished;
1504 } DecryptedPartition;
1505
1506 struct DecryptedImage {
1507 DecryptedPartition *decrypted;
1508 size_t n_decrypted;
1509 size_t n_allocated;
1510 };
1511 #endif
1512
1513 DecryptedImage* decrypted_image_unref(DecryptedImage* d) {
1514 #if HAVE_LIBCRYPTSETUP
1515 size_t i;
1516 int r;
1517
1518 if (!d)
1519 return NULL;
1520
1521 for (i = 0; i < d->n_decrypted; i++) {
1522 DecryptedPartition *p = d->decrypted + i;
1523
1524 if (p->device && p->name && !p->relinquished) {
1525 r = sym_crypt_deactivate_by_name(p->device, p->name, 0);
1526 if (r < 0)
1527 log_debug_errno(r, "Failed to deactivate encrypted partition %s", p->name);
1528 }
1529
1530 if (p->device)
1531 sym_crypt_free(p->device);
1532 free(p->name);
1533 }
1534
1535 free(d);
1536 #endif
1537 return NULL;
1538 }
1539
1540 #if HAVE_LIBCRYPTSETUP
1541
1542 static int make_dm_name_and_node(const void *original_node, const char *suffix, char **ret_name, char **ret_node) {
1543 _cleanup_free_ char *name = NULL, *node = NULL;
1544 const char *base;
1545
1546 assert(original_node);
1547 assert(suffix);
1548 assert(ret_name);
1549 assert(ret_node);
1550
1551 base = strrchr(original_node, '/');
1552 if (!base)
1553 base = original_node;
1554 else
1555 base++;
1556 if (isempty(base))
1557 return -EINVAL;
1558
1559 name = strjoin(base, suffix);
1560 if (!name)
1561 return -ENOMEM;
1562 if (!filename_is_valid(name))
1563 return -EINVAL;
1564
1565 node = path_join(sym_crypt_get_dir(), name);
1566 if (!node)
1567 return -ENOMEM;
1568
1569 *ret_name = TAKE_PTR(name);
1570 *ret_node = TAKE_PTR(node);
1571
1572 return 0;
1573 }
1574
1575 static int decrypt_partition(
1576 DissectedPartition *m,
1577 const char *passphrase,
1578 DissectImageFlags flags,
1579 DecryptedImage *d) {
1580
1581 _cleanup_free_ char *node = NULL, *name = NULL;
1582 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
1583 int r;
1584
1585 assert(m);
1586 assert(d);
1587
1588 if (!m->found || !m->node || !m->fstype)
1589 return 0;
1590
1591 if (!streq(m->fstype, "crypto_LUKS"))
1592 return 0;
1593
1594 if (!passphrase)
1595 return -ENOKEY;
1596
1597 r = dlopen_cryptsetup();
1598 if (r < 0)
1599 return r;
1600
1601 r = make_dm_name_and_node(m->node, "-decrypted", &name, &node);
1602 if (r < 0)
1603 return r;
1604
1605 if (!GREEDY_REALLOC0(d->decrypted, d->n_allocated, d->n_decrypted + 1))
1606 return -ENOMEM;
1607
1608 r = sym_crypt_init(&cd, m->node);
1609 if (r < 0)
1610 return log_debug_errno(r, "Failed to initialize dm-crypt: %m");
1611
1612 cryptsetup_enable_logging(cd);
1613
1614 r = sym_crypt_load(cd, CRYPT_LUKS, NULL);
1615 if (r < 0)
1616 return log_debug_errno(r, "Failed to load LUKS metadata: %m");
1617
1618 r = sym_crypt_activate_by_passphrase(cd, name, CRYPT_ANY_SLOT, passphrase, strlen(passphrase),
1619 ((flags & DISSECT_IMAGE_READ_ONLY) ? CRYPT_ACTIVATE_READONLY : 0) |
1620 ((flags & DISSECT_IMAGE_DISCARD_ON_CRYPTO) ? CRYPT_ACTIVATE_ALLOW_DISCARDS : 0));
1621 if (r < 0) {
1622 log_debug_errno(r, "Failed to activate LUKS device: %m");
1623 return r == -EPERM ? -EKEYREJECTED : r;
1624 }
1625
1626 d->decrypted[d->n_decrypted++] = (DecryptedPartition) {
1627 .name = TAKE_PTR(name),
1628 .device = TAKE_PTR(cd),
1629 };
1630
1631 m->decrypted_node = TAKE_PTR(node);
1632
1633 return 0;
1634 }
1635
1636 static int verity_can_reuse(
1637 const VeritySettings *verity,
1638 const char *name,
1639 struct crypt_device **ret_cd) {
1640
1641 /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
1642 _cleanup_free_ char *root_hash_existing = NULL;
1643 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
1644 struct crypt_params_verity crypt_params = {};
1645 size_t root_hash_existing_size;
1646 int r;
1647
1648 assert(verity);
1649 assert(name);
1650 assert(ret_cd);
1651
1652 r = sym_crypt_init_by_name(&cd, name);
1653 if (r < 0)
1654 return log_debug_errno(r, "Error opening verity device, crypt_init_by_name failed: %m");
1655
1656 r = sym_crypt_get_verity_info(cd, &crypt_params);
1657 if (r < 0)
1658 return log_debug_errno(r, "Error opening verity device, crypt_get_verity_info failed: %m");
1659
1660 root_hash_existing_size = verity->root_hash_size;
1661 root_hash_existing = malloc0(root_hash_existing_size);
1662 if (!root_hash_existing)
1663 return -ENOMEM;
1664
1665 r = sym_crypt_volume_key_get(cd, CRYPT_ANY_SLOT, root_hash_existing, &root_hash_existing_size, NULL, 0);
1666 if (r < 0)
1667 return log_debug_errno(r, "Error opening verity device, crypt_volume_key_get failed: %m");
1668 if (verity->root_hash_size != root_hash_existing_size ||
1669 memcmp(root_hash_existing, verity->root_hash, verity->root_hash_size) != 0)
1670 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but root hashes are different.");
1671
1672 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1673 /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the
1674 * same settings, so that a previous unsigned mount will not be reused if the user asks to use
1675 * signing for the new one, and vice versa. */
1676 if (!!verity->root_hash_sig != !!(crypt_params.flags & CRYPT_VERITY_ROOT_HASH_SIGNATURE))
1677 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but signature settings are not the same.");
1678 #endif
1679
1680 *ret_cd = TAKE_PTR(cd);
1681 return 0;
1682 }
1683
1684 static inline void dm_deferred_remove_clean(char *name) {
1685 if (!name)
1686 return;
1687
1688 (void) sym_crypt_deactivate_by_name(NULL, name, CRYPT_DEACTIVATE_DEFERRED);
1689 free(name);
1690 }
1691 DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean);
1692
1693 static int verity_partition(
1694 PartitionDesignator designator,
1695 DissectedPartition *m,
1696 DissectedPartition *v,
1697 const VeritySettings *verity,
1698 DissectImageFlags flags,
1699 DecryptedImage *d) {
1700
1701 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
1702 _cleanup_(dm_deferred_remove_cleanp) char *restore_deferred_remove = NULL;
1703 _cleanup_free_ char *node = NULL, *name = NULL;
1704 int r;
1705
1706 assert(m);
1707 assert(v || (verity && verity->data_path));
1708
1709 if (!verity || !verity->root_hash)
1710 return 0;
1711 if (!((verity->designator < 0 && designator == PARTITION_ROOT) ||
1712 (verity->designator == designator)))
1713 return 0;
1714
1715 if (!m->found || !m->node || !m->fstype)
1716 return 0;
1717 if (!verity->data_path) {
1718 if (!v->found || !v->node || !v->fstype)
1719 return 0;
1720
1721 if (!streq(v->fstype, "DM_verity_hash"))
1722 return 0;
1723 }
1724
1725 r = dlopen_cryptsetup();
1726 if (r < 0)
1727 return r;
1728
1729 if (FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) {
1730 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
1731 _cleanup_free_ char *root_hash_encoded = NULL;
1732
1733 root_hash_encoded = hexmem(verity->root_hash, verity->root_hash_size);
1734 if (!root_hash_encoded)
1735 return -ENOMEM;
1736
1737 r = make_dm_name_and_node(root_hash_encoded, "-verity", &name, &node);
1738 } else
1739 r = make_dm_name_and_node(m->node, "-verity", &name, &node);
1740 if (r < 0)
1741 return r;
1742
1743 r = sym_crypt_init(&cd, verity->data_path ?: v->node);
1744 if (r < 0)
1745 return r;
1746
1747 cryptsetup_enable_logging(cd);
1748
1749 r = sym_crypt_load(cd, CRYPT_VERITY, NULL);
1750 if (r < 0)
1751 return r;
1752
1753 r = sym_crypt_set_data_device(cd, m->node);
1754 if (r < 0)
1755 return r;
1756
1757 if (!GREEDY_REALLOC0(d->decrypted, d->n_allocated, d->n_decrypted + 1))
1758 return -ENOMEM;
1759
1760 /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
1761 * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
1762 * retry a few times before giving up. */
1763 for (unsigned i = 0; i < N_DEVICE_NODE_LIST_ATTEMPTS; i++) {
1764 if (verity->root_hash_sig) {
1765 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1766 r = sym_crypt_activate_by_signed_key(
1767 cd,
1768 name,
1769 verity->root_hash,
1770 verity->root_hash_size,
1771 verity->root_hash_sig,
1772 verity->root_hash_sig_size,
1773 CRYPT_ACTIVATE_READONLY);
1774 #else
1775 r = log_debug_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
1776 "Activation of verity device with signature requested, but not supported by %s due to missing crypt_activate_by_signed_key().", program_invocation_short_name);
1777 #endif
1778 } else
1779 r = sym_crypt_activate_by_volume_key(
1780 cd,
1781 name,
1782 verity->root_hash,
1783 verity->root_hash_size,
1784 CRYPT_ACTIVATE_READONLY);
1785 /* libdevmapper can return EINVAL when the device is already in the activation stage.
1786 * There's no way to distinguish this situation from a genuine error due to invalid
1787 * parameters, so immediately fall back to activating the device with a unique name.
1788 * Improvements in libcrypsetup can ensure this never happens:
1789 * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
1790 if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
1791 return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
1792 if (!IN_SET(r,
1793 0, /* Success */
1794 -EEXIST, /* Volume is already open and ready to be used */
1795 -EBUSY, /* Volume is being opened but not ready, crypt_init_by_name can fetch details */
1796 -ENODEV /* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */))
1797 return r;
1798 if (IN_SET(r, -EEXIST, -EBUSY)) {
1799 struct crypt_device *existing_cd = NULL;
1800
1801 if (!restore_deferred_remove){
1802 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
1803 r = dm_deferred_remove_cancel(name);
1804 /* If activation returns EBUSY there might be no deferred removal to cancel, that's fine */
1805 if (r < 0 && r != -ENXIO)
1806 return log_debug_errno(r, "Disabling automated deferred removal for verity device %s failed: %m", node);
1807 if (r == 0) {
1808 restore_deferred_remove = strdup(name);
1809 if (!restore_deferred_remove)
1810 return -ENOMEM;
1811 }
1812 }
1813
1814 r = verity_can_reuse(verity, name, &existing_cd);
1815 /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
1816 if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
1817 return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
1818 if (!IN_SET(r, 0, -ENODEV, -ENOENT, -EBUSY))
1819 return log_debug_errno(r, "Checking whether existing verity device %s can be reused failed: %m", node);
1820 if (r == 0) {
1821 /* devmapper might say that the device exists, but the devlink might not yet have been
1822 * created. Check and wait for the udev event in that case. */
1823 r = device_wait_for_devlink(node, "block", usec_add(now(CLOCK_MONOTONIC), 100 * USEC_PER_MSEC), NULL);
1824 /* Fallback to activation with a unique device if it's taking too long */
1825 if (r == -ETIMEDOUT)
1826 break;
1827 if (r < 0)
1828 return r;
1829
1830 if (cd)
1831 sym_crypt_free(cd);
1832 cd = existing_cd;
1833 }
1834 }
1835 if (r == 0)
1836 break;
1837
1838 /* Device is being opened by another process, but it has not finished yet, yield for 2ms */
1839 (void) usleep(2 * USEC_PER_MSEC);
1840 }
1841
1842 /* An existing verity device was reported by libcryptsetup/libdevmapper, but we can't use it at this time.
1843 * Fall back to activating it with a unique device name. */
1844 if (r != 0 && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
1845 return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
1846
1847 /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
1848 restore_deferred_remove = mfree(restore_deferred_remove);
1849
1850 d->decrypted[d->n_decrypted++] = (DecryptedPartition) {
1851 .name = TAKE_PTR(name),
1852 .device = TAKE_PTR(cd),
1853 };
1854
1855 m->decrypted_node = TAKE_PTR(node);
1856
1857 return 0;
1858 }
1859 #endif
1860
1861 int dissected_image_decrypt(
1862 DissectedImage *m,
1863 const char *passphrase,
1864 const VeritySettings *verity,
1865 DissectImageFlags flags,
1866 DecryptedImage **ret) {
1867
1868 #if HAVE_LIBCRYPTSETUP
1869 _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL;
1870 int r;
1871 #endif
1872
1873 assert(m);
1874 assert(!verity || verity->root_hash || verity->root_hash_size == 0);
1875
1876 /* Returns:
1877 *
1878 * = 0 → There was nothing to decrypt
1879 * > 0 → Decrypted successfully
1880 * -ENOKEY → There's something to decrypt but no key was supplied
1881 * -EKEYREJECTED → Passed key was not correct
1882 */
1883
1884 if (verity && verity->root_hash && verity->root_hash_size < sizeof(sd_id128_t))
1885 return -EINVAL;
1886
1887 if (!m->encrypted && !m->verity) {
1888 *ret = NULL;
1889 return 0;
1890 }
1891
1892 #if HAVE_LIBCRYPTSETUP
1893 d = new0(DecryptedImage, 1);
1894 if (!d)
1895 return -ENOMEM;
1896
1897 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
1898 DissectedPartition *p = m->partitions + i;
1899 PartitionDesignator k;
1900
1901 if (!p->found)
1902 continue;
1903
1904 r = decrypt_partition(p, passphrase, flags, d);
1905 if (r < 0)
1906 return r;
1907
1908 k = PARTITION_VERITY_OF(i);
1909 if (k >= 0) {
1910 r = verity_partition(i, p, m->partitions + k, verity, flags | DISSECT_IMAGE_VERITY_SHARE, d);
1911 if (r < 0)
1912 return r;
1913 }
1914
1915 if (!p->decrypted_fstype && p->decrypted_node) {
1916 r = probe_filesystem(p->decrypted_node, &p->decrypted_fstype);
1917 if (r < 0 && r != -EUCLEAN)
1918 return r;
1919 }
1920 }
1921
1922 *ret = TAKE_PTR(d);
1923
1924 return 1;
1925 #else
1926 return -EOPNOTSUPP;
1927 #endif
1928 }
1929
1930 int dissected_image_decrypt_interactively(
1931 DissectedImage *m,
1932 const char *passphrase,
1933 const VeritySettings *verity,
1934 DissectImageFlags flags,
1935 DecryptedImage **ret) {
1936
1937 _cleanup_strv_free_erase_ char **z = NULL;
1938 int n = 3, r;
1939
1940 if (passphrase)
1941 n--;
1942
1943 for (;;) {
1944 r = dissected_image_decrypt(m, passphrase, verity, flags, ret);
1945 if (r >= 0)
1946 return r;
1947 if (r == -EKEYREJECTED)
1948 log_error_errno(r, "Incorrect passphrase, try again!");
1949 else if (r != -ENOKEY)
1950 return log_error_errno(r, "Failed to decrypt image: %m");
1951
1952 if (--n < 0)
1953 return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED),
1954 "Too many retries.");
1955
1956 z = strv_free(z);
1957
1958 r = ask_password_auto("Please enter image passphrase:", NULL, "dissect", "dissect", USEC_INFINITY, 0, &z);
1959 if (r < 0)
1960 return log_error_errno(r, "Failed to query for passphrase: %m");
1961
1962 passphrase = z[0];
1963 }
1964 }
1965
1966 int decrypted_image_relinquish(DecryptedImage *d) {
1967
1968 #if HAVE_LIBCRYPTSETUP
1969 size_t i;
1970 int r;
1971 #endif
1972
1973 assert(d);
1974
1975 /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a boolean so
1976 * that we don't clean it up ourselves either anymore */
1977
1978 #if HAVE_LIBCRYPTSETUP
1979 for (i = 0; i < d->n_decrypted; i++) {
1980 DecryptedPartition *p = d->decrypted + i;
1981
1982 if (p->relinquished)
1983 continue;
1984
1985 r = sym_crypt_deactivate_by_name(NULL, p->name, CRYPT_DEACTIVATE_DEFERRED);
1986 if (r < 0)
1987 return log_debug_errno(r, "Failed to mark %s for auto-removal: %m", p->name);
1988
1989 p->relinquished = true;
1990 }
1991 #endif
1992
1993 return 0;
1994 }
1995
1996 static char *build_auxiliary_path(const char *image, const char *suffix) {
1997 const char *e;
1998 char *n;
1999
2000 assert(image);
2001 assert(suffix);
2002
2003 e = endswith(image, ".raw");
2004 if (!e)
2005 return strjoin(e, suffix);
2006
2007 n = new(char, e - image + strlen(suffix) + 1);
2008 if (!n)
2009 return NULL;
2010
2011 strcpy(mempcpy(n, image, e - image), suffix);
2012 return n;
2013 }
2014
2015 void verity_settings_done(VeritySettings *v) {
2016 assert(v);
2017
2018 v->root_hash = mfree(v->root_hash);
2019 v->root_hash_size = 0;
2020
2021 v->root_hash_sig = mfree(v->root_hash_sig);
2022 v->root_hash_sig_size = 0;
2023
2024 v->data_path = mfree(v->data_path);
2025 }
2026
2027 int verity_settings_load(
2028 VeritySettings *verity,
2029 const char *image,
2030 const char *root_hash_path,
2031 const char *root_hash_sig_path) {
2032
2033 _cleanup_free_ void *root_hash = NULL, *root_hash_sig = NULL;
2034 size_t root_hash_size = 0, root_hash_sig_size = 0;
2035 _cleanup_free_ char *verity_data_path = NULL;
2036 PartitionDesignator designator;
2037 int r;
2038
2039 assert(verity);
2040 assert(image);
2041 assert(verity->designator < 0 || IN_SET(verity->designator, PARTITION_ROOT, PARTITION_USR));
2042
2043 /* If we are asked to load the root hash for a device node, exit early */
2044 if (is_device_path(image))
2045 return 0;
2046
2047 designator = verity->designator;
2048
2049 /* We only fill in what isn't already filled in */
2050
2051 if (!verity->root_hash) {
2052 _cleanup_free_ char *text = NULL;
2053
2054 if (root_hash_path) {
2055 /* If explicitly specified it takes precedence */
2056 r = read_one_line_file(root_hash_path, &text);
2057 if (r < 0)
2058 return r;
2059
2060 if (designator < 0)
2061 designator = PARTITION_ROOT;
2062 } else {
2063 /* Otherwise look for xattr and separate file, and first for the data for root and if
2064 * that doesn't exist for /usr */
2065
2066 if (designator < 0 || designator == PARTITION_ROOT) {
2067 r = getxattr_malloc(image, "user.verity.roothash", &text, true);
2068 if (r < 0) {
2069 _cleanup_free_ char *p = NULL;
2070
2071 if (!IN_SET(r, -ENODATA, -ENOENT) && !ERRNO_IS_NOT_SUPPORTED(r))
2072 return r;
2073
2074 p = build_auxiliary_path(image, ".roothash");
2075 if (!p)
2076 return -ENOMEM;
2077
2078 r = read_one_line_file(p, &text);
2079 if (r < 0 && r != -ENOENT)
2080 return r;
2081 }
2082
2083 if (text)
2084 designator = PARTITION_ROOT;
2085 }
2086
2087 if (!text && (designator < 0 || designator == PARTITION_USR)) {
2088 /* So in the "roothash" xattr/file name above the "root" of course primarily
2089 * refers to the root of the Verity Merkle tree. But coincidentally it also
2090 * is the hash for the *root* file system, i.e. the "root" neatly refers to
2091 * two distinct concepts called "root". Taking benefit of this happy
2092 * coincidence we call the file with the root hash for the /usr/ file system
2093 * `usrhash`, because `usrroothash` or `rootusrhash` would just be too
2094 * confusing. We thus drop the reference to the root of the Merkle tree, and
2095 * just indicate which file system it's about. */
2096 r = getxattr_malloc(image, "user.verity.usrhash", &text, true);
2097 if (r < 0) {
2098 _cleanup_free_ char *p = NULL;
2099
2100 if (!IN_SET(r, -ENODATA, -ENOENT) && !ERRNO_IS_NOT_SUPPORTED(r))
2101 return r;
2102
2103 p = build_auxiliary_path(image, ".usrhash");
2104 if (!p)
2105 return -ENOMEM;
2106
2107 r = read_one_line_file(p, &text);
2108 if (r < 0 && r != -ENOENT)
2109 return r;
2110 }
2111
2112 if (text)
2113 designator = PARTITION_USR;
2114 }
2115 }
2116
2117 if (text) {
2118 r = unhexmem(text, strlen(text), &root_hash, &root_hash_size);
2119 if (r < 0)
2120 return r;
2121 if (root_hash_size < sizeof(sd_id128_t))
2122 return -EINVAL;
2123 }
2124 }
2125
2126 if ((root_hash || verity->root_hash) && !verity->root_hash_sig) {
2127 if (root_hash_sig_path) {
2128 r = read_full_file(root_hash_sig_path, (char**) &root_hash_sig, &root_hash_sig_size);
2129 if (r < 0 && r != -ENOENT)
2130 return r;
2131
2132 if (designator < 0)
2133 designator = PARTITION_ROOT;
2134 } else {
2135 if (designator < 0 || designator == PARTITION_ROOT) {
2136 _cleanup_free_ char *p = NULL;
2137
2138 /* Follow naming convention recommended by the relevant RFC:
2139 * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
2140 p = build_auxiliary_path(image, ".roothash.p7s");
2141 if (!p)
2142 return -ENOMEM;
2143
2144 r = read_full_file(p, (char**) &root_hash_sig, &root_hash_sig_size);
2145 if (r < 0 && r != -ENOENT)
2146 return r;
2147 if (r >= 0)
2148 designator = PARTITION_ROOT;
2149 }
2150
2151 if (!root_hash_sig && (designator < 0 || designator == PARTITION_USR)) {
2152 _cleanup_free_ char *p = NULL;
2153
2154 p = build_auxiliary_path(image, ".usrhash.p7s");
2155 if (!p)
2156 return -ENOMEM;
2157
2158 r = read_full_file(p, (char**) &root_hash_sig, &root_hash_sig_size);
2159 if (r < 0 && r != -ENOENT)
2160 return r;
2161 if (r >= 0)
2162 designator = PARTITION_USR;
2163 }
2164 }
2165
2166 if (root_hash_sig && root_hash_sig_size == 0) /* refuse empty size signatures */
2167 return -EINVAL;
2168 }
2169
2170 if (!verity->data_path) {
2171 _cleanup_free_ char *p = NULL;
2172
2173 p = build_auxiliary_path(image, ".verity");
2174 if (!p)
2175 return -ENOMEM;
2176
2177 if (access(p, F_OK) < 0) {
2178 if (errno != ENOENT)
2179 return -errno;
2180 } else
2181 verity_data_path = TAKE_PTR(p);
2182 }
2183
2184 if (root_hash) {
2185 verity->root_hash = TAKE_PTR(root_hash);
2186 verity->root_hash_size = root_hash_size;
2187 }
2188
2189 if (root_hash_sig) {
2190 verity->root_hash_sig = TAKE_PTR(root_hash_sig);
2191 verity->root_hash_sig_size = root_hash_sig_size;
2192 }
2193
2194 if (verity_data_path)
2195 verity->data_path = TAKE_PTR(verity_data_path);
2196
2197 if (verity->designator < 0)
2198 verity->designator = designator;
2199
2200 return 1;
2201 }
2202
2203 int dissected_image_acquire_metadata(DissectedImage *m) {
2204
2205 enum {
2206 META_HOSTNAME,
2207 META_MACHINE_ID,
2208 META_MACHINE_INFO,
2209 META_OS_RELEASE,
2210 _META_MAX,
2211 };
2212
2213 static const char *const paths[_META_MAX] = {
2214 [META_HOSTNAME] = "/etc/hostname\0",
2215 [META_MACHINE_ID] = "/etc/machine-id\0",
2216 [META_MACHINE_INFO] = "/etc/machine-info\0",
2217 [META_OS_RELEASE] = ("/etc/os-release\0"
2218 "/usr/lib/os-release\0"),
2219 };
2220
2221 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL;
2222 _cleanup_close_pair_ int error_pipe[2] = { -1, -1 };
2223 _cleanup_(rmdir_and_freep) char *t = NULL;
2224 _cleanup_(sigkill_waitp) pid_t child = 0;
2225 sd_id128_t machine_id = SD_ID128_NULL;
2226 _cleanup_free_ char *hostname = NULL;
2227 unsigned n_meta_initialized = 0, k;
2228 int fds[2 * _META_MAX], r, v;
2229 ssize_t n;
2230
2231 BLOCK_SIGNALS(SIGCHLD);
2232
2233 assert(m);
2234
2235 for (; n_meta_initialized < _META_MAX; n_meta_initialized ++)
2236 if (pipe2(fds + 2*n_meta_initialized, O_CLOEXEC) < 0) {
2237 r = -errno;
2238 goto finish;
2239 }
2240
2241 r = mkdtemp_malloc("/tmp/dissect-XXXXXX", &t);
2242 if (r < 0)
2243 goto finish;
2244
2245 if (pipe2(error_pipe, O_CLOEXEC) < 0) {
2246 r = -errno;
2247 goto finish;
2248 }
2249
2250 r = safe_fork("(sd-dissect)", FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE, &child);
2251 if (r < 0)
2252 goto finish;
2253 if (r == 0) {
2254 error_pipe[0] = safe_close(error_pipe[0]);
2255
2256 r = dissected_image_mount(m, t, UID_INVALID, DISSECT_IMAGE_READ_ONLY|DISSECT_IMAGE_MOUNT_ROOT_ONLY|DISSECT_IMAGE_VALIDATE_OS);
2257 if (r < 0) {
2258 /* Let parent know the error */
2259 (void) write(error_pipe[1], &r, sizeof(r));
2260
2261 log_debug_errno(r, "Failed to mount dissected image: %m");
2262 _exit(EXIT_FAILURE);
2263 }
2264
2265 for (k = 0; k < _META_MAX; k++) {
2266 _cleanup_close_ int fd = -ENOENT;
2267 const char *p;
2268
2269 fds[2*k] = safe_close(fds[2*k]);
2270
2271 NULSTR_FOREACH(p, paths[k]) {
2272 fd = chase_symlinks_and_open(p, t, CHASE_PREFIX_ROOT, O_RDONLY|O_CLOEXEC|O_NOCTTY, NULL);
2273 if (fd >= 0)
2274 break;
2275 }
2276 if (fd < 0) {
2277 log_debug_errno(fd, "Failed to read %s file of image, ignoring: %m", paths[k]);
2278 fds[2*k+1] = safe_close(fds[2*k+1]);
2279 continue;
2280 }
2281
2282 r = copy_bytes(fd, fds[2*k+1], (uint64_t) -1, 0);
2283 if (r < 0) {
2284 (void) write(error_pipe[1], &r, sizeof(r));
2285 _exit(EXIT_FAILURE);
2286 }
2287
2288 fds[2*k+1] = safe_close(fds[2*k+1]);
2289 }
2290
2291 _exit(EXIT_SUCCESS);
2292 }
2293
2294 error_pipe[1] = safe_close(error_pipe[1]);
2295
2296 for (k = 0; k < _META_MAX; k++) {
2297 _cleanup_fclose_ FILE *f = NULL;
2298
2299 fds[2*k+1] = safe_close(fds[2*k+1]);
2300
2301 f = take_fdopen(&fds[2*k], "r");
2302 if (!f) {
2303 r = -errno;
2304 goto finish;
2305 }
2306
2307 switch (k) {
2308
2309 case META_HOSTNAME:
2310 r = read_etc_hostname_stream(f, &hostname);
2311 if (r < 0)
2312 log_debug_errno(r, "Failed to read /etc/hostname: %m");
2313
2314 break;
2315
2316 case META_MACHINE_ID: {
2317 _cleanup_free_ char *line = NULL;
2318
2319 r = read_line(f, LONG_LINE_MAX, &line);
2320 if (r < 0)
2321 log_debug_errno(r, "Failed to read /etc/machine-id: %m");
2322 else if (r == 33) {
2323 r = sd_id128_from_string(line, &machine_id);
2324 if (r < 0)
2325 log_debug_errno(r, "Image contains invalid /etc/machine-id: %s", line);
2326 } else if (r == 0)
2327 log_debug("/etc/machine-id file is empty.");
2328 else if (streq(line, "uninitialized"))
2329 log_debug("/etc/machine-id file is uninitialized (likely aborted first boot).");
2330 else
2331 log_debug("/etc/machine-id has unexpected length %i.", r);
2332
2333 break;
2334 }
2335
2336 case META_MACHINE_INFO:
2337 r = load_env_file_pairs(f, "machine-info", &machine_info);
2338 if (r < 0)
2339 log_debug_errno(r, "Failed to read /etc/machine-info: %m");
2340
2341 break;
2342
2343 case META_OS_RELEASE:
2344 r = load_env_file_pairs(f, "os-release", &os_release);
2345 if (r < 0)
2346 log_debug_errno(r, "Failed to read OS release file: %m");
2347
2348 break;
2349 }
2350 }
2351
2352 r = wait_for_terminate_and_check("(sd-dissect)", child, 0);
2353 child = 0;
2354 if (r < 0)
2355 return r;
2356
2357 n = read(error_pipe[0], &v, sizeof(v));
2358 if (n < 0)
2359 return -errno;
2360 if (n == sizeof(v))
2361 return v; /* propagate error sent to us from child */
2362 if (n != 0)
2363 return -EIO;
2364
2365 if (r != EXIT_SUCCESS)
2366 return -EPROTO;
2367
2368 free_and_replace(m->hostname, hostname);
2369 m->machine_id = machine_id;
2370 strv_free_and_replace(m->machine_info, machine_info);
2371 strv_free_and_replace(m->os_release, os_release);
2372
2373 finish:
2374 for (k = 0; k < n_meta_initialized; k++)
2375 safe_close_pair(fds + 2*k);
2376
2377 return r;
2378 }
2379
2380 int dissect_image_and_warn(
2381 int fd,
2382 const char *name,
2383 const VeritySettings *verity,
2384 const MountOptions *mount_options,
2385 DissectImageFlags flags,
2386 DissectedImage **ret) {
2387
2388 _cleanup_free_ char *buffer = NULL;
2389 int r;
2390
2391 if (!name) {
2392 r = fd_get_path(fd, &buffer);
2393 if (r < 0)
2394 return r;
2395
2396 name = buffer;
2397 }
2398
2399 r = dissect_image(fd, verity, mount_options, flags, ret);
2400 switch (r) {
2401
2402 case -EOPNOTSUPP:
2403 return log_error_errno(r, "Dissecting images is not supported, compiled without blkid support.");
2404
2405 case -ENOPKG:
2406 return log_error_errno(r, "Couldn't identify a suitable partition table or file system in '%s'.", name);
2407
2408 case -EADDRNOTAVAIL:
2409 return log_error_errno(r, "No root partition for specified root hash found in '%s'.", name);
2410
2411 case -ENOTUNIQ:
2412 return log_error_errno(r, "Multiple suitable root partitions found in image '%s'.", name);
2413
2414 case -ENXIO:
2415 return log_error_errno(r, "No suitable root partition found in image '%s'.", name);
2416
2417 case -EPROTONOSUPPORT:
2418 return log_error_errno(r, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name);
2419
2420 default:
2421 if (r < 0)
2422 return log_error_errno(r, "Failed to dissect image '%s': %m", name);
2423
2424 return r;
2425 }
2426 }
2427
2428 bool dissected_image_can_do_verity(const DissectedImage *image, PartitionDesignator partition_designator) {
2429 if (image->single_file_system)
2430 return partition_designator == PARTITION_ROOT && image->can_verity;
2431
2432 return PARTITION_VERITY_OF(partition_designator) >= 0;
2433 }
2434
2435 bool dissected_image_has_verity(const DissectedImage *image, PartitionDesignator partition_designator) {
2436 int k;
2437
2438 if (image->single_file_system)
2439 return partition_designator == PARTITION_ROOT && image->verity;
2440
2441 k = PARTITION_VERITY_OF(partition_designator);
2442 return k >= 0 && image->partitions[k].found;
2443 }
2444
2445 MountOptions* mount_options_free_all(MountOptions *options) {
2446 MountOptions *m;
2447
2448 while ((m = options)) {
2449 LIST_REMOVE(mount_options, options, m);
2450 free(m->options);
2451 free(m);
2452 }
2453
2454 return NULL;
2455 }
2456
2457 const char* mount_options_from_designator(const MountOptions *options, PartitionDesignator designator) {
2458 const MountOptions *m;
2459
2460 LIST_FOREACH(mount_options, m, options)
2461 if (designator == m->partition_designator && !isempty(m->options))
2462 return m->options;
2463
2464 return NULL;
2465 }
2466
2467 int mount_image_privately_interactively(
2468 const char *image,
2469 DissectImageFlags flags,
2470 char **ret_directory,
2471 LoopDevice **ret_loop_device,
2472 DecryptedImage **ret_decrypted_image) {
2473
2474 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
2475 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
2476 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
2477 _cleanup_(rmdir_and_freep) char *created_dir = NULL;
2478 _cleanup_free_ char *temp = NULL;
2479 int r;
2480
2481 /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
2482 * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
2483 * easily. */
2484
2485 assert(image);
2486 assert(ret_directory);
2487 assert(ret_loop_device);
2488 assert(ret_decrypted_image);
2489
2490 r = tempfn_random_child(NULL, program_invocation_short_name, &temp);
2491 if (r < 0)
2492 return log_error_errno(r, "Failed to generate temporary mount directory: %m");
2493
2494 r = loop_device_make_by_path(
2495 image,
2496 FLAGS_SET(flags, DISSECT_IMAGE_READ_ONLY) ? O_RDONLY : O_RDWR,
2497 FLAGS_SET(flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN,
2498 &d);
2499 if (r < 0)
2500 return log_error_errno(r, "Failed to set up loopback device: %m");
2501
2502 r = dissect_image_and_warn(d->fd, image, NULL, NULL, flags, &dissected_image);
2503 if (r < 0)
2504 return r;
2505
2506 r = dissected_image_decrypt_interactively(dissected_image, NULL, NULL, flags, &decrypted_image);
2507 if (r < 0)
2508 return r;
2509
2510 r = detach_mount_namespace();
2511 if (r < 0)
2512 return log_error_errno(r, "Failed to detach mount namespace: %m");
2513
2514 r = mkdir_p(temp, 0700);
2515 if (r < 0)
2516 return log_error_errno(r, "Failed to create mount point: %m");
2517
2518 created_dir = TAKE_PTR(temp);
2519
2520 r = dissected_image_mount_and_warn(dissected_image, created_dir, UID_INVALID, flags);
2521 if (r < 0)
2522 return r;
2523
2524 if (decrypted_image) {
2525 r = decrypted_image_relinquish(decrypted_image);
2526 if (r < 0)
2527 return log_error_errno(r, "Failed to relinquish DM devices: %m");
2528 }
2529
2530 loop_device_relinquish(d);
2531
2532 *ret_directory = TAKE_PTR(created_dir);
2533 *ret_loop_device = TAKE_PTR(d);
2534 *ret_decrypted_image = TAKE_PTR(decrypted_image);
2535
2536 return 0;
2537 }
2538
2539 static const char *const partition_designator_table[] = {
2540 [PARTITION_ROOT] = "root",
2541 [PARTITION_ROOT_SECONDARY] = "root-secondary",
2542 [PARTITION_USR] = "usr",
2543 [PARTITION_USR_SECONDARY] = "usr-secondary",
2544 [PARTITION_HOME] = "home",
2545 [PARTITION_SRV] = "srv",
2546 [PARTITION_ESP] = "esp",
2547 [PARTITION_XBOOTLDR] = "xbootldr",
2548 [PARTITION_SWAP] = "swap",
2549 [PARTITION_ROOT_VERITY] = "root-verity",
2550 [PARTITION_ROOT_SECONDARY_VERITY] = "root-secondary-verity",
2551 [PARTITION_USR_VERITY] = "usr-verity",
2552 [PARTITION_USR_SECONDARY_VERITY] = "usr-secondary-verity",
2553 [PARTITION_TMP] = "tmp",
2554 [PARTITION_VAR] = "var",
2555 };
2556
2557 int verity_dissect_and_mount(const char *src, const char *dest, const MountOptions *options) {
2558 _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL;
2559 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
2560 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
2561 _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT;
2562 DissectImageFlags dissect_image_flags;
2563 int r;
2564
2565 assert(src);
2566 assert(dest);
2567
2568 r = verity_settings_load(&verity, src, NULL, NULL);
2569 if (r < 0)
2570 return log_debug_errno(r, "Failed to load root hash: %m");
2571
2572 dissect_image_flags = verity.data_path ? DISSECT_IMAGE_NO_PARTITION_TABLE : 0;
2573
2574 r = loop_device_make_by_path(
2575 src,
2576 -1,
2577 verity.data_path ? 0 : LO_FLAGS_PARTSCAN,
2578 &loop_device);
2579 if (r < 0)
2580 return log_debug_errno(r, "Failed to create loop device for image: %m");
2581
2582 r = dissect_image(
2583 loop_device->fd,
2584 &verity,
2585 options,
2586 dissect_image_flags,
2587 &dissected_image);
2588 /* No partition table? Might be a single-filesystem image, try again */
2589 if (!verity.data_path && r == -ENOPKG)
2590 r = dissect_image(
2591 loop_device->fd,
2592 &verity,
2593 options,
2594 dissect_image_flags|DISSECT_IMAGE_NO_PARTITION_TABLE,
2595 &dissected_image);
2596 if (r < 0)
2597 return log_debug_errno(r, "Failed to dissect image: %m");
2598
2599 r = dissected_image_decrypt(
2600 dissected_image,
2601 NULL,
2602 &verity,
2603 dissect_image_flags,
2604 &decrypted_image);
2605 if (r < 0)
2606 return log_debug_errno(r, "Failed to decrypt dissected image: %m");
2607
2608 r = mkdir_p_label(dest, 0755);
2609 if (r < 0)
2610 return log_debug_errno(r, "Failed to create destination directory %s: %m", dest);
2611 r = umount_recursive(dest, 0);
2612 if (r < 0)
2613 return log_debug_errno(r, "Failed to umount under destination directory %s: %m", dest);
2614
2615 r = dissected_image_mount(dissected_image, dest, UID_INVALID, dissect_image_flags);
2616 if (r < 0)
2617 return log_debug_errno(r, "Failed to mount image: %m");
2618
2619 if (decrypted_image) {
2620 r = decrypted_image_relinquish(decrypted_image);
2621 if (r < 0)
2622 return log_debug_errno(r, "Failed to relinquish decrypted image: %m");
2623 }
2624
2625 loop_device_relinquish(loop_device);
2626
2627 return 0;
2628 }
2629
2630 DEFINE_STRING_TABLE_LOOKUP(partition_designator, PartitionDesignator);
Unnamed repository; edit this file 'description' to name the repository.