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/blkpg.h>
8 #include <linux/dm-ioctl.h>
9 #include <linux/loop.h>
10 #include <sys/mount.h>
11 #include <sys/prctl.h>
12 #include <sys/wait.h>
13 #include <sysexits.h>
14
15 #if HAVE_OPENSSL
16 #include <openssl/err.h>
17 #include <openssl/pem.h>
18 #include <openssl/x509.h>
19 #endif
20
21 #include "sd-device.h"
22 #include "sd-id128.h"
23
24 #include "architecture.h"
25 #include "ask-password-api.h"
26 #include "blkid-util.h"
27 #include "blockdev-util.h"
28 #include "chase-symlinks.h"
29 #include "conf-files.h"
30 #include "copy.h"
31 #include "cryptsetup-util.h"
32 #include "def.h"
33 #include "device-nodes.h"
34 #include "device-util.h"
35 #include "discover-image.h"
36 #include "dissect-image.h"
37 #include "dm-util.h"
38 #include "env-file.h"
39 #include "env-util.h"
40 #include "extension-release.h"
41 #include "fd-util.h"
42 #include "fileio.h"
43 #include "fs-util.h"
44 #include "fsck-util.h"
45 #include "gpt.h"
46 #include "hexdecoct.h"
47 #include "hostname-setup.h"
48 #include "id128-util.h"
49 #include "import-util.h"
50 #include "io-util.h"
51 #include "mkdir-label.h"
52 #include "mount-util.h"
53 #include "mountpoint-util.h"
54 #include "namespace-util.h"
55 #include "nulstr-util.h"
56 #include "openssl-util.h"
57 #include "os-util.h"
58 #include "path-util.h"
59 #include "process-util.h"
60 #include "raw-clone.h"
61 #include "resize-fs.h"
62 #include "signal-util.h"
63 #include "stat-util.h"
64 #include "stdio-util.h"
65 #include "string-table.h"
66 #include "string-util.h"
67 #include "strv.h"
68 #include "tmpfile-util.h"
69 #include "udev-util.h"
70 #include "user-util.h"
71 #include "xattr-util.h"
72
73 /* how many times to wait for the device nodes to appear */
74 #define N_DEVICE_NODE_LIST_ATTEMPTS 10
75
probe_filesystem(const char * node,char ** ret_fstype)76 int probe_filesystem(const char *node, char **ret_fstype) {
77 /* Try to find device content type and return it in *ret_fstype. If nothing is found,
78 * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and an
79 * different error otherwise. */
80
81 #if HAVE_BLKID
82 _cleanup_(blkid_free_probep) blkid_probe b = NULL;
83 const char *fstype;
84 int r;
85
86 errno = 0;
87 b = blkid_new_probe_from_filename(node);
88 if (!b)
89 return errno_or_else(ENOMEM);
90
91 blkid_probe_enable_superblocks(b, 1);
92 blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE);
93
94 errno = 0;
95 r = blkid_do_safeprobe(b);
96 if (r == 1)
97 goto not_found;
98 if (r == -2)
99 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN),
100 "Results ambiguous for partition %s", node);
101 if (r != 0)
102 return log_debug_errno(errno_or_else(EIO), "Failed to probe partition %s: %m", node);
103
104 (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
105
106 if (fstype) {
107 char *t;
108
109 log_debug("Probed fstype '%s' on partition %s.", fstype, node);
110
111 t = strdup(fstype);
112 if (!t)
113 return -ENOMEM;
114
115 *ret_fstype = t;
116 return 1;
117 }
118
119 not_found:
120 log_debug("No type detected on partition %s", node);
121 *ret_fstype = NULL;
122 return 0;
123 #else
124 return -EOPNOTSUPP;
125 #endif
126 }
127
128 #if HAVE_BLKID
check_partition_flags(const char * node,unsigned long long pflags,unsigned long long supported)129 static void check_partition_flags(
130 const char *node,
131 unsigned long long pflags,
132 unsigned long long supported) {
133
134 assert(node);
135
136 /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
137 pflags &= ~(supported | GPT_FLAG_REQUIRED_PARTITION | GPT_FLAG_NO_BLOCK_IO_PROTOCOL | GPT_FLAG_LEGACY_BIOS_BOOTABLE);
138
139 if (pflags == 0)
140 return;
141
142 /* If there are other bits set, then log about it, to make things discoverable */
143 for (unsigned i = 0; i < sizeof(pflags) * 8; i++) {
144 unsigned long long bit = 1ULL << i;
145 if (!FLAGS_SET(pflags, bit))
146 continue;
147
148 log_debug("Unexpected partition flag %llu set on %s!", bit, node);
149 }
150 }
151 #endif
152
dissected_partition_done(DissectedPartition * p)153 static void dissected_partition_done(DissectedPartition *p) {
154 assert(p);
155
156 free(p->fstype);
157 free(p->node);
158 free(p->label);
159 free(p->decrypted_fstype);
160 free(p->decrypted_node);
161 free(p->mount_options);
162
163 *p = (DissectedPartition) {
164 .partno = -1,
165 .architecture = _ARCHITECTURE_INVALID,
166 };
167 }
168
169 #if HAVE_BLKID
ioctl_partition_add(int fd,const char * name,int nr,uint64_t start,uint64_t size)170 static int ioctl_partition_add(
171 int fd,
172 const char *name,
173 int nr,
174 uint64_t start,
175 uint64_t size) {
176
177 assert(fd >= 0);
178 assert(name);
179 assert(nr > 0);
180
181 struct blkpg_partition bp = {
182 .pno = nr,
183 .start = start,
184 .length = size,
185 };
186
187 struct blkpg_ioctl_arg ba = {
188 .op = BLKPG_ADD_PARTITION,
189 .data = &bp,
190 .datalen = sizeof(bp),
191 };
192
193 if (strlen(name) >= sizeof(bp.devname))
194 return -EINVAL;
195
196 strcpy(bp.devname, name);
197
198 return RET_NERRNO(ioctl(fd, BLKPG, &ba));
199 }
200
make_partition_devname(const char * whole_devname,int nr,char ** ret)201 static int make_partition_devname(
202 const char *whole_devname,
203 int nr,
204 char **ret) {
205
206 bool need_p;
207
208 assert(whole_devname);
209 assert(nr > 0);
210
211 /* Given a whole block device node name (e.g. /dev/sda or /dev/loop7) generate a partition device
212 * name (e.g. /dev/sda7 or /dev/loop7p5). The rule the kernel uses is simple: if whole block device
213 * node name ends in a digit, then suffix a 'p', followed by the partition number. Otherwise, just
214 * suffix the partition number without any 'p'. */
215
216 if (isempty(whole_devname)) /* Make sure there *is* a last char */
217 return -EINVAL;
218
219 need_p = strchr(DIGITS, whole_devname[strlen(whole_devname)-1]); /* Last char a digit? */
220
221 return asprintf(ret, "%s%s%i", whole_devname, need_p ? "p" : "", nr);
222 }
223 #endif
224
dissect_image(int fd,const VeritySettings * verity,const MountOptions * mount_options,uint64_t diskseq,uint64_t uevent_seqnum_not_before,usec_t timestamp_not_before,DissectImageFlags flags,DissectedImage ** ret)225 int dissect_image(
226 int fd,
227 const VeritySettings *verity,
228 const MountOptions *mount_options,
229 uint64_t diskseq,
230 uint64_t uevent_seqnum_not_before,
231 usec_t timestamp_not_before,
232 DissectImageFlags flags,
233 DissectedImage **ret) {
234
235 #if HAVE_BLKID
236 sd_id128_t root_uuid = SD_ID128_NULL, root_verity_uuid = SD_ID128_NULL;
237 sd_id128_t usr_uuid = SD_ID128_NULL, usr_verity_uuid = SD_ID128_NULL;
238 bool is_gpt, is_mbr, multiple_generic = false,
239 generic_rw = false, /* initialize to appease gcc */
240 generic_growfs = false;
241 _cleanup_(sd_device_unrefp) sd_device *d = NULL;
242 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
243 _cleanup_(blkid_free_probep) blkid_probe b = NULL;
244 _cleanup_free_ char *generic_node = NULL;
245 sd_id128_t generic_uuid = SD_ID128_NULL;
246 const char *pttype = NULL, *sysname = NULL, *devname = NULL;
247 blkid_partlist pl;
248 int r, generic_nr = -1, n_partitions;
249 struct stat st;
250
251 assert(fd >= 0);
252 assert(ret);
253 assert(!verity || verity->designator < 0 || IN_SET(verity->designator, PARTITION_ROOT, PARTITION_USR));
254 assert(!verity || verity->root_hash || verity->root_hash_size == 0);
255 assert(!verity || verity->root_hash_sig || verity->root_hash_sig_size == 0);
256 assert(!verity || (verity->root_hash || !verity->root_hash_sig));
257 assert(!((flags & DISSECT_IMAGE_GPT_ONLY) && (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)));
258
259 /* Probes a disk image, and returns information about what it found in *ret.
260 *
261 * Returns -ENOPKG if no suitable partition table or file system could be found.
262 * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found.
263 * Returns -ENXIO if we couldn't find any partition suitable as root or /usr partition
264 * Returns -ENOTUNIQ if we only found multiple generic partitions and thus don't know what to do with that */
265
266 if (verity && verity->root_hash) {
267 sd_id128_t fsuuid, vuuid;
268
269 /* If a root hash is supplied, then we use the root partition that has a UUID that match the
270 * first 128bit of the root hash. And we use the verity partition that has a UUID that match
271 * the final 128bit. */
272
273 if (verity->root_hash_size < sizeof(sd_id128_t))
274 return -EINVAL;
275
276 memcpy(&fsuuid, verity->root_hash, sizeof(sd_id128_t));
277 memcpy(&vuuid, (const uint8_t*) verity->root_hash + verity->root_hash_size - sizeof(sd_id128_t), sizeof(sd_id128_t));
278
279 if (sd_id128_is_null(fsuuid))
280 return -EINVAL;
281 if (sd_id128_is_null(vuuid))
282 return -EINVAL;
283
284 /* If the verity data declares it's for the /usr partition, then search for that, in all
285 * other cases assume it's for the root partition. */
286 if (verity->designator == PARTITION_USR) {
287 usr_uuid = fsuuid;
288 usr_verity_uuid = vuuid;
289 } else {
290 root_uuid = fsuuid;
291 root_verity_uuid = vuuid;
292 }
293 }
294
295 if (fstat(fd, &st) < 0)
296 return -errno;
297
298 if (!S_ISBLK(st.st_mode))
299 return -ENOTBLK;
300
301 r = sd_device_new_from_stat_rdev(&d, &st);
302 if (r < 0)
303 return r;
304
305 b = blkid_new_probe();
306 if (!b)
307 return -ENOMEM;
308
309 errno = 0;
310 r = blkid_probe_set_device(b, fd, 0, 0);
311 if (r != 0)
312 return errno_or_else(ENOMEM);
313
314 if ((flags & DISSECT_IMAGE_GPT_ONLY) == 0) {
315 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
316 blkid_probe_enable_superblocks(b, 1);
317 blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE|BLKID_SUBLKS_USAGE);
318 }
319
320 blkid_probe_enable_partitions(b, 1);
321 blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS);
322
323 errno = 0;
324 r = blkid_do_safeprobe(b);
325 if (IN_SET(r, -2, 1))
326 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG), "Failed to identify any partition table.");
327 if (r != 0)
328 return errno_or_else(EIO);
329
330 m = new(DissectedImage, 1);
331 if (!m)
332 return -ENOMEM;
333
334 *m = (DissectedImage) {
335 .has_init_system = -1,
336 };
337
338 r = sd_device_get_sysname(d, &sysname);
339 if (r < 0)
340 return log_debug_errno(r, "Failed to get device sysname: %m");
341 if (startswith(sysname, "loop")) {
342 _cleanup_free_ char *name_stripped = NULL;
343 const char *full_path;
344
345 r = sd_device_get_sysattr_value(d, "loop/backing_file", &full_path);
346 if (r < 0)
347 log_debug_errno(r, "Failed to lookup image name via loop device backing file sysattr, ignoring: %m");
348 else {
349 r = raw_strip_suffixes(basename(full_path), &name_stripped);
350 if (r < 0)
351 return r;
352 }
353
354 free_and_replace(m->image_name, name_stripped);
355 } else {
356 r = free_and_strdup(&m->image_name, sysname);
357 if (r < 0)
358 return r;
359 }
360 r = sd_device_get_devname(d, &devname);
361 if (r < 0)
362 return log_debug_errno(r, "Failed to get device devname: %m");
363
364 if (!image_name_is_valid(m->image_name)) {
365 log_debug("Image name %s is not valid, ignoring", strempty(m->image_name));
366 m->image_name = mfree(m->image_name);
367 }
368
369 if ((!(flags & DISSECT_IMAGE_GPT_ONLY) &&
370 (flags & DISSECT_IMAGE_GENERIC_ROOT)) ||
371 (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)) {
372 const char *usage = NULL;
373
374 /* If flags permit this, also allow using non-partitioned single-filesystem images */
375
376 (void) blkid_probe_lookup_value(b, "USAGE", &usage, NULL);
377 if (STRPTR_IN_SET(usage, "filesystem", "crypto")) {
378 _cleanup_free_ char *t = NULL, *n = NULL, *o = NULL;
379 const char *fstype = NULL, *options = NULL;
380
381 /* OK, we have found a file system, that's our root partition then. */
382 (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
383
384 if (fstype) {
385 t = strdup(fstype);
386 if (!t)
387 return -ENOMEM;
388 }
389
390 n = strdup(devname);
391 if (!n)
392 return -ENOMEM;
393
394 m->single_file_system = true;
395 m->encrypted = streq_ptr(fstype, "crypto_LUKS");
396
397 m->has_verity = verity && verity->data_path;
398 m->verity_ready = m->has_verity &&
399 verity->root_hash &&
400 (verity->designator < 0 || verity->designator == PARTITION_ROOT);
401
402 m->has_verity_sig = false; /* signature not embedded, must be specified */
403 m->verity_sig_ready = m->verity_ready &&
404 verity->root_hash_sig;
405
406 options = mount_options_from_designator(mount_options, PARTITION_ROOT);
407 if (options) {
408 o = strdup(options);
409 if (!o)
410 return -ENOMEM;
411 }
412
413 m->partitions[PARTITION_ROOT] = (DissectedPartition) {
414 .found = true,
415 .rw = !m->verity_ready && !fstype_is_ro(fstype),
416 .partno = -1,
417 .architecture = _ARCHITECTURE_INVALID,
418 .fstype = TAKE_PTR(t),
419 .node = TAKE_PTR(n),
420 .mount_options = TAKE_PTR(o),
421 .offset = 0,
422 .size = UINT64_MAX,
423 };
424
425 *ret = TAKE_PTR(m);
426 return 0;
427 }
428 }
429
430 (void) blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL);
431 if (!pttype)
432 return -ENOPKG;
433
434 is_gpt = streq_ptr(pttype, "gpt");
435 is_mbr = streq_ptr(pttype, "dos");
436
437 if (!is_gpt && ((flags & DISSECT_IMAGE_GPT_ONLY) || !is_mbr))
438 return -ENOPKG;
439
440 /* We support external verity data partitions only if the image has no partition table */
441 if (verity && verity->data_path)
442 return -EBADR;
443
444 /* Safety check: refuse block devices that carry a partition table but for which the kernel doesn't
445 * do partition scanning. */
446 r = blockdev_partscan_enabled(fd);
447 if (r < 0)
448 return r;
449 if (r == 0)
450 return -EPROTONOSUPPORT;
451
452 errno = 0;
453 pl = blkid_probe_get_partitions(b);
454 if (!pl)
455 return errno_or_else(ENOMEM);
456
457 errno = 0;
458 n_partitions = blkid_partlist_numof_partitions(pl);
459 if (n_partitions < 0)
460 return errno_or_else(EIO);
461
462 for (int i = 0; i < n_partitions; i++) {
463 _cleanup_free_ char *node = NULL;
464 unsigned long long pflags;
465 blkid_loff_t start, size;
466 blkid_partition pp;
467 int nr;
468
469 errno = 0;
470 pp = blkid_partlist_get_partition(pl, i);
471 if (!pp)
472 return errno_or_else(EIO);
473
474 pflags = blkid_partition_get_flags(pp);
475
476 errno = 0;
477 nr = blkid_partition_get_partno(pp);
478 if (nr < 0)
479 return errno_or_else(EIO);
480
481 errno = 0;
482 start = blkid_partition_get_start(pp);
483 if (start < 0)
484 return errno_or_else(EIO);
485
486 assert((uint64_t) start < UINT64_MAX/512);
487
488 errno = 0;
489 size = blkid_partition_get_size(pp);
490 if (size < 0)
491 return errno_or_else(EIO);
492
493 assert((uint64_t) size < UINT64_MAX/512);
494
495 r = make_partition_devname(devname, nr, &node);
496 if (r < 0)
497 return r;
498
499 /* So here's the thing: after the main ("whole") block device popped up it might take a while
500 * before the kernel fully probed the partition table. Waiting for that to finish is icky in
501 * userspace. So here's what we do instead. We issue the BLKPG_ADD_PARTITION ioctl to add the
502 * partition ourselves, racing against the kernel. Good thing is: if this call fails with
503 * EBUSY then the kernel was quicker than us, and that's totally OK, the outcome is good for
504 * us: the device node will exist. If OTOH our call was successful we won the race. Which is
505 * also good as the outcome is the same: the partition block device exists, and we can use
506 * it.
507 *
508 * Kernel returns EBUSY if there's already a partition by that number or an overlapping
509 * partition already existent. */
510
511 r = ioctl_partition_add(fd, node, nr, (uint64_t) start * 512, (uint64_t) size * 512);
512 if (r < 0) {
513 if (r != -EBUSY)
514 return log_debug_errno(r, "BLKPG_ADD_PARTITION failed: %m");
515
516 log_debug_errno(r, "Kernel was quicker than us in adding partition %i.", nr);
517 } else
518 log_debug("We were quicker than kernel in adding partition %i.", nr);
519
520 if (is_gpt) {
521 PartitionDesignator designator = _PARTITION_DESIGNATOR_INVALID;
522 Architecture architecture = _ARCHITECTURE_INVALID;
523 const char *stype, *sid, *fstype = NULL, *label;
524 sd_id128_t type_id, id;
525 bool rw = true, growfs = false;
526
527 sid = blkid_partition_get_uuid(pp);
528 if (!sid)
529 continue;
530 if (sd_id128_from_string(sid, &id) < 0)
531 continue;
532
533 stype = blkid_partition_get_type_string(pp);
534 if (!stype)
535 continue;
536 if (sd_id128_from_string(stype, &type_id) < 0)
537 continue;
538
539 label = blkid_partition_get_name(pp); /* libblkid returns NULL here if empty */
540
541 if (sd_id128_equal(type_id, GPT_HOME)) {
542
543 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY|GPT_FLAG_GROWFS);
544
545 if (pflags & GPT_FLAG_NO_AUTO)
546 continue;
547
548 designator = PARTITION_HOME;
549 rw = !(pflags & GPT_FLAG_READ_ONLY);
550 growfs = FLAGS_SET(pflags, GPT_FLAG_GROWFS);
551
552 } else if (sd_id128_equal(type_id, GPT_SRV)) {
553
554 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY|GPT_FLAG_GROWFS);
555
556 if (pflags & GPT_FLAG_NO_AUTO)
557 continue;
558
559 designator = PARTITION_SRV;
560 rw = !(pflags & GPT_FLAG_READ_ONLY);
561 growfs = FLAGS_SET(pflags, GPT_FLAG_GROWFS);
562
563 } else if (sd_id128_equal(type_id, GPT_ESP)) {
564
565 /* Note that we don't check the GPT_FLAG_NO_AUTO flag for the ESP, as it is
566 * not defined there. We instead check the GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as
567 * recommended by the UEFI spec (See "12.3.3 Number and Location of System
568 * Partitions"). */
569
570 if (pflags & GPT_FLAG_NO_BLOCK_IO_PROTOCOL)
571 continue;
572
573 designator = PARTITION_ESP;
574 fstype = "vfat";
575
576 } else if (sd_id128_equal(type_id, GPT_XBOOTLDR)) {
577
578 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY|GPT_FLAG_GROWFS);
579
580 if (pflags & GPT_FLAG_NO_AUTO)
581 continue;
582
583 designator = PARTITION_XBOOTLDR;
584 rw = !(pflags & GPT_FLAG_READ_ONLY);
585 growfs = FLAGS_SET(pflags, GPT_FLAG_GROWFS);
586
587 } else if (gpt_partition_type_is_root(type_id)) {
588
589 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY|GPT_FLAG_GROWFS);
590
591 if (pflags & GPT_FLAG_NO_AUTO)
592 continue;
593
594 /* If a root ID is specified, ignore everything but the root id */
595 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
596 continue;
597
598 assert_se((architecture = gpt_partition_type_uuid_to_arch(type_id)) >= 0);
599 designator = PARTITION_ROOT_OF_ARCH(architecture);
600 rw = !(pflags & GPT_FLAG_READ_ONLY);
601 growfs = FLAGS_SET(pflags, GPT_FLAG_GROWFS);
602
603 } else if (gpt_partition_type_is_root_verity(type_id)) {
604
605 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
606
607 if (pflags & GPT_FLAG_NO_AUTO)
608 continue;
609
610 m->has_verity = true;
611
612 /* If no verity configuration is specified, then don't do verity */
613 if (!verity)
614 continue;
615 if (verity->designator >= 0 && verity->designator != PARTITION_ROOT)
616 continue;
617
618 /* If root hash is specified, then ignore everything but the root id */
619 if (!sd_id128_is_null(root_verity_uuid) && !sd_id128_equal(root_verity_uuid, id))
620 continue;
621
622 assert_se((architecture = gpt_partition_type_uuid_to_arch(type_id)) >= 0);
623 designator = PARTITION_VERITY_OF(PARTITION_ROOT_OF_ARCH(architecture));
624 fstype = "DM_verity_hash";
625 rw = false;
626
627 } else if (gpt_partition_type_is_root_verity_sig(type_id)) {
628
629 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
630
631 if (pflags & GPT_FLAG_NO_AUTO)
632 continue;
633
634 m->has_verity_sig = true;
635
636 /* If root hash is specified explicitly, then ignore any embedded signature */
637 if (!verity)
638 continue;
639 if (verity->designator >= 0 && verity->designator != PARTITION_ROOT)
640 continue;
641 if (verity->root_hash)
642 continue;
643
644 assert_se((architecture = gpt_partition_type_uuid_to_arch(type_id)) >= 0);
645 designator = PARTITION_VERITY_SIG_OF(PARTITION_ROOT_OF_ARCH(architecture));
646 fstype = "verity_hash_signature";
647 rw = false;
648
649 } else if (gpt_partition_type_is_usr(type_id)) {
650
651 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY|GPT_FLAG_GROWFS);
652
653 if (pflags & GPT_FLAG_NO_AUTO)
654 continue;
655
656 /* If a usr ID is specified, ignore everything but the usr id */
657 if (!sd_id128_is_null(usr_uuid) && !sd_id128_equal(usr_uuid, id))
658 continue;
659
660 assert_se((architecture = gpt_partition_type_uuid_to_arch(type_id)) >= 0);
661 designator = PARTITION_USR_OF_ARCH(architecture);
662 rw = !(pflags & GPT_FLAG_READ_ONLY);
663 growfs = FLAGS_SET(pflags, GPT_FLAG_GROWFS);
664
665 } else if (gpt_partition_type_is_usr_verity(type_id)) {
666
667 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
668
669 if (pflags & GPT_FLAG_NO_AUTO)
670 continue;
671
672 m->has_verity = true;
673
674 if (!verity)
675 continue;
676 if (verity->designator >= 0 && verity->designator != PARTITION_USR)
677 continue;
678
679 /* If usr hash is specified, then ignore everything but the usr id */
680 if (!sd_id128_is_null(usr_verity_uuid) && !sd_id128_equal(usr_verity_uuid, id))
681 continue;
682
683 assert_se((architecture = gpt_partition_type_uuid_to_arch(type_id)) >= 0);
684 designator = PARTITION_VERITY_OF(PARTITION_USR_OF_ARCH(architecture));
685 fstype = "DM_verity_hash";
686 rw = false;
687
688 } else if (gpt_partition_type_is_usr_verity_sig(type_id)) {
689
690 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
691
692 if (pflags & GPT_FLAG_NO_AUTO)
693 continue;
694
695 m->has_verity_sig = true;
696
697 /* If usr hash is specified explicitly, then ignore any embedded signature */
698 if (!verity)
699 continue;
700 if (verity->designator >= 0 && verity->designator != PARTITION_USR)
701 continue;
702 if (verity->root_hash)
703 continue;
704
705 assert_se((architecture = gpt_partition_type_uuid_to_arch(type_id)) >= 0);
706 designator = PARTITION_VERITY_SIG_OF(PARTITION_USR_OF_ARCH(architecture));
707 fstype = "verity_hash_signature";
708 rw = false;
709
710 } else if (sd_id128_equal(type_id, GPT_SWAP)) {
711
712 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO);
713
714 if (pflags & GPT_FLAG_NO_AUTO)
715 continue;
716
717 designator = PARTITION_SWAP;
718
719 } else if (sd_id128_equal(type_id, GPT_LINUX_GENERIC)) {
720
721 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY|GPT_FLAG_GROWFS);
722
723 if (pflags & GPT_FLAG_NO_AUTO)
724 continue;
725
726 if (generic_node)
727 multiple_generic = true;
728 else {
729 generic_nr = nr;
730 generic_rw = !(pflags & GPT_FLAG_READ_ONLY);
731 generic_growfs = FLAGS_SET(pflags, GPT_FLAG_GROWFS);
732 generic_uuid = id;
733 generic_node = strdup(node);
734 if (!generic_node)
735 return -ENOMEM;
736 }
737
738 } else if (sd_id128_equal(type_id, GPT_TMP)) {
739
740 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY|GPT_FLAG_GROWFS);
741
742 if (pflags & GPT_FLAG_NO_AUTO)
743 continue;
744
745 designator = PARTITION_TMP;
746 rw = !(pflags & GPT_FLAG_READ_ONLY);
747 growfs = FLAGS_SET(pflags, GPT_FLAG_GROWFS);
748
749 } else if (sd_id128_equal(type_id, GPT_VAR)) {
750
751 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY|GPT_FLAG_GROWFS);
752
753 if (pflags & GPT_FLAG_NO_AUTO)
754 continue;
755
756 if (!FLAGS_SET(flags, DISSECT_IMAGE_RELAX_VAR_CHECK)) {
757 sd_id128_t var_uuid;
758
759 /* For /var we insist that the uuid of the partition matches the
760 * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
761 * ID. Why? Unlike the other partitions /var is inherently
762 * installation specific, hence we need to be careful not to mount it
763 * in the wrong installation. By hashing the partition UUID from
764 * /etc/machine-id we can securely bind the partition to the
765 * installation. */
766
767 r = sd_id128_get_machine_app_specific(GPT_VAR, &var_uuid);
768 if (r < 0)
769 return r;
770
771 if (!sd_id128_equal(var_uuid, id)) {
772 log_debug("Found a /var/ partition, but its UUID didn't match our expectations, ignoring.");
773 continue;
774 }
775 }
776
777 designator = PARTITION_VAR;
778 rw = !(pflags & GPT_FLAG_READ_ONLY);
779 growfs = FLAGS_SET(pflags, GPT_FLAG_GROWFS);
780 }
781
782 if (designator != _PARTITION_DESIGNATOR_INVALID) {
783 _cleanup_free_ char *t = NULL, *n = NULL, *o = NULL, *l = NULL;
784 const char *options = NULL;
785
786 if (m->partitions[designator].found) {
787 /* For most partition types the first one we see wins. Except for the
788 * rootfs and /usr, where we do a version compare of the label, and
789 * let the newest version win. This permits a simple A/B versioning
790 * scheme in OS images. */
791
792 if (!PARTITION_DESIGNATOR_VERSIONED(designator) ||
793 strverscmp_improved(m->partitions[designator].label, label) >= 0)
794 continue;
795
796 dissected_partition_done(m->partitions + designator);
797 }
798
799 if (fstype) {
800 t = strdup(fstype);
801 if (!t)
802 return -ENOMEM;
803 }
804
805 n = strdup(node);
806 if (!n)
807 return -ENOMEM;
808
809 if (label) {
810 l = strdup(label);
811 if (!l)
812 return -ENOMEM;
813 }
814
815 options = mount_options_from_designator(mount_options, designator);
816 if (options) {
817 o = strdup(options);
818 if (!o)
819 return -ENOMEM;
820 }
821
822 m->partitions[designator] = (DissectedPartition) {
823 .found = true,
824 .partno = nr,
825 .rw = rw,
826 .growfs = growfs,
827 .architecture = architecture,
828 .node = TAKE_PTR(n),
829 .fstype = TAKE_PTR(t),
830 .label = TAKE_PTR(l),
831 .uuid = id,
832 .mount_options = TAKE_PTR(o),
833 .offset = (uint64_t) start * 512,
834 .size = (uint64_t) size * 512,
835 };
836 }
837
838 } else if (is_mbr) {
839
840 switch (blkid_partition_get_type(pp)) {
841
842 case 0x83: /* Linux partition */
843
844 if (pflags != 0x80) /* Bootable flag */
845 continue;
846
847 if (generic_node)
848 multiple_generic = true;
849 else {
850 generic_nr = nr;
851 generic_rw = true;
852 generic_growfs = false;
853 generic_node = strdup(node);
854 if (!generic_node)
855 return -ENOMEM;
856 }
857
858 break;
859
860 case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
861 _cleanup_free_ char *n = NULL, *o = NULL;
862 sd_id128_t id = SD_ID128_NULL;
863 const char *sid, *options = NULL;
864
865 /* First one wins */
866 if (m->partitions[PARTITION_XBOOTLDR].found)
867 continue;
868
869 sid = blkid_partition_get_uuid(pp);
870 if (sid)
871 (void) sd_id128_from_string(sid, &id);
872
873 n = strdup(node);
874 if (!n)
875 return -ENOMEM;
876
877 options = mount_options_from_designator(mount_options, PARTITION_XBOOTLDR);
878 if (options) {
879 o = strdup(options);
880 if (!o)
881 return -ENOMEM;
882 }
883
884 m->partitions[PARTITION_XBOOTLDR] = (DissectedPartition) {
885 .found = true,
886 .partno = nr,
887 .rw = true,
888 .growfs = false,
889 .architecture = _ARCHITECTURE_INVALID,
890 .node = TAKE_PTR(n),
891 .uuid = id,
892 .mount_options = TAKE_PTR(o),
893 .offset = (uint64_t) start * 512,
894 .size = (uint64_t) size * 512,
895 };
896
897 break;
898 }}
899 }
900 }
901
902 if (m->partitions[PARTITION_ROOT].found) {
903 /* If we found the primary arch, then invalidate the secondary and other arch to avoid any
904 * ambiguities, since we never want to mount the secondary or other arch in this case. */
905 m->partitions[PARTITION_ROOT_SECONDARY].found = false;
906 m->partitions[PARTITION_ROOT_SECONDARY_VERITY].found = false;
907 m->partitions[PARTITION_ROOT_SECONDARY_VERITY_SIG].found = false;
908 m->partitions[PARTITION_USR_SECONDARY].found = false;
909 m->partitions[PARTITION_USR_SECONDARY_VERITY].found = false;
910 m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG].found = false;
911
912 m->partitions[PARTITION_ROOT_OTHER].found = false;
913 m->partitions[PARTITION_ROOT_OTHER_VERITY].found = false;
914 m->partitions[PARTITION_ROOT_OTHER_VERITY_SIG].found = false;
915 m->partitions[PARTITION_USR_OTHER].found = false;
916 m->partitions[PARTITION_USR_OTHER_VERITY].found = false;
917 m->partitions[PARTITION_USR_OTHER_VERITY_SIG].found = false;
918
919 } else if (m->partitions[PARTITION_ROOT_VERITY].found ||
920 m->partitions[PARTITION_ROOT_VERITY_SIG].found)
921 return -EADDRNOTAVAIL; /* Verity found but no matching rootfs? Something is off, refuse. */
922
923 else if (m->partitions[PARTITION_ROOT_SECONDARY].found) {
924
925 /* No root partition found but there's one for the secondary architecture? Then upgrade
926 * secondary arch to first and invalidate the other arch. */
927
928 log_debug("No root partition found of the native architecture, falling back to a root "
929 "partition of the secondary architecture.");
930
931 m->partitions[PARTITION_ROOT] = m->partitions[PARTITION_ROOT_SECONDARY];
932 zero(m->partitions[PARTITION_ROOT_SECONDARY]);
933 m->partitions[PARTITION_ROOT_VERITY] = m->partitions[PARTITION_ROOT_SECONDARY_VERITY];
934 zero(m->partitions[PARTITION_ROOT_SECONDARY_VERITY]);
935 m->partitions[PARTITION_ROOT_VERITY_SIG] = m->partitions[PARTITION_ROOT_SECONDARY_VERITY_SIG];
936 zero(m->partitions[PARTITION_ROOT_SECONDARY_VERITY_SIG]);
937
938 m->partitions[PARTITION_USR] = m->partitions[PARTITION_USR_SECONDARY];
939 zero(m->partitions[PARTITION_USR_SECONDARY]);
940 m->partitions[PARTITION_USR_VERITY] = m->partitions[PARTITION_USR_SECONDARY_VERITY];
941 zero(m->partitions[PARTITION_USR_SECONDARY_VERITY]);
942 m->partitions[PARTITION_USR_VERITY_SIG] = m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG];
943 zero(m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG]);
944
945 m->partitions[PARTITION_ROOT_OTHER].found = false;
946 m->partitions[PARTITION_ROOT_OTHER_VERITY].found = false;
947 m->partitions[PARTITION_ROOT_OTHER_VERITY_SIG].found = false;
948 m->partitions[PARTITION_USR_OTHER].found = false;
949 m->partitions[PARTITION_USR_OTHER_VERITY].found = false;
950 m->partitions[PARTITION_USR_OTHER_VERITY_SIG].found = false;
951
952 } else if (m->partitions[PARTITION_ROOT_SECONDARY_VERITY].found ||
953 m->partitions[PARTITION_ROOT_SECONDARY_VERITY_SIG].found)
954 return -EADDRNOTAVAIL; /* as above */
955
956 else if (m->partitions[PARTITION_ROOT_OTHER].found) {
957
958 /* No root or secondary partition found but there's one for another architecture? Then
959 * upgrade the other architecture to first. */
960
961 log_debug("No root partition found of the native architecture or the secondary architecture, "
962 "falling back to a root partition of a non-native architecture (%s).",
963 architecture_to_string(m->partitions[PARTITION_ROOT_OTHER].architecture));
964
965 m->partitions[PARTITION_ROOT] = m->partitions[PARTITION_ROOT_OTHER];
966 zero(m->partitions[PARTITION_ROOT_OTHER]);
967 m->partitions[PARTITION_ROOT_VERITY] = m->partitions[PARTITION_ROOT_OTHER_VERITY];
968 zero(m->partitions[PARTITION_ROOT_OTHER_VERITY]);
969 m->partitions[PARTITION_ROOT_VERITY_SIG] = m->partitions[PARTITION_ROOT_OTHER_VERITY_SIG];
970 zero(m->partitions[PARTITION_ROOT_OTHER_VERITY_SIG]);
971
972 m->partitions[PARTITION_USR] = m->partitions[PARTITION_USR_OTHER];
973 zero(m->partitions[PARTITION_USR_OTHER]);
974 m->partitions[PARTITION_USR_VERITY] = m->partitions[PARTITION_USR_OTHER_VERITY];
975 zero(m->partitions[PARTITION_USR_OTHER_VERITY]);
976 m->partitions[PARTITION_USR_VERITY_SIG] = m->partitions[PARTITION_USR_OTHER_VERITY_SIG];
977 zero(m->partitions[PARTITION_USR_OTHER_VERITY_SIG]);
978 }
979
980 /* Hmm, we found a signature partition but no Verity data? Something is off. */
981 if (m->partitions[PARTITION_ROOT_VERITY_SIG].found && !m->partitions[PARTITION_ROOT_VERITY].found)
982 return -EADDRNOTAVAIL;
983
984 if (m->partitions[PARTITION_USR].found) {
985 /* Invalidate secondary and other arch /usr/ if we found the primary arch */
986 m->partitions[PARTITION_USR_SECONDARY].found = false;
987 m->partitions[PARTITION_USR_SECONDARY_VERITY].found = false;
988 m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG].found = false;
989
990 m->partitions[PARTITION_USR_OTHER].found = false;
991 m->partitions[PARTITION_USR_OTHER_VERITY].found = false;
992 m->partitions[PARTITION_USR_OTHER_VERITY_SIG].found = false;
993
994 } else if (m->partitions[PARTITION_USR_VERITY].found ||
995 m->partitions[PARTITION_USR_VERITY_SIG].found)
996 return -EADDRNOTAVAIL; /* as above */
997
998 else if (m->partitions[PARTITION_USR_SECONDARY].found) {
999
1000 log_debug("No usr partition found of the native architecture, falling back to a usr "
1001 "partition of the secondary architecture.");
1002
1003 /* Upgrade secondary arch to primary */
1004 m->partitions[PARTITION_USR] = m->partitions[PARTITION_USR_SECONDARY];
1005 zero(m->partitions[PARTITION_USR_SECONDARY]);
1006 m->partitions[PARTITION_USR_VERITY] = m->partitions[PARTITION_USR_SECONDARY_VERITY];
1007 zero(m->partitions[PARTITION_USR_SECONDARY_VERITY]);
1008 m->partitions[PARTITION_USR_VERITY_SIG] = m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG];
1009 zero(m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG]);
1010
1011 m->partitions[PARTITION_USR_OTHER].found = false;
1012 m->partitions[PARTITION_USR_OTHER_VERITY].found = false;
1013 m->partitions[PARTITION_USR_OTHER_VERITY_SIG].found = false;
1014
1015 } else if (m->partitions[PARTITION_USR_SECONDARY_VERITY].found ||
1016 m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG].found)
1017 return -EADDRNOTAVAIL; /* as above */
1018
1019 else if (m->partitions[PARTITION_USR_OTHER].found) {
1020
1021 log_debug("No usr partition found of the native architecture or the secondary architecture, "
1022 "falling back to a usr partition of a non-native architecture (%s).",
1023 architecture_to_string(m->partitions[PARTITION_ROOT_OTHER].architecture));
1024
1025 /* Upgrade other arch to primary */
1026 m->partitions[PARTITION_USR] = m->partitions[PARTITION_USR_OTHER];
1027 zero(m->partitions[PARTITION_USR_OTHER]);
1028 m->partitions[PARTITION_USR_VERITY] = m->partitions[PARTITION_USR_OTHER_VERITY];
1029 zero(m->partitions[PARTITION_USR_OTHER_VERITY]);
1030 m->partitions[PARTITION_USR_VERITY_SIG] = m->partitions[PARTITION_USR_OTHER_VERITY_SIG];
1031 zero(m->partitions[PARTITION_USR_OTHER_VERITY_SIG]);
1032 }
1033
1034 /* Hmm, we found a signature partition but no Verity data? Something is off. */
1035 if (m->partitions[PARTITION_USR_VERITY_SIG].found && !m->partitions[PARTITION_USR_VERITY].found)
1036 return -EADDRNOTAVAIL;
1037
1038 /* If root and /usr are combined then insist that the architecture matches */
1039 if (m->partitions[PARTITION_ROOT].found &&
1040 m->partitions[PARTITION_USR].found &&
1041 (m->partitions[PARTITION_ROOT].architecture >= 0 &&
1042 m->partitions[PARTITION_USR].architecture >= 0 &&
1043 m->partitions[PARTITION_ROOT].architecture != m->partitions[PARTITION_USR].architecture))
1044 return -EADDRNOTAVAIL;
1045
1046 if (!m->partitions[PARTITION_ROOT].found &&
1047 !m->partitions[PARTITION_USR].found &&
1048 (flags & DISSECT_IMAGE_GENERIC_ROOT) &&
1049 (!verity || !verity->root_hash || verity->designator != PARTITION_USR)) {
1050
1051 /* OK, we found nothing usable, then check if there's a single generic partition, and use
1052 * that. If the root hash was set however, then we won't fall back to a generic node, because
1053 * the root hash decides. */
1054
1055 /* If we didn't find a properly marked root partition, but we did find a single suitable
1056 * generic Linux partition, then use this as root partition, if the caller asked for it. */
1057 if (multiple_generic)
1058 return -ENOTUNIQ;
1059
1060 /* If we didn't find a generic node, then we can't fix this up either */
1061 if (generic_node) {
1062 _cleanup_free_ char *o = NULL;
1063 const char *options;
1064
1065 options = mount_options_from_designator(mount_options, PARTITION_ROOT);
1066 if (options) {
1067 o = strdup(options);
1068 if (!o)
1069 return -ENOMEM;
1070 }
1071
1072 assert(generic_nr >= 0);
1073 m->partitions[PARTITION_ROOT] = (DissectedPartition) {
1074 .found = true,
1075 .rw = generic_rw,
1076 .growfs = generic_growfs,
1077 .partno = generic_nr,
1078 .architecture = _ARCHITECTURE_INVALID,
1079 .node = TAKE_PTR(generic_node),
1080 .uuid = generic_uuid,
1081 .mount_options = TAKE_PTR(o),
1082 .offset = UINT64_MAX,
1083 .size = UINT64_MAX,
1084 };
1085 }
1086 }
1087
1088 /* Check if we have a root fs if we are told to do check. /usr alone is fine too, but only if appropriate flag for that is set too */
1089 if (FLAGS_SET(flags, DISSECT_IMAGE_REQUIRE_ROOT) &&
1090 !(m->partitions[PARTITION_ROOT].found || (m->partitions[PARTITION_USR].found && FLAGS_SET(flags, DISSECT_IMAGE_USR_NO_ROOT))))
1091 return -ENXIO;
1092
1093 if (m->partitions[PARTITION_ROOT_VERITY].found) {
1094 /* We only support one verity partition per image, i.e. can't do for both /usr and root fs */
1095 if (m->partitions[PARTITION_USR_VERITY].found)
1096 return -ENOTUNIQ;
1097
1098 /* We don't support verity enabled root with a split out /usr. Neither with nor without
1099 * verity there. (Note that we do support verity-less root with verity-full /usr, though.) */
1100 if (m->partitions[PARTITION_USR].found)
1101 return -EADDRNOTAVAIL;
1102 }
1103
1104 if (verity) {
1105 /* If a verity designator is specified, then insist that the matching partition exists */
1106 if (verity->designator >= 0 && !m->partitions[verity->designator].found)
1107 return -EADDRNOTAVAIL;
1108
1109 if (verity->root_hash) {
1110 /* If we have an explicit root hash and found the partitions for it, then we are ready to use
1111 * Verity, set things up for it */
1112
1113 if (verity->designator < 0 || verity->designator == PARTITION_ROOT) {
1114 if (!m->partitions[PARTITION_ROOT_VERITY].found || !m->partitions[PARTITION_ROOT].found)
1115 return -EADDRNOTAVAIL;
1116
1117 /* If we found a verity setup, then the root partition is necessarily read-only. */
1118 m->partitions[PARTITION_ROOT].rw = false;
1119 m->verity_ready = true;
1120
1121 } else {
1122 assert(verity->designator == PARTITION_USR);
1123
1124 if (!m->partitions[PARTITION_USR_VERITY].found || !m->partitions[PARTITION_USR].found)
1125 return -EADDRNOTAVAIL;
1126
1127 m->partitions[PARTITION_USR].rw = false;
1128 m->verity_ready = true;
1129 }
1130
1131 if (m->verity_ready)
1132 m->verity_sig_ready = verity->root_hash_sig;
1133
1134 } else if (m->partitions[verity->designator == PARTITION_USR ? PARTITION_USR_VERITY_SIG : PARTITION_ROOT_VERITY_SIG].found) {
1135
1136 /* If we found an embedded signature partition, we are ready, too. */
1137
1138 m->verity_ready = m->verity_sig_ready = true;
1139 m->partitions[verity->designator == PARTITION_USR ? PARTITION_USR : PARTITION_ROOT].rw = false;
1140 }
1141 }
1142
1143 blkid_free_probe(b);
1144 b = NULL;
1145
1146 /* Fill in file system types if we don't know them yet. */
1147 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
1148 DissectedPartition *p = m->partitions + i;
1149
1150 if (!p->found)
1151 continue;
1152
1153 if (!p->fstype && p->node) {
1154 r = probe_filesystem(p->node, &p->fstype);
1155 if (r < 0 && r != -EUCLEAN)
1156 return r;
1157 }
1158
1159 if (streq_ptr(p->fstype, "crypto_LUKS"))
1160 m->encrypted = true;
1161
1162 if (p->fstype && fstype_is_ro(p->fstype))
1163 p->rw = false;
1164
1165 if (!p->rw)
1166 p->growfs = false;
1167 }
1168
1169 *ret = TAKE_PTR(m);
1170 return 0;
1171 #else
1172 return -EOPNOTSUPP;
1173 #endif
1174 }
1175
dissected_image_unref(DissectedImage * m)1176 DissectedImage* dissected_image_unref(DissectedImage *m) {
1177 if (!m)
1178 return NULL;
1179
1180 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++)
1181 dissected_partition_done(m->partitions + i);
1182
1183 free(m->image_name);
1184 free(m->hostname);
1185 strv_free(m->machine_info);
1186 strv_free(m->os_release);
1187 strv_free(m->extension_release);
1188
1189 return mfree(m);
1190 }
1191
is_loop_device(const char * path)1192 static int is_loop_device(const char *path) {
1193 char s[SYS_BLOCK_PATH_MAX("/../loop/")];
1194 struct stat st;
1195
1196 assert(path);
1197
1198 if (stat(path, &st) < 0)
1199 return -errno;
1200
1201 if (!S_ISBLK(st.st_mode))
1202 return -ENOTBLK;
1203
1204 xsprintf_sys_block_path(s, "/loop/", st.st_dev);
1205 if (access(s, F_OK) < 0) {
1206 if (errno != ENOENT)
1207 return -errno;
1208
1209 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
1210 xsprintf_sys_block_path(s, "/../loop/", st.st_dev);
1211 if (access(s, F_OK) < 0)
1212 return errno == ENOENT ? false : -errno;
1213 }
1214
1215 return true;
1216 }
1217
run_fsck(const char * node,const char * fstype)1218 static int run_fsck(const char *node, const char *fstype) {
1219 int r, exit_status;
1220 pid_t pid;
1221
1222 assert(node);
1223 assert(fstype);
1224
1225 r = fsck_exists(fstype);
1226 if (r < 0) {
1227 log_debug_errno(r, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype);
1228 return 0;
1229 }
1230 if (r == 0) {
1231 log_debug("Not checking partition %s, as fsck for %s does not exist.", node, fstype);
1232 return 0;
1233 }
1234
1235 r = safe_fork("(fsck)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG|FORK_NULL_STDIO, &pid);
1236 if (r < 0)
1237 return log_debug_errno(r, "Failed to fork off fsck: %m");
1238 if (r == 0) {
1239 /* Child */
1240 execl("/sbin/fsck", "/sbin/fsck", "-aT", node, NULL);
1241 log_open();
1242 log_debug_errno(errno, "Failed to execl() fsck: %m");
1243 _exit(FSCK_OPERATIONAL_ERROR);
1244 }
1245
1246 exit_status = wait_for_terminate_and_check("fsck", pid, 0);
1247 if (exit_status < 0)
1248 return log_debug_errno(exit_status, "Failed to fork off /sbin/fsck: %m");
1249
1250 if ((exit_status & ~FSCK_ERROR_CORRECTED) != FSCK_SUCCESS) {
1251 log_debug("fsck failed with exit status %i.", exit_status);
1252
1253 if ((exit_status & (FSCK_SYSTEM_SHOULD_REBOOT|FSCK_ERRORS_LEFT_UNCORRECTED)) != 0)
1254 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN), "File system is corrupted, refusing.");
1255
1256 log_debug("Ignoring fsck error.");
1257 }
1258
1259 return 0;
1260 }
1261
fs_grow(const char * node_path,const char * mount_path)1262 static int fs_grow(const char *node_path, const char *mount_path) {
1263 _cleanup_close_ int mount_fd = -1, node_fd = -1;
1264 uint64_t size, newsize;
1265 int r;
1266
1267 node_fd = open(node_path, O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY);
1268 if (node_fd < 0)
1269 return log_debug_errno(errno, "Failed to open node device %s: %m", node_path);
1270
1271 if (ioctl(node_fd, BLKGETSIZE64, &size) != 0)
1272 return log_debug_errno(errno, "Failed to get block device size of %s: %m", node_path);
1273
1274 mount_fd = open(mount_path, O_RDONLY|O_DIRECTORY|O_CLOEXEC);
1275 if (mount_fd < 0)
1276 return log_debug_errno(errno, "Failed to open mountd file system %s: %m", mount_path);
1277
1278 log_debug("Resizing \"%s\" to %"PRIu64" bytes...", mount_path, size);
1279 r = resize_fs(mount_fd, size, &newsize);
1280 if (r < 0)
1281 return log_debug_errno(r, "Failed to resize \"%s\" to %"PRIu64" bytes: %m", mount_path, size);
1282
1283 if (newsize == size)
1284 log_debug("Successfully resized \"%s\" to %s bytes.",
1285 mount_path, FORMAT_BYTES(newsize));
1286 else {
1287 assert(newsize < size);
1288 log_debug("Successfully resized \"%s\" to %s bytes (%"PRIu64" bytes lost due to blocksize).",
1289 mount_path, FORMAT_BYTES(newsize), size - newsize);
1290 }
1291
1292 return 0;
1293 }
1294
mount_partition(DissectedPartition * m,const char * where,const char * directory,uid_t uid_shift,uid_t uid_range,DissectImageFlags flags)1295 static int mount_partition(
1296 DissectedPartition *m,
1297 const char *where,
1298 const char *directory,
1299 uid_t uid_shift,
1300 uid_t uid_range,
1301 DissectImageFlags flags) {
1302
1303 _cleanup_free_ char *chased = NULL, *options = NULL;
1304 const char *p, *node, *fstype;
1305 bool rw, remap_uid_gid = false;
1306 int r;
1307
1308 assert(m);
1309 assert(where);
1310
1311 /* Use decrypted node and matching fstype if available, otherwise use the original device */
1312 node = m->decrypted_node ?: m->node;
1313 fstype = m->decrypted_node ? m->decrypted_fstype: m->fstype;
1314
1315 if (!m->found || !node)
1316 return 0;
1317 if (!fstype)
1318 return -EAFNOSUPPORT;
1319
1320 /* We are looking at an encrypted partition? This either means stacked encryption, or the caller
1321 * didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error for this
1322 * case. */
1323 if (streq(fstype, "crypto_LUKS"))
1324 return -EUNATCH;
1325
1326 rw = m->rw && !(flags & DISSECT_IMAGE_MOUNT_READ_ONLY);
1327
1328 if (FLAGS_SET(flags, DISSECT_IMAGE_FSCK) && rw) {
1329 r = run_fsck(node, fstype);
1330 if (r < 0)
1331 return r;
1332 }
1333
1334 if (directory) {
1335 /* Automatically create missing mount points inside the image, if necessary. */
1336 r = mkdir_p_root(where, directory, uid_shift, (gid_t) uid_shift, 0755);
1337 if (r < 0 && r != -EROFS)
1338 return r;
1339
1340 r = chase_symlinks(directory, where, CHASE_PREFIX_ROOT, &chased, NULL);
1341 if (r < 0)
1342 return r;
1343
1344 p = chased;
1345 } else {
1346 /* Create top-level mount if missing – but only if this is asked for. This won't modify the
1347 * image (as the branch above does) but the host hierarchy, and the created directory might
1348 * survive our mount in the host hierarchy hence. */
1349 if (FLAGS_SET(flags, DISSECT_IMAGE_MKDIR)) {
1350 r = mkdir_p(where, 0755);
1351 if (r < 0)
1352 return r;
1353 }
1354
1355 p = where;
1356 }
1357
1358 /* If requested, turn on discard support. */
1359 if (fstype_can_discard(fstype) &&
1360 ((flags & DISSECT_IMAGE_DISCARD) ||
1361 ((flags & DISSECT_IMAGE_DISCARD_ON_LOOP) && is_loop_device(m->node) > 0))) {
1362 options = strdup("discard");
1363 if (!options)
1364 return -ENOMEM;
1365 }
1366
1367 if (uid_is_valid(uid_shift) && uid_shift != 0) {
1368
1369 if (fstype_can_uid_gid(fstype)) {
1370 _cleanup_free_ char *uid_option = NULL;
1371
1372 if (asprintf(&uid_option, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0)
1373 return -ENOMEM;
1374
1375 if (!strextend_with_separator(&options, ",", uid_option))
1376 return -ENOMEM;
1377 } else if (FLAGS_SET(flags, DISSECT_IMAGE_MOUNT_IDMAPPED))
1378 remap_uid_gid = true;
1379 }
1380
1381 if (!isempty(m->mount_options))
1382 if (!strextend_with_separator(&options, ",", m->mount_options))
1383 return -ENOMEM;
1384
1385 /* So, when you request MS_RDONLY from ext4, then this means nothing. It happily still writes to the
1386 * backing storage. What's worse, the BLKRO[GS]ET flag and (in case of loopback devices)
1387 * LO_FLAGS_READ_ONLY don't mean anything, they affect userspace accesses only, and write accesses
1388 * from the upper file system still get propagated through to the underlying file system,
1389 * unrestricted. To actually get ext4/xfs/btrfs to stop writing to the device we need to specify
1390 * "norecovery" as mount option, in addition to MS_RDONLY. Yes, this sucks, since it means we need to
1391 * carry a per file system table here.
1392 *
1393 * Note that this means that we might not be able to mount corrupted file systems as read-only
1394 * anymore (since in some cases the kernel implementations will refuse mounting when corrupted,
1395 * read-only and "norecovery" is specified). But I think for the case of automatically determined
1396 * mount options for loopback devices this is the right choice, since otherwise using the same
1397 * loopback file twice even in read-only mode, is going to fail badly sooner or later. The usecase of
1398 * making reuse of the immutable images "just work" is more relevant to us than having read-only
1399 * access that actually modifies stuff work on such image files. Or to say this differently: if
1400 * people want their file systems to be fixed up they should just open them in writable mode, where
1401 * all these problems don't exist. */
1402 if (!rw && STRPTR_IN_SET(fstype, "ext3", "ext4", "xfs", "btrfs"))
1403 if (!strextend_with_separator(&options, ",", "norecovery"))
1404 return -ENOMEM;
1405
1406 r = mount_nofollow_verbose(LOG_DEBUG, node, p, fstype, MS_NODEV|(rw ? 0 : MS_RDONLY), options);
1407 if (r < 0)
1408 return r;
1409
1410 if (rw && m->growfs && FLAGS_SET(flags, DISSECT_IMAGE_GROWFS))
1411 (void) fs_grow(node, p);
1412
1413 if (remap_uid_gid) {
1414 r = remount_idmap(p, uid_shift, uid_range, REMOUNT_IDMAP_HOST_ROOT);
1415 if (r < 0)
1416 return r;
1417 }
1418
1419 return 1;
1420 }
1421
mount_root_tmpfs(const char * where,uid_t uid_shift,DissectImageFlags flags)1422 static int mount_root_tmpfs(const char *where, uid_t uid_shift, DissectImageFlags flags) {
1423 _cleanup_free_ char *options = NULL;
1424 int r;
1425
1426 assert(where);
1427
1428 /* For images that contain /usr/ but no rootfs, let's mount rootfs as tmpfs */
1429
1430 if (FLAGS_SET(flags, DISSECT_IMAGE_MKDIR)) {
1431 r = mkdir_p(where, 0755);
1432 if (r < 0)
1433 return r;
1434 }
1435
1436 if (uid_is_valid(uid_shift)) {
1437 if (asprintf(&options, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0)
1438 return -ENOMEM;
1439 }
1440
1441 r = mount_nofollow_verbose(LOG_DEBUG, "rootfs", where, "tmpfs", MS_NODEV, options);
1442 if (r < 0)
1443 return r;
1444
1445 return 1;
1446 }
1447
dissected_image_mount(DissectedImage * m,const char * where,uid_t uid_shift,uid_t uid_range,DissectImageFlags flags)1448 int dissected_image_mount(
1449 DissectedImage *m,
1450 const char *where,
1451 uid_t uid_shift,
1452 uid_t uid_range,
1453 DissectImageFlags flags) {
1454
1455 int r, xbootldr_mounted;
1456
1457 assert(m);
1458 assert(where);
1459
1460 /* Returns:
1461 *
1462 * -ENXIO → No root partition found
1463 * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release/extension-release file found
1464 * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet
1465 * -EUCLEAN → fsck for file system failed
1466 * -EBUSY → File system already mounted/used elsewhere (kernel)
1467 * -EAFNOSUPPORT → File system type not supported or not known
1468 */
1469
1470 if (!(m->partitions[PARTITION_ROOT].found ||
1471 (m->partitions[PARTITION_USR].found && FLAGS_SET(flags, DISSECT_IMAGE_USR_NO_ROOT))))
1472 return -ENXIO; /* Require a root fs or at least a /usr/ fs (the latter is subject to a flag of its own) */
1473
1474 if ((flags & DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY) == 0) {
1475
1476 /* First mount the root fs. If there's none we use a tmpfs. */
1477 if (m->partitions[PARTITION_ROOT].found)
1478 r = mount_partition(m->partitions + PARTITION_ROOT, where, NULL, uid_shift, uid_range, flags);
1479 else
1480 r = mount_root_tmpfs(where, uid_shift, flags);
1481 if (r < 0)
1482 return r;
1483
1484 /* For us mounting root always means mounting /usr as well */
1485 r = mount_partition(m->partitions + PARTITION_USR, where, "/usr", uid_shift, uid_range, flags);
1486 if (r < 0)
1487 return r;
1488
1489 if ((flags & (DISSECT_IMAGE_VALIDATE_OS|DISSECT_IMAGE_VALIDATE_OS_EXT)) != 0) {
1490 /* If either one of the validation flags are set, ensure that the image qualifies
1491 * as one or the other (or both). */
1492 bool ok = false;
1493
1494 if (FLAGS_SET(flags, DISSECT_IMAGE_VALIDATE_OS)) {
1495 r = path_is_os_tree(where);
1496 if (r < 0)
1497 return r;
1498 if (r > 0)
1499 ok = true;
1500 }
1501 if (!ok && FLAGS_SET(flags, DISSECT_IMAGE_VALIDATE_OS_EXT)) {
1502 r = path_is_extension_tree(where, m->image_name);
1503 if (r < 0)
1504 return r;
1505 if (r > 0)
1506 ok = true;
1507 }
1508
1509 if (!ok)
1510 return -ENOMEDIUM;
1511 }
1512 }
1513
1514 if (flags & DISSECT_IMAGE_MOUNT_ROOT_ONLY)
1515 return 0;
1516
1517 r = mount_partition(m->partitions + PARTITION_HOME, where, "/home", uid_shift, uid_range, flags);
1518 if (r < 0)
1519 return r;
1520
1521 r = mount_partition(m->partitions + PARTITION_SRV, where, "/srv", uid_shift, uid_range, flags);
1522 if (r < 0)
1523 return r;
1524
1525 r = mount_partition(m->partitions + PARTITION_VAR, where, "/var", uid_shift, uid_range, flags);
1526 if (r < 0)
1527 return r;
1528
1529 r = mount_partition(m->partitions + PARTITION_TMP, where, "/var/tmp", uid_shift, uid_range, flags);
1530 if (r < 0)
1531 return r;
1532
1533 xbootldr_mounted = mount_partition(m->partitions + PARTITION_XBOOTLDR, where, "/boot", uid_shift, uid_range, flags);
1534 if (xbootldr_mounted < 0)
1535 return xbootldr_mounted;
1536
1537 if (m->partitions[PARTITION_ESP].found) {
1538 int esp_done = false;
1539
1540 /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it
1541 * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */
1542
1543 r = chase_symlinks("/efi", where, CHASE_PREFIX_ROOT, NULL, NULL);
1544 if (r < 0) {
1545 if (r != -ENOENT)
1546 return r;
1547
1548 /* /efi doesn't exist. Let's see if /boot is suitable then */
1549
1550 if (!xbootldr_mounted) {
1551 _cleanup_free_ char *p = NULL;
1552
1553 r = chase_symlinks("/boot", where, CHASE_PREFIX_ROOT, &p, NULL);
1554 if (r < 0) {
1555 if (r != -ENOENT)
1556 return r;
1557 } else if (dir_is_empty(p, /* ignore_hidden_or_backup= */ false) > 0) {
1558 /* It exists and is an empty directory. Let's mount the ESP there. */
1559 r = mount_partition(m->partitions + PARTITION_ESP, where, "/boot", uid_shift, uid_range, flags);
1560 if (r < 0)
1561 return r;
1562
1563 esp_done = true;
1564 }
1565 }
1566 }
1567
1568 if (!esp_done) {
1569 /* OK, let's mount the ESP now to /efi (possibly creating the dir if missing) */
1570
1571 r = mount_partition(m->partitions + PARTITION_ESP, where, "/efi", uid_shift, uid_range, flags);
1572 if (r < 0)
1573 return r;
1574 }
1575 }
1576
1577 return 0;
1578 }
1579
dissected_image_mount_and_warn(DissectedImage * m,const char * where,uid_t uid_shift,uid_t uid_range,DissectImageFlags flags)1580 int dissected_image_mount_and_warn(
1581 DissectedImage *m,
1582 const char *where,
1583 uid_t uid_shift,
1584 uid_t uid_range,
1585 DissectImageFlags flags) {
1586
1587 int r;
1588
1589 assert(m);
1590 assert(where);
1591
1592 r = dissected_image_mount(m, where, uid_shift, uid_range, flags);
1593 if (r == -ENXIO)
1594 return log_error_errno(r, "Not root file system found in image.");
1595 if (r == -EMEDIUMTYPE)
1596 return log_error_errno(r, "No suitable os-release/extension-release file in image found.");
1597 if (r == -EUNATCH)
1598 return log_error_errno(r, "Encrypted file system discovered, but decryption not requested.");
1599 if (r == -EUCLEAN)
1600 return log_error_errno(r, "File system check on image failed.");
1601 if (r == -EBUSY)
1602 return log_error_errno(r, "File system already mounted elsewhere.");
1603 if (r == -EAFNOSUPPORT)
1604 return log_error_errno(r, "File system type not supported or not known.");
1605 if (r < 0)
1606 return log_error_errno(r, "Failed to mount image: %m");
1607
1608 return r;
1609 }
1610
1611 #if HAVE_LIBCRYPTSETUP
1612 typedef struct DecryptedPartition {
1613 struct crypt_device *device;
1614 char *name;
1615 bool relinquished;
1616 } DecryptedPartition;
1617
1618 struct DecryptedImage {
1619 DecryptedPartition *decrypted;
1620 size_t n_decrypted;
1621 };
1622 #endif
1623
decrypted_image_unref(DecryptedImage * d)1624 DecryptedImage* decrypted_image_unref(DecryptedImage* d) {
1625 #if HAVE_LIBCRYPTSETUP
1626 int r;
1627
1628 if (!d)
1629 return NULL;
1630
1631 for (size_t i = 0; i < d->n_decrypted; i++) {
1632 DecryptedPartition *p = d->decrypted + i;
1633
1634 if (p->device && p->name && !p->relinquished) {
1635 r = sym_crypt_deactivate_by_name(p->device, p->name, 0);
1636 if (r < 0)
1637 log_debug_errno(r, "Failed to deactivate encrypted partition %s", p->name);
1638 }
1639
1640 if (p->device)
1641 sym_crypt_free(p->device);
1642 free(p->name);
1643 }
1644
1645 free(d->decrypted);
1646 free(d);
1647 #endif
1648 return NULL;
1649 }
1650
1651 #if HAVE_LIBCRYPTSETUP
1652
make_dm_name_and_node(const void * original_node,const char * suffix,char ** ret_name,char ** ret_node)1653 static int make_dm_name_and_node(const void *original_node, const char *suffix, char **ret_name, char **ret_node) {
1654 _cleanup_free_ char *name = NULL, *node = NULL;
1655 const char *base;
1656
1657 assert(original_node);
1658 assert(suffix);
1659 assert(ret_name);
1660 assert(ret_node);
1661
1662 base = strrchr(original_node, '/');
1663 if (!base)
1664 base = original_node;
1665 else
1666 base++;
1667 if (isempty(base))
1668 return -EINVAL;
1669
1670 name = strjoin(base, suffix);
1671 if (!name)
1672 return -ENOMEM;
1673 if (!filename_is_valid(name))
1674 return -EINVAL;
1675
1676 node = path_join(sym_crypt_get_dir(), name);
1677 if (!node)
1678 return -ENOMEM;
1679
1680 *ret_name = TAKE_PTR(name);
1681 *ret_node = TAKE_PTR(node);
1682
1683 return 0;
1684 }
1685
decrypt_partition(DissectedPartition * m,const char * passphrase,DissectImageFlags flags,DecryptedImage * d)1686 static int decrypt_partition(
1687 DissectedPartition *m,
1688 const char *passphrase,
1689 DissectImageFlags flags,
1690 DecryptedImage *d) {
1691
1692 _cleanup_free_ char *node = NULL, *name = NULL;
1693 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
1694 int r;
1695
1696 assert(m);
1697 assert(d);
1698
1699 if (!m->found || !m->node || !m->fstype)
1700 return 0;
1701
1702 if (!streq(m->fstype, "crypto_LUKS"))
1703 return 0;
1704
1705 if (!passphrase)
1706 return -ENOKEY;
1707
1708 r = dlopen_cryptsetup();
1709 if (r < 0)
1710 return r;
1711
1712 r = make_dm_name_and_node(m->node, "-decrypted", &name, &node);
1713 if (r < 0)
1714 return r;
1715
1716 if (!GREEDY_REALLOC0(d->decrypted, d->n_decrypted + 1))
1717 return -ENOMEM;
1718
1719 r = sym_crypt_init(&cd, m->node);
1720 if (r < 0)
1721 return log_debug_errno(r, "Failed to initialize dm-crypt: %m");
1722
1723 cryptsetup_enable_logging(cd);
1724
1725 r = sym_crypt_load(cd, CRYPT_LUKS, NULL);
1726 if (r < 0)
1727 return log_debug_errno(r, "Failed to load LUKS metadata: %m");
1728
1729 r = sym_crypt_activate_by_passphrase(cd, name, CRYPT_ANY_SLOT, passphrase, strlen(passphrase),
1730 ((flags & DISSECT_IMAGE_DEVICE_READ_ONLY) ? CRYPT_ACTIVATE_READONLY : 0) |
1731 ((flags & DISSECT_IMAGE_DISCARD_ON_CRYPTO) ? CRYPT_ACTIVATE_ALLOW_DISCARDS : 0));
1732 if (r < 0) {
1733 log_debug_errno(r, "Failed to activate LUKS device: %m");
1734 return r == -EPERM ? -EKEYREJECTED : r;
1735 }
1736
1737 d->decrypted[d->n_decrypted++] = (DecryptedPartition) {
1738 .name = TAKE_PTR(name),
1739 .device = TAKE_PTR(cd),
1740 };
1741
1742 m->decrypted_node = TAKE_PTR(node);
1743
1744 return 0;
1745 }
1746
verity_can_reuse(const VeritySettings * verity,const char * name,struct crypt_device ** ret_cd)1747 static int verity_can_reuse(
1748 const VeritySettings *verity,
1749 const char *name,
1750 struct crypt_device **ret_cd) {
1751
1752 /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
1753 _cleanup_free_ char *root_hash_existing = NULL;
1754 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
1755 struct crypt_params_verity crypt_params = {};
1756 size_t root_hash_existing_size;
1757 int r;
1758
1759 assert(verity);
1760 assert(name);
1761 assert(ret_cd);
1762
1763 r = sym_crypt_init_by_name(&cd, name);
1764 if (r < 0)
1765 return log_debug_errno(r, "Error opening verity device, crypt_init_by_name failed: %m");
1766
1767 cryptsetup_enable_logging(cd);
1768
1769 r = sym_crypt_get_verity_info(cd, &crypt_params);
1770 if (r < 0)
1771 return log_debug_errno(r, "Error opening verity device, crypt_get_verity_info failed: %m");
1772
1773 root_hash_existing_size = verity->root_hash_size;
1774 root_hash_existing = malloc0(root_hash_existing_size);
1775 if (!root_hash_existing)
1776 return -ENOMEM;
1777
1778 r = sym_crypt_volume_key_get(cd, CRYPT_ANY_SLOT, root_hash_existing, &root_hash_existing_size, NULL, 0);
1779 if (r < 0)
1780 return log_debug_errno(r, "Error opening verity device, crypt_volume_key_get failed: %m");
1781 if (verity->root_hash_size != root_hash_existing_size ||
1782 memcmp(root_hash_existing, verity->root_hash, verity->root_hash_size) != 0)
1783 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but root hashes are different.");
1784
1785 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1786 /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the
1787 * same settings, so that a previous unsigned mount will not be reused if the user asks to use
1788 * signing for the new one, and vice versa. */
1789 if (!!verity->root_hash_sig != !!(crypt_params.flags & CRYPT_VERITY_ROOT_HASH_SIGNATURE))
1790 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but signature settings are not the same.");
1791 #endif
1792
1793 *ret_cd = TAKE_PTR(cd);
1794 return 0;
1795 }
1796
dm_deferred_remove_clean(char * name)1797 static inline char* dm_deferred_remove_clean(char *name) {
1798 if (!name)
1799 return NULL;
1800
1801 (void) sym_crypt_deactivate_by_name(NULL, name, CRYPT_DEACTIVATE_DEFERRED);
1802 return mfree(name);
1803 }
1804 DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean);
1805
validate_signature_userspace(const VeritySettings * verity)1806 static int validate_signature_userspace(const VeritySettings *verity) {
1807 #if HAVE_OPENSSL
1808 _cleanup_(sk_X509_free_allp) STACK_OF(X509) *sk = NULL;
1809 _cleanup_strv_free_ char **certs = NULL;
1810 _cleanup_(PKCS7_freep) PKCS7 *p7 = NULL;
1811 _cleanup_free_ char *s = NULL;
1812 _cleanup_(BIO_freep) BIO *bio = NULL; /* 'bio' must be freed first, 's' second, hence keep this order
1813 * of declaration in place, please */
1814 const unsigned char *d;
1815 int r;
1816
1817 assert(verity);
1818 assert(verity->root_hash);
1819 assert(verity->root_hash_sig);
1820
1821 /* Because installing a signature certificate into the kernel chain is so messy, let's optionally do
1822 * userspace validation. */
1823
1824 r = conf_files_list_nulstr(&certs, ".crt", NULL, CONF_FILES_REGULAR|CONF_FILES_FILTER_MASKED, CONF_PATHS_NULSTR("verity.d"));
1825 if (r < 0)
1826 return log_debug_errno(r, "Failed to enumerate certificates: %m");
1827 if (strv_isempty(certs)) {
1828 log_debug("No userspace dm-verity certificates found.");
1829 return 0;
1830 }
1831
1832 d = verity->root_hash_sig;
1833 p7 = d2i_PKCS7(NULL, &d, (long) verity->root_hash_sig_size);
1834 if (!p7)
1835 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to parse PKCS7 DER signature data.");
1836
1837 s = hexmem(verity->root_hash, verity->root_hash_size);
1838 if (!s)
1839 return log_oom_debug();
1840
1841 bio = BIO_new_mem_buf(s, strlen(s));
1842 if (!bio)
1843 return log_oom_debug();
1844
1845 sk = sk_X509_new_null();
1846 if (!sk)
1847 return log_oom_debug();
1848
1849 STRV_FOREACH(i, certs) {
1850 _cleanup_(X509_freep) X509 *c = NULL;
1851 _cleanup_fclose_ FILE *f = NULL;
1852
1853 f = fopen(*i, "re");
1854 if (!f) {
1855 log_debug_errno(errno, "Failed to open '%s', ignoring: %m", *i);
1856 continue;
1857 }
1858
1859 c = PEM_read_X509(f, NULL, NULL, NULL);
1860 if (!c) {
1861 log_debug("Failed to load X509 certificate '%s', ignoring.", *i);
1862 continue;
1863 }
1864
1865 if (sk_X509_push(sk, c) == 0)
1866 return log_oom_debug();
1867
1868 TAKE_PTR(c);
1869 }
1870
1871 r = PKCS7_verify(p7, sk, NULL, bio, NULL, PKCS7_NOINTERN|PKCS7_NOVERIFY);
1872 if (r)
1873 log_debug("Userspace PKCS#7 validation succeeded.");
1874 else
1875 log_debug("Userspace PKCS#7 validation failed: %s", ERR_error_string(ERR_get_error(), NULL));
1876
1877 return r;
1878 #else
1879 log_debug("Not doing client-side validation of dm-verity root hash signatures, OpenSSL support disabled.");
1880 return 0;
1881 #endif
1882 }
1883
do_crypt_activate_verity(struct crypt_device * cd,const char * name,const VeritySettings * verity)1884 static int do_crypt_activate_verity(
1885 struct crypt_device *cd,
1886 const char *name,
1887 const VeritySettings *verity) {
1888
1889 bool check_signature;
1890 int r;
1891
1892 assert(cd);
1893 assert(name);
1894 assert(verity);
1895
1896 if (verity->root_hash_sig) {
1897 r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIGNATURE");
1898 if (r < 0 && r != -ENXIO)
1899 log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIGNATURE");
1900
1901 check_signature = r != 0;
1902 } else
1903 check_signature = false;
1904
1905 if (check_signature) {
1906
1907 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1908 /* First, if we have support for signed keys in the kernel, then try that first. */
1909 r = sym_crypt_activate_by_signed_key(
1910 cd,
1911 name,
1912 verity->root_hash,
1913 verity->root_hash_size,
1914 verity->root_hash_sig,
1915 verity->root_hash_sig_size,
1916 CRYPT_ACTIVATE_READONLY);
1917 if (r >= 0)
1918 return r;
1919
1920 log_debug("Validation of dm-verity signature failed via the kernel, trying userspace validation instead.");
1921 #else
1922 log_debug("Activation of verity device with signature requested, but not supported via the kernel by %s due to missing crypt_activate_by_signed_key(), trying userspace validation instead.",
1923 program_invocation_short_name);
1924 #endif
1925
1926 /* So this didn't work via the kernel, then let's try userspace validation instead. If that
1927 * works we'll try to activate without telling the kernel the signature. */
1928
1929 r = validate_signature_userspace(verity);
1930 if (r < 0)
1931 return r;
1932 if (r == 0)
1933 return log_debug_errno(SYNTHETIC_ERRNO(ENOKEY),
1934 "Activation of signed Verity volume worked neither via the kernel nor in userspace, can't activate.");
1935 }
1936
1937 return sym_crypt_activate_by_volume_key(
1938 cd,
1939 name,
1940 verity->root_hash,
1941 verity->root_hash_size,
1942 CRYPT_ACTIVATE_READONLY);
1943 }
1944
verity_partition(PartitionDesignator designator,DissectedPartition * m,DissectedPartition * v,const VeritySettings * verity,DissectImageFlags flags,DecryptedImage * d)1945 static int verity_partition(
1946 PartitionDesignator designator,
1947 DissectedPartition *m,
1948 DissectedPartition *v,
1949 const VeritySettings *verity,
1950 DissectImageFlags flags,
1951 DecryptedImage *d) {
1952
1953 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
1954 _cleanup_(dm_deferred_remove_cleanp) char *restore_deferred_remove = NULL;
1955 _cleanup_free_ char *node = NULL, *name = NULL;
1956 int r;
1957
1958 assert(m);
1959 assert(v || (verity && verity->data_path));
1960
1961 if (!verity || !verity->root_hash)
1962 return 0;
1963 if (!((verity->designator < 0 && designator == PARTITION_ROOT) ||
1964 (verity->designator == designator)))
1965 return 0;
1966
1967 if (!m->found || !m->node || !m->fstype)
1968 return 0;
1969 if (!verity->data_path) {
1970 if (!v->found || !v->node || !v->fstype)
1971 return 0;
1972
1973 if (!streq(v->fstype, "DM_verity_hash"))
1974 return 0;
1975 }
1976
1977 r = dlopen_cryptsetup();
1978 if (r < 0)
1979 return r;
1980
1981 if (FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) {
1982 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
1983 _cleanup_free_ char *root_hash_encoded = NULL;
1984
1985 root_hash_encoded = hexmem(verity->root_hash, verity->root_hash_size);
1986 if (!root_hash_encoded)
1987 return -ENOMEM;
1988
1989 r = make_dm_name_and_node(root_hash_encoded, "-verity", &name, &node);
1990 } else
1991 r = make_dm_name_and_node(m->node, "-verity", &name, &node);
1992 if (r < 0)
1993 return r;
1994
1995 r = sym_crypt_init(&cd, verity->data_path ?: v->node);
1996 if (r < 0)
1997 return r;
1998
1999 cryptsetup_enable_logging(cd);
2000
2001 r = sym_crypt_load(cd, CRYPT_VERITY, NULL);
2002 if (r < 0)
2003 return r;
2004
2005 r = sym_crypt_set_data_device(cd, m->node);
2006 if (r < 0)
2007 return r;
2008
2009 if (!GREEDY_REALLOC0(d->decrypted, d->n_decrypted + 1))
2010 return -ENOMEM;
2011
2012 /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
2013 * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
2014 * retry a few times before giving up. */
2015 for (unsigned i = 0; i < N_DEVICE_NODE_LIST_ATTEMPTS; i++) {
2016
2017 r = do_crypt_activate_verity(cd, name, verity);
2018 /* libdevmapper can return EINVAL when the device is already in the activation stage.
2019 * There's no way to distinguish this situation from a genuine error due to invalid
2020 * parameters, so immediately fall back to activating the device with a unique name.
2021 * Improvements in libcrypsetup can ensure this never happens:
2022 * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
2023 if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
2024 return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
2025 if (!IN_SET(r,
2026 0, /* Success */
2027 -EEXIST, /* Volume is already open and ready to be used */
2028 -EBUSY, /* Volume is being opened but not ready, crypt_init_by_name can fetch details */
2029 -ENODEV /* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */))
2030 return r;
2031 if (IN_SET(r, -EEXIST, -EBUSY)) {
2032 struct crypt_device *existing_cd = NULL;
2033
2034 if (!restore_deferred_remove){
2035 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
2036 r = dm_deferred_remove_cancel(name);
2037 /* If activation returns EBUSY there might be no deferred removal to cancel, that's fine */
2038 if (r < 0 && r != -ENXIO)
2039 return log_debug_errno(r, "Disabling automated deferred removal for verity device %s failed: %m", node);
2040 if (r == 0) {
2041 restore_deferred_remove = strdup(name);
2042 if (!restore_deferred_remove)
2043 return -ENOMEM;
2044 }
2045 }
2046
2047 r = verity_can_reuse(verity, name, &existing_cd);
2048 /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
2049 if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
2050 return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
2051 if (!IN_SET(r, 0, -ENODEV, -ENOENT, -EBUSY))
2052 return log_debug_errno(r, "Checking whether existing verity device %s can be reused failed: %m", node);
2053 if (r == 0) {
2054 /* devmapper might say that the device exists, but the devlink might not yet have been
2055 * created. Check and wait for the udev event in that case. */
2056 r = device_wait_for_devlink(node, "block", usec_add(now(CLOCK_MONOTONIC), 100 * USEC_PER_MSEC), NULL);
2057 /* Fallback to activation with a unique device if it's taking too long */
2058 if (r == -ETIMEDOUT)
2059 break;
2060 if (r < 0)
2061 return r;
2062
2063 if (cd)
2064 sym_crypt_free(cd);
2065 cd = existing_cd;
2066 }
2067 }
2068 if (r == 0)
2069 break;
2070
2071 /* Device is being opened by another process, but it has not finished yet, yield for 2ms */
2072 (void) usleep(2 * USEC_PER_MSEC);
2073 }
2074
2075 /* An existing verity device was reported by libcryptsetup/libdevmapper, but we can't use it at this time.
2076 * Fall back to activating it with a unique device name. */
2077 if (r != 0 && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
2078 return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
2079
2080 /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
2081 restore_deferred_remove = mfree(restore_deferred_remove);
2082
2083 d->decrypted[d->n_decrypted++] = (DecryptedPartition) {
2084 .name = TAKE_PTR(name),
2085 .device = TAKE_PTR(cd),
2086 };
2087
2088 m->decrypted_node = TAKE_PTR(node);
2089
2090 return 0;
2091 }
2092 #endif
2093
dissected_image_decrypt(DissectedImage * m,const char * passphrase,const VeritySettings * verity,DissectImageFlags flags,DecryptedImage ** ret)2094 int dissected_image_decrypt(
2095 DissectedImage *m,
2096 const char *passphrase,
2097 const VeritySettings *verity,
2098 DissectImageFlags flags,
2099 DecryptedImage **ret) {
2100
2101 #if HAVE_LIBCRYPTSETUP
2102 _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL;
2103 int r;
2104 #endif
2105
2106 assert(m);
2107 assert(!verity || verity->root_hash || verity->root_hash_size == 0);
2108
2109 /* Returns:
2110 *
2111 * = 0 → There was nothing to decrypt
2112 * > 0 → Decrypted successfully
2113 * -ENOKEY → There's something to decrypt but no key was supplied
2114 * -EKEYREJECTED → Passed key was not correct
2115 */
2116
2117 if (verity && verity->root_hash && verity->root_hash_size < sizeof(sd_id128_t))
2118 return -EINVAL;
2119
2120 if (!m->encrypted && !m->verity_ready) {
2121 *ret = NULL;
2122 return 0;
2123 }
2124
2125 #if HAVE_LIBCRYPTSETUP
2126 d = new0(DecryptedImage, 1);
2127 if (!d)
2128 return -ENOMEM;
2129
2130 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
2131 DissectedPartition *p = m->partitions + i;
2132 PartitionDesignator k;
2133
2134 if (!p->found)
2135 continue;
2136
2137 r = decrypt_partition(p, passphrase, flags, d);
2138 if (r < 0)
2139 return r;
2140
2141 k = PARTITION_VERITY_OF(i);
2142 if (k >= 0) {
2143 r = verity_partition(i, p, m->partitions + k, verity, flags | DISSECT_IMAGE_VERITY_SHARE, d);
2144 if (r < 0)
2145 return r;
2146 }
2147
2148 if (!p->decrypted_fstype && p->decrypted_node) {
2149 r = probe_filesystem(p->decrypted_node, &p->decrypted_fstype);
2150 if (r < 0 && r != -EUCLEAN)
2151 return r;
2152 }
2153 }
2154
2155 *ret = TAKE_PTR(d);
2156
2157 return 1;
2158 #else
2159 return -EOPNOTSUPP;
2160 #endif
2161 }
2162
dissected_image_decrypt_interactively(DissectedImage * m,const char * passphrase,const VeritySettings * verity,DissectImageFlags flags,DecryptedImage ** ret)2163 int dissected_image_decrypt_interactively(
2164 DissectedImage *m,
2165 const char *passphrase,
2166 const VeritySettings *verity,
2167 DissectImageFlags flags,
2168 DecryptedImage **ret) {
2169
2170 _cleanup_strv_free_erase_ char **z = NULL;
2171 int n = 3, r;
2172
2173 if (passphrase)
2174 n--;
2175
2176 for (;;) {
2177 r = dissected_image_decrypt(m, passphrase, verity, flags, ret);
2178 if (r >= 0)
2179 return r;
2180 if (r == -EKEYREJECTED)
2181 log_error_errno(r, "Incorrect passphrase, try again!");
2182 else if (r != -ENOKEY)
2183 return log_error_errno(r, "Failed to decrypt image: %m");
2184
2185 if (--n < 0)
2186 return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED),
2187 "Too many retries.");
2188
2189 z = strv_free(z);
2190
2191 r = ask_password_auto("Please enter image passphrase:", NULL, "dissect", "dissect", "dissect.passphrase", USEC_INFINITY, 0, &z);
2192 if (r < 0)
2193 return log_error_errno(r, "Failed to query for passphrase: %m");
2194
2195 passphrase = z[0];
2196 }
2197 }
2198
decrypted_image_relinquish(DecryptedImage * d)2199 int decrypted_image_relinquish(DecryptedImage *d) {
2200 assert(d);
2201
2202 /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a
2203 * boolean so that we don't clean it up ourselves either anymore */
2204
2205 #if HAVE_LIBCRYPTSETUP
2206 int r;
2207
2208 for (size_t i = 0; i < d->n_decrypted; i++) {
2209 DecryptedPartition *p = d->decrypted + i;
2210
2211 if (p->relinquished)
2212 continue;
2213
2214 r = sym_crypt_deactivate_by_name(NULL, p->name, CRYPT_DEACTIVATE_DEFERRED);
2215 if (r < 0)
2216 return log_debug_errno(r, "Failed to mark %s for auto-removal: %m", p->name);
2217
2218 p->relinquished = true;
2219 }
2220 #endif
2221
2222 return 0;
2223 }
2224
build_auxiliary_path(const char * image,const char * suffix)2225 static char *build_auxiliary_path(const char *image, const char *suffix) {
2226 const char *e;
2227 char *n;
2228
2229 assert(image);
2230 assert(suffix);
2231
2232 e = endswith(image, ".raw");
2233 if (!e)
2234 return strjoin(e, suffix);
2235
2236 n = new(char, e - image + strlen(suffix) + 1);
2237 if (!n)
2238 return NULL;
2239
2240 strcpy(mempcpy(n, image, e - image), suffix);
2241 return n;
2242 }
2243
verity_settings_done(VeritySettings * v)2244 void verity_settings_done(VeritySettings *v) {
2245 assert(v);
2246
2247 v->root_hash = mfree(v->root_hash);
2248 v->root_hash_size = 0;
2249
2250 v->root_hash_sig = mfree(v->root_hash_sig);
2251 v->root_hash_sig_size = 0;
2252
2253 v->data_path = mfree(v->data_path);
2254 }
2255
verity_settings_load(VeritySettings * verity,const char * image,const char * root_hash_path,const char * root_hash_sig_path)2256 int verity_settings_load(
2257 VeritySettings *verity,
2258 const char *image,
2259 const char *root_hash_path,
2260 const char *root_hash_sig_path) {
2261
2262 _cleanup_free_ void *root_hash = NULL, *root_hash_sig = NULL;
2263 size_t root_hash_size = 0, root_hash_sig_size = 0;
2264 _cleanup_free_ char *verity_data_path = NULL;
2265 PartitionDesignator designator;
2266 int r;
2267
2268 assert(verity);
2269 assert(image);
2270 assert(verity->designator < 0 || IN_SET(verity->designator, PARTITION_ROOT, PARTITION_USR));
2271
2272 /* If we are asked to load the root hash for a device node, exit early */
2273 if (is_device_path(image))
2274 return 0;
2275
2276 r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIDECAR");
2277 if (r < 0 && r != -ENXIO)
2278 log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIDECAR, ignoring: %m");
2279 if (r == 0)
2280 return 0;
2281
2282 designator = verity->designator;
2283
2284 /* We only fill in what isn't already filled in */
2285
2286 if (!verity->root_hash) {
2287 _cleanup_free_ char *text = NULL;
2288
2289 if (root_hash_path) {
2290 /* If explicitly specified it takes precedence */
2291 r = read_one_line_file(root_hash_path, &text);
2292 if (r < 0)
2293 return r;
2294
2295 if (designator < 0)
2296 designator = PARTITION_ROOT;
2297 } else {
2298 /* Otherwise look for xattr and separate file, and first for the data for root and if
2299 * that doesn't exist for /usr */
2300
2301 if (designator < 0 || designator == PARTITION_ROOT) {
2302 r = getxattr_malloc(image, "user.verity.roothash", &text);
2303 if (r < 0) {
2304 _cleanup_free_ char *p = NULL;
2305
2306 if (!IN_SET(r, -ENODATA, -ENOENT) && !ERRNO_IS_NOT_SUPPORTED(r))
2307 return r;
2308
2309 p = build_auxiliary_path(image, ".roothash");
2310 if (!p)
2311 return -ENOMEM;
2312
2313 r = read_one_line_file(p, &text);
2314 if (r < 0 && r != -ENOENT)
2315 return r;
2316 }
2317
2318 if (text)
2319 designator = PARTITION_ROOT;
2320 }
2321
2322 if (!text && (designator < 0 || designator == PARTITION_USR)) {
2323 /* So in the "roothash" xattr/file name above the "root" of course primarily
2324 * refers to the root of the Verity Merkle tree. But coincidentally it also
2325 * is the hash for the *root* file system, i.e. the "root" neatly refers to
2326 * two distinct concepts called "root". Taking benefit of this happy
2327 * coincidence we call the file with the root hash for the /usr/ file system
2328 * `usrhash`, because `usrroothash` or `rootusrhash` would just be too
2329 * confusing. We thus drop the reference to the root of the Merkle tree, and
2330 * just indicate which file system it's about. */
2331 r = getxattr_malloc(image, "user.verity.usrhash", &text);
2332 if (r < 0) {
2333 _cleanup_free_ char *p = NULL;
2334
2335 if (!IN_SET(r, -ENODATA, -ENOENT) && !ERRNO_IS_NOT_SUPPORTED(r))
2336 return r;
2337
2338 p = build_auxiliary_path(image, ".usrhash");
2339 if (!p)
2340 return -ENOMEM;
2341
2342 r = read_one_line_file(p, &text);
2343 if (r < 0 && r != -ENOENT)
2344 return r;
2345 }
2346
2347 if (text)
2348 designator = PARTITION_USR;
2349 }
2350 }
2351
2352 if (text) {
2353 r = unhexmem(text, strlen(text), &root_hash, &root_hash_size);
2354 if (r < 0)
2355 return r;
2356 if (root_hash_size < sizeof(sd_id128_t))
2357 return -EINVAL;
2358 }
2359 }
2360
2361 if ((root_hash || verity->root_hash) && !verity->root_hash_sig) {
2362 if (root_hash_sig_path) {
2363 r = read_full_file(root_hash_sig_path, (char**) &root_hash_sig, &root_hash_sig_size);
2364 if (r < 0 && r != -ENOENT)
2365 return r;
2366
2367 if (designator < 0)
2368 designator = PARTITION_ROOT;
2369 } else {
2370 if (designator < 0 || designator == PARTITION_ROOT) {
2371 _cleanup_free_ char *p = NULL;
2372
2373 /* Follow naming convention recommended by the relevant RFC:
2374 * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
2375 p = build_auxiliary_path(image, ".roothash.p7s");
2376 if (!p)
2377 return -ENOMEM;
2378
2379 r = read_full_file(p, (char**) &root_hash_sig, &root_hash_sig_size);
2380 if (r < 0 && r != -ENOENT)
2381 return r;
2382 if (r >= 0)
2383 designator = PARTITION_ROOT;
2384 }
2385
2386 if (!root_hash_sig && (designator < 0 || designator == PARTITION_USR)) {
2387 _cleanup_free_ char *p = NULL;
2388
2389 p = build_auxiliary_path(image, ".usrhash.p7s");
2390 if (!p)
2391 return -ENOMEM;
2392
2393 r = read_full_file(p, (char**) &root_hash_sig, &root_hash_sig_size);
2394 if (r < 0 && r != -ENOENT)
2395 return r;
2396 if (r >= 0)
2397 designator = PARTITION_USR;
2398 }
2399 }
2400
2401 if (root_hash_sig && root_hash_sig_size == 0) /* refuse empty size signatures */
2402 return -EINVAL;
2403 }
2404
2405 if (!verity->data_path) {
2406 _cleanup_free_ char *p = NULL;
2407
2408 p = build_auxiliary_path(image, ".verity");
2409 if (!p)
2410 return -ENOMEM;
2411
2412 if (access(p, F_OK) < 0) {
2413 if (errno != ENOENT)
2414 return -errno;
2415 } else
2416 verity_data_path = TAKE_PTR(p);
2417 }
2418
2419 if (root_hash) {
2420 verity->root_hash = TAKE_PTR(root_hash);
2421 verity->root_hash_size = root_hash_size;
2422 }
2423
2424 if (root_hash_sig) {
2425 verity->root_hash_sig = TAKE_PTR(root_hash_sig);
2426 verity->root_hash_sig_size = root_hash_sig_size;
2427 }
2428
2429 if (verity_data_path)
2430 verity->data_path = TAKE_PTR(verity_data_path);
2431
2432 if (verity->designator < 0)
2433 verity->designator = designator;
2434
2435 return 1;
2436 }
2437
dissected_image_load_verity_sig_partition(DissectedImage * m,int fd,VeritySettings * verity)2438 int dissected_image_load_verity_sig_partition(
2439 DissectedImage *m,
2440 int fd,
2441 VeritySettings *verity) {
2442
2443 _cleanup_free_ void *root_hash = NULL, *root_hash_sig = NULL;
2444 _cleanup_(json_variant_unrefp) JsonVariant *v = NULL;
2445 size_t root_hash_size, root_hash_sig_size;
2446 _cleanup_free_ char *buf = NULL;
2447 PartitionDesignator d;
2448 DissectedPartition *p;
2449 JsonVariant *rh, *sig;
2450 ssize_t n;
2451 char *e;
2452 int r;
2453
2454 assert(m);
2455 assert(fd >= 0);
2456 assert(verity);
2457
2458 if (verity->root_hash && verity->root_hash_sig) /* Already loaded? */
2459 return 0;
2460
2461 r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_EMBEDDED");
2462 if (r < 0 && r != -ENXIO)
2463 log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_EMBEDDED, ignoring: %m");
2464 if (r == 0)
2465 return 0;
2466
2467 d = PARTITION_VERITY_SIG_OF(verity->designator < 0 ? PARTITION_ROOT : verity->designator);
2468 assert(d >= 0);
2469
2470 p = m->partitions + d;
2471 if (!p->found)
2472 return 0;
2473 if (p->offset == UINT64_MAX || p->size == UINT64_MAX)
2474 return -EINVAL;
2475
2476 if (p->size > 4*1024*1024) /* Signature data cannot possible be larger than 4M, refuse that */
2477 return -EFBIG;
2478
2479 buf = new(char, p->size+1);
2480 if (!buf)
2481 return -ENOMEM;
2482
2483 n = pread(fd, buf, p->size, p->offset);
2484 if (n < 0)
2485 return -ENOMEM;
2486 if ((uint64_t) n != p->size)
2487 return -EIO;
2488
2489 e = memchr(buf, 0, p->size);
2490 if (e) {
2491 /* If we found a NUL byte then the rest of the data must be NUL too */
2492 if (!memeqzero(e, p->size - (e - buf)))
2493 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature data contains embedded NUL byte.");
2494 } else
2495 buf[p->size] = 0;
2496
2497 r = json_parse(buf, 0, &v, NULL, NULL);
2498 if (r < 0)
2499 return log_debug_errno(r, "Failed to parse signature JSON data: %m");
2500
2501 rh = json_variant_by_key(v, "rootHash");
2502 if (!rh)
2503 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature JSON object lacks 'rootHash' field.");
2504 if (!json_variant_is_string(rh))
2505 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "'rootHash' field of signature JSON object is not a string.");
2506
2507 r = unhexmem(json_variant_string(rh), SIZE_MAX, &root_hash, &root_hash_size);
2508 if (r < 0)
2509 return log_debug_errno(r, "Failed to parse root hash field: %m");
2510
2511 /* Check if specified root hash matches if it is specified */
2512 if (verity->root_hash &&
2513 memcmp_nn(verity->root_hash, verity->root_hash_size, root_hash, root_hash_size) != 0) {
2514 _cleanup_free_ char *a = NULL, *b = NULL;
2515
2516 a = hexmem(root_hash, root_hash_size);
2517 b = hexmem(verity->root_hash, verity->root_hash_size);
2518
2519 return log_debug_errno(r, "Root hash in signature JSON data (%s) doesn't match configured hash (%s).", strna(a), strna(b));
2520 }
2521
2522 sig = json_variant_by_key(v, "signature");
2523 if (!sig)
2524 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature JSON object lacks 'signature' field.");
2525 if (!json_variant_is_string(sig))
2526 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "'signature' field of signature JSON object is not a string.");
2527
2528 r = unbase64mem(json_variant_string(sig), SIZE_MAX, &root_hash_sig, &root_hash_sig_size);
2529 if (r < 0)
2530 return log_debug_errno(r, "Failed to parse signature field: %m");
2531
2532 free_and_replace(verity->root_hash, root_hash);
2533 verity->root_hash_size = root_hash_size;
2534
2535 free_and_replace(verity->root_hash_sig, root_hash_sig);
2536 verity->root_hash_sig_size = root_hash_sig_size;
2537
2538 return 1;
2539 }
2540
dissected_image_acquire_metadata(DissectedImage * m,DissectImageFlags extra_flags)2541 int dissected_image_acquire_metadata(DissectedImage *m, DissectImageFlags extra_flags) {
2542
2543 enum {
2544 META_HOSTNAME,
2545 META_MACHINE_ID,
2546 META_MACHINE_INFO,
2547 META_OS_RELEASE,
2548 META_EXTENSION_RELEASE,
2549 META_HAS_INIT_SYSTEM,
2550 _META_MAX,
2551 };
2552
2553 static const char *const paths[_META_MAX] = {
2554 [META_HOSTNAME] = "/etc/hostname\0",
2555 [META_MACHINE_ID] = "/etc/machine-id\0",
2556 [META_MACHINE_INFO] = "/etc/machine-info\0",
2557 [META_OS_RELEASE] = ("/etc/os-release\0"
2558 "/usr/lib/os-release\0"),
2559 [META_EXTENSION_RELEASE] = "extension-release\0", /* Used only for logging. */
2560 [META_HAS_INIT_SYSTEM] = "has-init-system\0", /* ditto */
2561 };
2562
2563 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL, **extension_release = NULL;
2564 _cleanup_close_pair_ int error_pipe[2] = { -1, -1 };
2565 _cleanup_(rmdir_and_freep) char *t = NULL;
2566 _cleanup_(sigkill_waitp) pid_t child = 0;
2567 sd_id128_t machine_id = SD_ID128_NULL;
2568 _cleanup_free_ char *hostname = NULL;
2569 unsigned n_meta_initialized = 0;
2570 int fds[2 * _META_MAX], r, v;
2571 int has_init_system = -1;
2572 ssize_t n;
2573
2574 BLOCK_SIGNALS(SIGCHLD);
2575
2576 assert(m);
2577
2578 for (; n_meta_initialized < _META_MAX; n_meta_initialized ++) {
2579 if (!paths[n_meta_initialized]) {
2580 fds[2*n_meta_initialized] = fds[2*n_meta_initialized+1] = -1;
2581 continue;
2582 }
2583
2584 if (pipe2(fds + 2*n_meta_initialized, O_CLOEXEC) < 0) {
2585 r = -errno;
2586 goto finish;
2587 }
2588 }
2589
2590 r = mkdtemp_malloc("/tmp/dissect-XXXXXX", &t);
2591 if (r < 0)
2592 goto finish;
2593
2594 if (pipe2(error_pipe, O_CLOEXEC) < 0) {
2595 r = -errno;
2596 goto finish;
2597 }
2598
2599 r = safe_fork("(sd-dissect)", FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE, &child);
2600 if (r < 0)
2601 goto finish;
2602 if (r == 0) {
2603 /* Child in a new mount namespace */
2604 error_pipe[0] = safe_close(error_pipe[0]);
2605
2606 r = dissected_image_mount(
2607 m,
2608 t,
2609 UID_INVALID,
2610 UID_INVALID,
2611 extra_flags |
2612 DISSECT_IMAGE_READ_ONLY |
2613 DISSECT_IMAGE_MOUNT_ROOT_ONLY |
2614 DISSECT_IMAGE_USR_NO_ROOT);
2615 if (r < 0) {
2616 log_debug_errno(r, "Failed to mount dissected image: %m");
2617 goto inner_fail;
2618 }
2619
2620 for (unsigned k = 0; k < _META_MAX; k++) {
2621 _cleanup_close_ int fd = -ENOENT;
2622 const char *p;
2623
2624 if (!paths[k])
2625 continue;
2626
2627 fds[2*k] = safe_close(fds[2*k]);
2628
2629 switch (k) {
2630
2631 case META_EXTENSION_RELEASE:
2632 /* As per the os-release spec, if the image is an extension it will have a file
2633 * named after the image name in extension-release.d/ - we use the image name
2634 * and try to resolve it with the extension-release helpers, as sometimes
2635 * the image names are mangled on deployment and do not match anymore.
2636 * Unlike other paths this is not fixed, and the image name
2637 * can be mangled on deployment, so by calling into the helper
2638 * we allow a fallback that matches on the first extension-release
2639 * file found in the directory, if one named after the image cannot
2640 * be found first. */
2641 r = open_extension_release(t, m->image_name, NULL, &fd);
2642 if (r < 0)
2643 fd = r; /* Propagate the error. */
2644 break;
2645
2646 case META_HAS_INIT_SYSTEM: {
2647 bool found = false;
2648
2649 FOREACH_STRING(init,
2650 "/usr/lib/systemd/systemd", /* systemd on /usr merged system */
2651 "/lib/systemd/systemd", /* systemd on /usr non-merged systems */
2652 "/sbin/init") { /* traditional path the Linux kernel invokes */
2653
2654 r = chase_symlinks(init, t, CHASE_PREFIX_ROOT, NULL, NULL);
2655 if (r < 0) {
2656 if (r != -ENOENT)
2657 log_debug_errno(r, "Failed to resolve %s, ignoring: %m", init);
2658 } else {
2659 found = true;
2660 break;
2661 }
2662 }
2663
2664 r = loop_write(fds[2*k+1], &found, sizeof(found), false);
2665 if (r < 0)
2666 goto inner_fail;
2667
2668 continue;
2669 }
2670
2671 default:
2672 NULSTR_FOREACH(p, paths[k]) {
2673 fd = chase_symlinks_and_open(p, t, CHASE_PREFIX_ROOT, O_RDONLY|O_CLOEXEC|O_NOCTTY, NULL);
2674 if (fd >= 0)
2675 break;
2676 }
2677 }
2678
2679 if (fd < 0) {
2680 log_debug_errno(fd, "Failed to read %s file of image, ignoring: %m", paths[k]);
2681 fds[2*k+1] = safe_close(fds[2*k+1]);
2682 continue;
2683 }
2684
2685 r = copy_bytes(fd, fds[2*k+1], UINT64_MAX, 0);
2686 if (r < 0)
2687 goto inner_fail;
2688
2689 fds[2*k+1] = safe_close(fds[2*k+1]);
2690 }
2691
2692 _exit(EXIT_SUCCESS);
2693
2694 inner_fail:
2695 /* Let parent know the error */
2696 (void) write(error_pipe[1], &r, sizeof(r));
2697 _exit(EXIT_FAILURE);
2698 }
2699
2700 error_pipe[1] = safe_close(error_pipe[1]);
2701
2702 for (unsigned k = 0; k < _META_MAX; k++) {
2703 _cleanup_fclose_ FILE *f = NULL;
2704
2705 if (!paths[k])
2706 continue;
2707
2708 fds[2*k+1] = safe_close(fds[2*k+1]);
2709
2710 f = take_fdopen(&fds[2*k], "r");
2711 if (!f) {
2712 r = -errno;
2713 goto finish;
2714 }
2715
2716 switch (k) {
2717
2718 case META_HOSTNAME:
2719 r = read_etc_hostname_stream(f, &hostname);
2720 if (r < 0)
2721 log_debug_errno(r, "Failed to read /etc/hostname of image: %m");
2722
2723 break;
2724
2725 case META_MACHINE_ID: {
2726 _cleanup_free_ char *line = NULL;
2727
2728 r = read_line(f, LONG_LINE_MAX, &line);
2729 if (r < 0)
2730 log_debug_errno(r, "Failed to read /etc/machine-id of image: %m");
2731 else if (r == 33) {
2732 r = sd_id128_from_string(line, &machine_id);
2733 if (r < 0)
2734 log_debug_errno(r, "Image contains invalid /etc/machine-id: %s", line);
2735 } else if (r == 0)
2736 log_debug("/etc/machine-id file of image is empty.");
2737 else if (streq(line, "uninitialized"))
2738 log_debug("/etc/machine-id file of image is uninitialized (likely aborted first boot).");
2739 else
2740 log_debug("/etc/machine-id file of image has unexpected length %i.", r);
2741
2742 break;
2743 }
2744
2745 case META_MACHINE_INFO:
2746 r = load_env_file_pairs(f, "machine-info", &machine_info);
2747 if (r < 0)
2748 log_debug_errno(r, "Failed to read /etc/machine-info of image: %m");
2749
2750 break;
2751
2752 case META_OS_RELEASE:
2753 r = load_env_file_pairs(f, "os-release", &os_release);
2754 if (r < 0)
2755 log_debug_errno(r, "Failed to read OS release file of image: %m");
2756
2757 break;
2758
2759 case META_EXTENSION_RELEASE:
2760 r = load_env_file_pairs(f, "extension-release", &extension_release);
2761 if (r < 0)
2762 log_debug_errno(r, "Failed to read extension release file of image: %m");
2763
2764 break;
2765
2766 case META_HAS_INIT_SYSTEM: {
2767 bool b = false;
2768 size_t nr;
2769
2770 errno = 0;
2771 nr = fread(&b, 1, sizeof(b), f);
2772 if (nr != sizeof(b))
2773 log_debug_errno(errno_or_else(EIO), "Failed to read has-init-system boolean: %m");
2774 else
2775 has_init_system = b;
2776
2777 break;
2778 }}
2779 }
2780
2781 r = wait_for_terminate_and_check("(sd-dissect)", child, 0);
2782 child = 0;
2783 if (r < 0)
2784 return r;
2785
2786 n = read(error_pipe[0], &v, sizeof(v));
2787 if (n < 0)
2788 return -errno;
2789 if (n == sizeof(v))
2790 return v; /* propagate error sent to us from child */
2791 if (n != 0)
2792 return -EIO;
2793
2794 if (r != EXIT_SUCCESS)
2795 return -EPROTO;
2796
2797 free_and_replace(m->hostname, hostname);
2798 m->machine_id = machine_id;
2799 strv_free_and_replace(m->machine_info, machine_info);
2800 strv_free_and_replace(m->os_release, os_release);
2801 strv_free_and_replace(m->extension_release, extension_release);
2802 m->has_init_system = has_init_system;
2803
2804 finish:
2805 for (unsigned k = 0; k < n_meta_initialized; k++)
2806 safe_close_pair(fds + 2*k);
2807
2808 return r;
2809 }
2810
dissect_image_and_warn(int fd,const char * name,const VeritySettings * verity,const MountOptions * mount_options,uint64_t diskseq,uint64_t uevent_seqnum_not_before,usec_t timestamp_not_before,DissectImageFlags flags,DissectedImage ** ret)2811 int dissect_image_and_warn(
2812 int fd,
2813 const char *name,
2814 const VeritySettings *verity,
2815 const MountOptions *mount_options,
2816 uint64_t diskseq,
2817 uint64_t uevent_seqnum_not_before,
2818 usec_t timestamp_not_before,
2819 DissectImageFlags flags,
2820 DissectedImage **ret) {
2821
2822 _cleanup_free_ char *buffer = NULL;
2823 int r;
2824
2825 if (!name) {
2826 r = fd_get_path(fd, &buffer);
2827 if (r < 0)
2828 return r;
2829
2830 name = buffer;
2831 }
2832
2833 r = dissect_image(fd, verity, mount_options, diskseq, uevent_seqnum_not_before, timestamp_not_before, flags, ret);
2834 switch (r) {
2835
2836 case -EOPNOTSUPP:
2837 return log_error_errno(r, "Dissecting images is not supported, compiled without blkid support.");
2838
2839 case -ENOPKG:
2840 return log_error_errno(r, "%s: Couldn't identify a suitable partition table or file system.", name);
2841
2842 case -ENOMEDIUM:
2843 return log_error_errno(r, "%s: The image does not pass validation.", name);
2844
2845 case -EADDRNOTAVAIL:
2846 return log_error_errno(r, "%s: No root partition for specified root hash found.", name);
2847
2848 case -ENOTUNIQ:
2849 return log_error_errno(r, "%s: Multiple suitable root partitions found in image.", name);
2850
2851 case -ENXIO:
2852 return log_error_errno(r, "%s: No suitable root partition found in image.", name);
2853
2854 case -EPROTONOSUPPORT:
2855 return log_error_errno(r, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name);
2856
2857 case -ENOTBLK:
2858 return log_error_errno(r, "%s: Image is not a block device.", name);
2859
2860 case -EBADR:
2861 return log_error_errno(r,
2862 "Combining partitioned images (such as '%s') with external Verity data (such as '%s') not supported. "
2863 "(Consider setting $SYSTEMD_DISSECT_VERITY_SIDECAR=0 to disable automatic discovery of external Verity data.)",
2864 name, strna(verity ? verity->data_path : NULL));
2865
2866 default:
2867 if (r < 0)
2868 return log_error_errno(r, "Failed to dissect image '%s': %m", name);
2869
2870 return r;
2871 }
2872 }
2873
dissected_image_verity_candidate(const DissectedImage * image,PartitionDesignator partition_designator)2874 bool dissected_image_verity_candidate(const DissectedImage *image, PartitionDesignator partition_designator) {
2875 assert(image);
2876
2877 /* Checks if this partition could theoretically do Verity. For non-partitioned images this only works
2878 * if there's an external verity file supplied, for which we can consult .has_verity. For partitioned
2879 * images we only check the partition type.
2880 *
2881 * This call is used to decide whether to suppress or show a verity column in tabular output of the
2882 * image. */
2883
2884 if (image->single_file_system)
2885 return partition_designator == PARTITION_ROOT && image->has_verity;
2886
2887 return PARTITION_VERITY_OF(partition_designator) >= 0;
2888 }
2889
dissected_image_verity_ready(const DissectedImage * image,PartitionDesignator partition_designator)2890 bool dissected_image_verity_ready(const DissectedImage *image, PartitionDesignator partition_designator) {
2891 PartitionDesignator k;
2892
2893 assert(image);
2894
2895 /* Checks if this partition has verity data available that we can activate. For non-partitioned this
2896 * works for the root partition, for others only if the associated verity partition was found. */
2897
2898 if (!image->verity_ready)
2899 return false;
2900
2901 if (image->single_file_system)
2902 return partition_designator == PARTITION_ROOT;
2903
2904 k = PARTITION_VERITY_OF(partition_designator);
2905 return k >= 0 && image->partitions[k].found;
2906 }
2907
dissected_image_verity_sig_ready(const DissectedImage * image,PartitionDesignator partition_designator)2908 bool dissected_image_verity_sig_ready(const DissectedImage *image, PartitionDesignator partition_designator) {
2909 PartitionDesignator k;
2910
2911 assert(image);
2912
2913 /* Checks if this partition has verity signature data available that we can use. */
2914
2915 if (!image->verity_sig_ready)
2916 return false;
2917
2918 if (image->single_file_system)
2919 return partition_designator == PARTITION_ROOT;
2920
2921 k = PARTITION_VERITY_SIG_OF(partition_designator);
2922 return k >= 0 && image->partitions[k].found;
2923 }
2924
mount_options_free_all(MountOptions * options)2925 MountOptions* mount_options_free_all(MountOptions *options) {
2926 MountOptions *m;
2927
2928 while ((m = options)) {
2929 LIST_REMOVE(mount_options, options, m);
2930 free(m->options);
2931 free(m);
2932 }
2933
2934 return NULL;
2935 }
2936
mount_options_from_designator(const MountOptions * options,PartitionDesignator designator)2937 const char* mount_options_from_designator(const MountOptions *options, PartitionDesignator designator) {
2938 LIST_FOREACH(mount_options, m, options)
2939 if (designator == m->partition_designator && !isempty(m->options))
2940 return m->options;
2941
2942 return NULL;
2943 }
2944
mount_image_privately_interactively(const char * image,DissectImageFlags flags,char ** ret_directory,LoopDevice ** ret_loop_device,DecryptedImage ** ret_decrypted_image)2945 int mount_image_privately_interactively(
2946 const char *image,
2947 DissectImageFlags flags,
2948 char **ret_directory,
2949 LoopDevice **ret_loop_device,
2950 DecryptedImage **ret_decrypted_image) {
2951
2952 _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT;
2953 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
2954 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
2955 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
2956 _cleanup_(rmdir_and_freep) char *created_dir = NULL;
2957 _cleanup_free_ char *temp = NULL;
2958 int r;
2959
2960 /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
2961 * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
2962 * easily. */
2963
2964 assert(image);
2965 assert(ret_directory);
2966 assert(ret_loop_device);
2967 assert(ret_decrypted_image);
2968
2969 r = verity_settings_load(&verity, image, NULL, NULL);
2970 if (r < 0)
2971 return log_error_errno(r, "Failed to load root hash data: %m");
2972
2973 r = tempfn_random_child(NULL, program_invocation_short_name, &temp);
2974 if (r < 0)
2975 return log_error_errno(r, "Failed to generate temporary mount directory: %m");
2976
2977 r = loop_device_make_by_path(
2978 image,
2979 FLAGS_SET(flags, DISSECT_IMAGE_DEVICE_READ_ONLY) ? O_RDONLY : O_RDWR,
2980 FLAGS_SET(flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN,
2981 &d);
2982 if (r < 0)
2983 return log_error_errno(r, "Failed to set up loopback device for %s: %m", image);
2984
2985 /* Make sure udevd doesn't issue BLKRRPART behind our backs */
2986 r = loop_device_flock(d, LOCK_SH);
2987 if (r < 0)
2988 return r;
2989
2990 r = dissect_image_and_warn(d->fd, image, &verity, NULL, d->diskseq, d->uevent_seqnum_not_before, d->timestamp_not_before, flags, &dissected_image);
2991 if (r < 0)
2992 return r;
2993
2994 r = dissected_image_load_verity_sig_partition(dissected_image, d->fd, &verity);
2995 if (r < 0)
2996 return r;
2997
2998 r = dissected_image_decrypt_interactively(dissected_image, NULL, &verity, flags, &decrypted_image);
2999 if (r < 0)
3000 return r;
3001
3002 r = detach_mount_namespace();
3003 if (r < 0)
3004 return log_error_errno(r, "Failed to detach mount namespace: %m");
3005
3006 r = mkdir_p(temp, 0700);
3007 if (r < 0)
3008 return log_error_errno(r, "Failed to create mount point: %m");
3009
3010 created_dir = TAKE_PTR(temp);
3011
3012 r = dissected_image_mount_and_warn(dissected_image, created_dir, UID_INVALID, UID_INVALID, flags);
3013 if (r < 0)
3014 return r;
3015
3016 r = loop_device_flock(d, LOCK_UN);
3017 if (r < 0)
3018 return r;
3019
3020 if (decrypted_image) {
3021 r = decrypted_image_relinquish(decrypted_image);
3022 if (r < 0)
3023 return log_error_errno(r, "Failed to relinquish DM devices: %m");
3024 }
3025
3026 loop_device_relinquish(d);
3027
3028 *ret_directory = TAKE_PTR(created_dir);
3029 *ret_loop_device = TAKE_PTR(d);
3030 *ret_decrypted_image = TAKE_PTR(decrypted_image);
3031
3032 return 0;
3033 }
3034
3035 static const char *const partition_designator_table[] = {
3036 [PARTITION_ROOT] = "root",
3037 [PARTITION_ROOT_SECONDARY] = "root-secondary",
3038 [PARTITION_ROOT_OTHER] = "root-other",
3039 [PARTITION_USR] = "usr",
3040 [PARTITION_USR_SECONDARY] = "usr-secondary",
3041 [PARTITION_USR_OTHER] = "usr-other",
3042 [PARTITION_HOME] = "home",
3043 [PARTITION_SRV] = "srv",
3044 [PARTITION_ESP] = "esp",
3045 [PARTITION_XBOOTLDR] = "xbootldr",
3046 [PARTITION_SWAP] = "swap",
3047 [PARTITION_ROOT_VERITY] = "root-verity",
3048 [PARTITION_ROOT_SECONDARY_VERITY] = "root-secondary-verity",
3049 [PARTITION_ROOT_OTHER_VERITY] = "root-other-verity",
3050 [PARTITION_USR_VERITY] = "usr-verity",
3051 [PARTITION_USR_SECONDARY_VERITY] = "usr-secondary-verity",
3052 [PARTITION_USR_OTHER_VERITY] = "usr-other-verity",
3053 [PARTITION_ROOT_VERITY_SIG] = "root-verity-sig",
3054 [PARTITION_ROOT_SECONDARY_VERITY_SIG] = "root-secondary-verity-sig",
3055 [PARTITION_ROOT_OTHER_VERITY_SIG] = "root-other-verity-sig",
3056 [PARTITION_USR_VERITY_SIG] = "usr-verity-sig",
3057 [PARTITION_USR_SECONDARY_VERITY_SIG] = "usr-secondary-verity-sig",
3058 [PARTITION_USR_OTHER_VERITY_SIG] = "usr-other-verity-sig",
3059 [PARTITION_TMP] = "tmp",
3060 [PARTITION_VAR] = "var",
3061 };
3062
verity_dissect_and_mount(int src_fd,const char * src,const char * dest,const MountOptions * options,const char * required_host_os_release_id,const char * required_host_os_release_version_id,const char * required_host_os_release_sysext_level,const char * required_sysext_scope)3063 int verity_dissect_and_mount(
3064 int src_fd,
3065 const char *src,
3066 const char *dest,
3067 const MountOptions *options,
3068 const char *required_host_os_release_id,
3069 const char *required_host_os_release_version_id,
3070 const char *required_host_os_release_sysext_level,
3071 const char *required_sysext_scope) {
3072
3073 _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL;
3074 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
3075 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
3076 _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT;
3077 DissectImageFlags dissect_image_flags;
3078 int r;
3079
3080 assert(src);
3081 assert(dest);
3082
3083 /* We might get an FD for the image, but we use the original path to look for the dm-verity files */
3084 r = verity_settings_load(&verity, src, NULL, NULL);
3085 if (r < 0)
3086 return log_debug_errno(r, "Failed to load root hash: %m");
3087
3088 dissect_image_flags = verity.data_path ? DISSECT_IMAGE_NO_PARTITION_TABLE : 0;
3089
3090 /* Note that we don't use loop_device_make here, as the FD is most likely O_PATH which would not be
3091 * accepted by LOOP_CONFIGURE, so just let loop_device_make_by_path reopen it as a regular FD. */
3092 r = loop_device_make_by_path(
3093 src_fd >= 0 ? FORMAT_PROC_FD_PATH(src_fd) : src,
3094 -1,
3095 verity.data_path ? 0 : LO_FLAGS_PARTSCAN,
3096 &loop_device);
3097 if (r < 0)
3098 return log_debug_errno(r, "Failed to create loop device for image: %m");
3099
3100 r = loop_device_flock(loop_device, LOCK_SH);
3101 if (r < 0)
3102 return log_debug_errno(r, "Failed to lock loop device: %m");
3103
3104 r = dissect_image(
3105 loop_device->fd,
3106 &verity,
3107 options,
3108 loop_device->diskseq,
3109 loop_device->uevent_seqnum_not_before,
3110 loop_device->timestamp_not_before,
3111 dissect_image_flags,
3112 &dissected_image);
3113 /* No partition table? Might be a single-filesystem image, try again */
3114 if (!verity.data_path && r == -ENOPKG)
3115 r = dissect_image(
3116 loop_device->fd,
3117 &verity,
3118 options,
3119 loop_device->diskseq,
3120 loop_device->uevent_seqnum_not_before,
3121 loop_device->timestamp_not_before,
3122 dissect_image_flags | DISSECT_IMAGE_NO_PARTITION_TABLE,
3123 &dissected_image);
3124 if (r < 0)
3125 return log_debug_errno(r, "Failed to dissect image: %m");
3126
3127 r = dissected_image_load_verity_sig_partition(dissected_image, loop_device->fd, &verity);
3128 if (r < 0)
3129 return r;
3130
3131 r = dissected_image_decrypt(
3132 dissected_image,
3133 NULL,
3134 &verity,
3135 dissect_image_flags,
3136 &decrypted_image);
3137 if (r < 0)
3138 return log_debug_errno(r, "Failed to decrypt dissected image: %m");
3139
3140 r = mkdir_p_label(dest, 0755);
3141 if (r < 0)
3142 return log_debug_errno(r, "Failed to create destination directory %s: %m", dest);
3143 r = umount_recursive(dest, 0);
3144 if (r < 0)
3145 return log_debug_errno(r, "Failed to umount under destination directory %s: %m", dest);
3146
3147 r = dissected_image_mount(dissected_image, dest, UID_INVALID, UID_INVALID, dissect_image_flags);
3148 if (r < 0)
3149 return log_debug_errno(r, "Failed to mount image: %m");
3150
3151 r = loop_device_flock(loop_device, LOCK_UN);
3152 if (r < 0)
3153 return log_debug_errno(r, "Failed to unlock loopback device: %m");
3154
3155 /* If we got os-release values from the caller, then we need to match them with the image's
3156 * extension-release.d/ content. Return -EINVAL if there's any mismatch.
3157 * First, check the distro ID. If that matches, then check the new SYSEXT_LEVEL value if
3158 * available, or else fallback to VERSION_ID. If neither is present (eg: rolling release),
3159 * then a simple match on the ID will be performed. */
3160 if (!isempty(required_host_os_release_id)) {
3161 _cleanup_strv_free_ char **extension_release = NULL;
3162
3163 r = load_extension_release_pairs(dest, dissected_image->image_name, &extension_release);
3164 if (r < 0)
3165 return log_debug_errno(r, "Failed to parse image %s extension-release metadata: %m", dissected_image->image_name);
3166
3167 r = extension_release_validate(
3168 dissected_image->image_name,
3169 required_host_os_release_id,
3170 required_host_os_release_version_id,
3171 required_host_os_release_sysext_level,
3172 required_sysext_scope,
3173 extension_release);
3174 if (r == 0)
3175 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE), "Image %s extension-release metadata does not match the root's", dissected_image->image_name);
3176 if (r < 0)
3177 return log_debug_errno(r, "Failed to compare image %s extension-release metadata with the root's os-release: %m", dissected_image->image_name);
3178 }
3179
3180 if (decrypted_image) {
3181 r = decrypted_image_relinquish(decrypted_image);
3182 if (r < 0)
3183 return log_debug_errno(r, "Failed to relinquish decrypted image: %m");
3184 }
3185
3186 loop_device_relinquish(loop_device);
3187
3188 return 0;
3189 }
3190
3191 DEFINE_STRING_TABLE_LOOKUP(partition_designator, PartitionDesignator);
3192