]> 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: rework how we wait for partitions
[thirdparty/systemd.git] / src / shared / dissect-image.c
1 /* SPDX-License-Identifier: LGPL-2.1+ */
2
3 #if HAVE_VALGRIND_MEMCHECK_H
4 #include <valgrind/memcheck.h>
5 #endif
6
7 #include <linux/dm-ioctl.h>
8 #include <linux/loop.h>
9 #include <sys/mount.h>
10 #include <sys/prctl.h>
11 #include <sys/wait.h>
12 #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-util.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 #endif
372
373 #define DEVICE_TIMEOUT_USEC (45 * USEC_PER_SEC)
374
375 int dissect_image(
376 int fd,
377 const VeritySettings *verity,
378 const MountOptions *mount_options,
379 DissectImageFlags flags,
380 DissectedImage **ret) {
381
382 #if HAVE_BLKID
383 sd_id128_t root_uuid = SD_ID128_NULL, root_verity_uuid = SD_ID128_NULL,
384 usr_uuid = SD_ID128_NULL, usr_verity_uuid = SD_ID128_NULL;
385 bool is_gpt, is_mbr, generic_rw, multiple_generic = false;
386 _cleanup_(sd_device_unrefp) sd_device *d = NULL;
387 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
388 _cleanup_(blkid_free_probep) blkid_probe b = NULL;
389 _cleanup_free_ char *generic_node = NULL;
390 sd_id128_t generic_uuid = SD_ID128_NULL;
391 const char *pttype = NULL;
392 blkid_partlist pl;
393 int r, generic_nr, n_partitions;
394 struct stat st;
395 usec_t deadline;
396
397 assert(fd >= 0);
398 assert(ret);
399 assert(!verity || verity->root_hash || verity->root_hash_size == 0);
400 assert(!((flags & DISSECT_IMAGE_GPT_ONLY) && (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)));
401
402 /* Probes a disk image, and returns information about what it found in *ret.
403 *
404 * Returns -ENOPKG if no suitable partition table or file system could be found.
405 * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found. */
406
407 if (verity && verity->root_hash) {
408 sd_id128_t fsuuid, vuuid;
409
410 /* If a root hash is supplied, then we use the root partition that has a UUID that match the
411 * first 128bit of the root hash. And we use the verity partition that has a UUID that match
412 * the final 128bit. */
413
414 if (verity->root_hash_size < sizeof(sd_id128_t))
415 return -EINVAL;
416
417 memcpy(&fsuuid, verity->root_hash, sizeof(sd_id128_t));
418 memcpy(&vuuid, (const uint8_t*) verity->root_hash + verity->root_hash_size - sizeof(sd_id128_t), sizeof(sd_id128_t));
419
420 if (sd_id128_is_null(fsuuid))
421 return -EINVAL;
422 if (sd_id128_is_null(vuuid))
423 return -EINVAL;
424
425 /* If the verity data declares it's for the /usr partition, then search for that, in all
426 * other cases assume it's for the root partition. */
427 if (verity->designator == PARTITION_USR) {
428 usr_uuid = fsuuid;
429 usr_verity_uuid = vuuid;
430 } else {
431 root_uuid = fsuuid;
432 root_verity_uuid = vuuid;
433 }
434 }
435
436 if (fstat(fd, &st) < 0)
437 return -errno;
438
439 if (!S_ISBLK(st.st_mode))
440 return -ENOTBLK;
441
442 r = sd_device_new_from_devnum(&d, 'b', st.st_rdev);
443 if (r < 0)
444 return r;
445
446 if (!FLAGS_SET(flags, DISSECT_IMAGE_NO_UDEV)) {
447 _cleanup_(sd_device_unrefp) sd_device *initialized = NULL;
448
449 /* If udev support is enabled, then let's wait for the device to be initialized before we doing anything. */
450
451 r = device_wait_for_initialization(d, "block", DEVICE_TIMEOUT_USEC, &initialized);
452 if (r < 0)
453 return r;
454
455 sd_device_unref(d);
456 d = TAKE_PTR(initialized);
457 }
458
459 b = blkid_new_probe();
460 if (!b)
461 return -ENOMEM;
462
463 errno = 0;
464 r = blkid_probe_set_device(b, fd, 0, 0);
465 if (r != 0)
466 return errno_or_else(ENOMEM);
467
468 if ((flags & DISSECT_IMAGE_GPT_ONLY) == 0) {
469 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
470 blkid_probe_enable_superblocks(b, 1);
471 blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE|BLKID_SUBLKS_USAGE);
472 }
473
474 blkid_probe_enable_partitions(b, 1);
475 blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS);
476
477 errno = 0;
478 r = blkid_do_safeprobe(b);
479 if (IN_SET(r, -2, 1))
480 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG), "Failed to identify any partition table.");
481 if (r != 0)
482 return errno_or_else(EIO);
483
484 m = new0(DissectedImage, 1);
485 if (!m)
486 return -ENOMEM;
487
488 if ((!(flags & DISSECT_IMAGE_GPT_ONLY) &&
489 (flags & DISSECT_IMAGE_REQUIRE_ROOT)) ||
490 (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)) {
491 const char *usage = NULL;
492
493 /* If flags permit this, also allow using non-partitioned single-filesystem images */
494
495 (void) blkid_probe_lookup_value(b, "USAGE", &usage, NULL);
496 if (STRPTR_IN_SET(usage, "filesystem", "crypto")) {
497 const char *fstype = NULL, *options = NULL, *devname = NULL;
498 _cleanup_free_ char *t = NULL, *n = NULL, *o = NULL;
499
500 /* OK, we have found a file system, that's our root partition then. */
501 (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
502
503 if (fstype) {
504 t = strdup(fstype);
505 if (!t)
506 return -ENOMEM;
507 }
508
509 r = sd_device_get_devname(d, &devname);
510 if (r < 0)
511 return r;
512
513 n = strdup(devname);
514 if (!n)
515 return -ENOMEM;
516
517 m->single_file_system = true;
518 m->verity = verity && verity->root_hash && verity->data_path && (verity->designator < 0 || verity->designator == PARTITION_ROOT);
519 m->can_verity = verity && verity->data_path;
520
521 options = mount_options_from_designator(mount_options, PARTITION_ROOT);
522 if (options) {
523 o = strdup(options);
524 if (!o)
525 return -ENOMEM;
526 }
527
528 m->partitions[PARTITION_ROOT] = (DissectedPartition) {
529 .found = true,
530 .rw = !m->verity,
531 .partno = -1,
532 .architecture = _ARCHITECTURE_INVALID,
533 .fstype = TAKE_PTR(t),
534 .node = TAKE_PTR(n),
535 .mount_options = TAKE_PTR(o),
536 };
537
538 m->encrypted = streq_ptr(fstype, "crypto_LUKS");
539
540 *ret = TAKE_PTR(m);
541 return 0;
542 }
543 }
544
545 (void) blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL);
546 if (!pttype)
547 return -ENOPKG;
548
549 is_gpt = streq_ptr(pttype, "gpt");
550 is_mbr = streq_ptr(pttype, "dos");
551
552 if (!is_gpt && ((flags & DISSECT_IMAGE_GPT_ONLY) || !is_mbr))
553 return -ENOPKG;
554
555 /* Safety check: refuse block devices that carry a partition table but for which the kernel doesn't
556 * do partition scanning. */
557 r = blockdev_partscan_enabled(fd);
558 if (r < 0)
559 return r;
560 if (r == 0)
561 return -EPROTONOSUPPORT;
562
563 errno = 0;
564 pl = blkid_probe_get_partitions(b);
565 if (!pl)
566 return errno_or_else(ENOMEM);
567
568 errno = 0;
569 n_partitions = blkid_partlist_numof_partitions(pl);
570 if (n_partitions < 0)
571 return errno_or_else(EIO);
572
573 deadline = usec_add(now(CLOCK_MONOTONIC), DEVICE_TIMEOUT_USEC);
574 for (int i = 0; i < n_partitions; i++) {
575 _cleanup_(sd_device_unrefp) sd_device *q = NULL;
576 unsigned long long pflags;
577 blkid_partition pp;
578 const char *node;
579 int nr;
580
581 errno = 0;
582 pp = blkid_partlist_get_partition(pl, i);
583 if (!pp)
584 return errno_or_else(EIO);
585
586 r = wait_for_partition_device(d, pp, deadline, &q);
587 if (r < 0)
588 return r;
589
590 r = sd_device_get_devname(q, &node);
591 if (r < 0)
592 return r;
593
594 pflags = blkid_partition_get_flags(pp);
595
596 errno = 0;
597 nr = blkid_partition_get_partno(pp);
598 if (nr < 0)
599 return errno_or_else(EIO);
600
601 if (is_gpt) {
602 PartitionDesignator designator = _PARTITION_DESIGNATOR_INVALID;
603 int architecture = _ARCHITECTURE_INVALID;
604 const char *stype, *sid, *fstype = NULL;
605 sd_id128_t type_id, id;
606 bool rw = true;
607
608 sid = blkid_partition_get_uuid(pp);
609 if (!sid)
610 continue;
611 if (sd_id128_from_string(sid, &id) < 0)
612 continue;
613
614 stype = blkid_partition_get_type_string(pp);
615 if (!stype)
616 continue;
617 if (sd_id128_from_string(stype, &type_id) < 0)
618 continue;
619
620 if (sd_id128_equal(type_id, GPT_HOME)) {
621
622 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
623
624 if (pflags & GPT_FLAG_NO_AUTO)
625 continue;
626
627 designator = PARTITION_HOME;
628 rw = !(pflags & GPT_FLAG_READ_ONLY);
629
630 } else if (sd_id128_equal(type_id, GPT_SRV)) {
631
632 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
633
634 if (pflags & GPT_FLAG_NO_AUTO)
635 continue;
636
637 designator = PARTITION_SRV;
638 rw = !(pflags & GPT_FLAG_READ_ONLY);
639
640 } else if (sd_id128_equal(type_id, GPT_ESP)) {
641
642 /* Note that we don't check the GPT_FLAG_NO_AUTO flag for the ESP, as it is
643 * not defined there. We instead check the GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as
644 * recommended by the UEFI spec (See "12.3.3 Number and Location of System
645 * Partitions"). */
646
647 if (pflags & GPT_FLAG_NO_BLOCK_IO_PROTOCOL)
648 continue;
649
650 designator = PARTITION_ESP;
651 fstype = "vfat";
652
653 } else if (sd_id128_equal(type_id, GPT_XBOOTLDR)) {
654
655 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
656
657 if (pflags & GPT_FLAG_NO_AUTO)
658 continue;
659
660 designator = PARTITION_XBOOTLDR;
661 rw = !(pflags & GPT_FLAG_READ_ONLY);
662 }
663 #ifdef GPT_ROOT_NATIVE
664 else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE)) {
665
666 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
667
668 if (pflags & GPT_FLAG_NO_AUTO)
669 continue;
670
671 /* If a root ID is specified, ignore everything but the root id */
672 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
673 continue;
674
675 designator = PARTITION_ROOT;
676 architecture = native_architecture();
677 rw = !(pflags & GPT_FLAG_READ_ONLY);
678
679 } else if (sd_id128_equal(type_id, GPT_ROOT_NATIVE_VERITY)) {
680
681 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
682
683 if (pflags & GPT_FLAG_NO_AUTO)
684 continue;
685
686 m->can_verity = true;
687
688 /* Ignore verity unless a root hash is specified */
689 if (sd_id128_is_null(root_verity_uuid) || !sd_id128_equal(root_verity_uuid, id))
690 continue;
691
692 designator = PARTITION_ROOT_VERITY;
693 fstype = "DM_verity_hash";
694 architecture = native_architecture();
695 rw = false;
696 }
697 #endif
698 #ifdef GPT_ROOT_SECONDARY
699 else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY)) {
700
701 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
702
703 if (pflags & GPT_FLAG_NO_AUTO)
704 continue;
705
706 /* If a root ID is specified, ignore everything but the root id */
707 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
708 continue;
709
710 designator = PARTITION_ROOT_SECONDARY;
711 architecture = SECONDARY_ARCHITECTURE;
712 rw = !(pflags & GPT_FLAG_READ_ONLY);
713
714 } else if (sd_id128_equal(type_id, GPT_ROOT_SECONDARY_VERITY)) {
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 m->can_verity = true;
722
723 /* Ignore verity unless root has is specified */
724 if (sd_id128_is_null(root_verity_uuid) || !sd_id128_equal(root_verity_uuid, id))
725 continue;
726
727 designator = PARTITION_ROOT_SECONDARY_VERITY;
728 fstype = "DM_verity_hash";
729 architecture = SECONDARY_ARCHITECTURE;
730 rw = false;
731 }
732 #endif
733 #ifdef GPT_USR_NATIVE
734 else if (sd_id128_equal(type_id, GPT_USR_NATIVE)) {
735
736 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
737
738 if (pflags & GPT_FLAG_NO_AUTO)
739 continue;
740
741 /* If a usr ID is specified, ignore everything but the usr id */
742 if (!sd_id128_is_null(usr_uuid) && !sd_id128_equal(usr_uuid, id))
743 continue;
744
745 designator = PARTITION_USR;
746 architecture = native_architecture();
747 rw = !(pflags & GPT_FLAG_READ_ONLY);
748
749 } else if (sd_id128_equal(type_id, GPT_USR_NATIVE_VERITY)) {
750
751 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
752
753 if (pflags & GPT_FLAG_NO_AUTO)
754 continue;
755
756 m->can_verity = true;
757
758 /* Ignore verity unless a usr hash is specified */
759 if (sd_id128_is_null(usr_verity_uuid) || !sd_id128_equal(usr_verity_uuid, id))
760 continue;
761
762 designator = PARTITION_USR_VERITY;
763 fstype = "DM_verity_hash";
764 architecture = native_architecture();
765 rw = false;
766 }
767 #endif
768 #ifdef GPT_USR_SECONDARY
769 else if (sd_id128_equal(type_id, GPT_USR_SECONDARY)) {
770
771 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
772
773 if (pflags & GPT_FLAG_NO_AUTO)
774 continue;
775
776 /* If a usr ID is specified, ignore everything but the usr id */
777 if (!sd_id128_is_null(usr_uuid) && !sd_id128_equal(usr_uuid, id))
778 continue;
779
780 designator = PARTITION_USR_SECONDARY;
781 architecture = SECONDARY_ARCHITECTURE;
782 rw = !(pflags & GPT_FLAG_READ_ONLY);
783
784 } else if (sd_id128_equal(type_id, GPT_USR_SECONDARY_VERITY)) {
785
786 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
787
788 if (pflags & GPT_FLAG_NO_AUTO)
789 continue;
790
791 m->can_verity = true;
792
793 /* Ignore verity unless usr has is specified */
794 if (sd_id128_is_null(usr_verity_uuid) || !sd_id128_equal(usr_verity_uuid, id))
795 continue;
796
797 designator = PARTITION_USR_SECONDARY_VERITY;
798 fstype = "DM_verity_hash";
799 architecture = SECONDARY_ARCHITECTURE;
800 rw = false;
801 }
802 #endif
803 else if (sd_id128_equal(type_id, GPT_SWAP)) {
804
805 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO);
806
807 if (pflags & GPT_FLAG_NO_AUTO)
808 continue;
809
810 designator = PARTITION_SWAP;
811 fstype = "swap";
812
813 } else if (sd_id128_equal(type_id, GPT_LINUX_GENERIC)) {
814
815 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
816
817 if (pflags & GPT_FLAG_NO_AUTO)
818 continue;
819
820 if (generic_node)
821 multiple_generic = true;
822 else {
823 generic_nr = nr;
824 generic_rw = !(pflags & GPT_FLAG_READ_ONLY);
825 generic_uuid = id;
826 generic_node = strdup(node);
827 if (!generic_node)
828 return -ENOMEM;
829 }
830
831 } else if (sd_id128_equal(type_id, GPT_TMP)) {
832
833 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
834
835 if (pflags & GPT_FLAG_NO_AUTO)
836 continue;
837
838 designator = PARTITION_TMP;
839 rw = !(pflags & GPT_FLAG_READ_ONLY);
840
841 } else if (sd_id128_equal(type_id, GPT_VAR)) {
842
843 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
844
845 if (pflags & GPT_FLAG_NO_AUTO)
846 continue;
847
848 if (!FLAGS_SET(flags, DISSECT_IMAGE_RELAX_VAR_CHECK)) {
849 sd_id128_t var_uuid;
850
851 /* For /var we insist that the uuid of the partition matches the
852 * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
853 * ID. Why? Unlike the other partitions /var is inherently
854 * installation specific, hence we need to be careful not to mount it
855 * in the wrong installation. By hashing the partition UUID from
856 * /etc/machine-id we can securely bind the partition to the
857 * installation. */
858
859 r = sd_id128_get_machine_app_specific(GPT_VAR, &var_uuid);
860 if (r < 0)
861 return r;
862
863 if (!sd_id128_equal(var_uuid, id)) {
864 log_debug("Found a /var/ partition, but its UUID didn't match our expectations, ignoring.");
865 continue;
866 }
867 }
868
869 designator = PARTITION_VAR;
870 rw = !(pflags & GPT_FLAG_READ_ONLY);
871 }
872
873 if (designator != _PARTITION_DESIGNATOR_INVALID) {
874 _cleanup_free_ char *t = NULL, *n = NULL, *o = NULL;
875 const char *options = NULL;
876
877 /* First one wins */
878 if (m->partitions[designator].found)
879 continue;
880
881 if (fstype) {
882 t = strdup(fstype);
883 if (!t)
884 return -ENOMEM;
885 }
886
887 n = strdup(node);
888 if (!n)
889 return -ENOMEM;
890
891 options = mount_options_from_designator(mount_options, designator);
892 if (options) {
893 o = strdup(options);
894 if (!o)
895 return -ENOMEM;
896 }
897
898 m->partitions[designator] = (DissectedPartition) {
899 .found = true,
900 .partno = nr,
901 .rw = rw,
902 .architecture = architecture,
903 .node = TAKE_PTR(n),
904 .fstype = TAKE_PTR(t),
905 .uuid = id,
906 .mount_options = TAKE_PTR(o),
907 };
908 }
909
910 } else if (is_mbr) {
911
912 switch (blkid_partition_get_type(pp)) {
913
914 case 0x83: /* Linux partition */
915
916 if (pflags != 0x80) /* Bootable flag */
917 continue;
918
919 if (generic_node)
920 multiple_generic = true;
921 else {
922 generic_nr = nr;
923 generic_rw = true;
924 generic_node = strdup(node);
925 if (!generic_node)
926 return -ENOMEM;
927 }
928
929 break;
930
931 case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
932 _cleanup_free_ char *n = NULL, *o = NULL;
933 sd_id128_t id = SD_ID128_NULL;
934 const char *sid, *options = NULL;
935
936 /* First one wins */
937 if (m->partitions[PARTITION_XBOOTLDR].found)
938 continue;
939
940 sid = blkid_partition_get_uuid(pp);
941 if (sid)
942 (void) sd_id128_from_string(sid, &id);
943
944 n = strdup(node);
945 if (!n)
946 return -ENOMEM;
947
948 options = mount_options_from_designator(mount_options, PARTITION_XBOOTLDR);
949 if (options) {
950 o = strdup(options);
951 if (!o)
952 return -ENOMEM;
953 }
954
955 m->partitions[PARTITION_XBOOTLDR] = (DissectedPartition) {
956 .found = true,
957 .partno = nr,
958 .rw = true,
959 .architecture = _ARCHITECTURE_INVALID,
960 .node = TAKE_PTR(n),
961 .uuid = id,
962 .mount_options = TAKE_PTR(o),
963 };
964
965 break;
966 }}
967 }
968 }
969
970 if (m->partitions[PARTITION_ROOT].found) {
971 /* If we found the primary arch, then invalidate the secondary arch to avoid any ambiguities,
972 * since we never want to mount the secondary arch in this case. */
973 m->partitions[PARTITION_ROOT_SECONDARY].found = false;
974 m->partitions[PARTITION_ROOT_SECONDARY_VERITY].found = false;
975 m->partitions[PARTITION_USR_SECONDARY].found = false;
976 m->partitions[PARTITION_USR_SECONDARY_VERITY].found = false;
977 } else {
978 /* No root partition found? Then let's see if ther's one for the secondary architecture. And if not
979 * either, then check if there's a single generic one, and use that. */
980
981 if (m->partitions[PARTITION_ROOT_VERITY].found)
982 return -EADDRNOTAVAIL;
983
984 /* We didn't find a primary architecture root, but we found a primary architecture /usr? Refuse that for now. */
985 if (m->partitions[PARTITION_USR].found || m->partitions[PARTITION_USR_VERITY].found)
986 return -EADDRNOTAVAIL;
987
988 if (m->partitions[PARTITION_ROOT_SECONDARY].found) {
989 /* Upgrade secondary arch to first */
990 m->partitions[PARTITION_ROOT] = m->partitions[PARTITION_ROOT_SECONDARY];
991 zero(m->partitions[PARTITION_ROOT_SECONDARY]);
992 m->partitions[PARTITION_ROOT_VERITY] = m->partitions[PARTITION_ROOT_SECONDARY_VERITY];
993 zero(m->partitions[PARTITION_ROOT_SECONDARY_VERITY]);
994
995 m->partitions[PARTITION_USR] = m->partitions[PARTITION_USR_SECONDARY];
996 zero(m->partitions[PARTITION_USR_SECONDARY]);
997 m->partitions[PARTITION_USR_VERITY] = m->partitions[PARTITION_USR_SECONDARY_VERITY];
998 zero(m->partitions[PARTITION_USR_SECONDARY_VERITY]);
999
1000 } else if (flags & DISSECT_IMAGE_REQUIRE_ROOT) {
1001 _cleanup_free_ char *o = NULL;
1002 const char *options = NULL;
1003
1004 /* If the root hash was set, then we won't fall back to a generic node, because the
1005 * root hash decides. */
1006 if (verity && verity->root_hash)
1007 return -EADDRNOTAVAIL;
1008
1009 /* If we didn't find a generic node, then we can't fix this up either */
1010 if (!generic_node)
1011 return -ENXIO;
1012
1013 /* If we didn't find a properly marked root partition, but we did find a single suitable
1014 * generic Linux partition, then use this as root partition, if the caller asked for it. */
1015 if (multiple_generic)
1016 return -ENOTUNIQ;
1017
1018 options = mount_options_from_designator(mount_options, PARTITION_ROOT);
1019 if (options) {
1020 o = strdup(options);
1021 if (!o)
1022 return -ENOMEM;
1023 }
1024
1025 m->partitions[PARTITION_ROOT] = (DissectedPartition) {
1026 .found = true,
1027 .rw = generic_rw,
1028 .partno = generic_nr,
1029 .architecture = _ARCHITECTURE_INVALID,
1030 .node = TAKE_PTR(generic_node),
1031 .uuid = generic_uuid,
1032 .mount_options = TAKE_PTR(o),
1033 };
1034 }
1035 }
1036
1037 /* Refuse if we found a verity partition for /usr but no matching file system partition */
1038 if (!m->partitions[PARTITION_USR].found && m->partitions[PARTITION_USR_VERITY].found)
1039 return -EADDRNOTAVAIL;
1040
1041 /* Combinations of verity /usr with verity-less root is OK, but the reverse is not */
1042 if (m->partitions[PARTITION_ROOT_VERITY].found && m->partitions[PARTITION_USR].found && !m->partitions[PARTITION_USR_VERITY].found)
1043 return -EADDRNOTAVAIL;
1044
1045 if (verity && verity->root_hash) {
1046 if (verity->designator < 0 || verity->designator == PARTITION_ROOT) {
1047 if (!m->partitions[PARTITION_ROOT_VERITY].found || !m->partitions[PARTITION_ROOT].found)
1048 return -EADDRNOTAVAIL;
1049
1050 /* If we found a verity setup, then the root partition is necessarily read-only. */
1051 m->partitions[PARTITION_ROOT].rw = false;
1052 m->verity = true;
1053 }
1054
1055 if (verity->designator == PARTITION_USR) {
1056 if (!m->partitions[PARTITION_USR_VERITY].found || !m->partitions[PARTITION_USR].found)
1057 return -EADDRNOTAVAIL;
1058
1059 m->partitions[PARTITION_USR].rw = false;
1060 m->verity = true;
1061 }
1062 }
1063
1064 blkid_free_probe(b);
1065 b = NULL;
1066
1067 /* Fill in file system types if we don't know them yet. */
1068 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
1069 DissectedPartition *p = m->partitions + i;
1070
1071 if (!p->found)
1072 continue;
1073
1074 if (!p->fstype && p->node) {
1075 r = probe_filesystem(p->node, &p->fstype);
1076 if (r < 0 && r != -EUCLEAN)
1077 return r;
1078 }
1079
1080 if (streq_ptr(p->fstype, "crypto_LUKS"))
1081 m->encrypted = true;
1082
1083 if (p->fstype && fstype_is_ro(p->fstype))
1084 p->rw = false;
1085 }
1086
1087 *ret = TAKE_PTR(m);
1088 return 0;
1089 #else
1090 return -EOPNOTSUPP;
1091 #endif
1092 }
1093
1094 DissectedImage* dissected_image_unref(DissectedImage *m) {
1095 if (!m)
1096 return NULL;
1097
1098 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
1099 free(m->partitions[i].fstype);
1100 free(m->partitions[i].node);
1101 free(m->partitions[i].decrypted_fstype);
1102 free(m->partitions[i].decrypted_node);
1103 free(m->partitions[i].mount_options);
1104 }
1105
1106 free(m->hostname);
1107 strv_free(m->machine_info);
1108 strv_free(m->os_release);
1109
1110 return mfree(m);
1111 }
1112
1113 static int is_loop_device(const char *path) {
1114 char s[SYS_BLOCK_PATH_MAX("/../loop/")];
1115 struct stat st;
1116
1117 assert(path);
1118
1119 if (stat(path, &st) < 0)
1120 return -errno;
1121
1122 if (!S_ISBLK(st.st_mode))
1123 return -ENOTBLK;
1124
1125 xsprintf_sys_block_path(s, "/loop/", st.st_dev);
1126 if (access(s, F_OK) < 0) {
1127 if (errno != ENOENT)
1128 return -errno;
1129
1130 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
1131 xsprintf_sys_block_path(s, "/../loop/", st.st_dev);
1132 if (access(s, F_OK) < 0)
1133 return errno == ENOENT ? false : -errno;
1134 }
1135
1136 return true;
1137 }
1138
1139 static int run_fsck(const char *node, const char *fstype) {
1140 int r, exit_status;
1141 pid_t pid;
1142
1143 assert(node);
1144 assert(fstype);
1145
1146 r = fsck_exists(fstype);
1147 if (r < 0) {
1148 log_debug_errno(r, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype);
1149 return 0;
1150 }
1151 if (r == 0) {
1152 log_debug("Not checking partition %s, as fsck for %s does not exist.", node, fstype);
1153 return 0;
1154 }
1155
1156 r = safe_fork("(fsck)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG|FORK_NULL_STDIO, &pid);
1157 if (r < 0)
1158 return log_debug_errno(r, "Failed to fork off fsck: %m");
1159 if (r == 0) {
1160 /* Child */
1161 execl("/sbin/fsck", "/sbin/fsck", "-aT", node, NULL);
1162 log_debug_errno(errno, "Failed to execl() fsck: %m");
1163 _exit(FSCK_OPERATIONAL_ERROR);
1164 }
1165
1166 exit_status = wait_for_terminate_and_check("fsck", pid, 0);
1167 if (exit_status < 0)
1168 return log_debug_errno(exit_status, "Failed to fork off /sbin/fsck: %m");
1169
1170 if ((exit_status & ~FSCK_ERROR_CORRECTED) != FSCK_SUCCESS) {
1171 log_debug("fsck failed with exit status %i.", exit_status);
1172
1173 if ((exit_status & (FSCK_SYSTEM_SHOULD_REBOOT|FSCK_ERRORS_LEFT_UNCORRECTED)) != 0)
1174 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN), "File system is corrupted, refusing.");
1175
1176 log_debug("Ignoring fsck error.");
1177 }
1178
1179 return 0;
1180 }
1181
1182 static int mount_partition(
1183 DissectedPartition *m,
1184 const char *where,
1185 const char *directory,
1186 uid_t uid_shift,
1187 DissectImageFlags flags) {
1188
1189 _cleanup_free_ char *chased = NULL, *options = NULL;
1190 const char *p, *node, *fstype;
1191 bool rw;
1192 int r;
1193
1194 assert(m);
1195 assert(where);
1196
1197 /* Use decrypted node and matching fstype if available, otherwise use the original device */
1198 node = m->decrypted_node ?: m->node;
1199 fstype = m->decrypted_node ? m->decrypted_fstype: m->fstype;
1200
1201 if (!m->found || !node)
1202 return 0;
1203 if (!fstype)
1204 return -EAFNOSUPPORT;
1205
1206 /* 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. */
1207 if (streq(fstype, "crypto_LUKS"))
1208 return -EUNATCH;
1209
1210 rw = m->rw && !(flags & DISSECT_IMAGE_READ_ONLY);
1211
1212 if (FLAGS_SET(flags, DISSECT_IMAGE_FSCK) && rw) {
1213 r = run_fsck(node, fstype);
1214 if (r < 0)
1215 return r;
1216 }
1217
1218 if (directory) {
1219 if (!FLAGS_SET(flags, DISSECT_IMAGE_READ_ONLY)) {
1220 /* Automatically create missing mount points, if necessary. */
1221 r = mkdir_p_root(where, directory, uid_shift, (gid_t) uid_shift, 0755);
1222 if (r < 0)
1223 return r;
1224 }
1225
1226 r = chase_symlinks(directory, where, CHASE_PREFIX_ROOT, &chased, NULL);
1227 if (r < 0)
1228 return r;
1229
1230 p = chased;
1231 } else
1232 p = where;
1233
1234 /* If requested, turn on discard support. */
1235 if (fstype_can_discard(fstype) &&
1236 ((flags & DISSECT_IMAGE_DISCARD) ||
1237 ((flags & DISSECT_IMAGE_DISCARD_ON_LOOP) && is_loop_device(m->node) > 0))) {
1238 options = strdup("discard");
1239 if (!options)
1240 return -ENOMEM;
1241 }
1242
1243 if (uid_is_valid(uid_shift) && uid_shift != 0 && fstype_can_uid_gid(fstype)) {
1244 _cleanup_free_ char *uid_option = NULL;
1245
1246 if (asprintf(&uid_option, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0)
1247 return -ENOMEM;
1248
1249 if (!strextend_with_separator(&options, ",", uid_option, NULL))
1250 return -ENOMEM;
1251 }
1252
1253 if (!isempty(m->mount_options))
1254 if (!strextend_with_separator(&options, ",", m->mount_options, NULL))
1255 return -ENOMEM;
1256
1257 if (FLAGS_SET(flags, DISSECT_IMAGE_MKDIR)) {
1258 r = mkdir_p(p, 0755);
1259 if (r < 0)
1260 return r;
1261 }
1262
1263 r = mount_nofollow_verbose(LOG_DEBUG, node, p, fstype, MS_NODEV|(rw ? 0 : MS_RDONLY), options);
1264 if (r < 0)
1265 return r;
1266
1267 return 1;
1268 }
1269
1270 int dissected_image_mount(DissectedImage *m, const char *where, uid_t uid_shift, DissectImageFlags flags) {
1271 int r, xbootldr_mounted;
1272
1273 assert(m);
1274 assert(where);
1275
1276 /* Returns:
1277 *
1278 * -ENXIO → No root partition found
1279 * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release file found
1280 * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet
1281 * -EUCLEAN → fsck for file system failed
1282 * -EBUSY → File system already mounted/used elsewhere (kernel)
1283 * -EAFNOSUPPORT → File system type not supported or not known
1284 */
1285
1286 if (!m->partitions[PARTITION_ROOT].found)
1287 return -ENXIO;
1288
1289 if ((flags & DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY) == 0) {
1290 r = mount_partition(m->partitions + PARTITION_ROOT, where, NULL, uid_shift, flags);
1291 if (r < 0)
1292 return r;
1293 }
1294
1295 /* Mask DISSECT_IMAGE_MKDIR for all subdirs: the idea is that only the top-level mount point is
1296 * created if needed, but the image itself not modified. */
1297 flags &= ~DISSECT_IMAGE_MKDIR;
1298
1299 if ((flags & DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY) == 0) {
1300 /* For us mounting root always means mounting /usr as well */
1301 r = mount_partition(m->partitions + PARTITION_USR, where, "/usr", uid_shift, flags);
1302 if (r < 0)
1303 return r;
1304
1305 if (flags & DISSECT_IMAGE_VALIDATE_OS) {
1306 r = path_is_os_tree(where);
1307 if (r < 0)
1308 return r;
1309 if (r == 0)
1310 return -EMEDIUMTYPE;
1311 }
1312 }
1313
1314 if (flags & DISSECT_IMAGE_MOUNT_ROOT_ONLY)
1315 return 0;
1316
1317 r = mount_partition(m->partitions + PARTITION_HOME, where, "/home", uid_shift, flags);
1318 if (r < 0)
1319 return r;
1320
1321 r = mount_partition(m->partitions + PARTITION_SRV, where, "/srv", uid_shift, flags);
1322 if (r < 0)
1323 return r;
1324
1325 r = mount_partition(m->partitions + PARTITION_VAR, where, "/var", uid_shift, flags);
1326 if (r < 0)
1327 return r;
1328
1329 r = mount_partition(m->partitions + PARTITION_TMP, where, "/var/tmp", uid_shift, flags);
1330 if (r < 0)
1331 return r;
1332
1333 xbootldr_mounted = mount_partition(m->partitions + PARTITION_XBOOTLDR, where, "/boot", uid_shift, flags);
1334 if (xbootldr_mounted < 0)
1335 return xbootldr_mounted;
1336
1337 if (m->partitions[PARTITION_ESP].found) {
1338 int esp_done = false;
1339
1340 /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it
1341 * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */
1342
1343 r = chase_symlinks("/efi", where, CHASE_PREFIX_ROOT, NULL, NULL);
1344 if (r < 0) {
1345 if (r != -ENOENT)
1346 return r;
1347
1348 /* /efi doesn't exist. Let's see if /boot is suitable then */
1349
1350 if (!xbootldr_mounted) {
1351 _cleanup_free_ char *p = NULL;
1352
1353 r = chase_symlinks("/boot", where, CHASE_PREFIX_ROOT, &p, NULL);
1354 if (r < 0) {
1355 if (r != -ENOENT)
1356 return r;
1357 } else if (dir_is_empty(p) > 0) {
1358 /* It exists and is an empty directory. Let's mount the ESP there. */
1359 r = mount_partition(m->partitions + PARTITION_ESP, where, "/boot", uid_shift, flags);
1360 if (r < 0)
1361 return r;
1362
1363 esp_done = true;
1364 }
1365 }
1366 }
1367
1368 if (!esp_done) {
1369 /* OK, let's mount the ESP now to /efi (possibly creating the dir if missing) */
1370
1371 r = mount_partition(m->partitions + PARTITION_ESP, where, "/efi", uid_shift, flags);
1372 if (r < 0)
1373 return r;
1374 }
1375 }
1376
1377 return 0;
1378 }
1379
1380 int dissected_image_mount_and_warn(DissectedImage *m, const char *where, uid_t uid_shift, DissectImageFlags flags) {
1381 int r;
1382
1383 assert(m);
1384 assert(where);
1385
1386 r = dissected_image_mount(m, where, uid_shift, flags);
1387 if (r == -ENXIO)
1388 return log_error_errno(r, "Not root file system found in image.");
1389 if (r == -EMEDIUMTYPE)
1390 return log_error_errno(r, "No suitable os-release file in image found.");
1391 if (r == -EUNATCH)
1392 return log_error_errno(r, "Encrypted file system discovered, but decryption not requested.");
1393 if (r == -EUCLEAN)
1394 return log_error_errno(r, "File system check on image failed.");
1395 if (r == -EBUSY)
1396 return log_error_errno(r, "File system already mounted elsewhere.");
1397 if (r == -EAFNOSUPPORT)
1398 return log_error_errno(r, "File system type not supported or not known.");
1399 if (r < 0)
1400 return log_error_errno(r, "Failed to mount image: %m");
1401
1402 return r;
1403 }
1404
1405 #if HAVE_LIBCRYPTSETUP
1406 typedef struct DecryptedPartition {
1407 struct crypt_device *device;
1408 char *name;
1409 bool relinquished;
1410 } DecryptedPartition;
1411
1412 struct DecryptedImage {
1413 DecryptedPartition *decrypted;
1414 size_t n_decrypted;
1415 size_t n_allocated;
1416 };
1417 #endif
1418
1419 DecryptedImage* decrypted_image_unref(DecryptedImage* d) {
1420 #if HAVE_LIBCRYPTSETUP
1421 size_t i;
1422 int r;
1423
1424 if (!d)
1425 return NULL;
1426
1427 for (i = 0; i < d->n_decrypted; i++) {
1428 DecryptedPartition *p = d->decrypted + i;
1429
1430 if (p->device && p->name && !p->relinquished) {
1431 r = sym_crypt_deactivate_by_name(p->device, p->name, 0);
1432 if (r < 0)
1433 log_debug_errno(r, "Failed to deactivate encrypted partition %s", p->name);
1434 }
1435
1436 if (p->device)
1437 sym_crypt_free(p->device);
1438 free(p->name);
1439 }
1440
1441 free(d);
1442 #endif
1443 return NULL;
1444 }
1445
1446 #if HAVE_LIBCRYPTSETUP
1447
1448 static int make_dm_name_and_node(const void *original_node, const char *suffix, char **ret_name, char **ret_node) {
1449 _cleanup_free_ char *name = NULL, *node = NULL;
1450 const char *base;
1451
1452 assert(original_node);
1453 assert(suffix);
1454 assert(ret_name);
1455 assert(ret_node);
1456
1457 base = strrchr(original_node, '/');
1458 if (!base)
1459 base = original_node;
1460 else
1461 base++;
1462 if (isempty(base))
1463 return -EINVAL;
1464
1465 name = strjoin(base, suffix);
1466 if (!name)
1467 return -ENOMEM;
1468 if (!filename_is_valid(name))
1469 return -EINVAL;
1470
1471 node = path_join(sym_crypt_get_dir(), name);
1472 if (!node)
1473 return -ENOMEM;
1474
1475 *ret_name = TAKE_PTR(name);
1476 *ret_node = TAKE_PTR(node);
1477
1478 return 0;
1479 }
1480
1481 static int decrypt_partition(
1482 DissectedPartition *m,
1483 const char *passphrase,
1484 DissectImageFlags flags,
1485 DecryptedImage *d) {
1486
1487 _cleanup_free_ char *node = NULL, *name = NULL;
1488 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
1489 int r;
1490
1491 assert(m);
1492 assert(d);
1493
1494 if (!m->found || !m->node || !m->fstype)
1495 return 0;
1496
1497 if (!streq(m->fstype, "crypto_LUKS"))
1498 return 0;
1499
1500 if (!passphrase)
1501 return -ENOKEY;
1502
1503 r = dlopen_cryptsetup();
1504 if (r < 0)
1505 return r;
1506
1507 r = make_dm_name_and_node(m->node, "-decrypted", &name, &node);
1508 if (r < 0)
1509 return r;
1510
1511 if (!GREEDY_REALLOC0(d->decrypted, d->n_allocated, d->n_decrypted + 1))
1512 return -ENOMEM;
1513
1514 r = sym_crypt_init(&cd, m->node);
1515 if (r < 0)
1516 return log_debug_errno(r, "Failed to initialize dm-crypt: %m");
1517
1518 cryptsetup_enable_logging(cd);
1519
1520 r = sym_crypt_load(cd, CRYPT_LUKS, NULL);
1521 if (r < 0)
1522 return log_debug_errno(r, "Failed to load LUKS metadata: %m");
1523
1524 r = sym_crypt_activate_by_passphrase(cd, name, CRYPT_ANY_SLOT, passphrase, strlen(passphrase),
1525 ((flags & DISSECT_IMAGE_READ_ONLY) ? CRYPT_ACTIVATE_READONLY : 0) |
1526 ((flags & DISSECT_IMAGE_DISCARD_ON_CRYPTO) ? CRYPT_ACTIVATE_ALLOW_DISCARDS : 0));
1527 if (r < 0) {
1528 log_debug_errno(r, "Failed to activate LUKS device: %m");
1529 return r == -EPERM ? -EKEYREJECTED : r;
1530 }
1531
1532 d->decrypted[d->n_decrypted++] = (DecryptedPartition) {
1533 .name = TAKE_PTR(name),
1534 .device = TAKE_PTR(cd),
1535 };
1536
1537 m->decrypted_node = TAKE_PTR(node);
1538
1539 return 0;
1540 }
1541
1542 static int verity_can_reuse(
1543 const VeritySettings *verity,
1544 const char *name,
1545 struct crypt_device **ret_cd) {
1546
1547 /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
1548 _cleanup_free_ char *root_hash_existing = NULL;
1549 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
1550 struct crypt_params_verity crypt_params = {};
1551 size_t root_hash_existing_size;
1552 int r;
1553
1554 assert(verity);
1555 assert(name);
1556 assert(ret_cd);
1557
1558 r = sym_crypt_init_by_name(&cd, name);
1559 if (r < 0)
1560 return log_debug_errno(r, "Error opening verity device, crypt_init_by_name failed: %m");
1561
1562 r = sym_crypt_get_verity_info(cd, &crypt_params);
1563 if (r < 0)
1564 return log_debug_errno(r, "Error opening verity device, crypt_get_verity_info failed: %m");
1565
1566 root_hash_existing_size = verity->root_hash_size;
1567 root_hash_existing = malloc0(root_hash_existing_size);
1568 if (!root_hash_existing)
1569 return -ENOMEM;
1570
1571 r = sym_crypt_volume_key_get(cd, CRYPT_ANY_SLOT, root_hash_existing, &root_hash_existing_size, NULL, 0);
1572 if (r < 0)
1573 return log_debug_errno(r, "Error opening verity device, crypt_volume_key_get failed: %m");
1574 if (verity->root_hash_size != root_hash_existing_size ||
1575 memcmp(root_hash_existing, verity->root_hash, verity->root_hash_size) != 0)
1576 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but root hashes are different.");
1577
1578 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1579 /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the
1580 * same settings, so that a previous unsigned mount will not be reused if the user asks to use
1581 * signing for the new one, and viceversa. */
1582 if (!!verity->root_hash_sig != !!(crypt_params.flags & CRYPT_VERITY_ROOT_HASH_SIGNATURE))
1583 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but signature settings are not the same.");
1584 #endif
1585
1586 *ret_cd = TAKE_PTR(cd);
1587 return 0;
1588 }
1589
1590 static inline void dm_deferred_remove_clean(char *name) {
1591 if (!name)
1592 return;
1593
1594 (void) sym_crypt_deactivate_by_name(NULL, name, CRYPT_DEACTIVATE_DEFERRED);
1595 free(name);
1596 }
1597 DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean);
1598
1599 static int verity_partition(
1600 PartitionDesignator designator,
1601 DissectedPartition *m,
1602 DissectedPartition *v,
1603 const VeritySettings *verity,
1604 DissectImageFlags flags,
1605 DecryptedImage *d) {
1606
1607 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
1608 _cleanup_(dm_deferred_remove_cleanp) char *restore_deferred_remove = NULL;
1609 _cleanup_free_ char *node = NULL, *name = NULL;
1610 int r;
1611
1612 assert(m);
1613 assert(v || (verity && verity->data_path));
1614
1615 if (!verity || !verity->root_hash)
1616 return 0;
1617 if (!((verity->designator < 0 && designator == PARTITION_ROOT) ||
1618 (verity->designator == designator)))
1619 return 0;
1620
1621 if (!m->found || !m->node || !m->fstype)
1622 return 0;
1623 if (!verity->data_path) {
1624 if (!v->found || !v->node || !v->fstype)
1625 return 0;
1626
1627 if (!streq(v->fstype, "DM_verity_hash"))
1628 return 0;
1629 }
1630
1631 r = dlopen_cryptsetup();
1632 if (r < 0)
1633 return r;
1634
1635 if (FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) {
1636 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
1637 _cleanup_free_ char *root_hash_encoded = NULL;
1638
1639 root_hash_encoded = hexmem(verity->root_hash, verity->root_hash_size);
1640 if (!root_hash_encoded)
1641 return -ENOMEM;
1642
1643 r = make_dm_name_and_node(root_hash_encoded, "-verity", &name, &node);
1644 } else
1645 r = make_dm_name_and_node(m->node, "-verity", &name, &node);
1646 if (r < 0)
1647 return r;
1648
1649 r = sym_crypt_init(&cd, verity->data_path ?: v->node);
1650 if (r < 0)
1651 return r;
1652
1653 cryptsetup_enable_logging(cd);
1654
1655 r = sym_crypt_load(cd, CRYPT_VERITY, NULL);
1656 if (r < 0)
1657 return r;
1658
1659 r = sym_crypt_set_data_device(cd, m->node);
1660 if (r < 0)
1661 return r;
1662
1663 if (!GREEDY_REALLOC0(d->decrypted, d->n_allocated, d->n_decrypted + 1))
1664 return -ENOMEM;
1665
1666 /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
1667 * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
1668 * retry a few times before giving up. */
1669 for (unsigned i = 0; i < N_DEVICE_NODE_LIST_ATTEMPTS; i++) {
1670 if (verity->root_hash_sig) {
1671 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1672 r = sym_crypt_activate_by_signed_key(
1673 cd,
1674 name,
1675 verity->root_hash,
1676 verity->root_hash_size,
1677 verity->root_hash_sig,
1678 verity->root_hash_sig_size,
1679 CRYPT_ACTIVATE_READONLY);
1680 #else
1681 r = log_debug_errno(SYNTHETIC_ERRNO(EOPNOTSUPP),
1682 "Activation of verity device with signature requested, but not supported by %s due to missing crypt_activate_by_signed_key().", program_invocation_short_name);
1683 #endif
1684 } else
1685 r = sym_crypt_activate_by_volume_key(
1686 cd,
1687 name,
1688 verity->root_hash,
1689 verity->root_hash_size,
1690 CRYPT_ACTIVATE_READONLY);
1691 /* libdevmapper can return EINVAL when the device is already in the activation stage.
1692 * There's no way to distinguish this situation from a genuine error due to invalid
1693 * parameters, so immediately fall back to activating the device with a unique name.
1694 * Improvements in libcrypsetup can ensure this never happens:
1695 * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
1696 if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
1697 return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
1698 if (!IN_SET(r,
1699 0, /* Success */
1700 -EEXIST, /* Volume is already open and ready to be used */
1701 -EBUSY, /* Volume is being opened but not ready, crypt_init_by_name can fetch details */
1702 -ENODEV /* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */))
1703 return r;
1704 if (IN_SET(r, -EEXIST, -EBUSY)) {
1705 struct crypt_device *existing_cd = NULL;
1706
1707 if (!restore_deferred_remove){
1708 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
1709 r = dm_deferred_remove_cancel(name);
1710 /* If activation returns EBUSY there might be no deferred removal to cancel, that's fine */
1711 if (r < 0 && r != -ENXIO)
1712 return log_debug_errno(r, "Disabling automated deferred removal for verity device %s failed: %m", node);
1713 if (r == 0) {
1714 restore_deferred_remove = strdup(name);
1715 if (!restore_deferred_remove)
1716 return -ENOMEM;
1717 }
1718 }
1719
1720 r = verity_can_reuse(verity, name, &existing_cd);
1721 /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
1722 if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
1723 return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
1724 if (!IN_SET(r, 0, -ENODEV, -ENOENT, -EBUSY))
1725 return log_debug_errno(r, "Checking whether existing verity device %s can be reused failed: %m", node);
1726 if (r == 0) {
1727 /* devmapper might say that the device exists, but the devlink might not yet have been
1728 * created. Check and wait for the udev event in that case. */
1729 r = device_wait_for_devlink(node, "block", 100 * USEC_PER_MSEC, NULL);
1730 /* Fallback to activation with a unique device if it's taking too long */
1731 if (r == -ETIMEDOUT)
1732 break;
1733 if (r < 0)
1734 return r;
1735
1736 if (cd)
1737 sym_crypt_free(cd);
1738 cd = existing_cd;
1739 }
1740 }
1741 if (r == 0)
1742 break;
1743
1744 /* Device is being opened by another process, but it has not finished yet, yield for 2ms */
1745 (void) usleep(2 * USEC_PER_MSEC);
1746 }
1747
1748 /* An existing verity device was reported by libcryptsetup/libdevmapper, but we can't use it at this time.
1749 * Fall back to activating it with a unique device name. */
1750 if (r != 0 && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
1751 return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
1752
1753 /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
1754 restore_deferred_remove = mfree(restore_deferred_remove);
1755
1756 d->decrypted[d->n_decrypted++] = (DecryptedPartition) {
1757 .name = TAKE_PTR(name),
1758 .device = TAKE_PTR(cd),
1759 };
1760
1761 m->decrypted_node = TAKE_PTR(node);
1762
1763 return 0;
1764 }
1765 #endif
1766
1767 int dissected_image_decrypt(
1768 DissectedImage *m,
1769 const char *passphrase,
1770 const VeritySettings *verity,
1771 DissectImageFlags flags,
1772 DecryptedImage **ret) {
1773
1774 #if HAVE_LIBCRYPTSETUP
1775 _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL;
1776 int r;
1777 #endif
1778
1779 assert(m);
1780 assert(!verity || verity->root_hash || verity->root_hash_size == 0);
1781
1782 /* Returns:
1783 *
1784 * = 0 → There was nothing to decrypt
1785 * > 0 → Decrypted successfully
1786 * -ENOKEY → There's something to decrypt but no key was supplied
1787 * -EKEYREJECTED → Passed key was not correct
1788 */
1789
1790 if (verity && verity->root_hash && verity->root_hash_size < sizeof(sd_id128_t))
1791 return -EINVAL;
1792
1793 if (!m->encrypted && !m->verity) {
1794 *ret = NULL;
1795 return 0;
1796 }
1797
1798 #if HAVE_LIBCRYPTSETUP
1799 d = new0(DecryptedImage, 1);
1800 if (!d)
1801 return -ENOMEM;
1802
1803 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
1804 DissectedPartition *p = m->partitions + i;
1805 PartitionDesignator k;
1806
1807 if (!p->found)
1808 continue;
1809
1810 r = decrypt_partition(p, passphrase, flags, d);
1811 if (r < 0)
1812 return r;
1813
1814 k = PARTITION_VERITY_OF(i);
1815 if (k >= 0) {
1816 r = verity_partition(i, p, m->partitions + k, verity, flags | DISSECT_IMAGE_VERITY_SHARE, d);
1817 if (r < 0)
1818 return r;
1819 }
1820
1821 if (!p->decrypted_fstype && p->decrypted_node) {
1822 r = probe_filesystem(p->decrypted_node, &p->decrypted_fstype);
1823 if (r < 0 && r != -EUCLEAN)
1824 return r;
1825 }
1826 }
1827
1828 *ret = TAKE_PTR(d);
1829
1830 return 1;
1831 #else
1832 return -EOPNOTSUPP;
1833 #endif
1834 }
1835
1836 int dissected_image_decrypt_interactively(
1837 DissectedImage *m,
1838 const char *passphrase,
1839 const VeritySettings *verity,
1840 DissectImageFlags flags,
1841 DecryptedImage **ret) {
1842
1843 _cleanup_strv_free_erase_ char **z = NULL;
1844 int n = 3, r;
1845
1846 if (passphrase)
1847 n--;
1848
1849 for (;;) {
1850 r = dissected_image_decrypt(m, passphrase, verity, flags, ret);
1851 if (r >= 0)
1852 return r;
1853 if (r == -EKEYREJECTED)
1854 log_error_errno(r, "Incorrect passphrase, try again!");
1855 else if (r != -ENOKEY)
1856 return log_error_errno(r, "Failed to decrypt image: %m");
1857
1858 if (--n < 0)
1859 return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED),
1860 "Too many retries.");
1861
1862 z = strv_free(z);
1863
1864 r = ask_password_auto("Please enter image passphrase:", NULL, "dissect", "dissect", USEC_INFINITY, 0, &z);
1865 if (r < 0)
1866 return log_error_errno(r, "Failed to query for passphrase: %m");
1867
1868 passphrase = z[0];
1869 }
1870 }
1871
1872 int decrypted_image_relinquish(DecryptedImage *d) {
1873
1874 #if HAVE_LIBCRYPTSETUP
1875 size_t i;
1876 int r;
1877 #endif
1878
1879 assert(d);
1880
1881 /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a boolean so
1882 * that we don't clean it up ourselves either anymore */
1883
1884 #if HAVE_LIBCRYPTSETUP
1885 for (i = 0; i < d->n_decrypted; i++) {
1886 DecryptedPartition *p = d->decrypted + i;
1887
1888 if (p->relinquished)
1889 continue;
1890
1891 r = sym_crypt_deactivate_by_name(NULL, p->name, CRYPT_DEACTIVATE_DEFERRED);
1892 if (r < 0)
1893 return log_debug_errno(r, "Failed to mark %s for auto-removal: %m", p->name);
1894
1895 p->relinquished = true;
1896 }
1897 #endif
1898
1899 return 0;
1900 }
1901
1902 static char *build_auxiliary_path(const char *image, const char *suffix) {
1903 const char *e;
1904 char *n;
1905
1906 assert(image);
1907 assert(suffix);
1908
1909 e = endswith(image, ".raw");
1910 if (!e)
1911 return strjoin(e, suffix);
1912
1913 n = new(char, e - image + strlen(suffix) + 1);
1914 if (!n)
1915 return NULL;
1916
1917 strcpy(mempcpy(n, image, e - image), suffix);
1918 return n;
1919 }
1920
1921 void verity_settings_done(VeritySettings *v) {
1922 assert(v);
1923
1924 v->root_hash = mfree(v->root_hash);
1925 v->root_hash_size = 0;
1926
1927 v->root_hash_sig = mfree(v->root_hash_sig);
1928 v->root_hash_sig_size = 0;
1929
1930 v->data_path = mfree(v->data_path);
1931 }
1932
1933 int verity_settings_load(
1934 VeritySettings *verity,
1935 const char *image,
1936 const char *root_hash_path,
1937 const char *root_hash_sig_path) {
1938
1939 _cleanup_free_ void *root_hash = NULL, *root_hash_sig = NULL;
1940 size_t root_hash_size = 0, root_hash_sig_size = 0;
1941 _cleanup_free_ char *verity_data_path = NULL;
1942 PartitionDesignator designator;
1943 int r;
1944
1945 assert(verity);
1946 assert(image);
1947 assert(verity->designator < 0 || IN_SET(verity->designator, PARTITION_ROOT, PARTITION_USR));
1948
1949 /* If we are asked to load the root hash for a device node, exit early */
1950 if (is_device_path(image))
1951 return 0;
1952
1953 designator = verity->designator;
1954
1955 /* We only fill in what isn't already filled in */
1956
1957 if (!verity->root_hash) {
1958 _cleanup_free_ char *text = NULL;
1959
1960 if (root_hash_path) {
1961 /* If explicitly specified it takes precedence */
1962 r = read_one_line_file(root_hash_path, &text);
1963 if (r < 0)
1964 return r;
1965
1966 if (designator < 0)
1967 designator = PARTITION_ROOT;
1968 } else {
1969 /* Otherwise look for xattr and separate file, and first for the data for root and if
1970 * that doesn't exist for /usr */
1971
1972 if (designator < 0 || designator == PARTITION_ROOT) {
1973 r = getxattr_malloc(image, "user.verity.roothash", &text, true);
1974 if (r < 0) {
1975 _cleanup_free_ char *p = NULL;
1976
1977 if (!IN_SET(r, -ENODATA, -ENOENT) && !ERRNO_IS_NOT_SUPPORTED(r))
1978 return r;
1979
1980 p = build_auxiliary_path(image, ".roothash");
1981 if (!p)
1982 return -ENOMEM;
1983
1984 r = read_one_line_file(p, &text);
1985 if (r < 0 && r != -ENOENT)
1986 return r;
1987 }
1988
1989 if (text)
1990 designator = PARTITION_ROOT;
1991 }
1992
1993 if (!text && (designator < 0 || designator == PARTITION_USR)) {
1994 /* So in the "roothash" xattr/file name above the "root" of course primarily
1995 * refers to the root of the Verity Merkle tree. But coincidentally it also
1996 * is the hash for the *root* file system, i.e. the "root" neatly refers to
1997 * two distinct concepts called "root". Taking benefit of this happy
1998 * coincidence we call the file with the root hash for the /usr/ file system
1999 * `usrhash`, because `usrroothash` or `rootusrhash` would just be too
2000 * confusing. We thus drop the reference to the root of the Merkle tree, and
2001 * just indicate which file system it's about. */
2002 r = getxattr_malloc(image, "user.verity.usrhash", &text, true);
2003 if (r < 0) {
2004 _cleanup_free_ char *p = NULL;
2005
2006 if (!IN_SET(r, -ENODATA, -ENOENT) && !ERRNO_IS_NOT_SUPPORTED(r))
2007 return r;
2008
2009 p = build_auxiliary_path(image, ".usrhash");
2010 if (!p)
2011 return -ENOMEM;
2012
2013 r = read_one_line_file(p, &text);
2014 if (r < 0 && r != -ENOENT)
2015 return r;
2016 }
2017
2018 if (text)
2019 designator = PARTITION_USR;
2020 }
2021 }
2022
2023 if (text) {
2024 r = unhexmem(text, strlen(text), &root_hash, &root_hash_size);
2025 if (r < 0)
2026 return r;
2027 if (root_hash_size < sizeof(sd_id128_t))
2028 return -EINVAL;
2029 }
2030 }
2031
2032 if (verity->root_hash && !verity->root_hash_sig) {
2033 if (root_hash_sig_path) {
2034 r = read_full_file_full(AT_FDCWD, root_hash_sig_path, 0, (char**) &root_hash_sig, &root_hash_sig_size);
2035 if (r < 0 && r != -ENOENT)
2036 return r;
2037
2038 if (designator < 0)
2039 designator = PARTITION_ROOT;
2040 } else {
2041 if (designator < 0 || designator == PARTITION_ROOT) {
2042 _cleanup_free_ char *p = NULL;
2043
2044 /* Follow naming convention recommended by the relevant RFC:
2045 * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
2046 p = build_auxiliary_path(image, ".roothash.p7s");
2047 if (!p)
2048 return -ENOMEM;
2049
2050 r = read_full_file_full(AT_FDCWD, p, 0, (char**) &root_hash_sig, &root_hash_sig_size);
2051 if (r < 0 && r != -ENOENT)
2052 return r;
2053 if (r >= 0)
2054 designator = PARTITION_ROOT;
2055 }
2056
2057 if (!root_hash_sig && (designator < 0 || designator == PARTITION_USR)) {
2058 _cleanup_free_ char *p = NULL;
2059
2060 p = build_auxiliary_path(image, ".usrhash.p7s");
2061 if (!p)
2062 return -ENOMEM;
2063
2064 r = read_full_file_full(AT_FDCWD, p, 0, (char**) &root_hash_sig, &root_hash_sig_size);
2065 if (r < 0 && r != -ENOENT)
2066 return r;
2067 if (r >= 0)
2068 designator = PARTITION_USR;
2069 }
2070 }
2071
2072 if (root_hash_sig && root_hash_sig_size == 0) /* refuse empty size signatures */
2073 return -EINVAL;
2074 }
2075
2076 if (!verity->data_path) {
2077 _cleanup_free_ char *p = NULL;
2078
2079 p = build_auxiliary_path(image, ".verity");
2080 if (!p)
2081 return -ENOMEM;
2082
2083 if (access(p, F_OK) < 0) {
2084 if (errno != ENOENT)
2085 return -errno;
2086 } else
2087 verity_data_path = TAKE_PTR(p);
2088 }
2089
2090 if (root_hash) {
2091 verity->root_hash = TAKE_PTR(root_hash);
2092 verity->root_hash_size = root_hash_size;
2093 }
2094
2095 if (root_hash_sig) {
2096 verity->root_hash_sig = TAKE_PTR(root_hash_sig);
2097 verity->root_hash_sig_size = root_hash_sig_size;
2098 }
2099
2100 if (verity_data_path)
2101 verity->data_path = TAKE_PTR(verity_data_path);
2102
2103 if (verity->designator < 0)
2104 verity->designator = designator;
2105
2106 return 1;
2107 }
2108
2109 int dissected_image_acquire_metadata(DissectedImage *m) {
2110
2111 enum {
2112 META_HOSTNAME,
2113 META_MACHINE_ID,
2114 META_MACHINE_INFO,
2115 META_OS_RELEASE,
2116 _META_MAX,
2117 };
2118
2119 static const char *const paths[_META_MAX] = {
2120 [META_HOSTNAME] = "/etc/hostname\0",
2121 [META_MACHINE_ID] = "/etc/machine-id\0",
2122 [META_MACHINE_INFO] = "/etc/machine-info\0",
2123 [META_OS_RELEASE] = "/etc/os-release\0"
2124 "/usr/lib/os-release\0",
2125 };
2126
2127 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL;
2128 _cleanup_close_pair_ int error_pipe[2] = { -1, -1 };
2129 _cleanup_(rmdir_and_freep) char *t = NULL;
2130 _cleanup_(sigkill_waitp) pid_t child = 0;
2131 sd_id128_t machine_id = SD_ID128_NULL;
2132 _cleanup_free_ char *hostname = NULL;
2133 unsigned n_meta_initialized = 0, k;
2134 int fds[2 * _META_MAX], r, v;
2135 ssize_t n;
2136
2137 BLOCK_SIGNALS(SIGCHLD);
2138
2139 assert(m);
2140
2141 for (; n_meta_initialized < _META_MAX; n_meta_initialized ++)
2142 if (pipe2(fds + 2*n_meta_initialized, O_CLOEXEC) < 0) {
2143 r = -errno;
2144 goto finish;
2145 }
2146
2147 r = mkdtemp_malloc("/tmp/dissect-XXXXXX", &t);
2148 if (r < 0)
2149 goto finish;
2150
2151 if (pipe2(error_pipe, O_CLOEXEC) < 0) {
2152 r = -errno;
2153 goto finish;
2154 }
2155
2156 r = safe_fork("(sd-dissect)", FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE, &child);
2157 if (r < 0)
2158 goto finish;
2159 if (r == 0) {
2160 error_pipe[0] = safe_close(error_pipe[0]);
2161
2162 r = dissected_image_mount(m, t, UID_INVALID, DISSECT_IMAGE_READ_ONLY|DISSECT_IMAGE_MOUNT_ROOT_ONLY|DISSECT_IMAGE_VALIDATE_OS);
2163 if (r < 0) {
2164 /* Let parent know the error */
2165 (void) write(error_pipe[1], &r, sizeof(r));
2166
2167 log_debug_errno(r, "Failed to mount dissected image: %m");
2168 _exit(EXIT_FAILURE);
2169 }
2170
2171 for (k = 0; k < _META_MAX; k++) {
2172 _cleanup_close_ int fd = -ENOENT;
2173 const char *p;
2174
2175 fds[2*k] = safe_close(fds[2*k]);
2176
2177 NULSTR_FOREACH(p, paths[k]) {
2178 fd = chase_symlinks_and_open(p, t, CHASE_PREFIX_ROOT, O_RDONLY|O_CLOEXEC|O_NOCTTY, NULL);
2179 if (fd >= 0)
2180 break;
2181 }
2182 if (fd < 0) {
2183 log_debug_errno(fd, "Failed to read %s file of image, ignoring: %m", paths[k]);
2184 fds[2*k+1] = safe_close(fds[2*k+1]);
2185 continue;
2186 }
2187
2188 r = copy_bytes(fd, fds[2*k+1], (uint64_t) -1, 0);
2189 if (r < 0) {
2190 (void) write(error_pipe[1], &r, sizeof(r));
2191 _exit(EXIT_FAILURE);
2192 }
2193
2194 fds[2*k+1] = safe_close(fds[2*k+1]);
2195 }
2196
2197 _exit(EXIT_SUCCESS);
2198 }
2199
2200 error_pipe[1] = safe_close(error_pipe[1]);
2201
2202 for (k = 0; k < _META_MAX; k++) {
2203 _cleanup_fclose_ FILE *f = NULL;
2204
2205 fds[2*k+1] = safe_close(fds[2*k+1]);
2206
2207 f = take_fdopen(&fds[2*k], "r");
2208 if (!f) {
2209 r = -errno;
2210 goto finish;
2211 }
2212
2213 switch (k) {
2214
2215 case META_HOSTNAME:
2216 r = read_etc_hostname_stream(f, &hostname);
2217 if (r < 0)
2218 log_debug_errno(r, "Failed to read /etc/hostname: %m");
2219
2220 break;
2221
2222 case META_MACHINE_ID: {
2223 _cleanup_free_ char *line = NULL;
2224
2225 r = read_line(f, LONG_LINE_MAX, &line);
2226 if (r < 0)
2227 log_debug_errno(r, "Failed to read /etc/machine-id: %m");
2228 else if (r == 33) {
2229 r = sd_id128_from_string(line, &machine_id);
2230 if (r < 0)
2231 log_debug_errno(r, "Image contains invalid /etc/machine-id: %s", line);
2232 } else if (r == 0)
2233 log_debug("/etc/machine-id file is empty.");
2234 else if (streq(line, "uninitialized"))
2235 log_debug("/etc/machine-id file is uninitialized (likely aborted first boot).");
2236 else
2237 log_debug("/etc/machine-id has unexpected length %i.", r);
2238
2239 break;
2240 }
2241
2242 case META_MACHINE_INFO:
2243 r = load_env_file_pairs(f, "machine-info", &machine_info);
2244 if (r < 0)
2245 log_debug_errno(r, "Failed to read /etc/machine-info: %m");
2246
2247 break;
2248
2249 case META_OS_RELEASE:
2250 r = load_env_file_pairs(f, "os-release", &os_release);
2251 if (r < 0)
2252 log_debug_errno(r, "Failed to read OS release file: %m");
2253
2254 break;
2255 }
2256 }
2257
2258 r = wait_for_terminate_and_check("(sd-dissect)", child, 0);
2259 child = 0;
2260 if (r < 0)
2261 return r;
2262
2263 n = read(error_pipe[0], &v, sizeof(v));
2264 if (n < 0)
2265 return -errno;
2266 if (n == sizeof(v))
2267 return v; /* propagate error sent to us from child */
2268 if (n != 0)
2269 return -EIO;
2270
2271 if (r != EXIT_SUCCESS)
2272 return -EPROTO;
2273
2274 free_and_replace(m->hostname, hostname);
2275 m->machine_id = machine_id;
2276 strv_free_and_replace(m->machine_info, machine_info);
2277 strv_free_and_replace(m->os_release, os_release);
2278
2279 finish:
2280 for (k = 0; k < n_meta_initialized; k++)
2281 safe_close_pair(fds + 2*k);
2282
2283 return r;
2284 }
2285
2286 int dissect_image_and_warn(
2287 int fd,
2288 const char *name,
2289 const VeritySettings *verity,
2290 const MountOptions *mount_options,
2291 DissectImageFlags flags,
2292 DissectedImage **ret) {
2293
2294 _cleanup_free_ char *buffer = NULL;
2295 int r;
2296
2297 if (!name) {
2298 r = fd_get_path(fd, &buffer);
2299 if (r < 0)
2300 return r;
2301
2302 name = buffer;
2303 }
2304
2305 r = dissect_image(fd, verity, mount_options, flags, ret);
2306 switch (r) {
2307
2308 case -EOPNOTSUPP:
2309 return log_error_errno(r, "Dissecting images is not supported, compiled without blkid support.");
2310
2311 case -ENOPKG:
2312 return log_error_errno(r, "Couldn't identify a suitable partition table or file system in '%s'.", name);
2313
2314 case -EADDRNOTAVAIL:
2315 return log_error_errno(r, "No root partition for specified root hash found in '%s'.", name);
2316
2317 case -ENOTUNIQ:
2318 return log_error_errno(r, "Multiple suitable root partitions found in image '%s'.", name);
2319
2320 case -ENXIO:
2321 return log_error_errno(r, "No suitable root partition found in image '%s'.", name);
2322
2323 case -EPROTONOSUPPORT:
2324 return log_error_errno(r, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name);
2325
2326 default:
2327 if (r < 0)
2328 return log_error_errno(r, "Failed to dissect image '%s': %m", name);
2329
2330 return r;
2331 }
2332 }
2333
2334 bool dissected_image_can_do_verity(const DissectedImage *image, PartitionDesignator partition_designator) {
2335 if (image->single_file_system)
2336 return partition_designator == PARTITION_ROOT && image->can_verity;
2337
2338 return PARTITION_VERITY_OF(partition_designator) >= 0;
2339 }
2340
2341 bool dissected_image_has_verity(const DissectedImage *image, PartitionDesignator partition_designator) {
2342 int k;
2343
2344 if (image->single_file_system)
2345 return partition_designator == PARTITION_ROOT && image->verity;
2346
2347 k = PARTITION_VERITY_OF(partition_designator);
2348 return k >= 0 && image->partitions[k].found;
2349 }
2350
2351 MountOptions* mount_options_free_all(MountOptions *options) {
2352 MountOptions *m;
2353
2354 while ((m = options)) {
2355 LIST_REMOVE(mount_options, options, m);
2356 free(m->options);
2357 free(m);
2358 }
2359
2360 return NULL;
2361 }
2362
2363 const char* mount_options_from_designator(const MountOptions *options, PartitionDesignator designator) {
2364 const MountOptions *m;
2365
2366 LIST_FOREACH(mount_options, m, options)
2367 if (designator == m->partition_designator && !isempty(m->options))
2368 return m->options;
2369
2370 return NULL;
2371 }
2372
2373 int mount_image_privately_interactively(
2374 const char *image,
2375 DissectImageFlags flags,
2376 char **ret_directory,
2377 LoopDevice **ret_loop_device,
2378 DecryptedImage **ret_decrypted_image) {
2379
2380 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
2381 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
2382 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
2383 _cleanup_(rmdir_and_freep) char *created_dir = NULL;
2384 _cleanup_free_ char *temp = NULL;
2385 int r;
2386
2387 /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
2388 * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
2389 * easily. */
2390
2391 assert(image);
2392 assert(ret_directory);
2393 assert(ret_loop_device);
2394 assert(ret_decrypted_image);
2395
2396 r = tempfn_random_child(NULL, program_invocation_short_name, &temp);
2397 if (r < 0)
2398 return log_error_errno(r, "Failed to generate temporary mount directory: %m");
2399
2400 r = loop_device_make_by_path(
2401 image,
2402 FLAGS_SET(flags, DISSECT_IMAGE_READ_ONLY) ? O_RDONLY : O_RDWR,
2403 FLAGS_SET(flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN,
2404 &d);
2405 if (r < 0)
2406 return log_error_errno(r, "Failed to set up loopback device: %m");
2407
2408 r = dissect_image_and_warn(d->fd, image, NULL, NULL, flags, &dissected_image);
2409 if (r < 0)
2410 return r;
2411
2412 r = dissected_image_decrypt_interactively(dissected_image, NULL, NULL, flags, &decrypted_image);
2413 if (r < 0)
2414 return r;
2415
2416 r = detach_mount_namespace();
2417 if (r < 0)
2418 return log_error_errno(r, "Failed to detach mount namespace: %m");
2419
2420 r = mkdir_p(temp, 0700);
2421 if (r < 0)
2422 return log_error_errno(r, "Failed to create mount point: %m");
2423
2424 created_dir = TAKE_PTR(temp);
2425
2426 r = dissected_image_mount_and_warn(dissected_image, created_dir, UID_INVALID, flags);
2427 if (r < 0)
2428 return r;
2429
2430 if (decrypted_image) {
2431 r = decrypted_image_relinquish(decrypted_image);
2432 if (r < 0)
2433 return log_error_errno(r, "Failed to relinquish DM devices: %m");
2434 }
2435
2436 loop_device_relinquish(d);
2437
2438 *ret_directory = TAKE_PTR(created_dir);
2439 *ret_loop_device = TAKE_PTR(d);
2440 *ret_decrypted_image = TAKE_PTR(decrypted_image);
2441
2442 return 0;
2443 }
2444
2445 static const char *const partition_designator_table[] = {
2446 [PARTITION_ROOT] = "root",
2447 [PARTITION_ROOT_SECONDARY] = "root-secondary",
2448 [PARTITION_USR] = "usr",
2449 [PARTITION_USR_SECONDARY] = "usr-secondary",
2450 [PARTITION_HOME] = "home",
2451 [PARTITION_SRV] = "srv",
2452 [PARTITION_ESP] = "esp",
2453 [PARTITION_XBOOTLDR] = "xbootldr",
2454 [PARTITION_SWAP] = "swap",
2455 [PARTITION_ROOT_VERITY] = "root-verity",
2456 [PARTITION_ROOT_SECONDARY_VERITY] = "root-secondary-verity",
2457 [PARTITION_USR_VERITY] = "usr-verity",
2458 [PARTITION_USR_SECONDARY_VERITY] = "usr-secondary-verity",
2459 [PARTITION_TMP] = "tmp",
2460 [PARTITION_VAR] = "var",
2461 };
2462
2463 DEFINE_STRING_TABLE_LOOKUP(partition_designator, PartitionDesignator);
Unnamed repository; edit this file 'description' to name the repository.