]> git.ipfire.org Git - thirdparty/systemd.git/blame - src/shared/dissect-image.c
dissect-image: reuse LoopDevice.node in dissect_image()
[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) {
08fe0a53 726 _cleanup_free_ char *t = NULL, *n = 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
748 n = strdup(node);
749 if (!n)
750 return -ENOMEM;
751
08fe0a53
LP
752 if (label) {
753 l = strdup(label);
754 if (!l)
755 return -ENOMEM;
756 }
757
f5215bc8 758 options = mount_options_from_designator(mount_options, designator);
18d73705
LB
759 if (options) {
760 o = strdup(options);
761 if (!o)
762 return -ENOMEM;
763 }
764
8c1be37e
LP
765 m->partitions[designator] = (DissectedPartition) {
766 .found = true,
767 .partno = nr,
768 .rw = rw,
de98f631 769 .growfs = growfs,
8c1be37e 770 .architecture = architecture,
1cc6c93a
YW
771 .node = TAKE_PTR(n),
772 .fstype = TAKE_PTR(t),
08fe0a53 773 .label = TAKE_PTR(l),
be30ad41 774 .uuid = id,
18d73705 775 .mount_options = TAKE_PTR(o),
88b3300f
LP
776 .offset = (uint64_t) start * 512,
777 .size = (uint64_t) size * 512,
8c1be37e 778 };
8c1be37e
LP
779 }
780
781 } else if (is_mbr) {
782
a8c47660 783 switch (blkid_partition_get_type(pp)) {
8c1be37e 784
a8c47660
LP
785 case 0x83: /* Linux partition */
786
787 if (pflags != 0x80) /* Bootable flag */
788 continue;
8c1be37e 789
a8c47660
LP
790 if (generic_node)
791 multiple_generic = true;
792 else {
793 generic_nr = nr;
794 generic_rw = true;
de98f631 795 generic_growfs = false;
a8c47660
LP
796 generic_node = strdup(node);
797 if (!generic_node)
798 return -ENOMEM;
799 }
800
801 break;
802
803 case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
18d73705 804 _cleanup_free_ char *n = NULL, *o = NULL;
a8c47660 805 sd_id128_t id = SD_ID128_NULL;
18d73705 806 const char *sid, *options = NULL;
a8c47660
LP
807
808 /* First one wins */
234c2e16 809 if (m->partitions[PARTITION_XBOOTLDR].found)
a8c47660
LP
810 continue;
811
812 sid = blkid_partition_get_uuid(pp);
813 if (sid)
814 (void) sd_id128_from_string(sid, &id);
815
816 n = strdup(node);
817 if (!n)
8c1be37e 818 return -ENOMEM;
a8c47660 819
f5215bc8 820 options = mount_options_from_designator(mount_options, PARTITION_XBOOTLDR);
18d73705
LB
821 if (options) {
822 o = strdup(options);
823 if (!o)
824 return -ENOMEM;
825 }
826
a8c47660
LP
827 m->partitions[PARTITION_XBOOTLDR] = (DissectedPartition) {
828 .found = true,
829 .partno = nr,
830 .rw = true,
de98f631 831 .growfs = false,
a8c47660
LP
832 .architecture = _ARCHITECTURE_INVALID,
833 .node = TAKE_PTR(n),
834 .uuid = id,
18d73705 835 .mount_options = TAKE_PTR(o),
88b3300f
LP
836 .offset = (uint64_t) start * 512,
837 .size = (uint64_t) size * 512,
a8c47660
LP
838 };
839
840 break;
841 }}
8c1be37e
LP
842 }
843 }
844
74cb2db9 845 if (m->partitions[PARTITION_ROOT].found) {
49ae9d91
DDM
846 /* If we found the primary arch, then invalidate the secondary and other arch to avoid any
847 * ambiguities, since we never want to mount the secondary or other arch in this case. */
74cb2db9
LP
848 m->partitions[PARTITION_ROOT_SECONDARY].found = false;
849 m->partitions[PARTITION_ROOT_SECONDARY_VERITY].found = false;
8ee9615e 850 m->partitions[PARTITION_ROOT_SECONDARY_VERITY_SIG].found = false;
aee36b4e
LP
851 m->partitions[PARTITION_USR_SECONDARY].found = false;
852 m->partitions[PARTITION_USR_SECONDARY_VERITY].found = false;
8ee9615e 853 m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG].found = false;
8c1be37e 854
49ae9d91
DDM
855 m->partitions[PARTITION_ROOT_OTHER].found = false;
856 m->partitions[PARTITION_ROOT_OTHER_VERITY].found = false;
857 m->partitions[PARTITION_ROOT_OTHER_VERITY_SIG].found = false;
858 m->partitions[PARTITION_USR_OTHER].found = false;
859 m->partitions[PARTITION_USR_OTHER_VERITY].found = false;
860 m->partitions[PARTITION_USR_OTHER_VERITY_SIG].found = false;
861
8ee9615e
LP
862 } else if (m->partitions[PARTITION_ROOT_VERITY].found ||
863 m->partitions[PARTITION_ROOT_VERITY_SIG].found)
7cf66030 864 return -EADDRNOTAVAIL; /* Verity found but no matching rootfs? Something is off, refuse. */
4623e8e6 865
7cf66030 866 else if (m->partitions[PARTITION_ROOT_SECONDARY].found) {
aee36b4e 867
7cf66030 868 /* No root partition found but there's one for the secondary architecture? Then upgrade
49ae9d91
DDM
869 * secondary arch to first and invalidate the other arch. */
870
871 log_debug("No root partition found of the native architecture, falling back to a root "
872 "partition of the secondary architecture.");
4623e8e6 873
7cf66030
LP
874 m->partitions[PARTITION_ROOT] = m->partitions[PARTITION_ROOT_SECONDARY];
875 zero(m->partitions[PARTITION_ROOT_SECONDARY]);
876 m->partitions[PARTITION_ROOT_VERITY] = m->partitions[PARTITION_ROOT_SECONDARY_VERITY];
877 zero(m->partitions[PARTITION_ROOT_SECONDARY_VERITY]);
8ee9615e
LP
878 m->partitions[PARTITION_ROOT_VERITY_SIG] = m->partitions[PARTITION_ROOT_SECONDARY_VERITY_SIG];
879 zero(m->partitions[PARTITION_ROOT_SECONDARY_VERITY_SIG]);
aee36b4e 880
7cf66030
LP
881 m->partitions[PARTITION_USR] = m->partitions[PARTITION_USR_SECONDARY];
882 zero(m->partitions[PARTITION_USR_SECONDARY]);
883 m->partitions[PARTITION_USR_VERITY] = m->partitions[PARTITION_USR_SECONDARY_VERITY];
884 zero(m->partitions[PARTITION_USR_SECONDARY_VERITY]);
8ee9615e
LP
885 m->partitions[PARTITION_USR_VERITY_SIG] = m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG];
886 zero(m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG]);
e0f9e7bd 887
49ae9d91
DDM
888 m->partitions[PARTITION_ROOT_OTHER].found = false;
889 m->partitions[PARTITION_ROOT_OTHER_VERITY].found = false;
890 m->partitions[PARTITION_ROOT_OTHER_VERITY_SIG].found = false;
891 m->partitions[PARTITION_USR_OTHER].found = false;
892 m->partitions[PARTITION_USR_OTHER_VERITY].found = false;
893 m->partitions[PARTITION_USR_OTHER_VERITY_SIG].found = false;
894
8ee9615e
LP
895 } else if (m->partitions[PARTITION_ROOT_SECONDARY_VERITY].found ||
896 m->partitions[PARTITION_ROOT_SECONDARY_VERITY_SIG].found)
7cf66030 897 return -EADDRNOTAVAIL; /* as above */
18d73705 898
49ae9d91
DDM
899 else if (m->partitions[PARTITION_ROOT_OTHER].found) {
900
901 /* No root or secondary partition found but there's one for another architecture? Then
902 * upgrade the other architecture to first. */
903
904 log_debug("No root partition found of the native architecture or the secondary architecture, "
905 "falling back to a root partition of a non-native architecture (%s).",
906 architecture_to_string(m->partitions[PARTITION_ROOT_OTHER].architecture));
907
908 m->partitions[PARTITION_ROOT] = m->partitions[PARTITION_ROOT_OTHER];
909 zero(m->partitions[PARTITION_ROOT_OTHER]);
910 m->partitions[PARTITION_ROOT_VERITY] = m->partitions[PARTITION_ROOT_OTHER_VERITY];
911 zero(m->partitions[PARTITION_ROOT_OTHER_VERITY]);
912 m->partitions[PARTITION_ROOT_VERITY_SIG] = m->partitions[PARTITION_ROOT_OTHER_VERITY_SIG];
913 zero(m->partitions[PARTITION_ROOT_OTHER_VERITY_SIG]);
914
915 m->partitions[PARTITION_USR] = m->partitions[PARTITION_USR_OTHER];
916 zero(m->partitions[PARTITION_USR_OTHER]);
917 m->partitions[PARTITION_USR_VERITY] = m->partitions[PARTITION_USR_OTHER_VERITY];
918 zero(m->partitions[PARTITION_USR_OTHER_VERITY]);
919 m->partitions[PARTITION_USR_VERITY_SIG] = m->partitions[PARTITION_USR_OTHER_VERITY_SIG];
920 zero(m->partitions[PARTITION_USR_OTHER_VERITY_SIG]);
921 }
922
8ee9615e
LP
923 /* Hmm, we found a signature partition but no Verity data? Something is off. */
924 if (m->partitions[PARTITION_ROOT_VERITY_SIG].found && !m->partitions[PARTITION_ROOT_VERITY].found)
925 return -EADDRNOTAVAIL;
7cf66030 926
8ee9615e 927 if (m->partitions[PARTITION_USR].found) {
49ae9d91 928 /* Invalidate secondary and other arch /usr/ if we found the primary arch */
7cf66030
LP
929 m->partitions[PARTITION_USR_SECONDARY].found = false;
930 m->partitions[PARTITION_USR_SECONDARY_VERITY].found = false;
8ee9615e 931 m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG].found = false;
7cf66030 932
49ae9d91
DDM
933 m->partitions[PARTITION_USR_OTHER].found = false;
934 m->partitions[PARTITION_USR_OTHER_VERITY].found = false;
935 m->partitions[PARTITION_USR_OTHER_VERITY_SIG].found = false;
936
8ee9615e
LP
937 } else if (m->partitions[PARTITION_USR_VERITY].found ||
938 m->partitions[PARTITION_USR_VERITY_SIG].found)
7cf66030 939 return -EADDRNOTAVAIL; /* as above */
8c1be37e 940
7cf66030 941 else if (m->partitions[PARTITION_USR_SECONDARY].found) {
e0f9e7bd 942
49ae9d91
DDM
943 log_debug("No usr partition found of the native architecture, falling back to a usr "
944 "partition of the secondary architecture.");
945
7cf66030
LP
946 /* Upgrade secondary arch to primary */
947 m->partitions[PARTITION_USR] = m->partitions[PARTITION_USR_SECONDARY];
948 zero(m->partitions[PARTITION_USR_SECONDARY]);
949 m->partitions[PARTITION_USR_VERITY] = m->partitions[PARTITION_USR_SECONDARY_VERITY];
950 zero(m->partitions[PARTITION_USR_SECONDARY_VERITY]);
8ee9615e
LP
951 m->partitions[PARTITION_USR_VERITY_SIG] = m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG];
952 zero(m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG]);
7cf66030 953
49ae9d91
DDM
954 m->partitions[PARTITION_USR_OTHER].found = false;
955 m->partitions[PARTITION_USR_OTHER_VERITY].found = false;
956 m->partitions[PARTITION_USR_OTHER_VERITY_SIG].found = false;
957
8ee9615e
LP
958 } else if (m->partitions[PARTITION_USR_SECONDARY_VERITY].found ||
959 m->partitions[PARTITION_USR_SECONDARY_VERITY_SIG].found)
7cf66030
LP
960 return -EADDRNOTAVAIL; /* as above */
961
49ae9d91
DDM
962 else if (m->partitions[PARTITION_USR_OTHER].found) {
963
964 log_debug("No usr partition found of the native architecture or the secondary architecture, "
965 "falling back to a usr partition of a non-native architecture (%s).",
966 architecture_to_string(m->partitions[PARTITION_ROOT_OTHER].architecture));
967
968 /* Upgrade other arch to primary */
969 m->partitions[PARTITION_USR] = m->partitions[PARTITION_USR_OTHER];
970 zero(m->partitions[PARTITION_USR_OTHER]);
971 m->partitions[PARTITION_USR_VERITY] = m->partitions[PARTITION_USR_OTHER_VERITY];
972 zero(m->partitions[PARTITION_USR_OTHER_VERITY]);
973 m->partitions[PARTITION_USR_VERITY_SIG] = m->partitions[PARTITION_USR_OTHER_VERITY_SIG];
974 zero(m->partitions[PARTITION_USR_OTHER_VERITY_SIG]);
975 }
976
8ee9615e
LP
977 /* Hmm, we found a signature partition but no Verity data? Something is off. */
978 if (m->partitions[PARTITION_USR_VERITY_SIG].found && !m->partitions[PARTITION_USR_VERITY].found)
979 return -EADDRNOTAVAIL;
980
cb241a69
LP
981 /* If root and /usr are combined then insist that the architecture matches */
982 if (m->partitions[PARTITION_ROOT].found &&
983 m->partitions[PARTITION_USR].found &&
984 (m->partitions[PARTITION_ROOT].architecture >= 0 &&
985 m->partitions[PARTITION_USR].architecture >= 0 &&
986 m->partitions[PARTITION_ROOT].architecture != m->partitions[PARTITION_USR].architecture))
987 return -EADDRNOTAVAIL;
988
4ab51780
LP
989 if (!m->partitions[PARTITION_ROOT].found &&
990 !m->partitions[PARTITION_USR].found &&
991 (flags & DISSECT_IMAGE_GENERIC_ROOT) &&
00a8b34f 992 (!verity || !verity->root_hash || verity->designator != PARTITION_USR)) {
7cf66030 993
1b010ae7 994 /* OK, we found nothing usable, then check if there's a single generic partition, and use
4b5de5dd
LP
995 * that. If the root hash was set however, then we won't fall back to a generic node, because
996 * the root hash decides. */
7cf66030
LP
997
998 /* If we didn't find a properly marked root partition, but we did find a single suitable
999 * generic Linux partition, then use this as root partition, if the caller asked for it. */
1000 if (multiple_generic)
1001 return -ENOTUNIQ;
1002
4b5de5dd
LP
1003 /* If we didn't find a generic node, then we can't fix this up either */
1004 if (generic_node) {
1005 _cleanup_free_ char *o = NULL;
1006 const char *options;
8c1be37e 1007
f5215bc8 1008 options = mount_options_from_designator(mount_options, PARTITION_ROOT);
18d73705
LB
1009 if (options) {
1010 o = strdup(options);
1011 if (!o)
1012 return -ENOMEM;
1013 }
1014
1f8fb21c 1015 assert(generic_nr >= 0);
8c1be37e
LP
1016 m->partitions[PARTITION_ROOT] = (DissectedPartition) {
1017 .found = true,
1018 .rw = generic_rw,
de98f631 1019 .growfs = generic_growfs,
8c1be37e
LP
1020 .partno = generic_nr,
1021 .architecture = _ARCHITECTURE_INVALID,
1cc6c93a 1022 .node = TAKE_PTR(generic_node),
be30ad41 1023 .uuid = generic_uuid,
18d73705 1024 .mount_options = TAKE_PTR(o),
88b3300f
LP
1025 .offset = UINT64_MAX,
1026 .size = UINT64_MAX,
8c1be37e 1027 };
e0f9e7bd 1028 }
8c1be37e
LP
1029 }
1030
4b5de5dd
LP
1031 /* 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 */
1032 if (FLAGS_SET(flags, DISSECT_IMAGE_REQUIRE_ROOT) &&
1033 !(m->partitions[PARTITION_ROOT].found || (m->partitions[PARTITION_USR].found && FLAGS_SET(flags, DISSECT_IMAGE_USR_NO_ROOT))))
1034 return -ENXIO;
1035
7b32164f
LP
1036 if (m->partitions[PARTITION_ROOT_VERITY].found) {
1037 /* We only support one verity partition per image, i.e. can't do for both /usr and root fs */
1038 if (m->partitions[PARTITION_USR_VERITY].found)
1039 return -ENOTUNIQ;
1040
1041 /* We don't support verity enabled root with a split out /usr. Neither with nor without
1042 * verity there. (Note that we do support verity-less root with verity-full /usr, though.) */
1043 if (m->partitions[PARTITION_USR].found)
1044 return -EADDRNOTAVAIL;
1045 }
aee36b4e 1046
1903defc
LP
1047 if (verity) {
1048 /* If a verity designator is specified, then insist that the matching partition exists */
1049 if (verity->designator >= 0 && !m->partitions[verity->designator].found)
1050 return -EADDRNOTAVAIL;
aee36b4e 1051
1903defc 1052 if (verity->root_hash) {
8ee9615e
LP
1053 /* If we have an explicit root hash and found the partitions for it, then we are ready to use
1054 * Verity, set things up for it */
1055
1903defc
LP
1056 if (verity->designator < 0 || verity->designator == PARTITION_ROOT) {
1057 if (!m->partitions[PARTITION_ROOT_VERITY].found || !m->partitions[PARTITION_ROOT].found)
1058 return -EADDRNOTAVAIL;
4623e8e6 1059
1903defc
LP
1060 /* If we found a verity setup, then the root partition is necessarily read-only. */
1061 m->partitions[PARTITION_ROOT].rw = false;
1062 m->verity_ready = true;
1903defc 1063
f9e0bb61
LP
1064 } else {
1065 assert(verity->designator == PARTITION_USR);
1066
1903defc
LP
1067 if (!m->partitions[PARTITION_USR_VERITY].found || !m->partitions[PARTITION_USR].found)
1068 return -EADDRNOTAVAIL;
4623e8e6 1069
1903defc
LP
1070 m->partitions[PARTITION_USR].rw = false;
1071 m->verity_ready = true;
1072 }
8ee9615e
LP
1073
1074 if (m->verity_ready)
b98416e1 1075 m->verity_sig_ready = verity->root_hash_sig;
8ee9615e
LP
1076
1077 } else if (m->partitions[verity->designator == PARTITION_USR ? PARTITION_USR_VERITY_SIG : PARTITION_ROOT_VERITY_SIG].found) {
1078
1079 /* If we found an embedded signature partition, we are ready, too. */
1080
1081 m->verity_ready = m->verity_sig_ready = true;
1082 m->partitions[verity->designator == PARTITION_USR ? PARTITION_USR : PARTITION_ROOT].rw = false;
aee36b4e 1083 }
4623e8e6
LP
1084 }
1085
18b5886e
LP
1086 blkid_free_probe(b);
1087 b = NULL;
1088
8c1be37e 1089 /* Fill in file system types if we don't know them yet. */
569a0e42 1090 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
18b5886e 1091 DissectedPartition *p = m->partitions + i;
8c1be37e 1092
18b5886e 1093 if (!p->found)
8c1be37e
LP
1094 continue;
1095
18b5886e
LP
1096 if (!p->fstype && p->node) {
1097 r = probe_filesystem(p->node, &p->fstype);
7cc84b2c 1098 if (r < 0 && r != -EUCLEAN)
18b5886e 1099 return r;
8c1be37e
LP
1100 }
1101
18b5886e
LP
1102 if (streq_ptr(p->fstype, "crypto_LUKS"))
1103 m->encrypted = true;
896f937f
LP
1104
1105 if (p->fstype && fstype_is_ro(p->fstype))
1106 p->rw = false;
de98f631
LP
1107
1108 if (!p->rw)
1109 p->growfs = false;
8c1be37e
LP
1110 }
1111
1cc6c93a 1112 *ret = TAKE_PTR(m);
8c1be37e
LP
1113 return 0;
1114#else
1115 return -EOPNOTSUPP;
1116#endif
1117}
1118
1119DissectedImage* dissected_image_unref(DissectedImage *m) {
8c1be37e
LP
1120 if (!m)
1121 return NULL;
1122
08fe0a53 1123 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++)
234c2e16 1124 dissected_partition_done(m->partitions + i);
8c1be37e 1125
593fe6c0 1126 free(m->image_name);
3b925504
LP
1127 free(m->hostname);
1128 strv_free(m->machine_info);
1129 strv_free(m->os_release);
7718ac97 1130 strv_free(m->extension_release);
3b925504 1131
5fecf46d 1132 return mfree(m);
8c1be37e
LP
1133}
1134
18b5886e 1135static int is_loop_device(const char *path) {
553e15f2 1136 char s[SYS_BLOCK_PATH_MAX("/../loop/")];
18b5886e
LP
1137 struct stat st;
1138
1139 assert(path);
1140
1141 if (stat(path, &st) < 0)
1142 return -errno;
1143
1144 if (!S_ISBLK(st.st_mode))
1145 return -ENOTBLK;
1146
553e15f2 1147 xsprintf_sys_block_path(s, "/loop/", st.st_dev);
18b5886e
LP
1148 if (access(s, F_OK) < 0) {
1149 if (errno != ENOENT)
1150 return -errno;
1151
1152 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
553e15f2 1153 xsprintf_sys_block_path(s, "/../loop/", st.st_dev);
18b5886e
LP
1154 if (access(s, F_OK) < 0)
1155 return errno == ENOENT ? false : -errno;
1156 }
1157
1158 return true;
1159}
1160
cf32c486
LP
1161static int run_fsck(const char *node, const char *fstype) {
1162 int r, exit_status;
1163 pid_t pid;
1164
1165 assert(node);
1166 assert(fstype);
1167
1168 r = fsck_exists(fstype);
1169 if (r < 0) {
1170 log_debug_errno(r, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype);
1171 return 0;
1172 }
1173 if (r == 0) {
1174 log_debug("Not checking partition %s, as fsck for %s does not exist.", node, fstype);
1175 return 0;
1176 }
1177
1178 r = safe_fork("(fsck)", FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG|FORK_NULL_STDIO, &pid);
1179 if (r < 0)
1180 return log_debug_errno(r, "Failed to fork off fsck: %m");
1181 if (r == 0) {
1182 /* Child */
1183 execl("/sbin/fsck", "/sbin/fsck", "-aT", node, NULL);
7e0ed2e9 1184 log_open();
cf32c486
LP
1185 log_debug_errno(errno, "Failed to execl() fsck: %m");
1186 _exit(FSCK_OPERATIONAL_ERROR);
1187 }
1188
1189 exit_status = wait_for_terminate_and_check("fsck", pid, 0);
1190 if (exit_status < 0)
1191 return log_debug_errno(exit_status, "Failed to fork off /sbin/fsck: %m");
1192
1193 if ((exit_status & ~FSCK_ERROR_CORRECTED) != FSCK_SUCCESS) {
1194 log_debug("fsck failed with exit status %i.", exit_status);
1195
1196 if ((exit_status & (FSCK_SYSTEM_SHOULD_REBOOT|FSCK_ERRORS_LEFT_UNCORRECTED)) != 0)
1197 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN), "File system is corrupted, refusing.");
1198
1199 log_debug("Ignoring fsck error.");
1200 }
1201
1202 return 0;
1203}
1204
81939d9d
LP
1205static int fs_grow(const char *node_path, const char *mount_path) {
1206 _cleanup_close_ int mount_fd = -1, node_fd = -1;
81939d9d
LP
1207 uint64_t size, newsize;
1208 int r;
1209
1210 node_fd = open(node_path, O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY);
1211 if (node_fd < 0)
1212 return log_debug_errno(errno, "Failed to open node device %s: %m", node_path);
1213
1214 if (ioctl(node_fd, BLKGETSIZE64, &size) != 0)
1215 return log_debug_errno(errno, "Failed to get block device size of %s: %m", node_path);
1216
1217 mount_fd = open(mount_path, O_RDONLY|O_DIRECTORY|O_CLOEXEC);
1218 if (mount_fd < 0)
1219 return log_debug_errno(errno, "Failed to open mountd file system %s: %m", mount_path);
1220
1221 log_debug("Resizing \"%s\" to %"PRIu64" bytes...", mount_path, size);
1222 r = resize_fs(mount_fd, size, &newsize);
1223 if (r < 0)
1224 return log_debug_errno(r, "Failed to resize \"%s\" to %"PRIu64" bytes: %m", mount_path, size);
1225
1226 if (newsize == size)
1227 log_debug("Successfully resized \"%s\" to %s bytes.",
2b59bf51 1228 mount_path, FORMAT_BYTES(newsize));
81939d9d
LP
1229 else {
1230 assert(newsize < size);
1231 log_debug("Successfully resized \"%s\" to %s bytes (%"PRIu64" bytes lost due to blocksize).",
2b59bf51 1232 mount_path, FORMAT_BYTES(newsize), size - newsize);
81939d9d
LP
1233 }
1234
1235 return 0;
1236}
1237
18b5886e
LP
1238static int mount_partition(
1239 DissectedPartition *m,
1240 const char *where,
1241 const char *directory,
2d3a5a73 1242 uid_t uid_shift,
21b61b1d 1243 uid_t uid_range,
18b5886e
LP
1244 DissectImageFlags flags) {
1245
2d3a5a73
LP
1246 _cleanup_free_ char *chased = NULL, *options = NULL;
1247 const char *p, *node, *fstype;
21b61b1d 1248 bool rw, remap_uid_gid = false;
2eedfd2d 1249 int r;
8c1be37e
LP
1250
1251 assert(m);
1252 assert(where);
1253
4dc28665 1254 /* Use decrypted node and matching fstype if available, otherwise use the original device */
18b5886e 1255 node = m->decrypted_node ?: m->node;
4dc28665 1256 fstype = m->decrypted_node ? m->decrypted_fstype: m->fstype;
18b5886e 1257
4dc28665 1258 if (!m->found || !node)
8c1be37e 1259 return 0;
4dc28665
LP
1260 if (!fstype)
1261 return -EAFNOSUPPORT;
8c1be37e 1262
68ac5118
ZJS
1263 /* We are looking at an encrypted partition? This either means stacked encryption, or the caller
1264 * didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error for this
1265 * case. */
4dc28665 1266 if (streq(fstype, "crypto_LUKS"))
fa45d12c 1267 return -EUNATCH;
18b5886e 1268
ef9c184d 1269 rw = m->rw && !(flags & DISSECT_IMAGE_MOUNT_READ_ONLY);
8c1be37e 1270
cf32c486
LP
1271 if (FLAGS_SET(flags, DISSECT_IMAGE_FSCK) && rw) {
1272 r = run_fsck(node, fstype);
1273 if (r < 0)
1274 return r;
1275 }
1276
2eedfd2d 1277 if (directory) {
334eb5b0
LP
1278 /* Automatically create missing mount points inside the image, if necessary. */
1279 r = mkdir_p_root(where, directory, uid_shift, (gid_t) uid_shift, 0755);
1280 if (r < 0 && r != -EROFS)
1281 return r;
1f0f82f1 1282
a5648b80 1283 r = chase_symlinks(directory, where, CHASE_PREFIX_ROOT, &chased, NULL);
2eedfd2d
LP
1284 if (r < 0)
1285 return r;
1286
1287 p = chased;
9842905e
LP
1288 } else {
1289 /* Create top-level mount if missing – but only if this is asked for. This won't modify the
1290 * image (as the branch above does) but the host hierarchy, and the created directory might
1291 * survive our mount in the host hierarchy hence. */
1292 if (FLAGS_SET(flags, DISSECT_IMAGE_MKDIR)) {
1293 r = mkdir_p(where, 0755);
1294 if (r < 0)
1295 return r;
1296 }
1297
8c1be37e 1298 p = where;
9842905e 1299 }
8c1be37e 1300
18b5886e 1301 /* If requested, turn on discard support. */
154d2269 1302 if (fstype_can_discard(fstype) &&
18b5886e 1303 ((flags & DISSECT_IMAGE_DISCARD) ||
3afda7c7 1304 ((flags & DISSECT_IMAGE_DISCARD_ON_LOOP) && is_loop_device(m->node) > 0))) {
2d3a5a73
LP
1305 options = strdup("discard");
1306 if (!options)
1307 return -ENOMEM;
1308 }
1309
21b61b1d 1310 if (uid_is_valid(uid_shift) && uid_shift != 0) {
2d3a5a73 1311
21b61b1d
LP
1312 if (fstype_can_uid_gid(fstype)) {
1313 _cleanup_free_ char *uid_option = NULL;
2d3a5a73 1314
21b61b1d
LP
1315 if (asprintf(&uid_option, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0)
1316 return -ENOMEM;
1317
1318 if (!strextend_with_separator(&options, ",", uid_option))
1319 return -ENOMEM;
1320 } else if (FLAGS_SET(flags, DISSECT_IMAGE_MOUNT_IDMAPPED))
1321 remap_uid_gid = true;
2d3a5a73 1322 }
8c1be37e 1323
18d73705 1324 if (!isempty(m->mount_options))
c2bc710b 1325 if (!strextend_with_separator(&options, ",", m->mount_options))
18d73705
LB
1326 return -ENOMEM;
1327
b620bf33
LP
1328 /* So, when you request MS_RDONLY from ext4, then this means nothing. It happily still writes to the
1329 * backing storage. What's worse, the BLKRO[GS]ET flag and (in case of loopback devices)
1330 * LO_FLAGS_READ_ONLY don't mean anything, they affect userspace accesses only, and write accesses
1331 * from the upper file system still get propagated through to the underlying file system,
1332 * unrestricted. To actually get ext4/xfs/btrfs to stop writing to the device we need to specify
1333 * "norecovery" as mount option, in addition to MS_RDONLY. Yes, this sucks, since it means we need to
1334 * carry a per file system table here.
1335 *
1336 * Note that this means that we might not be able to mount corrupted file systems as read-only
1337 * anymore (since in some cases the kernel implementations will refuse mounting when corrupted,
1338 * read-only and "norecovery" is specified). But I think for the case of automatically determined
1339 * mount options for loopback devices this is the right choice, since otherwise using the same
1340 * loopback file twice even in read-only mode, is going to fail badly sooner or later. The usecase of
1341 * making reuse of the immutable images "just work" is more relevant to us than having read-only
1342 * access that actually modifies stuff work on such image files. Or to say this differently: if
1343 * people want their file systems to be fixed up they should just open them in writable mode, where
1344 * all these problems don't exist. */
1345 if (!rw && STRPTR_IN_SET(fstype, "ext3", "ext4", "xfs", "btrfs"))
1346 if (!strextend_with_separator(&options, ",", "norecovery"))
1347 return -ENOMEM;
1348
511a8cfe 1349 r = mount_nofollow_verbose(LOG_DEBUG, node, p, fstype, MS_NODEV|(rw ? 0 : MS_RDONLY), options);
d9223c07
LP
1350 if (r < 0)
1351 return r;
1352
81939d9d
LP
1353 if (rw && m->growfs && FLAGS_SET(flags, DISSECT_IMAGE_GROWFS))
1354 (void) fs_grow(node, p);
1355
21b61b1d 1356 if (remap_uid_gid) {
2b2777ed 1357 r = remount_idmap(p, uid_shift, uid_range, UID_INVALID, REMOUNT_IDMAPPING_HOST_ROOT);
21b61b1d
LP
1358 if (r < 0)
1359 return r;
1360 }
1361
d9223c07 1362 return 1;
8c1be37e
LP
1363}
1364
7cf66030
LP
1365static int mount_root_tmpfs(const char *where, uid_t uid_shift, DissectImageFlags flags) {
1366 _cleanup_free_ char *options = NULL;
1367 int r;
1368
1369 assert(where);
1370
1371 /* For images that contain /usr/ but no rootfs, let's mount rootfs as tmpfs */
1372
1373 if (FLAGS_SET(flags, DISSECT_IMAGE_MKDIR)) {
1374 r = mkdir_p(where, 0755);
1375 if (r < 0)
1376 return r;
1377 }
1378
1379 if (uid_is_valid(uid_shift)) {
1380 if (asprintf(&options, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0)
1381 return -ENOMEM;
1382 }
1383
1384 r = mount_nofollow_verbose(LOG_DEBUG, "rootfs", where, "tmpfs", MS_NODEV, options);
1385 if (r < 0)
1386 return r;
1387
1388 return 1;
1389}
1390
21b61b1d
LP
1391int dissected_image_mount(
1392 DissectedImage *m,
1393 const char *where,
1394 uid_t uid_shift,
1395 uid_t uid_range,
1396 DissectImageFlags flags) {
1397
1f0f82f1 1398 int r, xbootldr_mounted;
8c1be37e
LP
1399
1400 assert(m);
1401 assert(where);
1402
fa45d12c
LP
1403 /* Returns:
1404 *
1405 * -ENXIO → No root partition found
7718ac97 1406 * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release/extension-release file found
fa45d12c
LP
1407 * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet
1408 * -EUCLEAN → fsck for file system failed
1409 * -EBUSY → File system already mounted/used elsewhere (kernel)
4dc28665 1410 * -EAFNOSUPPORT → File system type not supported or not known
fa45d12c
LP
1411 */
1412
7cf66030
LP
1413 if (!(m->partitions[PARTITION_ROOT].found ||
1414 (m->partitions[PARTITION_USR].found && FLAGS_SET(flags, DISSECT_IMAGE_USR_NO_ROOT))))
1415 return -ENXIO; /* Require a root fs or at least a /usr/ fs (the latter is subject to a flag of its own) */
8c1be37e 1416
2d3a5a73 1417 if ((flags & DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY) == 0) {
7cf66030
LP
1418
1419 /* First mount the root fs. If there's none we use a tmpfs. */
1420 if (m->partitions[PARTITION_ROOT].found)
21b61b1d 1421 r = mount_partition(m->partitions + PARTITION_ROOT, where, NULL, uid_shift, uid_range, flags);
7cf66030
LP
1422 else
1423 r = mount_root_tmpfs(where, uid_shift, flags);
2d3a5a73
LP
1424 if (r < 0)
1425 return r;
aee36b4e 1426
aee36b4e 1427 /* For us mounting root always means mounting /usr as well */
21b61b1d 1428 r = mount_partition(m->partitions + PARTITION_USR, where, "/usr", uid_shift, uid_range, flags);
aee36b4e
LP
1429 if (r < 0)
1430 return r;
03bcb6d4 1431
9ccb531a
LB
1432 if ((flags & (DISSECT_IMAGE_VALIDATE_OS|DISSECT_IMAGE_VALIDATE_OS_EXT)) != 0) {
1433 /* If either one of the validation flags are set, ensure that the image qualifies
1434 * as one or the other (or both). */
1435 bool ok = false;
1436
1437 if (FLAGS_SET(flags, DISSECT_IMAGE_VALIDATE_OS)) {
1438 r = path_is_os_tree(where);
1439 if (r < 0)
1440 return r;
1441 if (r > 0)
1442 ok = true;
1443 }
1444 if (!ok && FLAGS_SET(flags, DISSECT_IMAGE_VALIDATE_OS_EXT)) {
7718ac97
LB
1445 r = path_is_extension_tree(where, m->image_name);
1446 if (r < 0)
1447 return r;
9ccb531a
LB
1448 if (r > 0)
1449 ok = true;
7718ac97 1450 }
9ccb531a
LB
1451
1452 if (!ok)
1453 return -ENOMEDIUM;
03bcb6d4 1454 }
2d3a5a73
LP
1455 }
1456
705727fd 1457 if (flags & DISSECT_IMAGE_MOUNT_ROOT_ONLY)
2d3a5a73 1458 return 0;
8c1be37e 1459
21b61b1d 1460 r = mount_partition(m->partitions + PARTITION_HOME, where, "/home", uid_shift, uid_range, flags);
8c1be37e
LP
1461 if (r < 0)
1462 return r;
1463
21b61b1d 1464 r = mount_partition(m->partitions + PARTITION_SRV, where, "/srv", uid_shift, uid_range, flags);
8c1be37e
LP
1465 if (r < 0)
1466 return r;
1467
21b61b1d 1468 r = mount_partition(m->partitions + PARTITION_VAR, where, "/var", uid_shift, uid_range, flags);
d4dffb85
LP
1469 if (r < 0)
1470 return r;
1471
21b61b1d 1472 r = mount_partition(m->partitions + PARTITION_TMP, where, "/var/tmp", uid_shift, uid_range, flags);
d4dffb85
LP
1473 if (r < 0)
1474 return r;
1475
21b61b1d 1476 xbootldr_mounted = mount_partition(m->partitions + PARTITION_XBOOTLDR, where, "/boot", uid_shift, uid_range, flags);
1f0f82f1
LP
1477 if (xbootldr_mounted < 0)
1478 return xbootldr_mounted;
d9223c07 1479
8c1be37e 1480 if (m->partitions[PARTITION_ESP].found) {
1f0f82f1
LP
1481 int esp_done = false;
1482
d9223c07
LP
1483 /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it
1484 * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */
8c1be37e 1485
a5648b80 1486 r = chase_symlinks("/efi", where, CHASE_PREFIX_ROOT, NULL, NULL);
1f0f82f1
LP
1487 if (r < 0) {
1488 if (r != -ENOENT)
d9223c07 1489 return r;
8c1be37e 1490
1f0f82f1
LP
1491 /* /efi doesn't exist. Let's see if /boot is suitable then */
1492
1493 if (!xbootldr_mounted) {
1494 _cleanup_free_ char *p = NULL;
2eedfd2d 1495
1f0f82f1
LP
1496 r = chase_symlinks("/boot", where, CHASE_PREFIX_ROOT, &p, NULL);
1497 if (r < 0) {
1498 if (r != -ENOENT)
1499 return r;
db55bbf2 1500 } else if (dir_is_empty(p, /* ignore_hidden_or_backup= */ false) > 0) {
1f0f82f1 1501 /* It exists and is an empty directory. Let's mount the ESP there. */
21b61b1d 1502 r = mount_partition(m->partitions + PARTITION_ESP, where, "/boot", uid_shift, uid_range, flags);
1f0f82f1
LP
1503 if (r < 0)
1504 return r;
1505
1506 esp_done = true;
1507 }
2eedfd2d 1508 }
8c1be37e 1509 }
1f0f82f1
LP
1510
1511 if (!esp_done) {
1512 /* OK, let's mount the ESP now to /efi (possibly creating the dir if missing) */
1513
21b61b1d 1514 r = mount_partition(m->partitions + PARTITION_ESP, where, "/efi", uid_shift, uid_range, flags);
1f0f82f1
LP
1515 if (r < 0)
1516 return r;
1517 }
8c1be37e
LP
1518 }
1519
1520 return 0;
1521}
1522
21b61b1d
LP
1523int dissected_image_mount_and_warn(
1524 DissectedImage *m,
1525 const char *where,
1526 uid_t uid_shift,
1527 uid_t uid_range,
1528 DissectImageFlags flags) {
1529
af187ab2
LP
1530 int r;
1531
1532 assert(m);
1533 assert(where);
1534
21b61b1d 1535 r = dissected_image_mount(m, where, uid_shift, uid_range, flags);
af187ab2
LP
1536 if (r == -ENXIO)
1537 return log_error_errno(r, "Not root file system found in image.");
1538 if (r == -EMEDIUMTYPE)
7718ac97 1539 return log_error_errno(r, "No suitable os-release/extension-release file in image found.");
af187ab2
LP
1540 if (r == -EUNATCH)
1541 return log_error_errno(r, "Encrypted file system discovered, but decryption not requested.");
1542 if (r == -EUCLEAN)
1543 return log_error_errno(r, "File system check on image failed.");
1544 if (r == -EBUSY)
1545 return log_error_errno(r, "File system already mounted elsewhere.");
4dc28665
LP
1546 if (r == -EAFNOSUPPORT)
1547 return log_error_errno(r, "File system type not supported or not known.");
af187ab2
LP
1548 if (r < 0)
1549 return log_error_errno(r, "Failed to mount image: %m");
1550
1551 return r;
1552}
1553
349cc4a5 1554#if HAVE_LIBCRYPTSETUP
18b5886e
LP
1555typedef struct DecryptedPartition {
1556 struct crypt_device *device;
1557 char *name;
1558 bool relinquished;
1559} DecryptedPartition;
1560
1561struct DecryptedImage {
1562 DecryptedPartition *decrypted;
1563 size_t n_decrypted;
18b5886e
LP
1564};
1565#endif
1566
1567DecryptedImage* decrypted_image_unref(DecryptedImage* d) {
349cc4a5 1568#if HAVE_LIBCRYPTSETUP
18b5886e
LP
1569 int r;
1570
1571 if (!d)
1572 return NULL;
1573
67f63ee5 1574 for (size_t i = 0; i < d->n_decrypted; i++) {
18b5886e
LP
1575 DecryptedPartition *p = d->decrypted + i;
1576
1577 if (p->device && p->name && !p->relinquished) {
0d12936d 1578 r = sym_crypt_deactivate_by_name(p->device, p->name, 0);
18b5886e
LP
1579 if (r < 0)
1580 log_debug_errno(r, "Failed to deactivate encrypted partition %s", p->name);
1581 }
1582
1583 if (p->device)
0d12936d 1584 sym_crypt_free(p->device);
18b5886e
LP
1585 free(p->name);
1586 }
1587
f91861e4 1588 free(d->decrypted);
18b5886e
LP
1589 free(d);
1590#endif
1591 return NULL;
1592}
1593
349cc4a5 1594#if HAVE_LIBCRYPTSETUP
4623e8e6
LP
1595
1596static int make_dm_name_and_node(const void *original_node, const char *suffix, char **ret_name, char **ret_node) {
1597 _cleanup_free_ char *name = NULL, *node = NULL;
1598 const char *base;
1599
1600 assert(original_node);
1601 assert(suffix);
1602 assert(ret_name);
1603 assert(ret_node);
1604
1605 base = strrchr(original_node, '/');
1606 if (!base)
ac1f3ad0
LB
1607 base = original_node;
1608 else
1609 base++;
4623e8e6
LP
1610 if (isempty(base))
1611 return -EINVAL;
1612
1613 name = strjoin(base, suffix);
1614 if (!name)
1615 return -ENOMEM;
1616 if (!filename_is_valid(name))
1617 return -EINVAL;
1618
0d12936d 1619 node = path_join(sym_crypt_get_dir(), name);
4623e8e6
LP
1620 if (!node)
1621 return -ENOMEM;
1622
1cc6c93a
YW
1623 *ret_name = TAKE_PTR(name);
1624 *ret_node = TAKE_PTR(node);
4623e8e6 1625
4623e8e6
LP
1626 return 0;
1627}
1628
18b5886e
LP
1629static int decrypt_partition(
1630 DissectedPartition *m,
1631 const char *passphrase,
1632 DissectImageFlags flags,
1633 DecryptedImage *d) {
1634
1635 _cleanup_free_ char *node = NULL, *name = NULL;
0d12936d 1636 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
18b5886e
LP
1637 int r;
1638
1639 assert(m);
1640 assert(d);
1641
1642 if (!m->found || !m->node || !m->fstype)
1643 return 0;
1644
1645 if (!streq(m->fstype, "crypto_LUKS"))
1646 return 0;
1647
bdd73ac5
ZJS
1648 if (!passphrase)
1649 return -ENOKEY;
1650
0d12936d
LP
1651 r = dlopen_cryptsetup();
1652 if (r < 0)
1653 return r;
1654
4623e8e6
LP
1655 r = make_dm_name_and_node(m->node, "-decrypted", &name, &node);
1656 if (r < 0)
1657 return r;
18b5886e 1658
319a4f4b 1659 if (!GREEDY_REALLOC0(d->decrypted, d->n_decrypted + 1))
18b5886e
LP
1660 return -ENOMEM;
1661
0d12936d 1662 r = sym_crypt_init(&cd, m->node);
18b5886e 1663 if (r < 0)
715cbb81 1664 return log_debug_errno(r, "Failed to initialize dm-crypt: %m");
18b5886e 1665
efc3b12f 1666 cryptsetup_enable_logging(cd);
1887032f 1667
0d12936d 1668 r = sym_crypt_load(cd, CRYPT_LUKS, NULL);
294bd454
ZJS
1669 if (r < 0)
1670 return log_debug_errno(r, "Failed to load LUKS metadata: %m");
18b5886e 1671
0d12936d 1672 r = sym_crypt_activate_by_passphrase(cd, name, CRYPT_ANY_SLOT, passphrase, strlen(passphrase),
ef9c184d 1673 ((flags & DISSECT_IMAGE_DEVICE_READ_ONLY) ? CRYPT_ACTIVATE_READONLY : 0) |
0d12936d 1674 ((flags & DISSECT_IMAGE_DISCARD_ON_CRYPTO) ? CRYPT_ACTIVATE_ALLOW_DISCARDS : 0));
294bd454 1675 if (r < 0) {
715cbb81 1676 log_debug_errno(r, "Failed to activate LUKS device: %m");
294bd454 1677 return r == -EPERM ? -EKEYREJECTED : r;
18b5886e 1678 }
18b5886e 1679
94344385
LP
1680 d->decrypted[d->n_decrypted++] = (DecryptedPartition) {
1681 .name = TAKE_PTR(name),
1682 .device = TAKE_PTR(cd),
1683 };
18b5886e 1684
1cc6c93a 1685 m->decrypted_node = TAKE_PTR(node);
18b5886e
LP
1686
1687 return 0;
4623e8e6
LP
1688}
1689
89e62e0b
LP
1690static int verity_can_reuse(
1691 const VeritySettings *verity,
1692 const char *name,
1693 struct crypt_device **ret_cd) {
1694
ac1f3ad0
LB
1695 /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
1696 _cleanup_free_ char *root_hash_existing = NULL;
0d12936d 1697 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
ac1f3ad0 1698 struct crypt_params_verity crypt_params = {};
89e62e0b 1699 size_t root_hash_existing_size;
ac1f3ad0
LB
1700 int r;
1701
89e62e0b
LP
1702 assert(verity);
1703 assert(name);
ac1f3ad0
LB
1704 assert(ret_cd);
1705
0d12936d 1706 r = sym_crypt_init_by_name(&cd, name);
ac1f3ad0
LB
1707 if (r < 0)
1708 return log_debug_errno(r, "Error opening verity device, crypt_init_by_name failed: %m");
0d12936d 1709
c719805e
LP
1710 cryptsetup_enable_logging(cd);
1711
0d12936d 1712 r = sym_crypt_get_verity_info(cd, &crypt_params);
ac1f3ad0
LB
1713 if (r < 0)
1714 return log_debug_errno(r, "Error opening verity device, crypt_get_verity_info failed: %m");
0d12936d 1715
89e62e0b
LP
1716 root_hash_existing_size = verity->root_hash_size;
1717 root_hash_existing = malloc0(root_hash_existing_size);
ac1f3ad0
LB
1718 if (!root_hash_existing)
1719 return -ENOMEM;
0d12936d
LP
1720
1721 r = sym_crypt_volume_key_get(cd, CRYPT_ANY_SLOT, root_hash_existing, &root_hash_existing_size, NULL, 0);
ac1f3ad0
LB
1722 if (r < 0)
1723 return log_debug_errno(r, "Error opening verity device, crypt_volume_key_get failed: %m");
89e62e0b
LP
1724 if (verity->root_hash_size != root_hash_existing_size ||
1725 memcmp(root_hash_existing, verity->root_hash, verity->root_hash_size) != 0)
ac1f3ad0 1726 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but root hashes are different.");
89e62e0b 1727
ac1f3ad0 1728#if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
89e62e0b
LP
1729 /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the
1730 * same settings, so that a previous unsigned mount will not be reused if the user asks to use
28423d9a 1731 * signing for the new one, and vice versa. */
89e62e0b 1732 if (!!verity->root_hash_sig != !!(crypt_params.flags & CRYPT_VERITY_ROOT_HASH_SIGNATURE))
ac1f3ad0
LB
1733 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but signature settings are not the same.");
1734#endif
1735
1736 *ret_cd = TAKE_PTR(cd);
1737 return 0;
1738}
1739
75db809a 1740static inline char* dm_deferred_remove_clean(char *name) {
ac1f3ad0 1741 if (!name)
75db809a 1742 return NULL;
0d12936d
LP
1743
1744 (void) sym_crypt_deactivate_by_name(NULL, name, CRYPT_DEACTIVATE_DEFERRED);
75db809a 1745 return mfree(name);
ac1f3ad0
LB
1746}
1747DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean);
1748
c2fa92e7
LP
1749static int validate_signature_userspace(const VeritySettings *verity) {
1750#if HAVE_OPENSSL
1751 _cleanup_(sk_X509_free_allp) STACK_OF(X509) *sk = NULL;
1752 _cleanup_strv_free_ char **certs = NULL;
1753 _cleanup_(PKCS7_freep) PKCS7 *p7 = NULL;
1754 _cleanup_free_ char *s = NULL;
1755 _cleanup_(BIO_freep) BIO *bio = NULL; /* 'bio' must be freed first, 's' second, hence keep this order
1756 * of declaration in place, please */
1757 const unsigned char *d;
c2fa92e7
LP
1758 int r;
1759
1760 assert(verity);
1761 assert(verity->root_hash);
1762 assert(verity->root_hash_sig);
1763
1764 /* Because installing a signature certificate into the kernel chain is so messy, let's optionally do
1765 * userspace validation. */
1766
1767 r = conf_files_list_nulstr(&certs, ".crt", NULL, CONF_FILES_REGULAR|CONF_FILES_FILTER_MASKED, CONF_PATHS_NULSTR("verity.d"));
1768 if (r < 0)
1769 return log_debug_errno(r, "Failed to enumerate certificates: %m");
1770 if (strv_isempty(certs)) {
1771 log_debug("No userspace dm-verity certificates found.");
1772 return 0;
1773 }
1774
1775 d = verity->root_hash_sig;
1776 p7 = d2i_PKCS7(NULL, &d, (long) verity->root_hash_sig_size);
1777 if (!p7)
1778 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to parse PKCS7 DER signature data.");
1779
1780 s = hexmem(verity->root_hash, verity->root_hash_size);
1781 if (!s)
1782 return log_oom_debug();
1783
1784 bio = BIO_new_mem_buf(s, strlen(s));
1785 if (!bio)
1786 return log_oom_debug();
1787
1788 sk = sk_X509_new_null();
1789 if (!sk)
1790 return log_oom_debug();
1791
1792 STRV_FOREACH(i, certs) {
1793 _cleanup_(X509_freep) X509 *c = NULL;
1794 _cleanup_fclose_ FILE *f = NULL;
1795
1796 f = fopen(*i, "re");
1797 if (!f) {
1798 log_debug_errno(errno, "Failed to open '%s', ignoring: %m", *i);
1799 continue;
1800 }
1801
1802 c = PEM_read_X509(f, NULL, NULL, NULL);
1803 if (!c) {
1804 log_debug("Failed to load X509 certificate '%s', ignoring.", *i);
1805 continue;
1806 }
1807
1808 if (sk_X509_push(sk, c) == 0)
1809 return log_oom_debug();
1810
1811 TAKE_PTR(c);
1812 }
1813
1814 r = PKCS7_verify(p7, sk, NULL, bio, NULL, PKCS7_NOINTERN|PKCS7_NOVERIFY);
1815 if (r)
1816 log_debug("Userspace PKCS#7 validation succeeded.");
1817 else
1818 log_debug("Userspace PKCS#7 validation failed: %s", ERR_error_string(ERR_get_error(), NULL));
1819
1820 return r;
1821#else
1822 log_debug("Not doing client-side validation of dm-verity root hash signatures, OpenSSL support disabled.");
1823 return 0;
1824#endif
1825}
1826
1827static int do_crypt_activate_verity(
1828 struct crypt_device *cd,
1829 const char *name,
1830 const VeritySettings *verity) {
1831
1832 bool check_signature;
1833 int r;
1834
1835 assert(cd);
1836 assert(name);
1837 assert(verity);
1838
1839 if (verity->root_hash_sig) {
1840 r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIGNATURE");
1841 if (r < 0 && r != -ENXIO)
1842 log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIGNATURE");
1843
1844 check_signature = r != 0;
1845 } else
1846 check_signature = false;
1847
1848 if (check_signature) {
1849
1850#if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1851 /* First, if we have support for signed keys in the kernel, then try that first. */
1852 r = sym_crypt_activate_by_signed_key(
1853 cd,
1854 name,
1855 verity->root_hash,
1856 verity->root_hash_size,
1857 verity->root_hash_sig,
1858 verity->root_hash_sig_size,
1859 CRYPT_ACTIVATE_READONLY);
1860 if (r >= 0)
1861 return r;
1862
1863 log_debug("Validation of dm-verity signature failed via the kernel, trying userspace validation instead.");
1864#else
1865 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.",
1866 program_invocation_short_name);
1867#endif
1868
1869 /* So this didn't work via the kernel, then let's try userspace validation instead. If that
1870 * works we'll try to activate without telling the kernel the signature. */
1871
1872 r = validate_signature_userspace(verity);
1873 if (r < 0)
1874 return r;
1875 if (r == 0)
1876 return log_debug_errno(SYNTHETIC_ERRNO(ENOKEY),
1877 "Activation of signed Verity volume worked neither via the kernel nor in userspace, can't activate.");
1878 }
1879
1880 return sym_crypt_activate_by_volume_key(
1881 cd,
1882 name,
1883 verity->root_hash,
1884 verity->root_hash_size,
1885 CRYPT_ACTIVATE_READONLY);
1886}
1887
4623e8e6 1888static int verity_partition(
aee36b4e 1889 PartitionDesignator designator,
4623e8e6
LP
1890 DissectedPartition *m,
1891 DissectedPartition *v,
89e62e0b 1892 const VeritySettings *verity,
4623e8e6
LP
1893 DissectImageFlags flags,
1894 DecryptedImage *d) {
1895
0d12936d 1896 _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL;
ac1f3ad0 1897 _cleanup_(dm_deferred_remove_cleanp) char *restore_deferred_remove = NULL;
89e62e0b 1898 _cleanup_free_ char *node = NULL, *name = NULL;
4623e8e6
LP
1899 int r;
1900
1901 assert(m);
89e62e0b 1902 assert(v || (verity && verity->data_path));
4623e8e6 1903
89e62e0b 1904 if (!verity || !verity->root_hash)
4623e8e6 1905 return 0;
aee36b4e
LP
1906 if (!((verity->designator < 0 && designator == PARTITION_ROOT) ||
1907 (verity->designator == designator)))
1908 return 0;
4623e8e6
LP
1909
1910 if (!m->found || !m->node || !m->fstype)
1911 return 0;
89e62e0b 1912 if (!verity->data_path) {
e7cbe5cb
LB
1913 if (!v->found || !v->node || !v->fstype)
1914 return 0;
4623e8e6 1915
e7cbe5cb
LB
1916 if (!streq(v->fstype, "DM_verity_hash"))
1917 return 0;
1918 }
4623e8e6 1919
0d12936d
LP
1920 r = dlopen_cryptsetup();
1921 if (r < 0)
1922 return r;
1923
ac1f3ad0
LB
1924 if (FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) {
1925 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
1926 _cleanup_free_ char *root_hash_encoded = NULL;
0d12936d 1927
89e62e0b 1928 root_hash_encoded = hexmem(verity->root_hash, verity->root_hash_size);
ac1f3ad0
LB
1929 if (!root_hash_encoded)
1930 return -ENOMEM;
aee36b4e 1931
ac1f3ad0
LB
1932 r = make_dm_name_and_node(root_hash_encoded, "-verity", &name, &node);
1933 } else
1934 r = make_dm_name_and_node(m->node, "-verity", &name, &node);
4623e8e6
LP
1935 if (r < 0)
1936 return r;
1937
89e62e0b 1938 r = sym_crypt_init(&cd, verity->data_path ?: v->node);
4623e8e6
LP
1939 if (r < 0)
1940 return r;
1941
efc3b12f 1942 cryptsetup_enable_logging(cd);
1887032f 1943
0d12936d 1944 r = sym_crypt_load(cd, CRYPT_VERITY, NULL);
4623e8e6 1945 if (r < 0)
294bd454 1946 return r;
4623e8e6 1947
0d12936d 1948 r = sym_crypt_set_data_device(cd, m->node);
4623e8e6 1949 if (r < 0)
294bd454 1950 return r;
4623e8e6 1951
319a4f4b 1952 if (!GREEDY_REALLOC0(d->decrypted, d->n_decrypted + 1))
ac1f3ad0
LB
1953 return -ENOMEM;
1954
1955 /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
1956 * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
1957 * retry a few times before giving up. */
1958 for (unsigned i = 0; i < N_DEVICE_NODE_LIST_ATTEMPTS; i++) {
c2fa92e7
LP
1959
1960 r = do_crypt_activate_verity(cd, name, verity);
ac1f3ad0
LB
1961 /* libdevmapper can return EINVAL when the device is already in the activation stage.
1962 * There's no way to distinguish this situation from a genuine error due to invalid
2aed63f4 1963 * parameters, so immediately fall back to activating the device with a unique name.
89e62e0b
LP
1964 * Improvements in libcrypsetup can ensure this never happens:
1965 * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
ac1f3ad0 1966 if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
aee36b4e 1967 return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
9ecb5c10 1968 if (!IN_SET(r,
22043172 1969 0, /* Success */
9ecb5c10 1970 -EEXIST, /* Volume is already open and ready to be used */
22043172
LP
1971 -EBUSY, /* Volume is being opened but not ready, crypt_init_by_name can fetch details */
1972 -ENODEV /* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */))
ac1f3ad0 1973 return r;
9ecb5c10 1974 if (IN_SET(r, -EEXIST, -EBUSY)) {
ac1f3ad0 1975 struct crypt_device *existing_cd = NULL;
c2923fdc 1976
ac1f3ad0
LB
1977 if (!restore_deferred_remove){
1978 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
1979 r = dm_deferred_remove_cancel(name);
9ecb5c10
LB
1980 /* If activation returns EBUSY there might be no deferred removal to cancel, that's fine */
1981 if (r < 0 && r != -ENXIO)
ac1f3ad0 1982 return log_debug_errno(r, "Disabling automated deferred removal for verity device %s failed: %m", node);
9ecb5c10
LB
1983 if (r == 0) {
1984 restore_deferred_remove = strdup(name);
1985 if (!restore_deferred_remove)
1986 return -ENOMEM;
1987 }
ac1f3ad0 1988 }
c2923fdc 1989
89e62e0b 1990 r = verity_can_reuse(verity, name, &existing_cd);
ac1f3ad0
LB
1991 /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
1992 if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
aee36b4e 1993 return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
9ecb5c10 1994 if (!IN_SET(r, 0, -ENODEV, -ENOENT, -EBUSY))
ac1f3ad0
LB
1995 return log_debug_errno(r, "Checking whether existing verity device %s can be reused failed: %m", node);
1996 if (r == 0) {
2b660510
YW
1997 usec_t timeout_usec = 100 * USEC_PER_MSEC;
1998 const char *e;
1999
2000 /* On slower machines, like non-KVM vm, setting up device may take a long time.
2001 * Let's make the timeout configurable. */
2002 e = getenv("SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC");
2003 if (e) {
2004 usec_t t;
2005
2006 r = parse_sec(e, &t);
2007 if (r < 0)
2008 log_debug_errno(r,
2009 "Failed to parse timeout specified in $SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC, "
2010 "using the default timeout (%s).",
2011 FORMAT_TIMESPAN(timeout_usec, USEC_PER_MSEC));
2012 else
2013 timeout_usec = t;
2014 }
2015
c419b6f0
LB
2016 /* devmapper might say that the device exists, but the devlink might not yet have been
2017 * created. Check and wait for the udev event in that case. */
2b660510 2018 r = device_wait_for_devlink(node, "block", timeout_usec, NULL);
c419b6f0
LB
2019 /* Fallback to activation with a unique device if it's taking too long */
2020 if (r == -ETIMEDOUT)
2021 break;
2022 if (r < 0)
2023 return r;
2024
ac1f3ad0 2025 if (cd)
0d12936d 2026 sym_crypt_free(cd);
ac1f3ad0
LB
2027 cd = existing_cd;
2028 }
c2923fdc 2029 }
ac1f3ad0
LB
2030 if (r == 0)
2031 break;
ecab4c47
LB
2032
2033 /* Device is being opened by another process, but it has not finished yet, yield for 2ms */
2034 (void) usleep(2 * USEC_PER_MSEC);
ac1f3ad0
LB
2035 }
2036
ac1f3ad0
LB
2037 /* An existing verity device was reported by libcryptsetup/libdevmapper, but we can't use it at this time.
2038 * Fall back to activating it with a unique device name. */
2039 if (r != 0 && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE))
aee36b4e 2040 return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d);
ac1f3ad0
LB
2041
2042 /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
2043 restore_deferred_remove = mfree(restore_deferred_remove);
4623e8e6 2044
94344385
LP
2045 d->decrypted[d->n_decrypted++] = (DecryptedPartition) {
2046 .name = TAKE_PTR(name),
2047 .device = TAKE_PTR(cd),
2048 };
4623e8e6 2049
1cc6c93a 2050 m->decrypted_node = TAKE_PTR(node);
4623e8e6
LP
2051
2052 return 0;
18b5886e
LP
2053}
2054#endif
2055
2056int dissected_image_decrypt(
2057 DissectedImage *m,
2058 const char *passphrase,
89e62e0b 2059 const VeritySettings *verity,
18b5886e
LP
2060 DissectImageFlags flags,
2061 DecryptedImage **ret) {
2062
349cc4a5 2063#if HAVE_LIBCRYPTSETUP
49b5b3b4 2064 _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL;
18b5886e
LP
2065 int r;
2066#endif
2067
2068 assert(m);
89e62e0b 2069 assert(!verity || verity->root_hash || verity->root_hash_size == 0);
18b5886e
LP
2070
2071 /* Returns:
2072 *
2073 * = 0 → There was nothing to decrypt
2074 * > 0 → Decrypted successfully
d1c536f5 2075 * -ENOKEY → There's something to decrypt but no key was supplied
18b5886e
LP
2076 * -EKEYREJECTED → Passed key was not correct
2077 */
2078
89e62e0b 2079 if (verity && verity->root_hash && verity->root_hash_size < sizeof(sd_id128_t))
4623e8e6
LP
2080 return -EINVAL;
2081
c3c88d67 2082 if (!m->encrypted && !m->verity_ready) {
18b5886e
LP
2083 *ret = NULL;
2084 return 0;
2085 }
2086
349cc4a5 2087#if HAVE_LIBCRYPTSETUP
18b5886e
LP
2088 d = new0(DecryptedImage, 1);
2089 if (!d)
2090 return -ENOMEM;
2091
569a0e42 2092 for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) {
18b5886e 2093 DissectedPartition *p = m->partitions + i;
22043172 2094 PartitionDesignator k;
18b5886e
LP
2095
2096 if (!p->found)
2097 continue;
2098
2099 r = decrypt_partition(p, passphrase, flags, d);
2100 if (r < 0)
2101 return r;
2102
4623e8e6
LP
2103 k = PARTITION_VERITY_OF(i);
2104 if (k >= 0) {
aee36b4e 2105 r = verity_partition(i, p, m->partitions + k, verity, flags | DISSECT_IMAGE_VERITY_SHARE, d);
4623e8e6
LP
2106 if (r < 0)
2107 return r;
2108 }
2109
18b5886e
LP
2110 if (!p->decrypted_fstype && p->decrypted_node) {
2111 r = probe_filesystem(p->decrypted_node, &p->decrypted_fstype);
7cc84b2c 2112 if (r < 0 && r != -EUCLEAN)
18b5886e
LP
2113 return r;
2114 }
2115 }
2116
1cc6c93a 2117 *ret = TAKE_PTR(d);
18b5886e
LP
2118
2119 return 1;
2120#else
2121 return -EOPNOTSUPP;
2122#endif
2123}
2124
2125int dissected_image_decrypt_interactively(
2126 DissectedImage *m,
2127 const char *passphrase,
89e62e0b 2128 const VeritySettings *verity,
18b5886e
LP
2129 DissectImageFlags flags,
2130 DecryptedImage **ret) {
2131
2132 _cleanup_strv_free_erase_ char **z = NULL;
2133 int n = 3, r;
2134
2135 if (passphrase)
2136 n--;
2137
2138 for (;;) {
89e62e0b 2139 r = dissected_image_decrypt(m, passphrase, verity, flags, ret);
18b5886e
LP
2140 if (r >= 0)
2141 return r;
2142 if (r == -EKEYREJECTED)
2143 log_error_errno(r, "Incorrect passphrase, try again!");
fc95c359
YW
2144 else if (r != -ENOKEY)
2145 return log_error_errno(r, "Failed to decrypt image: %m");
18b5886e 2146
baaa35ad
ZJS
2147 if (--n < 0)
2148 return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED),
2149 "Too many retries.");
18b5886e
LP
2150
2151 z = strv_free(z);
2152
8806bb4b 2153 r = ask_password_auto("Please enter image passphrase:", NULL, "dissect", "dissect", "dissect.passphrase", USEC_INFINITY, 0, &z);
18b5886e
LP
2154 if (r < 0)
2155 return log_error_errno(r, "Failed to query for passphrase: %m");
2156
2157 passphrase = z[0];
2158 }
2159}
2160
18b5886e 2161int decrypted_image_relinquish(DecryptedImage *d) {
18b5886e
LP
2162 assert(d);
2163
67f63ee5
ZJS
2164 /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a
2165 * boolean so that we don't clean it up ourselves either anymore */
18b5886e 2166
349cc4a5 2167#if HAVE_LIBCRYPTSETUP
67f63ee5
ZJS
2168 int r;
2169
2170 for (size_t i = 0; i < d->n_decrypted; i++) {
18b5886e
LP
2171 DecryptedPartition *p = d->decrypted + i;
2172
2173 if (p->relinquished)
2174 continue;
2175
0d12936d 2176 r = sym_crypt_deactivate_by_name(NULL, p->name, CRYPT_DEACTIVATE_DEFERRED);
18b5886e
LP
2177 if (r < 0)
2178 return log_debug_errno(r, "Failed to mark %s for auto-removal: %m", p->name);
2179
2180 p->relinquished = true;
2181 }
2182#endif
2183
2184 return 0;
2185}
2186
89e62e0b
LP
2187static char *build_auxiliary_path(const char *image, const char *suffix) {
2188 const char *e;
2189 char *n;
2190
2191 assert(image);
2192 assert(suffix);
2193
2194 e = endswith(image, ".raw");
2195 if (!e)
2196 return strjoin(e, suffix);
2197
2198 n = new(char, e - image + strlen(suffix) + 1);
2199 if (!n)
2200 return NULL;
2201
2202 strcpy(mempcpy(n, image, e - image), suffix);
2203 return n;
2204}
2205
2206void verity_settings_done(VeritySettings *v) {
2207 assert(v);
2208
2209 v->root_hash = mfree(v->root_hash);
2210 v->root_hash_size = 0;
2211
2212 v->root_hash_sig = mfree(v->root_hash_sig);
2213 v->root_hash_sig_size = 0;
2214
2215 v->data_path = mfree(v->data_path);
2216}
2217
2218int verity_settings_load(
2219 VeritySettings *verity,
f5ea63a5
LP
2220 const char *image,
2221 const char *root_hash_path,
89e62e0b
LP
2222 const char *root_hash_sig_path) {
2223
2224 _cleanup_free_ void *root_hash = NULL, *root_hash_sig = NULL;
2225 size_t root_hash_size = 0, root_hash_sig_size = 0;
2226 _cleanup_free_ char *verity_data_path = NULL;
aee36b4e 2227 PartitionDesignator designator;
78ebe980
LP
2228 int r;
2229
89e62e0b 2230 assert(verity);
78ebe980 2231 assert(image);
aee36b4e 2232 assert(verity->designator < 0 || IN_SET(verity->designator, PARTITION_ROOT, PARTITION_USR));
78ebe980 2233
89e62e0b
LP
2234 /* If we are asked to load the root hash for a device node, exit early */
2235 if (is_device_path(image))
78ebe980 2236 return 0;
78ebe980 2237
d5fcc5b0
LP
2238 r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIDECAR");
2239 if (r < 0 && r != -ENXIO)
2240 log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIDECAR, ignoring: %m");
2241 if (r == 0)
2242 return 0;
2243
aee36b4e
LP
2244 designator = verity->designator;
2245
89e62e0b 2246 /* We only fill in what isn't already filled in */
c2923fdc 2247
89e62e0b 2248 if (!verity->root_hash) {
e7cbe5cb 2249 _cleanup_free_ char *text = NULL;
e7cbe5cb 2250
0389f4fa 2251 if (root_hash_path) {
aee36b4e 2252 /* If explicitly specified it takes precedence */
0389f4fa
LB
2253 r = read_one_line_file(root_hash_path, &text);
2254 if (r < 0)
e7cbe5cb 2255 return r;
aee36b4e
LP
2256
2257 if (designator < 0)
2258 designator = PARTITION_ROOT;
0389f4fa 2259 } else {
aee36b4e
LP
2260 /* Otherwise look for xattr and separate file, and first for the data for root and if
2261 * that doesn't exist for /usr */
0389f4fa 2262
aee36b4e 2263 if (designator < 0 || designator == PARTITION_ROOT) {
c53e07e2 2264 r = getxattr_malloc(image, "user.verity.roothash", &text);
aee36b4e
LP
2265 if (r < 0) {
2266 _cleanup_free_ char *p = NULL;
78ebe980 2267
aee36b4e
LP
2268 if (!IN_SET(r, -ENODATA, -ENOENT) && !ERRNO_IS_NOT_SUPPORTED(r))
2269 return r;
e7cbe5cb 2270
aee36b4e
LP
2271 p = build_auxiliary_path(image, ".roothash");
2272 if (!p)
2273 return -ENOMEM;
2274
2275 r = read_one_line_file(p, &text);
2276 if (r < 0 && r != -ENOENT)
2277 return r;
2278 }
2279
2280 if (text)
2281 designator = PARTITION_ROOT;
2282 }
2283
2284 if (!text && (designator < 0 || designator == PARTITION_USR)) {
2285 /* So in the "roothash" xattr/file name above the "root" of course primarily
2286 * refers to the root of the Verity Merkle tree. But coincidentally it also
2287 * is the hash for the *root* file system, i.e. the "root" neatly refers to
2288 * two distinct concepts called "root". Taking benefit of this happy
2289 * coincidence we call the file with the root hash for the /usr/ file system
2290 * `usrhash`, because `usrroothash` or `rootusrhash` would just be too
2291 * confusing. We thus drop the reference to the root of the Merkle tree, and
2292 * just indicate which file system it's about. */
c53e07e2 2293 r = getxattr_malloc(image, "user.verity.usrhash", &text);
aee36b4e
LP
2294 if (r < 0) {
2295 _cleanup_free_ char *p = NULL;
2296
2297 if (!IN_SET(r, -ENODATA, -ENOENT) && !ERRNO_IS_NOT_SUPPORTED(r))
2298 return r;
2299
2300 p = build_auxiliary_path(image, ".usrhash");
2301 if (!p)
2302 return -ENOMEM;
2303
2304 r = read_one_line_file(p, &text);
2305 if (r < 0 && r != -ENOENT)
2306 return r;
2307 }
2308
2309 if (text)
2310 designator = PARTITION_USR;
0389f4fa 2311 }
e7cbe5cb
LB
2312 }
2313
2314 if (text) {
89e62e0b 2315 r = unhexmem(text, strlen(text), &root_hash, &root_hash_size);
e7cbe5cb
LB
2316 if (r < 0)
2317 return r;
89e62e0b 2318 if (root_hash_size < sizeof(sd_id128_t))
e7cbe5cb
LB
2319 return -EINVAL;
2320 }
2321 }
2322
90f98986 2323 if ((root_hash || verity->root_hash) && !verity->root_hash_sig) {
aee36b4e 2324 if (root_hash_sig_path) {
ae9cf30b 2325 r = read_full_file(root_hash_sig_path, (char**) &root_hash_sig, &root_hash_sig_size);
aee36b4e
LP
2326 if (r < 0 && r != -ENOENT)
2327 return r;
2328
2329 if (designator < 0)
2330 designator = PARTITION_ROOT;
2331 } else {
2332 if (designator < 0 || designator == PARTITION_ROOT) {
2333 _cleanup_free_ char *p = NULL;
2334
2335 /* Follow naming convention recommended by the relevant RFC:
2336 * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
2337 p = build_auxiliary_path(image, ".roothash.p7s");
2338 if (!p)
2339 return -ENOMEM;
89e62e0b 2340
ae9cf30b 2341 r = read_full_file(p, (char**) &root_hash_sig, &root_hash_sig_size);
aee36b4e
LP
2342 if (r < 0 && r != -ENOENT)
2343 return r;
2344 if (r >= 0)
2345 designator = PARTITION_ROOT;
2346 }
2347
2348 if (!root_hash_sig && (designator < 0 || designator == PARTITION_USR)) {
2349 _cleanup_free_ char *p = NULL;
2350
2351 p = build_auxiliary_path(image, ".usrhash.p7s");
2352 if (!p)
2353 return -ENOMEM;
89e62e0b 2354
ae9cf30b 2355 r = read_full_file(p, (char**) &root_hash_sig, &root_hash_sig_size);
aee36b4e
LP
2356 if (r < 0 && r != -ENOENT)
2357 return r;
2358 if (r >= 0)
2359 designator = PARTITION_USR;
2360 }
89e62e0b
LP
2361 }
2362
aee36b4e 2363 if (root_hash_sig && root_hash_sig_size == 0) /* refuse empty size signatures */
89e62e0b
LP
2364 return -EINVAL;
2365 }
2366
2367 if (!verity->data_path) {
2368 _cleanup_free_ char *p = NULL;
2369
2370 p = build_auxiliary_path(image, ".verity");
2371 if (!p)
2372 return -ENOMEM;
2373
2374 if (access(p, F_OK) < 0) {
2375 if (errno != ENOENT)
2376 return -errno;
2377 } else
2378 verity_data_path = TAKE_PTR(p);
2379 }
2380
2381 if (root_hash) {
2382 verity->root_hash = TAKE_PTR(root_hash);
2383 verity->root_hash_size = root_hash_size;
2384 }
2385
2386 if (root_hash_sig) {
2387 verity->root_hash_sig = TAKE_PTR(root_hash_sig);
2388 verity->root_hash_sig_size = root_hash_sig_size;
e7cbe5cb 2389 }
89e62e0b
LP
2390
2391 if (verity_data_path)
2392 verity->data_path = TAKE_PTR(verity_data_path);
78ebe980 2393
aee36b4e
LP
2394 if (verity->designator < 0)
2395 verity->designator = designator;
2396
78ebe980
LP
2397 return 1;
2398}
2399
88b3300f
LP
2400int dissected_image_load_verity_sig_partition(
2401 DissectedImage *m,
2402 int fd,
2403 VeritySettings *verity) {
2404
2405 _cleanup_free_ void *root_hash = NULL, *root_hash_sig = NULL;
2406 _cleanup_(json_variant_unrefp) JsonVariant *v = NULL;
2407 size_t root_hash_size, root_hash_sig_size;
2408 _cleanup_free_ char *buf = NULL;
2409 PartitionDesignator d;
2410 DissectedPartition *p;
2411 JsonVariant *rh, *sig;
2412 ssize_t n;
2413 char *e;
2414 int r;
2415
2416 assert(m);
2417 assert(fd >= 0);
2418 assert(verity);
2419
2420 if (verity->root_hash && verity->root_hash_sig) /* Already loaded? */
2421 return 0;
2422
2423 r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_EMBEDDED");
2424 if (r < 0 && r != -ENXIO)
2425 log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_EMBEDDED, ignoring: %m");
2426 if (r == 0)
2427 return 0;
2428
2429 d = PARTITION_VERITY_SIG_OF(verity->designator < 0 ? PARTITION_ROOT : verity->designator);
2430 assert(d >= 0);
2431
2432 p = m->partitions + d;
2433 if (!p->found)
2434 return 0;
2435 if (p->offset == UINT64_MAX || p->size == UINT64_MAX)
2436 return -EINVAL;
2437
2438 if (p->size > 4*1024*1024) /* Signature data cannot possible be larger than 4M, refuse that */
2439 return -EFBIG;
2440
2441 buf = new(char, p->size+1);
2442 if (!buf)
2443 return -ENOMEM;
2444
2445 n = pread(fd, buf, p->size, p->offset);
2446 if (n < 0)
2447 return -ENOMEM;
2448 if ((uint64_t) n != p->size)
2449 return -EIO;
2450
2451 e = memchr(buf, 0, p->size);
2452 if (e) {
2453 /* If we found a NUL byte then the rest of the data must be NUL too */
2454 if (!memeqzero(e, p->size - (e - buf)))
2455 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature data contains embedded NUL byte.");
2456 } else
2457 buf[p->size] = 0;
2458
2459 r = json_parse(buf, 0, &v, NULL, NULL);
2460 if (r < 0)
2461 return log_debug_errno(r, "Failed to parse signature JSON data: %m");
2462
2463 rh = json_variant_by_key(v, "rootHash");
2464 if (!rh)
2465 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature JSON object lacks 'rootHash' field.");
2466 if (!json_variant_is_string(rh))
2467 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "'rootHash' field of signature JSON object is not a string.");
2468
2469 r = unhexmem(json_variant_string(rh), SIZE_MAX, &root_hash, &root_hash_size);
2470 if (r < 0)
2471 return log_debug_errno(r, "Failed to parse root hash field: %m");
2472
2473 /* Check if specified root hash matches if it is specified */
2474 if (verity->root_hash &&
2475 memcmp_nn(verity->root_hash, verity->root_hash_size, root_hash, root_hash_size) != 0) {
2476 _cleanup_free_ char *a = NULL, *b = NULL;
2477
2478 a = hexmem(root_hash, root_hash_size);
2479 b = hexmem(verity->root_hash, verity->root_hash_size);
2480
2481 return log_debug_errno(r, "Root hash in signature JSON data (%s) doesn't match configured hash (%s).", strna(a), strna(b));
2482 }
2483
2484 sig = json_variant_by_key(v, "signature");
2485 if (!sig)
2486 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature JSON object lacks 'signature' field.");
2487 if (!json_variant_is_string(sig))
2488 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "'signature' field of signature JSON object is not a string.");
2489
2490 r = unbase64mem(json_variant_string(sig), SIZE_MAX, &root_hash_sig, &root_hash_sig_size);
2491 if (r < 0)
2492 return log_debug_errno(r, "Failed to parse signature field: %m");
2493
2494 free_and_replace(verity->root_hash, root_hash);
2495 verity->root_hash_size = root_hash_size;
2496
2497 free_and_replace(verity->root_hash_sig, root_hash_sig);
2498 verity->root_hash_sig_size = root_hash_sig_size;
2499
2500 return 1;
2501}
2502
22847508 2503int dissected_image_acquire_metadata(DissectedImage *m, DissectImageFlags extra_flags) {
3b925504
LP
2504
2505 enum {
2506 META_HOSTNAME,
2507 META_MACHINE_ID,
2508 META_MACHINE_INFO,
2509 META_OS_RELEASE,
7718ac97 2510 META_EXTENSION_RELEASE,
a4e0d617 2511 META_HAS_INIT_SYSTEM,
3b925504
LP
2512 _META_MAX,
2513 };
2514
9a4b883b 2515 static const char *const paths[_META_MAX] = {
7718ac97
LB
2516 [META_HOSTNAME] = "/etc/hostname\0",
2517 [META_MACHINE_ID] = "/etc/machine-id\0",
2518 [META_MACHINE_INFO] = "/etc/machine-info\0",
9a4b883b 2519 [META_OS_RELEASE] = ("/etc/os-release\0"
22847508 2520 "/usr/lib/os-release\0"),
a4e0d617
LP
2521 [META_EXTENSION_RELEASE] = "extension-release\0", /* Used only for logging. */
2522 [META_HAS_INIT_SYSTEM] = "has-init-system\0", /* ditto */
3b925504
LP
2523 };
2524
7718ac97 2525 _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL, **extension_release = NULL;
af8219d5 2526 _cleanup_close_pair_ int error_pipe[2] = { -1, -1 };
3b925504
LP
2527 _cleanup_(rmdir_and_freep) char *t = NULL;
2528 _cleanup_(sigkill_waitp) pid_t child = 0;
2529 sd_id128_t machine_id = SD_ID128_NULL;
2530 _cleanup_free_ char *hostname = NULL;
67f63ee5 2531 unsigned n_meta_initialized = 0;
af8219d5 2532 int fds[2 * _META_MAX], r, v;
a4e0d617 2533 int has_init_system = -1;
af8219d5 2534 ssize_t n;
3b925504
LP
2535
2536 BLOCK_SIGNALS(SIGCHLD);
2537
2538 assert(m);
2539
7718ac97 2540 for (; n_meta_initialized < _META_MAX; n_meta_initialized ++) {
d9119c00
LP
2541 if (!paths[n_meta_initialized]) {
2542 fds[2*n_meta_initialized] = fds[2*n_meta_initialized+1] = -1;
7718ac97 2543 continue;
d9119c00
LP
2544 }
2545
3b925504
LP
2546 if (pipe2(fds + 2*n_meta_initialized, O_CLOEXEC) < 0) {
2547 r = -errno;
2548 goto finish;
2549 }
7718ac97 2550 }
3b925504
LP
2551
2552 r = mkdtemp_malloc("/tmp/dissect-XXXXXX", &t);
2553 if (r < 0)
2554 goto finish;
2555
af8219d5
LP
2556 if (pipe2(error_pipe, O_CLOEXEC) < 0) {
2557 r = -errno;
2558 goto finish;
2559 }
2560
e2047ba9 2561 r = safe_fork("(sd-dissect)", FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE, &child);
be39f6ee 2562 if (r < 0)
3b925504 2563 goto finish;
be39f6ee 2564 if (r == 0) {
a4e0d617 2565 /* Child in a new mount namespace */
af8219d5
LP
2566 error_pipe[0] = safe_close(error_pipe[0]);
2567
7cf66030
LP
2568 r = dissected_image_mount(
2569 m,
2570 t,
2571 UID_INVALID,
21b61b1d 2572 UID_INVALID,
22847508
ZJS
2573 extra_flags |
2574 DISSECT_IMAGE_READ_ONLY |
2575 DISSECT_IMAGE_MOUNT_ROOT_ONLY |
7cf66030 2576 DISSECT_IMAGE_USR_NO_ROOT);
429d4e41
LP
2577 if (r < 0) {
2578 log_debug_errno(r, "Failed to mount dissected image: %m");
03ae68f4 2579 goto inner_fail;
429d4e41 2580 }
3b925504 2581
67f63ee5 2582 for (unsigned k = 0; k < _META_MAX; k++) {
37e44c3f 2583 _cleanup_close_ int fd = -ENOENT;
3b925504
LP
2584 const char *p;
2585
7718ac97
LB
2586 if (!paths[k])
2587 continue;
2588
3b925504
LP
2589 fds[2*k] = safe_close(fds[2*k]);
2590
a4e0d617
LP
2591 switch (k) {
2592
2593 case META_EXTENSION_RELEASE:
9a4b883b
LB
2594 /* As per the os-release spec, if the image is an extension it will have a file
2595 * named after the image name in extension-release.d/ - we use the image name
2596 * and try to resolve it with the extension-release helpers, as sometimes
2597 * the image names are mangled on deployment and do not match anymore.
2598 * Unlike other paths this is not fixed, and the image name
2599 * can be mangled on deployment, so by calling into the helper
2600 * we allow a fallback that matches on the first extension-release
2601 * file found in the directory, if one named after the image cannot
2602 * be found first. */
2603 r = open_extension_release(t, m->image_name, NULL, &fd);
2604 if (r < 0)
2605 fd = r; /* Propagate the error. */
a4e0d617
LP
2606 break;
2607
2608 case META_HAS_INIT_SYSTEM: {
2609 bool found = false;
a4e0d617
LP
2610
2611 FOREACH_STRING(init,
2612 "/usr/lib/systemd/systemd", /* systemd on /usr merged system */
2613 "/lib/systemd/systemd", /* systemd on /usr non-merged systems */
2614 "/sbin/init") { /* traditional path the Linux kernel invokes */
2615
2616 r = chase_symlinks(init, t, CHASE_PREFIX_ROOT, NULL, NULL);
2617 if (r < 0) {
2618 if (r != -ENOENT)
2619 log_debug_errno(r, "Failed to resolve %s, ignoring: %m", init);
2620 } else {
2621 found = true;
2622 break;
2623 }
2624 }
2625
2626 r = loop_write(fds[2*k+1], &found, sizeof(found), false);
2627 if (r < 0)
2628 goto inner_fail;
2629
2630 continue;
2631 }
2632
2633 default:
9a4b883b
LB
2634 NULSTR_FOREACH(p, paths[k]) {
2635 fd = chase_symlinks_and_open(p, t, CHASE_PREFIX_ROOT, O_RDONLY|O_CLOEXEC|O_NOCTTY, NULL);
2636 if (fd >= 0)
2637 break;
2638 }
a4e0d617
LP
2639 }
2640
36952d19
LP
2641 if (fd < 0) {
2642 log_debug_errno(fd, "Failed to read %s file of image, ignoring: %m", paths[k]);
37e44c3f 2643 fds[2*k+1] = safe_close(fds[2*k+1]);
3b925504 2644 continue;
36952d19 2645 }
3b925504 2646
f5fbe71d 2647 r = copy_bytes(fd, fds[2*k+1], UINT64_MAX, 0);
a4e0d617
LP
2648 if (r < 0)
2649 goto inner_fail;
3b925504
LP
2650
2651 fds[2*k+1] = safe_close(fds[2*k+1]);
2652 }
2653
2654 _exit(EXIT_SUCCESS);
a4e0d617
LP
2655
2656 inner_fail:
03ae68f4 2657 /* Let parent know the error */
a4e0d617
LP
2658 (void) write(error_pipe[1], &r, sizeof(r));
2659 _exit(EXIT_FAILURE);
3b925504
LP
2660 }
2661
af8219d5
LP
2662 error_pipe[1] = safe_close(error_pipe[1]);
2663
67f63ee5 2664 for (unsigned k = 0; k < _META_MAX; k++) {
3b925504
LP
2665 _cleanup_fclose_ FILE *f = NULL;
2666
7718ac97
LB
2667 if (!paths[k])
2668 continue;
2669
3b925504
LP
2670 fds[2*k+1] = safe_close(fds[2*k+1]);
2671
4fa744a3 2672 f = take_fdopen(&fds[2*k], "r");
3b925504
LP
2673 if (!f) {
2674 r = -errno;
2675 goto finish;
2676 }
2677
3b925504
LP
2678 switch (k) {
2679
2680 case META_HOSTNAME:
2681 r = read_etc_hostname_stream(f, &hostname);
2682 if (r < 0)
f6048e5e 2683 log_debug_errno(r, "Failed to read /etc/hostname of image: %m");
3b925504
LP
2684
2685 break;
2686
2687 case META_MACHINE_ID: {
2688 _cleanup_free_ char *line = NULL;
2689
2690 r = read_line(f, LONG_LINE_MAX, &line);
2691 if (r < 0)
f6048e5e 2692 log_debug_errno(r, "Failed to read /etc/machine-id of image: %m");
3b925504
LP
2693 else if (r == 33) {
2694 r = sd_id128_from_string(line, &machine_id);
2695 if (r < 0)
2696 log_debug_errno(r, "Image contains invalid /etc/machine-id: %s", line);
2697 } else if (r == 0)
f6048e5e 2698 log_debug("/etc/machine-id file of image is empty.");
ab763cb2 2699 else if (streq(line, "uninitialized"))
f6048e5e 2700 log_debug("/etc/machine-id file of image is uninitialized (likely aborted first boot).");
3b925504 2701 else
f6048e5e 2702 log_debug("/etc/machine-id file of image has unexpected length %i.", r);
3b925504
LP
2703
2704 break;
2705 }
2706
2707 case META_MACHINE_INFO:
aa8fbc74 2708 r = load_env_file_pairs(f, "machine-info", &machine_info);
3b925504 2709 if (r < 0)
f6048e5e 2710 log_debug_errno(r, "Failed to read /etc/machine-info of image: %m");
3b925504
LP
2711
2712 break;
2713
2714 case META_OS_RELEASE:
aa8fbc74 2715 r = load_env_file_pairs(f, "os-release", &os_release);
3b925504 2716 if (r < 0)
f6048e5e 2717 log_debug_errno(r, "Failed to read OS release file of image: %m");
3b925504
LP
2718
2719 break;
7718ac97
LB
2720
2721 case META_EXTENSION_RELEASE:
2722 r = load_env_file_pairs(f, "extension-release", &extension_release);
2723 if (r < 0)
f6048e5e 2724 log_debug_errno(r, "Failed to read extension release file of image: %m");
7718ac97
LB
2725
2726 break;
a4e0d617
LP
2727
2728 case META_HAS_INIT_SYSTEM: {
2729 bool b = false;
2730 size_t nr;
2731
2732 errno = 0;
2733 nr = fread(&b, 1, sizeof(b), f);
2734 if (nr != sizeof(b))
2735 log_debug_errno(errno_or_else(EIO), "Failed to read has-init-system boolean: %m");
2736 else
2737 has_init_system = b;
2738
2739 break;
2740 }}
3b925504
LP
2741 }
2742
2e87a1fd 2743 r = wait_for_terminate_and_check("(sd-dissect)", child, 0);
3b925504 2744 child = 0;
2e87a1fd 2745 if (r < 0)
af8219d5
LP
2746 return r;
2747
2748 n = read(error_pipe[0], &v, sizeof(v));
2749 if (n < 0)
2750 return -errno;
2751 if (n == sizeof(v))
2752 return v; /* propagate error sent to us from child */
2753 if (n != 0)
2754 return -EIO;
2755
2e87a1fd
LP
2756 if (r != EXIT_SUCCESS)
2757 return -EPROTO;
3b925504
LP
2758
2759 free_and_replace(m->hostname, hostname);
2760 m->machine_id = machine_id;
2761 strv_free_and_replace(m->machine_info, machine_info);
2762 strv_free_and_replace(m->os_release, os_release);
7718ac97 2763 strv_free_and_replace(m->extension_release, extension_release);
a4e0d617 2764 m->has_init_system = has_init_system;
3b925504
LP
2765
2766finish:
67f63ee5 2767 for (unsigned k = 0; k < n_meta_initialized; k++)
3b925504
LP
2768 safe_close_pair(fds + 2*k);
2769
2770 return r;
2771}
2772
bad31660 2773int dissect_loop_device_and_warn(
bad31660 2774 const LoopDevice *loop,
89e62e0b 2775 const VeritySettings *verity,
18d73705 2776 const MountOptions *mount_options,
4526113f
LP
2777 DissectImageFlags flags,
2778 DissectedImage **ret) {
2779
64dd3a24 2780 const char *name;
4526113f
LP
2781 int r;
2782
bad31660
YW
2783 assert(loop);
2784 assert(loop->fd >= 0);
4526113f 2785
64dd3a24 2786 name = ASSERT_PTR(loop->backing_file ?: loop->node);
4526113f 2787
369de26f 2788 r = dissect_loop_device(loop, verity, mount_options, flags, ret);
4526113f
LP
2789 switch (r) {
2790
2791 case -EOPNOTSUPP:
2792 return log_error_errno(r, "Dissecting images is not supported, compiled without blkid support.");
2793
2794 case -ENOPKG:
48084df6
ZJS
2795 return log_error_errno(r, "%s: Couldn't identify a suitable partition table or file system.", name);
2796
2797 case -ENOMEDIUM:
2798 return log_error_errno(r, "%s: The image does not pass validation.", name);
4526113f
LP
2799
2800 case -EADDRNOTAVAIL:
48084df6 2801 return log_error_errno(r, "%s: No root partition for specified root hash found.", name);
4526113f
LP
2802
2803 case -ENOTUNIQ:
48084df6 2804 return log_error_errno(r, "%s: Multiple suitable root partitions found in image.", name);
4526113f
LP
2805
2806 case -ENXIO:
48084df6 2807 return log_error_errno(r, "%s: No suitable root partition found in image.", name);
4526113f
LP
2808
2809 case -EPROTONOSUPPORT:
2810 return log_error_errno(r, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name);
2811
48084df6
ZJS
2812 case -ENOTBLK:
2813 return log_error_errno(r, "%s: Image is not a block device.", name);
2814
a94aa2b9
LP
2815 case -EBADR:
2816 return log_error_errno(r,
2817 "Combining partitioned images (such as '%s') with external Verity data (such as '%s') not supported. "
2818 "(Consider setting $SYSTEMD_DISSECT_VERITY_SIDECAR=0 to disable automatic discovery of external Verity data.)",
2819 name, strna(verity ? verity->data_path : NULL));
2820
4526113f
LP
2821 default:
2822 if (r < 0)
2823 return log_error_errno(r, "Failed to dissect image '%s': %m", name);
2824
2825 return r;
2826 }
2827}
2828
49536766
LP
2829bool dissected_image_verity_candidate(const DissectedImage *image, PartitionDesignator partition_designator) {
2830 assert(image);
2831
2832 /* Checks if this partition could theoretically do Verity. For non-partitioned images this only works
2833 * if there's an external verity file supplied, for which we can consult .has_verity. For partitioned
2834 * images we only check the partition type.
2835 *
2836 * This call is used to decide whether to suppress or show a verity column in tabular output of the
2837 * image. */
2838
e7cbe5cb 2839 if (image->single_file_system)
c3c88d67 2840 return partition_designator == PARTITION_ROOT && image->has_verity;
e7cbe5cb
LB
2841
2842 return PARTITION_VERITY_OF(partition_designator) >= 0;
2843}
2844
49536766
LP
2845bool dissected_image_verity_ready(const DissectedImage *image, PartitionDesignator partition_designator) {
2846 PartitionDesignator k;
2847
2848 assert(image);
2849
2850 /* Checks if this partition has verity data available that we can activate. For non-partitioned this
2851 * works for the root partition, for others only if the associated verity partition was found. */
2852
2853 if (!image->verity_ready)
2854 return false;
e7cbe5cb
LB
2855
2856 if (image->single_file_system)
49536766 2857 return partition_designator == PARTITION_ROOT;
e7cbe5cb
LB
2858
2859 k = PARTITION_VERITY_OF(partition_designator);
2860 return k >= 0 && image->partitions[k].found;
2861}
2862
8ee9615e
LP
2863bool dissected_image_verity_sig_ready(const DissectedImage *image, PartitionDesignator partition_designator) {
2864 PartitionDesignator k;
2865
2866 assert(image);
2867
2868 /* Checks if this partition has verity signature data available that we can use. */
2869
2870 if (!image->verity_sig_ready)
2871 return false;
2872
2873 if (image->single_file_system)
2874 return partition_designator == PARTITION_ROOT;
2875
2876 k = PARTITION_VERITY_SIG_OF(partition_designator);
2877 return k >= 0 && image->partitions[k].found;
2878}
2879
18d73705
LB
2880MountOptions* mount_options_free_all(MountOptions *options) {
2881 MountOptions *m;
2882
2883 while ((m = options)) {
2884 LIST_REMOVE(mount_options, options, m);
2885 free(m->options);
2886 free(m);
2887 }
2888
2889 return NULL;
2890}
2891
569a0e42 2892const char* mount_options_from_designator(const MountOptions *options, PartitionDesignator designator) {
f5215bc8 2893 LIST_FOREACH(mount_options, m, options)
9ece6444 2894 if (designator == m->partition_designator && !isempty(m->options))
18d73705 2895 return m->options;
6aa05ebd 2896
18d73705
LB
2897 return NULL;
2898}
2899
6aa05ebd
LP
2900int mount_image_privately_interactively(
2901 const char *image,
2902 DissectImageFlags flags,
2903 char **ret_directory,
2904 LoopDevice **ret_loop_device,
2905 DecryptedImage **ret_decrypted_image) {
2906
27ec815e 2907 _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT;
6aa05ebd
LP
2908 _cleanup_(loop_device_unrefp) LoopDevice *d = NULL;
2909 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
2910 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
2911 _cleanup_(rmdir_and_freep) char *created_dir = NULL;
2912 _cleanup_free_ char *temp = NULL;
2913 int r;
2914
2915 /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
2916 * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
2917 * easily. */
2918
2919 assert(image);
2920 assert(ret_directory);
2921 assert(ret_loop_device);
2922 assert(ret_decrypted_image);
2923
27ec815e
LP
2924 r = verity_settings_load(&verity, image, NULL, NULL);
2925 if (r < 0)
2926 return log_error_errno(r, "Failed to load root hash data: %m");
2927
6aa05ebd
LP
2928 r = tempfn_random_child(NULL, program_invocation_short_name, &temp);
2929 if (r < 0)
2930 return log_error_errno(r, "Failed to generate temporary mount directory: %m");
2931
2932 r = loop_device_make_by_path(
2933 image,
ef9c184d 2934 FLAGS_SET(flags, DISSECT_IMAGE_DEVICE_READ_ONLY) ? O_RDONLY : O_RDWR,
6aa05ebd 2935 FLAGS_SET(flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN,
7f52206a 2936 LOCK_SH,
6aa05ebd
LP
2937 &d);
2938 if (r < 0)
7b87fe4c 2939 return log_error_errno(r, "Failed to set up loopback device for %s: %m", image);
6aa05ebd 2940
64dd3a24 2941 r = dissect_loop_device_and_warn(d, &verity, NULL, flags, &dissected_image);
6aa05ebd
LP
2942 if (r < 0)
2943 return r;
2944
88b3300f
LP
2945 r = dissected_image_load_verity_sig_partition(dissected_image, d->fd, &verity);
2946 if (r < 0)
2947 return r;
2948
27ec815e 2949 r = dissected_image_decrypt_interactively(dissected_image, NULL, &verity, flags, &decrypted_image);
6aa05ebd
LP
2950 if (r < 0)
2951 return r;
2952
2953 r = detach_mount_namespace();
2954 if (r < 0)
2955 return log_error_errno(r, "Failed to detach mount namespace: %m");
2956
2957 r = mkdir_p(temp, 0700);
2958 if (r < 0)
2959 return log_error_errno(r, "Failed to create mount point: %m");
2960
2961 created_dir = TAKE_PTR(temp);
2962
21b61b1d 2963 r = dissected_image_mount_and_warn(dissected_image, created_dir, UID_INVALID, UID_INVALID, flags);
6aa05ebd 2964 if (r < 0)
af187ab2 2965 return r;
6aa05ebd 2966
41bc4849
LP
2967 r = loop_device_flock(d, LOCK_UN);
2968 if (r < 0)
2969 return r;
2970
6aa05ebd
LP
2971 if (decrypted_image) {
2972 r = decrypted_image_relinquish(decrypted_image);
2973 if (r < 0)
2974 return log_error_errno(r, "Failed to relinquish DM devices: %m");
2975 }
2976
2977 loop_device_relinquish(d);
2978
2979 *ret_directory = TAKE_PTR(created_dir);
2980 *ret_loop_device = TAKE_PTR(d);
2981 *ret_decrypted_image = TAKE_PTR(decrypted_image);
2982
2983 return 0;
2984}
2985
8c1be37e 2986static const char *const partition_designator_table[] = {
68ac5118
ZJS
2987 [PARTITION_ROOT] = "root",
2988 [PARTITION_ROOT_SECONDARY] = "root-secondary",
2989 [PARTITION_ROOT_OTHER] = "root-other",
2990 [PARTITION_USR] = "usr",
2991 [PARTITION_USR_SECONDARY] = "usr-secondary",
2992 [PARTITION_USR_OTHER] = "usr-other",
2993 [PARTITION_HOME] = "home",
2994 [PARTITION_SRV] = "srv",
2995 [PARTITION_ESP] = "esp",
2996 [PARTITION_XBOOTLDR] = "xbootldr",
2997 [PARTITION_SWAP] = "swap",
2998 [PARTITION_ROOT_VERITY] = "root-verity",
2999 [PARTITION_ROOT_SECONDARY_VERITY] = "root-secondary-verity",
3000 [PARTITION_ROOT_OTHER_VERITY] = "root-other-verity",
3001 [PARTITION_USR_VERITY] = "usr-verity",
3002 [PARTITION_USR_SECONDARY_VERITY] = "usr-secondary-verity",
3003 [PARTITION_USR_OTHER_VERITY] = "usr-other-verity",
3004 [PARTITION_ROOT_VERITY_SIG] = "root-verity-sig",
8ee9615e 3005 [PARTITION_ROOT_SECONDARY_VERITY_SIG] = "root-secondary-verity-sig",
68ac5118
ZJS
3006 [PARTITION_ROOT_OTHER_VERITY_SIG] = "root-other-verity-sig",
3007 [PARTITION_USR_VERITY_SIG] = "usr-verity-sig",
3008 [PARTITION_USR_SECONDARY_VERITY_SIG] = "usr-secondary-verity-sig",
3009 [PARTITION_USR_OTHER_VERITY_SIG] = "usr-other-verity-sig",
3010 [PARTITION_TMP] = "tmp",
3011 [PARTITION_VAR] = "var",
8c1be37e
LP
3012};
3013
93f59701 3014int verity_dissect_and_mount(
cedf5b1a 3015 int src_fd,
93f59701
LB
3016 const char *src,
3017 const char *dest,
3018 const MountOptions *options,
3019 const char *required_host_os_release_id,
3020 const char *required_host_os_release_version_id,
60c5f700
LP
3021 const char *required_host_os_release_sysext_level,
3022 const char *required_sysext_scope) {
93f59701 3023
4beda316
LB
3024 _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL;
3025 _cleanup_(decrypted_image_unrefp) DecryptedImage *decrypted_image = NULL;
3026 _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL;
3027 _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT;
3028 DissectImageFlags dissect_image_flags;
3029 int r;
3030
3031 assert(src);
3032 assert(dest);
3033
cedf5b1a 3034 /* We might get an FD for the image, but we use the original path to look for the dm-verity files */
4beda316
LB
3035 r = verity_settings_load(&verity, src, NULL, NULL);
3036 if (r < 0)
3037 return log_debug_errno(r, "Failed to load root hash: %m");
3038
3039 dissect_image_flags = verity.data_path ? DISSECT_IMAGE_NO_PARTITION_TABLE : 0;
3040
cedf5b1a
LB
3041 /* Note that we don't use loop_device_make here, as the FD is most likely O_PATH which would not be
3042 * accepted by LOOP_CONFIGURE, so just let loop_device_make_by_path reopen it as a regular FD. */
4beda316 3043 r = loop_device_make_by_path(
cedf5b1a 3044 src_fd >= 0 ? FORMAT_PROC_FD_PATH(src_fd) : src,
4beda316
LB
3045 -1,
3046 verity.data_path ? 0 : LO_FLAGS_PARTSCAN,
7f52206a 3047 LOCK_SH,
4beda316
LB
3048 &loop_device);
3049 if (r < 0)
3050 return log_debug_errno(r, "Failed to create loop device for image: %m");
3051
bad31660
YW
3052 r = dissect_loop_device(
3053 loop_device,
4beda316
LB
3054 &verity,
3055 options,
3056 dissect_image_flags,
3057 &dissected_image);
3058 /* No partition table? Might be a single-filesystem image, try again */
3059 if (!verity.data_path && r == -ENOPKG)
bad31660
YW
3060 r = dissect_loop_device(
3061 loop_device,
4beda316
LB
3062 &verity,
3063 options,
75dc190d 3064 dissect_image_flags | DISSECT_IMAGE_NO_PARTITION_TABLE,
4beda316
LB
3065 &dissected_image);
3066 if (r < 0)
3067 return log_debug_errno(r, "Failed to dissect image: %m");
3068
88b3300f
LP
3069 r = dissected_image_load_verity_sig_partition(dissected_image, loop_device->fd, &verity);
3070 if (r < 0)
3071 return r;
3072
4beda316
LB
3073 r = dissected_image_decrypt(
3074 dissected_image,
3075 NULL,
3076 &verity,
3077 dissect_image_flags,
3078 &decrypted_image);
3079 if (r < 0)
3080 return log_debug_errno(r, "Failed to decrypt dissected image: %m");
3081
3082 r = mkdir_p_label(dest, 0755);
3083 if (r < 0)
3084 return log_debug_errno(r, "Failed to create destination directory %s: %m", dest);
3085 r = umount_recursive(dest, 0);
3086 if (r < 0)
3087 return log_debug_errno(r, "Failed to umount under destination directory %s: %m", dest);
3088
21b61b1d 3089 r = dissected_image_mount(dissected_image, dest, UID_INVALID, UID_INVALID, dissect_image_flags);
4beda316
LB
3090 if (r < 0)
3091 return log_debug_errno(r, "Failed to mount image: %m");
3092
41bc4849
LP
3093 r = loop_device_flock(loop_device, LOCK_UN);
3094 if (r < 0)
3095 return log_debug_errno(r, "Failed to unlock loopback device: %m");
3096
93f59701
LB
3097 /* If we got os-release values from the caller, then we need to match them with the image's
3098 * extension-release.d/ content. Return -EINVAL if there's any mismatch.
3099 * First, check the distro ID. If that matches, then check the new SYSEXT_LEVEL value if
37361f46
LB
3100 * available, or else fallback to VERSION_ID. If neither is present (eg: rolling release),
3101 * then a simple match on the ID will be performed. */
8b2dcbbd 3102 if (required_host_os_release_id) {
93f59701
LB
3103 _cleanup_strv_free_ char **extension_release = NULL;
3104
d30d86b7
YW
3105 assert(!isempty(required_host_os_release_id));
3106
93f59701
LB
3107 r = load_extension_release_pairs(dest, dissected_image->image_name, &extension_release);
3108 if (r < 0)
3109 return log_debug_errno(r, "Failed to parse image %s extension-release metadata: %m", dissected_image->image_name);
3110
3111 r = extension_release_validate(
60c5f700
LP
3112 dissected_image->image_name,
3113 required_host_os_release_id,
3114 required_host_os_release_version_id,
3115 required_host_os_release_sysext_level,
3116 required_sysext_scope,
3117 extension_release);
93f59701
LB
3118 if (r == 0)
3119 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE), "Image %s extension-release metadata does not match the root's", dissected_image->image_name);
3120 if (r < 0)
3121 return log_debug_errno(r, "Failed to compare image %s extension-release metadata with the root's os-release: %m", dissected_image->image_name);
3122 }
3123
4beda316
LB
3124 if (decrypted_image) {
3125 r = decrypted_image_relinquish(decrypted_image);
3126 if (r < 0)
3127 return log_debug_errno(r, "Failed to relinquish decrypted image: %m");
3128 }
3129
3130 loop_device_relinquish(loop_device);
3131
3132 return 0;
3133}
3134
569a0e42 3135DEFINE_STRING_TABLE_LOOKUP(partition_designator, PartitionDesignator);