]>
Commit | Line | Data |
---|---|---|
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" |
5228c58e | 28 | #include "btrfs-util.h" |
f461a28d | 29 | #include "chase.h" |
c2fa92e7 | 30 | #include "conf-files.h" |
28db6fbf | 31 | #include "constants.h" |
3b925504 | 32 | #include "copy.h" |
1e2f3230 | 33 | #include "cryptsetup-util.h" |
553e15f2 | 34 | #include "device-nodes.h" |
8437c059 | 35 | #include "device-util.h" |
f7725647 | 36 | #include "devnum-util.h" |
7718ac97 | 37 | #include "discover-image.h" |
8c1be37e | 38 | #include "dissect-image.h" |
a709a315 | 39 | #include "dm-util.h" |
686d13b9 | 40 | #include "env-file.h" |
88b3300f | 41 | #include "env-util.h" |
d51f8eb3 | 42 | #include "extension-util.h" |
18b5886e | 43 | #include "fd-util.h" |
78ebe980 | 44 | #include "fileio.h" |
2eedfd2d | 45 | #include "fs-util.h" |
cf32c486 | 46 | #include "fsck-util.h" |
8c1be37e | 47 | #include "gpt.h" |
78ebe980 | 48 | #include "hexdecoct.h" |
e2054217 | 49 | #include "hostname-setup.h" |
3b925504 | 50 | #include "id128-util.h" |
593fe6c0 | 51 | #include "import-util.h" |
a4e0d617 | 52 | #include "io-util.h" |
254e392e | 53 | #include "missing_mount.h" |
8d9a1d59 | 54 | #include "missing_syscall.h" |
35cd0ba5 | 55 | #include "mkdir-label.h" |
8c1be37e | 56 | #include "mount-util.h" |
e4de7287 | 57 | #include "mountpoint-util.h" |
6aa05ebd | 58 | #include "namespace-util.h" |
d8b4d14d | 59 | #include "nulstr-util.h" |
c2fa92e7 | 60 | #include "openssl-util.h" |
d58ad743 | 61 | #include "os-util.h" |
8c1be37e | 62 | #include "path-util.h" |
3b925504 LP |
63 | #include "process-util.h" |
64 | #include "raw-clone.h" | |
81939d9d | 65 | #include "resize-fs.h" |
3b925504 | 66 | #include "signal-util.h" |
05c4c59f | 67 | #include "sparse-endian.h" |
8c1be37e | 68 | #include "stat-util.h" |
18b5886e | 69 | #include "stdio-util.h" |
8c1be37e LP |
70 | #include "string-table.h" |
71 | #include "string-util.h" | |
2eedfd2d | 72 | #include "strv.h" |
e4de7287 | 73 | #include "tmpfile-util.h" |
a8040b6d | 74 | #include "udev-util.h" |
2d3a5a73 | 75 | #include "user-util.h" |
41488e1f | 76 | #include "xattr-util.h" |
8c1be37e | 77 | |
28e2641a FF |
78 | /* how many times to wait for the device nodes to appear */ |
79 | #define N_DEVICE_NODE_LIST_ATTEMPTS 10 | |
80 | ||
80ce8580 LP |
81 | int dissect_fstype_ok(const char *fstype) { |
82 | const char *e; | |
83 | bool b; | |
84 | ||
85 | /* When we automatically mount file systems, be a bit conservative by default what we are willing to | |
86 | * mount, just as an extra safety net to not mount with badly maintained legacy file system | |
87 | * drivers. */ | |
88 | ||
89 | e = secure_getenv("SYSTEMD_DISSECT_FILE_SYSTEMS"); | |
90 | if (e) { | |
91 | _cleanup_strv_free_ char **l = NULL; | |
92 | ||
93 | l = strv_split(e, ":"); | |
94 | if (!l) | |
95 | return -ENOMEM; | |
96 | ||
97 | b = strv_contains(l, fstype); | |
98 | } else | |
99 | b = STR_IN_SET(fstype, | |
100 | "btrfs", | |
101 | "erofs", | |
102 | "ext4", | |
ee6cf8ea | 103 | "f2fs", |
80ce8580 LP |
104 | "squashfs", |
105 | "vfat", | |
106 | "xfs"); | |
107 | if (b) | |
108 | return true; | |
109 | ||
110 | log_debug("File system type '%s' is not allowed to be mounted as result of automatic dissection.", fstype); | |
111 | return false; | |
112 | } | |
113 | ||
05c4c59f LP |
114 | int probe_sector_size(int fd, uint32_t *ret) { |
115 | ||
05c4c59f LP |
116 | /* Disk images might be for 512B or for 4096 sector sizes, let's try to auto-detect that by searching |
117 | * for the GPT headers at the relevant byte offsets */ | |
118 | ||
6b0651df | 119 | assert_cc(sizeof(GptHeader) == 92); |
05c4c59f LP |
120 | |
121 | /* We expect a sector size in the range 512…4096. The GPT header is located in the second | |
122 | * sector. Hence it could be at byte 512 at the earliest, and at byte 4096 at the latest. And we must | |
123 | * read with granularity of the largest sector size we care about. Which means 8K. */ | |
124 | uint8_t sectors[2 * 4096]; | |
125 | uint32_t found = 0; | |
126 | ssize_t n; | |
127 | ||
128 | assert(fd >= 0); | |
129 | assert(ret); | |
130 | ||
131 | n = pread(fd, sectors, sizeof(sectors), 0); | |
132 | if (n < 0) | |
133 | return -errno; | |
134 | if (n != sizeof(sectors)) /* too short? */ | |
135 | goto not_found; | |
136 | ||
137 | /* Let's see if we find the GPT partition header with various expected sector sizes */ | |
138 | for (uint32_t sz = 512; sz <= 4096; sz <<= 1) { | |
6b0651df | 139 | const GptHeader *p; |
05c4c59f LP |
140 | |
141 | assert(sizeof(sectors) >= sz * 2); | |
6b0651df | 142 | p = (const GptHeader*) (sectors + sz); |
05c4c59f | 143 | |
6b0651df | 144 | if (!gpt_header_has_signature(p)) |
05c4c59f LP |
145 | continue; |
146 | ||
147 | if (found != 0) | |
148 | return log_debug_errno(SYNTHETIC_ERRNO(ENOTUNIQ), | |
149 | "Detected valid partition table at offsets matching multiple sector sizes, refusing."); | |
150 | ||
151 | found = sz; | |
152 | } | |
153 | ||
154 | if (found != 0) { | |
155 | log_debug("Determined sector size %" PRIu32 " based on discovered partition table.", found); | |
156 | *ret = found; | |
157 | return 1; /* indicate we *did* find it */ | |
158 | } | |
159 | ||
160 | not_found: | |
161 | log_debug("Couldn't find any partition table to derive sector size of."); | |
162 | *ret = 512; /* pick the traditional default */ | |
163 | return 0; /* indicate we didn't find it */ | |
164 | } | |
165 | ||
81dde3d8 LP |
166 | int probe_sector_size_prefer_ioctl(int fd, uint32_t *ret) { |
167 | struct stat st; | |
168 | ||
169 | assert(fd >= 0); | |
170 | assert(ret); | |
171 | ||
172 | /* Just like probe_sector_size(), but if we are looking at a block device, will use the already | |
173 | * configured sector size rather than probing by contents */ | |
174 | ||
175 | if (fstat(fd, &st) < 0) | |
176 | return -errno; | |
177 | ||
178 | if (S_ISBLK(st.st_mode)) | |
179 | return blockdev_get_sector_size(fd, ret); | |
180 | ||
181 | return probe_sector_size(fd, ret); | |
182 | } | |
183 | ||
c80c9079 LP |
184 | int probe_filesystem_full( |
185 | int fd, | |
186 | const char *path, | |
187 | uint64_t offset, | |
188 | uint64_t size, | |
189 | char **ret_fstype) { | |
190 | ||
7cc84b2c | 191 | /* Try to find device content type and return it in *ret_fstype. If nothing is found, |
c80c9079 | 192 | * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and a |
7cc84b2c ZJS |
193 | * different error otherwise. */ |
194 | ||
349cc4a5 | 195 | #if HAVE_BLKID |
8e766630 | 196 | _cleanup_(blkid_free_probep) blkid_probe b = NULL; |
d2c6e79d | 197 | _cleanup_free_ char *path_by_fd = NULL; |
254d1313 | 198 | _cleanup_close_ int fd_close = -EBADF; |
18b5886e LP |
199 | const char *fstype; |
200 | int r; | |
201 | ||
d2c6e79d YW |
202 | assert(fd >= 0 || path); |
203 | assert(ret_fstype); | |
204 | ||
205 | if (fd < 0) { | |
206 | fd_close = open(path, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY); | |
207 | if (fd_close < 0) | |
208 | return -errno; | |
209 | ||
210 | fd = fd_close; | |
211 | } | |
212 | ||
213 | if (!path) { | |
214 | r = fd_get_path(fd, &path_by_fd); | |
215 | if (r < 0) | |
216 | return r; | |
217 | ||
218 | path = path_by_fd; | |
219 | } | |
220 | ||
c80c9079 LP |
221 | if (size == 0) /* empty size? nothing found! */ |
222 | goto not_found; | |
223 | ||
d2c6e79d | 224 | b = blkid_new_probe(); |
18b5886e | 225 | if (!b) |
d2c6e79d YW |
226 | return -ENOMEM; |
227 | ||
4d49f44f LP |
228 | /* The Linux kernel maintains separate block device caches for main ("whole") and partition block |
229 | * devices, which means making a change to one might not be reflected immediately when reading via | |
230 | * the other. That's massively confusing when mixing accesses to such devices. Let's address this in | |
231 | * a limited way: when probing a file system that is not at the beginning of the block device we | |
232 | * apparently probe a partition via the main block device, and in that case let's first flush the | |
233 | * main block device cache, so that we get the data that the per-partition block device last | |
234 | * sync'ed on. | |
235 | * | |
236 | * This only works under the assumption that any tools that write to the partition block devices | |
237 | * issue an syncfs()/fsync() on the device after making changes. Typically file system formatting | |
238 | * tools that write a superblock onto a partition block device do that, however. */ | |
239 | if (offset != 0) | |
240 | if (ioctl(fd, BLKFLSBUF, 0) < 0) | |
241 | log_debug_errno(errno, "Failed to flush block device cache, ignoring: %m"); | |
242 | ||
d2c6e79d | 243 | errno = 0; |
c80c9079 LP |
244 | r = blkid_probe_set_device( |
245 | b, | |
246 | fd, | |
247 | offset, | |
248 | size == UINT64_MAX ? 0 : size); /* when blkid sees size=0 it understands "everything". We prefer using UINT64_MAX for that */ | |
d2c6e79d | 249 | if (r != 0) |
66855de7 | 250 | return errno_or_else(ENOMEM); |
18b5886e LP |
251 | |
252 | blkid_probe_enable_superblocks(b, 1); | |
253 | blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE); | |
254 | ||
255 | errno = 0; | |
256 | r = blkid_do_safeprobe(b); | |
2e3944b8 | 257 | if (r == _BLKID_SAFEPROBE_NOT_FOUND) |
18b5886e | 258 | goto not_found; |
2e3944b8 | 259 | if (r == _BLKID_SAFEPROBE_AMBIGUOUS) |
58dfbfbd | 260 | return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN), |
d2c6e79d | 261 | "Results ambiguous for partition %s", path); |
2e3944b8 | 262 | if (r == _BLKID_SAFEPROBE_ERROR) |
d2c6e79d | 263 | return log_debug_errno(errno_or_else(EIO), "Failed to probe partition %s: %m", path); |
18b5886e | 264 | |
2e3944b8 LP |
265 | assert(r == _BLKID_SAFEPROBE_FOUND); |
266 | ||
18b5886e LP |
267 | (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL); |
268 | ||
269 | if (fstype) { | |
270 | char *t; | |
271 | ||
d2c6e79d | 272 | log_debug("Probed fstype '%s' on partition %s.", fstype, path); |
1cefb9a6 | 273 | |
18b5886e LP |
274 | t = strdup(fstype); |
275 | if (!t) | |
276 | return -ENOMEM; | |
277 | ||
278 | *ret_fstype = t; | |
279 | return 1; | |
280 | } | |
281 | ||
282 | not_found: | |
d2c6e79d | 283 | log_debug("No type detected on partition %s", path); |
18b5886e LP |
284 | *ret_fstype = NULL; |
285 | return 0; | |
d1c536f5 ZJS |
286 | #else |
287 | return -EOPNOTSUPP; | |
a75e27eb | 288 | #endif |
d1c536f5 | 289 | } |
18b5886e | 290 | |
40c10d3f | 291 | #if HAVE_BLKID |
cd22d856 LP |
292 | static int image_policy_may_use( |
293 | const ImagePolicy *policy, | |
294 | PartitionDesignator designator) { | |
295 | ||
296 | PartitionPolicyFlags f; | |
297 | ||
298 | /* For each partition we find in the partition table do a first check if it may exist at all given | |
299 | * the policy, or if it shall be ignored. */ | |
300 | ||
301 | f = image_policy_get_exhaustively(policy, designator); | |
302 | if (f < 0) | |
303 | return f; | |
304 | ||
305 | if ((f & _PARTITION_POLICY_USE_MASK) == PARTITION_POLICY_ABSENT) | |
306 | /* only flag set in policy is "absent"? then this partition may not exist at all */ | |
307 | return log_debug_errno( | |
308 | SYNTHETIC_ERRNO(ERFKILL), | |
309 | "Partition of designator '%s' exists, but not allowed by policy, refusing.", | |
310 | partition_designator_to_string(designator)); | |
311 | if ((f & _PARTITION_POLICY_USE_MASK & ~PARTITION_POLICY_ABSENT) == PARTITION_POLICY_UNUSED) { | |
312 | /* only "unused" or "unused" + "absent" are set? then don't use it */ | |
313 | log_debug("Partition of designator '%s' exists, and policy dictates to ignore it, doing so.", | |
314 | partition_designator_to_string(designator)); | |
315 | return false; /* ignore! */ | |
316 | } | |
317 | ||
318 | return true; /* use! */ | |
319 | } | |
320 | ||
321 | static int image_policy_check_protection( | |
322 | const ImagePolicy *policy, | |
323 | PartitionDesignator designator, | |
324 | PartitionPolicyFlags found_flags) { | |
325 | ||
326 | PartitionPolicyFlags policy_flags; | |
327 | ||
328 | /* Checks if the flags in the policy for the designated partition overlap the flags of what we found */ | |
329 | ||
330 | if (found_flags < 0) | |
331 | return found_flags; | |
332 | ||
333 | policy_flags = image_policy_get_exhaustively(policy, designator); | |
334 | if (policy_flags < 0) | |
335 | return policy_flags; | |
336 | ||
337 | if ((found_flags & policy_flags) == 0) { | |
338 | _cleanup_free_ char *found_flags_string = NULL, *policy_flags_string = NULL; | |
339 | ||
340 | (void) partition_policy_flags_to_string(found_flags, /* simplify= */ true, &found_flags_string); | |
341 | (void) partition_policy_flags_to_string(policy_flags, /* simplify= */ true, &policy_flags_string); | |
342 | ||
343 | return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL), "Partition %s discovered with policy '%s' but '%s' was required, refusing.", | |
344 | partition_designator_to_string(designator), | |
345 | strnull(found_flags_string), strnull(policy_flags_string)); | |
346 | } | |
347 | ||
348 | return 0; | |
349 | } | |
350 | ||
351 | static int image_policy_check_partition_flags( | |
352 | const ImagePolicy *policy, | |
353 | PartitionDesignator designator, | |
354 | uint64_t gpt_flags) { | |
355 | ||
356 | PartitionPolicyFlags policy_flags; | |
357 | bool b; | |
358 | ||
359 | /* Checks if the partition flags in the policy match reality */ | |
360 | ||
361 | policy_flags = image_policy_get_exhaustively(policy, designator); | |
362 | if (policy_flags < 0) | |
363 | return policy_flags; | |
364 | ||
365 | b = FLAGS_SET(gpt_flags, SD_GPT_FLAG_READ_ONLY); | |
366 | if ((policy_flags & _PARTITION_POLICY_READ_ONLY_MASK) == (b ? PARTITION_POLICY_READ_ONLY_OFF : PARTITION_POLICY_READ_ONLY_ON)) | |
367 | return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL), "Partition %s has 'read-only' flag incorrectly set (must be %s, is %s), refusing.", | |
368 | partition_designator_to_string(designator), | |
369 | one_zero(!b), one_zero(b)); | |
370 | ||
371 | b = FLAGS_SET(gpt_flags, SD_GPT_FLAG_GROWFS); | |
372 | if ((policy_flags & _PARTITION_POLICY_GROWFS_MASK) == (b ? PARTITION_POLICY_GROWFS_OFF : PARTITION_POLICY_GROWFS_ON)) | |
373 | return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL), "Partition %s has 'growfs' flag incorrectly set (must be %s, is %s), refusing.", | |
374 | partition_designator_to_string(designator), | |
375 | one_zero(!b), one_zero(b)); | |
376 | ||
377 | return 0; | |
378 | } | |
379 | ||
380 | static int dissected_image_probe_filesystems( | |
381 | DissectedImage *m, | |
382 | int fd, | |
383 | const ImagePolicy *policy) { | |
384 | ||
698bb074 YW |
385 | int r; |
386 | ||
387 | assert(m); | |
388 | ||
389 | /* Fill in file system types if we don't know them yet. */ | |
390 | ||
391 | for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) { | |
392 | DissectedPartition *p = m->partitions + i; | |
cd22d856 | 393 | PartitionPolicyFlags found_flags; |
698bb074 YW |
394 | |
395 | if (!p->found) | |
396 | continue; | |
397 | ||
c80c9079 LP |
398 | if (!p->fstype) { |
399 | /* If we have an fd referring to the partition block device, use that. Otherwise go | |
400 | * via the whole block device or backing regular file, and read via offset. */ | |
401 | if (p->mount_node_fd >= 0) | |
402 | r = probe_filesystem_full(p->mount_node_fd, p->node, 0, UINT64_MAX, &p->fstype); | |
403 | else | |
404 | r = probe_filesystem_full(fd, p->node, p->offset, p->size, &p->fstype); | |
405 | if (r < 0) | |
698bb074 YW |
406 | return r; |
407 | } | |
408 | ||
cd22d856 | 409 | if (streq_ptr(p->fstype, "crypto_LUKS")) { |
698bb074 | 410 | m->encrypted = true; |
cd22d856 LP |
411 | found_flags = PARTITION_POLICY_ENCRYPTED; /* found this one, and its definitely encrypted */ |
412 | } else | |
413 | /* found it, but it's definitely not encrypted, hence mask the encrypted flag, but | |
414 | * set all other ways that indicate "present". */ | |
415 | found_flags = PARTITION_POLICY_UNPROTECTED|PARTITION_POLICY_VERITY|PARTITION_POLICY_SIGNED; | |
698bb074 YW |
416 | |
417 | if (p->fstype && fstype_is_ro(p->fstype)) | |
418 | p->rw = false; | |
419 | ||
420 | if (!p->rw) | |
421 | p->growfs = false; | |
cd22d856 LP |
422 | |
423 | /* We might have learnt more about the file system now (i.e. whether it is encrypted or not), | |
424 | * hence we need to validate this against policy again, to see if the policy still matches | |
425 | * with this new information. Note that image_policy_check_protection() will check for | |
426 | * overlap between what's allowed in the policy and what we pass as 'found_policy' here. In | |
427 | * the unencrypted case we thus might pass an overly unspecific mask here (i.e. unprotected | |
428 | * OR verity OR signed), but that's fine since the earlier policy check already checked more | |
429 | * specific which of those three cases where OK. Keep in mind that this function here only | |
430 | * looks at specific partitions (and thus can only deduce encryption or not) but not the | |
431 | * overall partition table (and thus cannot deduce verity or not). The earlier dissection | |
432 | * checks already did the relevant checks that look at the whole partition table, and | |
433 | * enforced policy there as needed. */ | |
434 | r = image_policy_check_protection(policy, i, found_flags); | |
435 | if (r < 0) | |
436 | return r; | |
698bb074 YW |
437 | } |
438 | ||
439 | return 0; | |
440 | } | |
441 | ||
0f7c9a3d LP |
442 | static void check_partition_flags( |
443 | const char *node, | |
444 | unsigned long long pflags, | |
445 | unsigned long long supported) { | |
446 | ||
447 | assert(node); | |
448 | ||
449 | /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */ | |
92e72028 ZJS |
450 | pflags &= ~(supported | |
451 | SD_GPT_FLAG_REQUIRED_PARTITION | | |
452 | SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL | | |
453 | SD_GPT_FLAG_LEGACY_BIOS_BOOTABLE); | |
0f7c9a3d LP |
454 | |
455 | if (pflags == 0) | |
456 | return; | |
457 | ||
458 | /* If there are other bits set, then log about it, to make things discoverable */ | |
459 | for (unsigned i = 0; i < sizeof(pflags) * 8; i++) { | |
460 | unsigned long long bit = 1ULL << i; | |
461 | if (!FLAGS_SET(pflags, bit)) | |
462 | continue; | |
463 | ||
464 | log_debug("Unexpected partition flag %llu set on %s!", bit, node); | |
465 | } | |
466 | } | |
467 | ||
00e29505 YW |
468 | static int dissected_image_new(const char *path, DissectedImage **ret) { |
469 | _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL; | |
470 | _cleanup_free_ char *name = NULL; | |
471 | int r; | |
472 | ||
473 | assert(ret); | |
474 | ||
475 | if (path) { | |
476 | _cleanup_free_ char *filename = NULL; | |
477 | ||
478 | r = path_extract_filename(path, &filename); | |
479 | if (r < 0) | |
480 | return r; | |
481 | ||
482 | r = raw_strip_suffixes(filename, &name); | |
483 | if (r < 0) | |
484 | return r; | |
485 | ||
486 | if (!image_name_is_valid(name)) { | |
487 | log_debug("Image name %s is not valid, ignoring.", strna(name)); | |
488 | name = mfree(name); | |
489 | } | |
490 | } | |
491 | ||
492 | m = new(DissectedImage, 1); | |
493 | if (!m) | |
494 | return -ENOMEM; | |
495 | ||
496 | *m = (DissectedImage) { | |
497 | .has_init_system = -1, | |
498 | .image_name = TAKE_PTR(name), | |
499 | }; | |
500 | ||
babd5b08 YW |
501 | for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) |
502 | m->partitions[i] = DISSECTED_PARTITION_NULL; | |
503 | ||
00e29505 YW |
504 | *ret = TAKE_PTR(m); |
505 | return 0; | |
506 | } | |
507 | #endif | |
508 | ||
234c2e16 | 509 | static void dissected_partition_done(DissectedPartition *p) { |
1b010ae7 | 510 | assert(p); |
786e3a52 | 511 | |
1b010ae7 LP |
512 | free(p->fstype); |
513 | free(p->node); | |
514 | free(p->label); | |
515 | free(p->decrypted_fstype); | |
516 | free(p->decrypted_node); | |
517 | free(p->mount_options); | |
f7725647 | 518 | safe_close(p->mount_node_fd); |
8d9a1d59 | 519 | safe_close(p->fsmount_fd); |
786e3a52 | 520 | |
babd5b08 | 521 | *p = DISSECTED_PARTITION_NULL; |
1b010ae7 | 522 | } |
786e3a52 | 523 | |
1b010ae7 | 524 | #if HAVE_BLKID |
1b010ae7 LP |
525 | static int make_partition_devname( |
526 | const char *whole_devname, | |
1a81ddef | 527 | uint64_t diskseq, |
1b010ae7 | 528 | int nr, |
1a81ddef | 529 | DissectImageFlags flags, |
1b010ae7 | 530 | char **ret) { |
786e3a52 | 531 | |
1a81ddef LP |
532 | _cleanup_free_ char *s = NULL; |
533 | int r; | |
786e3a52 | 534 | |
1b010ae7 | 535 | assert(whole_devname); |
1a81ddef LP |
536 | assert(nr != 0); /* zero is not a valid partition nr */ |
537 | assert(ret); | |
aae22eb3 | 538 | |
c12a0d6d | 539 | if (!FLAGS_SET(flags, DISSECT_IMAGE_DISKSEQ_DEVNODE) || diskseq == 0) { |
4ba86848 | 540 | |
1a81ddef LP |
541 | /* Given a whole block device node name (e.g. /dev/sda or /dev/loop7) generate a partition |
542 | * device name (e.g. /dev/sda7 or /dev/loop7p5). The rule the kernel uses is simple: if whole | |
543 | * block device node name ends in a digit, then suffix a 'p', followed by the partition | |
544 | * number. Otherwise, just suffix the partition number without any 'p'. */ | |
08fe0a53 | 545 | |
1a81ddef LP |
546 | if (nr < 0) { /* whole disk? */ |
547 | s = strdup(whole_devname); | |
548 | if (!s) | |
549 | return -ENOMEM; | |
550 | } else { | |
551 | size_t l = strlen(whole_devname); | |
552 | if (l < 1) /* underflow check for the subtraction below */ | |
553 | return -EINVAL; | |
08fe0a53 | 554 | |
1a81ddef LP |
555 | bool need_p = ascii_isdigit(whole_devname[l-1]); /* Last char a digit? */ |
556 | ||
557 | if (asprintf(&s, "%s%s%i", whole_devname, need_p ? "p" : "", nr) < 0) | |
558 | return -ENOMEM; | |
559 | } | |
560 | } else { | |
561 | if (nr < 0) /* whole disk? */ | |
562 | r = asprintf(&s, "/dev/disk/by-diskseq/%" PRIu64, diskseq); | |
563 | else | |
564 | r = asprintf(&s, "/dev/disk/by-diskseq/%" PRIu64 "-part%i", diskseq, nr); | |
565 | if (r < 0) | |
566 | return -ENOMEM; | |
567 | } | |
568 | ||
569 | *ret = TAKE_PTR(s); | |
570 | return 0; | |
08fe0a53 LP |
571 | } |
572 | ||
1a81ddef LP |
573 | static int open_partition( |
574 | const char *node, | |
575 | bool is_partition, | |
576 | const LoopDevice *loop) { | |
577 | ||
f7725647 | 578 | _cleanup_(sd_device_unrefp) sd_device *dev = NULL; |
254d1313 | 579 | _cleanup_close_ int fd = -EBADF; |
f7725647 YW |
580 | dev_t devnum; |
581 | int r; | |
582 | ||
583 | assert(node); | |
584 | assert(loop); | |
585 | ||
586 | fd = open(node, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY); | |
587 | if (fd < 0) | |
588 | return -errno; | |
589 | ||
590 | /* Check if the block device is a child of (or equivalent to) the originally provided one. */ | |
591 | r = block_device_new_from_fd(fd, is_partition ? BLOCK_DEVICE_LOOKUP_WHOLE_DISK : 0, &dev); | |
592 | if (r < 0) | |
593 | return r; | |
594 | ||
595 | r = sd_device_get_devnum(dev, &devnum); | |
596 | if (r < 0) | |
597 | return r; | |
598 | ||
599 | if (loop->devno != devnum) | |
600 | return -ENXIO; | |
601 | ||
602 | /* Also check diskseq. */ | |
c12a0d6d | 603 | if (loop->diskseq != 0) { |
f7725647 YW |
604 | uint64_t diskseq; |
605 | ||
606 | r = fd_get_diskseq(fd, &diskseq); | |
607 | if (r < 0) | |
608 | return r; | |
609 | ||
610 | if (loop->diskseq != diskseq) | |
611 | return -ENXIO; | |
612 | } | |
613 | ||
614 | log_debug("Opened %s (fd=%i, whole_block_devnum=" DEVNUM_FORMAT_STR ", diskseq=%" PRIu64 ").", | |
615 | node, fd, DEVNUM_FORMAT_VAL(loop->devno), loop->diskseq); | |
616 | return TAKE_FD(fd); | |
617 | } | |
618 | ||
22e932f4 DDM |
619 | static int compare_arch(Architecture a, Architecture b) { |
620 | if (a == b) | |
621 | return 0; | |
622 | ||
623 | if (a == native_architecture()) | |
624 | return 1; | |
625 | ||
626 | if (b == native_architecture()) | |
627 | return -1; | |
628 | ||
629 | #ifdef ARCHITECTURE_SECONDARY | |
630 | if (a == ARCHITECTURE_SECONDARY) | |
631 | return 1; | |
632 | ||
633 | if (b == ARCHITECTURE_SECONDARY) | |
634 | return -1; | |
635 | #endif | |
636 | ||
637 | return 0; | |
638 | } | |
639 | ||
08f14be4 YW |
640 | static int dissect_image( |
641 | DissectedImage *m, | |
4526113f | 642 | int fd, |
0b214aa0 | 643 | const char *devname, |
89e62e0b | 644 | const VeritySettings *verity, |
18d73705 | 645 | const MountOptions *mount_options, |
84be0c71 | 646 | const ImagePolicy *policy, |
08f14be4 | 647 | DissectImageFlags flags) { |
8c1be37e | 648 | |
62ea0ed0 | 649 | sd_id128_t root_uuid = SD_ID128_NULL, root_verity_uuid = SD_ID128_NULL; |
62ea0ed0 | 650 | sd_id128_t usr_uuid = SD_ID128_NULL, usr_verity_uuid = SD_ID128_NULL; |
1f8fb21c | 651 | bool is_gpt, is_mbr, multiple_generic = false, |
de98f631 LP |
652 | generic_rw = false, /* initialize to appease gcc */ |
653 | generic_growfs = false; | |
8e766630 | 654 | _cleanup_(blkid_free_probep) blkid_probe b = NULL; |
8c1be37e | 655 | _cleanup_free_ char *generic_node = NULL; |
be30ad41 | 656 | sd_id128_t generic_uuid = SD_ID128_NULL; |
b387778c | 657 | const char *pttype = NULL, *sptuuid = NULL; |
8c1be37e | 658 | blkid_partlist pl; |
1f8fb21c | 659 | int r, generic_nr = -1, n_partitions; |
8c1be37e | 660 | |
08f14be4 | 661 | assert(m); |
8c1be37e | 662 | assert(fd >= 0); |
0b214aa0 | 663 | assert(devname); |
a0bff7ea | 664 | assert(!verity || verity->designator < 0 || IN_SET(verity->designator, PARTITION_ROOT, PARTITION_USR)); |
89e62e0b | 665 | assert(!verity || verity->root_hash || verity->root_hash_size == 0); |
a0bff7ea LP |
666 | assert(!verity || verity->root_hash_sig || verity->root_hash_sig_size == 0); |
667 | assert(!verity || (verity->root_hash || !verity->root_hash_sig)); | |
e7cbe5cb | 668 | assert(!((flags & DISSECT_IMAGE_GPT_ONLY) && (flags & DISSECT_IMAGE_NO_PARTITION_TABLE))); |
22ee78a8 | 669 | assert(m->sector_size > 0); |
8c1be37e LP |
670 | |
671 | /* Probes a disk image, and returns information about what it found in *ret. | |
672 | * | |
4623e8e6 | 673 | * Returns -ENOPKG if no suitable partition table or file system could be found. |
2679f407 LP |
674 | * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found. |
675 | * Returns -ENXIO if we couldn't find any partition suitable as root or /usr partition | |
97ce55e3 LP |
676 | * Returns -ENOTUNIQ if we only found multiple generic partitions and thus don't know what to do with that |
677 | * Returns -ERFKILL if image doesn't match image policy | |
678 | * Returns -EBADR if verity data was provided externally for an image that has a GPT partition table (i.e. is not just a naked fs) | |
679 | * Returns -EPROTONOSUPPORT if DISSECT_IMAGE_ADD_PARTITION_DEVICES is set but the block device does not have partition logic enabled | |
680 | * Returns -ENOMSG if we didn't find a single usable partition (and DISSECT_IMAGE_REFUSE_EMPTY is set) */ | |
4623e8e6 | 681 | |
c12a0d6d | 682 | uint64_t diskseq = m->loop ? m->loop->diskseq : 0; |
1a81ddef | 683 | |
89e62e0b | 684 | if (verity && verity->root_hash) { |
aee36b4e LP |
685 | sd_id128_t fsuuid, vuuid; |
686 | ||
687 | /* If a root hash is supplied, then we use the root partition that has a UUID that match the | |
da890466 ZJS |
688 | * first 128-bit of the root hash. And we use the verity partition that has a UUID that match |
689 | * the final 128-bit. */ | |
4623e8e6 | 690 | |
89e62e0b | 691 | if (verity->root_hash_size < sizeof(sd_id128_t)) |
4623e8e6 LP |
692 | return -EINVAL; |
693 | ||
aee36b4e LP |
694 | memcpy(&fsuuid, verity->root_hash, sizeof(sd_id128_t)); |
695 | memcpy(&vuuid, (const uint8_t*) verity->root_hash + verity->root_hash_size - sizeof(sd_id128_t), sizeof(sd_id128_t)); | |
4623e8e6 | 696 | |
aee36b4e | 697 | if (sd_id128_is_null(fsuuid)) |
4623e8e6 | 698 | return -EINVAL; |
aee36b4e | 699 | if (sd_id128_is_null(vuuid)) |
4623e8e6 | 700 | return -EINVAL; |
aee36b4e LP |
701 | |
702 | /* If the verity data declares it's for the /usr partition, then search for that, in all | |
703 | * other cases assume it's for the root partition. */ | |
704 | if (verity->designator == PARTITION_USR) { | |
705 | usr_uuid = fsuuid; | |
706 | usr_verity_uuid = vuuid; | |
707 | } else { | |
708 | root_uuid = fsuuid; | |
709 | root_verity_uuid = vuuid; | |
710 | } | |
4623e8e6 | 711 | } |
8c1be37e | 712 | |
8c1be37e LP |
713 | b = blkid_new_probe(); |
714 | if (!b) | |
715 | return -ENOMEM; | |
716 | ||
717 | errno = 0; | |
718 | r = blkid_probe_set_device(b, fd, 0, 0); | |
b382db9f | 719 | if (r != 0) |
66855de7 | 720 | return errno_or_else(ENOMEM); |
8c1be37e | 721 | |
22ee78a8 LP |
722 | errno = 0; |
723 | r = blkid_probe_set_sectorsize(b, m->sector_size); | |
724 | if (r != 0) | |
725 | return errno_or_else(EIO); | |
726 | ||
9b6deb03 LP |
727 | if ((flags & DISSECT_IMAGE_GPT_ONLY) == 0) { |
728 | /* Look for file system superblocks, unless we only shall look for GPT partition tables */ | |
729 | blkid_probe_enable_superblocks(b, 1); | |
b387778c | 730 | blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE|BLKID_SUBLKS_USAGE|BLKID_SUBLKS_UUID); |
9b6deb03 LP |
731 | } |
732 | ||
8c1be37e LP |
733 | blkid_probe_enable_partitions(b, 1); |
734 | blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS); | |
735 | ||
736 | errno = 0; | |
737 | r = blkid_do_safeprobe(b); | |
2e3944b8 | 738 | if (r == _BLKID_SAFEPROBE_ERROR) |
66855de7 | 739 | return errno_or_else(EIO); |
2e3944b8 LP |
740 | if (IN_SET(r, _BLKID_SAFEPROBE_AMBIGUOUS, _BLKID_SAFEPROBE_NOT_FOUND)) |
741 | return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG), "Failed to identify any partition table."); | |
742 | ||
743 | assert(r == _BLKID_SAFEPROBE_FOUND); | |
8c1be37e | 744 | |
e7cbe5cb | 745 | if ((!(flags & DISSECT_IMAGE_GPT_ONLY) && |
4b5de5dd | 746 | (flags & DISSECT_IMAGE_GENERIC_ROOT)) || |
e7cbe5cb | 747 | (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)) { |
9b6deb03 | 748 | const char *usage = NULL; |
8c1be37e | 749 | |
aee36b4e LP |
750 | /* If flags permit this, also allow using non-partitioned single-filesystem images */ |
751 | ||
9b6deb03 LP |
752 | (void) blkid_probe_lookup_value(b, "USAGE", &usage, NULL); |
753 | if (STRPTR_IN_SET(usage, "filesystem", "crypto")) { | |
18d73705 | 754 | _cleanup_free_ char *t = NULL, *n = NULL, *o = NULL; |
b387778c | 755 | const char *fstype = NULL, *options = NULL, *suuid = NULL; |
254d1313 | 756 | _cleanup_close_ int mount_node_fd = -EBADF; |
b387778c | 757 | sd_id128_t uuid = SD_ID128_NULL; |
cd22d856 LP |
758 | PartitionPolicyFlags found_flags; |
759 | bool encrypted; | |
760 | ||
761 | /* OK, we have found a file system, that's our root partition then. */ | |
762 | ||
763 | r = image_policy_may_use(policy, PARTITION_ROOT); | |
764 | if (r < 0) | |
765 | return r; | |
766 | if (r == 0) /* policy says ignore this, so we ignore it */ | |
767 | return -ENOPKG; | |
768 | ||
769 | (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL); | |
770 | (void) blkid_probe_lookup_value(b, "UUID", &suuid, NULL); | |
771 | ||
772 | encrypted = streq_ptr(fstype, "crypto_LUKS"); | |
773 | ||
774 | if (verity_settings_data_covers(verity, PARTITION_ROOT)) | |
775 | found_flags = verity->root_hash_sig ? PARTITION_POLICY_SIGNED : PARTITION_POLICY_VERITY; | |
776 | else | |
777 | found_flags = encrypted ? PARTITION_POLICY_ENCRYPTED : PARTITION_POLICY_UNPROTECTED; | |
778 | ||
779 | r = image_policy_check_protection(policy, PARTITION_ROOT, found_flags); | |
780 | if (r < 0) | |
781 | return r; | |
782 | ||
783 | r = image_policy_check_partition_flags(policy, PARTITION_ROOT, 0); /* we have no gpt partition flags, hence check against all bits off */ | |
784 | if (r < 0) | |
785 | return r; | |
f7725647 | 786 | |
73d88b80 | 787 | if (FLAGS_SET(flags, DISSECT_IMAGE_PIN_PARTITION_DEVICES)) { |
f7725647 YW |
788 | mount_node_fd = open_partition(devname, /* is_partition = */ false, m->loop); |
789 | if (mount_node_fd < 0) | |
790 | return mount_node_fd; | |
791 | } | |
8c1be37e | 792 | |
9b6deb03 LP |
793 | if (fstype) { |
794 | t = strdup(fstype); | |
795 | if (!t) | |
796 | return -ENOMEM; | |
797 | } | |
798 | ||
b387778c LP |
799 | if (suuid) { |
800 | /* blkid will return FAT's serial number as UUID, hence it is quite possible | |
801 | * that parsing this will fail. We'll ignore the ID, since it's just too | |
802 | * short to be useful as tru identifier. */ | |
803 | r = sd_id128_from_string(suuid, &uuid); | |
804 | if (r < 0) | |
805 | log_debug_errno(r, "Failed to parse file system UUID '%s', ignoring: %m", suuid); | |
806 | } | |
807 | ||
1a81ddef LP |
808 | r = make_partition_devname(devname, diskseq, -1, flags, &n); |
809 | if (r < 0) | |
810 | return r; | |
6c544d14 | 811 | |
e7cbe5cb | 812 | m->single_file_system = true; |
cd22d856 | 813 | m->encrypted = encrypted; |
c3c88d67 LP |
814 | |
815 | m->has_verity = verity && verity->data_path; | |
c2534821 | 816 | m->verity_ready = verity_settings_data_covers(verity, PARTITION_ROOT); |
e7cbe5cb | 817 | |
8ee9615e | 818 | m->has_verity_sig = false; /* signature not embedded, must be specified */ |
c2534821 | 819 | m->verity_sig_ready = m->verity_ready && verity->root_hash_sig; |
8ee9615e | 820 | |
b387778c LP |
821 | m->image_uuid = uuid; |
822 | ||
f5215bc8 | 823 | options = mount_options_from_designator(mount_options, PARTITION_ROOT); |
18d73705 LB |
824 | if (options) { |
825 | o = strdup(options); | |
826 | if (!o) | |
827 | return -ENOMEM; | |
828 | } | |
829 | ||
9b6deb03 LP |
830 | m->partitions[PARTITION_ROOT] = (DissectedPartition) { |
831 | .found = true, | |
e0d53d52 | 832 | .rw = !m->verity_ready && !fstype_is_ro(fstype), |
9b6deb03 LP |
833 | .partno = -1, |
834 | .architecture = _ARCHITECTURE_INVALID, | |
1cc6c93a YW |
835 | .fstype = TAKE_PTR(t), |
836 | .node = TAKE_PTR(n), | |
18d73705 | 837 | .mount_options = TAKE_PTR(o), |
f7725647 | 838 | .mount_node_fd = TAKE_FD(mount_node_fd), |
88b3300f LP |
839 | .offset = 0, |
840 | .size = UINT64_MAX, | |
8d9a1d59 | 841 | .fsmount_fd = -EBADF, |
9b6deb03 | 842 | }; |
8c1be37e | 843 | |
9b6deb03 LP |
844 | return 0; |
845 | } | |
8c1be37e LP |
846 | } |
847 | ||
848 | (void) blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL); | |
849 | if (!pttype) | |
850 | return -ENOPKG; | |
851 | ||
852 | is_gpt = streq_ptr(pttype, "gpt"); | |
853 | is_mbr = streq_ptr(pttype, "dos"); | |
854 | ||
9b6deb03 | 855 | if (!is_gpt && ((flags & DISSECT_IMAGE_GPT_ONLY) || !is_mbr)) |
8c1be37e LP |
856 | return -ENOPKG; |
857 | ||
0903fd26 LP |
858 | /* We support external verity data partitions only if the image has no partition table */ |
859 | if (verity && verity->data_path) | |
860 | return -EBADR; | |
861 | ||
73d88b80 | 862 | if (FLAGS_SET(flags, DISSECT_IMAGE_ADD_PARTITION_DEVICES)) { |
08f14be4 YW |
863 | /* Safety check: refuse block devices that carry a partition table but for which the kernel doesn't |
864 | * do partition scanning. */ | |
865 | r = blockdev_partscan_enabled(fd); | |
866 | if (r < 0) | |
867 | return r; | |
868 | if (r == 0) | |
869 | return -EPROTONOSUPPORT; | |
870 | } | |
4ba86848 | 871 | |
b387778c LP |
872 | (void) blkid_probe_lookup_value(b, "PTUUID", &sptuuid, NULL); |
873 | if (sptuuid) { | |
874 | r = sd_id128_from_string(sptuuid, &m->image_uuid); | |
875 | if (r < 0) | |
876 | log_debug_errno(r, "Failed to parse partition table UUID '%s', ignoring: %m", sptuuid); | |
877 | } | |
878 | ||
8c1be37e LP |
879 | errno = 0; |
880 | pl = blkid_probe_get_partitions(b); | |
b382db9f | 881 | if (!pl) |
66855de7 | 882 | return errno_or_else(ENOMEM); |
8c1be37e | 883 | |
4ba86848 LP |
884 | errno = 0; |
885 | n_partitions = blkid_partlist_numof_partitions(pl); | |
886 | if (n_partitions < 0) | |
887 | return errno_or_else(EIO); | |
8c1be37e | 888 | |
4ba86848 | 889 | for (int i = 0; i < n_partitions; i++) { |
1b010ae7 | 890 | _cleanup_free_ char *node = NULL; |
9b6deb03 | 891 | unsigned long long pflags; |
88b3300f | 892 | blkid_loff_t start, size; |
8c1be37e | 893 | blkid_partition pp; |
8c1be37e LP |
894 | int nr; |
895 | ||
4ba86848 LP |
896 | errno = 0; |
897 | pp = blkid_partlist_get_partition(pl, i); | |
898 | if (!pp) | |
899 | return errno_or_else(EIO); | |
aae22eb3 | 900 | |
9b6deb03 | 901 | pflags = blkid_partition_get_flags(pp); |
8c1be37e | 902 | |
4ba86848 | 903 | errno = 0; |
8c1be37e LP |
904 | nr = blkid_partition_get_partno(pp); |
905 | if (nr < 0) | |
4ba86848 | 906 | return errno_or_else(EIO); |
8c1be37e | 907 | |
88b3300f LP |
908 | errno = 0; |
909 | start = blkid_partition_get_start(pp); | |
910 | if (start < 0) | |
911 | return errno_or_else(EIO); | |
912 | ||
913 | assert((uint64_t) start < UINT64_MAX/512); | |
914 | ||
915 | errno = 0; | |
916 | size = blkid_partition_get_size(pp); | |
917 | if (size < 0) | |
918 | return errno_or_else(EIO); | |
919 | ||
920 | assert((uint64_t) size < UINT64_MAX/512); | |
921 | ||
1a81ddef LP |
922 | /* While probing we need the non-diskseq device node name to access the thing, hence mask off |
923 | * DISSECT_IMAGE_DISKSEQ_DEVNODE. */ | |
924 | r = make_partition_devname(devname, diskseq, nr, flags & ~DISSECT_IMAGE_DISKSEQ_DEVNODE, &node); | |
1b010ae7 LP |
925 | if (r < 0) |
926 | return r; | |
927 | ||
928 | /* So here's the thing: after the main ("whole") block device popped up it might take a while | |
929 | * before the kernel fully probed the partition table. Waiting for that to finish is icky in | |
930 | * userspace. So here's what we do instead. We issue the BLKPG_ADD_PARTITION ioctl to add the | |
931 | * partition ourselves, racing against the kernel. Good thing is: if this call fails with | |
932 | * EBUSY then the kernel was quicker than us, and that's totally OK, the outcome is good for | |
933 | * us: the device node will exist. If OTOH our call was successful we won the race. Which is | |
934 | * also good as the outcome is the same: the partition block device exists, and we can use | |
935 | * it. | |
936 | * | |
937 | * Kernel returns EBUSY if there's already a partition by that number or an overlapping | |
938 | * partition already existent. */ | |
939 | ||
73d88b80 | 940 | if (FLAGS_SET(flags, DISSECT_IMAGE_ADD_PARTITION_DEVICES)) { |
08f14be4 YW |
941 | r = block_device_add_partition(fd, node, nr, (uint64_t) start * 512, (uint64_t) size * 512); |
942 | if (r < 0) { | |
943 | if (r != -EBUSY) | |
944 | return log_debug_errno(r, "BLKPG_ADD_PARTITION failed: %m"); | |
1b010ae7 | 945 | |
08f14be4 YW |
946 | log_debug_errno(r, "Kernel was quicker than us in adding partition %i.", nr); |
947 | } else | |
948 | log_debug("We were quicker than kernel in adding partition %i.", nr); | |
949 | } | |
1b010ae7 | 950 | |
8c1be37e | 951 | if (is_gpt) { |
e3b9a5ff | 952 | const char *fstype = NULL, *label; |
4623e8e6 | 953 | sd_id128_t type_id, id; |
22e932f4 | 954 | GptPartitionType type; |
de98f631 | 955 | bool rw = true, growfs = false; |
8c1be37e | 956 | |
e3b9a5ff LP |
957 | r = blkid_partition_get_uuid_id128(pp, &id); |
958 | if (r < 0) { | |
959 | log_debug_errno(r, "Failed to read partition UUID, ignoring: %m"); | |
4623e8e6 | 960 | continue; |
e3b9a5ff | 961 | } |
4623e8e6 | 962 | |
e3b9a5ff LP |
963 | r = blkid_partition_get_type_id128(pp, &type_id); |
964 | if (r < 0) { | |
965 | log_debug_errno(r, "Failed to read partition type UUID, ignoring: %m"); | |
8c1be37e | 966 | continue; |
e3b9a5ff | 967 | } |
8c1be37e | 968 | |
22e932f4 DDM |
969 | type = gpt_partition_type_from_uuid(type_id); |
970 | ||
08fe0a53 LP |
971 | label = blkid_partition_get_name(pp); /* libblkid returns NULL here if empty */ |
972 | ||
fb3921b8 LP |
973 | if (IN_SET(type.designator, |
974 | PARTITION_HOME, | |
975 | PARTITION_SRV, | |
976 | PARTITION_XBOOTLDR, | |
977 | PARTITION_TMP)) { | |
a48dd347 | 978 | |
92e72028 ZJS |
979 | check_partition_flags(node, pflags, |
980 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS); | |
0f7c9a3d | 981 | |
92e72028 | 982 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
a48dd347 LP |
983 | continue; |
984 | ||
92e72028 ZJS |
985 | rw = !(pflags & SD_GPT_FLAG_READ_ONLY); |
986 | growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS); | |
aee36b4e | 987 | |
22e932f4 | 988 | } else if (type.designator == PARTITION_ESP) { |
a48dd347 | 989 | |
92e72028 ZJS |
990 | /* Note that we don't check the SD_GPT_FLAG_NO_AUTO flag for the ESP, as it is |
991 | * not defined there. We instead check the SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as | |
aee36b4e LP |
992 | * recommended by the UEFI spec (See "12.3.3 Number and Location of System |
993 | * Partitions"). */ | |
a48dd347 | 994 | |
92e72028 | 995 | if (pflags & SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL) |
a48dd347 LP |
996 | continue; |
997 | ||
8c1be37e | 998 | fstype = "vfat"; |
a8c47660 | 999 | |
22e932f4 | 1000 | } else if (type.designator == PARTITION_ROOT) { |
4623e8e6 | 1001 | |
92e72028 ZJS |
1002 | check_partition_flags(node, pflags, |
1003 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS); | |
0f7c9a3d | 1004 | |
92e72028 | 1005 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
a48dd347 LP |
1006 | continue; |
1007 | ||
4623e8e6 LP |
1008 | /* If a root ID is specified, ignore everything but the root id */ |
1009 | if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id)) | |
1010 | continue; | |
1011 | ||
92e72028 ZJS |
1012 | rw = !(pflags & SD_GPT_FLAG_READ_ONLY); |
1013 | growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS); | |
aee36b4e | 1014 | |
22e932f4 | 1015 | } else if (type.designator == PARTITION_ROOT_VERITY) { |
4623e8e6 | 1016 | |
92e72028 ZJS |
1017 | check_partition_flags(node, pflags, |
1018 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY); | |
0f7c9a3d | 1019 | |
92e72028 | 1020 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
a48dd347 LP |
1021 | continue; |
1022 | ||
c3c88d67 | 1023 | m->has_verity = true; |
4623e8e6 | 1024 | |
8ee9615e LP |
1025 | /* If no verity configuration is specified, then don't do verity */ |
1026 | if (!verity) | |
1027 | continue; | |
1028 | if (verity->designator >= 0 && verity->designator != PARTITION_ROOT) | |
1029 | continue; | |
1030 | ||
1031 | /* If root hash is specified, then ignore everything but the root id */ | |
1032 | if (!sd_id128_is_null(root_verity_uuid) && !sd_id128_equal(root_verity_uuid, id)) | |
4623e8e6 LP |
1033 | continue; |
1034 | ||
4623e8e6 | 1035 | fstype = "DM_verity_hash"; |
4623e8e6 | 1036 | rw = false; |
8ee9615e | 1037 | |
22e932f4 | 1038 | } else if (type.designator == PARTITION_ROOT_VERITY_SIG) { |
8ee9615e | 1039 | |
92e72028 ZJS |
1040 | check_partition_flags(node, pflags, |
1041 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY); | |
8ee9615e | 1042 | |
92e72028 | 1043 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
8ee9615e LP |
1044 | continue; |
1045 | ||
1046 | m->has_verity_sig = true; | |
1047 | ||
8ee9615e LP |
1048 | if (!verity) |
1049 | continue; | |
1050 | if (verity->designator >= 0 && verity->designator != PARTITION_ROOT) | |
1051 | continue; | |
8ee9615e | 1052 | |
8ee9615e | 1053 | fstype = "verity_hash_signature"; |
4623e8e6 | 1054 | rw = false; |
8ee9615e | 1055 | |
22e932f4 | 1056 | } else if (type.designator == PARTITION_USR) { |
aee36b4e | 1057 | |
92e72028 ZJS |
1058 | check_partition_flags(node, pflags, |
1059 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS); | |
aee36b4e | 1060 | |
92e72028 | 1061 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
aee36b4e LP |
1062 | continue; |
1063 | ||
1064 | /* If a usr ID is specified, ignore everything but the usr id */ | |
1065 | if (!sd_id128_is_null(usr_uuid) && !sd_id128_equal(usr_uuid, id)) | |
1066 | continue; | |
1067 | ||
92e72028 ZJS |
1068 | rw = !(pflags & SD_GPT_FLAG_READ_ONLY); |
1069 | growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS); | |
aee36b4e | 1070 | |
22e932f4 | 1071 | } else if (type.designator == PARTITION_USR_VERITY) { |
aee36b4e | 1072 | |
92e72028 ZJS |
1073 | check_partition_flags(node, pflags, |
1074 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY); | |
aee36b4e | 1075 | |
92e72028 | 1076 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
aee36b4e LP |
1077 | continue; |
1078 | ||
c3c88d67 | 1079 | m->has_verity = true; |
aee36b4e | 1080 | |
8ee9615e LP |
1081 | if (!verity) |
1082 | continue; | |
1083 | if (verity->designator >= 0 && verity->designator != PARTITION_USR) | |
1084 | continue; | |
1085 | ||
1086 | /* If usr hash is specified, then ignore everything but the usr id */ | |
1087 | if (!sd_id128_is_null(usr_verity_uuid) && !sd_id128_equal(usr_verity_uuid, id)) | |
aee36b4e LP |
1088 | continue; |
1089 | ||
aee36b4e | 1090 | fstype = "DM_verity_hash"; |
aee36b4e | 1091 | rw = false; |
8ee9615e | 1092 | |
22e932f4 | 1093 | } else if (type.designator == PARTITION_USR_VERITY_SIG) { |
8ee9615e | 1094 | |
92e72028 ZJS |
1095 | check_partition_flags(node, pflags, |
1096 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY); | |
8ee9615e | 1097 | |
92e72028 | 1098 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
8ee9615e LP |
1099 | continue; |
1100 | ||
1101 | m->has_verity_sig = true; | |
1102 | ||
8ee9615e LP |
1103 | if (!verity) |
1104 | continue; | |
1105 | if (verity->designator >= 0 && verity->designator != PARTITION_USR) | |
1106 | continue; | |
8ee9615e | 1107 | |
8ee9615e | 1108 | fstype = "verity_hash_signature"; |
aee36b4e | 1109 | rw = false; |
8ee9615e | 1110 | |
22e932f4 | 1111 | } else if (type.designator == PARTITION_SWAP) { |
a48dd347 | 1112 | |
92e72028 | 1113 | check_partition_flags(node, pflags, SD_GPT_FLAG_NO_AUTO); |
0f7c9a3d | 1114 | |
92e72028 | 1115 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
a48dd347 LP |
1116 | continue; |
1117 | ||
41aca66b LP |
1118 | /* Note: we don't set fstype = "swap" here, because we still need to probe if |
1119 | * it might be encrypted (i.e. fstype "crypt_LUKS") or unencrypted | |
1120 | * (i.e. fstype "swap"), and the only way to figure that out is via fstype | |
1121 | * probing. */ | |
df4524cb | 1122 | |
df655bf3 DDM |
1123 | /* We don't have a designator for SD_GPT_LINUX_GENERIC so check the UUID instead. */ |
1124 | } else if (sd_id128_equal(type.uuid, SD_GPT_LINUX_GENERIC)) { | |
8c1be37e | 1125 | |
92e72028 ZJS |
1126 | check_partition_flags(node, pflags, |
1127 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS); | |
0f7c9a3d | 1128 | |
92e72028 | 1129 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
a48dd347 LP |
1130 | continue; |
1131 | ||
8c1be37e LP |
1132 | if (generic_node) |
1133 | multiple_generic = true; | |
1134 | else { | |
1135 | generic_nr = nr; | |
92e72028 ZJS |
1136 | generic_rw = !(pflags & SD_GPT_FLAG_READ_ONLY); |
1137 | generic_growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS); | |
be30ad41 | 1138 | generic_uuid = id; |
a5590886 | 1139 | generic_node = TAKE_PTR(node); |
8c1be37e | 1140 | } |
d4dffb85 | 1141 | |
22e932f4 | 1142 | } else if (type.designator == PARTITION_VAR) { |
d4dffb85 | 1143 | |
92e72028 ZJS |
1144 | check_partition_flags(node, pflags, |
1145 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS); | |
0f7c9a3d | 1146 | |
92e72028 | 1147 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
d4dffb85 LP |
1148 | continue; |
1149 | ||
1150 | if (!FLAGS_SET(flags, DISSECT_IMAGE_RELAX_VAR_CHECK)) { | |
1151 | sd_id128_t var_uuid; | |
1152 | ||
1153 | /* For /var we insist that the uuid of the partition matches the | |
1154 | * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine | |
1155 | * ID. Why? Unlike the other partitions /var is inherently | |
1156 | * installation specific, hence we need to be careful not to mount it | |
1157 | * in the wrong installation. By hashing the partition UUID from | |
1158 | * /etc/machine-id we can securely bind the partition to the | |
1159 | * installation. */ | |
1160 | ||
92e72028 | 1161 | r = sd_id128_get_machine_app_specific(SD_GPT_VAR, &var_uuid); |
d4dffb85 LP |
1162 | if (r < 0) |
1163 | return r; | |
1164 | ||
1165 | if (!sd_id128_equal(var_uuid, id)) { | |
a52efa81 YW |
1166 | log_debug("Found a /var/ partition, but its UUID didn't match our expectations " |
1167 | "(found: " SD_ID128_UUID_FORMAT_STR ", expected: " SD_ID128_UUID_FORMAT_STR "), ignoring.", | |
1168 | SD_ID128_FORMAT_VAL(id), SD_ID128_FORMAT_VAL(var_uuid)); | |
d4dffb85 LP |
1169 | continue; |
1170 | } | |
1171 | } | |
1172 | ||
92e72028 ZJS |
1173 | rw = !(pflags & SD_GPT_FLAG_READ_ONLY); |
1174 | growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS); | |
8c1be37e LP |
1175 | } |
1176 | ||
22e932f4 | 1177 | if (type.designator != _PARTITION_DESIGNATOR_INVALID) { |
1a81ddef | 1178 | _cleanup_free_ char *t = NULL, *o = NULL, *l = NULL, *n = NULL; |
254d1313 | 1179 | _cleanup_close_ int mount_node_fd = -EBADF; |
18d73705 | 1180 | const char *options = NULL; |
8c1be37e | 1181 | |
cd22d856 LP |
1182 | r = image_policy_may_use(policy, type.designator); |
1183 | if (r < 0) | |
1184 | return r; | |
1185 | if (r == 0) { | |
1186 | /* Policy says: ignore; Remember this fact, so that we later can distinguish between "found but ignored" and "not found at all" */ | |
1187 | ||
1188 | if (!m->partitions[type.designator].found) | |
1189 | m->partitions[type.designator].ignored = true; | |
1190 | ||
1191 | continue; | |
1192 | } | |
1193 | ||
22e932f4 | 1194 | if (m->partitions[type.designator].found) { |
ca6cdd26 LP |
1195 | int c; |
1196 | ||
08fe0a53 LP |
1197 | /* For most partition types the first one we see wins. Except for the |
1198 | * rootfs and /usr, where we do a version compare of the label, and | |
1199 | * let the newest version win. This permits a simple A/B versioning | |
1200 | * scheme in OS images. */ | |
1201 | ||
ca6cdd26 LP |
1202 | c = compare_arch(type.arch, m->partitions[type.designator].architecture); |
1203 | if (c < 0) /* the arch we already found is better than the one we found now */ | |
22e932f4 | 1204 | continue; |
ca6cdd26 LP |
1205 | if (c == 0 && /* same arch? then go by version in label */ |
1206 | (!partition_designator_is_versioned(type.designator) || | |
1207 | strverscmp_improved(label, m->partitions[type.designator].label) <= 0)) | |
08fe0a53 LP |
1208 | continue; |
1209 | ||
22e932f4 | 1210 | dissected_partition_done(m->partitions + type.designator); |
08fe0a53 | 1211 | } |
8c1be37e | 1212 | |
73d88b80 | 1213 | if (FLAGS_SET(flags, DISSECT_IMAGE_PIN_PARTITION_DEVICES) && |
f52261a0 | 1214 | type.designator != PARTITION_SWAP) { |
f7725647 YW |
1215 | mount_node_fd = open_partition(node, /* is_partition = */ true, m->loop); |
1216 | if (mount_node_fd < 0) | |
1217 | return mount_node_fd; | |
1218 | } | |
1219 | ||
1a81ddef LP |
1220 | r = make_partition_devname(devname, diskseq, nr, flags, &n); |
1221 | if (r < 0) | |
1222 | return r; | |
1223 | ||
8c1be37e LP |
1224 | if (fstype) { |
1225 | t = strdup(fstype); | |
1226 | if (!t) | |
1227 | return -ENOMEM; | |
1228 | } | |
1229 | ||
08fe0a53 LP |
1230 | if (label) { |
1231 | l = strdup(label); | |
1232 | if (!l) | |
1233 | return -ENOMEM; | |
1234 | } | |
1235 | ||
22e932f4 | 1236 | options = mount_options_from_designator(mount_options, type.designator); |
18d73705 LB |
1237 | if (options) { |
1238 | o = strdup(options); | |
1239 | if (!o) | |
1240 | return -ENOMEM; | |
1241 | } | |
1242 | ||
22e932f4 | 1243 | m->partitions[type.designator] = (DissectedPartition) { |
8c1be37e LP |
1244 | .found = true, |
1245 | .partno = nr, | |
1246 | .rw = rw, | |
de98f631 | 1247 | .growfs = growfs, |
22e932f4 | 1248 | .architecture = type.arch, |
1a81ddef | 1249 | .node = TAKE_PTR(n), |
1cc6c93a | 1250 | .fstype = TAKE_PTR(t), |
08fe0a53 | 1251 | .label = TAKE_PTR(l), |
be30ad41 | 1252 | .uuid = id, |
18d73705 | 1253 | .mount_options = TAKE_PTR(o), |
f7725647 | 1254 | .mount_node_fd = TAKE_FD(mount_node_fd), |
88b3300f LP |
1255 | .offset = (uint64_t) start * 512, |
1256 | .size = (uint64_t) size * 512, | |
d90b03f8 | 1257 | .gpt_flags = pflags, |
8d9a1d59 | 1258 | .fsmount_fd = -EBADF, |
8c1be37e | 1259 | }; |
8c1be37e LP |
1260 | } |
1261 | ||
1262 | } else if (is_mbr) { | |
1263 | ||
a8c47660 | 1264 | switch (blkid_partition_get_type(pp)) { |
8c1be37e | 1265 | |
a8c47660 LP |
1266 | case 0x83: /* Linux partition */ |
1267 | ||
1268 | if (pflags != 0x80) /* Bootable flag */ | |
1269 | continue; | |
8c1be37e | 1270 | |
a8c47660 LP |
1271 | if (generic_node) |
1272 | multiple_generic = true; | |
1273 | else { | |
1274 | generic_nr = nr; | |
1275 | generic_rw = true; | |
de98f631 | 1276 | generic_growfs = false; |
a5590886 | 1277 | generic_node = TAKE_PTR(node); |
a8c47660 LP |
1278 | } |
1279 | ||
1280 | break; | |
1281 | ||
1282 | case 0xEA: { /* Boot Loader Spec extended $BOOT partition */ | |
254d1313 | 1283 | _cleanup_close_ int mount_node_fd = -EBADF; |
1a81ddef | 1284 | _cleanup_free_ char *o = NULL, *n = NULL; |
a8c47660 | 1285 | sd_id128_t id = SD_ID128_NULL; |
e3b9a5ff | 1286 | const char *options = NULL; |
a8c47660 | 1287 | |
cd22d856 LP |
1288 | r = image_policy_may_use(policy, PARTITION_XBOOTLDR); |
1289 | if (r < 0) | |
1290 | return r; | |
1291 | if (r == 0) { /* policy says: ignore */ | |
1292 | if (!m->partitions[PARTITION_XBOOTLDR].found) | |
1293 | m->partitions[PARTITION_XBOOTLDR].ignored = true; | |
1294 | ||
1295 | continue; | |
1296 | } | |
1297 | ||
a8c47660 | 1298 | /* First one wins */ |
234c2e16 | 1299 | if (m->partitions[PARTITION_XBOOTLDR].found) |
a8c47660 LP |
1300 | continue; |
1301 | ||
73d88b80 | 1302 | if (FLAGS_SET(flags, DISSECT_IMAGE_PIN_PARTITION_DEVICES)) { |
f7725647 YW |
1303 | mount_node_fd = open_partition(node, /* is_partition = */ true, m->loop); |
1304 | if (mount_node_fd < 0) | |
1305 | return mount_node_fd; | |
1306 | } | |
1307 | ||
e3b9a5ff | 1308 | (void) blkid_partition_get_uuid_id128(pp, &id); |
a8c47660 | 1309 | |
1a81ddef LP |
1310 | r = make_partition_devname(devname, diskseq, nr, flags, &n); |
1311 | if (r < 0) | |
1312 | return r; | |
1313 | ||
f5215bc8 | 1314 | options = mount_options_from_designator(mount_options, PARTITION_XBOOTLDR); |
18d73705 LB |
1315 | if (options) { |
1316 | o = strdup(options); | |
1317 | if (!o) | |
1318 | return -ENOMEM; | |
1319 | } | |
1320 | ||
a8c47660 LP |
1321 | m->partitions[PARTITION_XBOOTLDR] = (DissectedPartition) { |
1322 | .found = true, | |
1323 | .partno = nr, | |
1324 | .rw = true, | |
de98f631 | 1325 | .growfs = false, |
a8c47660 | 1326 | .architecture = _ARCHITECTURE_INVALID, |
1a81ddef | 1327 | .node = TAKE_PTR(n), |
a8c47660 | 1328 | .uuid = id, |
18d73705 | 1329 | .mount_options = TAKE_PTR(o), |
f7725647 | 1330 | .mount_node_fd = TAKE_FD(mount_node_fd), |
88b3300f LP |
1331 | .offset = (uint64_t) start * 512, |
1332 | .size = (uint64_t) size * 512, | |
8d9a1d59 | 1333 | .fsmount_fd = -EBADF, |
a8c47660 LP |
1334 | }; |
1335 | ||
1336 | break; | |
1337 | }} | |
8c1be37e LP |
1338 | } |
1339 | } | |
1340 | ||
22e932f4 DDM |
1341 | if (!m->partitions[PARTITION_ROOT].found && |
1342 | (m->partitions[PARTITION_ROOT_VERITY].found || | |
1343 | m->partitions[PARTITION_ROOT_VERITY_SIG].found)) | |
1344 | return -EADDRNOTAVAIL; /* Verity found but no matching rootfs? Something is off, refuse. */ | |
49ae9d91 | 1345 | |
8ee9615e LP |
1346 | /* Hmm, we found a signature partition but no Verity data? Something is off. */ |
1347 | if (m->partitions[PARTITION_ROOT_VERITY_SIG].found && !m->partitions[PARTITION_ROOT_VERITY].found) | |
1348 | return -EADDRNOTAVAIL; | |
7cf66030 | 1349 | |
22e932f4 DDM |
1350 | if (!m->partitions[PARTITION_USR].found && |
1351 | (m->partitions[PARTITION_USR_VERITY].found || | |
1352 | m->partitions[PARTITION_USR_VERITY_SIG].found)) | |
1353 | return -EADDRNOTAVAIL; /* as above */ | |
7cf66030 | 1354 | |
22e932f4 | 1355 | /* as above */ |
8ee9615e LP |
1356 | if (m->partitions[PARTITION_USR_VERITY_SIG].found && !m->partitions[PARTITION_USR_VERITY].found) |
1357 | return -EADDRNOTAVAIL; | |
1358 | ||
cb241a69 LP |
1359 | /* If root and /usr are combined then insist that the architecture matches */ |
1360 | if (m->partitions[PARTITION_ROOT].found && | |
1361 | m->partitions[PARTITION_USR].found && | |
1362 | (m->partitions[PARTITION_ROOT].architecture >= 0 && | |
1363 | m->partitions[PARTITION_USR].architecture >= 0 && | |
1364 | m->partitions[PARTITION_ROOT].architecture != m->partitions[PARTITION_USR].architecture)) | |
1365 | return -EADDRNOTAVAIL; | |
1366 | ||
4ab51780 LP |
1367 | if (!m->partitions[PARTITION_ROOT].found && |
1368 | !m->partitions[PARTITION_USR].found && | |
1369 | (flags & DISSECT_IMAGE_GENERIC_ROOT) && | |
00a8b34f | 1370 | (!verity || !verity->root_hash || verity->designator != PARTITION_USR)) { |
7cf66030 | 1371 | |
1b010ae7 | 1372 | /* OK, we found nothing usable, then check if there's a single generic partition, and use |
4b5de5dd LP |
1373 | * that. If the root hash was set however, then we won't fall back to a generic node, because |
1374 | * the root hash decides. */ | |
7cf66030 LP |
1375 | |
1376 | /* If we didn't find a properly marked root partition, but we did find a single suitable | |
1377 | * generic Linux partition, then use this as root partition, if the caller asked for it. */ | |
1378 | if (multiple_generic) | |
1379 | return -ENOTUNIQ; | |
1380 | ||
4b5de5dd LP |
1381 | /* If we didn't find a generic node, then we can't fix this up either */ |
1382 | if (generic_node) { | |
cd22d856 | 1383 | r = image_policy_may_use(policy, PARTITION_ROOT); |
1a81ddef LP |
1384 | if (r < 0) |
1385 | return r; | |
cd22d856 LP |
1386 | if (r == 0) |
1387 | /* Policy says: ignore; remember that we did */ | |
1388 | m->partitions[PARTITION_ROOT].ignored = true; | |
1389 | else { | |
1390 | _cleanup_close_ int mount_node_fd = -EBADF; | |
1391 | _cleanup_free_ char *o = NULL, *n = NULL; | |
1392 | const char *options; | |
1a81ddef | 1393 | |
cd22d856 LP |
1394 | if (FLAGS_SET(flags, DISSECT_IMAGE_PIN_PARTITION_DEVICES)) { |
1395 | mount_node_fd = open_partition(generic_node, /* is_partition = */ true, m->loop); | |
1396 | if (mount_node_fd < 0) | |
1397 | return mount_node_fd; | |
1398 | } | |
18d73705 | 1399 | |
cd22d856 LP |
1400 | r = make_partition_devname(devname, diskseq, generic_nr, flags, &n); |
1401 | if (r < 0) | |
1402 | return r; | |
1403 | ||
1404 | options = mount_options_from_designator(mount_options, PARTITION_ROOT); | |
1405 | if (options) { | |
1406 | o = strdup(options); | |
1407 | if (!o) | |
1408 | return -ENOMEM; | |
1409 | } | |
1410 | ||
1411 | assert(generic_nr >= 0); | |
1412 | m->partitions[PARTITION_ROOT] = (DissectedPartition) { | |
1413 | .found = true, | |
1414 | .rw = generic_rw, | |
1415 | .growfs = generic_growfs, | |
1416 | .partno = generic_nr, | |
1417 | .architecture = _ARCHITECTURE_INVALID, | |
1418 | .node = TAKE_PTR(n), | |
1419 | .uuid = generic_uuid, | |
1420 | .mount_options = TAKE_PTR(o), | |
1421 | .mount_node_fd = TAKE_FD(mount_node_fd), | |
1422 | .offset = UINT64_MAX, | |
1423 | .size = UINT64_MAX, | |
8d9a1d59 | 1424 | .fsmount_fd = -EBADF, |
cd22d856 LP |
1425 | }; |
1426 | } | |
e0f9e7bd | 1427 | } |
8c1be37e LP |
1428 | } |
1429 | ||
4b5de5dd LP |
1430 | /* 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 */ |
1431 | if (FLAGS_SET(flags, DISSECT_IMAGE_REQUIRE_ROOT) && | |
1432 | !(m->partitions[PARTITION_ROOT].found || (m->partitions[PARTITION_USR].found && FLAGS_SET(flags, DISSECT_IMAGE_USR_NO_ROOT)))) | |
1433 | return -ENXIO; | |
1434 | ||
7b32164f LP |
1435 | if (m->partitions[PARTITION_ROOT_VERITY].found) { |
1436 | /* We only support one verity partition per image, i.e. can't do for both /usr and root fs */ | |
1437 | if (m->partitions[PARTITION_USR_VERITY].found) | |
1438 | return -ENOTUNIQ; | |
1439 | ||
1440 | /* We don't support verity enabled root with a split out /usr. Neither with nor without | |
1441 | * verity there. (Note that we do support verity-less root with verity-full /usr, though.) */ | |
1442 | if (m->partitions[PARTITION_USR].found) | |
1443 | return -EADDRNOTAVAIL; | |
1444 | } | |
aee36b4e | 1445 | |
1903defc LP |
1446 | if (verity) { |
1447 | /* If a verity designator is specified, then insist that the matching partition exists */ | |
1448 | if (verity->designator >= 0 && !m->partitions[verity->designator].found) | |
1449 | return -EADDRNOTAVAIL; | |
aee36b4e | 1450 | |
01086e76 VD |
1451 | bool have_verity_sig_partition; |
1452 | if (verity->designator >= 0) | |
1453 | have_verity_sig_partition = m->partitions[verity->designator == PARTITION_USR ? PARTITION_USR_VERITY_SIG : PARTITION_ROOT_VERITY_SIG].found; | |
1454 | else | |
1455 | have_verity_sig_partition = m->partitions[PARTITION_USR_VERITY_SIG].found || m->partitions[PARTITION_ROOT_VERITY_SIG].found; | |
3dd73ea7 | 1456 | |
1903defc | 1457 | if (verity->root_hash) { |
8ee9615e LP |
1458 | /* If we have an explicit root hash and found the partitions for it, then we are ready to use |
1459 | * Verity, set things up for it */ | |
1460 | ||
1903defc LP |
1461 | if (verity->designator < 0 || verity->designator == PARTITION_ROOT) { |
1462 | if (!m->partitions[PARTITION_ROOT_VERITY].found || !m->partitions[PARTITION_ROOT].found) | |
1463 | return -EADDRNOTAVAIL; | |
4623e8e6 | 1464 | |
1903defc LP |
1465 | /* If we found a verity setup, then the root partition is necessarily read-only. */ |
1466 | m->partitions[PARTITION_ROOT].rw = false; | |
1467 | m->verity_ready = true; | |
1903defc | 1468 | |
f9e0bb61 LP |
1469 | } else { |
1470 | assert(verity->designator == PARTITION_USR); | |
1471 | ||
1903defc LP |
1472 | if (!m->partitions[PARTITION_USR_VERITY].found || !m->partitions[PARTITION_USR].found) |
1473 | return -EADDRNOTAVAIL; | |
4623e8e6 | 1474 | |
1903defc LP |
1475 | m->partitions[PARTITION_USR].rw = false; |
1476 | m->verity_ready = true; | |
1477 | } | |
8ee9615e LP |
1478 | |
1479 | if (m->verity_ready) | |
3dd73ea7 | 1480 | m->verity_sig_ready = verity->root_hash_sig || have_verity_sig_partition; |
8ee9615e | 1481 | |
3dd73ea7 | 1482 | } else if (have_verity_sig_partition) { |
8ee9615e LP |
1483 | |
1484 | /* If we found an embedded signature partition, we are ready, too. */ | |
1485 | ||
1486 | m->verity_ready = m->verity_sig_ready = true; | |
01086e76 VD |
1487 | if (verity->designator >= 0) |
1488 | m->partitions[verity->designator == PARTITION_USR ? PARTITION_USR : PARTITION_ROOT].rw = false; | |
1489 | else if (m->partitions[PARTITION_USR_VERITY_SIG].found) | |
1490 | m->partitions[PARTITION_USR].rw = false; | |
1491 | else if (m->partitions[PARTITION_ROOT_VERITY_SIG].found) | |
1492 | m->partitions[PARTITION_ROOT].rw = false; | |
aee36b4e | 1493 | } |
4623e8e6 LP |
1494 | } |
1495 | ||
598fd4da LP |
1496 | bool any = false; |
1497 | ||
cd22d856 LP |
1498 | /* After we discovered all partitions let's see if the verity requirements match the policy. (Note: |
1499 | * we don't check encryption requirements here, because we haven't probed the file system yet, hence | |
1500 | * don't know if this is encrypted or not) */ | |
1501 | for (PartitionDesignator di = 0; di < _PARTITION_DESIGNATOR_MAX; di++) { | |
1502 | PartitionDesignator vi, si; | |
1503 | PartitionPolicyFlags found_flags; | |
1504 | ||
598fd4da LP |
1505 | any = any || m->partitions[di].found; |
1506 | ||
cd22d856 LP |
1507 | vi = partition_verity_of(di); |
1508 | si = partition_verity_sig_of(di); | |
1509 | ||
1510 | /* Determine the verity protection level for this partition. */ | |
1511 | found_flags = m->partitions[di].found ? | |
1512 | (vi >= 0 && m->partitions[vi].found ? | |
1513 | (si >= 0 && m->partitions[si].found ? PARTITION_POLICY_SIGNED : PARTITION_POLICY_VERITY) : | |
1514 | PARTITION_POLICY_ENCRYPTED|PARTITION_POLICY_UNPROTECTED) : | |
1515 | (m->partitions[di].ignored ? PARTITION_POLICY_UNUSED : PARTITION_POLICY_ABSENT); | |
1516 | ||
1517 | r = image_policy_check_protection(policy, di, found_flags); | |
1518 | if (r < 0) | |
1519 | return r; | |
1520 | ||
1521 | if (m->partitions[di].found) { | |
1522 | r = image_policy_check_partition_flags(policy, di, m->partitions[di].gpt_flags); | |
1523 | if (r < 0) | |
1524 | return r; | |
1525 | } | |
1526 | } | |
1527 | ||
598fd4da LP |
1528 | if (!any && !FLAGS_SET(flags, DISSECT_IMAGE_ALLOW_EMPTY)) |
1529 | return -ENOMSG; | |
1530 | ||
cd22d856 | 1531 | r = dissected_image_probe_filesystems(m, fd, policy); |
c80c9079 LP |
1532 | if (r < 0) |
1533 | return r; | |
1534 | ||
08f14be4 YW |
1535 | return 0; |
1536 | } | |
1537 | #endif | |
18b5886e | 1538 | |
08f14be4 YW |
1539 | int dissect_image_file( |
1540 | const char *path, | |
1541 | const VeritySettings *verity, | |
1542 | const MountOptions *mount_options, | |
84be0c71 | 1543 | const ImagePolicy *image_policy, |
08f14be4 YW |
1544 | DissectImageFlags flags, |
1545 | DissectedImage **ret) { | |
8c1be37e | 1546 | |
08f14be4 YW |
1547 | #if HAVE_BLKID |
1548 | _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL; | |
254d1313 | 1549 | _cleanup_close_ int fd = -EBADF; |
51778dea | 1550 | struct stat st; |
08f14be4 | 1551 | int r; |
8c1be37e | 1552 | |
08f14be4 | 1553 | assert(path); |
8c1be37e | 1554 | |
08f14be4 YW |
1555 | fd = open(path, O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY); |
1556 | if (fd < 0) | |
1557 | return -errno; | |
896f937f | 1558 | |
51778dea LP |
1559 | if (fstat(fd, &st) < 0) |
1560 | return -errno; | |
1561 | ||
1562 | r = stat_verify_regular(&st); | |
08f14be4 YW |
1563 | if (r < 0) |
1564 | return r; | |
de98f631 | 1565 | |
08f14be4 YW |
1566 | r = dissected_image_new(path, &m); |
1567 | if (r < 0) | |
1568 | return r; | |
1569 | ||
51778dea LP |
1570 | m->image_size = st.st_size; |
1571 | ||
22ee78a8 LP |
1572 | r = probe_sector_size(fd, &m->sector_size); |
1573 | if (r < 0) | |
1574 | return r; | |
1575 | ||
84be0c71 | 1576 | r = dissect_image(m, fd, path, verity, mount_options, image_policy, flags); |
698bb074 YW |
1577 | if (r < 0) |
1578 | return r; | |
8c1be37e | 1579 | |
93a8a85b LP |
1580 | if (ret) |
1581 | *ret = TAKE_PTR(m); | |
8c1be37e LP |
1582 | return 0; |
1583 | #else | |
1584 | return -EOPNOTSUPP; | |
1585 | #endif | |
1586 | } | |
1587 | ||
4953e39c ZJS |
1588 | int dissect_log_error(int log_level, int r, const char *name, const VeritySettings *verity) { |
1589 | assert(log_level >= 0 && log_level <= LOG_DEBUG); | |
7cd7a195 LP |
1590 | assert(name); |
1591 | ||
1592 | switch (r) { | |
1593 | ||
1594 | case 0 ... INT_MAX: /* success! */ | |
1595 | return r; | |
1596 | ||
1597 | case -EOPNOTSUPP: | |
0214ead6 | 1598 | return log_full_errno(log_level, r, "Dissecting images is not supported, compiled without blkid support."); |
7cd7a195 LP |
1599 | |
1600 | case -ENOPKG: | |
0214ead6 | 1601 | return log_full_errno(log_level, r, "%s: Couldn't identify a suitable partition table or file system.", name); |
7cd7a195 LP |
1602 | |
1603 | case -ENOMEDIUM: | |
0214ead6 | 1604 | return log_full_errno(log_level, r, "%s: The image does not pass os-release/extension-release validation.", name); |
7cd7a195 LP |
1605 | |
1606 | case -EADDRNOTAVAIL: | |
0214ead6 | 1607 | return log_full_errno(log_level, r, "%s: No root partition for specified root hash found.", name); |
7cd7a195 LP |
1608 | |
1609 | case -ENOTUNIQ: | |
0214ead6 | 1610 | return log_full_errno(log_level, r, "%s: Multiple suitable root partitions found in image.", name); |
7cd7a195 LP |
1611 | |
1612 | case -ENXIO: | |
0214ead6 | 1613 | return log_full_errno(log_level, r, "%s: No suitable root partition found in image.", name); |
7cd7a195 LP |
1614 | |
1615 | case -EPROTONOSUPPORT: | |
0214ead6 | 1616 | return log_full_errno(log_level, r, "Device '%s' is a loopback block device with partition scanning turned off, please turn it on.", name); |
7cd7a195 LP |
1617 | |
1618 | case -ENOTBLK: | |
0214ead6 | 1619 | return log_full_errno(log_level, r, "%s: Image is not a block device.", name); |
7cd7a195 LP |
1620 | |
1621 | case -EBADR: | |
0214ead6 QH |
1622 | return log_full_errno(log_level, r, |
1623 | "Combining partitioned images (such as '%s') with external Verity data (such as '%s') not supported. " | |
1624 | "(Consider setting $SYSTEMD_DISSECT_VERITY_SIDECAR=0 to disable automatic discovery of external Verity data.)", | |
1625 | name, strna(verity ? verity->data_path : NULL)); | |
7cd7a195 LP |
1626 | |
1627 | case -ERFKILL: | |
0214ead6 | 1628 | return log_full_errno(log_level, r, "%s: image does not match image policy.", name); |
7cd7a195 | 1629 | |
598fd4da | 1630 | case -ENOMSG: |
0214ead6 | 1631 | return log_full_errno(log_level, r, "%s: no suitable partitions found.", name); |
598fd4da | 1632 | |
7cd7a195 | 1633 | default: |
0214ead6 | 1634 | return log_full_errno(log_level, r, "%s: cannot dissect image: %m", name); |
7cd7a195 LP |
1635 | } |
1636 | } | |
1637 | ||
1638 | int dissect_image_file_and_warn( | |
1639 | const char *path, | |
1640 | const VeritySettings *verity, | |
1641 | const MountOptions *mount_options, | |
1642 | const ImagePolicy *image_policy, | |
1643 | DissectImageFlags flags, | |
1644 | DissectedImage **ret) { | |
1645 | ||
1646 | return dissect_log_error( | |
4953e39c | 1647 | LOG_ERR, |
7cd7a195 LP |
1648 | dissect_image_file(path, verity, mount_options, image_policy, flags, ret), |
1649 | path, | |
1650 | verity); | |
1651 | } | |
1652 | ||
8c1be37e | 1653 | DissectedImage* dissected_image_unref(DissectedImage *m) { |
8c1be37e LP |
1654 | if (!m) |
1655 | return NULL; | |
1656 | ||
ac1e1b5f | 1657 | /* First, clear dissected partitions. */ |
08fe0a53 | 1658 | for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) |
234c2e16 | 1659 | dissected_partition_done(m->partitions + i); |
8c1be37e | 1660 | |
ac1e1b5f YW |
1661 | /* Second, free decrypted images. This must be after dissected_partition_done(), as freeing |
1662 | * DecryptedImage may try to deactivate partitions. */ | |
1663 | decrypted_image_unref(m->decrypted_image); | |
1664 | ||
1e63dc4f YW |
1665 | /* Third, unref LoopDevice. This must be called after the above two, as freeing LoopDevice may try to |
1666 | * remove existing partitions on the loopback block device. */ | |
1667 | loop_device_unref(m->loop); | |
1668 | ||
593fe6c0 | 1669 | free(m->image_name); |
3b925504 LP |
1670 | free(m->hostname); |
1671 | strv_free(m->machine_info); | |
1672 | strv_free(m->os_release); | |
fab22946 | 1673 | strv_free(m->initrd_release); |
a81fe93e LP |
1674 | strv_free(m->confext_release); |
1675 | strv_free(m->sysext_release); | |
3b925504 | 1676 | |
5fecf46d | 1677 | return mfree(m); |
8c1be37e LP |
1678 | } |
1679 | ||
18b5886e | 1680 | static int is_loop_device(const char *path) { |
553e15f2 | 1681 | char s[SYS_BLOCK_PATH_MAX("/../loop/")]; |
18b5886e LP |
1682 | struct stat st; |
1683 | ||
1684 | assert(path); | |
1685 | ||
1686 | if (stat(path, &st) < 0) | |
1687 | return -errno; | |
1688 | ||
1689 | if (!S_ISBLK(st.st_mode)) | |
1690 | return -ENOTBLK; | |
1691 | ||
553e15f2 | 1692 | xsprintf_sys_block_path(s, "/loop/", st.st_dev); |
18b5886e LP |
1693 | if (access(s, F_OK) < 0) { |
1694 | if (errno != ENOENT) | |
1695 | return -errno; | |
1696 | ||
1697 | /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */ | |
553e15f2 | 1698 | xsprintf_sys_block_path(s, "/../loop/", st.st_dev); |
18b5886e LP |
1699 | if (access(s, F_OK) < 0) |
1700 | return errno == ENOENT ? false : -errno; | |
1701 | } | |
1702 | ||
1703 | return true; | |
1704 | } | |
1705 | ||
f8ab7812 | 1706 | static int run_fsck(int node_fd, const char *fstype) { |
cf32c486 LP |
1707 | int r, exit_status; |
1708 | pid_t pid; | |
1709 | ||
f8ab7812 | 1710 | assert(node_fd >= 0); |
cf32c486 LP |
1711 | assert(fstype); |
1712 | ||
13556724 | 1713 | r = fsck_exists_for_fstype(fstype); |
cf32c486 LP |
1714 | if (r < 0) { |
1715 | log_debug_errno(r, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype); | |
1716 | return 0; | |
1717 | } | |
1718 | if (r == 0) { | |
f8ab7812 | 1719 | log_debug("Not checking partition %s, as fsck for %s does not exist.", FORMAT_PROC_FD_PATH(node_fd), fstype); |
cf32c486 LP |
1720 | return 0; |
1721 | } | |
1722 | ||
f8ab7812 LP |
1723 | r = safe_fork_full( |
1724 | "(fsck)", | |
911f8f01 | 1725 | NULL, |
f8ab7812 | 1726 | &node_fd, 1, /* Leave the node fd open */ |
e9ccae31 | 1727 | FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG_SIGTERM|FORK_REARRANGE_STDIO|FORK_CLOEXEC_OFF, |
f8ab7812 | 1728 | &pid); |
cf32c486 LP |
1729 | if (r < 0) |
1730 | return log_debug_errno(r, "Failed to fork off fsck: %m"); | |
1731 | if (r == 0) { | |
1732 | /* Child */ | |
360c9cdc | 1733 | execlp("fsck", "fsck", "-aT", FORMAT_PROC_FD_PATH(node_fd), NULL); |
7e0ed2e9 | 1734 | log_open(); |
cf32c486 LP |
1735 | log_debug_errno(errno, "Failed to execl() fsck: %m"); |
1736 | _exit(FSCK_OPERATIONAL_ERROR); | |
1737 | } | |
1738 | ||
1739 | exit_status = wait_for_terminate_and_check("fsck", pid, 0); | |
1740 | if (exit_status < 0) | |
360c9cdc | 1741 | return log_debug_errno(exit_status, "Failed to fork off fsck: %m"); |
cf32c486 LP |
1742 | |
1743 | if ((exit_status & ~FSCK_ERROR_CORRECTED) != FSCK_SUCCESS) { | |
1744 | log_debug("fsck failed with exit status %i.", exit_status); | |
1745 | ||
1746 | if ((exit_status & (FSCK_SYSTEM_SHOULD_REBOOT|FSCK_ERRORS_LEFT_UNCORRECTED)) != 0) | |
1747 | return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN), "File system is corrupted, refusing."); | |
1748 | ||
1749 | log_debug("Ignoring fsck error."); | |
1750 | } | |
1751 | ||
1752 | return 0; | |
1753 | } | |
1754 | ||
8d9a1d59 LP |
1755 | static int fs_grow(const char *node_path, int mount_fd, const char *mount_path) { |
1756 | _cleanup_close_ int _mount_fd = -EBADF, node_fd = -EBADF; | |
81939d9d | 1757 | uint64_t size, newsize; |
8d9a1d59 | 1758 | const char *id; |
81939d9d LP |
1759 | int r; |
1760 | ||
8d9a1d59 LP |
1761 | assert(node_path); |
1762 | assert(mount_fd >= 0 || mount_path); | |
1763 | ||
81939d9d LP |
1764 | node_fd = open(node_path, O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY); |
1765 | if (node_fd < 0) | |
1766 | return log_debug_errno(errno, "Failed to open node device %s: %m", node_path); | |
1767 | ||
01db9c85 LP |
1768 | r = blockdev_get_device_size(node_fd, &size); |
1769 | if (r < 0) | |
1770 | return log_debug_errno(r, "Failed to get block device size of %s: %m", node_path); | |
81939d9d | 1771 | |
8d9a1d59 LP |
1772 | if (mount_fd < 0) { |
1773 | assert(mount_path); | |
1774 | ||
1775 | _mount_fd = open(mount_path, O_RDONLY|O_DIRECTORY|O_CLOEXEC); | |
1776 | if (_mount_fd < 0) | |
1777 | return log_debug_errno(errno, "Failed to open mounted file system %s: %m", mount_path); | |
1778 | ||
1779 | mount_fd = _mount_fd; | |
1780 | } else { | |
1781 | mount_fd = fd_reopen_condition(mount_fd, O_RDONLY|O_DIRECTORY|O_CLOEXEC, O_RDONLY|O_DIRECTORY|O_CLOEXEC, &_mount_fd); | |
1782 | if (mount_fd < 0) | |
1783 | return log_debug_errno(errno, "Failed to reopen mount node: %m"); | |
1784 | } | |
1785 | ||
1786 | id = mount_path ?: node_path; | |
81939d9d | 1787 | |
8d9a1d59 | 1788 | log_debug("Resizing \"%s\" to %"PRIu64" bytes...", id, size); |
81939d9d LP |
1789 | r = resize_fs(mount_fd, size, &newsize); |
1790 | if (r < 0) | |
8d9a1d59 | 1791 | return log_debug_errno(r, "Failed to resize \"%s\" to %"PRIu64" bytes: %m", id, size); |
81939d9d LP |
1792 | |
1793 | if (newsize == size) | |
1794 | log_debug("Successfully resized \"%s\" to %s bytes.", | |
8d9a1d59 | 1795 | id, FORMAT_BYTES(newsize)); |
81939d9d LP |
1796 | else { |
1797 | assert(newsize < size); | |
1798 | log_debug("Successfully resized \"%s\" to %s bytes (%"PRIu64" bytes lost due to blocksize).", | |
8d9a1d59 | 1799 | id, FORMAT_BYTES(newsize), size - newsize); |
81939d9d LP |
1800 | } |
1801 | ||
1802 | return 0; | |
1803 | } | |
1804 | ||
254e392e LP |
1805 | int partition_pick_mount_options( |
1806 | PartitionDesignator d, | |
1807 | const char *fstype, | |
1808 | bool rw, | |
1809 | bool discard, | |
1810 | char **ret_options, | |
1811 | unsigned long *ret_ms_flags) { | |
1812 | ||
1813 | _cleanup_free_ char *options = NULL; | |
1814 | ||
1815 | assert(ret_options); | |
1816 | ||
1817 | /* Selects a baseline of bind mount flags, that should always apply. | |
1818 | * | |
1819 | * Firstly, we set MS_NODEV universally on all mounts, since we don't want to allow device nodes outside of /dev/. | |
1820 | * | |
1821 | * On /var/tmp/ we'll also set MS_NOSUID, same as we set for /tmp/ on the host. | |
1822 | * | |
1823 | * On the ESP and XBOOTLDR partitions we'll also disable symlinks, and execution. These file systems | |
1824 | * are generally untrusted (i.e. not encrypted or authenticated), and typically VFAT hence we should | |
1825 | * be as restrictive as possible, and this shouldn't hurt, since the functionality is not available | |
1826 | * there anyway. */ | |
1827 | ||
1828 | unsigned long flags = MS_NODEV; | |
1829 | ||
1830 | if (!rw) | |
1831 | flags |= MS_RDONLY; | |
1832 | ||
1833 | switch (d) { | |
1834 | ||
1835 | case PARTITION_ESP: | |
1836 | case PARTITION_XBOOTLDR: | |
1837 | flags |= MS_NOSUID|MS_NOEXEC|ms_nosymfollow_supported(); | |
1838 | ||
6eda6f7e LP |
1839 | /* The ESP might contain a pre-boot random seed. Let's make this unaccessible to regular |
1840 | * userspace. ESP/XBOOTLDR is almost certainly VFAT, hence if we don't know assume it is. */ | |
1841 | if (!fstype || fstype_can_umask(fstype)) | |
254e392e LP |
1842 | if (!strextend_with_separator(&options, ",", "umask=0077")) |
1843 | return -ENOMEM; | |
1844 | break; | |
1845 | ||
1846 | case PARTITION_TMP: | |
1847 | flags |= MS_NOSUID; | |
1848 | break; | |
1849 | ||
1850 | default: | |
1851 | break; | |
1852 | } | |
1853 | ||
1854 | /* So, when you request MS_RDONLY from ext4, then this means nothing. It happily still writes to the | |
1855 | * backing storage. What's worse, the BLKRO[GS]ET flag and (in case of loopback devices) | |
1856 | * LO_FLAGS_READ_ONLY don't mean anything, they affect userspace accesses only, and write accesses | |
1857 | * from the upper file system still get propagated through to the underlying file system, | |
1858 | * unrestricted. To actually get ext4/xfs/btrfs to stop writing to the device we need to specify | |
1859 | * "norecovery" as mount option, in addition to MS_RDONLY. Yes, this sucks, since it means we need to | |
1860 | * carry a per file system table here. | |
1861 | * | |
1862 | * Note that this means that we might not be able to mount corrupted file systems as read-only | |
1863 | * anymore (since in some cases the kernel implementations will refuse mounting when corrupted, | |
1864 | * read-only and "norecovery" is specified). But I think for the case of automatically determined | |
1865 | * mount options for loopback devices this is the right choice, since otherwise using the same | |
7227dd81 | 1866 | * loopback file twice even in read-only mode, is going to fail badly sooner or later. The use case of |
254e392e LP |
1867 | * making reuse of the immutable images "just work" is more relevant to us than having read-only |
1868 | * access that actually modifies stuff work on such image files. Or to say this differently: if | |
1869 | * people want their file systems to be fixed up they should just open them in writable mode, where | |
1870 | * all these problems don't exist. */ | |
034ebc47 | 1871 | if (!rw && fstype && fstype_can_norecovery(fstype)) |
254e392e LP |
1872 | if (!strextend_with_separator(&options, ",", "norecovery")) |
1873 | return -ENOMEM; | |
1874 | ||
1875 | if (discard && fstype && fstype_can_discard(fstype)) | |
1876 | if (!strextend_with_separator(&options, ",", "discard")) | |
1877 | return -ENOMEM; | |
1878 | ||
1879 | if (!ret_ms_flags) /* Fold flags into option string if ret_flags specified as NULL */ | |
1880 | if (!strextend_with_separator(&options, ",", | |
1881 | FLAGS_SET(flags, MS_RDONLY) ? "ro" : "rw", | |
1882 | FLAGS_SET(flags, MS_NODEV) ? "nodev" : "dev", | |
1883 | FLAGS_SET(flags, MS_NOSUID) ? "nosuid" : "suid", | |
1884 | FLAGS_SET(flags, MS_NOEXEC) ? "noexec" : "exec", | |
1885 | FLAGS_SET(flags, MS_NOSYMFOLLOW) ? "nosymfollow" : NULL)) | |
1886 | /* NB: we suppress 'symfollow' here, since it's the default, and old /bin/mount might not know it */ | |
1887 | return -ENOMEM; | |
1888 | ||
1889 | if (ret_ms_flags) | |
1890 | *ret_ms_flags = flags; | |
1891 | ||
1892 | *ret_options = TAKE_PTR(options); | |
1893 | return 0; | |
1894 | } | |
1895 | ||
8d9a1d59 LP |
1896 | static bool need_user_mapping(uid_t uid_shift, uid_t uid_range) { |
1897 | ||
1898 | if (!uid_is_valid(uid_shift)) | |
1899 | return false; | |
1900 | ||
1901 | return uid_shift != 0 || uid_range != UINT32_MAX; | |
1902 | } | |
1903 | ||
18b5886e | 1904 | static int mount_partition( |
254e392e | 1905 | PartitionDesignator d, |
18b5886e LP |
1906 | DissectedPartition *m, |
1907 | const char *where, | |
1908 | const char *directory, | |
2d3a5a73 | 1909 | uid_t uid_shift, |
21b61b1d | 1910 | uid_t uid_range, |
8d9a1d59 | 1911 | int userns_fd, |
18b5886e LP |
1912 | DissectImageFlags flags) { |
1913 | ||
2d3a5a73 | 1914 | _cleanup_free_ char *chased = NULL, *options = NULL; |
8d9a1d59 LP |
1915 | const char *p = NULL, *node, *fstype = NULL; |
1916 | bool rw, discard, grow; | |
254e392e | 1917 | unsigned long ms_flags; |
2eedfd2d | 1918 | int r; |
8c1be37e LP |
1919 | |
1920 | assert(m); | |
8c1be37e | 1921 | |
8d9a1d59 | 1922 | if (!m->found) |
f7725647 YW |
1923 | return 0; |
1924 | ||
8d9a1d59 LP |
1925 | /* Check the various combinations when we can't do anything anymore */ |
1926 | if (m->fsmount_fd < 0 && m->mount_node_fd < 0) | |
1927 | return 0; | |
1928 | if (m->fsmount_fd >= 0 && !where) | |
1929 | return 0; | |
1930 | if (!where && m->mount_node_fd < 0) | |
1931 | return 0; | |
18b5886e | 1932 | |
8d9a1d59 LP |
1933 | if (m->fsmount_fd < 0) { |
1934 | fstype = dissected_partition_fstype(m); | |
1935 | if (!fstype) | |
1936 | return -EAFNOSUPPORT; | |
8c1be37e | 1937 | |
8d9a1d59 LP |
1938 | /* We are looking at an encrypted partition? This either means stacked encryption, or the |
1939 | * caller didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error | |
1940 | * for this case. */ | |
1941 | if (streq(fstype, "crypto_LUKS")) | |
1942 | return -EUNATCH; | |
18b5886e | 1943 | |
8d9a1d59 LP |
1944 | r = dissect_fstype_ok(fstype); |
1945 | if (r < 0) | |
1946 | return r; | |
1947 | if (!r) | |
1948 | return -EIDRM; /* Recognizable error */ | |
1949 | } | |
14ce2467 | 1950 | |
8d9a1d59 | 1951 | node = m->mount_node_fd < 0 ? NULL : FORMAT_PROC_FD_PATH(m->mount_node_fd); |
ef9c184d | 1952 | rw = m->rw && !(flags & DISSECT_IMAGE_MOUNT_READ_ONLY); |
8c1be37e | 1953 | |
254e392e | 1954 | discard = ((flags & DISSECT_IMAGE_DISCARD) || |
8d9a1d59 | 1955 | ((flags & DISSECT_IMAGE_DISCARD_ON_LOOP) && (m->node && is_loop_device(m->node) > 0))); |
254e392e | 1956 | |
8d9a1d59 LP |
1957 | grow = rw && m->growfs && FLAGS_SET(flags, DISSECT_IMAGE_GROWFS); |
1958 | ||
1959 | if (FLAGS_SET(flags, DISSECT_IMAGE_FSCK) && rw && m->mount_node_fd >= 0 && m->fsmount_fd < 0) { | |
f8ab7812 | 1960 | r = run_fsck(m->mount_node_fd, fstype); |
cf32c486 LP |
1961 | if (r < 0) |
1962 | return r; | |
1963 | } | |
1964 | ||
8d9a1d59 LP |
1965 | if (where) { |
1966 | if (directory) { | |
1967 | /* Automatically create missing mount points inside the image, if necessary. */ | |
1968 | r = mkdir_p_root(where, directory, uid_shift, (gid_t) uid_shift, 0755, NULL); | |
1969 | if (r < 0 && r != -EROFS) | |
1970 | return r; | |
2eedfd2d | 1971 | |
8d9a1d59 | 1972 | r = chase(directory, where, CHASE_PREFIX_ROOT, &chased, NULL); |
9842905e LP |
1973 | if (r < 0) |
1974 | return r; | |
9842905e | 1975 | |
8d9a1d59 LP |
1976 | p = chased; |
1977 | } else { | |
1978 | /* Create top-level mount if missing – but only if this is asked for. This won't modify the | |
1979 | * image (as the branch above does) but the host hierarchy, and the created directory might | |
1980 | * survive our mount in the host hierarchy hence. */ | |
1981 | if (FLAGS_SET(flags, DISSECT_IMAGE_MKDIR)) { | |
1982 | r = mkdir_p(where, 0755); | |
1983 | if (r < 0) | |
1984 | return r; | |
1985 | } | |
8c1be37e | 1986 | |
8d9a1d59 LP |
1987 | p = where; |
1988 | } | |
1989 | } | |
2d3a5a73 | 1990 | |
8d9a1d59 LP |
1991 | if (m->fsmount_fd < 0) { |
1992 | r = partition_pick_mount_options(d, fstype, rw, discard, &options, &ms_flags); | |
1993 | if (r < 0) | |
1994 | return r; | |
2d3a5a73 | 1995 | |
8d9a1d59 | 1996 | if (need_user_mapping(uid_shift, uid_range) && fstype_can_uid_gid(fstype)) { |
21b61b1d | 1997 | _cleanup_free_ char *uid_option = NULL; |
2d3a5a73 | 1998 | |
21b61b1d LP |
1999 | if (asprintf(&uid_option, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0) |
2000 | return -ENOMEM; | |
2001 | ||
2002 | if (!strextend_with_separator(&options, ",", uid_option)) | |
2003 | return -ENOMEM; | |
8d9a1d59 LP |
2004 | |
2005 | userns_fd = -EBADF; /* Not needed */ | |
2006 | } | |
2007 | ||
2008 | if (!isempty(m->mount_options)) | |
2009 | if (!strextend_with_separator(&options, ",", m->mount_options)) | |
2010 | return -ENOMEM; | |
2d3a5a73 | 2011 | } |
8c1be37e | 2012 | |
8d9a1d59 LP |
2013 | if (p) { |
2014 | if (m->fsmount_fd >= 0) { | |
2015 | /* Case #1: Attach existing fsmount fd to the file system */ | |
18d73705 | 2016 | |
7c83d42e LB |
2017 | r = mount_exchange_graceful( |
2018 | m->fsmount_fd, | |
2019 | p, | |
2020 | FLAGS_SET(flags, DISSECT_IMAGE_TRY_ATOMIC_MOUNT_EXCHANGE)); | |
2021 | if (r < 0) | |
2022 | return log_debug_errno(r, "Failed to mount image on '%s': %m", p); | |
8d9a1d59 LP |
2023 | |
2024 | } else { | |
2025 | assert(node); | |
d9223c07 | 2026 | |
8d9a1d59 LP |
2027 | /* Case #2: Mount directly into place */ |
2028 | r = mount_nofollow_verbose(LOG_DEBUG, node, p, fstype, ms_flags, options); | |
2029 | if (r < 0) | |
2030 | return r; | |
81939d9d | 2031 | |
8d9a1d59 LP |
2032 | if (grow) |
2033 | (void) fs_grow(node, -EBADF, p); | |
2034 | ||
2035 | if (userns_fd >= 0) { | |
dba4fa89 | 2036 | r = remount_idmap_fd(STRV_MAKE(p), userns_fd); |
8d9a1d59 LP |
2037 | if (r < 0) |
2038 | return r; | |
2039 | } | |
2040 | } | |
2041 | } else { | |
2042 | assert(node); | |
2043 | ||
2044 | /* Case #3: Create fsmount fd */ | |
2045 | ||
2046 | m->fsmount_fd = make_fsmount(LOG_DEBUG, node, fstype, ms_flags, options, userns_fd); | |
2047 | if (m->fsmount_fd < 0) | |
2048 | return m->fsmount_fd; | |
2049 | ||
2050 | if (grow) | |
2051 | (void) fs_grow(node, m->fsmount_fd, NULL); | |
21b61b1d LP |
2052 | } |
2053 | ||
d9223c07 | 2054 | return 1; |
8c1be37e LP |
2055 | } |
2056 | ||
8d9a1d59 | 2057 | static int mount_root_tmpfs(const char *where, uid_t uid_shift, uid_t uid_range, DissectImageFlags flags) { |
7cf66030 LP |
2058 | _cleanup_free_ char *options = NULL; |
2059 | int r; | |
2060 | ||
2061 | assert(where); | |
2062 | ||
2063 | /* For images that contain /usr/ but no rootfs, let's mount rootfs as tmpfs */ | |
2064 | ||
2065 | if (FLAGS_SET(flags, DISSECT_IMAGE_MKDIR)) { | |
2066 | r = mkdir_p(where, 0755); | |
2067 | if (r < 0) | |
2068 | return r; | |
2069 | } | |
2070 | ||
8d9a1d59 | 2071 | if (need_user_mapping(uid_shift, uid_range)) { |
7cf66030 LP |
2072 | if (asprintf(&options, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0) |
2073 | return -ENOMEM; | |
2074 | } | |
2075 | ||
2076 | r = mount_nofollow_verbose(LOG_DEBUG, "rootfs", where, "tmpfs", MS_NODEV, options); | |
2077 | if (r < 0) | |
2078 | return r; | |
2079 | ||
2080 | return 1; | |
2081 | } | |
2082 | ||
1d96dae7 ZJS |
2083 | static int mount_point_is_available(const char *where, const char *path, bool missing_ok) { |
2084 | _cleanup_free_ char *p = NULL; | |
2085 | int r; | |
2086 | ||
2087 | /* Check whether <path> is suitable as a mountpoint, i.e. is an empty directory | |
2088 | * or does not exist at all (when missing_ok). */ | |
2089 | ||
2090 | r = chase(path, where, CHASE_PREFIX_ROOT, &p, NULL); | |
2091 | if (r == -ENOENT) | |
2092 | return missing_ok; | |
2093 | if (r < 0) | |
2094 | return log_debug_errno(r, "Failed to chase \"%s\": %m", path); | |
2095 | ||
2096 | r = dir_is_empty(p, /* ignore_hidden_or_backup= */ false); | |
2097 | if (r == -ENOTDIR) | |
2098 | return false; | |
2099 | if (r < 0) | |
2100 | return log_debug_errno(r, "Failed to check directory \"%s\": %m", p); | |
e36c6210 | 2101 | return r > 0; |
1d96dae7 ZJS |
2102 | } |
2103 | ||
21b61b1d LP |
2104 | int dissected_image_mount( |
2105 | DissectedImage *m, | |
2106 | const char *where, | |
2107 | uid_t uid_shift, | |
2108 | uid_t uid_range, | |
8d9a1d59 | 2109 | int userns_fd, |
21b61b1d LP |
2110 | DissectImageFlags flags) { |
2111 | ||
8d9a1d59 | 2112 | _cleanup_close_ int my_userns_fd = -EBADF; |
1d96dae7 | 2113 | int r; |
8c1be37e LP |
2114 | |
2115 | assert(m); | |
8c1be37e | 2116 | |
8d9a1d59 LP |
2117 | /* If 'where' is NULL then we'll use the new mount API to create fsmount() fds for the mounts and |
2118 | * store them in DissectedPartition.fsmount_fd. | |
2119 | * | |
2120 | * If 'where' is not NULL then we'll either mount the partitions to the right places ourselves, | |
2121 | * or use DissectedPartition.fsmount_fd and bind it to the right places. | |
2122 | * | |
2123 | * This allows splitting the setting up up the superblocks and the binding to file systems paths into | |
2124 | * two distinct and differently privileged components: one that gets the fsmount fds, and the other | |
2125 | * that then applies them. | |
2126 | * | |
2127 | * Returns: | |
fa45d12c LP |
2128 | * |
2129 | * -ENXIO → No root partition found | |
7718ac97 | 2130 | * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release/extension-release file found |
fa45d12c LP |
2131 | * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet |
2132 | * -EUCLEAN → fsck for file system failed | |
2133 | * -EBUSY → File system already mounted/used elsewhere (kernel) | |
4dc28665 | 2134 | * -EAFNOSUPPORT → File system type not supported or not known |
80ce8580 | 2135 | * -EIDRM → File system is not among allowlisted "common" file systems |
fa45d12c LP |
2136 | */ |
2137 | ||
8d9a1d59 LP |
2138 | if (!where && (flags & (DISSECT_IMAGE_VALIDATE_OS|DISSECT_IMAGE_VALIDATE_OS_EXT)) != 0) |
2139 | return -EOPNOTSUPP; /* for now, not supported */ | |
2140 | ||
7cf66030 LP |
2141 | if (!(m->partitions[PARTITION_ROOT].found || |
2142 | (m->partitions[PARTITION_USR].found && FLAGS_SET(flags, DISSECT_IMAGE_USR_NO_ROOT)))) | |
2143 | return -ENXIO; /* Require a root fs or at least a /usr/ fs (the latter is subject to a flag of its own) */ | |
8c1be37e | 2144 | |
8d9a1d59 LP |
2145 | if (userns_fd < 0 && need_user_mapping(uid_shift, uid_range) && FLAGS_SET(flags, DISSECT_IMAGE_MOUNT_IDMAPPED)) { |
2146 | ||
2147 | my_userns_fd = make_userns(uid_shift, uid_range, UID_INVALID, REMOUNT_IDMAPPING_HOST_ROOT); | |
2148 | if (my_userns_fd < 0) | |
2149 | return my_userns_fd; | |
2150 | ||
2151 | userns_fd = my_userns_fd; | |
2152 | } | |
2153 | ||
2d3a5a73 | 2154 | if ((flags & DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY) == 0) { |
7cf66030 LP |
2155 | |
2156 | /* First mount the root fs. If there's none we use a tmpfs. */ | |
8d9a1d59 LP |
2157 | if (m->partitions[PARTITION_ROOT].found) { |
2158 | r = mount_partition(PARTITION_ROOT, m->partitions + PARTITION_ROOT, where, NULL, uid_shift, uid_range, userns_fd, flags); | |
2159 | if (r < 0) | |
2160 | return r; | |
2161 | ||
2162 | } else if (where) { | |
2163 | r = mount_root_tmpfs(where, uid_shift, uid_range, flags); | |
2164 | if (r < 0) | |
2165 | return r; | |
2166 | } | |
aee36b4e | 2167 | |
aee36b4e | 2168 | /* For us mounting root always means mounting /usr as well */ |
8d9a1d59 | 2169 | r = mount_partition(PARTITION_USR, m->partitions + PARTITION_USR, where, "/usr", uid_shift, uid_range, userns_fd, flags); |
aee36b4e LP |
2170 | if (r < 0) |
2171 | return r; | |
8d9a1d59 | 2172 | } |
03bcb6d4 | 2173 | |
8d9a1d59 LP |
2174 | if ((flags & DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY) == 0 && |
2175 | (flags & (DISSECT_IMAGE_VALIDATE_OS|DISSECT_IMAGE_VALIDATE_OS_EXT)) != 0) { | |
2176 | /* If either one of the validation flags are set, ensure that the image qualifies as | |
2177 | * one or the other (or both). */ | |
2178 | bool ok = false; | |
9ccb531a | 2179 | |
8d9a1d59 LP |
2180 | assert(where); |
2181 | ||
2182 | if (FLAGS_SET(flags, DISSECT_IMAGE_VALIDATE_OS)) { | |
2183 | r = path_is_os_tree(where); | |
2184 | if (r < 0) | |
2185 | return r; | |
2186 | if (r > 0) | |
2187 | ok = true; | |
2188 | } | |
a8e8bcfb | 2189 | if (!ok && FLAGS_SET(flags, DISSECT_IMAGE_VALIDATE_OS_EXT) && m->image_name) { |
8d9a1d59 LP |
2190 | r = extension_has_forbidden_content(where); |
2191 | if (r < 0) | |
2192 | return r; | |
2193 | if (r == 0) { | |
2194 | r = path_is_extension_tree(IMAGE_SYSEXT, where, m->image_name, FLAGS_SET(flags, DISSECT_IMAGE_RELAX_EXTENSION_CHECK)); | |
2195 | if (r == 0) | |
2196 | r = path_is_extension_tree(IMAGE_CONFEXT, where, m->image_name, FLAGS_SET(flags, DISSECT_IMAGE_RELAX_EXTENSION_CHECK)); | |
9ccb531a LB |
2197 | if (r < 0) |
2198 | return r; | |
2199 | if (r > 0) | |
2200 | ok = true; | |
2201 | } | |
03bcb6d4 | 2202 | } |
8d9a1d59 LP |
2203 | |
2204 | if (!ok) | |
2205 | return -ENOMEDIUM; | |
2d3a5a73 LP |
2206 | } |
2207 | ||
705727fd | 2208 | if (flags & DISSECT_IMAGE_MOUNT_ROOT_ONLY) |
2d3a5a73 | 2209 | return 0; |
8c1be37e | 2210 | |
8d9a1d59 | 2211 | r = mount_partition(PARTITION_HOME, m->partitions + PARTITION_HOME, where, "/home", uid_shift, uid_range, userns_fd, flags); |
8c1be37e LP |
2212 | if (r < 0) |
2213 | return r; | |
2214 | ||
8d9a1d59 | 2215 | r = mount_partition(PARTITION_SRV, m->partitions + PARTITION_SRV, where, "/srv", uid_shift, uid_range, userns_fd, flags); |
8c1be37e LP |
2216 | if (r < 0) |
2217 | return r; | |
2218 | ||
8d9a1d59 | 2219 | r = mount_partition(PARTITION_VAR, m->partitions + PARTITION_VAR, where, "/var", uid_shift, uid_range, userns_fd, flags); |
d4dffb85 LP |
2220 | if (r < 0) |
2221 | return r; | |
2222 | ||
8d9a1d59 | 2223 | r = mount_partition(PARTITION_TMP, m->partitions + PARTITION_TMP, where, "/var/tmp", uid_shift, uid_range, userns_fd, flags); |
d4dffb85 LP |
2224 | if (r < 0) |
2225 | return r; | |
2226 | ||
8d9a1d59 LP |
2227 | int slash_boot_is_available = 0; |
2228 | if (where) { | |
2229 | r = slash_boot_is_available = mount_point_is_available(where, "/boot", /* missing_ok = */ true); | |
2230 | if (r < 0) | |
2231 | return r; | |
2232 | } | |
2233 | if (!where || slash_boot_is_available) { | |
2234 | r = mount_partition(PARTITION_XBOOTLDR, m->partitions + PARTITION_XBOOTLDR, where, "/boot", uid_shift, uid_range, userns_fd, flags); | |
1d96dae7 ZJS |
2235 | if (r < 0) |
2236 | return r; | |
2237 | slash_boot_is_available = !r; | |
2238 | } | |
d9223c07 | 2239 | |
8c1be37e | 2240 | if (m->partitions[PARTITION_ESP].found) { |
1d96dae7 | 2241 | const char *esp_path = NULL; |
1f0f82f1 | 2242 | |
8d9a1d59 LP |
2243 | if (where) { |
2244 | /* Mount the ESP to /boot/ if it exists and is empty and we didn't already mount the | |
2245 | * XBOOTLDR partition into it. Otherwise, use /efi instead, but only if it exists | |
2246 | * and is empty. */ | |
8c1be37e | 2247 | |
8d9a1d59 LP |
2248 | if (slash_boot_is_available) { |
2249 | r = mount_point_is_available(where, "/boot", /* missing_ok = */ false); | |
2250 | if (r < 0) | |
2251 | return r; | |
2252 | if (r > 0) | |
2253 | esp_path = "/boot"; | |
2254 | } | |
1f0f82f1 | 2255 | |
8d9a1d59 LP |
2256 | if (!esp_path) { |
2257 | r = mount_point_is_available(where, "/efi", /* missing_ok = */ true); | |
2258 | if (r < 0) | |
2259 | return r; | |
2260 | if (r > 0) | |
2261 | esp_path = "/efi"; | |
2262 | } | |
1d96dae7 | 2263 | } |
1f0f82f1 | 2264 | |
8d9a1d59 LP |
2265 | /* OK, let's mount the ESP now (possibly creating the dir if missing) */ |
2266 | r = mount_partition(PARTITION_ESP, m->partitions + PARTITION_ESP, where, esp_path, uid_shift, uid_range, userns_fd, flags); | |
2267 | if (r < 0) | |
2268 | return r; | |
8c1be37e LP |
2269 | } |
2270 | ||
2271 | return 0; | |
2272 | } | |
2273 | ||
21b61b1d LP |
2274 | int dissected_image_mount_and_warn( |
2275 | DissectedImage *m, | |
2276 | const char *where, | |
2277 | uid_t uid_shift, | |
2278 | uid_t uid_range, | |
8d9a1d59 | 2279 | int userns_fd, |
21b61b1d LP |
2280 | DissectImageFlags flags) { |
2281 | ||
af187ab2 LP |
2282 | int r; |
2283 | ||
2284 | assert(m); | |
af187ab2 | 2285 | |
8d9a1d59 | 2286 | r = dissected_image_mount(m, where, uid_shift, uid_range, userns_fd, flags); |
af187ab2 LP |
2287 | if (r == -ENXIO) |
2288 | return log_error_errno(r, "Not root file system found in image."); | |
2289 | if (r == -EMEDIUMTYPE) | |
7718ac97 | 2290 | return log_error_errno(r, "No suitable os-release/extension-release file in image found."); |
af187ab2 LP |
2291 | if (r == -EUNATCH) |
2292 | return log_error_errno(r, "Encrypted file system discovered, but decryption not requested."); | |
2293 | if (r == -EUCLEAN) | |
2294 | return log_error_errno(r, "File system check on image failed."); | |
2295 | if (r == -EBUSY) | |
2296 | return log_error_errno(r, "File system already mounted elsewhere."); | |
4dc28665 LP |
2297 | if (r == -EAFNOSUPPORT) |
2298 | return log_error_errno(r, "File system type not supported or not known."); | |
80ce8580 LP |
2299 | if (r == -EIDRM) |
2300 | return log_error_errno(r, "File system is too uncommon, refused."); | |
af187ab2 LP |
2301 | if (r < 0) |
2302 | return log_error_errno(r, "Failed to mount image: %m"); | |
2303 | ||
2304 | return r; | |
2305 | } | |
2306 | ||
349cc4a5 | 2307 | #if HAVE_LIBCRYPTSETUP |
9321ad51 | 2308 | struct DecryptedPartition { |
18b5886e LP |
2309 | struct crypt_device *device; |
2310 | char *name; | |
2311 | bool relinquished; | |
9321ad51 YW |
2312 | }; |
2313 | #endif | |
2314 | ||
2315 | typedef struct DecryptedPartition DecryptedPartition; | |
18b5886e LP |
2316 | |
2317 | struct DecryptedImage { | |
9321ad51 | 2318 | unsigned n_ref; |
18b5886e LP |
2319 | DecryptedPartition *decrypted; |
2320 | size_t n_decrypted; | |
18b5886e | 2321 | }; |
18b5886e | 2322 | |
9321ad51 | 2323 | static DecryptedImage* decrypted_image_free(DecryptedImage *d) { |
349cc4a5 | 2324 | #if HAVE_LIBCRYPTSETUP |
18b5886e LP |
2325 | int r; |
2326 | ||
2327 | if (!d) | |
2328 | return NULL; | |
2329 | ||
67f63ee5 | 2330 | for (size_t i = 0; i < d->n_decrypted; i++) { |
18b5886e LP |
2331 | DecryptedPartition *p = d->decrypted + i; |
2332 | ||
2333 | if (p->device && p->name && !p->relinquished) { | |
5228c58e DDM |
2334 | _cleanup_free_ char *node = NULL; |
2335 | ||
2336 | node = path_join("/dev/mapper", p->name); | |
2337 | if (node) { | |
2338 | r = btrfs_forget_device(node); | |
2339 | if (r < 0 && r != -ENOENT) | |
2340 | log_debug_errno(r, "Failed to forget btrfs device %s, ignoring: %m", node); | |
2341 | } else | |
2342 | log_oom_debug(); | |
2343 | ||
ea16d7f4 YW |
2344 | /* Let's deactivate lazily, as the dm volume may be already/still used by other processes. */ |
2345 | r = sym_crypt_deactivate_by_name(p->device, p->name, CRYPT_DEACTIVATE_DEFERRED); | |
18b5886e LP |
2346 | if (r < 0) |
2347 | log_debug_errno(r, "Failed to deactivate encrypted partition %s", p->name); | |
2348 | } | |
2349 | ||
2350 | if (p->device) | |
0d12936d | 2351 | sym_crypt_free(p->device); |
18b5886e LP |
2352 | free(p->name); |
2353 | } | |
2354 | ||
f91861e4 | 2355 | free(d->decrypted); |
18b5886e LP |
2356 | free(d); |
2357 | #endif | |
2358 | return NULL; | |
2359 | } | |
2360 | ||
9321ad51 YW |
2361 | DEFINE_TRIVIAL_REF_UNREF_FUNC(DecryptedImage, decrypted_image, decrypted_image_free); |
2362 | ||
349cc4a5 | 2363 | #if HAVE_LIBCRYPTSETUP |
9321ad51 YW |
2364 | static int decrypted_image_new(DecryptedImage **ret) { |
2365 | _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL; | |
2366 | ||
2367 | assert(ret); | |
2368 | ||
2369 | d = new(DecryptedImage, 1); | |
2370 | if (!d) | |
2371 | return -ENOMEM; | |
2372 | ||
2373 | *d = (DecryptedImage) { | |
2374 | .n_ref = 1, | |
2375 | }; | |
2376 | ||
2377 | *ret = TAKE_PTR(d); | |
2378 | return 0; | |
2379 | } | |
4623e8e6 LP |
2380 | |
2381 | static int make_dm_name_and_node(const void *original_node, const char *suffix, char **ret_name, char **ret_node) { | |
2382 | _cleanup_free_ char *name = NULL, *node = NULL; | |
2383 | const char *base; | |
2384 | ||
2385 | assert(original_node); | |
2386 | assert(suffix); | |
2387 | assert(ret_name); | |
2388 | assert(ret_node); | |
2389 | ||
2390 | base = strrchr(original_node, '/'); | |
2391 | if (!base) | |
ac1f3ad0 LB |
2392 | base = original_node; |
2393 | else | |
2394 | base++; | |
4623e8e6 LP |
2395 | if (isempty(base)) |
2396 | return -EINVAL; | |
2397 | ||
2398 | name = strjoin(base, suffix); | |
2399 | if (!name) | |
2400 | return -ENOMEM; | |
2401 | if (!filename_is_valid(name)) | |
2402 | return -EINVAL; | |
2403 | ||
0d12936d | 2404 | node = path_join(sym_crypt_get_dir(), name); |
4623e8e6 LP |
2405 | if (!node) |
2406 | return -ENOMEM; | |
2407 | ||
1cc6c93a YW |
2408 | *ret_name = TAKE_PTR(name); |
2409 | *ret_node = TAKE_PTR(node); | |
4623e8e6 | 2410 | |
4623e8e6 LP |
2411 | return 0; |
2412 | } | |
2413 | ||
18b5886e LP |
2414 | static int decrypt_partition( |
2415 | DissectedPartition *m, | |
2416 | const char *passphrase, | |
2417 | DissectImageFlags flags, | |
2418 | DecryptedImage *d) { | |
2419 | ||
2420 | _cleanup_free_ char *node = NULL, *name = NULL; | |
0d12936d | 2421 | _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL; |
254d1313 | 2422 | _cleanup_close_ int fd = -EBADF; |
18b5886e LP |
2423 | int r; |
2424 | ||
2425 | assert(m); | |
2426 | assert(d); | |
2427 | ||
2428 | if (!m->found || !m->node || !m->fstype) | |
2429 | return 0; | |
2430 | ||
2431 | if (!streq(m->fstype, "crypto_LUKS")) | |
2432 | return 0; | |
2433 | ||
bdd73ac5 ZJS |
2434 | if (!passphrase) |
2435 | return -ENOKEY; | |
2436 | ||
0d12936d LP |
2437 | r = dlopen_cryptsetup(); |
2438 | if (r < 0) | |
2439 | return r; | |
2440 | ||
4623e8e6 LP |
2441 | r = make_dm_name_and_node(m->node, "-decrypted", &name, &node); |
2442 | if (r < 0) | |
2443 | return r; | |
18b5886e | 2444 | |
319a4f4b | 2445 | if (!GREEDY_REALLOC0(d->decrypted, d->n_decrypted + 1)) |
18b5886e LP |
2446 | return -ENOMEM; |
2447 | ||
0d12936d | 2448 | r = sym_crypt_init(&cd, m->node); |
18b5886e | 2449 | if (r < 0) |
715cbb81 | 2450 | return log_debug_errno(r, "Failed to initialize dm-crypt: %m"); |
18b5886e | 2451 | |
efc3b12f | 2452 | cryptsetup_enable_logging(cd); |
1887032f | 2453 | |
0d12936d | 2454 | r = sym_crypt_load(cd, CRYPT_LUKS, NULL); |
294bd454 ZJS |
2455 | if (r < 0) |
2456 | return log_debug_errno(r, "Failed to load LUKS metadata: %m"); | |
18b5886e | 2457 | |
0d12936d | 2458 | r = sym_crypt_activate_by_passphrase(cd, name, CRYPT_ANY_SLOT, passphrase, strlen(passphrase), |
ef9c184d | 2459 | ((flags & DISSECT_IMAGE_DEVICE_READ_ONLY) ? CRYPT_ACTIVATE_READONLY : 0) | |
0d12936d | 2460 | ((flags & DISSECT_IMAGE_DISCARD_ON_CRYPTO) ? CRYPT_ACTIVATE_ALLOW_DISCARDS : 0)); |
294bd454 | 2461 | if (r < 0) { |
715cbb81 | 2462 | log_debug_errno(r, "Failed to activate LUKS device: %m"); |
294bd454 | 2463 | return r == -EPERM ? -EKEYREJECTED : r; |
18b5886e | 2464 | } |
18b5886e | 2465 | |
f7725647 YW |
2466 | fd = open(node, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY); |
2467 | if (fd < 0) | |
2468 | return log_debug_errno(errno, "Failed to open %s: %m", node); | |
2469 | ||
94344385 LP |
2470 | d->decrypted[d->n_decrypted++] = (DecryptedPartition) { |
2471 | .name = TAKE_PTR(name), | |
2472 | .device = TAKE_PTR(cd), | |
2473 | }; | |
18b5886e | 2474 | |
1cc6c93a | 2475 | m->decrypted_node = TAKE_PTR(node); |
f7725647 | 2476 | close_and_replace(m->mount_node_fd, fd); |
18b5886e LP |
2477 | |
2478 | return 0; | |
4623e8e6 LP |
2479 | } |
2480 | ||
89e62e0b LP |
2481 | static int verity_can_reuse( |
2482 | const VeritySettings *verity, | |
2483 | const char *name, | |
2484 | struct crypt_device **ret_cd) { | |
2485 | ||
ac1f3ad0 LB |
2486 | /* If the same volume was already open, check that the root hashes match, and reuse it if they do */ |
2487 | _cleanup_free_ char *root_hash_existing = NULL; | |
0d12936d | 2488 | _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL; |
ac1f3ad0 | 2489 | struct crypt_params_verity crypt_params = {}; |
89e62e0b | 2490 | size_t root_hash_existing_size; |
ac1f3ad0 LB |
2491 | int r; |
2492 | ||
89e62e0b LP |
2493 | assert(verity); |
2494 | assert(name); | |
ac1f3ad0 LB |
2495 | assert(ret_cd); |
2496 | ||
0d12936d | 2497 | r = sym_crypt_init_by_name(&cd, name); |
ac1f3ad0 LB |
2498 | if (r < 0) |
2499 | return log_debug_errno(r, "Error opening verity device, crypt_init_by_name failed: %m"); | |
0d12936d | 2500 | |
c719805e LP |
2501 | cryptsetup_enable_logging(cd); |
2502 | ||
0d12936d | 2503 | r = sym_crypt_get_verity_info(cd, &crypt_params); |
ac1f3ad0 LB |
2504 | if (r < 0) |
2505 | return log_debug_errno(r, "Error opening verity device, crypt_get_verity_info failed: %m"); | |
0d12936d | 2506 | |
89e62e0b LP |
2507 | root_hash_existing_size = verity->root_hash_size; |
2508 | root_hash_existing = malloc0(root_hash_existing_size); | |
ac1f3ad0 LB |
2509 | if (!root_hash_existing) |
2510 | return -ENOMEM; | |
0d12936d LP |
2511 | |
2512 | r = sym_crypt_volume_key_get(cd, CRYPT_ANY_SLOT, root_hash_existing, &root_hash_existing_size, NULL, 0); | |
ac1f3ad0 LB |
2513 | if (r < 0) |
2514 | return log_debug_errno(r, "Error opening verity device, crypt_volume_key_get failed: %m"); | |
89e62e0b LP |
2515 | if (verity->root_hash_size != root_hash_existing_size || |
2516 | memcmp(root_hash_existing, verity->root_hash, verity->root_hash_size) != 0) | |
ac1f3ad0 | 2517 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but root hashes are different."); |
89e62e0b | 2518 | |
ac1f3ad0 | 2519 | #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY |
89e62e0b LP |
2520 | /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the |
2521 | * same settings, so that a previous unsigned mount will not be reused if the user asks to use | |
28423d9a | 2522 | * signing for the new one, and vice versa. */ |
89e62e0b | 2523 | if (!!verity->root_hash_sig != !!(crypt_params.flags & CRYPT_VERITY_ROOT_HASH_SIGNATURE)) |
ac1f3ad0 LB |
2524 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but signature settings are not the same."); |
2525 | #endif | |
2526 | ||
2527 | *ret_cd = TAKE_PTR(cd); | |
2528 | return 0; | |
2529 | } | |
2530 | ||
cf1ab844 | 2531 | static char* dm_deferred_remove_clean(char *name) { |
ac1f3ad0 | 2532 | if (!name) |
75db809a | 2533 | return NULL; |
0d12936d LP |
2534 | |
2535 | (void) sym_crypt_deactivate_by_name(NULL, name, CRYPT_DEACTIVATE_DEFERRED); | |
75db809a | 2536 | return mfree(name); |
ac1f3ad0 LB |
2537 | } |
2538 | DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean); | |
2539 | ||
c2fa92e7 LP |
2540 | static int validate_signature_userspace(const VeritySettings *verity) { |
2541 | #if HAVE_OPENSSL | |
2542 | _cleanup_(sk_X509_free_allp) STACK_OF(X509) *sk = NULL; | |
2543 | _cleanup_strv_free_ char **certs = NULL; | |
2544 | _cleanup_(PKCS7_freep) PKCS7 *p7 = NULL; | |
2545 | _cleanup_free_ char *s = NULL; | |
2546 | _cleanup_(BIO_freep) BIO *bio = NULL; /* 'bio' must be freed first, 's' second, hence keep this order | |
2547 | * of declaration in place, please */ | |
2548 | const unsigned char *d; | |
c2fa92e7 LP |
2549 | int r; |
2550 | ||
2551 | assert(verity); | |
2552 | assert(verity->root_hash); | |
2553 | assert(verity->root_hash_sig); | |
2554 | ||
2555 | /* Because installing a signature certificate into the kernel chain is so messy, let's optionally do | |
2556 | * userspace validation. */ | |
2557 | ||
2558 | r = conf_files_list_nulstr(&certs, ".crt", NULL, CONF_FILES_REGULAR|CONF_FILES_FILTER_MASKED, CONF_PATHS_NULSTR("verity.d")); | |
2559 | if (r < 0) | |
2560 | return log_debug_errno(r, "Failed to enumerate certificates: %m"); | |
2561 | if (strv_isempty(certs)) { | |
2562 | log_debug("No userspace dm-verity certificates found."); | |
2563 | return 0; | |
2564 | } | |
2565 | ||
2566 | d = verity->root_hash_sig; | |
2567 | p7 = d2i_PKCS7(NULL, &d, (long) verity->root_hash_sig_size); | |
2568 | if (!p7) | |
2569 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to parse PKCS7 DER signature data."); | |
2570 | ||
2571 | s = hexmem(verity->root_hash, verity->root_hash_size); | |
2572 | if (!s) | |
2573 | return log_oom_debug(); | |
2574 | ||
2575 | bio = BIO_new_mem_buf(s, strlen(s)); | |
2576 | if (!bio) | |
2577 | return log_oom_debug(); | |
2578 | ||
2579 | sk = sk_X509_new_null(); | |
2580 | if (!sk) | |
2581 | return log_oom_debug(); | |
2582 | ||
2583 | STRV_FOREACH(i, certs) { | |
2584 | _cleanup_(X509_freep) X509 *c = NULL; | |
2585 | _cleanup_fclose_ FILE *f = NULL; | |
2586 | ||
2587 | f = fopen(*i, "re"); | |
2588 | if (!f) { | |
2589 | log_debug_errno(errno, "Failed to open '%s', ignoring: %m", *i); | |
2590 | continue; | |
2591 | } | |
2592 | ||
2593 | c = PEM_read_X509(f, NULL, NULL, NULL); | |
2594 | if (!c) { | |
2595 | log_debug("Failed to load X509 certificate '%s', ignoring.", *i); | |
2596 | continue; | |
2597 | } | |
2598 | ||
2599 | if (sk_X509_push(sk, c) == 0) | |
2600 | return log_oom_debug(); | |
2601 | ||
2602 | TAKE_PTR(c); | |
2603 | } | |
2604 | ||
2605 | r = PKCS7_verify(p7, sk, NULL, bio, NULL, PKCS7_NOINTERN|PKCS7_NOVERIFY); | |
2606 | if (r) | |
2607 | log_debug("Userspace PKCS#7 validation succeeded."); | |
2608 | else | |
2609 | log_debug("Userspace PKCS#7 validation failed: %s", ERR_error_string(ERR_get_error(), NULL)); | |
2610 | ||
2611 | return r; | |
2612 | #else | |
2613 | log_debug("Not doing client-side validation of dm-verity root hash signatures, OpenSSL support disabled."); | |
2614 | return 0; | |
2615 | #endif | |
2616 | } | |
2617 | ||
2618 | static int do_crypt_activate_verity( | |
2619 | struct crypt_device *cd, | |
2620 | const char *name, | |
2621 | const VeritySettings *verity) { | |
2622 | ||
2623 | bool check_signature; | |
ace07128 | 2624 | int r, k; |
c2fa92e7 LP |
2625 | |
2626 | assert(cd); | |
2627 | assert(name); | |
2628 | assert(verity); | |
2629 | ||
2630 | if (verity->root_hash_sig) { | |
2631 | r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIGNATURE"); | |
2632 | if (r < 0 && r != -ENXIO) | |
2633 | log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIGNATURE"); | |
2634 | ||
2635 | check_signature = r != 0; | |
2636 | } else | |
2637 | check_signature = false; | |
2638 | ||
2639 | if (check_signature) { | |
2640 | ||
2641 | #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY | |
2642 | /* First, if we have support for signed keys in the kernel, then try that first. */ | |
2643 | r = sym_crypt_activate_by_signed_key( | |
2644 | cd, | |
2645 | name, | |
2646 | verity->root_hash, | |
2647 | verity->root_hash_size, | |
2648 | verity->root_hash_sig, | |
2649 | verity->root_hash_sig_size, | |
2650 | CRYPT_ACTIVATE_READONLY); | |
2651 | if (r >= 0) | |
2652 | return r; | |
2653 | ||
ace07128 | 2654 | log_debug_errno(r, "Validation of dm-verity signature failed via the kernel, trying userspace validation instead: %m"); |
c2fa92e7 LP |
2655 | #else |
2656 | 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.", | |
2657 | program_invocation_short_name); | |
ace07128 | 2658 | r = 0; /* Set for the propagation below */ |
c2fa92e7 LP |
2659 | #endif |
2660 | ||
2661 | /* So this didn't work via the kernel, then let's try userspace validation instead. If that | |
2662 | * works we'll try to activate without telling the kernel the signature. */ | |
2663 | ||
ace07128 LB |
2664 | /* Preferably propagate the original kernel error, so that the fallback logic can work, |
2665 | * as the device-mapper is finicky around concurrent activations of the same volume */ | |
2666 | k = validate_signature_userspace(verity); | |
2667 | if (k < 0) | |
2668 | return r < 0 ? r : k; | |
2669 | if (k == 0) | |
2670 | return log_debug_errno(r < 0 ? r : SYNTHETIC_ERRNO(ENOKEY), | |
c2fa92e7 LP |
2671 | "Activation of signed Verity volume worked neither via the kernel nor in userspace, can't activate."); |
2672 | } | |
2673 | ||
2674 | return sym_crypt_activate_by_volume_key( | |
2675 | cd, | |
2676 | name, | |
2677 | verity->root_hash, | |
2678 | verity->root_hash_size, | |
2679 | CRYPT_ACTIVATE_READONLY); | |
2680 | } | |
2681 | ||
ad361a50 YW |
2682 | static usec_t verity_timeout(void) { |
2683 | usec_t t = 100 * USEC_PER_MSEC; | |
2684 | const char *e; | |
2685 | int r; | |
2686 | ||
2687 | /* On slower machines, like non-KVM vm, setting up device may take a long time. | |
2688 | * Let's make the timeout configurable. */ | |
2689 | ||
2690 | e = getenv("SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC"); | |
2691 | if (!e) | |
2692 | return t; | |
2693 | ||
2694 | r = parse_sec(e, &t); | |
2695 | if (r < 0) | |
2696 | log_debug_errno(r, | |
2697 | "Failed to parse timeout specified in $SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC, " | |
2698 | "using the default timeout (%s).", | |
2699 | FORMAT_TIMESPAN(t, USEC_PER_MSEC)); | |
2700 | ||
2701 | return t; | |
2702 | } | |
2703 | ||
4623e8e6 | 2704 | static int verity_partition( |
aee36b4e | 2705 | PartitionDesignator designator, |
4623e8e6 LP |
2706 | DissectedPartition *m, |
2707 | DissectedPartition *v, | |
89e62e0b | 2708 | const VeritySettings *verity, |
4623e8e6 LP |
2709 | DissectImageFlags flags, |
2710 | DecryptedImage *d) { | |
2711 | ||
0d12936d | 2712 | _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL; |
89e62e0b | 2713 | _cleanup_free_ char *node = NULL, *name = NULL; |
254d1313 | 2714 | _cleanup_close_ int mount_node_fd = -EBADF; |
4623e8e6 LP |
2715 | int r; |
2716 | ||
2717 | assert(m); | |
89e62e0b | 2718 | assert(v || (verity && verity->data_path)); |
4623e8e6 | 2719 | |
89e62e0b | 2720 | if (!verity || !verity->root_hash) |
4623e8e6 | 2721 | return 0; |
aee36b4e LP |
2722 | if (!((verity->designator < 0 && designator == PARTITION_ROOT) || |
2723 | (verity->designator == designator))) | |
2724 | return 0; | |
4623e8e6 LP |
2725 | |
2726 | if (!m->found || !m->node || !m->fstype) | |
2727 | return 0; | |
89e62e0b | 2728 | if (!verity->data_path) { |
e7cbe5cb LB |
2729 | if (!v->found || !v->node || !v->fstype) |
2730 | return 0; | |
4623e8e6 | 2731 | |
e7cbe5cb LB |
2732 | if (!streq(v->fstype, "DM_verity_hash")) |
2733 | return 0; | |
2734 | } | |
4623e8e6 | 2735 | |
0d12936d LP |
2736 | r = dlopen_cryptsetup(); |
2737 | if (r < 0) | |
2738 | return r; | |
2739 | ||
ac1f3ad0 LB |
2740 | if (FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) { |
2741 | /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */ | |
2742 | _cleanup_free_ char *root_hash_encoded = NULL; | |
0d12936d | 2743 | |
89e62e0b | 2744 | root_hash_encoded = hexmem(verity->root_hash, verity->root_hash_size); |
ac1f3ad0 LB |
2745 | if (!root_hash_encoded) |
2746 | return -ENOMEM; | |
aee36b4e | 2747 | |
ac1f3ad0 LB |
2748 | r = make_dm_name_and_node(root_hash_encoded, "-verity", &name, &node); |
2749 | } else | |
2750 | r = make_dm_name_and_node(m->node, "-verity", &name, &node); | |
4623e8e6 LP |
2751 | if (r < 0) |
2752 | return r; | |
2753 | ||
89e62e0b | 2754 | r = sym_crypt_init(&cd, verity->data_path ?: v->node); |
4623e8e6 LP |
2755 | if (r < 0) |
2756 | return r; | |
2757 | ||
efc3b12f | 2758 | cryptsetup_enable_logging(cd); |
1887032f | 2759 | |
0d12936d | 2760 | r = sym_crypt_load(cd, CRYPT_VERITY, NULL); |
4623e8e6 | 2761 | if (r < 0) |
294bd454 | 2762 | return r; |
4623e8e6 | 2763 | |
0d12936d | 2764 | r = sym_crypt_set_data_device(cd, m->node); |
4623e8e6 | 2765 | if (r < 0) |
294bd454 | 2766 | return r; |
4623e8e6 | 2767 | |
319a4f4b | 2768 | if (!GREEDY_REALLOC0(d->decrypted, d->n_decrypted + 1)) |
ac1f3ad0 LB |
2769 | return -ENOMEM; |
2770 | ||
2771 | /* If activating fails because the device already exists, check the metadata and reuse it if it matches. | |
2772 | * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time, | |
2773 | * retry a few times before giving up. */ | |
2774 | for (unsigned i = 0; i < N_DEVICE_NODE_LIST_ATTEMPTS; i++) { | |
758a3aeb | 2775 | _cleanup_(dm_deferred_remove_cleanp) char *restore_deferred_remove = NULL; |
1d369d78 | 2776 | _cleanup_(sym_crypt_freep) struct crypt_device *existing_cd = NULL; |
254d1313 | 2777 | _cleanup_close_ int fd = -EBADF; |
c2fa92e7 | 2778 | |
f80015ff YW |
2779 | /* First, check if the device already exists. */ |
2780 | fd = open(node, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY); | |
2781 | if (fd < 0 && !ERRNO_IS_DEVICE_ABSENT(errno)) | |
2782 | return log_debug_errno(errno, "Failed to open verity device %s: %m", node); | |
2783 | if (fd >= 0) | |
2784 | goto check; /* The device already exists. Let's check it. */ | |
2785 | ||
2786 | /* The symlink to the device node does not exist yet. Assume not activated, and let's activate it. */ | |
c2fa92e7 | 2787 | r = do_crypt_activate_verity(cd, name, verity); |
1d369d78 | 2788 | if (r >= 0) |
f7725647 | 2789 | goto try_open; /* The device is activated. Let's open it. */ |
ac1f3ad0 LB |
2790 | /* libdevmapper can return EINVAL when the device is already in the activation stage. |
2791 | * There's no way to distinguish this situation from a genuine error due to invalid | |
2aed63f4 | 2792 | * parameters, so immediately fall back to activating the device with a unique name. |
89e62e0b LP |
2793 | * Improvements in libcrypsetup can ensure this never happens: |
2794 | * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */ | |
ac1f3ad0 | 2795 | if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) |
0f75b0c5 | 2796 | break; |
1d369d78 YW |
2797 | if (r == -ENODEV) /* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */ |
2798 | goto try_again; | |
2799 | if (!IN_SET(r, | |
2800 | -EEXIST, /* Volume has already been opened and ready to be used. */ | |
2801 | -EBUSY /* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */)) | |
2802 | return log_debug_errno(r, "Failed to activate verity device %s: %m", node); | |
2803 | ||
f80015ff | 2804 | check: |
758a3aeb YW |
2805 | /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */ |
2806 | r = dm_deferred_remove_cancel(name); | |
2807 | /* -EBUSY and -ENXIO: the device has already been removed or being removed. We cannot | |
2808 | * use the device, try to open again. See target_message() in drivers/md/dm-ioctl.c | |
2809 | * and dm_cancel_deferred_remove() in drivers/md/dm.c */ | |
2810 | if (IN_SET(r, -EBUSY, -ENXIO)) | |
2811 | goto try_again; | |
2812 | if (r < 0) | |
2813 | return log_debug_errno(r, "Failed to disable automated deferred removal for verity device %s: %m", node); | |
12f5fbdf | 2814 | |
758a3aeb YW |
2815 | restore_deferred_remove = strdup(name); |
2816 | if (!restore_deferred_remove) | |
2817 | return log_oom_debug(); | |
c2923fdc | 2818 | |
1d369d78 YW |
2819 | r = verity_can_reuse(verity, name, &existing_cd); |
2820 | /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */ | |
2821 | if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) | |
2822 | break; | |
2823 | if (IN_SET(r, | |
2824 | -ENOENT, /* Removed?? */ | |
2825 | -EBUSY, /* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */ | |
2826 | -ENODEV /* Volume is being opened but not ready, crypt_init_by_name() would fail, try to open again. */ )) | |
2827 | goto try_again; | |
2828 | if (r < 0) | |
2829 | return log_debug_errno(r, "Failed to check if existing verity device %s can be reused: %m", node); | |
2830 | ||
f80015ff | 2831 | if (fd < 0) { |
1d369d78 YW |
2832 | /* devmapper might say that the device exists, but the devlink might not yet have been |
2833 | * created. Check and wait for the udev event in that case. */ | |
2834 | r = device_wait_for_devlink(node, "block", verity_timeout(), NULL); | |
2835 | /* Fallback to activation with a unique device if it's taking too long */ | |
2836 | if (r == -ETIMEDOUT && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) | |
0f75b0c5 | 2837 | break; |
1d369d78 YW |
2838 | if (r < 0) |
2839 | return log_debug_errno(r, "Failed to wait device node symlink %s: %m", node); | |
c2923fdc | 2840 | } |
ecab4c47 | 2841 | |
f7725647 YW |
2842 | try_open: |
2843 | if (fd < 0) { | |
2844 | /* Now, the device is activated and devlink is created. Let's open it. */ | |
2845 | fd = open(node, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY); | |
2846 | if (fd < 0) { | |
2847 | if (!ERRNO_IS_DEVICE_ABSENT(errno)) | |
2848 | return log_debug_errno(errno, "Failed to open verity device %s: %m", node); | |
2849 | ||
2850 | /* The device has already been removed?? */ | |
2851 | goto try_again; | |
2852 | } | |
2853 | } | |
2854 | ||
758a3aeb YW |
2855 | /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */ |
2856 | restore_deferred_remove = mfree(restore_deferred_remove); | |
2857 | ||
f7725647 | 2858 | mount_node_fd = TAKE_FD(fd); |
1d369d78 YW |
2859 | if (existing_cd) |
2860 | crypt_free_and_replace(cd, existing_cd); | |
2861 | ||
2862 | goto success; | |
2863 | ||
2864 | try_again: | |
2865 | /* Device is being removed by another process. Let's wait for a while. */ | |
4251512e | 2866 | (void) usleep_safe(2 * USEC_PER_MSEC); |
ac1f3ad0 LB |
2867 | } |
2868 | ||
0f75b0c5 YW |
2869 | /* All trials failed or a conflicting verity device exists. Let's try to activate with a unique name. */ |
2870 | if (FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) { | |
2871 | /* Before trying to activate with unique name, we need to free crypt_device object. | |
2872 | * Otherwise, we get error from libcryptsetup like the following: | |
2873 | * ------ | |
2874 | * systemd[1234]: Cannot use device /dev/loop5 which is in use (already mapped or mounted). | |
2875 | * ------ | |
2876 | */ | |
2877 | sym_crypt_free(cd); | |
2878 | cd = NULL; | |
aee36b4e | 2879 | return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d); |
0f75b0c5 | 2880 | } |
ac1f3ad0 | 2881 | |
9972e6d6 YW |
2882 | return log_debug_errno(SYNTHETIC_ERRNO(EBUSY), "All attempts to activate verity device %s failed.", name); |
2883 | ||
2884 | success: | |
94344385 LP |
2885 | d->decrypted[d->n_decrypted++] = (DecryptedPartition) { |
2886 | .name = TAKE_PTR(name), | |
2887 | .device = TAKE_PTR(cd), | |
2888 | }; | |
4623e8e6 | 2889 | |
1cc6c93a | 2890 | m->decrypted_node = TAKE_PTR(node); |
f7725647 | 2891 | close_and_replace(m->mount_node_fd, mount_node_fd); |
4623e8e6 LP |
2892 | |
2893 | return 0; | |
18b5886e LP |
2894 | } |
2895 | #endif | |
2896 | ||
2897 | int dissected_image_decrypt( | |
2898 | DissectedImage *m, | |
2899 | const char *passphrase, | |
89e62e0b | 2900 | const VeritySettings *verity, |
e330f97a | 2901 | DissectImageFlags flags) { |
18b5886e | 2902 | |
349cc4a5 | 2903 | #if HAVE_LIBCRYPTSETUP |
49b5b3b4 | 2904 | _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL; |
18b5886e LP |
2905 | int r; |
2906 | #endif | |
2907 | ||
2908 | assert(m); | |
89e62e0b | 2909 | assert(!verity || verity->root_hash || verity->root_hash_size == 0); |
18b5886e LP |
2910 | |
2911 | /* Returns: | |
2912 | * | |
2913 | * = 0 → There was nothing to decrypt | |
2914 | * > 0 → Decrypted successfully | |
d1c536f5 | 2915 | * -ENOKEY → There's something to decrypt but no key was supplied |
18b5886e LP |
2916 | * -EKEYREJECTED → Passed key was not correct |
2917 | */ | |
2918 | ||
89e62e0b | 2919 | if (verity && verity->root_hash && verity->root_hash_size < sizeof(sd_id128_t)) |
4623e8e6 LP |
2920 | return -EINVAL; |
2921 | ||
e330f97a | 2922 | if (!m->encrypted && !m->verity_ready) |
18b5886e | 2923 | return 0; |
18b5886e | 2924 | |
349cc4a5 | 2925 | #if HAVE_LIBCRYPTSETUP |
9321ad51 YW |
2926 | r = decrypted_image_new(&d); |
2927 | if (r < 0) | |
2928 | return r; | |
18b5886e | 2929 | |
569a0e42 | 2930 | for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) { |
18b5886e | 2931 | DissectedPartition *p = m->partitions + i; |
22043172 | 2932 | PartitionDesignator k; |
18b5886e LP |
2933 | |
2934 | if (!p->found) | |
2935 | continue; | |
2936 | ||
2937 | r = decrypt_partition(p, passphrase, flags, d); | |
2938 | if (r < 0) | |
2939 | return r; | |
2940 | ||
dd894023 | 2941 | k = partition_verity_of(i); |
4623e8e6 | 2942 | if (k >= 0) { |
aee36b4e | 2943 | r = verity_partition(i, p, m->partitions + k, verity, flags | DISSECT_IMAGE_VERITY_SHARE, d); |
4623e8e6 LP |
2944 | if (r < 0) |
2945 | return r; | |
2946 | } | |
2947 | ||
d2c6e79d | 2948 | if (!p->decrypted_fstype && p->mount_node_fd >= 0 && p->decrypted_node) { |
c80c9079 | 2949 | r = probe_filesystem_full(p->mount_node_fd, p->decrypted_node, 0, UINT64_MAX, &p->decrypted_fstype); |
7cc84b2c | 2950 | if (r < 0 && r != -EUCLEAN) |
18b5886e LP |
2951 | return r; |
2952 | } | |
2953 | } | |
2954 | ||
ac1e1b5f | 2955 | m->decrypted_image = TAKE_PTR(d); |
18b5886e LP |
2956 | |
2957 | return 1; | |
2958 | #else | |
2959 | return -EOPNOTSUPP; | |
2960 | #endif | |
2961 | } | |
2962 | ||
2963 | int dissected_image_decrypt_interactively( | |
2964 | DissectedImage *m, | |
2965 | const char *passphrase, | |
89e62e0b | 2966 | const VeritySettings *verity, |
e330f97a | 2967 | DissectImageFlags flags) { |
18b5886e LP |
2968 | |
2969 | _cleanup_strv_free_erase_ char **z = NULL; | |
2970 | int n = 3, r; | |
2971 | ||
2972 | if (passphrase) | |
2973 | n--; | |
2974 | ||
2975 | for (;;) { | |
e330f97a | 2976 | r = dissected_image_decrypt(m, passphrase, verity, flags); |
18b5886e LP |
2977 | if (r >= 0) |
2978 | return r; | |
2979 | if (r == -EKEYREJECTED) | |
2980 | log_error_errno(r, "Incorrect passphrase, try again!"); | |
fc95c359 YW |
2981 | else if (r != -ENOKEY) |
2982 | return log_error_errno(r, "Failed to decrypt image: %m"); | |
18b5886e | 2983 | |
baaa35ad ZJS |
2984 | if (--n < 0) |
2985 | return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED), | |
2986 | "Too many retries."); | |
18b5886e LP |
2987 | |
2988 | z = strv_free(z); | |
2989 | ||
8806bb4b | 2990 | r = ask_password_auto("Please enter image passphrase:", NULL, "dissect", "dissect", "dissect.passphrase", USEC_INFINITY, 0, &z); |
18b5886e LP |
2991 | if (r < 0) |
2992 | return log_error_errno(r, "Failed to query for passphrase: %m"); | |
2993 | ||
2994 | passphrase = z[0]; | |
2995 | } | |
2996 | } | |
2997 | ||
e330f97a | 2998 | static int decrypted_image_relinquish(DecryptedImage *d) { |
18b5886e LP |
2999 | assert(d); |
3000 | ||
67f63ee5 ZJS |
3001 | /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a |
3002 | * boolean so that we don't clean it up ourselves either anymore */ | |
18b5886e | 3003 | |
349cc4a5 | 3004 | #if HAVE_LIBCRYPTSETUP |
67f63ee5 ZJS |
3005 | int r; |
3006 | ||
3007 | for (size_t i = 0; i < d->n_decrypted; i++) { | |
18b5886e LP |
3008 | DecryptedPartition *p = d->decrypted + i; |
3009 | ||
3010 | if (p->relinquished) | |
3011 | continue; | |
3012 | ||
0d12936d | 3013 | r = sym_crypt_deactivate_by_name(NULL, p->name, CRYPT_DEACTIVATE_DEFERRED); |
18b5886e LP |
3014 | if (r < 0) |
3015 | return log_debug_errno(r, "Failed to mark %s for auto-removal: %m", p->name); | |
3016 | ||
3017 | p->relinquished = true; | |
3018 | } | |
3019 | #endif | |
3020 | ||
3021 | return 0; | |
3022 | } | |
3023 | ||
f906075a YW |
3024 | int dissected_image_relinquish(DissectedImage *m) { |
3025 | int r; | |
3026 | ||
3027 | assert(m); | |
3028 | ||
3029 | if (m->decrypted_image) { | |
3030 | r = decrypted_image_relinquish(m->decrypted_image); | |
3031 | if (r < 0) | |
3032 | return r; | |
3033 | } | |
3034 | ||
3035 | if (m->loop) | |
3036 | loop_device_relinquish(m->loop); | |
3037 | ||
3038 | return 0; | |
3039 | } | |
3040 | ||
89e62e0b LP |
3041 | static char *build_auxiliary_path(const char *image, const char *suffix) { |
3042 | const char *e; | |
3043 | char *n; | |
3044 | ||
3045 | assert(image); | |
3046 | assert(suffix); | |
3047 | ||
3048 | e = endswith(image, ".raw"); | |
3049 | if (!e) | |
3050 | return strjoin(e, suffix); | |
3051 | ||
3052 | n = new(char, e - image + strlen(suffix) + 1); | |
3053 | if (!n) | |
3054 | return NULL; | |
3055 | ||
3056 | strcpy(mempcpy(n, image, e - image), suffix); | |
3057 | return n; | |
3058 | } | |
3059 | ||
3060 | void verity_settings_done(VeritySettings *v) { | |
3061 | assert(v); | |
3062 | ||
3063 | v->root_hash = mfree(v->root_hash); | |
3064 | v->root_hash_size = 0; | |
3065 | ||
3066 | v->root_hash_sig = mfree(v->root_hash_sig); | |
3067 | v->root_hash_sig_size = 0; | |
3068 | ||
3069 | v->data_path = mfree(v->data_path); | |
3070 | } | |
3071 | ||
3072 | int verity_settings_load( | |
3073 | VeritySettings *verity, | |
f5ea63a5 LP |
3074 | const char *image, |
3075 | const char *root_hash_path, | |
89e62e0b LP |
3076 | const char *root_hash_sig_path) { |
3077 | ||
3078 | _cleanup_free_ void *root_hash = NULL, *root_hash_sig = NULL; | |
3079 | size_t root_hash_size = 0, root_hash_sig_size = 0; | |
3080 | _cleanup_free_ char *verity_data_path = NULL; | |
aee36b4e | 3081 | PartitionDesignator designator; |
78ebe980 LP |
3082 | int r; |
3083 | ||
89e62e0b | 3084 | assert(verity); |
78ebe980 | 3085 | assert(image); |
aee36b4e | 3086 | assert(verity->designator < 0 || IN_SET(verity->designator, PARTITION_ROOT, PARTITION_USR)); |
78ebe980 | 3087 | |
89e62e0b LP |
3088 | /* If we are asked to load the root hash for a device node, exit early */ |
3089 | if (is_device_path(image)) | |
78ebe980 | 3090 | return 0; |
78ebe980 | 3091 | |
d5fcc5b0 LP |
3092 | r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIDECAR"); |
3093 | if (r < 0 && r != -ENXIO) | |
3094 | log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIDECAR, ignoring: %m"); | |
3095 | if (r == 0) | |
3096 | return 0; | |
3097 | ||
aee36b4e LP |
3098 | designator = verity->designator; |
3099 | ||
89e62e0b | 3100 | /* We only fill in what isn't already filled in */ |
c2923fdc | 3101 | |
89e62e0b | 3102 | if (!verity->root_hash) { |
e7cbe5cb | 3103 | _cleanup_free_ char *text = NULL; |
e7cbe5cb | 3104 | |
0389f4fa | 3105 | if (root_hash_path) { |
aee36b4e | 3106 | /* If explicitly specified it takes precedence */ |
0389f4fa LB |
3107 | r = read_one_line_file(root_hash_path, &text); |
3108 | if (r < 0) | |
e7cbe5cb | 3109 | return r; |
aee36b4e LP |
3110 | |
3111 | if (designator < 0) | |
3112 | designator = PARTITION_ROOT; | |
0389f4fa | 3113 | } else { |
aee36b4e LP |
3114 | /* Otherwise look for xattr and separate file, and first for the data for root and if |
3115 | * that doesn't exist for /usr */ | |
0389f4fa | 3116 | |
aee36b4e | 3117 | if (designator < 0 || designator == PARTITION_ROOT) { |
c53e07e2 | 3118 | r = getxattr_malloc(image, "user.verity.roothash", &text); |
aee36b4e LP |
3119 | if (r < 0) { |
3120 | _cleanup_free_ char *p = NULL; | |
78ebe980 | 3121 | |
00675c36 | 3122 | if (r != -ENOENT && !ERRNO_IS_XATTR_ABSENT(r)) |
aee36b4e | 3123 | return r; |
e7cbe5cb | 3124 | |
aee36b4e LP |
3125 | p = build_auxiliary_path(image, ".roothash"); |
3126 | if (!p) | |
3127 | return -ENOMEM; | |
3128 | ||
3129 | r = read_one_line_file(p, &text); | |
3130 | if (r < 0 && r != -ENOENT) | |
3131 | return r; | |
3132 | } | |
3133 | ||
3134 | if (text) | |
3135 | designator = PARTITION_ROOT; | |
3136 | } | |
3137 | ||
3138 | if (!text && (designator < 0 || designator == PARTITION_USR)) { | |
3139 | /* So in the "roothash" xattr/file name above the "root" of course primarily | |
3140 | * refers to the root of the Verity Merkle tree. But coincidentally it also | |
3141 | * is the hash for the *root* file system, i.e. the "root" neatly refers to | |
3142 | * two distinct concepts called "root". Taking benefit of this happy | |
3143 | * coincidence we call the file with the root hash for the /usr/ file system | |
3144 | * `usrhash`, because `usrroothash` or `rootusrhash` would just be too | |
3145 | * confusing. We thus drop the reference to the root of the Merkle tree, and | |
3146 | * just indicate which file system it's about. */ | |
c53e07e2 | 3147 | r = getxattr_malloc(image, "user.verity.usrhash", &text); |
aee36b4e LP |
3148 | if (r < 0) { |
3149 | _cleanup_free_ char *p = NULL; | |
3150 | ||
00675c36 | 3151 | if (r != -ENOENT && !ERRNO_IS_XATTR_ABSENT(r)) |
aee36b4e LP |
3152 | return r; |
3153 | ||
3154 | p = build_auxiliary_path(image, ".usrhash"); | |
3155 | if (!p) | |
3156 | return -ENOMEM; | |
3157 | ||
3158 | r = read_one_line_file(p, &text); | |
3159 | if (r < 0 && r != -ENOENT) | |
3160 | return r; | |
3161 | } | |
3162 | ||
3163 | if (text) | |
3164 | designator = PARTITION_USR; | |
0389f4fa | 3165 | } |
e7cbe5cb LB |
3166 | } |
3167 | ||
3168 | if (text) { | |
bdd2036e | 3169 | r = unhexmem(text, &root_hash, &root_hash_size); |
e7cbe5cb LB |
3170 | if (r < 0) |
3171 | return r; | |
89e62e0b | 3172 | if (root_hash_size < sizeof(sd_id128_t)) |
e7cbe5cb LB |
3173 | return -EINVAL; |
3174 | } | |
3175 | } | |
3176 | ||
90f98986 | 3177 | if ((root_hash || verity->root_hash) && !verity->root_hash_sig) { |
aee36b4e | 3178 | if (root_hash_sig_path) { |
ae9cf30b | 3179 | r = read_full_file(root_hash_sig_path, (char**) &root_hash_sig, &root_hash_sig_size); |
aee36b4e LP |
3180 | if (r < 0 && r != -ENOENT) |
3181 | return r; | |
3182 | ||
3183 | if (designator < 0) | |
3184 | designator = PARTITION_ROOT; | |
3185 | } else { | |
3186 | if (designator < 0 || designator == PARTITION_ROOT) { | |
3187 | _cleanup_free_ char *p = NULL; | |
3188 | ||
3189 | /* Follow naming convention recommended by the relevant RFC: | |
3190 | * https://tools.ietf.org/html/rfc5751#section-3.2.1 */ | |
3191 | p = build_auxiliary_path(image, ".roothash.p7s"); | |
3192 | if (!p) | |
3193 | return -ENOMEM; | |
89e62e0b | 3194 | |
ae9cf30b | 3195 | r = read_full_file(p, (char**) &root_hash_sig, &root_hash_sig_size); |
aee36b4e LP |
3196 | if (r < 0 && r != -ENOENT) |
3197 | return r; | |
3198 | if (r >= 0) | |
3199 | designator = PARTITION_ROOT; | |
3200 | } | |
3201 | ||
3202 | if (!root_hash_sig && (designator < 0 || designator == PARTITION_USR)) { | |
3203 | _cleanup_free_ char *p = NULL; | |
3204 | ||
3205 | p = build_auxiliary_path(image, ".usrhash.p7s"); | |
3206 | if (!p) | |
3207 | return -ENOMEM; | |
89e62e0b | 3208 | |
ae9cf30b | 3209 | r = read_full_file(p, (char**) &root_hash_sig, &root_hash_sig_size); |
aee36b4e LP |
3210 | if (r < 0 && r != -ENOENT) |
3211 | return r; | |
3212 | if (r >= 0) | |
3213 | designator = PARTITION_USR; | |
3214 | } | |
89e62e0b LP |
3215 | } |
3216 | ||
aee36b4e | 3217 | if (root_hash_sig && root_hash_sig_size == 0) /* refuse empty size signatures */ |
89e62e0b LP |
3218 | return -EINVAL; |
3219 | } | |
3220 | ||
3221 | if (!verity->data_path) { | |
3222 | _cleanup_free_ char *p = NULL; | |
3223 | ||
3224 | p = build_auxiliary_path(image, ".verity"); | |
3225 | if (!p) | |
3226 | return -ENOMEM; | |
3227 | ||
3228 | if (access(p, F_OK) < 0) { | |
3229 | if (errno != ENOENT) | |
3230 | return -errno; | |
3231 | } else | |
3232 | verity_data_path = TAKE_PTR(p); | |
3233 | } | |
3234 | ||
3235 | if (root_hash) { | |
3236 | verity->root_hash = TAKE_PTR(root_hash); | |
3237 | verity->root_hash_size = root_hash_size; | |
3238 | } | |
3239 | ||
3240 | if (root_hash_sig) { | |
3241 | verity->root_hash_sig = TAKE_PTR(root_hash_sig); | |
3242 | verity->root_hash_sig_size = root_hash_sig_size; | |
e7cbe5cb | 3243 | } |
89e62e0b LP |
3244 | |
3245 | if (verity_data_path) | |
3246 | verity->data_path = TAKE_PTR(verity_data_path); | |
78ebe980 | 3247 | |
aee36b4e LP |
3248 | if (verity->designator < 0) |
3249 | verity->designator = designator; | |
3250 | ||
78ebe980 LP |
3251 | return 1; |
3252 | } | |
3253 | ||
88b3300f LP |
3254 | int dissected_image_load_verity_sig_partition( |
3255 | DissectedImage *m, | |
3256 | int fd, | |
3257 | VeritySettings *verity) { | |
3258 | ||
3259 | _cleanup_free_ void *root_hash = NULL, *root_hash_sig = NULL; | |
3260 | _cleanup_(json_variant_unrefp) JsonVariant *v = NULL; | |
3261 | size_t root_hash_size, root_hash_sig_size; | |
3262 | _cleanup_free_ char *buf = NULL; | |
3263 | PartitionDesignator d; | |
3264 | DissectedPartition *p; | |
3265 | JsonVariant *rh, *sig; | |
3266 | ssize_t n; | |
3267 | char *e; | |
3268 | int r; | |
3269 | ||
3270 | assert(m); | |
3271 | assert(fd >= 0); | |
3272 | assert(verity); | |
3273 | ||
3274 | if (verity->root_hash && verity->root_hash_sig) /* Already loaded? */ | |
3275 | return 0; | |
3276 | ||
3277 | r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_EMBEDDED"); | |
3278 | if (r < 0 && r != -ENXIO) | |
3279 | log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_EMBEDDED, ignoring: %m"); | |
3280 | if (r == 0) | |
3281 | return 0; | |
3282 | ||
dd894023 | 3283 | d = partition_verity_sig_of(verity->designator < 0 ? PARTITION_ROOT : verity->designator); |
88b3300f LP |
3284 | assert(d >= 0); |
3285 | ||
3286 | p = m->partitions + d; | |
3287 | if (!p->found) | |
3288 | return 0; | |
3289 | if (p->offset == UINT64_MAX || p->size == UINT64_MAX) | |
3290 | return -EINVAL; | |
3291 | ||
3292 | if (p->size > 4*1024*1024) /* Signature data cannot possible be larger than 4M, refuse that */ | |
d31f8e0c | 3293 | return log_debug_errno(SYNTHETIC_ERRNO(EFBIG), "Verity signature partition is larger than 4M, refusing."); |
88b3300f LP |
3294 | |
3295 | buf = new(char, p->size+1); | |
3296 | if (!buf) | |
3297 | return -ENOMEM; | |
3298 | ||
3299 | n = pread(fd, buf, p->size, p->offset); | |
3300 | if (n < 0) | |
3301 | return -ENOMEM; | |
3302 | if ((uint64_t) n != p->size) | |
3303 | return -EIO; | |
3304 | ||
3305 | e = memchr(buf, 0, p->size); | |
3306 | if (e) { | |
3307 | /* If we found a NUL byte then the rest of the data must be NUL too */ | |
3308 | if (!memeqzero(e, p->size - (e - buf))) | |
3309 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature data contains embedded NUL byte."); | |
3310 | } else | |
3311 | buf[p->size] = 0; | |
3312 | ||
3313 | r = json_parse(buf, 0, &v, NULL, NULL); | |
3314 | if (r < 0) | |
3315 | return log_debug_errno(r, "Failed to parse signature JSON data: %m"); | |
3316 | ||
3317 | rh = json_variant_by_key(v, "rootHash"); | |
3318 | if (!rh) | |
3319 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature JSON object lacks 'rootHash' field."); | |
3320 | if (!json_variant_is_string(rh)) | |
3321 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "'rootHash' field of signature JSON object is not a string."); | |
3322 | ||
bdd2036e | 3323 | r = unhexmem(json_variant_string(rh), &root_hash, &root_hash_size); |
88b3300f LP |
3324 | if (r < 0) |
3325 | return log_debug_errno(r, "Failed to parse root hash field: %m"); | |
3326 | ||
3327 | /* Check if specified root hash matches if it is specified */ | |
3328 | if (verity->root_hash && | |
3329 | memcmp_nn(verity->root_hash, verity->root_hash_size, root_hash, root_hash_size) != 0) { | |
3330 | _cleanup_free_ char *a = NULL, *b = NULL; | |
3331 | ||
3332 | a = hexmem(root_hash, root_hash_size); | |
3333 | b = hexmem(verity->root_hash, verity->root_hash_size); | |
3334 | ||
3335 | return log_debug_errno(r, "Root hash in signature JSON data (%s) doesn't match configured hash (%s).", strna(a), strna(b)); | |
3336 | } | |
3337 | ||
3338 | sig = json_variant_by_key(v, "signature"); | |
3339 | if (!sig) | |
3340 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature JSON object lacks 'signature' field."); | |
3341 | if (!json_variant_is_string(sig)) | |
3342 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "'signature' field of signature JSON object is not a string."); | |
3343 | ||
bdd2036e | 3344 | r = unbase64mem(json_variant_string(sig), &root_hash_sig, &root_hash_sig_size); |
88b3300f LP |
3345 | if (r < 0) |
3346 | return log_debug_errno(r, "Failed to parse signature field: %m"); | |
3347 | ||
3348 | free_and_replace(verity->root_hash, root_hash); | |
3349 | verity->root_hash_size = root_hash_size; | |
3350 | ||
3351 | free_and_replace(verity->root_hash_sig, root_hash_sig); | |
3352 | verity->root_hash_sig_size = root_hash_sig_size; | |
3353 | ||
3354 | return 1; | |
3355 | } | |
3356 | ||
22847508 | 3357 | int dissected_image_acquire_metadata(DissectedImage *m, DissectImageFlags extra_flags) { |
3b925504 LP |
3358 | |
3359 | enum { | |
3360 | META_HOSTNAME, | |
3361 | META_MACHINE_ID, | |
3362 | META_MACHINE_INFO, | |
3363 | META_OS_RELEASE, | |
fab22946 | 3364 | META_INITRD_RELEASE, |
a81fe93e LP |
3365 | META_SYSEXT_RELEASE, |
3366 | META_CONFEXT_RELEASE, | |
a4e0d617 | 3367 | META_HAS_INIT_SYSTEM, |
3b925504 LP |
3368 | _META_MAX, |
3369 | }; | |
3370 | ||
9a4b883b | 3371 | static const char *const paths[_META_MAX] = { |
7718ac97 LB |
3372 | [META_HOSTNAME] = "/etc/hostname\0", |
3373 | [META_MACHINE_ID] = "/etc/machine-id\0", | |
3374 | [META_MACHINE_INFO] = "/etc/machine-info\0", | |
a81fe93e LP |
3375 | [META_OS_RELEASE] = "/etc/os-release\0" |
3376 | "/usr/lib/os-release\0", | |
3377 | [META_INITRD_RELEASE] = "/etc/initrd-release\0" | |
3378 | "/usr/lib/initrd-release\0", | |
3379 | [META_SYSEXT_RELEASE] = "sysext-release\0", /* String used only for logging. */ | |
3380 | [META_CONFEXT_RELEASE] = "confext-release\0", /* ditto */ | |
a4e0d617 | 3381 | [META_HAS_INIT_SYSTEM] = "has-init-system\0", /* ditto */ |
3b925504 LP |
3382 | }; |
3383 | ||
a81fe93e | 3384 | _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL, **initrd_release = NULL, **sysext_release = NULL, **confext_release = NULL; |
71136404 | 3385 | _cleanup_close_pair_ int error_pipe[2] = EBADF_PAIR; |
3b925504 LP |
3386 | _cleanup_(rmdir_and_freep) char *t = NULL; |
3387 | _cleanup_(sigkill_waitp) pid_t child = 0; | |
3388 | sd_id128_t machine_id = SD_ID128_NULL; | |
3389 | _cleanup_free_ char *hostname = NULL; | |
67f63ee5 | 3390 | unsigned n_meta_initialized = 0; |
af8219d5 | 3391 | int fds[2 * _META_MAX], r, v; |
a4e0d617 | 3392 | int has_init_system = -1; |
af8219d5 | 3393 | ssize_t n; |
3b925504 LP |
3394 | |
3395 | BLOCK_SIGNALS(SIGCHLD); | |
3396 | ||
3397 | assert(m); | |
3398 | ||
b3a9d980 | 3399 | for (; n_meta_initialized < _META_MAX; n_meta_initialized++) { |
3ee413e6 | 3400 | assert(paths[n_meta_initialized]); |
d9119c00 | 3401 | |
3b925504 LP |
3402 | if (pipe2(fds + 2*n_meta_initialized, O_CLOEXEC) < 0) { |
3403 | r = -errno; | |
3404 | goto finish; | |
3405 | } | |
7718ac97 | 3406 | } |
3b925504 LP |
3407 | |
3408 | r = mkdtemp_malloc("/tmp/dissect-XXXXXX", &t); | |
3409 | if (r < 0) | |
3410 | goto finish; | |
3411 | ||
af8219d5 LP |
3412 | if (pipe2(error_pipe, O_CLOEXEC) < 0) { |
3413 | r = -errno; | |
3414 | goto finish; | |
3415 | } | |
3416 | ||
e9ccae31 | 3417 | r = safe_fork("(sd-dissect)", FORK_RESET_SIGNALS|FORK_DEATHSIG_SIGTERM|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE, &child); |
be39f6ee | 3418 | if (r < 0) |
3b925504 | 3419 | goto finish; |
be39f6ee | 3420 | if (r == 0) { |
a4e0d617 | 3421 | /* Child in a new mount namespace */ |
af8219d5 LP |
3422 | error_pipe[0] = safe_close(error_pipe[0]); |
3423 | ||
7cf66030 LP |
3424 | r = dissected_image_mount( |
3425 | m, | |
3426 | t, | |
8d9a1d59 LP |
3427 | /* uid_shift= */ UID_INVALID, |
3428 | /* uid_range= */ UID_INVALID, | |
3429 | /* userns_fd= */ -EBADF, | |
22847508 ZJS |
3430 | extra_flags | |
3431 | DISSECT_IMAGE_READ_ONLY | | |
3432 | DISSECT_IMAGE_MOUNT_ROOT_ONLY | | |
7cf66030 | 3433 | DISSECT_IMAGE_USR_NO_ROOT); |
429d4e41 LP |
3434 | if (r < 0) { |
3435 | log_debug_errno(r, "Failed to mount dissected image: %m"); | |
03ae68f4 | 3436 | goto inner_fail; |
429d4e41 | 3437 | } |
3b925504 | 3438 | |
67f63ee5 | 3439 | for (unsigned k = 0; k < _META_MAX; k++) { |
37e44c3f | 3440 | _cleanup_close_ int fd = -ENOENT; |
3b925504 | 3441 | |
3ee413e6 | 3442 | assert(paths[k]); |
7718ac97 | 3443 | |
3b925504 LP |
3444 | fds[2*k] = safe_close(fds[2*k]); |
3445 | ||
a4e0d617 LP |
3446 | switch (k) { |
3447 | ||
a81fe93e | 3448 | case META_SYSEXT_RELEASE: |
a8e8bcfb LP |
3449 | if (!m->image_name) |
3450 | goto next; | |
3451 | ||
a81fe93e LP |
3452 | /* As per the os-release spec, if the image is an extension it will have a |
3453 | * file named after the image name in extension-release.d/ - we use the image | |
3454 | * name and try to resolve it with the extension-release helpers, as | |
3455 | * sometimes the image names are mangled on deployment and do not match | |
3456 | * anymore. Unlike other paths this is not fixed, and the image name can be | |
3457 | * mangled on deployment, so by calling into the helper we allow a fallback | |
3458 | * that matches on the first extension-release file found in the directory, | |
3459 | * if one named after the image cannot be found first. */ | |
3460 | r = open_extension_release( | |
3461 | t, | |
3462 | IMAGE_SYSEXT, | |
3463 | m->image_name, | |
3464 | /* relax_extension_release_check= */ false, | |
3465 | /* ret_path= */ NULL, | |
3466 | &fd); | |
3467 | if (r < 0) | |
3468 | fd = r; | |
3469 | break; | |
3470 | ||
3471 | case META_CONFEXT_RELEASE: | |
a8e8bcfb LP |
3472 | if (!m->image_name) |
3473 | goto next; | |
3474 | ||
a81fe93e LP |
3475 | /* As above */ |
3476 | r = open_extension_release( | |
3477 | t, | |
3478 | IMAGE_CONFEXT, | |
3479 | m->image_name, | |
3480 | /* relax_extension_release_check= */ false, | |
3481 | /* ret_path= */ NULL, | |
3482 | &fd); | |
9a4b883b | 3483 | if (r < 0) |
484d26da | 3484 | fd = r; |
484d26da | 3485 | |
a4e0d617 LP |
3486 | break; |
3487 | ||
3488 | case META_HAS_INIT_SYSTEM: { | |
3489 | bool found = false; | |
a4e0d617 LP |
3490 | |
3491 | FOREACH_STRING(init, | |
a81fe93e LP |
3492 | "/usr/lib/systemd/systemd", /* systemd on /usr/ merged system */ |
3493 | "/lib/systemd/systemd", /* systemd on /usr/ non-merged systems */ | |
a4e0d617 LP |
3494 | "/sbin/init") { /* traditional path the Linux kernel invokes */ |
3495 | ||
f461a28d | 3496 | r = chase(init, t, CHASE_PREFIX_ROOT, NULL, NULL); |
a4e0d617 LP |
3497 | if (r < 0) { |
3498 | if (r != -ENOENT) | |
3499 | log_debug_errno(r, "Failed to resolve %s, ignoring: %m", init); | |
3500 | } else { | |
3501 | found = true; | |
3502 | break; | |
3503 | } | |
3504 | } | |
3505 | ||
e22c60a9 | 3506 | r = loop_write(fds[2*k+1], &found, sizeof(found)); |
a4e0d617 LP |
3507 | if (r < 0) |
3508 | goto inner_fail; | |
3509 | ||
29b4db7e | 3510 | goto next; |
a4e0d617 LP |
3511 | } |
3512 | ||
3513 | default: | |
9a4b883b | 3514 | NULSTR_FOREACH(p, paths[k]) { |
f461a28d | 3515 | fd = chase_and_open(p, t, CHASE_PREFIX_ROOT, O_RDONLY|O_CLOEXEC|O_NOCTTY, NULL); |
9a4b883b LB |
3516 | if (fd >= 0) |
3517 | break; | |
3518 | } | |
a4e0d617 LP |
3519 | } |
3520 | ||
36952d19 LP |
3521 | if (fd < 0) { |
3522 | log_debug_errno(fd, "Failed to read %s file of image, ignoring: %m", paths[k]); | |
29b4db7e | 3523 | goto next; |
36952d19 | 3524 | } |
3b925504 | 3525 | |
f5fbe71d | 3526 | r = copy_bytes(fd, fds[2*k+1], UINT64_MAX, 0); |
a4e0d617 LP |
3527 | if (r < 0) |
3528 | goto inner_fail; | |
3b925504 | 3529 | |
29b4db7e | 3530 | next: |
3b925504 LP |
3531 | fds[2*k+1] = safe_close(fds[2*k+1]); |
3532 | } | |
3533 | ||
3534 | _exit(EXIT_SUCCESS); | |
a4e0d617 LP |
3535 | |
3536 | inner_fail: | |
03ae68f4 | 3537 | /* Let parent know the error */ |
a4e0d617 LP |
3538 | (void) write(error_pipe[1], &r, sizeof(r)); |
3539 | _exit(EXIT_FAILURE); | |
3b925504 LP |
3540 | } |
3541 | ||
af8219d5 LP |
3542 | error_pipe[1] = safe_close(error_pipe[1]); |
3543 | ||
67f63ee5 | 3544 | for (unsigned k = 0; k < _META_MAX; k++) { |
3b925504 LP |
3545 | _cleanup_fclose_ FILE *f = NULL; |
3546 | ||
3ee413e6 | 3547 | assert(paths[k]); |
7718ac97 | 3548 | |
3b925504 LP |
3549 | fds[2*k+1] = safe_close(fds[2*k+1]); |
3550 | ||
4fa744a3 | 3551 | f = take_fdopen(&fds[2*k], "r"); |
3b925504 LP |
3552 | if (!f) { |
3553 | r = -errno; | |
3554 | goto finish; | |
3555 | } | |
3556 | ||
3b925504 LP |
3557 | switch (k) { |
3558 | ||
3559 | case META_HOSTNAME: | |
3560 | r = read_etc_hostname_stream(f, &hostname); | |
3561 | if (r < 0) | |
f6048e5e | 3562 | log_debug_errno(r, "Failed to read /etc/hostname of image: %m"); |
3b925504 LP |
3563 | |
3564 | break; | |
3565 | ||
3566 | case META_MACHINE_ID: { | |
3567 | _cleanup_free_ char *line = NULL; | |
3568 | ||
3569 | r = read_line(f, LONG_LINE_MAX, &line); | |
3570 | if (r < 0) | |
f6048e5e | 3571 | log_debug_errno(r, "Failed to read /etc/machine-id of image: %m"); |
3b925504 LP |
3572 | else if (r == 33) { |
3573 | r = sd_id128_from_string(line, &machine_id); | |
3574 | if (r < 0) | |
3575 | log_debug_errno(r, "Image contains invalid /etc/machine-id: %s", line); | |
3576 | } else if (r == 0) | |
f6048e5e | 3577 | log_debug("/etc/machine-id file of image is empty."); |
ab763cb2 | 3578 | else if (streq(line, "uninitialized")) |
f6048e5e | 3579 | log_debug("/etc/machine-id file of image is uninitialized (likely aborted first boot)."); |
3b925504 | 3580 | else |
f6048e5e | 3581 | log_debug("/etc/machine-id file of image has unexpected length %i.", r); |
3b925504 LP |
3582 | |
3583 | break; | |
3584 | } | |
3585 | ||
3586 | case META_MACHINE_INFO: | |
aa8fbc74 | 3587 | r = load_env_file_pairs(f, "machine-info", &machine_info); |
3b925504 | 3588 | if (r < 0) |
f6048e5e | 3589 | log_debug_errno(r, "Failed to read /etc/machine-info of image: %m"); |
3b925504 LP |
3590 | |
3591 | break; | |
3592 | ||
3593 | case META_OS_RELEASE: | |
aa8fbc74 | 3594 | r = load_env_file_pairs(f, "os-release", &os_release); |
3b925504 | 3595 | if (r < 0) |
f6048e5e | 3596 | log_debug_errno(r, "Failed to read OS release file of image: %m"); |
3b925504 LP |
3597 | |
3598 | break; | |
7718ac97 | 3599 | |
fab22946 LP |
3600 | case META_INITRD_RELEASE: |
3601 | r = load_env_file_pairs(f, "initrd-release", &initrd_release); | |
3602 | if (r < 0) | |
3603 | log_debug_errno(r, "Failed to read initrd release file of image: %m"); | |
3604 | ||
3605 | break; | |
3606 | ||
a81fe93e LP |
3607 | case META_SYSEXT_RELEASE: |
3608 | r = load_env_file_pairs(f, "sysext-release", &sysext_release); | |
3609 | if (r < 0) | |
3610 | log_debug_errno(r, "Failed to read sysext release file of image: %m"); | |
484d26da | 3611 | |
a81fe93e LP |
3612 | break; |
3613 | ||
3614 | case META_CONFEXT_RELEASE: | |
3615 | r = load_env_file_pairs(f, "confext-release", &confext_release); | |
3616 | if (r < 0) | |
3617 | log_debug_errno(r, "Failed to read confext release file of image: %m"); | |
7718ac97 LB |
3618 | |
3619 | break; | |
a4e0d617 LP |
3620 | |
3621 | case META_HAS_INIT_SYSTEM: { | |
3622 | bool b = false; | |
3623 | size_t nr; | |
3624 | ||
3625 | errno = 0; | |
3626 | nr = fread(&b, 1, sizeof(b), f); | |
3627 | if (nr != sizeof(b)) | |
3628 | log_debug_errno(errno_or_else(EIO), "Failed to read has-init-system boolean: %m"); | |
3629 | else | |
3630 | has_init_system = b; | |
3631 | ||
3632 | break; | |
3633 | }} | |
3b925504 LP |
3634 | } |
3635 | ||
2e87a1fd | 3636 | r = wait_for_terminate_and_check("(sd-dissect)", child, 0); |
3b925504 | 3637 | child = 0; |
2e87a1fd | 3638 | if (r < 0) |
8d5e61db | 3639 | goto finish; |
af8219d5 LP |
3640 | |
3641 | n = read(error_pipe[0], &v, sizeof(v)); | |
8d5e61db LP |
3642 | if (n < 0) { |
3643 | r = -errno; | |
3644 | goto finish; | |
3645 | } | |
3646 | if (n == sizeof(v)) { | |
3647 | r = v; /* propagate error sent to us from child */ | |
3648 | goto finish; | |
3649 | } | |
3650 | if (n != 0) { | |
3651 | r = -EIO; | |
3652 | goto finish; | |
3653 | } | |
3654 | if (r != EXIT_SUCCESS) { | |
3655 | r = -EPROTO; | |
3656 | goto finish; | |
3657 | } | |
3b925504 LP |
3658 | |
3659 | free_and_replace(m->hostname, hostname); | |
3660 | m->machine_id = machine_id; | |
3661 | strv_free_and_replace(m->machine_info, machine_info); | |
3662 | strv_free_and_replace(m->os_release, os_release); | |
fab22946 | 3663 | strv_free_and_replace(m->initrd_release, initrd_release); |
a81fe93e LP |
3664 | strv_free_and_replace(m->sysext_release, sysext_release); |
3665 | strv_free_and_replace(m->confext_release, confext_release); | |
a4e0d617 | 3666 | m->has_init_system = has_init_system; |
3b925504 LP |
3667 | |
3668 | finish: | |
67f63ee5 | 3669 | for (unsigned k = 0; k < n_meta_initialized; k++) |
3b925504 LP |
3670 | safe_close_pair(fds + 2*k); |
3671 | ||
3672 | return r; | |
3673 | } | |
3674 | ||
2348043f LP |
3675 | Architecture dissected_image_architecture(DissectedImage *img) { |
3676 | assert(img); | |
3677 | ||
3678 | if (img->partitions[PARTITION_ROOT].found && | |
3679 | img->partitions[PARTITION_ROOT].architecture >= 0) | |
3680 | return img->partitions[PARTITION_ROOT].architecture; | |
3681 | ||
3682 | if (img->partitions[PARTITION_USR].found && | |
3683 | img->partitions[PARTITION_USR].architecture >= 0) | |
3684 | return img->partitions[PARTITION_USR].architecture; | |
3685 | ||
3686 | return _ARCHITECTURE_INVALID; | |
3687 | } | |
3688 | ||
1e63dc4f YW |
3689 | int dissect_loop_device( |
3690 | LoopDevice *loop, | |
3691 | const VeritySettings *verity, | |
3692 | const MountOptions *mount_options, | |
84be0c71 | 3693 | const ImagePolicy *image_policy, |
1e63dc4f YW |
3694 | DissectImageFlags flags, |
3695 | DissectedImage **ret) { | |
3696 | ||
08f14be4 | 3697 | #if HAVE_BLKID |
1e63dc4f YW |
3698 | _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL; |
3699 | int r; | |
3700 | ||
3701 | assert(loop); | |
1e63dc4f | 3702 | |
08f14be4 | 3703 | r = dissected_image_new(loop->backing_file ?: loop->node, &m); |
1e63dc4f YW |
3704 | if (r < 0) |
3705 | return r; | |
3706 | ||
3707 | m->loop = loop_device_ref(loop); | |
51778dea | 3708 | m->image_size = m->loop->device_size; |
22ee78a8 | 3709 | m->sector_size = m->loop->sector_size; |
1e63dc4f | 3710 | |
84be0c71 | 3711 | r = dissect_image(m, loop->fd, loop->node, verity, mount_options, image_policy, flags); |
08f14be4 | 3712 | if (r < 0) |
08f14be4 YW |
3713 | return r; |
3714 | ||
93a8a85b LP |
3715 | if (ret) |
3716 | *ret = TAKE_PTR(m); | |
3717 | ||
1e63dc4f | 3718 | return 0; |
08f14be4 YW |
3719 | #else |
3720 | return -EOPNOTSUPP; | |
3721 | #endif | |
1e63dc4f YW |
3722 | } |
3723 | ||
bad31660 | 3724 | int dissect_loop_device_and_warn( |
1e63dc4f | 3725 | LoopDevice *loop, |
89e62e0b | 3726 | const VeritySettings *verity, |
18d73705 | 3727 | const MountOptions *mount_options, |
84be0c71 | 3728 | const ImagePolicy *image_policy, |
4526113f LP |
3729 | DissectImageFlags flags, |
3730 | DissectedImage **ret) { | |
3731 | ||
bad31660 | 3732 | assert(loop); |
4526113f | 3733 | |
7cd7a195 | 3734 | return dissect_log_error( |
4953e39c | 3735 | LOG_ERR, |
7cd7a195 LP |
3736 | dissect_loop_device(loop, verity, mount_options, image_policy, flags, ret), |
3737 | loop->backing_file ?: loop->node, | |
3738 | verity); | |
4526113f | 3739 | |
4526113f LP |
3740 | } |
3741 | ||
49536766 LP |
3742 | bool dissected_image_verity_candidate(const DissectedImage *image, PartitionDesignator partition_designator) { |
3743 | assert(image); | |
3744 | ||
3745 | /* Checks if this partition could theoretically do Verity. For non-partitioned images this only works | |
3746 | * if there's an external verity file supplied, for which we can consult .has_verity. For partitioned | |
3747 | * images we only check the partition type. | |
3748 | * | |
3749 | * This call is used to decide whether to suppress or show a verity column in tabular output of the | |
3750 | * image. */ | |
3751 | ||
e7cbe5cb | 3752 | if (image->single_file_system) |
c3c88d67 | 3753 | return partition_designator == PARTITION_ROOT && image->has_verity; |
e7cbe5cb | 3754 | |
dd894023 | 3755 | return partition_verity_of(partition_designator) >= 0; |
e7cbe5cb LB |
3756 | } |
3757 | ||
49536766 LP |
3758 | bool dissected_image_verity_ready(const DissectedImage *image, PartitionDesignator partition_designator) { |
3759 | PartitionDesignator k; | |
3760 | ||
3761 | assert(image); | |
3762 | ||
3763 | /* Checks if this partition has verity data available that we can activate. For non-partitioned this | |
3764 | * works for the root partition, for others only if the associated verity partition was found. */ | |
3765 | ||
3766 | if (!image->verity_ready) | |
3767 | return false; | |
e7cbe5cb LB |
3768 | |
3769 | if (image->single_file_system) | |
49536766 | 3770 | return partition_designator == PARTITION_ROOT; |
e7cbe5cb | 3771 | |
dd894023 | 3772 | k = partition_verity_of(partition_designator); |
e7cbe5cb LB |
3773 | return k >= 0 && image->partitions[k].found; |
3774 | } | |
3775 | ||
8ee9615e LP |
3776 | bool dissected_image_verity_sig_ready(const DissectedImage *image, PartitionDesignator partition_designator) { |
3777 | PartitionDesignator k; | |
3778 | ||
3779 | assert(image); | |
3780 | ||
3781 | /* Checks if this partition has verity signature data available that we can use. */ | |
3782 | ||
3783 | if (!image->verity_sig_ready) | |
3784 | return false; | |
3785 | ||
3786 | if (image->single_file_system) | |
3787 | return partition_designator == PARTITION_ROOT; | |
3788 | ||
dd894023 | 3789 | k = partition_verity_sig_of(partition_designator); |
8ee9615e LP |
3790 | return k >= 0 && image->partitions[k].found; |
3791 | } | |
3792 | ||
18d73705 LB |
3793 | MountOptions* mount_options_free_all(MountOptions *options) { |
3794 | MountOptions *m; | |
3795 | ||
52e3671b | 3796 | while ((m = LIST_POP(mount_options, options))) { |
18d73705 LB |
3797 | free(m->options); |
3798 | free(m); | |
3799 | } | |
3800 | ||
3801 | return NULL; | |
3802 | } | |
3803 | ||
569a0e42 | 3804 | const char* mount_options_from_designator(const MountOptions *options, PartitionDesignator designator) { |
f5215bc8 | 3805 | LIST_FOREACH(mount_options, m, options) |
9ece6444 | 3806 | if (designator == m->partition_designator && !isempty(m->options)) |
18d73705 | 3807 | return m->options; |
6aa05ebd | 3808 | |
18d73705 LB |
3809 | return NULL; |
3810 | } | |
3811 | ||
6aa05ebd LP |
3812 | int mount_image_privately_interactively( |
3813 | const char *image, | |
84be0c71 | 3814 | const ImagePolicy *image_policy, |
6aa05ebd LP |
3815 | DissectImageFlags flags, |
3816 | char **ret_directory, | |
a133d2c3 | 3817 | int *ret_dir_fd, |
e330f97a | 3818 | LoopDevice **ret_loop_device) { |
6aa05ebd | 3819 | |
27ec815e | 3820 | _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT; |
6aa05ebd | 3821 | _cleanup_(loop_device_unrefp) LoopDevice *d = NULL; |
6aa05ebd | 3822 | _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL; |
a4b3e942 | 3823 | _cleanup_free_ char *dir = NULL; |
6aa05ebd LP |
3824 | int r; |
3825 | ||
3826 | /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This | |
3827 | * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image | |
3828 | * easily. */ | |
3829 | ||
3830 | assert(image); | |
6aa05ebd | 3831 | assert(ret_loop_device); |
6aa05ebd | 3832 | |
0b75493d | 3833 | /* We intend to mount this right-away, hence add the partitions if needed and pin them. */ |
73d88b80 LP |
3834 | flags |= DISSECT_IMAGE_ADD_PARTITION_DEVICES | |
3835 | DISSECT_IMAGE_PIN_PARTITION_DEVICES; | |
3836 | ||
27ec815e LP |
3837 | r = verity_settings_load(&verity, image, NULL, NULL); |
3838 | if (r < 0) | |
3839 | return log_error_errno(r, "Failed to load root hash data: %m"); | |
3840 | ||
6aa05ebd LP |
3841 | r = loop_device_make_by_path( |
3842 | image, | |
ef9c184d | 3843 | FLAGS_SET(flags, DISSECT_IMAGE_DEVICE_READ_ONLY) ? O_RDONLY : O_RDWR, |
22ee78a8 | 3844 | /* sector_size= */ UINT32_MAX, |
6aa05ebd | 3845 | FLAGS_SET(flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN, |
7f52206a | 3846 | LOCK_SH, |
6aa05ebd LP |
3847 | &d); |
3848 | if (r < 0) | |
7b87fe4c | 3849 | return log_error_errno(r, "Failed to set up loopback device for %s: %m", image); |
6aa05ebd | 3850 | |
84be0c71 LP |
3851 | r = dissect_loop_device_and_warn( |
3852 | d, | |
3853 | &verity, | |
3854 | /* mount_options= */ NULL, | |
3855 | image_policy, | |
3856 | flags, | |
3857 | &dissected_image); | |
6aa05ebd LP |
3858 | if (r < 0) |
3859 | return r; | |
3860 | ||
88b3300f LP |
3861 | r = dissected_image_load_verity_sig_partition(dissected_image, d->fd, &verity); |
3862 | if (r < 0) | |
3863 | return r; | |
3864 | ||
e330f97a | 3865 | r = dissected_image_decrypt_interactively(dissected_image, NULL, &verity, flags); |
6aa05ebd LP |
3866 | if (r < 0) |
3867 | return r; | |
3868 | ||
3869 | r = detach_mount_namespace(); | |
3870 | if (r < 0) | |
3871 | return log_error_errno(r, "Failed to detach mount namespace: %m"); | |
3872 | ||
a4b3e942 | 3873 | r = mkdir_p("/run/systemd/mount-rootfs", 0555); |
6aa05ebd LP |
3874 | if (r < 0) |
3875 | return log_error_errno(r, "Failed to create mount point: %m"); | |
3876 | ||
a4b3e942 LP |
3877 | r = dissected_image_mount_and_warn( |
3878 | dissected_image, | |
3879 | "/run/systemd/mount-rootfs", | |
3880 | /* uid_shift= */ UID_INVALID, | |
3881 | /* uid_range= */ UID_INVALID, | |
8d9a1d59 | 3882 | /* userns_fd= */ -EBADF, |
a4b3e942 | 3883 | flags); |
6aa05ebd | 3884 | if (r < 0) |
af187ab2 | 3885 | return r; |
6aa05ebd | 3886 | |
41bc4849 LP |
3887 | r = loop_device_flock(d, LOCK_UN); |
3888 | if (r < 0) | |
3889 | return r; | |
3890 | ||
3044d343 YW |
3891 | r = dissected_image_relinquish(dissected_image); |
3892 | if (r < 0) | |
3893 | return log_error_errno(r, "Failed to relinquish DM and loopback block devices: %m"); | |
6aa05ebd | 3894 | |
a4b3e942 LP |
3895 | if (ret_directory) { |
3896 | dir = strdup("/run/systemd/mount-rootfs"); | |
3897 | if (!dir) | |
3898 | return log_oom(); | |
3899 | } | |
3900 | ||
a133d2c3 DDM |
3901 | if (ret_dir_fd) { |
3902 | _cleanup_close_ int dir_fd = -EBADF; | |
3903 | ||
a4b3e942 | 3904 | dir_fd = open("/run/systemd/mount-rootfs", O_CLOEXEC|O_DIRECTORY); |
a133d2c3 DDM |
3905 | if (dir_fd < 0) |
3906 | return log_error_errno(errno, "Failed to open mount point directory: %m"); | |
3907 | ||
3908 | *ret_dir_fd = TAKE_FD(dir_fd); | |
3909 | } | |
3910 | ||
a4b3e942 LP |
3911 | if (ret_directory) |
3912 | *ret_directory = TAKE_PTR(dir); | |
6aa05ebd | 3913 | |
a4b3e942 | 3914 | *ret_loop_device = TAKE_PTR(d); |
6aa05ebd LP |
3915 | return 0; |
3916 | } | |
3917 | ||
06768b90 LB |
3918 | static bool mount_options_relax_extension_release_checks(const MountOptions *options) { |
3919 | if (!options) | |
3920 | return false; | |
3921 | ||
3922 | return string_contains_word(mount_options_from_designator(options, PARTITION_ROOT), ",", "x-systemd.relax-extension-release-check") || | |
3923 | string_contains_word(mount_options_from_designator(options, PARTITION_USR), ",", "x-systemd.relax-extension-release-check") || | |
3924 | string_contains_word(options->options, ",", "x-systemd.relax-extension-release-check"); | |
3925 | } | |
3926 | ||
93f59701 | 3927 | int verity_dissect_and_mount( |
cedf5b1a | 3928 | int src_fd, |
93f59701 LB |
3929 | const char *src, |
3930 | const char *dest, | |
3931 | const MountOptions *options, | |
84be0c71 | 3932 | const ImagePolicy *image_policy, |
93f59701 LB |
3933 | const char *required_host_os_release_id, |
3934 | const char *required_host_os_release_version_id, | |
60c5f700 | 3935 | const char *required_host_os_release_sysext_level, |
5897469a | 3936 | const char *required_host_os_release_confext_level, |
3e107272 LB |
3937 | const char *required_sysext_scope, |
3938 | DissectedImage **ret_image) { | |
93f59701 | 3939 | |
4beda316 | 3940 | _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL; |
4beda316 LB |
3941 | _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL; |
3942 | _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT; | |
3943 | DissectImageFlags dissect_image_flags; | |
06768b90 | 3944 | bool relax_extension_release_check; |
4beda316 LB |
3945 | int r; |
3946 | ||
3947 | assert(src); | |
3e107272 LB |
3948 | /* Verifying release metadata requires mounted image for now, so ensure the check is skipped when |
3949 | * opening an image without mounting it immediately (i.e.: 'dest' is NULL). */ | |
3950 | assert(!required_host_os_release_id || dest); | |
4beda316 | 3951 | |
06768b90 LB |
3952 | relax_extension_release_check = mount_options_relax_extension_release_checks(options); |
3953 | ||
cedf5b1a | 3954 | /* We might get an FD for the image, but we use the original path to look for the dm-verity files */ |
4beda316 LB |
3955 | r = verity_settings_load(&verity, src, NULL, NULL); |
3956 | if (r < 0) | |
3957 | return log_debug_errno(r, "Failed to load root hash: %m"); | |
3958 | ||
06768b90 | 3959 | dissect_image_flags = (verity.data_path ? DISSECT_IMAGE_NO_PARTITION_TABLE : 0) | |
484d26da | 3960 | (relax_extension_release_check ? DISSECT_IMAGE_RELAX_EXTENSION_CHECK : 0) | |
73d88b80 LP |
3961 | DISSECT_IMAGE_ADD_PARTITION_DEVICES | |
3962 | DISSECT_IMAGE_PIN_PARTITION_DEVICES; | |
4beda316 | 3963 | |
cedf5b1a LB |
3964 | /* Note that we don't use loop_device_make here, as the FD is most likely O_PATH which would not be |
3965 | * accepted by LOOP_CONFIGURE, so just let loop_device_make_by_path reopen it as a regular FD. */ | |
4beda316 | 3966 | r = loop_device_make_by_path( |
cedf5b1a | 3967 | src_fd >= 0 ? FORMAT_PROC_FD_PATH(src_fd) : src, |
22ee78a8 LP |
3968 | /* open_flags= */ -1, |
3969 | /* sector_size= */ UINT32_MAX, | |
4beda316 | 3970 | verity.data_path ? 0 : LO_FLAGS_PARTSCAN, |
7f52206a | 3971 | LOCK_SH, |
4beda316 LB |
3972 | &loop_device); |
3973 | if (r < 0) | |
3974 | return log_debug_errno(r, "Failed to create loop device for image: %m"); | |
3975 | ||
bad31660 YW |
3976 | r = dissect_loop_device( |
3977 | loop_device, | |
4beda316 LB |
3978 | &verity, |
3979 | options, | |
84be0c71 | 3980 | image_policy, |
4beda316 LB |
3981 | dissect_image_flags, |
3982 | &dissected_image); | |
3983 | /* No partition table? Might be a single-filesystem image, try again */ | |
3984 | if (!verity.data_path && r == -ENOPKG) | |
bad31660 YW |
3985 | r = dissect_loop_device( |
3986 | loop_device, | |
4beda316 LB |
3987 | &verity, |
3988 | options, | |
84be0c71 | 3989 | image_policy, |
75dc190d | 3990 | dissect_image_flags | DISSECT_IMAGE_NO_PARTITION_TABLE, |
4beda316 LB |
3991 | &dissected_image); |
3992 | if (r < 0) | |
3993 | return log_debug_errno(r, "Failed to dissect image: %m"); | |
3994 | ||
88b3300f LP |
3995 | r = dissected_image_load_verity_sig_partition(dissected_image, loop_device->fd, &verity); |
3996 | if (r < 0) | |
3997 | return r; | |
3998 | ||
4beda316 LB |
3999 | r = dissected_image_decrypt( |
4000 | dissected_image, | |
4001 | NULL, | |
4002 | &verity, | |
e330f97a | 4003 | dissect_image_flags); |
4beda316 LB |
4004 | if (r < 0) |
4005 | return log_debug_errno(r, "Failed to decrypt dissected image: %m"); | |
4006 | ||
3e107272 LB |
4007 | if (dest) { |
4008 | r = mkdir_p_label(dest, 0755); | |
4009 | if (r < 0) | |
4010 | return log_debug_errno(r, "Failed to create destination directory %s: %m", dest); | |
4011 | r = umount_recursive(dest, 0); | |
4012 | if (r < 0) | |
4013 | return log_debug_errno(r, "Failed to umount under destination directory %s: %m", dest); | |
4014 | } | |
4beda316 | 4015 | |
8d9a1d59 LP |
4016 | r = dissected_image_mount( |
4017 | dissected_image, | |
4018 | dest, | |
4019 | /* uid_shift= */ UID_INVALID, | |
4020 | /* uid_range= */ UID_INVALID, | |
4021 | /* userns_fd= */ -EBADF, | |
4022 | dissect_image_flags); | |
4beda316 LB |
4023 | if (r < 0) |
4024 | return log_debug_errno(r, "Failed to mount image: %m"); | |
4025 | ||
41bc4849 LP |
4026 | r = loop_device_flock(loop_device, LOCK_UN); |
4027 | if (r < 0) | |
4028 | return log_debug_errno(r, "Failed to unlock loopback device: %m"); | |
4029 | ||
93f59701 LB |
4030 | /* If we got os-release values from the caller, then we need to match them with the image's |
4031 | * extension-release.d/ content. Return -EINVAL if there's any mismatch. | |
4032 | * First, check the distro ID. If that matches, then check the new SYSEXT_LEVEL value if | |
37361f46 LB |
4033 | * available, or else fallback to VERSION_ID. If neither is present (eg: rolling release), |
4034 | * then a simple match on the ID will be performed. */ | |
8b2dcbbd | 4035 | if (required_host_os_release_id) { |
93f59701 | 4036 | _cleanup_strv_free_ char **extension_release = NULL; |
484d26da | 4037 | ImageClass class = IMAGE_SYSEXT; |
93f59701 | 4038 | |
d30d86b7 YW |
4039 | assert(!isempty(required_host_os_release_id)); |
4040 | ||
b60e0f57 | 4041 | r = load_extension_release_pairs(dest, IMAGE_SYSEXT, dissected_image->image_name, relax_extension_release_check, &extension_release); |
484d26da MG |
4042 | if (r == -ENOENT) { |
4043 | r = load_extension_release_pairs(dest, IMAGE_CONFEXT, dissected_image->image_name, relax_extension_release_check, &extension_release); | |
4044 | if (r >= 0) | |
4045 | class = IMAGE_CONFEXT; | |
4046 | } | |
93f59701 LB |
4047 | if (r < 0) |
4048 | return log_debug_errno(r, "Failed to parse image %s extension-release metadata: %m", dissected_image->image_name); | |
4049 | ||
4050 | r = extension_release_validate( | |
60c5f700 LP |
4051 | dissected_image->image_name, |
4052 | required_host_os_release_id, | |
4053 | required_host_os_release_version_id, | |
5897469a | 4054 | class == IMAGE_SYSEXT ? required_host_os_release_sysext_level : required_host_os_release_confext_level, |
60c5f700 | 4055 | required_sysext_scope, |
30dfe035 | 4056 | extension_release, |
484d26da | 4057 | class); |
93f59701 LB |
4058 | if (r == 0) |
4059 | return log_debug_errno(SYNTHETIC_ERRNO(ESTALE), "Image %s extension-release metadata does not match the root's", dissected_image->image_name); | |
4060 | if (r < 0) | |
4061 | return log_debug_errno(r, "Failed to compare image %s extension-release metadata with the root's os-release: %m", dissected_image->image_name); | |
4062 | } | |
4063 | ||
3044d343 YW |
4064 | r = dissected_image_relinquish(dissected_image); |
4065 | if (r < 0) | |
4066 | return log_debug_errno(r, "Failed to relinquish dissected image: %m"); | |
4beda316 | 4067 | |
3e107272 LB |
4068 | if (ret_image) |
4069 | *ret_image = TAKE_PTR(dissected_image); | |
4070 | ||
4beda316 LB |
4071 | return 0; |
4072 | } |