]> git.ipfire.org Git - thirdparty/systemd.git/blame - src/shared/dissect-image.c
dissect-image: split out verity_timeout()
[thirdparty/systemd.git] / src / shared / dissect-image.c
CommitLineData
db9ecf05 1/* SPDX-License-Identifier: LGPL-2.1-or-later */
8c1be37e 2
10c1b188
LP
3#if HAVE_VALGRIND_MEMCHECK_H
4#include <valgrind/memcheck.h>
5#endif
6
01234e1f
YW
7#include <linux/dm-ioctl.h>
8#include <linux/loop.h>
19df770f 9#include <sys/file.h>
8c1be37e 10#include <sys/mount.h>
3b925504
LP
11#include <sys/prctl.h>
12#include <sys/wait.h>
f5ea63a5 13#include <sysexits.h>
8c1be37e 14
c2fa92e7
LP
15#if HAVE_OPENSSL
16#include <openssl/err.h>
17#include <openssl/pem.h>
18#include <openssl/x509.h>
19#endif
20
3c1f2cee 21#include "sd-device.h"
dccca82b
LP
22#include "sd-id128.h"
23
8c1be37e 24#include "architecture.h"
18b5886e 25#include "ask-password-api.h"
8c1be37e 26#include "blkid-util.h"
18c528e9 27#include "blockdev-util.h"
f4351959 28#include "chase-symlinks.h"
c2fa92e7 29#include "conf-files.h"
3b925504 30#include "copy.h"
1e2f3230 31#include "cryptsetup-util.h"
3b925504 32#include "def.h"
553e15f2 33#include "device-nodes.h"
8437c059 34#include "device-util.h"
7718ac97 35#include "discover-image.h"
8c1be37e 36#include "dissect-image.h"
a709a315 37#include "dm-util.h"
686d13b9 38#include "env-file.h"
88b3300f 39#include "env-util.h"
93f59701 40#include "extension-release.h"
18b5886e 41#include "fd-util.h"
78ebe980 42#include "fileio.h"
2eedfd2d 43#include "fs-util.h"
cf32c486 44#include "fsck-util.h"
8c1be37e 45#include "gpt.h"
78ebe980 46#include "hexdecoct.h"
e2054217 47#include "hostname-setup.h"
3b925504 48#include "id128-util.h"
593fe6c0 49#include "import-util.h"
a4e0d617 50#include "io-util.h"
35cd0ba5 51#include "mkdir-label.h"
8c1be37e 52#include "mount-util.h"
e4de7287 53#include "mountpoint-util.h"
6aa05ebd 54#include "namespace-util.h"
d8b4d14d 55#include "nulstr-util.h"
c2fa92e7 56#include "openssl-util.h"
d58ad743 57#include "os-util.h"
8c1be37e 58#include "path-util.h"
3b925504
LP
59#include "process-util.h"
60#include "raw-clone.h"
81939d9d 61#include "resize-fs.h"
3b925504 62#include "signal-util.h"
8c1be37e 63#include "stat-util.h"
18b5886e 64#include "stdio-util.h"
8c1be37e
LP
65#include "string-table.h"
66#include "string-util.h"
2eedfd2d 67#include "strv.h"
e4de7287 68#include "tmpfile-util.h"
a8040b6d 69#include "udev-util.h"
2d3a5a73 70#include "user-util.h"
41488e1f 71#include "xattr-util.h"
8c1be37e 72
28e2641a
FF
73/* how many times to wait for the device nodes to appear */
74#define N_DEVICE_NODE_LIST_ATTEMPTS 10
75
c34b75a1 76int probe_filesystem(const char *node, char **ret_fstype) {
7cc84b2c 77 /* Try to find device content type and return it in *ret_fstype. If nothing is found,
5238e957 78 * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and an
7cc84b2c
ZJS
79 * different error otherwise. */
80
349cc4a5 81#if HAVE_BLKID
8e766630 82 _cleanup_(blkid_free_probep) blkid_probe b = NULL;
18b5886e
LP
83 const char *fstype;
84 int r;
85
995fa2e5 86 errno = 0;
18b5886e
LP
87 b = blkid_new_probe_from_filename(node);
88 if (!b)
66855de7 89 return errno_or_else(ENOMEM);
18b5886e
LP
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);
1cefb9a6 96 if (r == 1)
18b5886e 97 goto not_found;
58dfbfbd
LP
98 if (r == -2)
99 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN),
100 "Results ambiguous for partition %s", node);
b382db9f 101 if (r != 0)
1cefb9a6 102 return log_debug_errno(errno_or_else(EIO), "Failed to probe partition %s: %m", node);
18b5886e
LP
103
104 (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
105
106 if (fstype) {
107 char *t;
108
1cefb9a6
LP
109 log_debug("Probed fstype '%s' on partition %s.", fstype, node);
110
18b5886e
LP
111 t = strdup(fstype);
112 if (!t)
113 return -ENOMEM;
114
115 *ret_fstype = t;
116 return 1;
117 }
118
119not_found:
1cefb9a6 120 log_debug("No type detected on partition %s", node);
18b5886e
LP
121 *ret_fstype = NULL;
122 return 0;
d1c536f5
ZJS
123#else
124 return -EOPNOTSUPP;
a75e27eb 125#endif
d1c536f5 126}
18b5886e 127
40c10d3f 128#if HAVE_BLKID
0f7c9a3d
LP
129static 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}
1b010ae7 151#endif
0f7c9a3d 152
234c2e16 153static void dissected_partition_done(DissectedPartition *p) {
1b010ae7 154 assert(p);
786e3a52 155
1b010ae7
LP
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);
786e3a52 162
1b010ae7
LP
163 *p = (DissectedPartition) {
164 .partno = -1,
165 .architecture = _ARCHITECTURE_INVALID,
166 };
167}
786e3a52 168
1b010ae7 169#if HAVE_BLKID
1b010ae7
LP
170static int make_partition_devname(
171 const char *whole_devname,
172 int nr,
173 char **ret) {
786e3a52 174
1b010ae7 175 bool need_p;
786e3a52 176
1b010ae7
LP
177 assert(whole_devname);
178 assert(nr > 0);
aae22eb3 179
1b010ae7
LP
180 /* Given a whole block device node name (e.g. /dev/sda or /dev/loop7) generate a partition device
181 * name (e.g. /dev/sda7 or /dev/loop7p5). The rule the kernel uses is simple: if whole block device
182 * node name ends in a digit, then suffix a 'p', followed by the partition number. Otherwise, just
183 * suffix the partition number without any 'p'. */
4ba86848 184
1b010ae7
LP
185 if (isempty(whole_devname)) /* Make sure there *is* a last char */
186 return -EINVAL;
08fe0a53 187
ff25d338 188 need_p = ascii_isdigit(whole_devname[strlen(whole_devname)-1]); /* Last char a digit? */
08fe0a53 189
1b010ae7 190 return asprintf(ret, "%s%s%i", whole_devname, need_p ? "p" : "", nr);
08fe0a53 191}
1b010ae7 192#endif
08fe0a53 193
4526113f
LP
194int dissect_image(
195 int fd,
0b214aa0 196 const char *devname,
369de26f 197 const char *image_path,
89e62e0b 198 const VeritySettings *verity,
18d73705 199 const MountOptions *mount_options,
4526113f
LP
200 DissectImageFlags flags,
201 DissectedImage **ret) {
8c1be37e 202
349cc4a5 203#if HAVE_BLKID
62ea0ed0 204 sd_id128_t root_uuid = SD_ID128_NULL, root_verity_uuid = SD_ID128_NULL;
62ea0ed0 205 sd_id128_t usr_uuid = SD_ID128_NULL, usr_verity_uuid = SD_ID128_NULL;
1f8fb21c 206 bool is_gpt, is_mbr, multiple_generic = false,
de98f631
LP
207 generic_rw = false, /* initialize to appease gcc */
208 generic_growfs = false;
8c1be37e 209 _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL;
8e766630 210 _cleanup_(blkid_free_probep) blkid_probe b = NULL;
8c1be37e 211 _cleanup_free_ char *generic_node = NULL;
be30ad41 212 sd_id128_t generic_uuid = SD_ID128_NULL;
0b214aa0 213 const char *pttype = NULL;
8c1be37e 214 blkid_partlist pl;
1f8fb21c 215 int r, generic_nr = -1, n_partitions;
8c1be37e
LP
216
217 assert(fd >= 0);
0b214aa0 218 assert(devname);
8c1be37e 219 assert(ret);
a0bff7ea 220 assert(!verity || verity->designator < 0 || IN_SET(verity->designator, PARTITION_ROOT, PARTITION_USR));
89e62e0b 221 assert(!verity || verity->root_hash || verity->root_hash_size == 0);
a0bff7ea
LP
222 assert(!verity || verity->root_hash_sig || verity->root_hash_sig_size == 0);
223 assert(!verity || (verity->root_hash || !verity->root_hash_sig));
e7cbe5cb 224 assert(!((flags & DISSECT_IMAGE_GPT_ONLY) && (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)));
8c1be37e
LP
225
226 /* Probes a disk image, and returns information about what it found in *ret.
227 *
4623e8e6 228 * Returns -ENOPKG if no suitable partition table or file system could be found.
2679f407
LP
229 * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found.
230 * Returns -ENXIO if we couldn't find any partition suitable as root or /usr partition
231 * Returns -ENOTUNIQ if we only found multiple generic partitions and thus don't know what to do with that */
4623e8e6 232
89e62e0b 233 if (verity && verity->root_hash) {
aee36b4e
LP
234 sd_id128_t fsuuid, vuuid;
235
236 /* If a root hash is supplied, then we use the root partition that has a UUID that match the
237 * first 128bit of the root hash. And we use the verity partition that has a UUID that match
238 * the final 128bit. */
4623e8e6 239
89e62e0b 240 if (verity->root_hash_size < sizeof(sd_id128_t))
4623e8e6
LP
241 return -EINVAL;
242
aee36b4e
LP
243 memcpy(&fsuuid, verity->root_hash, sizeof(sd_id128_t));
244 memcpy(&vuuid, (const uint8_t*) verity->root_hash + verity->root_hash_size - sizeof(sd_id128_t), sizeof(sd_id128_t));
4623e8e6 245
aee36b4e 246 if (sd_id128_is_null(fsuuid))
4623e8e6 247 return -EINVAL;
aee36b4e 248 if (sd_id128_is_null(vuuid))
4623e8e6 249 return -EINVAL;
aee36b4e
LP
250
251 /* If the verity data declares it's for the /usr partition, then search for that, in all
252 * other cases assume it's for the root partition. */
253 if (verity->designator == PARTITION_USR) {
254 usr_uuid = fsuuid;
255 usr_verity_uuid = vuuid;
256 } else {
257 root_uuid = fsuuid;
258 root_verity_uuid = vuuid;
259 }
4623e8e6 260 }
8c1be37e 261
8c1be37e
LP
262 b = blkid_new_probe();
263 if (!b)
264 return -ENOMEM;
265
266 errno = 0;
267 r = blkid_probe_set_device(b, fd, 0, 0);
b382db9f 268 if (r != 0)
66855de7 269 return errno_or_else(ENOMEM);
8c1be37e 270
9b6deb03
LP
271 if ((flags & DISSECT_IMAGE_GPT_ONLY) == 0) {
272 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
273 blkid_probe_enable_superblocks(b, 1);
274 blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE|BLKID_SUBLKS_USAGE);
275 }
276
8c1be37e
LP
277 blkid_probe_enable_partitions(b, 1);
278 blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS);
279
280 errno = 0;
281 r = blkid_do_safeprobe(b);
59ba6d0c
LP
282 if (IN_SET(r, -2, 1))
283 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG), "Failed to identify any partition table.");
b382db9f 284 if (r != 0)
66855de7 285 return errno_or_else(EIO);
8c1be37e 286
a4e0d617 287 m = new(DissectedImage, 1);
8c1be37e
LP
288 if (!m)
289 return -ENOMEM;
290
a4e0d617
LP
291 *m = (DissectedImage) {
292 .has_init_system = -1,
293 };
294
369de26f
YW
295 if (image_path) {
296 _cleanup_free_ char *extracted_filename = NULL, *name_stripped = NULL;
297
298 r = path_extract_filename(image_path, &extracted_filename);
299 if (r < 0)
300 return r;
301
302 r = raw_strip_suffixes(extracted_filename, &name_stripped);
593fe6c0
LB
303 if (r < 0)
304 return r;
369de26f
YW
305
306 if (!image_name_is_valid(name_stripped))
307 log_debug("Image name %s is not valid, ignoring.", strna(name_stripped));
308 else
309 m->image_name = TAKE_PTR(name_stripped);
593fe6c0 310 }
369de26f 311
e7cbe5cb 312 if ((!(flags & DISSECT_IMAGE_GPT_ONLY) &&
4b5de5dd 313 (flags & DISSECT_IMAGE_GENERIC_ROOT)) ||
e7cbe5cb 314 (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)) {
9b6deb03 315 const char *usage = NULL;
8c1be37e 316
aee36b4e
LP
317 /* If flags permit this, also allow using non-partitioned single-filesystem images */
318
9b6deb03
LP
319 (void) blkid_probe_lookup_value(b, "USAGE", &usage, NULL);
320 if (STRPTR_IN_SET(usage, "filesystem", "crypto")) {
18d73705 321 _cleanup_free_ char *t = NULL, *n = NULL, *o = NULL;
1b010ae7 322 const char *fstype = NULL, *options = NULL;
8c1be37e 323
9b6deb03
LP
324 /* OK, we have found a file system, that's our root partition then. */
325 (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL);
8c1be37e 326
9b6deb03
LP
327 if (fstype) {
328 t = strdup(fstype);
329 if (!t)
330 return -ENOMEM;
331 }
332
6c544d14
LP
333 n = strdup(devname);
334 if (!n)
335 return -ENOMEM;
336
e7cbe5cb 337 m->single_file_system = true;
c3c88d67
LP
338 m->encrypted = streq_ptr(fstype, "crypto_LUKS");
339
340 m->has_verity = verity && verity->data_path;
341 m->verity_ready = m->has_verity &&
342 verity->root_hash &&
343 (verity->designator < 0 || verity->designator == PARTITION_ROOT);
e7cbe5cb 344
8ee9615e
LP
345 m->has_verity_sig = false; /* signature not embedded, must be specified */
346 m->verity_sig_ready = m->verity_ready &&
347 verity->root_hash_sig;
348
f5215bc8 349 options = mount_options_from_designator(mount_options, PARTITION_ROOT);
18d73705
LB
350 if (options) {
351 o = strdup(options);
352 if (!o)
353 return -ENOMEM;
354 }
355
9b6deb03
LP
356 m->partitions[PARTITION_ROOT] = (DissectedPartition) {
357 .found = true,
e0d53d52 358 .rw = !m->verity_ready && !fstype_is_ro(fstype),
9b6deb03
LP
359 .partno = -1,
360 .architecture = _ARCHITECTURE_INVALID,
1cc6c93a
YW
361 .fstype = TAKE_PTR(t),
362 .node = TAKE_PTR(n),
18d73705 363 .mount_options = TAKE_PTR(o),
88b3300f
LP
364 .offset = 0,
365 .size = UINT64_MAX,
9b6deb03 366 };
8c1be37e 367
1cc6c93a 368 *ret = TAKE_PTR(m);
9b6deb03
LP
369 return 0;
370 }
8c1be37e
LP
371 }
372
373 (void) blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL);
374 if (!pttype)
375 return -ENOPKG;
376
377 is_gpt = streq_ptr(pttype, "gpt");
378 is_mbr = streq_ptr(pttype, "dos");
379
9b6deb03 380 if (!is_gpt && ((flags & DISSECT_IMAGE_GPT_ONLY) || !is_mbr))
8c1be37e
LP
381 return -ENOPKG;
382
0903fd26
LP
383 /* We support external verity data partitions only if the image has no partition table */
384 if (verity && verity->data_path)
385 return -EBADR;
386
4ba86848
LP
387 /* Safety check: refuse block devices that carry a partition table but for which the kernel doesn't
388 * do partition scanning. */
389 r = blockdev_partscan_enabled(fd);
390 if (r < 0)
391 return r;
392 if (r == 0)
393 return -EPROTONOSUPPORT;
394
8c1be37e
LP
395 errno = 0;
396 pl = blkid_probe_get_partitions(b);
b382db9f 397 if (!pl)
66855de7 398 return errno_or_else(ENOMEM);
8c1be37e 399
4ba86848
LP
400 errno = 0;
401 n_partitions = blkid_partlist_numof_partitions(pl);
402 if (n_partitions < 0)
403 return errno_or_else(EIO);
8c1be37e 404
4ba86848 405 for (int i = 0; i < n_partitions; i++) {
1b010ae7 406 _cleanup_free_ char *node = NULL;
9b6deb03 407 unsigned long long pflags;
88b3300f 408 blkid_loff_t start, size;
8c1be37e 409 blkid_partition pp;
8c1be37e
LP
410 int nr;
411
4ba86848
LP
412 errno = 0;
413 pp = blkid_partlist_get_partition(pl, i);
414 if (!pp)
415 return errno_or_else(EIO);
aae22eb3 416
9b6deb03 417 pflags = blkid_partition_get_flags(pp);
8c1be37e 418
4ba86848 419 errno = 0;
8c1be37e
LP
420 nr = blkid_partition_get_partno(pp);
421 if (nr < 0)
4ba86848 422 return errno_or_else(EIO);
8c1be37e 423
88b3300f
LP
424 errno = 0;
425 start = blkid_partition_get_start(pp);
426 if (start < 0)
427 return errno_or_else(EIO);
428
429 assert((uint64_t) start < UINT64_MAX/512);
430
431 errno = 0;
432 size = blkid_partition_get_size(pp);
433 if (size < 0)
434 return errno_or_else(EIO);
435
436 assert((uint64_t) size < UINT64_MAX/512);
437
1b010ae7
LP
438 r = make_partition_devname(devname, nr, &node);
439 if (r < 0)
440 return r;
441
442 /* So here's the thing: after the main ("whole") block device popped up it might take a while
443 * before the kernel fully probed the partition table. Waiting for that to finish is icky in
444 * userspace. So here's what we do instead. We issue the BLKPG_ADD_PARTITION ioctl to add the
445 * partition ourselves, racing against the kernel. Good thing is: if this call fails with
446 * EBUSY then the kernel was quicker than us, and that's totally OK, the outcome is good for
447 * us: the device node will exist. If OTOH our call was successful we won the race. Which is
448 * also good as the outcome is the same: the partition block device exists, and we can use
449 * it.
450 *
451 * Kernel returns EBUSY if there's already a partition by that number or an overlapping
452 * partition already existent. */
453
d25697f5 454 r = block_device_add_partition(fd, node, nr, (uint64_t) start * 512, (uint64_t) size * 512);
1b010ae7
LP
455 if (r < 0) {
456 if (r != -EBUSY)
457 return log_debug_errno(r, "BLKPG_ADD_PARTITION failed: %m");
458
459 log_debug_errno(r, "Kernel was quicker than us in adding partition %i.", nr);
460 } else
461 log_debug("We were quicker than kernel in adding partition %i.", nr);
462
8c1be37e 463 if (is_gpt) {
569a0e42 464 PartitionDesignator designator = _PARTITION_DESIGNATOR_INVALID;
6b41a7b2 465 Architecture architecture = _ARCHITECTURE_INVALID;
08fe0a53 466 const char *stype, *sid, *fstype = NULL, *label;
4623e8e6 467 sd_id128_t type_id, id;
de98f631 468 bool rw = true, growfs = false;
8c1be37e 469
4623e8e6
LP
470 sid = blkid_partition_get_uuid(pp);
471 if (!sid)
472 continue;
473 if (sd_id128_from_string(sid, &id) < 0)
474 continue;
475
8c1be37e
LP
476 stype = blkid_partition_get_type_string(pp);
477 if (!stype)
478 continue;
8c1be37e
LP
479 if (sd_id128_from_string(stype, &type_id) < 0)
480 continue;
481
08fe0a53
LP
482 label = blkid_partition_get_name(pp); /* libblkid returns NULL here if empty */
483
8c1be37e 484 if (sd_id128_equal(type_id, GPT_HOME)) {
a48dd347 485
de98f631 486 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY|GPT_FLAG_GROWFS);
0f7c9a3d 487
a48dd347
LP
488 if (pflags & GPT_FLAG_NO_AUTO)
489 continue;
490
8c1be37e 491 designator = PARTITION_HOME;
9b6deb03 492 rw = !(pflags & GPT_FLAG_READ_ONLY);
de98f631 493 growfs = FLAGS_SET(pflags, GPT_FLAG_GROWFS);
aee36b4e 494
8c1be37e 495 } else if (sd_id128_equal(type_id, GPT_SRV)) {
a48dd347 496
de98f631 497 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY|GPT_FLAG_GROWFS);
0f7c9a3d 498
a48dd347
LP
499 if (pflags & GPT_FLAG_NO_AUTO)
500 continue;
501
8c1be37e 502 designator = PARTITION_SRV;
9b6deb03 503 rw = !(pflags & GPT_FLAG_READ_ONLY);
de98f631 504 growfs = FLAGS_SET(pflags, GPT_FLAG_GROWFS);
aee36b4e 505
8c1be37e 506 } else if (sd_id128_equal(type_id, GPT_ESP)) {
a48dd347 507
aee36b4e
LP
508 /* Note that we don't check the GPT_FLAG_NO_AUTO flag for the ESP, as it is
509 * not defined there. We instead check the GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as
510 * recommended by the UEFI spec (See "12.3.3 Number and Location of System
511 * Partitions"). */
a48dd347
LP
512
513 if (pflags & GPT_FLAG_NO_BLOCK_IO_PROTOCOL)
514 continue;
515
8c1be37e
LP
516 designator = PARTITION_ESP;
517 fstype = "vfat";
a8c47660
LP
518
519 } else if (sd_id128_equal(type_id, GPT_XBOOTLDR)) {
520
de98f631 521 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY|GPT_FLAG_GROWFS);
0f7c9a3d 522
a8c47660
LP
523 if (pflags & GPT_FLAG_NO_AUTO)
524 continue;
525
526 designator = PARTITION_XBOOTLDR;
527 rw = !(pflags & GPT_FLAG_READ_ONLY);
de98f631 528 growfs = FLAGS_SET(pflags, GPT_FLAG_GROWFS);
49ae9d91
DDM
529
530 } else if (gpt_partition_type_is_root(type_id)) {
4623e8e6 531
de98f631 532 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY|GPT_FLAG_GROWFS);
0f7c9a3d 533
a48dd347
LP
534 if (pflags & GPT_FLAG_NO_AUTO)
535 continue;
536
4623e8e6
LP
537 /* If a root ID is specified, ignore everything but the root id */
538 if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id))
539 continue;
540
49ae9d91
DDM
541 assert_se((architecture = gpt_partition_type_uuid_to_arch(type_id)) >= 0);
542 designator = PARTITION_ROOT_OF_ARCH(architecture);
9b6deb03 543 rw = !(pflags & GPT_FLAG_READ_ONLY);
de98f631 544 growfs = FLAGS_SET(pflags, GPT_FLAG_GROWFS);
aee36b4e 545
49ae9d91 546 } else if (gpt_partition_type_is_root_verity(type_id)) {
4623e8e6 547
0f7c9a3d
LP
548 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
549
a48dd347
LP
550 if (pflags & GPT_FLAG_NO_AUTO)
551 continue;
552
c3c88d67 553 m->has_verity = true;
4623e8e6 554
8ee9615e
LP
555 /* If no verity configuration is specified, then don't do verity */
556 if (!verity)
557 continue;
558 if (verity->designator >= 0 && verity->designator != PARTITION_ROOT)
559 continue;
560
561 /* If root hash is specified, then ignore everything but the root id */
562 if (!sd_id128_is_null(root_verity_uuid) && !sd_id128_equal(root_verity_uuid, id))
4623e8e6
LP
563 continue;
564
49ae9d91
DDM
565 assert_se((architecture = gpt_partition_type_uuid_to_arch(type_id)) >= 0);
566 designator = PARTITION_VERITY_OF(PARTITION_ROOT_OF_ARCH(architecture));
4623e8e6 567 fstype = "DM_verity_hash";
4623e8e6 568 rw = false;
8ee9615e 569
49ae9d91 570 } else if (gpt_partition_type_is_root_verity_sig(type_id)) {
8ee9615e
LP
571
572 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
573
574 if (pflags & GPT_FLAG_NO_AUTO)
575 continue;
576
577 m->has_verity_sig = true;
578
579 /* If root hash is specified explicitly, then ignore any embedded signature */
580 if (!verity)
581 continue;
582 if (verity->designator >= 0 && verity->designator != PARTITION_ROOT)
583 continue;
584 if (verity->root_hash)
585 continue;
586
49ae9d91
DDM
587 assert_se((architecture = gpt_partition_type_uuid_to_arch(type_id)) >= 0);
588 designator = PARTITION_VERITY_SIG_OF(PARTITION_ROOT_OF_ARCH(architecture));
8ee9615e 589 fstype = "verity_hash_signature";
4623e8e6 590 rw = false;
8ee9615e 591
49ae9d91 592 } else if (gpt_partition_type_is_usr(type_id)) {
aee36b4e 593
de98f631 594 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY|GPT_FLAG_GROWFS);
aee36b4e
LP
595
596 if (pflags & GPT_FLAG_NO_AUTO)
597 continue;
598
599 /* If a usr ID is specified, ignore everything but the usr id */
600 if (!sd_id128_is_null(usr_uuid) && !sd_id128_equal(usr_uuid, id))
601 continue;
602
49ae9d91
DDM
603 assert_se((architecture = gpt_partition_type_uuid_to_arch(type_id)) >= 0);
604 designator = PARTITION_USR_OF_ARCH(architecture);
aee36b4e 605 rw = !(pflags & GPT_FLAG_READ_ONLY);
de98f631 606 growfs = FLAGS_SET(pflags, GPT_FLAG_GROWFS);
aee36b4e 607
49ae9d91 608 } else if (gpt_partition_type_is_usr_verity(type_id)) {
aee36b4e
LP
609
610 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
611
612 if (pflags & GPT_FLAG_NO_AUTO)
613 continue;
614
c3c88d67 615 m->has_verity = true;
aee36b4e 616
8ee9615e
LP
617 if (!verity)
618 continue;
619 if (verity->designator >= 0 && verity->designator != PARTITION_USR)
620 continue;
621
622 /* If usr hash is specified, then ignore everything but the usr id */
623 if (!sd_id128_is_null(usr_verity_uuid) && !sd_id128_equal(usr_verity_uuid, id))
aee36b4e
LP
624 continue;
625
49ae9d91
DDM
626 assert_se((architecture = gpt_partition_type_uuid_to_arch(type_id)) >= 0);
627 designator = PARTITION_VERITY_OF(PARTITION_USR_OF_ARCH(architecture));
aee36b4e 628 fstype = "DM_verity_hash";
aee36b4e 629 rw = false;
8ee9615e 630
49ae9d91 631 } else if (gpt_partition_type_is_usr_verity_sig(type_id)) {
8ee9615e
LP
632
633 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY);
634
635 if (pflags & GPT_FLAG_NO_AUTO)
636 continue;
637
638 m->has_verity_sig = true;
639
640 /* If usr hash is specified explicitly, then ignore any embedded signature */
641 if (!verity)
642 continue;
643 if (verity->designator >= 0 && verity->designator != PARTITION_USR)
644 continue;
645 if (verity->root_hash)
646 continue;
647
49ae9d91
DDM
648 assert_se((architecture = gpt_partition_type_uuid_to_arch(type_id)) >= 0);
649 designator = PARTITION_VERITY_SIG_OF(PARTITION_USR_OF_ARCH(architecture));
8ee9615e 650 fstype = "verity_hash_signature";
aee36b4e 651 rw = false;
8ee9615e 652
49ae9d91 653 } else if (sd_id128_equal(type_id, GPT_SWAP)) {
a48dd347 654
0f7c9a3d
LP
655 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO);
656
a48dd347
LP
657 if (pflags & GPT_FLAG_NO_AUTO)
658 continue;
659
8c1be37e 660 designator = PARTITION_SWAP;
aee36b4e 661
8c1be37e
LP
662 } else if (sd_id128_equal(type_id, GPT_LINUX_GENERIC)) {
663
de98f631 664 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY|GPT_FLAG_GROWFS);
0f7c9a3d 665
a48dd347
LP
666 if (pflags & GPT_FLAG_NO_AUTO)
667 continue;
668
8c1be37e
LP
669 if (generic_node)
670 multiple_generic = true;
671 else {
672 generic_nr = nr;
9b6deb03 673 generic_rw = !(pflags & GPT_FLAG_READ_ONLY);
de98f631 674 generic_growfs = FLAGS_SET(pflags, GPT_FLAG_GROWFS);
be30ad41 675 generic_uuid = id;
8c1be37e
LP
676 generic_node = strdup(node);
677 if (!generic_node)
678 return -ENOMEM;
679 }
d4dffb85
LP
680
681 } else if (sd_id128_equal(type_id, GPT_TMP)) {
682
de98f631 683 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY|GPT_FLAG_GROWFS);
0f7c9a3d 684
d4dffb85
LP
685 if (pflags & GPT_FLAG_NO_AUTO)
686 continue;
687
688 designator = PARTITION_TMP;
689 rw = !(pflags & GPT_FLAG_READ_ONLY);
de98f631 690 growfs = FLAGS_SET(pflags, GPT_FLAG_GROWFS);
d4dffb85
LP
691
692 } else if (sd_id128_equal(type_id, GPT_VAR)) {
693
de98f631 694 check_partition_flags(node, pflags, GPT_FLAG_NO_AUTO|GPT_FLAG_READ_ONLY|GPT_FLAG_GROWFS);
0f7c9a3d 695
d4dffb85
LP
696 if (pflags & GPT_FLAG_NO_AUTO)
697 continue;
698
699 if (!FLAGS_SET(flags, DISSECT_IMAGE_RELAX_VAR_CHECK)) {
700 sd_id128_t var_uuid;
701
702 /* For /var we insist that the uuid of the partition matches the
703 * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
704 * ID. Why? Unlike the other partitions /var is inherently
705 * installation specific, hence we need to be careful not to mount it
706 * in the wrong installation. By hashing the partition UUID from
707 * /etc/machine-id we can securely bind the partition to the
708 * installation. */
709
710 r = sd_id128_get_machine_app_specific(GPT_VAR, &var_uuid);
711 if (r < 0)
712 return r;
713
714 if (!sd_id128_equal(var_uuid, id)) {
715 log_debug("Found a /var/ partition, but its UUID didn't match our expectations, ignoring.");
716 continue;
717 }
718 }
719
720 designator = PARTITION_VAR;
721 rw = !(pflags & GPT_FLAG_READ_ONLY);
de98f631 722 growfs = FLAGS_SET(pflags, GPT_FLAG_GROWFS);
8c1be37e
LP
723 }
724
725 if (designator != _PARTITION_DESIGNATOR_INVALID) {
0fb5036f 726 _cleanup_free_ char *t = NULL, *o = NULL, *l = NULL;
18d73705 727 const char *options = NULL;
8c1be37e 728
08fe0a53
LP
729 if (m->partitions[designator].found) {
730 /* For most partition types the first one we see wins. Except for the
731 * rootfs and /usr, where we do a version compare of the label, and
732 * let the newest version win. This permits a simple A/B versioning
733 * scheme in OS images. */
734
735 if (!PARTITION_DESIGNATOR_VERSIONED(designator) ||
234c2e16 736 strverscmp_improved(m->partitions[designator].label, label) >= 0)
08fe0a53
LP
737 continue;
738
234c2e16 739 dissected_partition_done(m->partitions + designator);
08fe0a53 740 }
8c1be37e
LP
741
742 if (fstype) {
743 t = strdup(fstype);
744 if (!t)
745 return -ENOMEM;
746 }
747
08fe0a53
LP
748 if (label) {
749 l = strdup(label);
750 if (!l)
751 return -ENOMEM;
752 }
753
f5215bc8 754 options = mount_options_from_designator(mount_options, designator);
18d73705
LB
755 if (options) {
756 o = strdup(options);
757 if (!o)
758 return -ENOMEM;
759 }
760
8c1be37e
LP
761 m->partitions[designator] = (DissectedPartition) {
762 .found = true,
763 .partno = nr,
764 .rw = rw,
de98f631 765 .growfs = growfs,
8c1be37e 766 .architecture = architecture,
0fb5036f 767 .node = TAKE_PTR(node),
1cc6c93a 768 .fstype = TAKE_PTR(t),
08fe0a53 769 .label = TAKE_PTR(l),
be30ad41 770 .uuid = id,
18d73705 771 .mount_options = TAKE_PTR(o),
88b3300f
LP
772 .offset = (uint64_t) start * 512,
773 .size = (uint64_t) size * 512,
8c1be37e 774 };
8c1be37e
LP
775 }
776
777 } else if (is_mbr) {
778
a8c47660 779 switch (blkid_partition_get_type(pp)) {
8c1be37e 780
a8c47660
LP
781 case 0x83: /* Linux partition */
782
783 if (pflags != 0x80) /* Bootable flag */
784 continue;
8c1be37e 785
a8c47660
LP
786 if (generic_node)
787 multiple_generic = true;
788 else {
789 generic_nr = nr;
790 generic_rw = true;
de98f631 791 generic_growfs = false;
a8c47660
LP
792 generic_node = strdup(node);
793 if (!generic_node)
794 return -ENOMEM;
795 }
796
797 break;
798
799 case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
0fb5036f 800 _cleanup_free_ char *o = NULL;
a8c47660 801 sd_id128_t id = SD_ID128_NULL;
18d73705 802 const char *sid, *options = NULL;
a8c47660
LP
803
804 /* First one wins */
234c2e16 805 if (m->partitions[PARTITION_XBOOTLDR].found)
a8c47660
LP
806 continue;
807
808 sid = blkid_partition_get_uuid(pp);
809 if (sid)
810 (void) sd_id128_from_string(sid, &id);
811
f5215bc8 812 options = mount_options_from_designator(mount_options, PARTITION_XBOOTLDR);
18d73705
LB
813 if (options) {
814 o = strdup(options);
815 if (!o)
816 return -ENOMEM;
817 }
818
a8c47660
LP
819 m->partitions[PARTITION_XBOOTLDR] = (DissectedPartition) {
820 .found = true,
821 .partno = nr,
822 .rw = true,
de98f631 823 .growfs = false,
a8c47660 824 .architecture = _ARCHITECTURE_INVALID,
0fb5036f 825 .node = TAKE_PTR(node),
a8c47660 826 .uuid = id,
18d73705 827 .mount_options = TAKE_PTR(o),
88b3300f
LP
828 .offset = (uint64_t) start * 512,
829 .size = (uint64_t) size * 512,
a8c47660
LP
830 };
831
832 break;
833 }}
8c1be37e
LP
834 }
835 }
836
74cb2db9 837 if (m->partitions[PARTITION_ROOT].found) {
49ae9d91
DDM
838 /* If we found the primary arch, then invalidate the secondary and other arch to avoid any
839 * ambiguities, since we never want to mount the secondary or other arch in this case. */
74cb2db9
LP
840 m->partitions[PARTITION_ROOT_SECONDARY].found = false;
841 m->partitions[PARTITION_ROOT_SECONDARY_VERITY].found = false;
8ee9615e 842 m->partitions[PARTITION_ROOT_SECONDARY_VERITY_SIG].found = false;
aee36b4e
LP
843 m->partitions[PARTITION_USR_SECONDARY].found = false;
844 m->partitions[PARTITION_USR_SECONDARY_VERITY].found = false;
8ee9615e 845 m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG].found = false;
8c1be37e 846
49ae9d91
DDM
847 m->partitions[PARTITION_ROOT_OTHER].found = false;
848 m->partitions[PARTITION_ROOT_OTHER_VERITY].found = false;
849 m->partitions[PARTITION_ROOT_OTHER_VERITY_SIG].found = false;
850 m->partitions[PARTITION_USR_OTHER].found = false;
851 m->partitions[PARTITION_USR_OTHER_VERITY].found = false;
852 m->partitions[PARTITION_USR_OTHER_VERITY_SIG].found = false;
853
8ee9615e
LP
854 } else if (m->partitions[PARTITION_ROOT_VERITY].found ||
855 m->partitions[PARTITION_ROOT_VERITY_SIG].found)
7cf66030 856 return -EADDRNOTAVAIL; /* Verity found but no matching rootfs? Something is off, refuse. */
4623e8e6 857
7cf66030 858 else if (m->partitions[PARTITION_ROOT_SECONDARY].found) {
aee36b4e 859
7cf66030 860 /* No root partition found but there's one for the secondary architecture? Then upgrade
49ae9d91
DDM
861 * secondary arch to first and invalidate the other arch. */
862
863 log_debug("No root partition found of the native architecture, falling back to a root "
864 "partition of the secondary architecture.");
4623e8e6 865
7cf66030
LP
866 m->partitions[PARTITION_ROOT] = m->partitions[PARTITION_ROOT_SECONDARY];
867 zero(m->partitions[PARTITION_ROOT_SECONDARY]);
868 m->partitions[PARTITION_ROOT_VERITY] = m->partitions[PARTITION_ROOT_SECONDARY_VERITY];
869 zero(m->partitions[PARTITION_ROOT_SECONDARY_VERITY]);
8ee9615e
LP
870 m->partitions[PARTITION_ROOT_VERITY_SIG] = m->partitions[PARTITION_ROOT_SECONDARY_VERITY_SIG];
871 zero(m->partitions[PARTITION_ROOT_SECONDARY_VERITY_SIG]);
aee36b4e 872
7cf66030
LP
873 m->partitions[PARTITION_USR] = m->partitions[PARTITION_USR_SECONDARY];
874 zero(m->partitions[PARTITION_USR_SECONDARY]);
875 m->partitions[PARTITION_USR_VERITY] = m->partitions[PARTITION_USR_SECONDARY_VERITY];
876 zero(m->partitions[PARTITION_USR_SECONDARY_VERITY]);
8ee9615e
LP
877 m->partitions[PARTITION_USR_VERITY_SIG] = m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG];
878 zero(m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG]);
e0f9e7bd 879
49ae9d91
DDM
880 m->partitions[PARTITION_ROOT_OTHER].found = false;
881 m->partitions[PARTITION_ROOT_OTHER_VERITY].found = false;
882 m->partitions[PARTITION_ROOT_OTHER_VERITY_SIG].found = false;
883 m->partitions[PARTITION_USR_OTHER].found = false;
884 m->partitions[PARTITION_USR_OTHER_VERITY].found = false;
885 m->partitions[PARTITION_USR_OTHER_VERITY_SIG].found = false;
886
8ee9615e
LP
887 } else if (m->partitions[PARTITION_ROOT_SECONDARY_VERITY].found ||
888 m->partitions[PARTITION_ROOT_SECONDARY_VERITY_SIG].found)
7cf66030 889 return -EADDRNOTAVAIL; /* as above */
18d73705 890
49ae9d91
DDM
891 else if (m->partitions[PARTITION_ROOT_OTHER].found) {
892
893 /* No root or secondary partition found but there's one for another architecture? Then
894 * upgrade the other architecture to first. */
895
896 log_debug("No root partition found of the native architecture or the secondary architecture, "
897 "falling back to a root partition of a non-native architecture (%s).",
898 architecture_to_string(m->partitions[PARTITION_ROOT_OTHER].architecture));
899
900 m->partitions[PARTITION_ROOT] = m->partitions[PARTITION_ROOT_OTHER];
901 zero(m->partitions[PARTITION_ROOT_OTHER]);
902 m->partitions[PARTITION_ROOT_VERITY] = m->partitions[PARTITION_ROOT_OTHER_VERITY];
903 zero(m->partitions[PARTITION_ROOT_OTHER_VERITY]);
904 m->partitions[PARTITION_ROOT_VERITY_SIG] = m->partitions[PARTITION_ROOT_OTHER_VERITY_SIG];
905 zero(m->partitions[PARTITION_ROOT_OTHER_VERITY_SIG]);
906
907 m->partitions[PARTITION_USR] = m->partitions[PARTITION_USR_OTHER];
908 zero(m->partitions[PARTITION_USR_OTHER]);
909 m->partitions[PARTITION_USR_VERITY] = m->partitions[PARTITION_USR_OTHER_VERITY];
910 zero(m->partitions[PARTITION_USR_OTHER_VERITY]);
911 m->partitions[PARTITION_USR_VERITY_SIG] = m->partitions[PARTITION_USR_OTHER_VERITY_SIG];
912 zero(m->partitions[PARTITION_USR_OTHER_VERITY_SIG]);
913 }
914
8ee9615e
LP
915 /* Hmm, we found a signature partition but no Verity data? Something is off. */
916 if (m->partitions[PARTITION_ROOT_VERITY_SIG].found && !m->partitions[PARTITION_ROOT_VERITY].found)
917 return -EADDRNOTAVAIL;
7cf66030 918
8ee9615e 919 if (m->partitions[PARTITION_USR].found) {
49ae9d91 920 /* Invalidate secondary and other arch /usr/ if we found the primary arch */
7cf66030
LP
921 m->partitions[PARTITION_USR_SECONDARY].found = false;
922 m->partitions[PARTITION_USR_SECONDARY_VERITY].found = false;
8ee9615e 923 m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG].found = false;
7cf66030 924
49ae9d91
DDM
925 m->partitions[PARTITION_USR_OTHER].found = false;
926 m->partitions[PARTITION_USR_OTHER_VERITY].found = false;
927 m->partitions[PARTITION_USR_OTHER_VERITY_SIG].found = false;
928
8ee9615e
LP
929 } else if (m->partitions[PARTITION_USR_VERITY].found ||
930 m->partitions[PARTITION_USR_VERITY_SIG].found)
7cf66030 931 return -EADDRNOTAVAIL; /* as above */
8c1be37e 932
7cf66030 933 else if (m->partitions[PARTITION_USR_SECONDARY].found) {
e0f9e7bd 934
49ae9d91
DDM
935 log_debug("No usr partition found of the native architecture, falling back to a usr "
936 "partition of the secondary architecture.");
937
7cf66030
LP
938 /* Upgrade secondary arch to primary */
939 m->partitions[PARTITION_USR] = m->partitions[PARTITION_USR_SECONDARY];
940 zero(m->partitions[PARTITION_USR_SECONDARY]);
941 m->partitions[PARTITION_USR_VERITY] = m->partitions[PARTITION_USR_SECONDARY_VERITY];
942 zero(m->partitions[PARTITION_USR_SECONDARY_VERITY]);
8ee9615e
LP
943 m->partitions[PARTITION_USR_VERITY_SIG] = m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG];
944 zero(m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG]);
7cf66030 945
49ae9d91
DDM
946 m->partitions[PARTITION_USR_OTHER].found = false;
947 m->partitions[PARTITION_USR_OTHER_VERITY].found = false;
948 m->partitions[PARTITION_USR_OTHER_VERITY_SIG].found = false;
949
8ee9615e
LP
950 } else if (m->partitions[PARTITION_USR_SECONDARY_VERITY].found ||
951 m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG].found)
7cf66030
LP
952 return -EADDRNOTAVAIL; /* as above */
953
49ae9d91
DDM
954 else if (m->partitions[PARTITION_USR_OTHER].found) {
955
956 log_debug("No usr partition found of the native architecture or the secondary architecture, "
957 "falling back to a usr partition of a non-native architecture (%s).",
958 architecture_to_string(m->partitions[PARTITION_ROOT_OTHER].architecture));
959
960 /* Upgrade other arch to primary */
961 m->partitions[PARTITION_USR] = m->partitions[PARTITION_USR_OTHER];
962 zero(m->partitions[PARTITION_USR_OTHER]);
963 m->partitions[PARTITION_USR_VERITY] = m->partitions[PARTITION_USR_OTHER_VERITY];
964 zero(m->partitions[PARTITION_USR_OTHER_VERITY]);
965 m->partitions[PARTITION_USR_VERITY_SIG] = m->partitions[PARTITION_USR_OTHER_VERITY_SIG];
966 zero(m->partitions[PARTITION_USR_OTHER_VERITY_SIG]);
967 }
968
8ee9615e
LP
969 /* Hmm, we found a signature partition but no Verity data? Something is off. */
970 if (m->partitions[PARTITION_USR_VERITY_SIG].found && !m->partitions[PARTITION_USR_VERITY].found)
971 return -EADDRNOTAVAIL;
972
cb241a69
LP
973 /* If root and /usr are combined then insist that the architecture matches */
974 if (m->partitions[PARTITION_ROOT].found &&
975 m->partitions[PARTITION_USR].found &&
976 (m->partitions[PARTITION_ROOT].architecture >= 0 &&
977 m->partitions[PARTITION_USR].architecture >= 0 &&
978 m->partitions[PARTITION_ROOT].architecture != m->partitions[PARTITION_USR].architecture))
979 return -EADDRNOTAVAIL;
980
4ab51780
LP
981 if (!m->partitions[PARTITION_ROOT].found &&
982 !m->partitions[PARTITION_USR].found &&
983 (flags & DISSECT_IMAGE_GENERIC_ROOT) &&
00a8b34f 984 (!verity || !verity->root_hash || verity->designator != PARTITION_USR)) {
7cf66030 985
1b010ae7 986 /* OK, we found nothing usable, then check if there's a single generic partition, and use
4b5de5dd
LP
987 * that. If the root hash was set however, then we won't fall back to a generic node, because
988 * the root hash decides. */
7cf66030
LP
989
990 /* If we didn't find a properly marked root partition, but we did find a single suitable
991 * generic Linux partition, then use this as root partition, if the caller asked for it. */
992 if (multiple_generic)
993 return -ENOTUNIQ;
994
4b5de5dd
LP
995 /* If we didn't find a generic node, then we can't fix this up either */
996 if (generic_node) {
997 _cleanup_free_ char *o = NULL;
998 const char *options;
8c1be37e 999
f5215bc8 1000 options = mount_options_from_designator(mount_options, PARTITION_ROOT);
18d73705
LB
1001 if (options) {
1002 o = strdup(options);
1003 if (!o)
1004 return -ENOMEM;
1005 }
1006
1f8fb21c 1007 assert(generic_nr >= 0);
8c1be37e
LP
1008 m->partitions[PARTITION_ROOT] = (DissectedPartition) {
1009 .found = true,
1010 .rw = generic_rw,
de98f631 1011 .growfs = generic_growfs,
8c1be37e
LP
1012 .partno = generic_nr,
1013 .architecture = _ARCHITECTURE_INVALID,
1cc6c93a 1014 .node = TAKE_PTR(generic_node),
be30ad41 1015 .uuid = generic_uuid,
18d73705 1016 .mount_options = TAKE_PTR(o),
88b3300f
LP
1017 .offset = UINT64_MAX,
1018 .size = UINT64_MAX,
8c1be37e 1019 };
e0f9e7bd 1020 }
8c1be37e
LP
1021 }
1022
4b5de5dd
LP
1023 /* 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 */
1024 if (FLAGS_SET(flags, DISSECT_IMAGE_REQUIRE_ROOT) &&
1025 !(m->partitions[PARTITION_ROOT].found || (m->partitions[PARTITION_USR].found && FLAGS_SET(flags, DISSECT_IMAGE_USR_NO_ROOT))))
1026 return -ENXIO;
1027
7b32164f
LP
1028 if (m->partitions[PARTITION_ROOT_VERITY].found) {
1029 /* We only support one verity partition per image, i.e. can't do for both /usr and root fs */
1030 if (m->partitions[PARTITION_USR_VERITY].found)
1031 return -ENOTUNIQ;
1032
1033 /* We don't support verity enabled root with a split out /usr. Neither with nor without
1034 * verity there. (Note that we do support verity-less root with verity-full /usr, though.) */
1035 if (m->partitions[PARTITION_USR].found)
1036 return -EADDRNOTAVAIL;
1037 }
aee36b4e 1038
1903defc
LP
1039 if (verity) {
1040 /* If a verity designator is specified, then insist that the matching partition exists */
1041 if (verity->designator >= 0 && !m->partitions[verity->designator].found)
1042 return -EADDRNOTAVAIL;
aee36b4e 1043
1903defc 1044 if (verity->root_hash) {
8ee9615e
LP
1045 /* If we have an explicit root hash and found the partitions for it, then we are ready to use
1046 * Verity, set things up for it */
1047
1903defc
LP
1048 if (verity->designator < 0 || verity->designator == PARTITION_ROOT) {
1049 if (!m->partitions[PARTITION_ROOT_VERITY].found || !m->partitions[PARTITION_ROOT].found)
1050 return -EADDRNOTAVAIL;
4623e8e6 1051
1903defc
LP
1052 /* If we found a verity setup, then the root partition is necessarily read-only. */
1053 m->partitions[PARTITION_ROOT].rw = false;
1054 m->verity_ready = true;
1903defc 1055
f9e0bb61
LP
1056 } else {
1057 assert(verity->designator == PARTITION_USR);
1058
1903defc
LP
1059 if (!m->partitions[PARTITION_USR_VERITY].found || !m->partitions[PARTITION_USR].found)
1060 return -EADDRNOTAVAIL;
4623e8e6 1061
1903defc
LP
1062 m->partitions[PARTITION_USR].rw = false;
1063 m->verity_ready = true;
1064 }
8ee9615e
LP
1065
1066 if (m->verity_ready)
b98416e1 1067 m->verity_sig_ready = verity->root_hash_sig;
8ee9615e
LP
1068
1069 } else if (m->partitions[verity->designator == PARTITION_USR ? PARTITION_USR_VERITY_SIG : PARTITION_ROOT_VERITY_SIG].found) {
1070
1071 /* If we found an embedded signature partition, we are ready, too. */
1072
1073 m->verity_ready = m->verity_sig_ready = true;
1074 m->partitions[verity->designator == PARTITION_USR ? PARTITION_USR : PARTITION_ROOT].rw = false;
aee36b4e 1075 }
4623e8e6
LP
1076 }
1077
18b5886e
LP
1078 blkid_free_probe(b);
1079 b = NULL;
1080
8c1be37e 1081 /* Fill in file system types if we don't know them yet. */
569a0e42 1082 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
18b5886e 1083 DissectedPartition *p = m->partitions + i;
8c1be37e 1084
18b5886e 1085 if (!p->found)
8c1be37e
LP
1086 continue;
1087
18b5886e
LP
1088 if (!p->fstype && p->node) {
1089 r = probe_filesystem(p->node, &p->fstype);
7cc84b2c 1090 if (r < 0 && r != -EUCLEAN)
18b5886e 1091 return r;
8c1be37e
LP
1092 }
1093
18b5886e
LP
1094 if (streq_ptr(p->fstype, "crypto_LUKS"))
1095 m->encrypted = true;
896f937f
LP
1096
1097 if (p->fstype && fstype_is_ro(p->fstype))
1098 p->rw = false;
de98f631
LP
1099
1100 if (!p->rw)
1101 p->growfs = false;
8c1be37e
LP
1102 }
1103
1cc6c93a 1104 *ret = TAKE_PTR(m);
8c1be37e
LP
1105 return 0;
1106#else
1107 return -EOPNOTSUPP;
1108#endif
1109}
1110
1111DissectedImage* dissected_image_unref(DissectedImage *m) {
8c1be37e
LP
1112 if (!m)
1113 return NULL;
1114
08fe0a53 1115 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++)
234c2e16 1116 dissected_partition_done(m->partitions + i);
8c1be37e 1117
593fe6c0 1118 free(m->image_name);
3b925504
LP
1119 free(m->hostname);
1120 strv_free(m->machine_info);
1121 strv_free(m->os_release);
7718ac97 1122 strv_free(m->extension_release);
3b925504 1123
5fecf46d 1124 return mfree(m);
8c1be37e
LP
1125}
1126
18b5886e 1127static int is_loop_device(const char *path) {
553e15f2 1128 char s[SYS_BLOCK_PATH_MAX("/../loop/")];
18b5886e
LP
1129 struct stat st;
1130
1131 assert(path);
1132
1133 if (stat(path, &st) < 0)
1134 return -errno;
1135
1136 if (!S_ISBLK(st.st_mode))
1137 return -ENOTBLK;
1138
553e15f2 1139 xsprintf_sys_block_path(s, "/loop/", st.st_dev);
18b5886e
LP
1140 if (access(s, F_OK) < 0) {
1141 if (errno != ENOENT)
1142 return -errno;
1143
1144 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
553e15f2 1145 xsprintf_sys_block_path(s, "/../loop/", st.st_dev);
18b5886e
LP
1146 if (access(s, F_OK) < 0)
1147 return errno == ENOENT ? false : -errno;
1148 }
1149
1150 return true;
1151}
1152
cf32c486
LP
1153static int run_fsck(const char *node, const char *fstype) {
1154 int r, exit_status;
1155 pid_t pid;
1156
1157 assert(node);
1158 assert(fstype);
1159
1160 r = fsck_exists(fstype);
1161 if (r < 0) {
1162 log_debug_errno(r, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype);
1163 return 0;
1164 }
1165 if (r == 0) {
1166 log_debug("Not checking partition %s, as fsck for %s does not exist.", node, fstype);
1167 return 0;
1168 }
1169
1170 r = safe_fork("(fsck)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG|FORK_NULL_STDIO, &pid);
1171 if (r < 0)
1172 return log_debug_errno(r, "Failed to fork off fsck: %m");
1173 if (r == 0) {
1174 /* Child */
1175 execl("/sbin/fsck", "/sbin/fsck", "-aT", node, NULL);
7e0ed2e9 1176 log_open();
cf32c486
LP
1177 log_debug_errno(errno, "Failed to execl() fsck: %m");
1178 _exit(FSCK_OPERATIONAL_ERROR);
1179 }
1180
1181 exit_status = wait_for_terminate_and_check("fsck", pid, 0);
1182 if (exit_status < 0)
1183 return log_debug_errno(exit_status, "Failed to fork off /sbin/fsck: %m");
1184
1185 if ((exit_status & ~FSCK_ERROR_CORRECTED) != FSCK_SUCCESS) {
1186 log_debug("fsck failed with exit status %i.", exit_status);
1187
1188 if ((exit_status & (FSCK_SYSTEM_SHOULD_REBOOT|FSCK_ERRORS_LEFT_UNCORRECTED)) != 0)
1189 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN), "File system is corrupted, refusing.");
1190
1191 log_debug("Ignoring fsck error.");
1192 }
1193
1194 return 0;
1195}
1196
81939d9d
LP
1197static int fs_grow(const char *node_path, const char *mount_path) {
1198 _cleanup_close_ int mount_fd = -1, node_fd = -1;
81939d9d
LP
1199 uint64_t size, newsize;
1200 int r;
1201
1202 node_fd = open(node_path, O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY);
1203 if (node_fd < 0)
1204 return log_debug_errno(errno, "Failed to open node device %s: %m", node_path);
1205
1206 if (ioctl(node_fd, BLKGETSIZE64, &size) != 0)
1207 return log_debug_errno(errno, "Failed to get block device size of %s: %m", node_path);
1208
1209 mount_fd = open(mount_path, O_RDONLY|O_DIRECTORY|O_CLOEXEC);
1210 if (mount_fd < 0)
1211 return log_debug_errno(errno, "Failed to open mountd file system %s: %m", mount_path);
1212
1213 log_debug("Resizing \"%s\" to %"PRIu64" bytes...", mount_path, size);
1214 r = resize_fs(mount_fd, size, &newsize);
1215 if (r < 0)
1216 return log_debug_errno(r, "Failed to resize \"%s\" to %"PRIu64" bytes: %m", mount_path, size);
1217
1218 if (newsize == size)
1219 log_debug("Successfully resized \"%s\" to %s bytes.",
2b59bf51 1220 mount_path, FORMAT_BYTES(newsize));
81939d9d
LP
1221 else {
1222 assert(newsize < size);
1223 log_debug("Successfully resized \"%s\" to %s bytes (%"PRIu64" bytes lost due to blocksize).",
2b59bf51 1224 mount_path, FORMAT_BYTES(newsize), size - newsize);
81939d9d
LP
1225 }
1226
1227 return 0;
1228}
1229
18b5886e
LP
1230static int mount_partition(
1231 DissectedPartition *m,
1232 const char *where,
1233 const char *directory,
2d3a5a73 1234 uid_t uid_shift,
21b61b1d 1235 uid_t uid_range,
18b5886e
LP
1236 DissectImageFlags flags) {
1237
2d3a5a73
LP
1238 _cleanup_free_ char *chased = NULL, *options = NULL;
1239 const char *p, *node, *fstype;
21b61b1d 1240 bool rw, remap_uid_gid = false;
2eedfd2d 1241 int r;
8c1be37e
LP
1242
1243 assert(m);
1244 assert(where);
1245
4dc28665 1246 /* Use decrypted node and matching fstype if available, otherwise use the original device */
18b5886e 1247 node = m->decrypted_node ?: m->node;
4dc28665 1248 fstype = m->decrypted_node ? m->decrypted_fstype: m->fstype;
18b5886e 1249
4dc28665 1250 if (!m->found || !node)
8c1be37e 1251 return 0;
4dc28665
LP
1252 if (!fstype)
1253 return -EAFNOSUPPORT;
8c1be37e 1254
68ac5118
ZJS
1255 /* We are looking at an encrypted partition? This either means stacked encryption, or the caller
1256 * didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error for this
1257 * case. */
4dc28665 1258 if (streq(fstype, "crypto_LUKS"))
fa45d12c 1259 return -EUNATCH;
18b5886e 1260
ef9c184d 1261 rw = m->rw && !(flags & DISSECT_IMAGE_MOUNT_READ_ONLY);
8c1be37e 1262
cf32c486
LP
1263 if (FLAGS_SET(flags, DISSECT_IMAGE_FSCK) && rw) {
1264 r = run_fsck(node, fstype);
1265 if (r < 0)
1266 return r;
1267 }
1268
2eedfd2d 1269 if (directory) {
334eb5b0
LP
1270 /* Automatically create missing mount points inside the image, if necessary. */
1271 r = mkdir_p_root(where, directory, uid_shift, (gid_t) uid_shift, 0755);
1272 if (r < 0 && r != -EROFS)
1273 return r;
1f0f82f1 1274
a5648b80 1275 r = chase_symlinks(directory, where, CHASE_PREFIX_ROOT, &chased, NULL);
2eedfd2d
LP
1276 if (r < 0)
1277 return r;
1278
1279 p = chased;
9842905e
LP
1280 } else {
1281 /* Create top-level mount if missing – but only if this is asked for. This won't modify the
1282 * image (as the branch above does) but the host hierarchy, and the created directory might
1283 * survive our mount in the host hierarchy hence. */
1284 if (FLAGS_SET(flags, DISSECT_IMAGE_MKDIR)) {
1285 r = mkdir_p(where, 0755);
1286 if (r < 0)
1287 return r;
1288 }
1289
8c1be37e 1290 p = where;
9842905e 1291 }
8c1be37e 1292
18b5886e 1293 /* If requested, turn on discard support. */
154d2269 1294 if (fstype_can_discard(fstype) &&
18b5886e 1295 ((flags & DISSECT_IMAGE_DISCARD) ||
3afda7c7 1296 ((flags & DISSECT_IMAGE_DISCARD_ON_LOOP) && is_loop_device(m->node) > 0))) {
2d3a5a73
LP
1297 options = strdup("discard");
1298 if (!options)
1299 return -ENOMEM;
1300 }
1301
21b61b1d 1302 if (uid_is_valid(uid_shift) && uid_shift != 0) {
2d3a5a73 1303
21b61b1d
LP
1304 if (fstype_can_uid_gid(fstype)) {
1305 _cleanup_free_ char *uid_option = NULL;
2d3a5a73 1306
21b61b1d
LP
1307 if (asprintf(&uid_option, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0)
1308 return -ENOMEM;
1309
1310 if (!strextend_with_separator(&options, ",", uid_option))
1311 return -ENOMEM;
1312 } else if (FLAGS_SET(flags, DISSECT_IMAGE_MOUNT_IDMAPPED))
1313 remap_uid_gid = true;
2d3a5a73 1314 }
8c1be37e 1315
18d73705 1316 if (!isempty(m->mount_options))
c2bc710b 1317 if (!strextend_with_separator(&options, ",", m->mount_options))
18d73705
LB
1318 return -ENOMEM;
1319
b620bf33
LP
1320 /* So, when you request MS_RDONLY from ext4, then this means nothing. It happily still writes to the
1321 * backing storage. What's worse, the BLKRO[GS]ET flag and (in case of loopback devices)
1322 * LO_FLAGS_READ_ONLY don't mean anything, they affect userspace accesses only, and write accesses
1323 * from the upper file system still get propagated through to the underlying file system,
1324 * unrestricted. To actually get ext4/xfs/btrfs to stop writing to the device we need to specify
1325 * "norecovery" as mount option, in addition to MS_RDONLY. Yes, this sucks, since it means we need to
1326 * carry a per file system table here.
1327 *
1328 * Note that this means that we might not be able to mount corrupted file systems as read-only
1329 * anymore (since in some cases the kernel implementations will refuse mounting when corrupted,
1330 * read-only and "norecovery" is specified). But I think for the case of automatically determined
1331 * mount options for loopback devices this is the right choice, since otherwise using the same
1332 * loopback file twice even in read-only mode, is going to fail badly sooner or later. The usecase of
1333 * making reuse of the immutable images "just work" is more relevant to us than having read-only
1334 * access that actually modifies stuff work on such image files. Or to say this differently: if
1335 * people want their file systems to be fixed up they should just open them in writable mode, where
1336 * all these problems don't exist. */
1337 if (!rw && STRPTR_IN_SET(fstype, "ext3", "ext4", "xfs", "btrfs"))
1338 if (!strextend_with_separator(&options, ",", "norecovery"))
1339 return -ENOMEM;
1340
511a8cfe 1341 r = mount_nofollow_verbose(LOG_DEBUG, node, p, fstype, MS_NODEV|(rw ? 0 : MS_RDONLY), options);
d9223c07
LP
1342 if (r < 0)
1343 return r;
1344
81939d9d
LP
1345 if (rw && m->growfs && FLAGS_SET(flags, DISSECT_IMAGE_GROWFS))
1346 (void) fs_grow(node, p);
1347
21b61b1d 1348 if (remap_uid_gid) {
2b2777ed 1349 r = remount_idmap(p, uid_shift, uid_range, UID_INVALID, REMOUNT_IDMAPPING_HOST_ROOT);
21b61b1d
LP
1350 if (r < 0)
1351 return r;
1352 }
1353
d9223c07 1354 return 1;
8c1be37e
LP
1355}
1356
7cf66030
LP
1357static int mount_root_tmpfs(const char *where, uid_t uid_shift, DissectImageFlags flags) {
1358 _cleanup_free_ char *options = NULL;
1359 int r;
1360
1361 assert(where);
1362
1363 /* For images that contain /usr/ but no rootfs, let's mount rootfs as tmpfs */
1364
1365 if (FLAGS_SET(flags, DISSECT_IMAGE_MKDIR)) {
1366 r = mkdir_p(where, 0755);
1367 if (r < 0)
1368 return r;
1369 }
1370
1371 if (uid_is_valid(uid_shift)) {
1372 if (asprintf(&options, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0)
1373 return -ENOMEM;
1374 }
1375
1376 r = mount_nofollow_verbose(LOG_DEBUG, "rootfs", where, "tmpfs", MS_NODEV, options);
1377 if (r < 0)
1378 return r;
1379
1380 return 1;
1381}
1382
21b61b1d
LP
1383int dissected_image_mount(
1384 DissectedImage *m,
1385 const char *where,
1386 uid_t uid_shift,
1387 uid_t uid_range,
1388 DissectImageFlags flags) {
1389
1f0f82f1 1390 int r, xbootldr_mounted;
8c1be37e
LP
1391
1392 assert(m);
1393 assert(where);
1394
fa45d12c
LP
1395 /* Returns:
1396 *
1397 * -ENXIO → No root partition found
7718ac97 1398 * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release/extension-release file found
fa45d12c
LP
1399 * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet
1400 * -EUCLEAN → fsck for file system failed
1401 * -EBUSY → File system already mounted/used elsewhere (kernel)
4dc28665 1402 * -EAFNOSUPPORT → File system type not supported or not known
fa45d12c
LP
1403 */
1404
7cf66030
LP
1405 if (!(m->partitions[PARTITION_ROOT].found ||
1406 (m->partitions[PARTITION_USR].found && FLAGS_SET(flags, DISSECT_IMAGE_USR_NO_ROOT))))
1407 return -ENXIO; /* Require a root fs or at least a /usr/ fs (the latter is subject to a flag of its own) */
8c1be37e 1408
2d3a5a73 1409 if ((flags & DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY) == 0) {
7cf66030
LP
1410
1411 /* First mount the root fs. If there's none we use a tmpfs. */
1412 if (m->partitions[PARTITION_ROOT].found)
21b61b1d 1413 r = mount_partition(m->partitions + PARTITION_ROOT, where, NULL, uid_shift, uid_range, flags);
7cf66030
LP
1414 else
1415 r = mount_root_tmpfs(where, uid_shift, flags);
2d3a5a73
LP
1416 if (r < 0)
1417 return r;
aee36b4e 1418
aee36b4e 1419 /* For us mounting root always means mounting /usr as well */
21b61b1d 1420 r = mount_partition(m->partitions + PARTITION_USR, where, "/usr", uid_shift, uid_range, flags);
aee36b4e
LP
1421 if (r < 0)
1422 return r;
03bcb6d4 1423
9ccb531a
LB
1424 if ((flags & (DISSECT_IMAGE_VALIDATE_OS|DISSECT_IMAGE_VALIDATE_OS_EXT)) != 0) {
1425 /* If either one of the validation flags are set, ensure that the image qualifies
1426 * as one or the other (or both). */
1427 bool ok = false;
1428
1429 if (FLAGS_SET(flags, DISSECT_IMAGE_VALIDATE_OS)) {
1430 r = path_is_os_tree(where);
1431 if (r < 0)
1432 return r;
1433 if (r > 0)
1434 ok = true;
1435 }
1436 if (!ok && FLAGS_SET(flags, DISSECT_IMAGE_VALIDATE_OS_EXT)) {
7718ac97
LB
1437 r = path_is_extension_tree(where, m->image_name);
1438 if (r < 0)
1439 return r;
9ccb531a
LB
1440 if (r > 0)
1441 ok = true;
7718ac97 1442 }
9ccb531a
LB
1443
1444 if (!ok)
1445 return -ENOMEDIUM;
03bcb6d4 1446 }
2d3a5a73
LP
1447 }
1448
705727fd 1449 if (flags & DISSECT_IMAGE_MOUNT_ROOT_ONLY)
2d3a5a73 1450 return 0;
8c1be37e 1451
21b61b1d 1452 r = mount_partition(m->partitions + PARTITION_HOME, where, "/home", uid_shift, uid_range, flags);
8c1be37e
LP
1453 if (r < 0)
1454 return r;
1455
21b61b1d 1456 r = mount_partition(m->partitions + PARTITION_SRV, where, "/srv", uid_shift, uid_range, flags);
8c1be37e
LP
1457 if (r < 0)
1458 return r;
1459
21b61b1d 1460 r = mount_partition(m->partitions + PARTITION_VAR, where, "/var", uid_shift, uid_range, flags);
d4dffb85
LP
1461 if (r < 0)
1462 return r;
1463
21b61b1d 1464 r = mount_partition(m->partitions + PARTITION_TMP, where, "/var/tmp", uid_shift, uid_range, flags);
d4dffb85
LP
1465 if (r < 0)
1466 return r;
1467
21b61b1d 1468 xbootldr_mounted = mount_partition(m->partitions + PARTITION_XBOOTLDR, where, "/boot", uid_shift, uid_range, flags);
1f0f82f1
LP
1469 if (xbootldr_mounted < 0)
1470 return xbootldr_mounted;
d9223c07 1471
8c1be37e 1472 if (m->partitions[PARTITION_ESP].found) {
1f0f82f1
LP
1473 int esp_done = false;
1474
d9223c07
LP
1475 /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it
1476 * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */
8c1be37e 1477
a5648b80 1478 r = chase_symlinks("/efi", where, CHASE_PREFIX_ROOT, NULL, NULL);
1f0f82f1
LP
1479 if (r < 0) {
1480 if (r != -ENOENT)
d9223c07 1481 return r;
8c1be37e 1482
1f0f82f1
LP
1483 /* /efi doesn't exist. Let's see if /boot is suitable then */
1484
1485 if (!xbootldr_mounted) {
1486 _cleanup_free_ char *p = NULL;
2eedfd2d 1487
1f0f82f1
LP
1488 r = chase_symlinks("/boot", where, CHASE_PREFIX_ROOT, &p, NULL);
1489 if (r < 0) {
1490 if (r != -ENOENT)
1491 return r;
db55bbf2 1492 } else if (dir_is_empty(p, /* ignore_hidden_or_backup= */ false) > 0) {
1f0f82f1 1493 /* It exists and is an empty directory. Let's mount the ESP there. */
21b61b1d 1494 r = mount_partition(m->partitions + PARTITION_ESP, where, "/boot", uid_shift, uid_range, flags);
1f0f82f1
LP
1495 if (r < 0)
1496 return r;
1497
1498 esp_done = true;
1499 }
2eedfd2d 1500 }
8c1be37e 1501 }
1f0f82f1
LP
1502
1503 if (!esp_done) {
1504 /* OK, let's mount the ESP now to /efi (possibly creating the dir if missing) */
1505
21b61b1d 1506 r = mount_partition(m->partitions + PARTITION_ESP, where, "/efi", uid_shift, uid_range, flags);
1f0f82f1
LP
1507 if (r < 0)
1508 return r;
1509 }
8c1be37e
LP
1510 }
1511
1512 return 0;
1513}
1514
21b61b1d
LP
1515int dissected_image_mount_and_warn(
1516 DissectedImage *m,
1517 const char *where,
1518 uid_t uid_shift,
1519 uid_t uid_range,
1520 DissectImageFlags flags) {
1521
af187ab2
LP
1522 int r;
1523
1524 assert(m);
1525 assert(where);
1526
21b61b1d 1527 r = dissected_image_mount(m, where, uid_shift, uid_range, flags);
af187ab2
LP
1528 if (r == -ENXIO)
1529 return log_error_errno(r, "Not root file system found in image.");
1530 if (r == -EMEDIUMTYPE)
7718ac97 1531 return log_error_errno(r, "No suitable os-release/extension-release file in image found.");
af187ab2
LP
1532 if (r == -EUNATCH)
1533 return log_error_errno(r, "Encrypted file system discovered, but decryption not requested.");
1534 if (r == -EUCLEAN)
1535 return log_error_errno(r, "File system check on image failed.");
1536 if (r == -EBUSY)
1537 return log_error_errno(r, "File system already mounted elsewhere.");
4dc28665
LP
1538 if (r == -EAFNOSUPPORT)
1539 return log_error_errno(r, "File system type not supported or not known.");
af187ab2
LP
1540 if (r < 0)
1541 return log_error_errno(r, "Failed to mount image: %m");
1542
1543 return r;
1544}
1545
349cc4a5 1546#if HAVE_LIBCRYPTSETUP
18b5886e
LP
1547typedef struct DecryptedPartition {
1548 struct crypt_device *device;
1549 char *name;
1550 bool relinquished;
1551} DecryptedPartition;
1552
1553struct DecryptedImage {
1554 DecryptedPartition *decrypted;
1555 size_t n_decrypted;
18b5886e
LP
1556};
1557#endif
1558
1559DecryptedImage* decrypted_image_unref(DecryptedImage* d) {
349cc4a5 1560#if HAVE_LIBCRYPTSETUP
18b5886e
LP
1561 int r;
1562
1563 if (!d)
1564 return NULL;
1565
67f63ee5 1566 for (size_t i = 0; i < d->n_decrypted; i++) {
18b5886e
LP
1567 DecryptedPartition *p = d->decrypted + i;
1568
1569 if (p->device && p->name && !p->relinquished) {
0d12936d 1570 r = sym_crypt_deactivate_by_name(p->device, p->name, 0);
18b5886e
LP
1571 if (r < 0)
1572 log_debug_errno(r, "Failed to deactivate encrypted partition %s", p->name);
1573 }
1574
1575 if (p->device)
0d12936d 1576 sym_crypt_free(p->device);
18b5886e
LP
1577 free(p->name);
1578 }
1579
f91861e4 1580 free(d->decrypted);
18b5886e
LP
1581 free(d);
1582#endif
1583 return NULL;
1584}
1585
349cc4a5 1586#if HAVE_LIBCRYPTSETUP
4623e8e6
LP
1587
1588static int make_dm_name_and_node(const void *original_node, const char *suffix, char **ret_name, char **ret_node) {
1589 _cleanup_free_ char *name = NULL, *node = NULL;
1590 const char *base;
1591
1592 assert(original_node);
1593 assert(suffix);
1594 assert(ret_name);
1595 assert(ret_node);
1596
1597 base = strrchr(original_node, '/');
1598 if (!base)
ac1f3ad0
LB
1599 base = original_node;
1600 else
1601 base++;
4623e8e6
LP
1602 if (isempty(base))
1603 return -EINVAL;
1604
1605 name = strjoin(base, suffix);
1606 if (!name)
1607 return -ENOMEM;
1608 if (!filename_is_valid(name))
1609 return -EINVAL;
1610
0d12936d 1611 node = path_join(sym_crypt_get_dir(), name);
4623e8e6
LP
1612 if (!node)
1613 return -ENOMEM;
1614
1cc6c93a
YW
1615 *ret_name = TAKE_PTR(name);
1616 *ret_node = TAKE_PTR(node);
4623e8e6 1617
4623e8e6
LP
1618 return 0;
1619}
1620
18b5886e
LP
1621static int decrypt_partition(
1622 DissectedPartition *m,
1623 const char *passphrase,
1624 DissectImageFlags flags,
1625 DecryptedImage *d) {
1626
1627 _cleanup_free_ char *node = NULL, *name = NULL;
0d12936d 1628 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
18b5886e
LP
1629 int r;
1630
1631 assert(m);
1632 assert(d);
1633
1634 if (!m->found || !m->node || !m->fstype)
1635 return 0;
1636
1637 if (!streq(m->fstype, "crypto_LUKS"))
1638 return 0;
1639
bdd73ac5
ZJS
1640 if (!passphrase)
1641 return -ENOKEY;
1642
0d12936d
LP
1643 r = dlopen_cryptsetup();
1644 if (r < 0)
1645 return r;
1646
4623e8e6
LP
1647 r = make_dm_name_and_node(m->node, "-decrypted", &name, &node);
1648 if (r < 0)
1649 return r;
18b5886e 1650
319a4f4b 1651 if (!GREEDY_REALLOC0(d->decrypted, d->n_decrypted + 1))
18b5886e
LP
1652 return -ENOMEM;
1653
0d12936d 1654 r = sym_crypt_init(&cd, m->node);
18b5886e 1655 if (r < 0)
715cbb81 1656 return log_debug_errno(r, "Failed to initialize dm-crypt: %m");
18b5886e 1657
efc3b12f 1658 cryptsetup_enable_logging(cd);
1887032f 1659
0d12936d 1660 r = sym_crypt_load(cd, CRYPT_LUKS, NULL);
294bd454
ZJS
1661 if (r < 0)
1662 return log_debug_errno(r, "Failed to load LUKS metadata: %m");
18b5886e 1663
0d12936d 1664 r = sym_crypt_activate_by_passphrase(cd, name, CRYPT_ANY_SLOT, passphrase, strlen(passphrase),
ef9c184d 1665 ((flags & DISSECT_IMAGE_DEVICE_READ_ONLY) ? CRYPT_ACTIVATE_READONLY : 0) |
0d12936d 1666 ((flags & DISSECT_IMAGE_DISCARD_ON_CRYPTO) ? CRYPT_ACTIVATE_ALLOW_DISCARDS : 0));
294bd454 1667 if (r < 0) {
715cbb81 1668 log_debug_errno(r, "Failed to activate LUKS device: %m");
294bd454 1669 return r == -EPERM ? -EKEYREJECTED : r;
18b5886e 1670 }
18b5886e 1671
94344385
LP
1672 d->decrypted[d->n_decrypted++] = (DecryptedPartition) {
1673 .name = TAKE_PTR(name),
1674 .device = TAKE_PTR(cd),
1675 };
18b5886e 1676
1cc6c93a 1677 m->decrypted_node = TAKE_PTR(node);
18b5886e
LP
1678
1679 return 0;
4623e8e6
LP
1680}
1681
89e62e0b
LP
1682static int verity_can_reuse(
1683 const VeritySettings *verity,
1684 const char *name,
1685 struct crypt_device **ret_cd) {
1686
ac1f3ad0
LB
1687 /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
1688 _cleanup_free_ char *root_hash_existing = NULL;
0d12936d 1689 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
ac1f3ad0 1690 struct crypt_params_verity crypt_params = {};
89e62e0b 1691 size_t root_hash_existing_size;
ac1f3ad0
LB
1692 int r;
1693
89e62e0b
LP
1694 assert(verity);
1695 assert(name);
ac1f3ad0
LB
1696 assert(ret_cd);
1697
0d12936d 1698 r = sym_crypt_init_by_name(&cd, name);
ac1f3ad0
LB
1699 if (r < 0)
1700 return log_debug_errno(r, "Error opening verity device, crypt_init_by_name failed: %m");
0d12936d 1701
c719805e
LP
1702 cryptsetup_enable_logging(cd);
1703
0d12936d 1704 r = sym_crypt_get_verity_info(cd, &crypt_params);
ac1f3ad0
LB
1705 if (r < 0)
1706 return log_debug_errno(r, "Error opening verity device, crypt_get_verity_info failed: %m");
0d12936d 1707
89e62e0b
LP
1708 root_hash_existing_size = verity->root_hash_size;
1709 root_hash_existing = malloc0(root_hash_existing_size);
ac1f3ad0
LB
1710 if (!root_hash_existing)
1711 return -ENOMEM;
0d12936d
LP
1712
1713 r = sym_crypt_volume_key_get(cd, CRYPT_ANY_SLOT, root_hash_existing, &root_hash_existing_size, NULL, 0);
ac1f3ad0
LB
1714 if (r < 0)
1715 return log_debug_errno(r, "Error opening verity device, crypt_volume_key_get failed: %m");
89e62e0b
LP
1716 if (verity->root_hash_size != root_hash_existing_size ||
1717 memcmp(root_hash_existing, verity->root_hash, verity->root_hash_size) != 0)
ac1f3ad0 1718 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but root hashes are different.");
89e62e0b 1719
ac1f3ad0 1720#if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
89e62e0b
LP
1721 /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the
1722 * same settings, so that a previous unsigned mount will not be reused if the user asks to use
28423d9a 1723 * signing for the new one, and vice versa. */
89e62e0b 1724 if (!!verity->root_hash_sig != !!(crypt_params.flags & CRYPT_VERITY_ROOT_HASH_SIGNATURE))
ac1f3ad0
LB
1725 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but signature settings are not the same.");
1726#endif
1727
1728 *ret_cd = TAKE_PTR(cd);
1729 return 0;
1730}
1731
75db809a 1732static inline char* dm_deferred_remove_clean(char *name) {
ac1f3ad0 1733 if (!name)
75db809a 1734 return NULL;
0d12936d
LP
1735
1736 (void) sym_crypt_deactivate_by_name(NULL, name, CRYPT_DEACTIVATE_DEFERRED);
75db809a 1737 return mfree(name);
ac1f3ad0
LB
1738}
1739DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean);
1740
c2fa92e7
LP
1741static int validate_signature_userspace(const VeritySettings *verity) {
1742#if HAVE_OPENSSL
1743 _cleanup_(sk_X509_free_allp) STACK_OF(X509) *sk = NULL;
1744 _cleanup_strv_free_ char **certs = NULL;
1745 _cleanup_(PKCS7_freep) PKCS7 *p7 = NULL;
1746 _cleanup_free_ char *s = NULL;
1747 _cleanup_(BIO_freep) BIO *bio = NULL; /* 'bio' must be freed first, 's' second, hence keep this order
1748 * of declaration in place, please */
1749 const unsigned char *d;
c2fa92e7
LP
1750 int r;
1751
1752 assert(verity);
1753 assert(verity->root_hash);
1754 assert(verity->root_hash_sig);
1755
1756 /* Because installing a signature certificate into the kernel chain is so messy, let's optionally do
1757 * userspace validation. */
1758
1759 r = conf_files_list_nulstr(&certs, ".crt", NULL, CONF_FILES_REGULAR|CONF_FILES_FILTER_MASKED, CONF_PATHS_NULSTR("verity.d"));
1760 if (r < 0)
1761 return log_debug_errno(r, "Failed to enumerate certificates: %m");
1762 if (strv_isempty(certs)) {
1763 log_debug("No userspace dm-verity certificates found.");
1764 return 0;
1765 }
1766
1767 d = verity->root_hash_sig;
1768 p7 = d2i_PKCS7(NULL, &d, (long) verity->root_hash_sig_size);
1769 if (!p7)
1770 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to parse PKCS7 DER signature data.");
1771
1772 s = hexmem(verity->root_hash, verity->root_hash_size);
1773 if (!s)
1774 return log_oom_debug();
1775
1776 bio = BIO_new_mem_buf(s, strlen(s));
1777 if (!bio)
1778 return log_oom_debug();
1779
1780 sk = sk_X509_new_null();
1781 if (!sk)
1782 return log_oom_debug();
1783
1784 STRV_FOREACH(i, certs) {
1785 _cleanup_(X509_freep) X509 *c = NULL;
1786 _cleanup_fclose_ FILE *f = NULL;
1787
1788 f = fopen(*i, "re");
1789 if (!f) {
1790 log_debug_errno(errno, "Failed to open '%s', ignoring: %m", *i);
1791 continue;
1792 }
1793
1794 c = PEM_read_X509(f, NULL, NULL, NULL);
1795 if (!c) {
1796 log_debug("Failed to load X509 certificate '%s', ignoring.", *i);
1797 continue;
1798 }
1799
1800 if (sk_X509_push(sk, c) == 0)
1801 return log_oom_debug();
1802
1803 TAKE_PTR(c);
1804 }
1805
1806 r = PKCS7_verify(p7, sk, NULL, bio, NULL, PKCS7_NOINTERN|PKCS7_NOVERIFY);
1807 if (r)
1808 log_debug("Userspace PKCS#7 validation succeeded.");
1809 else
1810 log_debug("Userspace PKCS#7 validation failed: %s", ERR_error_string(ERR_get_error(), NULL));
1811
1812 return r;
1813#else
1814 log_debug("Not doing client-side validation of dm-verity root hash signatures, OpenSSL support disabled.");
1815 return 0;
1816#endif
1817}
1818
1819static int do_crypt_activate_verity(
1820 struct crypt_device *cd,
1821 const char *name,
1822 const VeritySettings *verity) {
1823
1824 bool check_signature;
1825 int r;
1826
1827 assert(cd);
1828 assert(name);
1829 assert(verity);
1830
1831 if (verity->root_hash_sig) {
1832 r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIGNATURE");
1833 if (r < 0 && r != -ENXIO)
1834 log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIGNATURE");
1835
1836 check_signature = r != 0;
1837 } else
1838 check_signature = false;
1839
1840 if (check_signature) {
1841
1842#if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1843 /* First, if we have support for signed keys in the kernel, then try that first. */
1844 r = sym_crypt_activate_by_signed_key(
1845 cd,
1846 name,
1847 verity->root_hash,
1848 verity->root_hash_size,
1849 verity->root_hash_sig,
1850 verity->root_hash_sig_size,
1851 CRYPT_ACTIVATE_READONLY);
1852 if (r >= 0)
1853 return r;
1854
1855 log_debug("Validation of dm-verity signature failed via the kernel, trying userspace validation instead.");
1856#else
1857 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.",
1858 program_invocation_short_name);
1859#endif
1860
1861 /* So this didn't work via the kernel, then let's try userspace validation instead. If that
1862 * works we'll try to activate without telling the kernel the signature. */
1863
1864 r = validate_signature_userspace(verity);
1865 if (r < 0)
1866 return r;
1867 if (r == 0)
1868 return log_debug_errno(SYNTHETIC_ERRNO(ENOKEY),
1869 "Activation of signed Verity volume worked neither via the kernel nor in userspace, can't activate.");
1870 }
1871
1872 return sym_crypt_activate_by_volume_key(
1873 cd,
1874 name,
1875 verity->root_hash,
1876 verity->root_hash_size,
1877 CRYPT_ACTIVATE_READONLY);
1878}
1879
ad361a50
YW
1880static usec_t verity_timeout(void) {
1881 usec_t t = 100 * USEC_PER_MSEC;
1882 const char *e;
1883 int r;
1884
1885 /* On slower machines, like non-KVM vm, setting up device may take a long time.
1886 * Let's make the timeout configurable. */
1887
1888 e = getenv("SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC");
1889 if (!e)
1890 return t;
1891
1892 r = parse_sec(e, &t);
1893 if (r < 0)
1894 log_debug_errno(r,
1895 "Failed to parse timeout specified in $SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC, "
1896 "using the default timeout (%s).",
1897 FORMAT_TIMESPAN(t, USEC_PER_MSEC));
1898
1899 return t;
1900}
1901
4623e8e6 1902static int verity_partition(
aee36b4e 1903 PartitionDesignator designator,
4623e8e6
LP
1904 DissectedPartition *m,
1905 DissectedPartition *v,
89e62e0b 1906 const VeritySettings *verity,
4623e8e6
LP
1907 DissectImageFlags flags,
1908 DecryptedImage *d) {
1909
0d12936d 1910 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
ac1f3ad0 1911 _cleanup_(dm_deferred_remove_cleanp) char *restore_deferred_remove = NULL;
89e62e0b 1912 _cleanup_free_ char *node = NULL, *name = NULL;
4623e8e6
LP
1913 int r;
1914
1915 assert(m);
89e62e0b 1916 assert(v || (verity && verity->data_path));
4623e8e6 1917
89e62e0b 1918 if (!verity || !verity->root_hash)
4623e8e6 1919 return 0;
aee36b4e
LP
1920 if (!((verity->designator < 0 && designator == PARTITION_ROOT) ||
1921 (verity->designator == designator)))
1922 return 0;
4623e8e6
LP
1923
1924 if (!m->found || !m->node || !m->fstype)
1925 return 0;
89e62e0b 1926 if (!verity->data_path) {
e7cbe5cb
LB
1927 if (!v->found || !v->node || !v->fstype)
1928 return 0;
4623e8e6 1929
e7cbe5cb
LB
1930 if (!streq(v->fstype, "DM_verity_hash"))
1931 return 0;
1932 }
4623e8e6 1933
0d12936d
LP
1934 r = dlopen_cryptsetup();
1935 if (r < 0)
1936 return r;
1937
ac1f3ad0
LB
1938 if (FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) {
1939 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
1940 _cleanup_free_ char *root_hash_encoded = NULL;
0d12936d 1941
89e62e0b 1942 root_hash_encoded = hexmem(verity->root_hash, verity->root_hash_size);
ac1f3ad0
LB
1943 if (!root_hash_encoded)
1944 return -ENOMEM;
aee36b4e 1945
ac1f3ad0
LB
1946 r = make_dm_name_and_node(root_hash_encoded, "-verity", &name, &node);
1947 } else
1948 r = make_dm_name_and_node(m->node, "-verity", &name, &node);
4623e8e6
LP
1949 if (r < 0)
1950 return r;
1951
89e62e0b 1952 r = sym_crypt_init(&cd, verity->data_path ?: v->node);
4623e8e6
LP
1953 if (r < 0)
1954 return r;
1955
efc3b12f 1956 cryptsetup_enable_logging(cd);
1887032f 1957
0d12936d 1958 r = sym_crypt_load(cd, CRYPT_VERITY, NULL);
4623e8e6 1959 if (r < 0)
294bd454 1960 return r;
4623e8e6 1961
0d12936d 1962 r = sym_crypt_set_data_device(cd, m->node);
4623e8e6 1963 if (r < 0)
294bd454 1964 return r;
4623e8e6 1965
319a4f4b 1966 if (!GREEDY_REALLOC0(d->decrypted, d->n_decrypted + 1))
ac1f3ad0
LB
1967 return -ENOMEM;
1968
1969 /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
1970 * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
1971 * retry a few times before giving up. */
1972 for (unsigned i = 0; i < N_DEVICE_NODE_LIST_ATTEMPTS; i++) {
c2fa92e7
LP
1973
1974 r = do_crypt_activate_verity(cd, name, verity);
ac1f3ad0
LB
1975 /* libdevmapper can return EINVAL when the device is already in the activation stage.
1976 * There's no way to distinguish this situation from a genuine error due to invalid
2aed63f4 1977 * parameters, so immediately fall back to activating the device with a unique name.
89e62e0b
LP
1978 * Improvements in libcrypsetup can ensure this never happens:
1979 * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
ac1f3ad0 1980 if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
aee36b4e 1981 return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
5bf5013f
YW
1982 if (r < 0 && !IN_SET(r,
1983 -EEXIST, /* Volume is already open and ready to be used */
1984 -EBUSY, /* Volume is being opened but not ready, crypt_init_by_name can fetch details */
1985 -ENODEV /* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */))
ac1f3ad0 1986 return r;
9ecb5c10 1987 if (IN_SET(r, -EEXIST, -EBUSY)) {
041e2eda 1988 _cleanup_(sym_crypt_freep) struct crypt_device *existing_cd = NULL;
c2923fdc 1989
ac1f3ad0
LB
1990 if (!restore_deferred_remove){
1991 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
1992 r = dm_deferred_remove_cancel(name);
9ecb5c10
LB
1993 /* If activation returns EBUSY there might be no deferred removal to cancel, that's fine */
1994 if (r < 0 && r != -ENXIO)
ac1f3ad0 1995 return log_debug_errno(r, "Disabling automated deferred removal for verity device %s failed: %m", node);
5bf5013f 1996 if (r >= 0) {
9ecb5c10
LB
1997 restore_deferred_remove = strdup(name);
1998 if (!restore_deferred_remove)
1999 return -ENOMEM;
2000 }
ac1f3ad0 2001 }
c2923fdc 2002
89e62e0b 2003 r = verity_can_reuse(verity, name, &existing_cd);
ac1f3ad0
LB
2004 /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
2005 if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
aee36b4e 2006 return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
5bf5013f 2007 if (r < 0 && !IN_SET(r, -ENODEV, -ENOENT, -EBUSY))
ac1f3ad0 2008 return log_debug_errno(r, "Checking whether existing verity device %s can be reused failed: %m", node);
5bf5013f 2009 if (r >= 0) {
c419b6f0
LB
2010 /* devmapper might say that the device exists, but the devlink might not yet have been
2011 * created. Check and wait for the udev event in that case. */
ad361a50 2012 r = device_wait_for_devlink(node, "block", verity_timeout(), NULL);
c419b6f0
LB
2013 /* Fallback to activation with a unique device if it's taking too long */
2014 if (r == -ETIMEDOUT)
2015 break;
2016 if (r < 0)
2017 return r;
2018
041e2eda
YW
2019 sym_crypt_free(cd);
2020 cd = TAKE_PTR(existing_cd);
ac1f3ad0 2021 }
c2923fdc 2022 }
5bf5013f 2023 if (r >= 0)
ac1f3ad0 2024 break;
ecab4c47
LB
2025
2026 /* Device is being opened by another process, but it has not finished yet, yield for 2ms */
2027 (void) usleep(2 * USEC_PER_MSEC);
ac1f3ad0
LB
2028 }
2029
ac1f3ad0
LB
2030 /* An existing verity device was reported by libcryptsetup/libdevmapper, but we can't use it at this time.
2031 * Fall back to activating it with a unique device name. */
5bf5013f 2032 if (r < 0 && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
aee36b4e 2033 return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
ac1f3ad0
LB
2034
2035 /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
2036 restore_deferred_remove = mfree(restore_deferred_remove);
4623e8e6 2037
94344385
LP
2038 d->decrypted[d->n_decrypted++] = (DecryptedPartition) {
2039 .name = TAKE_PTR(name),
2040 .device = TAKE_PTR(cd),
2041 };
4623e8e6 2042
1cc6c93a 2043 m->decrypted_node = TAKE_PTR(node);
4623e8e6
LP
2044
2045 return 0;
18b5886e
LP
2046}
2047#endif
2048
2049int dissected_image_decrypt(
2050 DissectedImage *m,
2051 const char *passphrase,
89e62e0b 2052 const VeritySettings *verity,
18b5886e
LP
2053 DissectImageFlags flags,
2054 DecryptedImage **ret) {
2055
349cc4a5 2056#if HAVE_LIBCRYPTSETUP
49b5b3b4 2057 _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL;
18b5886e
LP
2058 int r;
2059#endif
2060
2061 assert(m);
89e62e0b 2062 assert(!verity || verity->root_hash || verity->root_hash_size == 0);
18b5886e
LP
2063
2064 /* Returns:
2065 *
2066 * = 0 → There was nothing to decrypt
2067 * > 0 → Decrypted successfully
d1c536f5 2068 * -ENOKEY → There's something to decrypt but no key was supplied
18b5886e
LP
2069 * -EKEYREJECTED → Passed key was not correct
2070 */
2071
89e62e0b 2072 if (verity && verity->root_hash && verity->root_hash_size < sizeof(sd_id128_t))
4623e8e6
LP
2073 return -EINVAL;
2074
c3c88d67 2075 if (!m->encrypted && !m->verity_ready) {
18b5886e
LP
2076 *ret = NULL;
2077 return 0;
2078 }
2079
349cc4a5 2080#if HAVE_LIBCRYPTSETUP
18b5886e
LP
2081 d = new0(DecryptedImage, 1);
2082 if (!d)
2083 return -ENOMEM;
2084
569a0e42 2085 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
18b5886e 2086 DissectedPartition *p = m->partitions + i;
22043172 2087 PartitionDesignator k;
18b5886e
LP
2088
2089 if (!p->found)
2090 continue;
2091
2092 r = decrypt_partition(p, passphrase, flags, d);
2093 if (r < 0)
2094 return r;
2095
4623e8e6
LP
2096 k = PARTITION_VERITY_OF(i);
2097 if (k >= 0) {
aee36b4e 2098 r = verity_partition(i, p, m->partitions + k, verity, flags | DISSECT_IMAGE_VERITY_SHARE, d);
4623e8e6
LP
2099 if (r < 0)
2100 return r;
2101 }
2102
18b5886e
LP
2103 if (!p->decrypted_fstype && p->decrypted_node) {
2104 r = probe_filesystem(p->decrypted_node, &p->decrypted_fstype);
7cc84b2c 2105 if (r < 0 && r != -EUCLEAN)
18b5886e
LP
2106 return r;
2107 }
2108 }
2109
1cc6c93a 2110 *ret = TAKE_PTR(d);
18b5886e
LP
2111
2112 return 1;
2113#else
2114 return -EOPNOTSUPP;
2115#endif
2116}
2117
2118int dissected_image_decrypt_interactively(
2119 DissectedImage *m,
2120 const char *passphrase,
89e62e0b 2121 const VeritySettings *verity,
18b5886e
LP
2122 DissectImageFlags flags,
2123 DecryptedImage **ret) {
2124
2125 _cleanup_strv_free_erase_ char **z = NULL;
2126 int n = 3, r;
2127
2128 if (passphrase)
2129 n--;
2130
2131 for (;;) {
89e62e0b 2132 r = dissected_image_decrypt(m, passphrase, verity, flags, ret);
18b5886e
LP
2133 if (r >= 0)
2134 return r;
2135 if (r == -EKEYREJECTED)
2136 log_error_errno(r, "Incorrect passphrase, try again!");
fc95c359
YW
2137 else if (r != -ENOKEY)
2138 return log_error_errno(r, "Failed to decrypt image: %m");
18b5886e 2139
baaa35ad
ZJS
2140 if (--n < 0)
2141 return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED),
2142 "Too many retries.");
18b5886e
LP
2143
2144 z = strv_free(z);
2145
8806bb4b 2146 r = ask_password_auto("Please enter image passphrase:", NULL, "dissect", "dissect", "dissect.passphrase", USEC_INFINITY, 0, &z);
18b5886e
LP
2147 if (r < 0)
2148 return log_error_errno(r, "Failed to query for passphrase: %m");
2149
2150 passphrase = z[0];
2151 }
2152}
2153
18b5886e 2154int decrypted_image_relinquish(DecryptedImage *d) {
18b5886e
LP
2155 assert(d);
2156
67f63ee5
ZJS
2157 /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a
2158 * boolean so that we don't clean it up ourselves either anymore */
18b5886e 2159
349cc4a5 2160#if HAVE_LIBCRYPTSETUP
67f63ee5
ZJS
2161 int r;
2162
2163 for (size_t i = 0; i < d->n_decrypted; i++) {
18b5886e
LP
2164 DecryptedPartition *p = d->decrypted + i;
2165
2166 if (p->relinquished)
2167 continue;
2168
0d12936d 2169 r = sym_crypt_deactivate_by_name(NULL, p->name, CRYPT_DEACTIVATE_DEFERRED);
18b5886e
LP
2170 if (r < 0)
2171 return log_debug_errno(r, "Failed to mark %s for auto-removal: %m", p->name);
2172
2173 p->relinquished = true;
2174 }
2175#endif
2176
2177 return 0;
2178}
2179
89e62e0b
LP
2180static char *build_auxiliary_path(const char *image, const char *suffix) {
2181 const char *e;
2182 char *n;
2183
2184 assert(image);
2185 assert(suffix);
2186
2187 e = endswith(image, ".raw");
2188 if (!e)
2189 return strjoin(e, suffix);
2190
2191 n = new(char, e - image + strlen(suffix) + 1);
2192 if (!n)
2193 return NULL;
2194
2195 strcpy(mempcpy(n, image, e - image), suffix);
2196 return n;
2197}
2198
2199void verity_settings_done(VeritySettings *v) {
2200 assert(v);
2201
2202 v->root_hash = mfree(v->root_hash);
2203 v->root_hash_size = 0;
2204
2205 v->root_hash_sig = mfree(v->root_hash_sig);
2206 v->root_hash_sig_size = 0;
2207
2208 v->data_path = mfree(v->data_path);
2209}
2210
2211int verity_settings_load(
2212 VeritySettings *verity,
f5ea63a5
LP
2213 const char *image,
2214 const char *root_hash_path,
89e62e0b
LP
2215 const char *root_hash_sig_path) {
2216
2217 _cleanup_free_ void *root_hash = NULL, *root_hash_sig = NULL;
2218 size_t root_hash_size = 0, root_hash_sig_size = 0;
2219 _cleanup_free_ char *verity_data_path = NULL;
aee36b4e 2220 PartitionDesignator designator;
78ebe980
LP
2221 int r;
2222
89e62e0b 2223 assert(verity);
78ebe980 2224 assert(image);
aee36b4e 2225 assert(verity->designator < 0 || IN_SET(verity->designator, PARTITION_ROOT, PARTITION_USR));
78ebe980 2226
89e62e0b
LP
2227 /* If we are asked to load the root hash for a device node, exit early */
2228 if (is_device_path(image))
78ebe980 2229 return 0;
78ebe980 2230
d5fcc5b0
LP
2231 r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIDECAR");
2232 if (r < 0 && r != -ENXIO)
2233 log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIDECAR, ignoring: %m");
2234 if (r == 0)
2235 return 0;
2236
aee36b4e
LP
2237 designator = verity->designator;
2238
89e62e0b 2239 /* We only fill in what isn't already filled in */
c2923fdc 2240
89e62e0b 2241 if (!verity->root_hash) {
e7cbe5cb 2242 _cleanup_free_ char *text = NULL;
e7cbe5cb 2243
0389f4fa 2244 if (root_hash_path) {
aee36b4e 2245 /* If explicitly specified it takes precedence */
0389f4fa
LB
2246 r = read_one_line_file(root_hash_path, &text);
2247 if (r < 0)
e7cbe5cb 2248 return r;
aee36b4e
LP
2249
2250 if (designator < 0)
2251 designator = PARTITION_ROOT;
0389f4fa 2252 } else {
aee36b4e
LP
2253 /* Otherwise look for xattr and separate file, and first for the data for root and if
2254 * that doesn't exist for /usr */
0389f4fa 2255
aee36b4e 2256 if (designator < 0 || designator == PARTITION_ROOT) {
c53e07e2 2257 r = getxattr_malloc(image, "user.verity.roothash", &text);
aee36b4e
LP
2258 if (r < 0) {
2259 _cleanup_free_ char *p = NULL;
78ebe980 2260
aee36b4e
LP
2261 if (!IN_SET(r, -ENODATA, -ENOENT) && !ERRNO_IS_NOT_SUPPORTED(r))
2262 return r;
e7cbe5cb 2263
aee36b4e
LP
2264 p = build_auxiliary_path(image, ".roothash");
2265 if (!p)
2266 return -ENOMEM;
2267
2268 r = read_one_line_file(p, &text);
2269 if (r < 0 && r != -ENOENT)
2270 return r;
2271 }
2272
2273 if (text)
2274 designator = PARTITION_ROOT;
2275 }
2276
2277 if (!text && (designator < 0 || designator == PARTITION_USR)) {
2278 /* So in the "roothash" xattr/file name above the "root" of course primarily
2279 * refers to the root of the Verity Merkle tree. But coincidentally it also
2280 * is the hash for the *root* file system, i.e. the "root" neatly refers to
2281 * two distinct concepts called "root". Taking benefit of this happy
2282 * coincidence we call the file with the root hash for the /usr/ file system
2283 * `usrhash`, because `usrroothash` or `rootusrhash` would just be too
2284 * confusing. We thus drop the reference to the root of the Merkle tree, and
2285 * just indicate which file system it's about. */
c53e07e2 2286 r = getxattr_malloc(image, "user.verity.usrhash", &text);
aee36b4e
LP
2287 if (r < 0) {
2288 _cleanup_free_ char *p = NULL;
2289
2290 if (!IN_SET(r, -ENODATA, -ENOENT) && !ERRNO_IS_NOT_SUPPORTED(r))
2291 return r;
2292
2293 p = build_auxiliary_path(image, ".usrhash");
2294 if (!p)
2295 return -ENOMEM;
2296
2297 r = read_one_line_file(p, &text);
2298 if (r < 0 && r != -ENOENT)
2299 return r;
2300 }
2301
2302 if (text)
2303 designator = PARTITION_USR;
0389f4fa 2304 }
e7cbe5cb
LB
2305 }
2306
2307 if (text) {
89e62e0b 2308 r = unhexmem(text, strlen(text), &root_hash, &root_hash_size);
e7cbe5cb
LB
2309 if (r < 0)
2310 return r;
89e62e0b 2311 if (root_hash_size < sizeof(sd_id128_t))
e7cbe5cb
LB
2312 return -EINVAL;
2313 }
2314 }
2315
90f98986 2316 if ((root_hash || verity->root_hash) && !verity->root_hash_sig) {
aee36b4e 2317 if (root_hash_sig_path) {
ae9cf30b 2318 r = read_full_file(root_hash_sig_path, (char**) &root_hash_sig, &root_hash_sig_size);
aee36b4e
LP
2319 if (r < 0 && r != -ENOENT)
2320 return r;
2321
2322 if (designator < 0)
2323 designator = PARTITION_ROOT;
2324 } else {
2325 if (designator < 0 || designator == PARTITION_ROOT) {
2326 _cleanup_free_ char *p = NULL;
2327
2328 /* Follow naming convention recommended by the relevant RFC:
2329 * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
2330 p = build_auxiliary_path(image, ".roothash.p7s");
2331 if (!p)
2332 return -ENOMEM;
89e62e0b 2333
ae9cf30b 2334 r = read_full_file(p, (char**) &root_hash_sig, &root_hash_sig_size);
aee36b4e
LP
2335 if (r < 0 && r != -ENOENT)
2336 return r;
2337 if (r >= 0)
2338 designator = PARTITION_ROOT;
2339 }
2340
2341 if (!root_hash_sig && (designator < 0 || designator == PARTITION_USR)) {
2342 _cleanup_free_ char *p = NULL;
2343
2344 p = build_auxiliary_path(image, ".usrhash.p7s");
2345 if (!p)
2346 return -ENOMEM;
89e62e0b 2347
ae9cf30b 2348 r = read_full_file(p, (char**) &root_hash_sig, &root_hash_sig_size);
aee36b4e
LP
2349 if (r < 0 && r != -ENOENT)
2350 return r;
2351 if (r >= 0)
2352 designator = PARTITION_USR;
2353 }
89e62e0b
LP
2354 }
2355
aee36b4e 2356 if (root_hash_sig && root_hash_sig_size == 0) /* refuse empty size signatures */
89e62e0b
LP
2357 return -EINVAL;
2358 }
2359
2360 if (!verity->data_path) {
2361 _cleanup_free_ char *p = NULL;
2362
2363 p = build_auxiliary_path(image, ".verity");
2364 if (!p)
2365 return -ENOMEM;
2366
2367 if (access(p, F_OK) < 0) {
2368 if (errno != ENOENT)
2369 return -errno;
2370 } else
2371 verity_data_path = TAKE_PTR(p);
2372 }
2373
2374 if (root_hash) {
2375 verity->root_hash = TAKE_PTR(root_hash);
2376 verity->root_hash_size = root_hash_size;
2377 }
2378
2379 if (root_hash_sig) {
2380 verity->root_hash_sig = TAKE_PTR(root_hash_sig);
2381 verity->root_hash_sig_size = root_hash_sig_size;
e7cbe5cb 2382 }
89e62e0b
LP
2383
2384 if (verity_data_path)
2385 verity->data_path = TAKE_PTR(verity_data_path);
78ebe980 2386
aee36b4e
LP
2387 if (verity->designator < 0)
2388 verity->designator = designator;
2389
78ebe980
LP
2390 return 1;
2391}
2392
88b3300f
LP
2393int dissected_image_load_verity_sig_partition(
2394 DissectedImage *m,
2395 int fd,
2396 VeritySettings *verity) {
2397
2398 _cleanup_free_ void *root_hash = NULL, *root_hash_sig = NULL;
2399 _cleanup_(json_variant_unrefp) JsonVariant *v = NULL;
2400 size_t root_hash_size, root_hash_sig_size;
2401 _cleanup_free_ char *buf = NULL;
2402 PartitionDesignator d;
2403 DissectedPartition *p;
2404 JsonVariant *rh, *sig;
2405 ssize_t n;
2406 char *e;
2407 int r;
2408
2409 assert(m);
2410 assert(fd >= 0);
2411 assert(verity);
2412
2413 if (verity->root_hash && verity->root_hash_sig) /* Already loaded? */
2414 return 0;
2415
2416 r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_EMBEDDED");
2417 if (r < 0 && r != -ENXIO)
2418 log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_EMBEDDED, ignoring: %m");
2419 if (r == 0)
2420 return 0;
2421
2422 d = PARTITION_VERITY_SIG_OF(verity->designator < 0 ? PARTITION_ROOT : verity->designator);
2423 assert(d >= 0);
2424
2425 p = m->partitions + d;
2426 if (!p->found)
2427 return 0;
2428 if (p->offset == UINT64_MAX || p->size == UINT64_MAX)
2429 return -EINVAL;
2430
2431 if (p->size > 4*1024*1024) /* Signature data cannot possible be larger than 4M, refuse that */
2432 return -EFBIG;
2433
2434 buf = new(char, p->size+1);
2435 if (!buf)
2436 return -ENOMEM;
2437
2438 n = pread(fd, buf, p->size, p->offset);
2439 if (n < 0)
2440 return -ENOMEM;
2441 if ((uint64_t) n != p->size)
2442 return -EIO;
2443
2444 e = memchr(buf, 0, p->size);
2445 if (e) {
2446 /* If we found a NUL byte then the rest of the data must be NUL too */
2447 if (!memeqzero(e, p->size - (e - buf)))
2448 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature data contains embedded NUL byte.");
2449 } else
2450 buf[p->size] = 0;
2451
2452 r = json_parse(buf, 0, &v, NULL, NULL);
2453 if (r < 0)
2454 return log_debug_errno(r, "Failed to parse signature JSON data: %m");
2455
2456 rh = json_variant_by_key(v, "rootHash");
2457 if (!rh)
2458 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature JSON object lacks 'rootHash' field.");
2459 if (!json_variant_is_string(rh))
2460 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "'rootHash' field of signature JSON object is not a string.");
2461
2462 r = unhexmem(json_variant_string(rh), SIZE_MAX, &root_hash, &root_hash_size);
2463 if (r < 0)
2464 return log_debug_errno(r, "Failed to parse root hash field: %m");
2465
2466 /* Check if specified root hash matches if it is specified */
2467 if (verity->root_hash &&
2468 memcmp_nn(verity->root_hash, verity->root_hash_size, root_hash, root_hash_size) != 0) {
2469 _cleanup_free_ char *a = NULL, *b = NULL;
2470
2471 a = hexmem(root_hash, root_hash_size);
2472 b = hexmem(verity->root_hash, verity->root_hash_size);
2473
2474 return log_debug_errno(r, "Root hash in signature JSON data (%s) doesn't match configured hash (%s).", strna(a), strna(b));
2475 }
2476
2477 sig = json_variant_by_key(v, "signature");
2478 if (!sig)
2479 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature JSON object lacks 'signature' field.");
2480 if (!json_variant_is_string(sig))
2481 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "'signature' field of signature JSON object is not a string.");
2482
2483 r = unbase64mem(json_variant_string(sig), SIZE_MAX, &root_hash_sig, &root_hash_sig_size);
2484 if (r < 0)
2485 return log_debug_errno(r, "Failed to parse signature field: %m");
2486
2487 free_and_replace(verity->root_hash, root_hash);
2488 verity->root_hash_size = root_hash_size;
2489
2490 free_and_replace(verity->root_hash_sig, root_hash_sig);
2491 verity->root_hash_sig_size = root_hash_sig_size;
2492
2493 return 1;
2494}
2495
22847508 2496int dissected_image_acquire_metadata(DissectedImage *m, DissectImageFlags extra_flags) {
3b925504
LP
2497
2498 enum {
2499 META_HOSTNAME,
2500 META_MACHINE_ID,
2501 META_MACHINE_INFO,
2502 META_OS_RELEASE,
7718ac97 2503 META_EXTENSION_RELEASE,
a4e0d617 2504 META_HAS_INIT_SYSTEM,
3b925504
LP
2505 _META_MAX,
2506 };
2507
9a4b883b 2508 static const char *const paths[_META_MAX] = {
7718ac97
LB
2509 [META_HOSTNAME] = "/etc/hostname\0",
2510 [META_MACHINE_ID] = "/etc/machine-id\0",
2511 [META_MACHINE_INFO] = "/etc/machine-info\0",
9a4b883b 2512 [META_OS_RELEASE] = ("/etc/os-release\0"
22847508 2513 "/usr/lib/os-release\0"),
a4e0d617
LP
2514 [META_EXTENSION_RELEASE] = "extension-release\0", /* Used only for logging. */
2515 [META_HAS_INIT_SYSTEM] = "has-init-system\0", /* ditto */
3b925504
LP
2516 };
2517
7718ac97 2518 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL, **extension_release = NULL;
af8219d5 2519 _cleanup_close_pair_ int error_pipe[2] = { -1, -1 };
3b925504
LP
2520 _cleanup_(rmdir_and_freep) char *t = NULL;
2521 _cleanup_(sigkill_waitp) pid_t child = 0;
2522 sd_id128_t machine_id = SD_ID128_NULL;
2523 _cleanup_free_ char *hostname = NULL;
67f63ee5 2524 unsigned n_meta_initialized = 0;
af8219d5 2525 int fds[2 * _META_MAX], r, v;
a4e0d617 2526 int has_init_system = -1;
af8219d5 2527 ssize_t n;
3b925504
LP
2528
2529 BLOCK_SIGNALS(SIGCHLD);
2530
2531 assert(m);
2532
7718ac97 2533 for (; n_meta_initialized < _META_MAX; n_meta_initialized ++) {
d9119c00
LP
2534 if (!paths[n_meta_initialized]) {
2535 fds[2*n_meta_initialized] = fds[2*n_meta_initialized+1] = -1;
7718ac97 2536 continue;
d9119c00
LP
2537 }
2538
3b925504
LP
2539 if (pipe2(fds + 2*n_meta_initialized, O_CLOEXEC) < 0) {
2540 r = -errno;
2541 goto finish;
2542 }
7718ac97 2543 }
3b925504
LP
2544
2545 r = mkdtemp_malloc("/tmp/dissect-XXXXXX", &t);
2546 if (r < 0)
2547 goto finish;
2548
af8219d5
LP
2549 if (pipe2(error_pipe, O_CLOEXEC) < 0) {
2550 r = -errno;
2551 goto finish;
2552 }
2553
e2047ba9 2554 r = safe_fork("(sd-dissect)", FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE, &child);
be39f6ee 2555 if (r < 0)
3b925504 2556 goto finish;
be39f6ee 2557 if (r == 0) {
a4e0d617 2558 /* Child in a new mount namespace */
af8219d5
LP
2559 error_pipe[0] = safe_close(error_pipe[0]);
2560
7cf66030
LP
2561 r = dissected_image_mount(
2562 m,
2563 t,
2564 UID_INVALID,
21b61b1d 2565 UID_INVALID,
22847508
ZJS
2566 extra_flags |
2567 DISSECT_IMAGE_READ_ONLY |
2568 DISSECT_IMAGE_MOUNT_ROOT_ONLY |
7cf66030 2569 DISSECT_IMAGE_USR_NO_ROOT);
429d4e41
LP
2570 if (r < 0) {
2571 log_debug_errno(r, "Failed to mount dissected image: %m");
03ae68f4 2572 goto inner_fail;
429d4e41 2573 }
3b925504 2574
67f63ee5 2575 for (unsigned k = 0; k < _META_MAX; k++) {
37e44c3f 2576 _cleanup_close_ int fd = -ENOENT;
3b925504
LP
2577 const char *p;
2578
7718ac97
LB
2579 if (!paths[k])
2580 continue;
2581
3b925504
LP
2582 fds[2*k] = safe_close(fds[2*k]);
2583
a4e0d617
LP
2584 switch (k) {
2585
2586 case META_EXTENSION_RELEASE:
9a4b883b
LB
2587 /* As per the os-release spec, if the image is an extension it will have a file
2588 * named after the image name in extension-release.d/ - we use the image name
2589 * and try to resolve it with the extension-release helpers, as sometimes
2590 * the image names are mangled on deployment and do not match anymore.
2591 * Unlike other paths this is not fixed, and the image name
2592 * can be mangled on deployment, so by calling into the helper
2593 * we allow a fallback that matches on the first extension-release
2594 * file found in the directory, if one named after the image cannot
2595 * be found first. */
2596 r = open_extension_release(t, m->image_name, NULL, &fd);
2597 if (r < 0)
2598 fd = r; /* Propagate the error. */
a4e0d617
LP
2599 break;
2600
2601 case META_HAS_INIT_SYSTEM: {
2602 bool found = false;
a4e0d617
LP
2603
2604 FOREACH_STRING(init,
2605 "/usr/lib/systemd/systemd", /* systemd on /usr merged system */
2606 "/lib/systemd/systemd", /* systemd on /usr non-merged systems */
2607 "/sbin/init") { /* traditional path the Linux kernel invokes */
2608
2609 r = chase_symlinks(init, t, CHASE_PREFIX_ROOT, NULL, NULL);
2610 if (r < 0) {
2611 if (r != -ENOENT)
2612 log_debug_errno(r, "Failed to resolve %s, ignoring: %m", init);
2613 } else {
2614 found = true;
2615 break;
2616 }
2617 }
2618
2619 r = loop_write(fds[2*k+1], &found, sizeof(found), false);
2620 if (r < 0)
2621 goto inner_fail;
2622
2623 continue;
2624 }
2625
2626 default:
9a4b883b
LB
2627 NULSTR_FOREACH(p, paths[k]) {
2628 fd = chase_symlinks_and_open(p, t, CHASE_PREFIX_ROOT, O_RDONLY|O_CLOEXEC|O_NOCTTY, NULL);
2629 if (fd >= 0)
2630 break;
2631 }
a4e0d617
LP
2632 }
2633
36952d19
LP
2634 if (fd < 0) {
2635 log_debug_errno(fd, "Failed to read %s file of image, ignoring: %m", paths[k]);
37e44c3f 2636 fds[2*k+1] = safe_close(fds[2*k+1]);
3b925504 2637 continue;
36952d19 2638 }
3b925504 2639
f5fbe71d 2640 r = copy_bytes(fd, fds[2*k+1], UINT64_MAX, 0);
a4e0d617
LP
2641 if (r < 0)
2642 goto inner_fail;
3b925504
LP
2643
2644 fds[2*k+1] = safe_close(fds[2*k+1]);
2645 }
2646
2647 _exit(EXIT_SUCCESS);
a4e0d617
LP
2648
2649 inner_fail:
03ae68f4 2650 /* Let parent know the error */
a4e0d617
LP
2651 (void) write(error_pipe[1], &r, sizeof(r));
2652 _exit(EXIT_FAILURE);
3b925504
LP
2653 }
2654
af8219d5
LP
2655 error_pipe[1] = safe_close(error_pipe[1]);
2656
67f63ee5 2657 for (unsigned k = 0; k < _META_MAX; k++) {
3b925504
LP
2658 _cleanup_fclose_ FILE *f = NULL;
2659
7718ac97
LB
2660 if (!paths[k])
2661 continue;
2662
3b925504
LP
2663 fds[2*k+1] = safe_close(fds[2*k+1]);
2664
4fa744a3 2665 f = take_fdopen(&fds[2*k], "r");
3b925504
LP
2666 if (!f) {
2667 r = -errno;
2668 goto finish;
2669 }
2670
3b925504
LP
2671 switch (k) {
2672
2673 case META_HOSTNAME:
2674 r = read_etc_hostname_stream(f, &hostname);
2675 if (r < 0)
f6048e5e 2676 log_debug_errno(r, "Failed to read /etc/hostname of image: %m");
3b925504
LP
2677
2678 break;
2679
2680 case META_MACHINE_ID: {
2681 _cleanup_free_ char *line = NULL;
2682
2683 r = read_line(f, LONG_LINE_MAX, &line);
2684 if (r < 0)
f6048e5e 2685 log_debug_errno(r, "Failed to read /etc/machine-id of image: %m");
3b925504
LP
2686 else if (r == 33) {
2687 r = sd_id128_from_string(line, &machine_id);
2688 if (r < 0)
2689 log_debug_errno(r, "Image contains invalid /etc/machine-id: %s", line);
2690 } else if (r == 0)
f6048e5e 2691 log_debug("/etc/machine-id file of image is empty.");
ab763cb2 2692 else if (streq(line, "uninitialized"))
f6048e5e 2693 log_debug("/etc/machine-id file of image is uninitialized (likely aborted first boot).");
3b925504 2694 else
f6048e5e 2695 log_debug("/etc/machine-id file of image has unexpected length %i.", r);
3b925504
LP
2696
2697 break;
2698 }
2699
2700 case META_MACHINE_INFO:
aa8fbc74 2701 r = load_env_file_pairs(f, "machine-info", &machine_info);
3b925504 2702 if (r < 0)
f6048e5e 2703 log_debug_errno(r, "Failed to read /etc/machine-info of image: %m");
3b925504
LP
2704
2705 break;
2706
2707 case META_OS_RELEASE:
aa8fbc74 2708 r = load_env_file_pairs(f, "os-release", &os_release);
3b925504 2709 if (r < 0)
f6048e5e 2710 log_debug_errno(r, "Failed to read OS release file of image: %m");
3b925504
LP
2711
2712 break;
7718ac97
LB
2713
2714 case META_EXTENSION_RELEASE:
2715 r = load_env_file_pairs(f, "extension-release", &extension_release);
2716 if (r < 0)
f6048e5e 2717 log_debug_errno(r, "Failed to read extension release file of image: %m");
7718ac97
LB
2718
2719 break;
a4e0d617
LP
2720
2721 case META_HAS_INIT_SYSTEM: {
2722 bool b = false;
2723 size_t nr;
2724
2725 errno = 0;
2726 nr = fread(&b, 1, sizeof(b), f);
2727 if (nr != sizeof(b))
2728 log_debug_errno(errno_or_else(EIO), "Failed to read has-init-system boolean: %m");
2729 else
2730 has_init_system = b;
2731
2732 break;
2733 }}
3b925504
LP
2734 }
2735
2e87a1fd 2736 r = wait_for_terminate_and_check("(sd-dissect)", child, 0);
3b925504 2737 child = 0;
2e87a1fd 2738 if (r < 0)
af8219d5
LP
2739 return r;
2740
2741 n = read(error_pipe[0], &v, sizeof(v));
2742 if (n < 0)
2743 return -errno;
2744 if (n == sizeof(v))
2745 return v; /* propagate error sent to us from child */
2746 if (n != 0)
2747 return -EIO;
2748
2e87a1fd
LP
2749 if (r != EXIT_SUCCESS)
2750 return -EPROTO;
3b925504
LP
2751
2752 free_and_replace(m->hostname, hostname);
2753 m->machine_id = machine_id;
2754 strv_free_and_replace(m->machine_info, machine_info);
2755 strv_free_and_replace(m->os_release, os_release);
7718ac97 2756 strv_free_and_replace(m->extension_release, extension_release);
a4e0d617 2757 m->has_init_system = has_init_system;
3b925504
LP
2758
2759finish:
67f63ee5 2760 for (unsigned k = 0; k < n_meta_initialized; k++)
3b925504
LP
2761 safe_close_pair(fds + 2*k);
2762
2763 return r;
2764}
2765
bad31660 2766int dissect_loop_device_and_warn(
bad31660 2767 const LoopDevice *loop,
89e62e0b 2768 const VeritySettings *verity,
18d73705 2769 const MountOptions *mount_options,
4526113f
LP
2770 DissectImageFlags flags,
2771 DissectedImage **ret) {
2772
64dd3a24 2773 const char *name;
4526113f
LP
2774 int r;
2775
bad31660
YW
2776 assert(loop);
2777 assert(loop->fd >= 0);
4526113f 2778
64dd3a24 2779 name = ASSERT_PTR(loop->backing_file ?: loop->node);
4526113f 2780
369de26f 2781 r = dissect_loop_device(loop, verity, mount_options, flags, ret);
4526113f
LP
2782 switch (r) {
2783
2784 case -EOPNOTSUPP:
2785 return log_error_errno(r, "Dissecting images is not supported, compiled without blkid support.");
2786
2787 case -ENOPKG:
48084df6
ZJS
2788 return log_error_errno(r, "%s: Couldn't identify a suitable partition table or file system.", name);
2789
2790 case -ENOMEDIUM:
2791 return log_error_errno(r, "%s: The image does not pass validation.", name);
4526113f
LP
2792
2793 case -EADDRNOTAVAIL:
48084df6 2794 return log_error_errno(r, "%s: No root partition for specified root hash found.", name);
4526113f
LP
2795
2796 case -ENOTUNIQ:
48084df6 2797 return log_error_errno(r, "%s: Multiple suitable root partitions found in image.", name);
4526113f
LP
2798
2799 case -ENXIO:
48084df6 2800 return log_error_errno(r, "%s: No suitable root partition found in image.", name);
4526113f
LP
2801
2802 case -EPROTONOSUPPORT:
2803 return log_error_errno(r, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name);
2804
48084df6
ZJS
2805 case -ENOTBLK:
2806 return log_error_errno(r, "%s: Image is not a block device.", name);
2807
a94aa2b9
LP
2808 case -EBADR:
2809 return log_error_errno(r,
2810 "Combining partitioned images (such as '%s') with external Verity data (such as '%s') not supported. "
2811 "(Consider setting $SYSTEMD_DISSECT_VERITY_SIDECAR=0 to disable automatic discovery of external Verity data.)",
2812 name, strna(verity ? verity->data_path : NULL));
2813
4526113f
LP
2814 default:
2815 if (r < 0)
2816 return log_error_errno(r, "Failed to dissect image '%s': %m", name);
2817
2818 return r;
2819 }
2820}
2821
49536766
LP
2822bool dissected_image_verity_candidate(const DissectedImage *image, PartitionDesignator partition_designator) {
2823 assert(image);
2824
2825 /* Checks if this partition could theoretically do Verity. For non-partitioned images this only works
2826 * if there's an external verity file supplied, for which we can consult .has_verity. For partitioned
2827 * images we only check the partition type.
2828 *
2829 * This call is used to decide whether to suppress or show a verity column in tabular output of the
2830 * image. */
2831
e7cbe5cb 2832 if (image->single_file_system)
c3c88d67 2833 return partition_designator == PARTITION_ROOT && image->has_verity;
e7cbe5cb
LB
2834
2835 return PARTITION_VERITY_OF(partition_designator) >= 0;
2836}
2837
49536766
LP
2838bool dissected_image_verity_ready(const DissectedImage *image, PartitionDesignator partition_designator) {
2839 PartitionDesignator k;
2840
2841 assert(image);
2842
2843 /* Checks if this partition has verity data available that we can activate. For non-partitioned this
2844 * works for the root partition, for others only if the associated verity partition was found. */
2845
2846 if (!image->verity_ready)
2847 return false;
e7cbe5cb
LB
2848
2849 if (image->single_file_system)
49536766 2850 return partition_designator == PARTITION_ROOT;
e7cbe5cb
LB
2851
2852 k = PARTITION_VERITY_OF(partition_designator);
2853 return k >= 0 && image->partitions[k].found;
2854}
2855
8ee9615e
LP
2856bool dissected_image_verity_sig_ready(const DissectedImage *image, PartitionDesignator partition_designator) {
2857 PartitionDesignator k;
2858
2859 assert(image);
2860
2861 /* Checks if this partition has verity signature data available that we can use. */
2862
2863 if (!image->verity_sig_ready)
2864 return false;
2865
2866 if (image->single_file_system)
2867 return partition_designator == PARTITION_ROOT;
2868
2869 k = PARTITION_VERITY_SIG_OF(partition_designator);
2870 return k >= 0 && image->partitions[k].found;
2871}
2872
18d73705
LB
2873MountOptions* mount_options_free_all(MountOptions *options) {
2874 MountOptions *m;
2875
2876 while ((m = options)) {
2877 LIST_REMOVE(mount_options, options, m);
2878 free(m->options);
2879 free(m);
2880 }
2881
2882 return NULL;
2883}
2884
569a0e42 2885const char* mount_options_from_designator(const MountOptions *options, PartitionDesignator designator) {
f5215bc8 2886 LIST_FOREACH(mount_options, m, options)
9ece6444 2887 if (designator == m->partition_designator && !isempty(m->options))
18d73705 2888 return m->options;
6aa05ebd 2889
18d73705
LB
2890 return NULL;
2891}
2892
6aa05ebd
LP
2893int mount_image_privately_interactively(
2894 const char *image,
2895 DissectImageFlags flags,
2896 char **ret_directory,
2897 LoopDevice **ret_loop_device,
2898 DecryptedImage **ret_decrypted_image) {
2899
27ec815e 2900 _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT;
6aa05ebd
LP
2901 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
2902 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
2903 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
2904 _cleanup_(rmdir_and_freep) char *created_dir = NULL;
2905 _cleanup_free_ char *temp = NULL;
2906 int r;
2907
2908 /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
2909 * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
2910 * easily. */
2911
2912 assert(image);
2913 assert(ret_directory);
2914 assert(ret_loop_device);
2915 assert(ret_decrypted_image);
2916
27ec815e
LP
2917 r = verity_settings_load(&verity, image, NULL, NULL);
2918 if (r < 0)
2919 return log_error_errno(r, "Failed to load root hash data: %m");
2920
6aa05ebd
LP
2921 r = tempfn_random_child(NULL, program_invocation_short_name, &temp);
2922 if (r < 0)
2923 return log_error_errno(r, "Failed to generate temporary mount directory: %m");
2924
2925 r = loop_device_make_by_path(
2926 image,
ef9c184d 2927 FLAGS_SET(flags, DISSECT_IMAGE_DEVICE_READ_ONLY) ? O_RDONLY : O_RDWR,
6aa05ebd 2928 FLAGS_SET(flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN,
7f52206a 2929 LOCK_SH,
6aa05ebd
LP
2930 &d);
2931 if (r < 0)
7b87fe4c 2932 return log_error_errno(r, "Failed to set up loopback device for %s: %m", image);
6aa05ebd 2933
64dd3a24 2934 r = dissect_loop_device_and_warn(d, &verity, NULL, flags, &dissected_image);
6aa05ebd
LP
2935 if (r < 0)
2936 return r;
2937
88b3300f
LP
2938 r = dissected_image_load_verity_sig_partition(dissected_image, d->fd, &verity);
2939 if (r < 0)
2940 return r;
2941
27ec815e 2942 r = dissected_image_decrypt_interactively(dissected_image, NULL, &verity, flags, &decrypted_image);
6aa05ebd
LP
2943 if (r < 0)
2944 return r;
2945
2946 r = detach_mount_namespace();
2947 if (r < 0)
2948 return log_error_errno(r, "Failed to detach mount namespace: %m");
2949
2950 r = mkdir_p(temp, 0700);
2951 if (r < 0)
2952 return log_error_errno(r, "Failed to create mount point: %m");
2953
2954 created_dir = TAKE_PTR(temp);
2955
21b61b1d 2956 r = dissected_image_mount_and_warn(dissected_image, created_dir, UID_INVALID, UID_INVALID, flags);
6aa05ebd 2957 if (r < 0)
af187ab2 2958 return r;
6aa05ebd 2959
41bc4849
LP
2960 r = loop_device_flock(d, LOCK_UN);
2961 if (r < 0)
2962 return r;
2963
6aa05ebd
LP
2964 if (decrypted_image) {
2965 r = decrypted_image_relinquish(decrypted_image);
2966 if (r < 0)
2967 return log_error_errno(r, "Failed to relinquish DM devices: %m");
2968 }
2969
2970 loop_device_relinquish(d);
2971
2972 *ret_directory = TAKE_PTR(created_dir);
2973 *ret_loop_device = TAKE_PTR(d);
2974 *ret_decrypted_image = TAKE_PTR(decrypted_image);
2975
2976 return 0;
2977}
2978
8c1be37e 2979static const char *const partition_designator_table[] = {
68ac5118
ZJS
2980 [PARTITION_ROOT] = "root",
2981 [PARTITION_ROOT_SECONDARY] = "root-secondary",
2982 [PARTITION_ROOT_OTHER] = "root-other",
2983 [PARTITION_USR] = "usr",
2984 [PARTITION_USR_SECONDARY] = "usr-secondary",
2985 [PARTITION_USR_OTHER] = "usr-other",
2986 [PARTITION_HOME] = "home",
2987 [PARTITION_SRV] = "srv",
2988 [PARTITION_ESP] = "esp",
2989 [PARTITION_XBOOTLDR] = "xbootldr",
2990 [PARTITION_SWAP] = "swap",
2991 [PARTITION_ROOT_VERITY] = "root-verity",
2992 [PARTITION_ROOT_SECONDARY_VERITY] = "root-secondary-verity",
2993 [PARTITION_ROOT_OTHER_VERITY] = "root-other-verity",
2994 [PARTITION_USR_VERITY] = "usr-verity",
2995 [PARTITION_USR_SECONDARY_VERITY] = "usr-secondary-verity",
2996 [PARTITION_USR_OTHER_VERITY] = "usr-other-verity",
2997 [PARTITION_ROOT_VERITY_SIG] = "root-verity-sig",
8ee9615e 2998 [PARTITION_ROOT_SECONDARY_VERITY_SIG] = "root-secondary-verity-sig",
68ac5118
ZJS
2999 [PARTITION_ROOT_OTHER_VERITY_SIG] = "root-other-verity-sig",
3000 [PARTITION_USR_VERITY_SIG] = "usr-verity-sig",
3001 [PARTITION_USR_SECONDARY_VERITY_SIG] = "usr-secondary-verity-sig",
3002 [PARTITION_USR_OTHER_VERITY_SIG] = "usr-other-verity-sig",
3003 [PARTITION_TMP] = "tmp",
3004 [PARTITION_VAR] = "var",
8c1be37e
LP
3005};
3006
93f59701 3007int verity_dissect_and_mount(
cedf5b1a 3008 int src_fd,
93f59701
LB
3009 const char *src,
3010 const char *dest,
3011 const MountOptions *options,
3012 const char *required_host_os_release_id,
3013 const char *required_host_os_release_version_id,
60c5f700
LP
3014 const char *required_host_os_release_sysext_level,
3015 const char *required_sysext_scope) {
93f59701 3016
4beda316
LB
3017 _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL;
3018 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
3019 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
3020 _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT;
3021 DissectImageFlags dissect_image_flags;
3022 int r;
3023
3024 assert(src);
3025 assert(dest);
3026
cedf5b1a 3027 /* We might get an FD for the image, but we use the original path to look for the dm-verity files */
4beda316
LB
3028 r = verity_settings_load(&verity, src, NULL, NULL);
3029 if (r < 0)
3030 return log_debug_errno(r, "Failed to load root hash: %m");
3031
3032 dissect_image_flags = verity.data_path ? DISSECT_IMAGE_NO_PARTITION_TABLE : 0;
3033
cedf5b1a
LB
3034 /* Note that we don't use loop_device_make here, as the FD is most likely O_PATH which would not be
3035 * accepted by LOOP_CONFIGURE, so just let loop_device_make_by_path reopen it as a regular FD. */
4beda316 3036 r = loop_device_make_by_path(
cedf5b1a 3037 src_fd >= 0 ? FORMAT_PROC_FD_PATH(src_fd) : src,
4beda316
LB
3038 -1,
3039 verity.data_path ? 0 : LO_FLAGS_PARTSCAN,
7f52206a 3040 LOCK_SH,
4beda316
LB
3041 &loop_device);
3042 if (r < 0)
3043 return log_debug_errno(r, "Failed to create loop device for image: %m");
3044
bad31660
YW
3045 r = dissect_loop_device(
3046 loop_device,
4beda316
LB
3047 &verity,
3048 options,
3049 dissect_image_flags,
3050 &dissected_image);
3051 /* No partition table? Might be a single-filesystem image, try again */
3052 if (!verity.data_path && r == -ENOPKG)
bad31660
YW
3053 r = dissect_loop_device(
3054 loop_device,
4beda316
LB
3055 &verity,
3056 options,
75dc190d 3057 dissect_image_flags | DISSECT_IMAGE_NO_PARTITION_TABLE,
4beda316
LB
3058 &dissected_image);
3059 if (r < 0)
3060 return log_debug_errno(r, "Failed to dissect image: %m");
3061
88b3300f
LP
3062 r = dissected_image_load_verity_sig_partition(dissected_image, loop_device->fd, &verity);
3063 if (r < 0)
3064 return r;
3065
4beda316
LB
3066 r = dissected_image_decrypt(
3067 dissected_image,
3068 NULL,
3069 &verity,
3070 dissect_image_flags,
3071 &decrypted_image);
3072 if (r < 0)
3073 return log_debug_errno(r, "Failed to decrypt dissected image: %m");
3074
3075 r = mkdir_p_label(dest, 0755);
3076 if (r < 0)
3077 return log_debug_errno(r, "Failed to create destination directory %s: %m", dest);
3078 r = umount_recursive(dest, 0);
3079 if (r < 0)
3080 return log_debug_errno(r, "Failed to umount under destination directory %s: %m", dest);
3081
21b61b1d 3082 r = dissected_image_mount(dissected_image, dest, UID_INVALID, UID_INVALID, dissect_image_flags);
4beda316
LB
3083 if (r < 0)
3084 return log_debug_errno(r, "Failed to mount image: %m");
3085
41bc4849
LP
3086 r = loop_device_flock(loop_device, LOCK_UN);
3087 if (r < 0)
3088 return log_debug_errno(r, "Failed to unlock loopback device: %m");
3089
93f59701
LB
3090 /* If we got os-release values from the caller, then we need to match them with the image's
3091 * extension-release.d/ content. Return -EINVAL if there's any mismatch.
3092 * First, check the distro ID. If that matches, then check the new SYSEXT_LEVEL value if
37361f46
LB
3093 * available, or else fallback to VERSION_ID. If neither is present (eg: rolling release),
3094 * then a simple match on the ID will be performed. */
8b2dcbbd 3095 if (required_host_os_release_id) {
93f59701
LB
3096 _cleanup_strv_free_ char **extension_release = NULL;
3097
d30d86b7
YW
3098 assert(!isempty(required_host_os_release_id));
3099
93f59701
LB
3100 r = load_extension_release_pairs(dest, dissected_image->image_name, &extension_release);
3101 if (r < 0)
3102 return log_debug_errno(r, "Failed to parse image %s extension-release metadata: %m", dissected_image->image_name);
3103
3104 r = extension_release_validate(
60c5f700
LP
3105 dissected_image->image_name,
3106 required_host_os_release_id,
3107 required_host_os_release_version_id,
3108 required_host_os_release_sysext_level,
3109 required_sysext_scope,
3110 extension_release);
93f59701
LB
3111 if (r == 0)
3112 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE), "Image %s extension-release metadata does not match the root's", dissected_image->image_name);
3113 if (r < 0)
3114 return log_debug_errno(r, "Failed to compare image %s extension-release metadata with the root's os-release: %m", dissected_image->image_name);
3115 }
3116
4beda316
LB
3117 if (decrypted_image) {
3118 r = decrypted_image_relinquish(decrypted_image);
3119 if (r < 0)
3120 return log_debug_errno(r, "Failed to relinquish decrypted image: %m");
3121 }
3122
3123 loop_device_relinquish(loop_device);
3124
3125 return 0;
3126}
3127
569a0e42 3128DEFINE_STRING_TABLE_LOOKUP(partition_designator, PartitionDesignator);