]>
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" |
f4351959 | 29 | #include "chase-symlinks.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" |
93f59701 | 42 | #include "extension-release.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" |
35cd0ba5 | 54 | #include "mkdir-label.h" |
8c1be37e | 55 | #include "mount-util.h" |
e4de7287 | 56 | #include "mountpoint-util.h" |
6aa05ebd | 57 | #include "namespace-util.h" |
d8b4d14d | 58 | #include "nulstr-util.h" |
c2fa92e7 | 59 | #include "openssl-util.h" |
d58ad743 | 60 | #include "os-util.h" |
8c1be37e | 61 | #include "path-util.h" |
3b925504 LP |
62 | #include "process-util.h" |
63 | #include "raw-clone.h" | |
81939d9d | 64 | #include "resize-fs.h" |
3b925504 | 65 | #include "signal-util.h" |
05c4c59f | 66 | #include "sparse-endian.h" |
8c1be37e | 67 | #include "stat-util.h" |
18b5886e | 68 | #include "stdio-util.h" |
8c1be37e LP |
69 | #include "string-table.h" |
70 | #include "string-util.h" | |
2eedfd2d | 71 | #include "strv.h" |
e4de7287 | 72 | #include "tmpfile-util.h" |
a8040b6d | 73 | #include "udev-util.h" |
2d3a5a73 | 74 | #include "user-util.h" |
41488e1f | 75 | #include "xattr-util.h" |
8c1be37e | 76 | |
28e2641a FF |
77 | /* how many times to wait for the device nodes to appear */ |
78 | #define N_DEVICE_NODE_LIST_ATTEMPTS 10 | |
79 | ||
80ce8580 LP |
80 | int dissect_fstype_ok(const char *fstype) { |
81 | const char *e; | |
82 | bool b; | |
83 | ||
84 | /* When we automatically mount file systems, be a bit conservative by default what we are willing to | |
85 | * mount, just as an extra safety net to not mount with badly maintained legacy file system | |
86 | * drivers. */ | |
87 | ||
88 | e = secure_getenv("SYSTEMD_DISSECT_FILE_SYSTEMS"); | |
89 | if (e) { | |
90 | _cleanup_strv_free_ char **l = NULL; | |
91 | ||
92 | l = strv_split(e, ":"); | |
93 | if (!l) | |
94 | return -ENOMEM; | |
95 | ||
96 | b = strv_contains(l, fstype); | |
97 | } else | |
98 | b = STR_IN_SET(fstype, | |
99 | "btrfs", | |
100 | "erofs", | |
101 | "ext4", | |
102 | "squashfs", | |
103 | "vfat", | |
104 | "xfs"); | |
105 | if (b) | |
106 | return true; | |
107 | ||
108 | log_debug("File system type '%s' is not allowed to be mounted as result of automatic dissection.", fstype); | |
109 | return false; | |
110 | } | |
111 | ||
05c4c59f LP |
112 | int probe_sector_size(int fd, uint32_t *ret) { |
113 | ||
114 | struct gpt_header { | |
115 | char signature[8]; | |
116 | le32_t revision; | |
117 | le32_t header_size; | |
118 | le32_t crc32; | |
119 | le32_t reserved; | |
120 | le64_t my_lba; | |
121 | le64_t alternate_lba; | |
122 | le64_t first_usable_lba; | |
123 | le64_t last_usable_lba; | |
124 | sd_id128_t disk_guid; | |
125 | le64_t partition_entry_lba; | |
126 | le32_t number_of_partition_entries; | |
127 | le32_t size_of_partition_entry; | |
128 | le32_t partition_entry_array_crc32; | |
129 | } _packed_; | |
130 | ||
131 | /* Disk images might be for 512B or for 4096 sector sizes, let's try to auto-detect that by searching | |
132 | * for the GPT headers at the relevant byte offsets */ | |
133 | ||
134 | assert_cc(sizeof(struct gpt_header) == 92); | |
135 | ||
136 | /* We expect a sector size in the range 512…4096. The GPT header is located in the second | |
137 | * sector. Hence it could be at byte 512 at the earliest, and at byte 4096 at the latest. And we must | |
138 | * read with granularity of the largest sector size we care about. Which means 8K. */ | |
139 | uint8_t sectors[2 * 4096]; | |
140 | uint32_t found = 0; | |
141 | ssize_t n; | |
142 | ||
143 | assert(fd >= 0); | |
144 | assert(ret); | |
145 | ||
146 | n = pread(fd, sectors, sizeof(sectors), 0); | |
147 | if (n < 0) | |
148 | return -errno; | |
149 | if (n != sizeof(sectors)) /* too short? */ | |
150 | goto not_found; | |
151 | ||
152 | /* Let's see if we find the GPT partition header with various expected sector sizes */ | |
153 | for (uint32_t sz = 512; sz <= 4096; sz <<= 1) { | |
154 | struct gpt_header *p; | |
155 | ||
156 | assert(sizeof(sectors) >= sz * 2); | |
157 | p = (struct gpt_header*) (sectors + sz); | |
158 | ||
159 | if (memcmp(p->signature, (const char[8]) { 'E', 'F', 'I', ' ', 'P', 'A', 'R', 'T' }, 8) != 0) | |
160 | continue; | |
161 | ||
162 | if (le32toh(p->revision) != UINT32_C(0x00010000)) /* the only known revision of the spec: 1.0 */ | |
163 | continue; | |
164 | ||
165 | if (le32toh(p->header_size) < sizeof(struct gpt_header)) | |
166 | continue; | |
167 | ||
168 | if (le32toh(p->header_size) > 4096) /* larger than a sector? something is off… */ | |
169 | continue; | |
170 | ||
171 | if (le64toh(p->my_lba) != 1) /* this sector must claim to be at sector offset 1 */ | |
172 | continue; | |
173 | ||
174 | if (found != 0) | |
175 | return log_debug_errno(SYNTHETIC_ERRNO(ENOTUNIQ), | |
176 | "Detected valid partition table at offsets matching multiple sector sizes, refusing."); | |
177 | ||
178 | found = sz; | |
179 | } | |
180 | ||
181 | if (found != 0) { | |
182 | log_debug("Determined sector size %" PRIu32 " based on discovered partition table.", found); | |
183 | *ret = found; | |
184 | return 1; /* indicate we *did* find it */ | |
185 | } | |
186 | ||
187 | not_found: | |
188 | log_debug("Couldn't find any partition table to derive sector size of."); | |
189 | *ret = 512; /* pick the traditional default */ | |
190 | return 0; /* indicate we didn't find it */ | |
191 | } | |
192 | ||
81dde3d8 LP |
193 | int probe_sector_size_prefer_ioctl(int fd, uint32_t *ret) { |
194 | struct stat st; | |
195 | ||
196 | assert(fd >= 0); | |
197 | assert(ret); | |
198 | ||
199 | /* Just like probe_sector_size(), but if we are looking at a block device, will use the already | |
200 | * configured sector size rather than probing by contents */ | |
201 | ||
202 | if (fstat(fd, &st) < 0) | |
203 | return -errno; | |
204 | ||
205 | if (S_ISBLK(st.st_mode)) | |
206 | return blockdev_get_sector_size(fd, ret); | |
207 | ||
208 | return probe_sector_size(fd, ret); | |
209 | } | |
210 | ||
c80c9079 LP |
211 | int probe_filesystem_full( |
212 | int fd, | |
213 | const char *path, | |
214 | uint64_t offset, | |
215 | uint64_t size, | |
216 | char **ret_fstype) { | |
217 | ||
7cc84b2c | 218 | /* Try to find device content type and return it in *ret_fstype. If nothing is found, |
c80c9079 | 219 | * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and a |
7cc84b2c ZJS |
220 | * different error otherwise. */ |
221 | ||
349cc4a5 | 222 | #if HAVE_BLKID |
8e766630 | 223 | _cleanup_(blkid_free_probep) blkid_probe b = NULL; |
d2c6e79d | 224 | _cleanup_free_ char *path_by_fd = NULL; |
254d1313 | 225 | _cleanup_close_ int fd_close = -EBADF; |
18b5886e LP |
226 | const char *fstype; |
227 | int r; | |
228 | ||
d2c6e79d YW |
229 | assert(fd >= 0 || path); |
230 | assert(ret_fstype); | |
231 | ||
232 | if (fd < 0) { | |
233 | fd_close = open(path, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY); | |
234 | if (fd_close < 0) | |
235 | return -errno; | |
236 | ||
237 | fd = fd_close; | |
238 | } | |
239 | ||
240 | if (!path) { | |
241 | r = fd_get_path(fd, &path_by_fd); | |
242 | if (r < 0) | |
243 | return r; | |
244 | ||
245 | path = path_by_fd; | |
246 | } | |
247 | ||
c80c9079 LP |
248 | if (size == 0) /* empty size? nothing found! */ |
249 | goto not_found; | |
250 | ||
d2c6e79d | 251 | b = blkid_new_probe(); |
18b5886e | 252 | if (!b) |
d2c6e79d YW |
253 | return -ENOMEM; |
254 | ||
255 | errno = 0; | |
c80c9079 LP |
256 | r = blkid_probe_set_device( |
257 | b, | |
258 | fd, | |
259 | offset, | |
260 | size == UINT64_MAX ? 0 : size); /* when blkid sees size=0 it understands "everything". We prefer using UINT64_MAX for that */ | |
d2c6e79d | 261 | if (r != 0) |
66855de7 | 262 | return errno_or_else(ENOMEM); |
18b5886e LP |
263 | |
264 | blkid_probe_enable_superblocks(b, 1); | |
265 | blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE); | |
266 | ||
267 | errno = 0; | |
268 | r = blkid_do_safeprobe(b); | |
2e3944b8 | 269 | if (r == _BLKID_SAFEPROBE_NOT_FOUND) |
18b5886e | 270 | goto not_found; |
2e3944b8 | 271 | if (r == _BLKID_SAFEPROBE_AMBIGUOUS) |
58dfbfbd | 272 | return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN), |
d2c6e79d | 273 | "Results ambiguous for partition %s", path); |
2e3944b8 | 274 | if (r == _BLKID_SAFEPROBE_ERROR) |
d2c6e79d | 275 | return log_debug_errno(errno_or_else(EIO), "Failed to probe partition %s: %m", path); |
18b5886e | 276 | |
2e3944b8 LP |
277 | assert(r == _BLKID_SAFEPROBE_FOUND); |
278 | ||
18b5886e LP |
279 | (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL); |
280 | ||
281 | if (fstype) { | |
282 | char *t; | |
283 | ||
d2c6e79d | 284 | log_debug("Probed fstype '%s' on partition %s.", fstype, path); |
1cefb9a6 | 285 | |
18b5886e LP |
286 | t = strdup(fstype); |
287 | if (!t) | |
288 | return -ENOMEM; | |
289 | ||
290 | *ret_fstype = t; | |
291 | return 1; | |
292 | } | |
293 | ||
294 | not_found: | |
d2c6e79d | 295 | log_debug("No type detected on partition %s", path); |
18b5886e LP |
296 | *ret_fstype = NULL; |
297 | return 0; | |
d1c536f5 ZJS |
298 | #else |
299 | return -EOPNOTSUPP; | |
a75e27eb | 300 | #endif |
d1c536f5 | 301 | } |
18b5886e | 302 | |
40c10d3f | 303 | #if HAVE_BLKID |
c80c9079 | 304 | static int dissected_image_probe_filesystems(DissectedImage *m, int fd) { |
698bb074 YW |
305 | int r; |
306 | ||
307 | assert(m); | |
308 | ||
309 | /* Fill in file system types if we don't know them yet. */ | |
310 | ||
311 | for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) { | |
312 | DissectedPartition *p = m->partitions + i; | |
313 | ||
314 | if (!p->found) | |
315 | continue; | |
316 | ||
c80c9079 LP |
317 | if (!p->fstype) { |
318 | /* If we have an fd referring to the partition block device, use that. Otherwise go | |
319 | * via the whole block device or backing regular file, and read via offset. */ | |
320 | if (p->mount_node_fd >= 0) | |
321 | r = probe_filesystem_full(p->mount_node_fd, p->node, 0, UINT64_MAX, &p->fstype); | |
322 | else | |
323 | r = probe_filesystem_full(fd, p->node, p->offset, p->size, &p->fstype); | |
324 | if (r < 0) | |
698bb074 YW |
325 | return r; |
326 | } | |
327 | ||
328 | if (streq_ptr(p->fstype, "crypto_LUKS")) | |
329 | m->encrypted = true; | |
330 | ||
331 | if (p->fstype && fstype_is_ro(p->fstype)) | |
332 | p->rw = false; | |
333 | ||
334 | if (!p->rw) | |
335 | p->growfs = false; | |
336 | } | |
337 | ||
338 | return 0; | |
339 | } | |
340 | ||
0f7c9a3d LP |
341 | static void check_partition_flags( |
342 | const char *node, | |
343 | unsigned long long pflags, | |
344 | unsigned long long supported) { | |
345 | ||
346 | assert(node); | |
347 | ||
348 | /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */ | |
92e72028 ZJS |
349 | pflags &= ~(supported | |
350 | SD_GPT_FLAG_REQUIRED_PARTITION | | |
351 | SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL | | |
352 | SD_GPT_FLAG_LEGACY_BIOS_BOOTABLE); | |
0f7c9a3d LP |
353 | |
354 | if (pflags == 0) | |
355 | return; | |
356 | ||
357 | /* If there are other bits set, then log about it, to make things discoverable */ | |
358 | for (unsigned i = 0; i < sizeof(pflags) * 8; i++) { | |
359 | unsigned long long bit = 1ULL << i; | |
360 | if (!FLAGS_SET(pflags, bit)) | |
361 | continue; | |
362 | ||
363 | log_debug("Unexpected partition flag %llu set on %s!", bit, node); | |
364 | } | |
365 | } | |
1b010ae7 | 366 | #endif |
0f7c9a3d | 367 | |
00e29505 YW |
368 | #if HAVE_BLKID |
369 | static int dissected_image_new(const char *path, DissectedImage **ret) { | |
370 | _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL; | |
371 | _cleanup_free_ char *name = NULL; | |
372 | int r; | |
373 | ||
374 | assert(ret); | |
375 | ||
376 | if (path) { | |
377 | _cleanup_free_ char *filename = NULL; | |
378 | ||
379 | r = path_extract_filename(path, &filename); | |
380 | if (r < 0) | |
381 | return r; | |
382 | ||
383 | r = raw_strip_suffixes(filename, &name); | |
384 | if (r < 0) | |
385 | return r; | |
386 | ||
387 | if (!image_name_is_valid(name)) { | |
388 | log_debug("Image name %s is not valid, ignoring.", strna(name)); | |
389 | name = mfree(name); | |
390 | } | |
391 | } | |
392 | ||
393 | m = new(DissectedImage, 1); | |
394 | if (!m) | |
395 | return -ENOMEM; | |
396 | ||
397 | *m = (DissectedImage) { | |
398 | .has_init_system = -1, | |
399 | .image_name = TAKE_PTR(name), | |
400 | }; | |
401 | ||
babd5b08 YW |
402 | for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) |
403 | m->partitions[i] = DISSECTED_PARTITION_NULL; | |
404 | ||
00e29505 YW |
405 | *ret = TAKE_PTR(m); |
406 | return 0; | |
407 | } | |
408 | #endif | |
409 | ||
234c2e16 | 410 | static void dissected_partition_done(DissectedPartition *p) { |
1b010ae7 | 411 | assert(p); |
786e3a52 | 412 | |
1b010ae7 LP |
413 | free(p->fstype); |
414 | free(p->node); | |
415 | free(p->label); | |
416 | free(p->decrypted_fstype); | |
417 | free(p->decrypted_node); | |
418 | free(p->mount_options); | |
f7725647 | 419 | safe_close(p->mount_node_fd); |
786e3a52 | 420 | |
babd5b08 | 421 | *p = DISSECTED_PARTITION_NULL; |
1b010ae7 | 422 | } |
786e3a52 | 423 | |
1b010ae7 | 424 | #if HAVE_BLKID |
1b010ae7 LP |
425 | static int make_partition_devname( |
426 | const char *whole_devname, | |
1a81ddef | 427 | uint64_t diskseq, |
1b010ae7 | 428 | int nr, |
1a81ddef | 429 | DissectImageFlags flags, |
1b010ae7 | 430 | char **ret) { |
786e3a52 | 431 | |
1a81ddef LP |
432 | _cleanup_free_ char *s = NULL; |
433 | int r; | |
786e3a52 | 434 | |
1b010ae7 | 435 | assert(whole_devname); |
1a81ddef LP |
436 | assert(nr != 0); /* zero is not a valid partition nr */ |
437 | assert(ret); | |
aae22eb3 | 438 | |
c12a0d6d | 439 | if (!FLAGS_SET(flags, DISSECT_IMAGE_DISKSEQ_DEVNODE) || diskseq == 0) { |
4ba86848 | 440 | |
1a81ddef LP |
441 | /* Given a whole block device node name (e.g. /dev/sda or /dev/loop7) generate a partition |
442 | * device name (e.g. /dev/sda7 or /dev/loop7p5). The rule the kernel uses is simple: if whole | |
443 | * block device node name ends in a digit, then suffix a 'p', followed by the partition | |
444 | * number. Otherwise, just suffix the partition number without any 'p'. */ | |
08fe0a53 | 445 | |
1a81ddef LP |
446 | if (nr < 0) { /* whole disk? */ |
447 | s = strdup(whole_devname); | |
448 | if (!s) | |
449 | return -ENOMEM; | |
450 | } else { | |
451 | size_t l = strlen(whole_devname); | |
452 | if (l < 1) /* underflow check for the subtraction below */ | |
453 | return -EINVAL; | |
08fe0a53 | 454 | |
1a81ddef LP |
455 | bool need_p = ascii_isdigit(whole_devname[l-1]); /* Last char a digit? */ |
456 | ||
457 | if (asprintf(&s, "%s%s%i", whole_devname, need_p ? "p" : "", nr) < 0) | |
458 | return -ENOMEM; | |
459 | } | |
460 | } else { | |
461 | if (nr < 0) /* whole disk? */ | |
462 | r = asprintf(&s, "/dev/disk/by-diskseq/%" PRIu64, diskseq); | |
463 | else | |
464 | r = asprintf(&s, "/dev/disk/by-diskseq/%" PRIu64 "-part%i", diskseq, nr); | |
465 | if (r < 0) | |
466 | return -ENOMEM; | |
467 | } | |
468 | ||
469 | *ret = TAKE_PTR(s); | |
470 | return 0; | |
08fe0a53 LP |
471 | } |
472 | ||
1a81ddef LP |
473 | static int open_partition( |
474 | const char *node, | |
475 | bool is_partition, | |
476 | const LoopDevice *loop) { | |
477 | ||
f7725647 | 478 | _cleanup_(sd_device_unrefp) sd_device *dev = NULL; |
254d1313 | 479 | _cleanup_close_ int fd = -EBADF; |
f7725647 YW |
480 | dev_t devnum; |
481 | int r; | |
482 | ||
483 | assert(node); | |
484 | assert(loop); | |
485 | ||
486 | fd = open(node, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY); | |
487 | if (fd < 0) | |
488 | return -errno; | |
489 | ||
490 | /* Check if the block device is a child of (or equivalent to) the originally provided one. */ | |
491 | r = block_device_new_from_fd(fd, is_partition ? BLOCK_DEVICE_LOOKUP_WHOLE_DISK : 0, &dev); | |
492 | if (r < 0) | |
493 | return r; | |
494 | ||
495 | r = sd_device_get_devnum(dev, &devnum); | |
496 | if (r < 0) | |
497 | return r; | |
498 | ||
499 | if (loop->devno != devnum) | |
500 | return -ENXIO; | |
501 | ||
502 | /* Also check diskseq. */ | |
c12a0d6d | 503 | if (loop->diskseq != 0) { |
f7725647 YW |
504 | uint64_t diskseq; |
505 | ||
506 | r = fd_get_diskseq(fd, &diskseq); | |
507 | if (r < 0) | |
508 | return r; | |
509 | ||
510 | if (loop->diskseq != diskseq) | |
511 | return -ENXIO; | |
512 | } | |
513 | ||
514 | log_debug("Opened %s (fd=%i, whole_block_devnum=" DEVNUM_FORMAT_STR ", diskseq=%" PRIu64 ").", | |
515 | node, fd, DEVNUM_FORMAT_VAL(loop->devno), loop->diskseq); | |
516 | return TAKE_FD(fd); | |
517 | } | |
518 | ||
22e932f4 DDM |
519 | static int compare_arch(Architecture a, Architecture b) { |
520 | if (a == b) | |
521 | return 0; | |
522 | ||
523 | if (a == native_architecture()) | |
524 | return 1; | |
525 | ||
526 | if (b == native_architecture()) | |
527 | return -1; | |
528 | ||
529 | #ifdef ARCHITECTURE_SECONDARY | |
530 | if (a == ARCHITECTURE_SECONDARY) | |
531 | return 1; | |
532 | ||
533 | if (b == ARCHITECTURE_SECONDARY) | |
534 | return -1; | |
535 | #endif | |
536 | ||
537 | return 0; | |
538 | } | |
539 | ||
08f14be4 YW |
540 | static int dissect_image( |
541 | DissectedImage *m, | |
4526113f | 542 | int fd, |
0b214aa0 | 543 | const char *devname, |
89e62e0b | 544 | const VeritySettings *verity, |
18d73705 | 545 | const MountOptions *mount_options, |
08f14be4 | 546 | DissectImageFlags flags) { |
8c1be37e | 547 | |
62ea0ed0 | 548 | sd_id128_t root_uuid = SD_ID128_NULL, root_verity_uuid = SD_ID128_NULL; |
62ea0ed0 | 549 | sd_id128_t usr_uuid = SD_ID128_NULL, usr_verity_uuid = SD_ID128_NULL; |
1f8fb21c | 550 | bool is_gpt, is_mbr, multiple_generic = false, |
de98f631 LP |
551 | generic_rw = false, /* initialize to appease gcc */ |
552 | generic_growfs = false; | |
8e766630 | 553 | _cleanup_(blkid_free_probep) blkid_probe b = NULL; |
8c1be37e | 554 | _cleanup_free_ char *generic_node = NULL; |
be30ad41 | 555 | sd_id128_t generic_uuid = SD_ID128_NULL; |
b387778c | 556 | const char *pttype = NULL, *sptuuid = NULL; |
8c1be37e | 557 | blkid_partlist pl; |
1f8fb21c | 558 | int r, generic_nr = -1, n_partitions; |
8c1be37e | 559 | |
08f14be4 | 560 | assert(m); |
8c1be37e | 561 | assert(fd >= 0); |
0b214aa0 | 562 | assert(devname); |
a0bff7ea | 563 | assert(!verity || verity->designator < 0 || IN_SET(verity->designator, PARTITION_ROOT, PARTITION_USR)); |
89e62e0b | 564 | assert(!verity || verity->root_hash || verity->root_hash_size == 0); |
a0bff7ea LP |
565 | assert(!verity || verity->root_hash_sig || verity->root_hash_sig_size == 0); |
566 | assert(!verity || (verity->root_hash || !verity->root_hash_sig)); | |
e7cbe5cb | 567 | assert(!((flags & DISSECT_IMAGE_GPT_ONLY) && (flags & DISSECT_IMAGE_NO_PARTITION_TABLE))); |
22ee78a8 | 568 | assert(m->sector_size > 0); |
8c1be37e LP |
569 | |
570 | /* Probes a disk image, and returns information about what it found in *ret. | |
571 | * | |
4623e8e6 | 572 | * Returns -ENOPKG if no suitable partition table or file system could be found. |
2679f407 LP |
573 | * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found. |
574 | * Returns -ENXIO if we couldn't find any partition suitable as root or /usr partition | |
575 | * Returns -ENOTUNIQ if we only found multiple generic partitions and thus don't know what to do with that */ | |
4623e8e6 | 576 | |
c12a0d6d | 577 | uint64_t diskseq = m->loop ? m->loop->diskseq : 0; |
1a81ddef | 578 | |
89e62e0b | 579 | if (verity && verity->root_hash) { |
aee36b4e LP |
580 | sd_id128_t fsuuid, vuuid; |
581 | ||
582 | /* If a root hash is supplied, then we use the root partition that has a UUID that match the | |
583 | * first 128bit of the root hash. And we use the verity partition that has a UUID that match | |
584 | * the final 128bit. */ | |
4623e8e6 | 585 | |
89e62e0b | 586 | if (verity->root_hash_size < sizeof(sd_id128_t)) |
4623e8e6 LP |
587 | return -EINVAL; |
588 | ||
aee36b4e LP |
589 | memcpy(&fsuuid, verity->root_hash, sizeof(sd_id128_t)); |
590 | memcpy(&vuuid, (const uint8_t*) verity->root_hash + verity->root_hash_size - sizeof(sd_id128_t), sizeof(sd_id128_t)); | |
4623e8e6 | 591 | |
aee36b4e | 592 | if (sd_id128_is_null(fsuuid)) |
4623e8e6 | 593 | return -EINVAL; |
aee36b4e | 594 | if (sd_id128_is_null(vuuid)) |
4623e8e6 | 595 | return -EINVAL; |
aee36b4e LP |
596 | |
597 | /* If the verity data declares it's for the /usr partition, then search for that, in all | |
598 | * other cases assume it's for the root partition. */ | |
599 | if (verity->designator == PARTITION_USR) { | |
600 | usr_uuid = fsuuid; | |
601 | usr_verity_uuid = vuuid; | |
602 | } else { | |
603 | root_uuid = fsuuid; | |
604 | root_verity_uuid = vuuid; | |
605 | } | |
4623e8e6 | 606 | } |
8c1be37e | 607 | |
8c1be37e LP |
608 | b = blkid_new_probe(); |
609 | if (!b) | |
610 | return -ENOMEM; | |
611 | ||
612 | errno = 0; | |
613 | r = blkid_probe_set_device(b, fd, 0, 0); | |
b382db9f | 614 | if (r != 0) |
66855de7 | 615 | return errno_or_else(ENOMEM); |
8c1be37e | 616 | |
22ee78a8 LP |
617 | errno = 0; |
618 | r = blkid_probe_set_sectorsize(b, m->sector_size); | |
619 | if (r != 0) | |
620 | return errno_or_else(EIO); | |
621 | ||
9b6deb03 LP |
622 | if ((flags & DISSECT_IMAGE_GPT_ONLY) == 0) { |
623 | /* Look for file system superblocks, unless we only shall look for GPT partition tables */ | |
624 | blkid_probe_enable_superblocks(b, 1); | |
b387778c | 625 | blkid_probe_set_superblocks_flags(b, BLKID_SUBLKS_TYPE|BLKID_SUBLKS_USAGE|BLKID_SUBLKS_UUID); |
9b6deb03 LP |
626 | } |
627 | ||
8c1be37e LP |
628 | blkid_probe_enable_partitions(b, 1); |
629 | blkid_probe_set_partitions_flags(b, BLKID_PARTS_ENTRY_DETAILS); | |
630 | ||
631 | errno = 0; | |
632 | r = blkid_do_safeprobe(b); | |
2e3944b8 | 633 | if (r == _BLKID_SAFEPROBE_ERROR) |
66855de7 | 634 | return errno_or_else(EIO); |
2e3944b8 LP |
635 | if (IN_SET(r, _BLKID_SAFEPROBE_AMBIGUOUS, _BLKID_SAFEPROBE_NOT_FOUND)) |
636 | return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG), "Failed to identify any partition table."); | |
637 | ||
638 | assert(r == _BLKID_SAFEPROBE_FOUND); | |
8c1be37e | 639 | |
e7cbe5cb | 640 | if ((!(flags & DISSECT_IMAGE_GPT_ONLY) && |
4b5de5dd | 641 | (flags & DISSECT_IMAGE_GENERIC_ROOT)) || |
e7cbe5cb | 642 | (flags & DISSECT_IMAGE_NO_PARTITION_TABLE)) { |
9b6deb03 | 643 | const char *usage = NULL; |
8c1be37e | 644 | |
aee36b4e LP |
645 | /* If flags permit this, also allow using non-partitioned single-filesystem images */ |
646 | ||
9b6deb03 LP |
647 | (void) blkid_probe_lookup_value(b, "USAGE", &usage, NULL); |
648 | if (STRPTR_IN_SET(usage, "filesystem", "crypto")) { | |
18d73705 | 649 | _cleanup_free_ char *t = NULL, *n = NULL, *o = NULL; |
b387778c | 650 | const char *fstype = NULL, *options = NULL, *suuid = NULL; |
254d1313 | 651 | _cleanup_close_ int mount_node_fd = -EBADF; |
b387778c | 652 | sd_id128_t uuid = SD_ID128_NULL; |
f7725647 | 653 | |
73d88b80 | 654 | if (FLAGS_SET(flags, DISSECT_IMAGE_PIN_PARTITION_DEVICES)) { |
f7725647 YW |
655 | mount_node_fd = open_partition(devname, /* is_partition = */ false, m->loop); |
656 | if (mount_node_fd < 0) | |
657 | return mount_node_fd; | |
658 | } | |
8c1be37e | 659 | |
9b6deb03 LP |
660 | /* OK, we have found a file system, that's our root partition then. */ |
661 | (void) blkid_probe_lookup_value(b, "TYPE", &fstype, NULL); | |
b387778c | 662 | (void) blkid_probe_lookup_value(b, "UUID", &suuid, NULL); |
8c1be37e | 663 | |
9b6deb03 LP |
664 | if (fstype) { |
665 | t = strdup(fstype); | |
666 | if (!t) | |
667 | return -ENOMEM; | |
668 | } | |
669 | ||
b387778c LP |
670 | if (suuid) { |
671 | /* blkid will return FAT's serial number as UUID, hence it is quite possible | |
672 | * that parsing this will fail. We'll ignore the ID, since it's just too | |
673 | * short to be useful as tru identifier. */ | |
674 | r = sd_id128_from_string(suuid, &uuid); | |
675 | if (r < 0) | |
676 | log_debug_errno(r, "Failed to parse file system UUID '%s', ignoring: %m", suuid); | |
677 | } | |
678 | ||
1a81ddef LP |
679 | r = make_partition_devname(devname, diskseq, -1, flags, &n); |
680 | if (r < 0) | |
681 | return r; | |
6c544d14 | 682 | |
e7cbe5cb | 683 | m->single_file_system = true; |
c3c88d67 LP |
684 | m->encrypted = streq_ptr(fstype, "crypto_LUKS"); |
685 | ||
686 | m->has_verity = verity && verity->data_path; | |
c2534821 | 687 | m->verity_ready = verity_settings_data_covers(verity, PARTITION_ROOT); |
e7cbe5cb | 688 | |
8ee9615e | 689 | m->has_verity_sig = false; /* signature not embedded, must be specified */ |
c2534821 | 690 | m->verity_sig_ready = m->verity_ready && verity->root_hash_sig; |
8ee9615e | 691 | |
b387778c LP |
692 | m->image_uuid = uuid; |
693 | ||
f5215bc8 | 694 | options = mount_options_from_designator(mount_options, PARTITION_ROOT); |
18d73705 LB |
695 | if (options) { |
696 | o = strdup(options); | |
697 | if (!o) | |
698 | return -ENOMEM; | |
699 | } | |
700 | ||
9b6deb03 LP |
701 | m->partitions[PARTITION_ROOT] = (DissectedPartition) { |
702 | .found = true, | |
e0d53d52 | 703 | .rw = !m->verity_ready && !fstype_is_ro(fstype), |
9b6deb03 LP |
704 | .partno = -1, |
705 | .architecture = _ARCHITECTURE_INVALID, | |
1cc6c93a YW |
706 | .fstype = TAKE_PTR(t), |
707 | .node = TAKE_PTR(n), | |
18d73705 | 708 | .mount_options = TAKE_PTR(o), |
f7725647 | 709 | .mount_node_fd = TAKE_FD(mount_node_fd), |
88b3300f LP |
710 | .offset = 0, |
711 | .size = UINT64_MAX, | |
9b6deb03 | 712 | }; |
8c1be37e | 713 | |
9b6deb03 LP |
714 | return 0; |
715 | } | |
8c1be37e LP |
716 | } |
717 | ||
718 | (void) blkid_probe_lookup_value(b, "PTTYPE", &pttype, NULL); | |
719 | if (!pttype) | |
720 | return -ENOPKG; | |
721 | ||
722 | is_gpt = streq_ptr(pttype, "gpt"); | |
723 | is_mbr = streq_ptr(pttype, "dos"); | |
724 | ||
9b6deb03 | 725 | if (!is_gpt && ((flags & DISSECT_IMAGE_GPT_ONLY) || !is_mbr)) |
8c1be37e LP |
726 | return -ENOPKG; |
727 | ||
0903fd26 LP |
728 | /* We support external verity data partitions only if the image has no partition table */ |
729 | if (verity && verity->data_path) | |
730 | return -EBADR; | |
731 | ||
73d88b80 | 732 | if (FLAGS_SET(flags, DISSECT_IMAGE_ADD_PARTITION_DEVICES)) { |
08f14be4 YW |
733 | /* Safety check: refuse block devices that carry a partition table but for which the kernel doesn't |
734 | * do partition scanning. */ | |
735 | r = blockdev_partscan_enabled(fd); | |
736 | if (r < 0) | |
737 | return r; | |
738 | if (r == 0) | |
739 | return -EPROTONOSUPPORT; | |
740 | } | |
4ba86848 | 741 | |
b387778c LP |
742 | (void) blkid_probe_lookup_value(b, "PTUUID", &sptuuid, NULL); |
743 | if (sptuuid) { | |
744 | r = sd_id128_from_string(sptuuid, &m->image_uuid); | |
745 | if (r < 0) | |
746 | log_debug_errno(r, "Failed to parse partition table UUID '%s', ignoring: %m", sptuuid); | |
747 | } | |
748 | ||
8c1be37e LP |
749 | errno = 0; |
750 | pl = blkid_probe_get_partitions(b); | |
b382db9f | 751 | if (!pl) |
66855de7 | 752 | return errno_or_else(ENOMEM); |
8c1be37e | 753 | |
4ba86848 LP |
754 | errno = 0; |
755 | n_partitions = blkid_partlist_numof_partitions(pl); | |
756 | if (n_partitions < 0) | |
757 | return errno_or_else(EIO); | |
8c1be37e | 758 | |
4ba86848 | 759 | for (int i = 0; i < n_partitions; i++) { |
1b010ae7 | 760 | _cleanup_free_ char *node = NULL; |
9b6deb03 | 761 | unsigned long long pflags; |
88b3300f | 762 | blkid_loff_t start, size; |
8c1be37e | 763 | blkid_partition pp; |
8c1be37e LP |
764 | int nr; |
765 | ||
4ba86848 LP |
766 | errno = 0; |
767 | pp = blkid_partlist_get_partition(pl, i); | |
768 | if (!pp) | |
769 | return errno_or_else(EIO); | |
aae22eb3 | 770 | |
9b6deb03 | 771 | pflags = blkid_partition_get_flags(pp); |
8c1be37e | 772 | |
4ba86848 | 773 | errno = 0; |
8c1be37e LP |
774 | nr = blkid_partition_get_partno(pp); |
775 | if (nr < 0) | |
4ba86848 | 776 | return errno_or_else(EIO); |
8c1be37e | 777 | |
88b3300f LP |
778 | errno = 0; |
779 | start = blkid_partition_get_start(pp); | |
780 | if (start < 0) | |
781 | return errno_or_else(EIO); | |
782 | ||
783 | assert((uint64_t) start < UINT64_MAX/512); | |
784 | ||
785 | errno = 0; | |
786 | size = blkid_partition_get_size(pp); | |
787 | if (size < 0) | |
788 | return errno_or_else(EIO); | |
789 | ||
790 | assert((uint64_t) size < UINT64_MAX/512); | |
791 | ||
1a81ddef LP |
792 | /* While probing we need the non-diskseq device node name to access the thing, hence mask off |
793 | * DISSECT_IMAGE_DISKSEQ_DEVNODE. */ | |
794 | r = make_partition_devname(devname, diskseq, nr, flags & ~DISSECT_IMAGE_DISKSEQ_DEVNODE, &node); | |
1b010ae7 LP |
795 | if (r < 0) |
796 | return r; | |
797 | ||
798 | /* So here's the thing: after the main ("whole") block device popped up it might take a while | |
799 | * before the kernel fully probed the partition table. Waiting for that to finish is icky in | |
800 | * userspace. So here's what we do instead. We issue the BLKPG_ADD_PARTITION ioctl to add the | |
801 | * partition ourselves, racing against the kernel. Good thing is: if this call fails with | |
802 | * EBUSY then the kernel was quicker than us, and that's totally OK, the outcome is good for | |
803 | * us: the device node will exist. If OTOH our call was successful we won the race. Which is | |
804 | * also good as the outcome is the same: the partition block device exists, and we can use | |
805 | * it. | |
806 | * | |
807 | * Kernel returns EBUSY if there's already a partition by that number or an overlapping | |
808 | * partition already existent. */ | |
809 | ||
73d88b80 | 810 | if (FLAGS_SET(flags, DISSECT_IMAGE_ADD_PARTITION_DEVICES)) { |
08f14be4 YW |
811 | r = block_device_add_partition(fd, node, nr, (uint64_t) start * 512, (uint64_t) size * 512); |
812 | if (r < 0) { | |
813 | if (r != -EBUSY) | |
814 | return log_debug_errno(r, "BLKPG_ADD_PARTITION failed: %m"); | |
1b010ae7 | 815 | |
08f14be4 YW |
816 | log_debug_errno(r, "Kernel was quicker than us in adding partition %i.", nr); |
817 | } else | |
818 | log_debug("We were quicker than kernel in adding partition %i.", nr); | |
819 | } | |
1b010ae7 | 820 | |
8c1be37e | 821 | if (is_gpt) { |
e3b9a5ff | 822 | const char *fstype = NULL, *label; |
4623e8e6 | 823 | sd_id128_t type_id, id; |
22e932f4 | 824 | GptPartitionType type; |
de98f631 | 825 | bool rw = true, growfs = false; |
8c1be37e | 826 | |
e3b9a5ff LP |
827 | r = blkid_partition_get_uuid_id128(pp, &id); |
828 | if (r < 0) { | |
829 | log_debug_errno(r, "Failed to read partition UUID, ignoring: %m"); | |
4623e8e6 | 830 | continue; |
e3b9a5ff | 831 | } |
4623e8e6 | 832 | |
e3b9a5ff LP |
833 | r = blkid_partition_get_type_id128(pp, &type_id); |
834 | if (r < 0) { | |
835 | log_debug_errno(r, "Failed to read partition type UUID, ignoring: %m"); | |
8c1be37e | 836 | continue; |
e3b9a5ff | 837 | } |
8c1be37e | 838 | |
22e932f4 DDM |
839 | type = gpt_partition_type_from_uuid(type_id); |
840 | ||
08fe0a53 LP |
841 | label = blkid_partition_get_name(pp); /* libblkid returns NULL here if empty */ |
842 | ||
fb3921b8 LP |
843 | if (IN_SET(type.designator, |
844 | PARTITION_HOME, | |
845 | PARTITION_SRV, | |
846 | PARTITION_XBOOTLDR, | |
847 | PARTITION_TMP)) { | |
a48dd347 | 848 | |
92e72028 ZJS |
849 | check_partition_flags(node, pflags, |
850 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS); | |
0f7c9a3d | 851 | |
92e72028 | 852 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
a48dd347 LP |
853 | continue; |
854 | ||
92e72028 ZJS |
855 | rw = !(pflags & SD_GPT_FLAG_READ_ONLY); |
856 | growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS); | |
aee36b4e | 857 | |
22e932f4 | 858 | } else if (type.designator == PARTITION_ESP) { |
a48dd347 | 859 | |
92e72028 ZJS |
860 | /* Note that we don't check the SD_GPT_FLAG_NO_AUTO flag for the ESP, as it is |
861 | * not defined there. We instead check the SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as | |
aee36b4e LP |
862 | * recommended by the UEFI spec (See "12.3.3 Number and Location of System |
863 | * Partitions"). */ | |
a48dd347 | 864 | |
92e72028 | 865 | if (pflags & SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL) |
a48dd347 LP |
866 | continue; |
867 | ||
8c1be37e | 868 | fstype = "vfat"; |
a8c47660 | 869 | |
22e932f4 | 870 | } else if (type.designator == PARTITION_ROOT) { |
4623e8e6 | 871 | |
92e72028 ZJS |
872 | check_partition_flags(node, pflags, |
873 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS); | |
0f7c9a3d | 874 | |
92e72028 | 875 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
a48dd347 LP |
876 | continue; |
877 | ||
4623e8e6 LP |
878 | /* If a root ID is specified, ignore everything but the root id */ |
879 | if (!sd_id128_is_null(root_uuid) && !sd_id128_equal(root_uuid, id)) | |
880 | continue; | |
881 | ||
92e72028 ZJS |
882 | rw = !(pflags & SD_GPT_FLAG_READ_ONLY); |
883 | growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS); | |
aee36b4e | 884 | |
22e932f4 | 885 | } else if (type.designator == PARTITION_ROOT_VERITY) { |
4623e8e6 | 886 | |
92e72028 ZJS |
887 | check_partition_flags(node, pflags, |
888 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY); | |
0f7c9a3d | 889 | |
92e72028 | 890 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
a48dd347 LP |
891 | continue; |
892 | ||
c3c88d67 | 893 | m->has_verity = true; |
4623e8e6 | 894 | |
8ee9615e LP |
895 | /* If no verity configuration is specified, then don't do verity */ |
896 | if (!verity) | |
897 | continue; | |
898 | if (verity->designator >= 0 && verity->designator != PARTITION_ROOT) | |
899 | continue; | |
900 | ||
901 | /* If root hash is specified, then ignore everything but the root id */ | |
902 | if (!sd_id128_is_null(root_verity_uuid) && !sd_id128_equal(root_verity_uuid, id)) | |
4623e8e6 LP |
903 | continue; |
904 | ||
4623e8e6 | 905 | fstype = "DM_verity_hash"; |
4623e8e6 | 906 | rw = false; |
8ee9615e | 907 | |
22e932f4 | 908 | } else if (type.designator == PARTITION_ROOT_VERITY_SIG) { |
8ee9615e | 909 | |
92e72028 ZJS |
910 | check_partition_flags(node, pflags, |
911 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY); | |
8ee9615e | 912 | |
92e72028 | 913 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
8ee9615e LP |
914 | continue; |
915 | ||
916 | m->has_verity_sig = true; | |
917 | ||
8ee9615e LP |
918 | if (!verity) |
919 | continue; | |
920 | if (verity->designator >= 0 && verity->designator != PARTITION_ROOT) | |
921 | continue; | |
8ee9615e | 922 | |
8ee9615e | 923 | fstype = "verity_hash_signature"; |
4623e8e6 | 924 | rw = false; |
8ee9615e | 925 | |
22e932f4 | 926 | } else if (type.designator == PARTITION_USR) { |
aee36b4e | 927 | |
92e72028 ZJS |
928 | check_partition_flags(node, pflags, |
929 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS); | |
aee36b4e | 930 | |
92e72028 | 931 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
aee36b4e LP |
932 | continue; |
933 | ||
934 | /* If a usr ID is specified, ignore everything but the usr id */ | |
935 | if (!sd_id128_is_null(usr_uuid) && !sd_id128_equal(usr_uuid, id)) | |
936 | continue; | |
937 | ||
92e72028 ZJS |
938 | rw = !(pflags & SD_GPT_FLAG_READ_ONLY); |
939 | growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS); | |
aee36b4e | 940 | |
22e932f4 | 941 | } else if (type.designator == PARTITION_USR_VERITY) { |
aee36b4e | 942 | |
92e72028 ZJS |
943 | check_partition_flags(node, pflags, |
944 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY); | |
aee36b4e | 945 | |
92e72028 | 946 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
aee36b4e LP |
947 | continue; |
948 | ||
c3c88d67 | 949 | m->has_verity = true; |
aee36b4e | 950 | |
8ee9615e LP |
951 | if (!verity) |
952 | continue; | |
953 | if (verity->designator >= 0 && verity->designator != PARTITION_USR) | |
954 | continue; | |
955 | ||
956 | /* If usr hash is specified, then ignore everything but the usr id */ | |
957 | if (!sd_id128_is_null(usr_verity_uuid) && !sd_id128_equal(usr_verity_uuid, id)) | |
aee36b4e LP |
958 | continue; |
959 | ||
aee36b4e | 960 | fstype = "DM_verity_hash"; |
aee36b4e | 961 | rw = false; |
8ee9615e | 962 | |
22e932f4 | 963 | } else if (type.designator == PARTITION_USR_VERITY_SIG) { |
8ee9615e | 964 | |
92e72028 ZJS |
965 | check_partition_flags(node, pflags, |
966 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY); | |
8ee9615e | 967 | |
92e72028 | 968 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
8ee9615e LP |
969 | continue; |
970 | ||
971 | m->has_verity_sig = true; | |
972 | ||
8ee9615e LP |
973 | if (!verity) |
974 | continue; | |
975 | if (verity->designator >= 0 && verity->designator != PARTITION_USR) | |
976 | continue; | |
8ee9615e | 977 | |
8ee9615e | 978 | fstype = "verity_hash_signature"; |
aee36b4e | 979 | rw = false; |
8ee9615e | 980 | |
22e932f4 | 981 | } else if (type.designator == PARTITION_SWAP) { |
a48dd347 | 982 | |
92e72028 | 983 | check_partition_flags(node, pflags, SD_GPT_FLAG_NO_AUTO); |
0f7c9a3d | 984 | |
92e72028 | 985 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
a48dd347 LP |
986 | continue; |
987 | ||
41aca66b LP |
988 | /* Note: we don't set fstype = "swap" here, because we still need to probe if |
989 | * it might be encrypted (i.e. fstype "crypt_LUKS") or unencrypted | |
990 | * (i.e. fstype "swap"), and the only way to figure that out is via fstype | |
991 | * probing. */ | |
df4524cb | 992 | |
df655bf3 DDM |
993 | /* We don't have a designator for SD_GPT_LINUX_GENERIC so check the UUID instead. */ |
994 | } else if (sd_id128_equal(type.uuid, SD_GPT_LINUX_GENERIC)) { | |
8c1be37e | 995 | |
92e72028 ZJS |
996 | check_partition_flags(node, pflags, |
997 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS); | |
0f7c9a3d | 998 | |
92e72028 | 999 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
a48dd347 LP |
1000 | continue; |
1001 | ||
8c1be37e LP |
1002 | if (generic_node) |
1003 | multiple_generic = true; | |
1004 | else { | |
1005 | generic_nr = nr; | |
92e72028 ZJS |
1006 | generic_rw = !(pflags & SD_GPT_FLAG_READ_ONLY); |
1007 | generic_growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS); | |
be30ad41 | 1008 | generic_uuid = id; |
a5590886 | 1009 | generic_node = TAKE_PTR(node); |
8c1be37e | 1010 | } |
d4dffb85 | 1011 | |
22e932f4 | 1012 | } else if (type.designator == PARTITION_VAR) { |
d4dffb85 | 1013 | |
92e72028 ZJS |
1014 | check_partition_flags(node, pflags, |
1015 | SD_GPT_FLAG_NO_AUTO | SD_GPT_FLAG_READ_ONLY | SD_GPT_FLAG_GROWFS); | |
0f7c9a3d | 1016 | |
92e72028 | 1017 | if (pflags & SD_GPT_FLAG_NO_AUTO) |
d4dffb85 LP |
1018 | continue; |
1019 | ||
1020 | if (!FLAGS_SET(flags, DISSECT_IMAGE_RELAX_VAR_CHECK)) { | |
1021 | sd_id128_t var_uuid; | |
1022 | ||
1023 | /* For /var we insist that the uuid of the partition matches the | |
1024 | * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine | |
1025 | * ID. Why? Unlike the other partitions /var is inherently | |
1026 | * installation specific, hence we need to be careful not to mount it | |
1027 | * in the wrong installation. By hashing the partition UUID from | |
1028 | * /etc/machine-id we can securely bind the partition to the | |
1029 | * installation. */ | |
1030 | ||
92e72028 | 1031 | r = sd_id128_get_machine_app_specific(SD_GPT_VAR, &var_uuid); |
d4dffb85 LP |
1032 | if (r < 0) |
1033 | return r; | |
1034 | ||
1035 | if (!sd_id128_equal(var_uuid, id)) { | |
a52efa81 YW |
1036 | log_debug("Found a /var/ partition, but its UUID didn't match our expectations " |
1037 | "(found: " SD_ID128_UUID_FORMAT_STR ", expected: " SD_ID128_UUID_FORMAT_STR "), ignoring.", | |
1038 | SD_ID128_FORMAT_VAL(id), SD_ID128_FORMAT_VAL(var_uuid)); | |
d4dffb85 LP |
1039 | continue; |
1040 | } | |
1041 | } | |
1042 | ||
92e72028 ZJS |
1043 | rw = !(pflags & SD_GPT_FLAG_READ_ONLY); |
1044 | growfs = FLAGS_SET(pflags, SD_GPT_FLAG_GROWFS); | |
8c1be37e LP |
1045 | } |
1046 | ||
22e932f4 | 1047 | if (type.designator != _PARTITION_DESIGNATOR_INVALID) { |
1a81ddef | 1048 | _cleanup_free_ char *t = NULL, *o = NULL, *l = NULL, *n = NULL; |
254d1313 | 1049 | _cleanup_close_ int mount_node_fd = -EBADF; |
18d73705 | 1050 | const char *options = NULL; |
8c1be37e | 1051 | |
22e932f4 | 1052 | if (m->partitions[type.designator].found) { |
08fe0a53 LP |
1053 | /* For most partition types the first one we see wins. Except for the |
1054 | * rootfs and /usr, where we do a version compare of the label, and | |
1055 | * let the newest version win. This permits a simple A/B versioning | |
1056 | * scheme in OS images. */ | |
1057 | ||
22e932f4 DDM |
1058 | if (compare_arch(type.arch, m->partitions[type.designator].architecture) <= 0) |
1059 | continue; | |
1060 | ||
1061 | if (!partition_designator_is_versioned(type.designator) || | |
1062 | strverscmp_improved(m->partitions[type.designator].label, label) >= 0) | |
08fe0a53 LP |
1063 | continue; |
1064 | ||
22e932f4 | 1065 | dissected_partition_done(m->partitions + type.designator); |
08fe0a53 | 1066 | } |
8c1be37e | 1067 | |
73d88b80 | 1068 | if (FLAGS_SET(flags, DISSECT_IMAGE_PIN_PARTITION_DEVICES) && |
f52261a0 | 1069 | type.designator != PARTITION_SWAP) { |
f7725647 YW |
1070 | mount_node_fd = open_partition(node, /* is_partition = */ true, m->loop); |
1071 | if (mount_node_fd < 0) | |
1072 | return mount_node_fd; | |
1073 | } | |
1074 | ||
1a81ddef LP |
1075 | r = make_partition_devname(devname, diskseq, nr, flags, &n); |
1076 | if (r < 0) | |
1077 | return r; | |
1078 | ||
8c1be37e LP |
1079 | if (fstype) { |
1080 | t = strdup(fstype); | |
1081 | if (!t) | |
1082 | return -ENOMEM; | |
1083 | } | |
1084 | ||
08fe0a53 LP |
1085 | if (label) { |
1086 | l = strdup(label); | |
1087 | if (!l) | |
1088 | return -ENOMEM; | |
1089 | } | |
1090 | ||
22e932f4 | 1091 | options = mount_options_from_designator(mount_options, type.designator); |
18d73705 LB |
1092 | if (options) { |
1093 | o = strdup(options); | |
1094 | if (!o) | |
1095 | return -ENOMEM; | |
1096 | } | |
1097 | ||
22e932f4 | 1098 | m->partitions[type.designator] = (DissectedPartition) { |
8c1be37e LP |
1099 | .found = true, |
1100 | .partno = nr, | |
1101 | .rw = rw, | |
de98f631 | 1102 | .growfs = growfs, |
22e932f4 | 1103 | .architecture = type.arch, |
1a81ddef | 1104 | .node = TAKE_PTR(n), |
1cc6c93a | 1105 | .fstype = TAKE_PTR(t), |
08fe0a53 | 1106 | .label = TAKE_PTR(l), |
be30ad41 | 1107 | .uuid = id, |
18d73705 | 1108 | .mount_options = TAKE_PTR(o), |
f7725647 | 1109 | .mount_node_fd = TAKE_FD(mount_node_fd), |
88b3300f LP |
1110 | .offset = (uint64_t) start * 512, |
1111 | .size = (uint64_t) size * 512, | |
d90b03f8 | 1112 | .gpt_flags = pflags, |
8c1be37e | 1113 | }; |
8c1be37e LP |
1114 | } |
1115 | ||
1116 | } else if (is_mbr) { | |
1117 | ||
a8c47660 | 1118 | switch (blkid_partition_get_type(pp)) { |
8c1be37e | 1119 | |
a8c47660 LP |
1120 | case 0x83: /* Linux partition */ |
1121 | ||
1122 | if (pflags != 0x80) /* Bootable flag */ | |
1123 | continue; | |
8c1be37e | 1124 | |
a8c47660 LP |
1125 | if (generic_node) |
1126 | multiple_generic = true; | |
1127 | else { | |
1128 | generic_nr = nr; | |
1129 | generic_rw = true; | |
de98f631 | 1130 | generic_growfs = false; |
a5590886 | 1131 | generic_node = TAKE_PTR(node); |
a8c47660 LP |
1132 | } |
1133 | ||
1134 | break; | |
1135 | ||
1136 | case 0xEA: { /* Boot Loader Spec extended $BOOT partition */ | |
254d1313 | 1137 | _cleanup_close_ int mount_node_fd = -EBADF; |
1a81ddef | 1138 | _cleanup_free_ char *o = NULL, *n = NULL; |
a8c47660 | 1139 | sd_id128_t id = SD_ID128_NULL; |
e3b9a5ff | 1140 | const char *options = NULL; |
a8c47660 LP |
1141 | |
1142 | /* First one wins */ | |
234c2e16 | 1143 | if (m->partitions[PARTITION_XBOOTLDR].found) |
a8c47660 LP |
1144 | continue; |
1145 | ||
73d88b80 | 1146 | if (FLAGS_SET(flags, DISSECT_IMAGE_PIN_PARTITION_DEVICES)) { |
f7725647 YW |
1147 | mount_node_fd = open_partition(node, /* is_partition = */ true, m->loop); |
1148 | if (mount_node_fd < 0) | |
1149 | return mount_node_fd; | |
1150 | } | |
1151 | ||
e3b9a5ff | 1152 | (void) blkid_partition_get_uuid_id128(pp, &id); |
a8c47660 | 1153 | |
1a81ddef LP |
1154 | r = make_partition_devname(devname, diskseq, nr, flags, &n); |
1155 | if (r < 0) | |
1156 | return r; | |
1157 | ||
f5215bc8 | 1158 | options = mount_options_from_designator(mount_options, PARTITION_XBOOTLDR); |
18d73705 LB |
1159 | if (options) { |
1160 | o = strdup(options); | |
1161 | if (!o) | |
1162 | return -ENOMEM; | |
1163 | } | |
1164 | ||
a8c47660 LP |
1165 | m->partitions[PARTITION_XBOOTLDR] = (DissectedPartition) { |
1166 | .found = true, | |
1167 | .partno = nr, | |
1168 | .rw = true, | |
de98f631 | 1169 | .growfs = false, |
a8c47660 | 1170 | .architecture = _ARCHITECTURE_INVALID, |
1a81ddef | 1171 | .node = TAKE_PTR(n), |
a8c47660 | 1172 | .uuid = id, |
18d73705 | 1173 | .mount_options = TAKE_PTR(o), |
f7725647 | 1174 | .mount_node_fd = TAKE_FD(mount_node_fd), |
88b3300f LP |
1175 | .offset = (uint64_t) start * 512, |
1176 | .size = (uint64_t) size * 512, | |
a8c47660 LP |
1177 | }; |
1178 | ||
1179 | break; | |
1180 | }} | |
8c1be37e LP |
1181 | } |
1182 | } | |
1183 | ||
22e932f4 DDM |
1184 | if (!m->partitions[PARTITION_ROOT].found && |
1185 | (m->partitions[PARTITION_ROOT_VERITY].found || | |
1186 | m->partitions[PARTITION_ROOT_VERITY_SIG].found)) | |
1187 | return -EADDRNOTAVAIL; /* Verity found but no matching rootfs? Something is off, refuse. */ | |
49ae9d91 | 1188 | |
8ee9615e LP |
1189 | /* Hmm, we found a signature partition but no Verity data? Something is off. */ |
1190 | if (m->partitions[PARTITION_ROOT_VERITY_SIG].found && !m->partitions[PARTITION_ROOT_VERITY].found) | |
1191 | return -EADDRNOTAVAIL; | |
7cf66030 | 1192 | |
22e932f4 DDM |
1193 | if (!m->partitions[PARTITION_USR].found && |
1194 | (m->partitions[PARTITION_USR_VERITY].found || | |
1195 | m->partitions[PARTITION_USR_VERITY_SIG].found)) | |
1196 | return -EADDRNOTAVAIL; /* as above */ | |
7cf66030 | 1197 | |
22e932f4 | 1198 | /* as above */ |
8ee9615e LP |
1199 | if (m->partitions[PARTITION_USR_VERITY_SIG].found && !m->partitions[PARTITION_USR_VERITY].found) |
1200 | return -EADDRNOTAVAIL; | |
1201 | ||
cb241a69 LP |
1202 | /* If root and /usr are combined then insist that the architecture matches */ |
1203 | if (m->partitions[PARTITION_ROOT].found && | |
1204 | m->partitions[PARTITION_USR].found && | |
1205 | (m->partitions[PARTITION_ROOT].architecture >= 0 && | |
1206 | m->partitions[PARTITION_USR].architecture >= 0 && | |
1207 | m->partitions[PARTITION_ROOT].architecture != m->partitions[PARTITION_USR].architecture)) | |
1208 | return -EADDRNOTAVAIL; | |
1209 | ||
4ab51780 LP |
1210 | if (!m->partitions[PARTITION_ROOT].found && |
1211 | !m->partitions[PARTITION_USR].found && | |
1212 | (flags & DISSECT_IMAGE_GENERIC_ROOT) && | |
00a8b34f | 1213 | (!verity || !verity->root_hash || verity->designator != PARTITION_USR)) { |
7cf66030 | 1214 | |
1b010ae7 | 1215 | /* OK, we found nothing usable, then check if there's a single generic partition, and use |
4b5de5dd LP |
1216 | * that. If the root hash was set however, then we won't fall back to a generic node, because |
1217 | * the root hash decides. */ | |
7cf66030 LP |
1218 | |
1219 | /* If we didn't find a properly marked root partition, but we did find a single suitable | |
1220 | * generic Linux partition, then use this as root partition, if the caller asked for it. */ | |
1221 | if (multiple_generic) | |
1222 | return -ENOTUNIQ; | |
1223 | ||
4b5de5dd LP |
1224 | /* If we didn't find a generic node, then we can't fix this up either */ |
1225 | if (generic_node) { | |
254d1313 | 1226 | _cleanup_close_ int mount_node_fd = -EBADF; |
1a81ddef | 1227 | _cleanup_free_ char *o = NULL, *n = NULL; |
4b5de5dd | 1228 | const char *options; |
8c1be37e | 1229 | |
73d88b80 | 1230 | if (FLAGS_SET(flags, DISSECT_IMAGE_PIN_PARTITION_DEVICES)) { |
f7725647 YW |
1231 | mount_node_fd = open_partition(generic_node, /* is_partition = */ true, m->loop); |
1232 | if (mount_node_fd < 0) | |
1233 | return mount_node_fd; | |
1234 | } | |
1235 | ||
1a81ddef LP |
1236 | r = make_partition_devname(devname, diskseq, generic_nr, flags, &n); |
1237 | if (r < 0) | |
1238 | return r; | |
1239 | ||
f5215bc8 | 1240 | options = mount_options_from_designator(mount_options, PARTITION_ROOT); |
18d73705 LB |
1241 | if (options) { |
1242 | o = strdup(options); | |
1243 | if (!o) | |
1244 | return -ENOMEM; | |
1245 | } | |
1246 | ||
1f8fb21c | 1247 | assert(generic_nr >= 0); |
8c1be37e LP |
1248 | m->partitions[PARTITION_ROOT] = (DissectedPartition) { |
1249 | .found = true, | |
1250 | .rw = generic_rw, | |
de98f631 | 1251 | .growfs = generic_growfs, |
8c1be37e LP |
1252 | .partno = generic_nr, |
1253 | .architecture = _ARCHITECTURE_INVALID, | |
1a81ddef | 1254 | .node = TAKE_PTR(n), |
be30ad41 | 1255 | .uuid = generic_uuid, |
18d73705 | 1256 | .mount_options = TAKE_PTR(o), |
f7725647 | 1257 | .mount_node_fd = TAKE_FD(mount_node_fd), |
88b3300f LP |
1258 | .offset = UINT64_MAX, |
1259 | .size = UINT64_MAX, | |
8c1be37e | 1260 | }; |
e0f9e7bd | 1261 | } |
8c1be37e LP |
1262 | } |
1263 | ||
4b5de5dd LP |
1264 | /* 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 */ |
1265 | if (FLAGS_SET(flags, DISSECT_IMAGE_REQUIRE_ROOT) && | |
1266 | !(m->partitions[PARTITION_ROOT].found || (m->partitions[PARTITION_USR].found && FLAGS_SET(flags, DISSECT_IMAGE_USR_NO_ROOT)))) | |
1267 | return -ENXIO; | |
1268 | ||
7b32164f LP |
1269 | if (m->partitions[PARTITION_ROOT_VERITY].found) { |
1270 | /* We only support one verity partition per image, i.e. can't do for both /usr and root fs */ | |
1271 | if (m->partitions[PARTITION_USR_VERITY].found) | |
1272 | return -ENOTUNIQ; | |
1273 | ||
1274 | /* We don't support verity enabled root with a split out /usr. Neither with nor without | |
1275 | * verity there. (Note that we do support verity-less root with verity-full /usr, though.) */ | |
1276 | if (m->partitions[PARTITION_USR].found) | |
1277 | return -EADDRNOTAVAIL; | |
1278 | } | |
aee36b4e | 1279 | |
1903defc LP |
1280 | if (verity) { |
1281 | /* If a verity designator is specified, then insist that the matching partition exists */ | |
1282 | if (verity->designator >= 0 && !m->partitions[verity->designator].found) | |
1283 | return -EADDRNOTAVAIL; | |
aee36b4e | 1284 | |
3dd73ea7 DDM |
1285 | bool have_verity_sig_partition = |
1286 | m->partitions[verity->designator == PARTITION_USR ? PARTITION_USR_VERITY_SIG : PARTITION_ROOT_VERITY_SIG].found; | |
1287 | ||
1903defc | 1288 | if (verity->root_hash) { |
8ee9615e LP |
1289 | /* If we have an explicit root hash and found the partitions for it, then we are ready to use |
1290 | * Verity, set things up for it */ | |
1291 | ||
1903defc LP |
1292 | if (verity->designator < 0 || verity->designator == PARTITION_ROOT) { |
1293 | if (!m->partitions[PARTITION_ROOT_VERITY].found || !m->partitions[PARTITION_ROOT].found) | |
1294 | return -EADDRNOTAVAIL; | |
4623e8e6 | 1295 | |
1903defc LP |
1296 | /* If we found a verity setup, then the root partition is necessarily read-only. */ |
1297 | m->partitions[PARTITION_ROOT].rw = false; | |
1298 | m->verity_ready = true; | |
1903defc | 1299 | |
f9e0bb61 LP |
1300 | } else { |
1301 | assert(verity->designator == PARTITION_USR); | |
1302 | ||
1903defc LP |
1303 | if (!m->partitions[PARTITION_USR_VERITY].found || !m->partitions[PARTITION_USR].found) |
1304 | return -EADDRNOTAVAIL; | |
4623e8e6 | 1305 | |
1903defc LP |
1306 | m->partitions[PARTITION_USR].rw = false; |
1307 | m->verity_ready = true; | |
1308 | } | |
8ee9615e LP |
1309 | |
1310 | if (m->verity_ready) | |
3dd73ea7 | 1311 | m->verity_sig_ready = verity->root_hash_sig || have_verity_sig_partition; |
8ee9615e | 1312 | |
3dd73ea7 | 1313 | } else if (have_verity_sig_partition) { |
8ee9615e LP |
1314 | |
1315 | /* If we found an embedded signature partition, we are ready, too. */ | |
1316 | ||
1317 | m->verity_ready = m->verity_sig_ready = true; | |
1318 | m->partitions[verity->designator == PARTITION_USR ? PARTITION_USR : PARTITION_ROOT].rw = false; | |
aee36b4e | 1319 | } |
4623e8e6 LP |
1320 | } |
1321 | ||
c80c9079 LP |
1322 | r = dissected_image_probe_filesystems(m, fd); |
1323 | if (r < 0) | |
1324 | return r; | |
1325 | ||
08f14be4 YW |
1326 | return 0; |
1327 | } | |
1328 | #endif | |
18b5886e | 1329 | |
08f14be4 YW |
1330 | int dissect_image_file( |
1331 | const char *path, | |
1332 | const VeritySettings *verity, | |
1333 | const MountOptions *mount_options, | |
1334 | DissectImageFlags flags, | |
1335 | DissectedImage **ret) { | |
8c1be37e | 1336 | |
08f14be4 YW |
1337 | #if HAVE_BLKID |
1338 | _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL; | |
254d1313 | 1339 | _cleanup_close_ int fd = -EBADF; |
08f14be4 | 1340 | int r; |
8c1be37e | 1341 | |
08f14be4 | 1342 | assert(path); |
08f14be4 | 1343 | assert(ret); |
8c1be37e | 1344 | |
08f14be4 YW |
1345 | fd = open(path, O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY); |
1346 | if (fd < 0) | |
1347 | return -errno; | |
896f937f | 1348 | |
08f14be4 YW |
1349 | r = fd_verify_regular(fd); |
1350 | if (r < 0) | |
1351 | return r; | |
de98f631 | 1352 | |
08f14be4 YW |
1353 | r = dissected_image_new(path, &m); |
1354 | if (r < 0) | |
1355 | return r; | |
1356 | ||
22ee78a8 LP |
1357 | r = probe_sector_size(fd, &m->sector_size); |
1358 | if (r < 0) | |
1359 | return r; | |
1360 | ||
08f14be4 | 1361 | r = dissect_image(m, fd, path, verity, mount_options, flags); |
698bb074 YW |
1362 | if (r < 0) |
1363 | return r; | |
8c1be37e | 1364 | |
1cc6c93a | 1365 | *ret = TAKE_PTR(m); |
8c1be37e LP |
1366 | return 0; |
1367 | #else | |
1368 | return -EOPNOTSUPP; | |
1369 | #endif | |
1370 | } | |
1371 | ||
1372 | DissectedImage* dissected_image_unref(DissectedImage *m) { | |
8c1be37e LP |
1373 | if (!m) |
1374 | return NULL; | |
1375 | ||
ac1e1b5f | 1376 | /* First, clear dissected partitions. */ |
08fe0a53 | 1377 | for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) |
234c2e16 | 1378 | dissected_partition_done(m->partitions + i); |
8c1be37e | 1379 | |
ac1e1b5f YW |
1380 | /* Second, free decrypted images. This must be after dissected_partition_done(), as freeing |
1381 | * DecryptedImage may try to deactivate partitions. */ | |
1382 | decrypted_image_unref(m->decrypted_image); | |
1383 | ||
1e63dc4f YW |
1384 | /* Third, unref LoopDevice. This must be called after the above two, as freeing LoopDevice may try to |
1385 | * remove existing partitions on the loopback block device. */ | |
1386 | loop_device_unref(m->loop); | |
1387 | ||
593fe6c0 | 1388 | free(m->image_name); |
3b925504 LP |
1389 | free(m->hostname); |
1390 | strv_free(m->machine_info); | |
1391 | strv_free(m->os_release); | |
fab22946 | 1392 | strv_free(m->initrd_release); |
7718ac97 | 1393 | strv_free(m->extension_release); |
3b925504 | 1394 | |
5fecf46d | 1395 | return mfree(m); |
8c1be37e LP |
1396 | } |
1397 | ||
18b5886e | 1398 | static int is_loop_device(const char *path) { |
553e15f2 | 1399 | char s[SYS_BLOCK_PATH_MAX("/../loop/")]; |
18b5886e LP |
1400 | struct stat st; |
1401 | ||
1402 | assert(path); | |
1403 | ||
1404 | if (stat(path, &st) < 0) | |
1405 | return -errno; | |
1406 | ||
1407 | if (!S_ISBLK(st.st_mode)) | |
1408 | return -ENOTBLK; | |
1409 | ||
553e15f2 | 1410 | xsprintf_sys_block_path(s, "/loop/", st.st_dev); |
18b5886e LP |
1411 | if (access(s, F_OK) < 0) { |
1412 | if (errno != ENOENT) | |
1413 | return -errno; | |
1414 | ||
1415 | /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */ | |
553e15f2 | 1416 | xsprintf_sys_block_path(s, "/../loop/", st.st_dev); |
18b5886e LP |
1417 | if (access(s, F_OK) < 0) |
1418 | return errno == ENOENT ? false : -errno; | |
1419 | } | |
1420 | ||
1421 | return true; | |
1422 | } | |
1423 | ||
f8ab7812 | 1424 | static int run_fsck(int node_fd, const char *fstype) { |
cf32c486 LP |
1425 | int r, exit_status; |
1426 | pid_t pid; | |
1427 | ||
f8ab7812 | 1428 | assert(node_fd >= 0); |
cf32c486 LP |
1429 | assert(fstype); |
1430 | ||
13556724 | 1431 | r = fsck_exists_for_fstype(fstype); |
cf32c486 LP |
1432 | if (r < 0) { |
1433 | log_debug_errno(r, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype); | |
1434 | return 0; | |
1435 | } | |
1436 | if (r == 0) { | |
f8ab7812 | 1437 | log_debug("Not checking partition %s, as fsck for %s does not exist.", FORMAT_PROC_FD_PATH(node_fd), fstype); |
cf32c486 LP |
1438 | return 0; |
1439 | } | |
1440 | ||
f8ab7812 LP |
1441 | r = safe_fork_full( |
1442 | "(fsck)", | |
911f8f01 | 1443 | NULL, |
f8ab7812 | 1444 | &node_fd, 1, /* Leave the node fd open */ |
911f8f01 | 1445 | FORK_RESET_SIGNALS|FORK_CLOSE_ALL_FDS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG|FORK_REARRANGE_STDIO|FORK_CLOEXEC_OFF, |
f8ab7812 | 1446 | &pid); |
cf32c486 LP |
1447 | if (r < 0) |
1448 | return log_debug_errno(r, "Failed to fork off fsck: %m"); | |
1449 | if (r == 0) { | |
1450 | /* Child */ | |
f8ab7812 | 1451 | execl("/sbin/fsck", "/sbin/fsck", "-aT", FORMAT_PROC_FD_PATH(node_fd), NULL); |
7e0ed2e9 | 1452 | log_open(); |
cf32c486 LP |
1453 | log_debug_errno(errno, "Failed to execl() fsck: %m"); |
1454 | _exit(FSCK_OPERATIONAL_ERROR); | |
1455 | } | |
1456 | ||
1457 | exit_status = wait_for_terminate_and_check("fsck", pid, 0); | |
1458 | if (exit_status < 0) | |
1459 | return log_debug_errno(exit_status, "Failed to fork off /sbin/fsck: %m"); | |
1460 | ||
1461 | if ((exit_status & ~FSCK_ERROR_CORRECTED) != FSCK_SUCCESS) { | |
1462 | log_debug("fsck failed with exit status %i.", exit_status); | |
1463 | ||
1464 | if ((exit_status & (FSCK_SYSTEM_SHOULD_REBOOT|FSCK_ERRORS_LEFT_UNCORRECTED)) != 0) | |
1465 | return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN), "File system is corrupted, refusing."); | |
1466 | ||
1467 | log_debug("Ignoring fsck error."); | |
1468 | } | |
1469 | ||
1470 | return 0; | |
1471 | } | |
1472 | ||
81939d9d | 1473 | static int fs_grow(const char *node_path, const char *mount_path) { |
254d1313 | 1474 | _cleanup_close_ int mount_fd = -EBADF, node_fd = -EBADF; |
81939d9d LP |
1475 | uint64_t size, newsize; |
1476 | int r; | |
1477 | ||
1478 | node_fd = open(node_path, O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_NOCTTY); | |
1479 | if (node_fd < 0) | |
1480 | return log_debug_errno(errno, "Failed to open node device %s: %m", node_path); | |
1481 | ||
1482 | if (ioctl(node_fd, BLKGETSIZE64, &size) != 0) | |
1483 | return log_debug_errno(errno, "Failed to get block device size of %s: %m", node_path); | |
1484 | ||
1485 | mount_fd = open(mount_path, O_RDONLY|O_DIRECTORY|O_CLOEXEC); | |
1486 | if (mount_fd < 0) | |
1487 | return log_debug_errno(errno, "Failed to open mountd file system %s: %m", mount_path); | |
1488 | ||
1489 | log_debug("Resizing \"%s\" to %"PRIu64" bytes...", mount_path, size); | |
1490 | r = resize_fs(mount_fd, size, &newsize); | |
1491 | if (r < 0) | |
1492 | return log_debug_errno(r, "Failed to resize \"%s\" to %"PRIu64" bytes: %m", mount_path, size); | |
1493 | ||
1494 | if (newsize == size) | |
1495 | log_debug("Successfully resized \"%s\" to %s bytes.", | |
2b59bf51 | 1496 | mount_path, FORMAT_BYTES(newsize)); |
81939d9d LP |
1497 | else { |
1498 | assert(newsize < size); | |
1499 | log_debug("Successfully resized \"%s\" to %s bytes (%"PRIu64" bytes lost due to blocksize).", | |
2b59bf51 | 1500 | mount_path, FORMAT_BYTES(newsize), size - newsize); |
81939d9d LP |
1501 | } |
1502 | ||
1503 | return 0; | |
1504 | } | |
1505 | ||
254e392e LP |
1506 | int partition_pick_mount_options( |
1507 | PartitionDesignator d, | |
1508 | const char *fstype, | |
1509 | bool rw, | |
1510 | bool discard, | |
1511 | char **ret_options, | |
1512 | unsigned long *ret_ms_flags) { | |
1513 | ||
1514 | _cleanup_free_ char *options = NULL; | |
1515 | ||
1516 | assert(ret_options); | |
1517 | ||
1518 | /* Selects a baseline of bind mount flags, that should always apply. | |
1519 | * | |
1520 | * Firstly, we set MS_NODEV universally on all mounts, since we don't want to allow device nodes outside of /dev/. | |
1521 | * | |
1522 | * On /var/tmp/ we'll also set MS_NOSUID, same as we set for /tmp/ on the host. | |
1523 | * | |
1524 | * On the ESP and XBOOTLDR partitions we'll also disable symlinks, and execution. These file systems | |
1525 | * are generally untrusted (i.e. not encrypted or authenticated), and typically VFAT hence we should | |
1526 | * be as restrictive as possible, and this shouldn't hurt, since the functionality is not available | |
1527 | * there anyway. */ | |
1528 | ||
1529 | unsigned long flags = MS_NODEV; | |
1530 | ||
1531 | if (!rw) | |
1532 | flags |= MS_RDONLY; | |
1533 | ||
1534 | switch (d) { | |
1535 | ||
1536 | case PARTITION_ESP: | |
1537 | case PARTITION_XBOOTLDR: | |
1538 | flags |= MS_NOSUID|MS_NOEXEC|ms_nosymfollow_supported(); | |
1539 | ||
6eda6f7e LP |
1540 | /* The ESP might contain a pre-boot random seed. Let's make this unaccessible to regular |
1541 | * userspace. ESP/XBOOTLDR is almost certainly VFAT, hence if we don't know assume it is. */ | |
1542 | if (!fstype || fstype_can_umask(fstype)) | |
254e392e LP |
1543 | if (!strextend_with_separator(&options, ",", "umask=0077")) |
1544 | return -ENOMEM; | |
1545 | break; | |
1546 | ||
1547 | case PARTITION_TMP: | |
1548 | flags |= MS_NOSUID; | |
1549 | break; | |
1550 | ||
1551 | default: | |
1552 | break; | |
1553 | } | |
1554 | ||
1555 | /* So, when you request MS_RDONLY from ext4, then this means nothing. It happily still writes to the | |
1556 | * backing storage. What's worse, the BLKRO[GS]ET flag and (in case of loopback devices) | |
1557 | * LO_FLAGS_READ_ONLY don't mean anything, they affect userspace accesses only, and write accesses | |
1558 | * from the upper file system still get propagated through to the underlying file system, | |
1559 | * unrestricted. To actually get ext4/xfs/btrfs to stop writing to the device we need to specify | |
1560 | * "norecovery" as mount option, in addition to MS_RDONLY. Yes, this sucks, since it means we need to | |
1561 | * carry a per file system table here. | |
1562 | * | |
1563 | * Note that this means that we might not be able to mount corrupted file systems as read-only | |
1564 | * anymore (since in some cases the kernel implementations will refuse mounting when corrupted, | |
1565 | * read-only and "norecovery" is specified). But I think for the case of automatically determined | |
1566 | * mount options for loopback devices this is the right choice, since otherwise using the same | |
1567 | * loopback file twice even in read-only mode, is going to fail badly sooner or later. The usecase of | |
1568 | * making reuse of the immutable images "just work" is more relevant to us than having read-only | |
1569 | * access that actually modifies stuff work on such image files. Or to say this differently: if | |
1570 | * people want their file systems to be fixed up they should just open them in writable mode, where | |
1571 | * all these problems don't exist. */ | |
034ebc47 | 1572 | if (!rw && fstype && fstype_can_norecovery(fstype)) |
254e392e LP |
1573 | if (!strextend_with_separator(&options, ",", "norecovery")) |
1574 | return -ENOMEM; | |
1575 | ||
1576 | if (discard && fstype && fstype_can_discard(fstype)) | |
1577 | if (!strextend_with_separator(&options, ",", "discard")) | |
1578 | return -ENOMEM; | |
1579 | ||
1580 | if (!ret_ms_flags) /* Fold flags into option string if ret_flags specified as NULL */ | |
1581 | if (!strextend_with_separator(&options, ",", | |
1582 | FLAGS_SET(flags, MS_RDONLY) ? "ro" : "rw", | |
1583 | FLAGS_SET(flags, MS_NODEV) ? "nodev" : "dev", | |
1584 | FLAGS_SET(flags, MS_NOSUID) ? "nosuid" : "suid", | |
1585 | FLAGS_SET(flags, MS_NOEXEC) ? "noexec" : "exec", | |
1586 | FLAGS_SET(flags, MS_NOSYMFOLLOW) ? "nosymfollow" : NULL)) | |
1587 | /* NB: we suppress 'symfollow' here, since it's the default, and old /bin/mount might not know it */ | |
1588 | return -ENOMEM; | |
1589 | ||
1590 | if (ret_ms_flags) | |
1591 | *ret_ms_flags = flags; | |
1592 | ||
1593 | *ret_options = TAKE_PTR(options); | |
1594 | return 0; | |
1595 | } | |
1596 | ||
18b5886e | 1597 | static int mount_partition( |
254e392e | 1598 | PartitionDesignator d, |
18b5886e LP |
1599 | DissectedPartition *m, |
1600 | const char *where, | |
1601 | const char *directory, | |
2d3a5a73 | 1602 | uid_t uid_shift, |
21b61b1d | 1603 | uid_t uid_range, |
18b5886e LP |
1604 | DissectImageFlags flags) { |
1605 | ||
2d3a5a73 | 1606 | _cleanup_free_ char *chased = NULL, *options = NULL; |
254e392e | 1607 | bool rw, discard, remap_uid_gid = false; |
2d3a5a73 | 1608 | const char *p, *node, *fstype; |
254e392e | 1609 | unsigned long ms_flags; |
2eedfd2d | 1610 | int r; |
8c1be37e LP |
1611 | |
1612 | assert(m); | |
1613 | assert(where); | |
1614 | ||
f7725647 YW |
1615 | if (m->mount_node_fd < 0) |
1616 | return 0; | |
1617 | ||
4dc28665 | 1618 | /* Use decrypted node and matching fstype if available, otherwise use the original device */ |
f7725647 | 1619 | node = FORMAT_PROC_FD_PATH(m->mount_node_fd); |
c571dee7 | 1620 | fstype = dissected_partition_fstype(m); |
18b5886e | 1621 | |
4dc28665 LP |
1622 | if (!fstype) |
1623 | return -EAFNOSUPPORT; | |
80ce8580 LP |
1624 | r = dissect_fstype_ok(fstype); |
1625 | if (r < 0) | |
1626 | return r; | |
1627 | if (!r) | |
1628 | return -EIDRM; /* Recognizable error */ | |
8c1be37e | 1629 | |
68ac5118 ZJS |
1630 | /* We are looking at an encrypted partition? This either means stacked encryption, or the caller |
1631 | * didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error for this | |
1632 | * case. */ | |
4dc28665 | 1633 | if (streq(fstype, "crypto_LUKS")) |
fa45d12c | 1634 | return -EUNATCH; |
18b5886e | 1635 | |
ef9c184d | 1636 | rw = m->rw && !(flags & DISSECT_IMAGE_MOUNT_READ_ONLY); |
8c1be37e | 1637 | |
254e392e LP |
1638 | discard = ((flags & DISSECT_IMAGE_DISCARD) || |
1639 | ((flags & DISSECT_IMAGE_DISCARD_ON_LOOP) && is_loop_device(m->node) > 0)); | |
1640 | ||
cf32c486 | 1641 | if (FLAGS_SET(flags, DISSECT_IMAGE_FSCK) && rw) { |
f8ab7812 | 1642 | r = run_fsck(m->mount_node_fd, fstype); |
cf32c486 LP |
1643 | if (r < 0) |
1644 | return r; | |
1645 | } | |
1646 | ||
2eedfd2d | 1647 | if (directory) { |
334eb5b0 LP |
1648 | /* Automatically create missing mount points inside the image, if necessary. */ |
1649 | r = mkdir_p_root(where, directory, uid_shift, (gid_t) uid_shift, 0755); | |
1650 | if (r < 0 && r != -EROFS) | |
1651 | return r; | |
1f0f82f1 | 1652 | |
a5648b80 | 1653 | r = chase_symlinks(directory, where, CHASE_PREFIX_ROOT, &chased, NULL); |
2eedfd2d LP |
1654 | if (r < 0) |
1655 | return r; | |
1656 | ||
1657 | p = chased; | |
9842905e LP |
1658 | } else { |
1659 | /* Create top-level mount if missing – but only if this is asked for. This won't modify the | |
1660 | * image (as the branch above does) but the host hierarchy, and the created directory might | |
1661 | * survive our mount in the host hierarchy hence. */ | |
1662 | if (FLAGS_SET(flags, DISSECT_IMAGE_MKDIR)) { | |
1663 | r = mkdir_p(where, 0755); | |
1664 | if (r < 0) | |
1665 | return r; | |
1666 | } | |
1667 | ||
8c1be37e | 1668 | p = where; |
9842905e | 1669 | } |
8c1be37e | 1670 | |
254e392e LP |
1671 | r = partition_pick_mount_options(d, dissected_partition_fstype(m), rw, discard, &options, &ms_flags); |
1672 | if (r < 0) | |
1673 | return r; | |
2d3a5a73 | 1674 | |
21b61b1d | 1675 | if (uid_is_valid(uid_shift) && uid_shift != 0) { |
2d3a5a73 | 1676 | |
21b61b1d LP |
1677 | if (fstype_can_uid_gid(fstype)) { |
1678 | _cleanup_free_ char *uid_option = NULL; | |
2d3a5a73 | 1679 | |
21b61b1d LP |
1680 | if (asprintf(&uid_option, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0) |
1681 | return -ENOMEM; | |
1682 | ||
1683 | if (!strextend_with_separator(&options, ",", uid_option)) | |
1684 | return -ENOMEM; | |
1685 | } else if (FLAGS_SET(flags, DISSECT_IMAGE_MOUNT_IDMAPPED)) | |
1686 | remap_uid_gid = true; | |
2d3a5a73 | 1687 | } |
8c1be37e | 1688 | |
18d73705 | 1689 | if (!isempty(m->mount_options)) |
c2bc710b | 1690 | if (!strextend_with_separator(&options, ",", m->mount_options)) |
18d73705 LB |
1691 | return -ENOMEM; |
1692 | ||
254e392e | 1693 | r = mount_nofollow_verbose(LOG_DEBUG, node, p, fstype, ms_flags, options); |
d9223c07 LP |
1694 | if (r < 0) |
1695 | return r; | |
1696 | ||
81939d9d LP |
1697 | if (rw && m->growfs && FLAGS_SET(flags, DISSECT_IMAGE_GROWFS)) |
1698 | (void) fs_grow(node, p); | |
1699 | ||
21b61b1d | 1700 | if (remap_uid_gid) { |
2b2777ed | 1701 | r = remount_idmap(p, uid_shift, uid_range, UID_INVALID, REMOUNT_IDMAPPING_HOST_ROOT); |
21b61b1d LP |
1702 | if (r < 0) |
1703 | return r; | |
1704 | } | |
1705 | ||
d9223c07 | 1706 | return 1; |
8c1be37e LP |
1707 | } |
1708 | ||
7cf66030 LP |
1709 | static int mount_root_tmpfs(const char *where, uid_t uid_shift, DissectImageFlags flags) { |
1710 | _cleanup_free_ char *options = NULL; | |
1711 | int r; | |
1712 | ||
1713 | assert(where); | |
1714 | ||
1715 | /* For images that contain /usr/ but no rootfs, let's mount rootfs as tmpfs */ | |
1716 | ||
1717 | if (FLAGS_SET(flags, DISSECT_IMAGE_MKDIR)) { | |
1718 | r = mkdir_p(where, 0755); | |
1719 | if (r < 0) | |
1720 | return r; | |
1721 | } | |
1722 | ||
1723 | if (uid_is_valid(uid_shift)) { | |
1724 | if (asprintf(&options, "uid=" UID_FMT ",gid=" GID_FMT, uid_shift, (gid_t) uid_shift) < 0) | |
1725 | return -ENOMEM; | |
1726 | } | |
1727 | ||
1728 | r = mount_nofollow_verbose(LOG_DEBUG, "rootfs", where, "tmpfs", MS_NODEV, options); | |
1729 | if (r < 0) | |
1730 | return r; | |
1731 | ||
1732 | return 1; | |
1733 | } | |
1734 | ||
21b61b1d LP |
1735 | int dissected_image_mount( |
1736 | DissectedImage *m, | |
1737 | const char *where, | |
1738 | uid_t uid_shift, | |
1739 | uid_t uid_range, | |
1740 | DissectImageFlags flags) { | |
1741 | ||
1f0f82f1 | 1742 | int r, xbootldr_mounted; |
8c1be37e LP |
1743 | |
1744 | assert(m); | |
1745 | assert(where); | |
1746 | ||
fa45d12c LP |
1747 | /* Returns: |
1748 | * | |
1749 | * -ENXIO → No root partition found | |
7718ac97 | 1750 | * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release/extension-release file found |
fa45d12c LP |
1751 | * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet |
1752 | * -EUCLEAN → fsck for file system failed | |
1753 | * -EBUSY → File system already mounted/used elsewhere (kernel) | |
4dc28665 | 1754 | * -EAFNOSUPPORT → File system type not supported or not known |
80ce8580 | 1755 | * -EIDRM → File system is not among allowlisted "common" file systems |
fa45d12c LP |
1756 | */ |
1757 | ||
7cf66030 LP |
1758 | if (!(m->partitions[PARTITION_ROOT].found || |
1759 | (m->partitions[PARTITION_USR].found && FLAGS_SET(flags, DISSECT_IMAGE_USR_NO_ROOT)))) | |
1760 | return -ENXIO; /* Require a root fs or at least a /usr/ fs (the latter is subject to a flag of its own) */ | |
8c1be37e | 1761 | |
2d3a5a73 | 1762 | if ((flags & DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY) == 0) { |
7cf66030 LP |
1763 | |
1764 | /* First mount the root fs. If there's none we use a tmpfs. */ | |
1765 | if (m->partitions[PARTITION_ROOT].found) | |
254e392e | 1766 | r = mount_partition(PARTITION_ROOT, m->partitions + PARTITION_ROOT, where, NULL, uid_shift, uid_range, flags); |
7cf66030 LP |
1767 | else |
1768 | r = mount_root_tmpfs(where, uid_shift, flags); | |
2d3a5a73 LP |
1769 | if (r < 0) |
1770 | return r; | |
aee36b4e | 1771 | |
aee36b4e | 1772 | /* For us mounting root always means mounting /usr as well */ |
254e392e | 1773 | r = mount_partition(PARTITION_USR, m->partitions + PARTITION_USR, where, "/usr", uid_shift, uid_range, flags); |
aee36b4e LP |
1774 | if (r < 0) |
1775 | return r; | |
03bcb6d4 | 1776 | |
9ccb531a LB |
1777 | if ((flags & (DISSECT_IMAGE_VALIDATE_OS|DISSECT_IMAGE_VALIDATE_OS_EXT)) != 0) { |
1778 | /* If either one of the validation flags are set, ensure that the image qualifies | |
1779 | * as one or the other (or both). */ | |
1780 | bool ok = false; | |
1781 | ||
1782 | if (FLAGS_SET(flags, DISSECT_IMAGE_VALIDATE_OS)) { | |
1783 | r = path_is_os_tree(where); | |
1784 | if (r < 0) | |
1785 | return r; | |
1786 | if (r > 0) | |
1787 | ok = true; | |
1788 | } | |
1789 | if (!ok && FLAGS_SET(flags, DISSECT_IMAGE_VALIDATE_OS_EXT)) { | |
06768b90 | 1790 | r = path_is_extension_tree(where, m->image_name, FLAGS_SET(flags, DISSECT_IMAGE_RELAX_SYSEXT_CHECK)); |
7718ac97 LB |
1791 | if (r < 0) |
1792 | return r; | |
9ccb531a LB |
1793 | if (r > 0) |
1794 | ok = true; | |
7718ac97 | 1795 | } |
9ccb531a LB |
1796 | |
1797 | if (!ok) | |
1798 | return -ENOMEDIUM; | |
03bcb6d4 | 1799 | } |
2d3a5a73 LP |
1800 | } |
1801 | ||
705727fd | 1802 | if (flags & DISSECT_IMAGE_MOUNT_ROOT_ONLY) |
2d3a5a73 | 1803 | return 0; |
8c1be37e | 1804 | |
254e392e | 1805 | r = mount_partition(PARTITION_HOME, m->partitions + PARTITION_HOME, where, "/home", uid_shift, uid_range, flags); |
8c1be37e LP |
1806 | if (r < 0) |
1807 | return r; | |
1808 | ||
254e392e | 1809 | r = mount_partition(PARTITION_SRV, m->partitions + PARTITION_SRV, where, "/srv", uid_shift, uid_range, flags); |
8c1be37e LP |
1810 | if (r < 0) |
1811 | return r; | |
1812 | ||
254e392e | 1813 | r = mount_partition(PARTITION_VAR, m->partitions + PARTITION_VAR, where, "/var", uid_shift, uid_range, flags); |
d4dffb85 LP |
1814 | if (r < 0) |
1815 | return r; | |
1816 | ||
254e392e | 1817 | r = mount_partition(PARTITION_TMP, m->partitions + PARTITION_TMP, where, "/var/tmp", uid_shift, uid_range, flags); |
d4dffb85 LP |
1818 | if (r < 0) |
1819 | return r; | |
1820 | ||
254e392e | 1821 | xbootldr_mounted = mount_partition(PARTITION_XBOOTLDR, m->partitions + PARTITION_XBOOTLDR, where, "/boot", uid_shift, uid_range, flags); |
1f0f82f1 LP |
1822 | if (xbootldr_mounted < 0) |
1823 | return xbootldr_mounted; | |
d9223c07 | 1824 | |
8c1be37e | 1825 | if (m->partitions[PARTITION_ESP].found) { |
1f0f82f1 LP |
1826 | int esp_done = false; |
1827 | ||
d9223c07 LP |
1828 | /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it |
1829 | * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */ | |
8c1be37e | 1830 | |
a5648b80 | 1831 | r = chase_symlinks("/efi", where, CHASE_PREFIX_ROOT, NULL, NULL); |
1f0f82f1 LP |
1832 | if (r < 0) { |
1833 | if (r != -ENOENT) | |
d9223c07 | 1834 | return r; |
8c1be37e | 1835 | |
1f0f82f1 LP |
1836 | /* /efi doesn't exist. Let's see if /boot is suitable then */ |
1837 | ||
1838 | if (!xbootldr_mounted) { | |
1839 | _cleanup_free_ char *p = NULL; | |
2eedfd2d | 1840 | |
1f0f82f1 LP |
1841 | r = chase_symlinks("/boot", where, CHASE_PREFIX_ROOT, &p, NULL); |
1842 | if (r < 0) { | |
1843 | if (r != -ENOENT) | |
1844 | return r; | |
db55bbf2 | 1845 | } else if (dir_is_empty(p, /* ignore_hidden_or_backup= */ false) > 0) { |
1f0f82f1 | 1846 | /* It exists and is an empty directory. Let's mount the ESP there. */ |
254e392e | 1847 | r = mount_partition(PARTITION_ESP, m->partitions + PARTITION_ESP, where, "/boot", uid_shift, uid_range, flags); |
1f0f82f1 LP |
1848 | if (r < 0) |
1849 | return r; | |
1850 | ||
1851 | esp_done = true; | |
1852 | } | |
2eedfd2d | 1853 | } |
8c1be37e | 1854 | } |
1f0f82f1 LP |
1855 | |
1856 | if (!esp_done) { | |
1857 | /* OK, let's mount the ESP now to /efi (possibly creating the dir if missing) */ | |
1858 | ||
254e392e | 1859 | r = mount_partition(PARTITION_ESP, m->partitions + PARTITION_ESP, where, "/efi", uid_shift, uid_range, flags); |
1f0f82f1 LP |
1860 | if (r < 0) |
1861 | return r; | |
1862 | } | |
8c1be37e LP |
1863 | } |
1864 | ||
1865 | return 0; | |
1866 | } | |
1867 | ||
21b61b1d LP |
1868 | int dissected_image_mount_and_warn( |
1869 | DissectedImage *m, | |
1870 | const char *where, | |
1871 | uid_t uid_shift, | |
1872 | uid_t uid_range, | |
1873 | DissectImageFlags flags) { | |
1874 | ||
af187ab2 LP |
1875 | int r; |
1876 | ||
1877 | assert(m); | |
1878 | assert(where); | |
1879 | ||
21b61b1d | 1880 | r = dissected_image_mount(m, where, uid_shift, uid_range, flags); |
af187ab2 LP |
1881 | if (r == -ENXIO) |
1882 | return log_error_errno(r, "Not root file system found in image."); | |
1883 | if (r == -EMEDIUMTYPE) | |
7718ac97 | 1884 | return log_error_errno(r, "No suitable os-release/extension-release file in image found."); |
af187ab2 LP |
1885 | if (r == -EUNATCH) |
1886 | return log_error_errno(r, "Encrypted file system discovered, but decryption not requested."); | |
1887 | if (r == -EUCLEAN) | |
1888 | return log_error_errno(r, "File system check on image failed."); | |
1889 | if (r == -EBUSY) | |
1890 | return log_error_errno(r, "File system already mounted elsewhere."); | |
4dc28665 LP |
1891 | if (r == -EAFNOSUPPORT) |
1892 | return log_error_errno(r, "File system type not supported or not known."); | |
80ce8580 LP |
1893 | if (r == -EIDRM) |
1894 | return log_error_errno(r, "File system is too uncommon, refused."); | |
af187ab2 LP |
1895 | if (r < 0) |
1896 | return log_error_errno(r, "Failed to mount image: %m"); | |
1897 | ||
1898 | return r; | |
1899 | } | |
1900 | ||
349cc4a5 | 1901 | #if HAVE_LIBCRYPTSETUP |
9321ad51 | 1902 | struct DecryptedPartition { |
18b5886e LP |
1903 | struct crypt_device *device; |
1904 | char *name; | |
1905 | bool relinquished; | |
9321ad51 YW |
1906 | }; |
1907 | #endif | |
1908 | ||
1909 | typedef struct DecryptedPartition DecryptedPartition; | |
18b5886e LP |
1910 | |
1911 | struct DecryptedImage { | |
9321ad51 | 1912 | unsigned n_ref; |
18b5886e LP |
1913 | DecryptedPartition *decrypted; |
1914 | size_t n_decrypted; | |
18b5886e | 1915 | }; |
18b5886e | 1916 | |
9321ad51 | 1917 | static DecryptedImage* decrypted_image_free(DecryptedImage *d) { |
349cc4a5 | 1918 | #if HAVE_LIBCRYPTSETUP |
18b5886e LP |
1919 | int r; |
1920 | ||
1921 | if (!d) | |
1922 | return NULL; | |
1923 | ||
67f63ee5 | 1924 | for (size_t i = 0; i < d->n_decrypted; i++) { |
18b5886e LP |
1925 | DecryptedPartition *p = d->decrypted + i; |
1926 | ||
1927 | if (p->device && p->name && !p->relinquished) { | |
5228c58e DDM |
1928 | _cleanup_free_ char *node = NULL; |
1929 | ||
1930 | node = path_join("/dev/mapper", p->name); | |
1931 | if (node) { | |
1932 | r = btrfs_forget_device(node); | |
1933 | if (r < 0 && r != -ENOENT) | |
1934 | log_debug_errno(r, "Failed to forget btrfs device %s, ignoring: %m", node); | |
1935 | } else | |
1936 | log_oom_debug(); | |
1937 | ||
ea16d7f4 YW |
1938 | /* Let's deactivate lazily, as the dm volume may be already/still used by other processes. */ |
1939 | r = sym_crypt_deactivate_by_name(p->device, p->name, CRYPT_DEACTIVATE_DEFERRED); | |
18b5886e LP |
1940 | if (r < 0) |
1941 | log_debug_errno(r, "Failed to deactivate encrypted partition %s", p->name); | |
1942 | } | |
1943 | ||
1944 | if (p->device) | |
0d12936d | 1945 | sym_crypt_free(p->device); |
18b5886e LP |
1946 | free(p->name); |
1947 | } | |
1948 | ||
f91861e4 | 1949 | free(d->decrypted); |
18b5886e LP |
1950 | free(d); |
1951 | #endif | |
1952 | return NULL; | |
1953 | } | |
1954 | ||
9321ad51 YW |
1955 | DEFINE_TRIVIAL_REF_UNREF_FUNC(DecryptedImage, decrypted_image, decrypted_image_free); |
1956 | ||
349cc4a5 | 1957 | #if HAVE_LIBCRYPTSETUP |
9321ad51 YW |
1958 | static int decrypted_image_new(DecryptedImage **ret) { |
1959 | _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL; | |
1960 | ||
1961 | assert(ret); | |
1962 | ||
1963 | d = new(DecryptedImage, 1); | |
1964 | if (!d) | |
1965 | return -ENOMEM; | |
1966 | ||
1967 | *d = (DecryptedImage) { | |
1968 | .n_ref = 1, | |
1969 | }; | |
1970 | ||
1971 | *ret = TAKE_PTR(d); | |
1972 | return 0; | |
1973 | } | |
4623e8e6 LP |
1974 | |
1975 | static int make_dm_name_and_node(const void *original_node, const char *suffix, char **ret_name, char **ret_node) { | |
1976 | _cleanup_free_ char *name = NULL, *node = NULL; | |
1977 | const char *base; | |
1978 | ||
1979 | assert(original_node); | |
1980 | assert(suffix); | |
1981 | assert(ret_name); | |
1982 | assert(ret_node); | |
1983 | ||
1984 | base = strrchr(original_node, '/'); | |
1985 | if (!base) | |
ac1f3ad0 LB |
1986 | base = original_node; |
1987 | else | |
1988 | base++; | |
4623e8e6 LP |
1989 | if (isempty(base)) |
1990 | return -EINVAL; | |
1991 | ||
1992 | name = strjoin(base, suffix); | |
1993 | if (!name) | |
1994 | return -ENOMEM; | |
1995 | if (!filename_is_valid(name)) | |
1996 | return -EINVAL; | |
1997 | ||
0d12936d | 1998 | node = path_join(sym_crypt_get_dir(), name); |
4623e8e6 LP |
1999 | if (!node) |
2000 | return -ENOMEM; | |
2001 | ||
1cc6c93a YW |
2002 | *ret_name = TAKE_PTR(name); |
2003 | *ret_node = TAKE_PTR(node); | |
4623e8e6 | 2004 | |
4623e8e6 LP |
2005 | return 0; |
2006 | } | |
2007 | ||
18b5886e LP |
2008 | static int decrypt_partition( |
2009 | DissectedPartition *m, | |
2010 | const char *passphrase, | |
2011 | DissectImageFlags flags, | |
2012 | DecryptedImage *d) { | |
2013 | ||
2014 | _cleanup_free_ char *node = NULL, *name = NULL; | |
0d12936d | 2015 | _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL; |
254d1313 | 2016 | _cleanup_close_ int fd = -EBADF; |
18b5886e LP |
2017 | int r; |
2018 | ||
2019 | assert(m); | |
2020 | assert(d); | |
2021 | ||
2022 | if (!m->found || !m->node || !m->fstype) | |
2023 | return 0; | |
2024 | ||
2025 | if (!streq(m->fstype, "crypto_LUKS")) | |
2026 | return 0; | |
2027 | ||
bdd73ac5 ZJS |
2028 | if (!passphrase) |
2029 | return -ENOKEY; | |
2030 | ||
0d12936d LP |
2031 | r = dlopen_cryptsetup(); |
2032 | if (r < 0) | |
2033 | return r; | |
2034 | ||
4623e8e6 LP |
2035 | r = make_dm_name_and_node(m->node, "-decrypted", &name, &node); |
2036 | if (r < 0) | |
2037 | return r; | |
18b5886e | 2038 | |
319a4f4b | 2039 | if (!GREEDY_REALLOC0(d->decrypted, d->n_decrypted + 1)) |
18b5886e LP |
2040 | return -ENOMEM; |
2041 | ||
0d12936d | 2042 | r = sym_crypt_init(&cd, m->node); |
18b5886e | 2043 | if (r < 0) |
715cbb81 | 2044 | return log_debug_errno(r, "Failed to initialize dm-crypt: %m"); |
18b5886e | 2045 | |
efc3b12f | 2046 | cryptsetup_enable_logging(cd); |
1887032f | 2047 | |
0d12936d | 2048 | r = sym_crypt_load(cd, CRYPT_LUKS, NULL); |
294bd454 ZJS |
2049 | if (r < 0) |
2050 | return log_debug_errno(r, "Failed to load LUKS metadata: %m"); | |
18b5886e | 2051 | |
0d12936d | 2052 | r = sym_crypt_activate_by_passphrase(cd, name, CRYPT_ANY_SLOT, passphrase, strlen(passphrase), |
ef9c184d | 2053 | ((flags & DISSECT_IMAGE_DEVICE_READ_ONLY) ? CRYPT_ACTIVATE_READONLY : 0) | |
0d12936d | 2054 | ((flags & DISSECT_IMAGE_DISCARD_ON_CRYPTO) ? CRYPT_ACTIVATE_ALLOW_DISCARDS : 0)); |
294bd454 | 2055 | if (r < 0) { |
715cbb81 | 2056 | log_debug_errno(r, "Failed to activate LUKS device: %m"); |
294bd454 | 2057 | return r == -EPERM ? -EKEYREJECTED : r; |
18b5886e | 2058 | } |
18b5886e | 2059 | |
f7725647 YW |
2060 | fd = open(node, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY); |
2061 | if (fd < 0) | |
2062 | return log_debug_errno(errno, "Failed to open %s: %m", node); | |
2063 | ||
94344385 LP |
2064 | d->decrypted[d->n_decrypted++] = (DecryptedPartition) { |
2065 | .name = TAKE_PTR(name), | |
2066 | .device = TAKE_PTR(cd), | |
2067 | }; | |
18b5886e | 2068 | |
1cc6c93a | 2069 | m->decrypted_node = TAKE_PTR(node); |
f7725647 | 2070 | close_and_replace(m->mount_node_fd, fd); |
18b5886e LP |
2071 | |
2072 | return 0; | |
4623e8e6 LP |
2073 | } |
2074 | ||
89e62e0b LP |
2075 | static int verity_can_reuse( |
2076 | const VeritySettings *verity, | |
2077 | const char *name, | |
2078 | struct crypt_device **ret_cd) { | |
2079 | ||
ac1f3ad0 LB |
2080 | /* If the same volume was already open, check that the root hashes match, and reuse it if they do */ |
2081 | _cleanup_free_ char *root_hash_existing = NULL; | |
0d12936d | 2082 | _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL; |
ac1f3ad0 | 2083 | struct crypt_params_verity crypt_params = {}; |
89e62e0b | 2084 | size_t root_hash_existing_size; |
ac1f3ad0 LB |
2085 | int r; |
2086 | ||
89e62e0b LP |
2087 | assert(verity); |
2088 | assert(name); | |
ac1f3ad0 LB |
2089 | assert(ret_cd); |
2090 | ||
0d12936d | 2091 | r = sym_crypt_init_by_name(&cd, name); |
ac1f3ad0 LB |
2092 | if (r < 0) |
2093 | return log_debug_errno(r, "Error opening verity device, crypt_init_by_name failed: %m"); | |
0d12936d | 2094 | |
c719805e LP |
2095 | cryptsetup_enable_logging(cd); |
2096 | ||
0d12936d | 2097 | r = sym_crypt_get_verity_info(cd, &crypt_params); |
ac1f3ad0 LB |
2098 | if (r < 0) |
2099 | return log_debug_errno(r, "Error opening verity device, crypt_get_verity_info failed: %m"); | |
0d12936d | 2100 | |
89e62e0b LP |
2101 | root_hash_existing_size = verity->root_hash_size; |
2102 | root_hash_existing = malloc0(root_hash_existing_size); | |
ac1f3ad0 LB |
2103 | if (!root_hash_existing) |
2104 | return -ENOMEM; | |
0d12936d LP |
2105 | |
2106 | r = sym_crypt_volume_key_get(cd, CRYPT_ANY_SLOT, root_hash_existing, &root_hash_existing_size, NULL, 0); | |
ac1f3ad0 LB |
2107 | if (r < 0) |
2108 | return log_debug_errno(r, "Error opening verity device, crypt_volume_key_get failed: %m"); | |
89e62e0b LP |
2109 | if (verity->root_hash_size != root_hash_existing_size || |
2110 | memcmp(root_hash_existing, verity->root_hash, verity->root_hash_size) != 0) | |
ac1f3ad0 | 2111 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but root hashes are different."); |
89e62e0b | 2112 | |
ac1f3ad0 | 2113 | #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY |
89e62e0b LP |
2114 | /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the |
2115 | * same settings, so that a previous unsigned mount will not be reused if the user asks to use | |
28423d9a | 2116 | * signing for the new one, and vice versa. */ |
89e62e0b | 2117 | if (!!verity->root_hash_sig != !!(crypt_params.flags & CRYPT_VERITY_ROOT_HASH_SIGNATURE)) |
ac1f3ad0 LB |
2118 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Error opening verity device, it already exists but signature settings are not the same."); |
2119 | #endif | |
2120 | ||
2121 | *ret_cd = TAKE_PTR(cd); | |
2122 | return 0; | |
2123 | } | |
2124 | ||
75db809a | 2125 | static inline char* dm_deferred_remove_clean(char *name) { |
ac1f3ad0 | 2126 | if (!name) |
75db809a | 2127 | return NULL; |
0d12936d LP |
2128 | |
2129 | (void) sym_crypt_deactivate_by_name(NULL, name, CRYPT_DEACTIVATE_DEFERRED); | |
75db809a | 2130 | return mfree(name); |
ac1f3ad0 LB |
2131 | } |
2132 | DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean); | |
2133 | ||
c2fa92e7 LP |
2134 | static int validate_signature_userspace(const VeritySettings *verity) { |
2135 | #if HAVE_OPENSSL | |
2136 | _cleanup_(sk_X509_free_allp) STACK_OF(X509) *sk = NULL; | |
2137 | _cleanup_strv_free_ char **certs = NULL; | |
2138 | _cleanup_(PKCS7_freep) PKCS7 *p7 = NULL; | |
2139 | _cleanup_free_ char *s = NULL; | |
2140 | _cleanup_(BIO_freep) BIO *bio = NULL; /* 'bio' must be freed first, 's' second, hence keep this order | |
2141 | * of declaration in place, please */ | |
2142 | const unsigned char *d; | |
c2fa92e7 LP |
2143 | int r; |
2144 | ||
2145 | assert(verity); | |
2146 | assert(verity->root_hash); | |
2147 | assert(verity->root_hash_sig); | |
2148 | ||
2149 | /* Because installing a signature certificate into the kernel chain is so messy, let's optionally do | |
2150 | * userspace validation. */ | |
2151 | ||
2152 | r = conf_files_list_nulstr(&certs, ".crt", NULL, CONF_FILES_REGULAR|CONF_FILES_FILTER_MASKED, CONF_PATHS_NULSTR("verity.d")); | |
2153 | if (r < 0) | |
2154 | return log_debug_errno(r, "Failed to enumerate certificates: %m"); | |
2155 | if (strv_isempty(certs)) { | |
2156 | log_debug("No userspace dm-verity certificates found."); | |
2157 | return 0; | |
2158 | } | |
2159 | ||
2160 | d = verity->root_hash_sig; | |
2161 | p7 = d2i_PKCS7(NULL, &d, (long) verity->root_hash_sig_size); | |
2162 | if (!p7) | |
2163 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to parse PKCS7 DER signature data."); | |
2164 | ||
2165 | s = hexmem(verity->root_hash, verity->root_hash_size); | |
2166 | if (!s) | |
2167 | return log_oom_debug(); | |
2168 | ||
2169 | bio = BIO_new_mem_buf(s, strlen(s)); | |
2170 | if (!bio) | |
2171 | return log_oom_debug(); | |
2172 | ||
2173 | sk = sk_X509_new_null(); | |
2174 | if (!sk) | |
2175 | return log_oom_debug(); | |
2176 | ||
2177 | STRV_FOREACH(i, certs) { | |
2178 | _cleanup_(X509_freep) X509 *c = NULL; | |
2179 | _cleanup_fclose_ FILE *f = NULL; | |
2180 | ||
2181 | f = fopen(*i, "re"); | |
2182 | if (!f) { | |
2183 | log_debug_errno(errno, "Failed to open '%s', ignoring: %m", *i); | |
2184 | continue; | |
2185 | } | |
2186 | ||
2187 | c = PEM_read_X509(f, NULL, NULL, NULL); | |
2188 | if (!c) { | |
2189 | log_debug("Failed to load X509 certificate '%s', ignoring.", *i); | |
2190 | continue; | |
2191 | } | |
2192 | ||
2193 | if (sk_X509_push(sk, c) == 0) | |
2194 | return log_oom_debug(); | |
2195 | ||
2196 | TAKE_PTR(c); | |
2197 | } | |
2198 | ||
2199 | r = PKCS7_verify(p7, sk, NULL, bio, NULL, PKCS7_NOINTERN|PKCS7_NOVERIFY); | |
2200 | if (r) | |
2201 | log_debug("Userspace PKCS#7 validation succeeded."); | |
2202 | else | |
2203 | log_debug("Userspace PKCS#7 validation failed: %s", ERR_error_string(ERR_get_error(), NULL)); | |
2204 | ||
2205 | return r; | |
2206 | #else | |
2207 | log_debug("Not doing client-side validation of dm-verity root hash signatures, OpenSSL support disabled."); | |
2208 | return 0; | |
2209 | #endif | |
2210 | } | |
2211 | ||
2212 | static int do_crypt_activate_verity( | |
2213 | struct crypt_device *cd, | |
2214 | const char *name, | |
2215 | const VeritySettings *verity) { | |
2216 | ||
2217 | bool check_signature; | |
2218 | int r; | |
2219 | ||
2220 | assert(cd); | |
2221 | assert(name); | |
2222 | assert(verity); | |
2223 | ||
2224 | if (verity->root_hash_sig) { | |
2225 | r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIGNATURE"); | |
2226 | if (r < 0 && r != -ENXIO) | |
2227 | log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIGNATURE"); | |
2228 | ||
2229 | check_signature = r != 0; | |
2230 | } else | |
2231 | check_signature = false; | |
2232 | ||
2233 | if (check_signature) { | |
2234 | ||
2235 | #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY | |
2236 | /* First, if we have support for signed keys in the kernel, then try that first. */ | |
2237 | r = sym_crypt_activate_by_signed_key( | |
2238 | cd, | |
2239 | name, | |
2240 | verity->root_hash, | |
2241 | verity->root_hash_size, | |
2242 | verity->root_hash_sig, | |
2243 | verity->root_hash_sig_size, | |
2244 | CRYPT_ACTIVATE_READONLY); | |
2245 | if (r >= 0) | |
2246 | return r; | |
2247 | ||
2248 | log_debug("Validation of dm-verity signature failed via the kernel, trying userspace validation instead."); | |
2249 | #else | |
2250 | 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.", | |
2251 | program_invocation_short_name); | |
2252 | #endif | |
2253 | ||
2254 | /* So this didn't work via the kernel, then let's try userspace validation instead. If that | |
2255 | * works we'll try to activate without telling the kernel the signature. */ | |
2256 | ||
2257 | r = validate_signature_userspace(verity); | |
2258 | if (r < 0) | |
2259 | return r; | |
2260 | if (r == 0) | |
2261 | return log_debug_errno(SYNTHETIC_ERRNO(ENOKEY), | |
2262 | "Activation of signed Verity volume worked neither via the kernel nor in userspace, can't activate."); | |
2263 | } | |
2264 | ||
2265 | return sym_crypt_activate_by_volume_key( | |
2266 | cd, | |
2267 | name, | |
2268 | verity->root_hash, | |
2269 | verity->root_hash_size, | |
2270 | CRYPT_ACTIVATE_READONLY); | |
2271 | } | |
2272 | ||
ad361a50 YW |
2273 | static usec_t verity_timeout(void) { |
2274 | usec_t t = 100 * USEC_PER_MSEC; | |
2275 | const char *e; | |
2276 | int r; | |
2277 | ||
2278 | /* On slower machines, like non-KVM vm, setting up device may take a long time. | |
2279 | * Let's make the timeout configurable. */ | |
2280 | ||
2281 | e = getenv("SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC"); | |
2282 | if (!e) | |
2283 | return t; | |
2284 | ||
2285 | r = parse_sec(e, &t); | |
2286 | if (r < 0) | |
2287 | log_debug_errno(r, | |
2288 | "Failed to parse timeout specified in $SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC, " | |
2289 | "using the default timeout (%s).", | |
2290 | FORMAT_TIMESPAN(t, USEC_PER_MSEC)); | |
2291 | ||
2292 | return t; | |
2293 | } | |
2294 | ||
4623e8e6 | 2295 | static int verity_partition( |
aee36b4e | 2296 | PartitionDesignator designator, |
4623e8e6 LP |
2297 | DissectedPartition *m, |
2298 | DissectedPartition *v, | |
89e62e0b | 2299 | const VeritySettings *verity, |
4623e8e6 LP |
2300 | DissectImageFlags flags, |
2301 | DecryptedImage *d) { | |
2302 | ||
0d12936d | 2303 | _cleanup_(sym_crypt_freep) struct crypt_device *cd = NULL; |
ac1f3ad0 | 2304 | _cleanup_(dm_deferred_remove_cleanp) char *restore_deferred_remove = NULL; |
89e62e0b | 2305 | _cleanup_free_ char *node = NULL, *name = NULL; |
254d1313 | 2306 | _cleanup_close_ int mount_node_fd = -EBADF; |
4623e8e6 LP |
2307 | int r; |
2308 | ||
2309 | assert(m); | |
89e62e0b | 2310 | assert(v || (verity && verity->data_path)); |
4623e8e6 | 2311 | |
89e62e0b | 2312 | if (!verity || !verity->root_hash) |
4623e8e6 | 2313 | return 0; |
aee36b4e LP |
2314 | if (!((verity->designator < 0 && designator == PARTITION_ROOT) || |
2315 | (verity->designator == designator))) | |
2316 | return 0; | |
4623e8e6 LP |
2317 | |
2318 | if (!m->found || !m->node || !m->fstype) | |
2319 | return 0; | |
89e62e0b | 2320 | if (!verity->data_path) { |
e7cbe5cb LB |
2321 | if (!v->found || !v->node || !v->fstype) |
2322 | return 0; | |
4623e8e6 | 2323 | |
e7cbe5cb LB |
2324 | if (!streq(v->fstype, "DM_verity_hash")) |
2325 | return 0; | |
2326 | } | |
4623e8e6 | 2327 | |
0d12936d LP |
2328 | r = dlopen_cryptsetup(); |
2329 | if (r < 0) | |
2330 | return r; | |
2331 | ||
ac1f3ad0 LB |
2332 | if (FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) { |
2333 | /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */ | |
2334 | _cleanup_free_ char *root_hash_encoded = NULL; | |
0d12936d | 2335 | |
89e62e0b | 2336 | root_hash_encoded = hexmem(verity->root_hash, verity->root_hash_size); |
ac1f3ad0 LB |
2337 | if (!root_hash_encoded) |
2338 | return -ENOMEM; | |
aee36b4e | 2339 | |
ac1f3ad0 LB |
2340 | r = make_dm_name_and_node(root_hash_encoded, "-verity", &name, &node); |
2341 | } else | |
2342 | r = make_dm_name_and_node(m->node, "-verity", &name, &node); | |
4623e8e6 LP |
2343 | if (r < 0) |
2344 | return r; | |
2345 | ||
89e62e0b | 2346 | r = sym_crypt_init(&cd, verity->data_path ?: v->node); |
4623e8e6 LP |
2347 | if (r < 0) |
2348 | return r; | |
2349 | ||
efc3b12f | 2350 | cryptsetup_enable_logging(cd); |
1887032f | 2351 | |
0d12936d | 2352 | r = sym_crypt_load(cd, CRYPT_VERITY, NULL); |
4623e8e6 | 2353 | if (r < 0) |
294bd454 | 2354 | return r; |
4623e8e6 | 2355 | |
0d12936d | 2356 | r = sym_crypt_set_data_device(cd, m->node); |
4623e8e6 | 2357 | if (r < 0) |
294bd454 | 2358 | return r; |
4623e8e6 | 2359 | |
319a4f4b | 2360 | if (!GREEDY_REALLOC0(d->decrypted, d->n_decrypted + 1)) |
ac1f3ad0 LB |
2361 | return -ENOMEM; |
2362 | ||
2363 | /* If activating fails because the device already exists, check the metadata and reuse it if it matches. | |
2364 | * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time, | |
2365 | * retry a few times before giving up. */ | |
2366 | for (unsigned i = 0; i < N_DEVICE_NODE_LIST_ATTEMPTS; i++) { | |
1d369d78 | 2367 | _cleanup_(sym_crypt_freep) struct crypt_device *existing_cd = NULL; |
254d1313 | 2368 | _cleanup_close_ int fd = -EBADF; |
c2fa92e7 | 2369 | |
f80015ff YW |
2370 | /* First, check if the device already exists. */ |
2371 | fd = open(node, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY); | |
2372 | if (fd < 0 && !ERRNO_IS_DEVICE_ABSENT(errno)) | |
2373 | return log_debug_errno(errno, "Failed to open verity device %s: %m", node); | |
2374 | if (fd >= 0) | |
2375 | goto check; /* The device already exists. Let's check it. */ | |
2376 | ||
2377 | /* The symlink to the device node does not exist yet. Assume not activated, and let's activate it. */ | |
c2fa92e7 | 2378 | r = do_crypt_activate_verity(cd, name, verity); |
1d369d78 | 2379 | if (r >= 0) |
f7725647 | 2380 | goto try_open; /* The device is activated. Let's open it. */ |
ac1f3ad0 LB |
2381 | /* libdevmapper can return EINVAL when the device is already in the activation stage. |
2382 | * There's no way to distinguish this situation from a genuine error due to invalid | |
2aed63f4 | 2383 | * parameters, so immediately fall back to activating the device with a unique name. |
89e62e0b LP |
2384 | * Improvements in libcrypsetup can ensure this never happens: |
2385 | * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */ | |
ac1f3ad0 | 2386 | if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) |
0f75b0c5 | 2387 | break; |
1d369d78 YW |
2388 | if (r == -ENODEV) /* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */ |
2389 | goto try_again; | |
2390 | if (!IN_SET(r, | |
2391 | -EEXIST, /* Volume has already been opened and ready to be used. */ | |
2392 | -EBUSY /* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */)) | |
2393 | return log_debug_errno(r, "Failed to activate verity device %s: %m", node); | |
2394 | ||
f80015ff | 2395 | check: |
1d369d78 YW |
2396 | if (!restore_deferred_remove){ |
2397 | /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */ | |
2398 | r = dm_deferred_remove_cancel(name); | |
12f5fbdf YW |
2399 | /* -EBUSY and -ENXIO: the device has already been removed or being removed. We cannot |
2400 | * use the device, try to open again. See target_message() in drivers/md/dm-ioctl.c | |
2401 | * and dm_cancel_deferred_remove() in drivers/md/dm.c */ | |
2402 | if (IN_SET(r, -EBUSY, -ENXIO)) | |
2403 | goto try_again; | |
2404 | if (r < 0) | |
1d369d78 | 2405 | return log_debug_errno(r, "Failed to disable automated deferred removal for verity device %s: %m", node); |
12f5fbdf YW |
2406 | |
2407 | restore_deferred_remove = strdup(name); | |
2408 | if (!restore_deferred_remove) | |
2409 | return log_oom_debug(); | |
1d369d78 | 2410 | } |
c2923fdc | 2411 | |
1d369d78 YW |
2412 | r = verity_can_reuse(verity, name, &existing_cd); |
2413 | /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */ | |
2414 | if (r == -EINVAL && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) | |
2415 | break; | |
2416 | if (IN_SET(r, | |
2417 | -ENOENT, /* Removed?? */ | |
2418 | -EBUSY, /* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */ | |
2419 | -ENODEV /* Volume is being opened but not ready, crypt_init_by_name() would fail, try to open again. */ )) | |
2420 | goto try_again; | |
2421 | if (r < 0) | |
2422 | return log_debug_errno(r, "Failed to check if existing verity device %s can be reused: %m", node); | |
2423 | ||
f80015ff | 2424 | if (fd < 0) { |
1d369d78 YW |
2425 | /* devmapper might say that the device exists, but the devlink might not yet have been |
2426 | * created. Check and wait for the udev event in that case. */ | |
2427 | r = device_wait_for_devlink(node, "block", verity_timeout(), NULL); | |
2428 | /* Fallback to activation with a unique device if it's taking too long */ | |
2429 | if (r == -ETIMEDOUT && FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) | |
0f75b0c5 | 2430 | break; |
1d369d78 YW |
2431 | if (r < 0) |
2432 | return log_debug_errno(r, "Failed to wait device node symlink %s: %m", node); | |
c2923fdc | 2433 | } |
ecab4c47 | 2434 | |
f7725647 YW |
2435 | try_open: |
2436 | if (fd < 0) { | |
2437 | /* Now, the device is activated and devlink is created. Let's open it. */ | |
2438 | fd = open(node, O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_NOCTTY); | |
2439 | if (fd < 0) { | |
2440 | if (!ERRNO_IS_DEVICE_ABSENT(errno)) | |
2441 | return log_debug_errno(errno, "Failed to open verity device %s: %m", node); | |
2442 | ||
2443 | /* The device has already been removed?? */ | |
2444 | goto try_again; | |
2445 | } | |
2446 | } | |
2447 | ||
2448 | mount_node_fd = TAKE_FD(fd); | |
1d369d78 YW |
2449 | if (existing_cd) |
2450 | crypt_free_and_replace(cd, existing_cd); | |
2451 | ||
2452 | goto success; | |
2453 | ||
2454 | try_again: | |
2455 | /* Device is being removed by another process. Let's wait for a while. */ | |
ecab4c47 | 2456 | (void) usleep(2 * USEC_PER_MSEC); |
ac1f3ad0 LB |
2457 | } |
2458 | ||
0f75b0c5 YW |
2459 | /* All trials failed or a conflicting verity device exists. Let's try to activate with a unique name. */ |
2460 | if (FLAGS_SET(flags, DISSECT_IMAGE_VERITY_SHARE)) { | |
2461 | /* Before trying to activate with unique name, we need to free crypt_device object. | |
2462 | * Otherwise, we get error from libcryptsetup like the following: | |
2463 | * ------ | |
2464 | * systemd[1234]: Cannot use device /dev/loop5 which is in use (already mapped or mounted). | |
2465 | * ------ | |
2466 | */ | |
2467 | sym_crypt_free(cd); | |
2468 | cd = NULL; | |
aee36b4e | 2469 | return verity_partition(designator, m, v, verity, flags & ~DISSECT_IMAGE_VERITY_SHARE, d); |
0f75b0c5 | 2470 | } |
ac1f3ad0 | 2471 | |
9972e6d6 YW |
2472 | return log_debug_errno(SYNTHETIC_ERRNO(EBUSY), "All attempts to activate verity device %s failed.", name); |
2473 | ||
2474 | success: | |
ac1f3ad0 LB |
2475 | /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */ |
2476 | restore_deferred_remove = mfree(restore_deferred_remove); | |
4623e8e6 | 2477 | |
94344385 LP |
2478 | d->decrypted[d->n_decrypted++] = (DecryptedPartition) { |
2479 | .name = TAKE_PTR(name), | |
2480 | .device = TAKE_PTR(cd), | |
2481 | }; | |
4623e8e6 | 2482 | |
1cc6c93a | 2483 | m->decrypted_node = TAKE_PTR(node); |
f7725647 | 2484 | close_and_replace(m->mount_node_fd, mount_node_fd); |
4623e8e6 LP |
2485 | |
2486 | return 0; | |
18b5886e LP |
2487 | } |
2488 | #endif | |
2489 | ||
2490 | int dissected_image_decrypt( | |
2491 | DissectedImage *m, | |
2492 | const char *passphrase, | |
89e62e0b | 2493 | const VeritySettings *verity, |
e330f97a | 2494 | DissectImageFlags flags) { |
18b5886e | 2495 | |
349cc4a5 | 2496 | #if HAVE_LIBCRYPTSETUP |
49b5b3b4 | 2497 | _cleanup_(decrypted_image_unrefp) DecryptedImage *d = NULL; |
18b5886e LP |
2498 | int r; |
2499 | #endif | |
2500 | ||
2501 | assert(m); | |
89e62e0b | 2502 | assert(!verity || verity->root_hash || verity->root_hash_size == 0); |
18b5886e LP |
2503 | |
2504 | /* Returns: | |
2505 | * | |
2506 | * = 0 → There was nothing to decrypt | |
2507 | * > 0 → Decrypted successfully | |
d1c536f5 | 2508 | * -ENOKEY → There's something to decrypt but no key was supplied |
18b5886e LP |
2509 | * -EKEYREJECTED → Passed key was not correct |
2510 | */ | |
2511 | ||
89e62e0b | 2512 | if (verity && verity->root_hash && verity->root_hash_size < sizeof(sd_id128_t)) |
4623e8e6 LP |
2513 | return -EINVAL; |
2514 | ||
e330f97a | 2515 | if (!m->encrypted && !m->verity_ready) |
18b5886e | 2516 | return 0; |
18b5886e | 2517 | |
349cc4a5 | 2518 | #if HAVE_LIBCRYPTSETUP |
9321ad51 YW |
2519 | r = decrypted_image_new(&d); |
2520 | if (r < 0) | |
2521 | return r; | |
18b5886e | 2522 | |
569a0e42 | 2523 | for (PartitionDesignator i = 0; i < _PARTITION_DESIGNATOR_MAX; i++) { |
18b5886e | 2524 | DissectedPartition *p = m->partitions + i; |
22043172 | 2525 | PartitionDesignator k; |
18b5886e LP |
2526 | |
2527 | if (!p->found) | |
2528 | continue; | |
2529 | ||
2530 | r = decrypt_partition(p, passphrase, flags, d); | |
2531 | if (r < 0) | |
2532 | return r; | |
2533 | ||
dd894023 | 2534 | k = partition_verity_of(i); |
4623e8e6 | 2535 | if (k >= 0) { |
aee36b4e | 2536 | r = verity_partition(i, p, m->partitions + k, verity, flags | DISSECT_IMAGE_VERITY_SHARE, d); |
4623e8e6 LP |
2537 | if (r < 0) |
2538 | return r; | |
2539 | } | |
2540 | ||
d2c6e79d | 2541 | if (!p->decrypted_fstype && p->mount_node_fd >= 0 && p->decrypted_node) { |
c80c9079 | 2542 | r = probe_filesystem_full(p->mount_node_fd, p->decrypted_node, 0, UINT64_MAX, &p->decrypted_fstype); |
7cc84b2c | 2543 | if (r < 0 && r != -EUCLEAN) |
18b5886e LP |
2544 | return r; |
2545 | } | |
2546 | } | |
2547 | ||
ac1e1b5f | 2548 | m->decrypted_image = TAKE_PTR(d); |
18b5886e LP |
2549 | |
2550 | return 1; | |
2551 | #else | |
2552 | return -EOPNOTSUPP; | |
2553 | #endif | |
2554 | } | |
2555 | ||
2556 | int dissected_image_decrypt_interactively( | |
2557 | DissectedImage *m, | |
2558 | const char *passphrase, | |
89e62e0b | 2559 | const VeritySettings *verity, |
e330f97a | 2560 | DissectImageFlags flags) { |
18b5886e LP |
2561 | |
2562 | _cleanup_strv_free_erase_ char **z = NULL; | |
2563 | int n = 3, r; | |
2564 | ||
2565 | if (passphrase) | |
2566 | n--; | |
2567 | ||
2568 | for (;;) { | |
e330f97a | 2569 | r = dissected_image_decrypt(m, passphrase, verity, flags); |
18b5886e LP |
2570 | if (r >= 0) |
2571 | return r; | |
2572 | if (r == -EKEYREJECTED) | |
2573 | log_error_errno(r, "Incorrect passphrase, try again!"); | |
fc95c359 YW |
2574 | else if (r != -ENOKEY) |
2575 | return log_error_errno(r, "Failed to decrypt image: %m"); | |
18b5886e | 2576 | |
baaa35ad ZJS |
2577 | if (--n < 0) |
2578 | return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED), | |
2579 | "Too many retries."); | |
18b5886e LP |
2580 | |
2581 | z = strv_free(z); | |
2582 | ||
8806bb4b | 2583 | r = ask_password_auto("Please enter image passphrase:", NULL, "dissect", "dissect", "dissect.passphrase", USEC_INFINITY, 0, &z); |
18b5886e LP |
2584 | if (r < 0) |
2585 | return log_error_errno(r, "Failed to query for passphrase: %m"); | |
2586 | ||
2587 | passphrase = z[0]; | |
2588 | } | |
2589 | } | |
2590 | ||
e330f97a | 2591 | static int decrypted_image_relinquish(DecryptedImage *d) { |
18b5886e LP |
2592 | assert(d); |
2593 | ||
67f63ee5 ZJS |
2594 | /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a |
2595 | * boolean so that we don't clean it up ourselves either anymore */ | |
18b5886e | 2596 | |
349cc4a5 | 2597 | #if HAVE_LIBCRYPTSETUP |
67f63ee5 ZJS |
2598 | int r; |
2599 | ||
2600 | for (size_t i = 0; i < d->n_decrypted; i++) { | |
18b5886e LP |
2601 | DecryptedPartition *p = d->decrypted + i; |
2602 | ||
2603 | if (p->relinquished) | |
2604 | continue; | |
2605 | ||
0d12936d | 2606 | r = sym_crypt_deactivate_by_name(NULL, p->name, CRYPT_DEACTIVATE_DEFERRED); |
18b5886e LP |
2607 | if (r < 0) |
2608 | return log_debug_errno(r, "Failed to mark %s for auto-removal: %m", p->name); | |
2609 | ||
2610 | p->relinquished = true; | |
2611 | } | |
2612 | #endif | |
2613 | ||
2614 | return 0; | |
2615 | } | |
2616 | ||
f906075a YW |
2617 | int dissected_image_relinquish(DissectedImage *m) { |
2618 | int r; | |
2619 | ||
2620 | assert(m); | |
2621 | ||
2622 | if (m->decrypted_image) { | |
2623 | r = decrypted_image_relinquish(m->decrypted_image); | |
2624 | if (r < 0) | |
2625 | return r; | |
2626 | } | |
2627 | ||
2628 | if (m->loop) | |
2629 | loop_device_relinquish(m->loop); | |
2630 | ||
2631 | return 0; | |
2632 | } | |
2633 | ||
89e62e0b LP |
2634 | static char *build_auxiliary_path(const char *image, const char *suffix) { |
2635 | const char *e; | |
2636 | char *n; | |
2637 | ||
2638 | assert(image); | |
2639 | assert(suffix); | |
2640 | ||
2641 | e = endswith(image, ".raw"); | |
2642 | if (!e) | |
2643 | return strjoin(e, suffix); | |
2644 | ||
2645 | n = new(char, e - image + strlen(suffix) + 1); | |
2646 | if (!n) | |
2647 | return NULL; | |
2648 | ||
2649 | strcpy(mempcpy(n, image, e - image), suffix); | |
2650 | return n; | |
2651 | } | |
2652 | ||
2653 | void verity_settings_done(VeritySettings *v) { | |
2654 | assert(v); | |
2655 | ||
2656 | v->root_hash = mfree(v->root_hash); | |
2657 | v->root_hash_size = 0; | |
2658 | ||
2659 | v->root_hash_sig = mfree(v->root_hash_sig); | |
2660 | v->root_hash_sig_size = 0; | |
2661 | ||
2662 | v->data_path = mfree(v->data_path); | |
2663 | } | |
2664 | ||
2665 | int verity_settings_load( | |
2666 | VeritySettings *verity, | |
f5ea63a5 LP |
2667 | const char *image, |
2668 | const char *root_hash_path, | |
89e62e0b LP |
2669 | const char *root_hash_sig_path) { |
2670 | ||
2671 | _cleanup_free_ void *root_hash = NULL, *root_hash_sig = NULL; | |
2672 | size_t root_hash_size = 0, root_hash_sig_size = 0; | |
2673 | _cleanup_free_ char *verity_data_path = NULL; | |
aee36b4e | 2674 | PartitionDesignator designator; |
78ebe980 LP |
2675 | int r; |
2676 | ||
89e62e0b | 2677 | assert(verity); |
78ebe980 | 2678 | assert(image); |
aee36b4e | 2679 | assert(verity->designator < 0 || IN_SET(verity->designator, PARTITION_ROOT, PARTITION_USR)); |
78ebe980 | 2680 | |
89e62e0b LP |
2681 | /* If we are asked to load the root hash for a device node, exit early */ |
2682 | if (is_device_path(image)) | |
78ebe980 | 2683 | return 0; |
78ebe980 | 2684 | |
d5fcc5b0 LP |
2685 | r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIDECAR"); |
2686 | if (r < 0 && r != -ENXIO) | |
2687 | log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIDECAR, ignoring: %m"); | |
2688 | if (r == 0) | |
2689 | return 0; | |
2690 | ||
aee36b4e LP |
2691 | designator = verity->designator; |
2692 | ||
89e62e0b | 2693 | /* We only fill in what isn't already filled in */ |
c2923fdc | 2694 | |
89e62e0b | 2695 | if (!verity->root_hash) { |
e7cbe5cb | 2696 | _cleanup_free_ char *text = NULL; |
e7cbe5cb | 2697 | |
0389f4fa | 2698 | if (root_hash_path) { |
aee36b4e | 2699 | /* If explicitly specified it takes precedence */ |
0389f4fa LB |
2700 | r = read_one_line_file(root_hash_path, &text); |
2701 | if (r < 0) | |
e7cbe5cb | 2702 | return r; |
aee36b4e LP |
2703 | |
2704 | if (designator < 0) | |
2705 | designator = PARTITION_ROOT; | |
0389f4fa | 2706 | } else { |
aee36b4e LP |
2707 | /* Otherwise look for xattr and separate file, and first for the data for root and if |
2708 | * that doesn't exist for /usr */ | |
0389f4fa | 2709 | |
aee36b4e | 2710 | if (designator < 0 || designator == PARTITION_ROOT) { |
c53e07e2 | 2711 | r = getxattr_malloc(image, "user.verity.roothash", &text); |
aee36b4e LP |
2712 | if (r < 0) { |
2713 | _cleanup_free_ char *p = NULL; | |
78ebe980 | 2714 | |
00675c36 | 2715 | if (r != -ENOENT && !ERRNO_IS_XATTR_ABSENT(r)) |
aee36b4e | 2716 | return r; |
e7cbe5cb | 2717 | |
aee36b4e LP |
2718 | p = build_auxiliary_path(image, ".roothash"); |
2719 | if (!p) | |
2720 | return -ENOMEM; | |
2721 | ||
2722 | r = read_one_line_file(p, &text); | |
2723 | if (r < 0 && r != -ENOENT) | |
2724 | return r; | |
2725 | } | |
2726 | ||
2727 | if (text) | |
2728 | designator = PARTITION_ROOT; | |
2729 | } | |
2730 | ||
2731 | if (!text && (designator < 0 || designator == PARTITION_USR)) { | |
2732 | /* So in the "roothash" xattr/file name above the "root" of course primarily | |
2733 | * refers to the root of the Verity Merkle tree. But coincidentally it also | |
2734 | * is the hash for the *root* file system, i.e. the "root" neatly refers to | |
2735 | * two distinct concepts called "root". Taking benefit of this happy | |
2736 | * coincidence we call the file with the root hash for the /usr/ file system | |
2737 | * `usrhash`, because `usrroothash` or `rootusrhash` would just be too | |
2738 | * confusing. We thus drop the reference to the root of the Merkle tree, and | |
2739 | * just indicate which file system it's about. */ | |
c53e07e2 | 2740 | r = getxattr_malloc(image, "user.verity.usrhash", &text); |
aee36b4e LP |
2741 | if (r < 0) { |
2742 | _cleanup_free_ char *p = NULL; | |
2743 | ||
00675c36 | 2744 | if (r != -ENOENT && !ERRNO_IS_XATTR_ABSENT(r)) |
aee36b4e LP |
2745 | return r; |
2746 | ||
2747 | p = build_auxiliary_path(image, ".usrhash"); | |
2748 | if (!p) | |
2749 | return -ENOMEM; | |
2750 | ||
2751 | r = read_one_line_file(p, &text); | |
2752 | if (r < 0 && r != -ENOENT) | |
2753 | return r; | |
2754 | } | |
2755 | ||
2756 | if (text) | |
2757 | designator = PARTITION_USR; | |
0389f4fa | 2758 | } |
e7cbe5cb LB |
2759 | } |
2760 | ||
2761 | if (text) { | |
89e62e0b | 2762 | r = unhexmem(text, strlen(text), &root_hash, &root_hash_size); |
e7cbe5cb LB |
2763 | if (r < 0) |
2764 | return r; | |
89e62e0b | 2765 | if (root_hash_size < sizeof(sd_id128_t)) |
e7cbe5cb LB |
2766 | return -EINVAL; |
2767 | } | |
2768 | } | |
2769 | ||
90f98986 | 2770 | if ((root_hash || verity->root_hash) && !verity->root_hash_sig) { |
aee36b4e | 2771 | if (root_hash_sig_path) { |
ae9cf30b | 2772 | r = read_full_file(root_hash_sig_path, (char**) &root_hash_sig, &root_hash_sig_size); |
aee36b4e LP |
2773 | if (r < 0 && r != -ENOENT) |
2774 | return r; | |
2775 | ||
2776 | if (designator < 0) | |
2777 | designator = PARTITION_ROOT; | |
2778 | } else { | |
2779 | if (designator < 0 || designator == PARTITION_ROOT) { | |
2780 | _cleanup_free_ char *p = NULL; | |
2781 | ||
2782 | /* Follow naming convention recommended by the relevant RFC: | |
2783 | * https://tools.ietf.org/html/rfc5751#section-3.2.1 */ | |
2784 | p = build_auxiliary_path(image, ".roothash.p7s"); | |
2785 | if (!p) | |
2786 | return -ENOMEM; | |
89e62e0b | 2787 | |
ae9cf30b | 2788 | r = read_full_file(p, (char**) &root_hash_sig, &root_hash_sig_size); |
aee36b4e LP |
2789 | if (r < 0 && r != -ENOENT) |
2790 | return r; | |
2791 | if (r >= 0) | |
2792 | designator = PARTITION_ROOT; | |
2793 | } | |
2794 | ||
2795 | if (!root_hash_sig && (designator < 0 || designator == PARTITION_USR)) { | |
2796 | _cleanup_free_ char *p = NULL; | |
2797 | ||
2798 | p = build_auxiliary_path(image, ".usrhash.p7s"); | |
2799 | if (!p) | |
2800 | return -ENOMEM; | |
89e62e0b | 2801 | |
ae9cf30b | 2802 | r = read_full_file(p, (char**) &root_hash_sig, &root_hash_sig_size); |
aee36b4e LP |
2803 | if (r < 0 && r != -ENOENT) |
2804 | return r; | |
2805 | if (r >= 0) | |
2806 | designator = PARTITION_USR; | |
2807 | } | |
89e62e0b LP |
2808 | } |
2809 | ||
aee36b4e | 2810 | if (root_hash_sig && root_hash_sig_size == 0) /* refuse empty size signatures */ |
89e62e0b LP |
2811 | return -EINVAL; |
2812 | } | |
2813 | ||
2814 | if (!verity->data_path) { | |
2815 | _cleanup_free_ char *p = NULL; | |
2816 | ||
2817 | p = build_auxiliary_path(image, ".verity"); | |
2818 | if (!p) | |
2819 | return -ENOMEM; | |
2820 | ||
2821 | if (access(p, F_OK) < 0) { | |
2822 | if (errno != ENOENT) | |
2823 | return -errno; | |
2824 | } else | |
2825 | verity_data_path = TAKE_PTR(p); | |
2826 | } | |
2827 | ||
2828 | if (root_hash) { | |
2829 | verity->root_hash = TAKE_PTR(root_hash); | |
2830 | verity->root_hash_size = root_hash_size; | |
2831 | } | |
2832 | ||
2833 | if (root_hash_sig) { | |
2834 | verity->root_hash_sig = TAKE_PTR(root_hash_sig); | |
2835 | verity->root_hash_sig_size = root_hash_sig_size; | |
e7cbe5cb | 2836 | } |
89e62e0b LP |
2837 | |
2838 | if (verity_data_path) | |
2839 | verity->data_path = TAKE_PTR(verity_data_path); | |
78ebe980 | 2840 | |
aee36b4e LP |
2841 | if (verity->designator < 0) |
2842 | verity->designator = designator; | |
2843 | ||
78ebe980 LP |
2844 | return 1; |
2845 | } | |
2846 | ||
88b3300f LP |
2847 | int dissected_image_load_verity_sig_partition( |
2848 | DissectedImage *m, | |
2849 | int fd, | |
2850 | VeritySettings *verity) { | |
2851 | ||
2852 | _cleanup_free_ void *root_hash = NULL, *root_hash_sig = NULL; | |
2853 | _cleanup_(json_variant_unrefp) JsonVariant *v = NULL; | |
2854 | size_t root_hash_size, root_hash_sig_size; | |
2855 | _cleanup_free_ char *buf = NULL; | |
2856 | PartitionDesignator d; | |
2857 | DissectedPartition *p; | |
2858 | JsonVariant *rh, *sig; | |
2859 | ssize_t n; | |
2860 | char *e; | |
2861 | int r; | |
2862 | ||
2863 | assert(m); | |
2864 | assert(fd >= 0); | |
2865 | assert(verity); | |
2866 | ||
2867 | if (verity->root_hash && verity->root_hash_sig) /* Already loaded? */ | |
2868 | return 0; | |
2869 | ||
2870 | r = getenv_bool_secure("SYSTEMD_DISSECT_VERITY_EMBEDDED"); | |
2871 | if (r < 0 && r != -ENXIO) | |
2872 | log_debug_errno(r, "Failed to parse $SYSTEMD_DISSECT_VERITY_EMBEDDED, ignoring: %m"); | |
2873 | if (r == 0) | |
2874 | return 0; | |
2875 | ||
dd894023 | 2876 | d = partition_verity_sig_of(verity->designator < 0 ? PARTITION_ROOT : verity->designator); |
88b3300f LP |
2877 | assert(d >= 0); |
2878 | ||
2879 | p = m->partitions + d; | |
2880 | if (!p->found) | |
2881 | return 0; | |
2882 | if (p->offset == UINT64_MAX || p->size == UINT64_MAX) | |
2883 | return -EINVAL; | |
2884 | ||
2885 | if (p->size > 4*1024*1024) /* Signature data cannot possible be larger than 4M, refuse that */ | |
2886 | return -EFBIG; | |
2887 | ||
2888 | buf = new(char, p->size+1); | |
2889 | if (!buf) | |
2890 | return -ENOMEM; | |
2891 | ||
2892 | n = pread(fd, buf, p->size, p->offset); | |
2893 | if (n < 0) | |
2894 | return -ENOMEM; | |
2895 | if ((uint64_t) n != p->size) | |
2896 | return -EIO; | |
2897 | ||
2898 | e = memchr(buf, 0, p->size); | |
2899 | if (e) { | |
2900 | /* If we found a NUL byte then the rest of the data must be NUL too */ | |
2901 | if (!memeqzero(e, p->size - (e - buf))) | |
2902 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature data contains embedded NUL byte."); | |
2903 | } else | |
2904 | buf[p->size] = 0; | |
2905 | ||
2906 | r = json_parse(buf, 0, &v, NULL, NULL); | |
2907 | if (r < 0) | |
2908 | return log_debug_errno(r, "Failed to parse signature JSON data: %m"); | |
2909 | ||
2910 | rh = json_variant_by_key(v, "rootHash"); | |
2911 | if (!rh) | |
2912 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature JSON object lacks 'rootHash' field."); | |
2913 | if (!json_variant_is_string(rh)) | |
2914 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "'rootHash' field of signature JSON object is not a string."); | |
2915 | ||
2916 | r = unhexmem(json_variant_string(rh), SIZE_MAX, &root_hash, &root_hash_size); | |
2917 | if (r < 0) | |
2918 | return log_debug_errno(r, "Failed to parse root hash field: %m"); | |
2919 | ||
2920 | /* Check if specified root hash matches if it is specified */ | |
2921 | if (verity->root_hash && | |
2922 | memcmp_nn(verity->root_hash, verity->root_hash_size, root_hash, root_hash_size) != 0) { | |
2923 | _cleanup_free_ char *a = NULL, *b = NULL; | |
2924 | ||
2925 | a = hexmem(root_hash, root_hash_size); | |
2926 | b = hexmem(verity->root_hash, verity->root_hash_size); | |
2927 | ||
2928 | return log_debug_errno(r, "Root hash in signature JSON data (%s) doesn't match configured hash (%s).", strna(a), strna(b)); | |
2929 | } | |
2930 | ||
2931 | sig = json_variant_by_key(v, "signature"); | |
2932 | if (!sig) | |
2933 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Signature JSON object lacks 'signature' field."); | |
2934 | if (!json_variant_is_string(sig)) | |
2935 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "'signature' field of signature JSON object is not a string."); | |
2936 | ||
2937 | r = unbase64mem(json_variant_string(sig), SIZE_MAX, &root_hash_sig, &root_hash_sig_size); | |
2938 | if (r < 0) | |
2939 | return log_debug_errno(r, "Failed to parse signature field: %m"); | |
2940 | ||
2941 | free_and_replace(verity->root_hash, root_hash); | |
2942 | verity->root_hash_size = root_hash_size; | |
2943 | ||
2944 | free_and_replace(verity->root_hash_sig, root_hash_sig); | |
2945 | verity->root_hash_sig_size = root_hash_sig_size; | |
2946 | ||
2947 | return 1; | |
2948 | } | |
2949 | ||
22847508 | 2950 | int dissected_image_acquire_metadata(DissectedImage *m, DissectImageFlags extra_flags) { |
3b925504 LP |
2951 | |
2952 | enum { | |
2953 | META_HOSTNAME, | |
2954 | META_MACHINE_ID, | |
2955 | META_MACHINE_INFO, | |
2956 | META_OS_RELEASE, | |
fab22946 | 2957 | META_INITRD_RELEASE, |
7718ac97 | 2958 | META_EXTENSION_RELEASE, |
a4e0d617 | 2959 | META_HAS_INIT_SYSTEM, |
3b925504 LP |
2960 | _META_MAX, |
2961 | }; | |
2962 | ||
9a4b883b | 2963 | static const char *const paths[_META_MAX] = { |
7718ac97 LB |
2964 | [META_HOSTNAME] = "/etc/hostname\0", |
2965 | [META_MACHINE_ID] = "/etc/machine-id\0", | |
2966 | [META_MACHINE_INFO] = "/etc/machine-info\0", | |
9a4b883b | 2967 | [META_OS_RELEASE] = ("/etc/os-release\0" |
22847508 | 2968 | "/usr/lib/os-release\0"), |
fab22946 LP |
2969 | [META_INITRD_RELEASE] = ("/etc/initrd-release\0" |
2970 | "/usr/lib/initrd-release\0"), | |
a4e0d617 LP |
2971 | [META_EXTENSION_RELEASE] = "extension-release\0", /* Used only for logging. */ |
2972 | [META_HAS_INIT_SYSTEM] = "has-init-system\0", /* ditto */ | |
3b925504 LP |
2973 | }; |
2974 | ||
fab22946 | 2975 | _cleanup_strv_free_ char **machine_info = NULL, **os_release = NULL, **initrd_release = NULL, **extension_release = NULL; |
19ee48a6 | 2976 | _cleanup_close_pair_ int error_pipe[2] = PIPE_EBADF; |
3b925504 LP |
2977 | _cleanup_(rmdir_and_freep) char *t = NULL; |
2978 | _cleanup_(sigkill_waitp) pid_t child = 0; | |
2979 | sd_id128_t machine_id = SD_ID128_NULL; | |
2980 | _cleanup_free_ char *hostname = NULL; | |
67f63ee5 | 2981 | unsigned n_meta_initialized = 0; |
af8219d5 | 2982 | int fds[2 * _META_MAX], r, v; |
a4e0d617 | 2983 | int has_init_system = -1; |
af8219d5 | 2984 | ssize_t n; |
3b925504 LP |
2985 | |
2986 | BLOCK_SIGNALS(SIGCHLD); | |
2987 | ||
2988 | assert(m); | |
2989 | ||
7718ac97 | 2990 | for (; n_meta_initialized < _META_MAX; n_meta_initialized ++) { |
d9119c00 | 2991 | if (!paths[n_meta_initialized]) { |
254d1313 | 2992 | fds[2*n_meta_initialized] = fds[2*n_meta_initialized+1] = -EBADF; |
7718ac97 | 2993 | continue; |
d9119c00 LP |
2994 | } |
2995 | ||
3b925504 LP |
2996 | if (pipe2(fds + 2*n_meta_initialized, O_CLOEXEC) < 0) { |
2997 | r = -errno; | |
2998 | goto finish; | |
2999 | } | |
7718ac97 | 3000 | } |
3b925504 LP |
3001 | |
3002 | r = mkdtemp_malloc("/tmp/dissect-XXXXXX", &t); | |
3003 | if (r < 0) | |
3004 | goto finish; | |
3005 | ||
af8219d5 LP |
3006 | if (pipe2(error_pipe, O_CLOEXEC) < 0) { |
3007 | r = -errno; | |
3008 | goto finish; | |
3009 | } | |
3010 | ||
e2047ba9 | 3011 | r = safe_fork("(sd-dissect)", FORK_RESET_SIGNALS|FORK_DEATHSIG|FORK_NEW_MOUNTNS|FORK_MOUNTNS_SLAVE, &child); |
be39f6ee | 3012 | if (r < 0) |
3b925504 | 3013 | goto finish; |
be39f6ee | 3014 | if (r == 0) { |
a4e0d617 | 3015 | /* Child in a new mount namespace */ |
af8219d5 LP |
3016 | error_pipe[0] = safe_close(error_pipe[0]); |
3017 | ||
7cf66030 LP |
3018 | r = dissected_image_mount( |
3019 | m, | |
3020 | t, | |
3021 | UID_INVALID, | |
21b61b1d | 3022 | UID_INVALID, |
22847508 ZJS |
3023 | extra_flags | |
3024 | DISSECT_IMAGE_READ_ONLY | | |
3025 | DISSECT_IMAGE_MOUNT_ROOT_ONLY | | |
7cf66030 | 3026 | DISSECT_IMAGE_USR_NO_ROOT); |
429d4e41 LP |
3027 | if (r < 0) { |
3028 | log_debug_errno(r, "Failed to mount dissected image: %m"); | |
03ae68f4 | 3029 | goto inner_fail; |
429d4e41 | 3030 | } |
3b925504 | 3031 | |
67f63ee5 | 3032 | for (unsigned k = 0; k < _META_MAX; k++) { |
37e44c3f | 3033 | _cleanup_close_ int fd = -ENOENT; |
3b925504 | 3034 | |
7718ac97 LB |
3035 | if (!paths[k]) |
3036 | continue; | |
3037 | ||
3b925504 LP |
3038 | fds[2*k] = safe_close(fds[2*k]); |
3039 | ||
a4e0d617 LP |
3040 | switch (k) { |
3041 | ||
3042 | case META_EXTENSION_RELEASE: | |
9a4b883b LB |
3043 | /* As per the os-release spec, if the image is an extension it will have a file |
3044 | * named after the image name in extension-release.d/ - we use the image name | |
3045 | * and try to resolve it with the extension-release helpers, as sometimes | |
3046 | * the image names are mangled on deployment and do not match anymore. | |
3047 | * Unlike other paths this is not fixed, and the image name | |
3048 | * can be mangled on deployment, so by calling into the helper | |
3049 | * we allow a fallback that matches on the first extension-release | |
3050 | * file found in the directory, if one named after the image cannot | |
3051 | * be found first. */ | |
06768b90 | 3052 | r = open_extension_release(t, m->image_name, /* relax_extension_release_check= */ false, NULL, &fd); |
9a4b883b LB |
3053 | if (r < 0) |
3054 | fd = r; /* Propagate the error. */ | |
a4e0d617 LP |
3055 | break; |
3056 | ||
3057 | case META_HAS_INIT_SYSTEM: { | |
3058 | bool found = false; | |
a4e0d617 LP |
3059 | |
3060 | FOREACH_STRING(init, | |
3061 | "/usr/lib/systemd/systemd", /* systemd on /usr merged system */ | |
3062 | "/lib/systemd/systemd", /* systemd on /usr non-merged systems */ | |
3063 | "/sbin/init") { /* traditional path the Linux kernel invokes */ | |
3064 | ||
3065 | r = chase_symlinks(init, t, CHASE_PREFIX_ROOT, NULL, NULL); | |
3066 | if (r < 0) { | |
3067 | if (r != -ENOENT) | |
3068 | log_debug_errno(r, "Failed to resolve %s, ignoring: %m", init); | |
3069 | } else { | |
3070 | found = true; | |
3071 | break; | |
3072 | } | |
3073 | } | |
3074 | ||
3075 | r = loop_write(fds[2*k+1], &found, sizeof(found), false); | |
3076 | if (r < 0) | |
3077 | goto inner_fail; | |
3078 | ||
3079 | continue; | |
3080 | } | |
3081 | ||
3082 | default: | |
9a4b883b LB |
3083 | NULSTR_FOREACH(p, paths[k]) { |
3084 | fd = chase_symlinks_and_open(p, t, CHASE_PREFIX_ROOT, O_RDONLY|O_CLOEXEC|O_NOCTTY, NULL); | |
3085 | if (fd >= 0) | |
3086 | break; | |
3087 | } | |
a4e0d617 LP |
3088 | } |
3089 | ||
36952d19 LP |
3090 | if (fd < 0) { |
3091 | log_debug_errno(fd, "Failed to read %s file of image, ignoring: %m", paths[k]); | |
37e44c3f | 3092 | fds[2*k+1] = safe_close(fds[2*k+1]); |
3b925504 | 3093 | continue; |
36952d19 | 3094 | } |
3b925504 | 3095 | |
f5fbe71d | 3096 | r = copy_bytes(fd, fds[2*k+1], UINT64_MAX, 0); |
a4e0d617 LP |
3097 | if (r < 0) |
3098 | goto inner_fail; | |
3b925504 LP |
3099 | |
3100 | fds[2*k+1] = safe_close(fds[2*k+1]); | |
3101 | } | |
3102 | ||
3103 | _exit(EXIT_SUCCESS); | |
a4e0d617 LP |
3104 | |
3105 | inner_fail: | |
03ae68f4 | 3106 | /* Let parent know the error */ |
a4e0d617 LP |
3107 | (void) write(error_pipe[1], &r, sizeof(r)); |
3108 | _exit(EXIT_FAILURE); | |
3b925504 LP |
3109 | } |
3110 | ||
af8219d5 LP |
3111 | error_pipe[1] = safe_close(error_pipe[1]); |
3112 | ||
67f63ee5 | 3113 | for (unsigned k = 0; k < _META_MAX; k++) { |
3b925504 LP |
3114 | _cleanup_fclose_ FILE *f = NULL; |
3115 | ||
7718ac97 LB |
3116 | if (!paths[k]) |
3117 | continue; | |
3118 | ||
3b925504 LP |
3119 | fds[2*k+1] = safe_close(fds[2*k+1]); |
3120 | ||
4fa744a3 | 3121 | f = take_fdopen(&fds[2*k], "r"); |
3b925504 LP |
3122 | if (!f) { |
3123 | r = -errno; | |
3124 | goto finish; | |
3125 | } | |
3126 | ||
3b925504 LP |
3127 | switch (k) { |
3128 | ||
3129 | case META_HOSTNAME: | |
3130 | r = read_etc_hostname_stream(f, &hostname); | |
3131 | if (r < 0) | |
f6048e5e | 3132 | log_debug_errno(r, "Failed to read /etc/hostname of image: %m"); |
3b925504 LP |
3133 | |
3134 | break; | |
3135 | ||
3136 | case META_MACHINE_ID: { | |
3137 | _cleanup_free_ char *line = NULL; | |
3138 | ||
3139 | r = read_line(f, LONG_LINE_MAX, &line); | |
3140 | if (r < 0) | |
f6048e5e | 3141 | log_debug_errno(r, "Failed to read /etc/machine-id of image: %m"); |
3b925504 LP |
3142 | else if (r == 33) { |
3143 | r = sd_id128_from_string(line, &machine_id); | |
3144 | if (r < 0) | |
3145 | log_debug_errno(r, "Image contains invalid /etc/machine-id: %s", line); | |
3146 | } else if (r == 0) | |
f6048e5e | 3147 | log_debug("/etc/machine-id file of image is empty."); |
ab763cb2 | 3148 | else if (streq(line, "uninitialized")) |
f6048e5e | 3149 | log_debug("/etc/machine-id file of image is uninitialized (likely aborted first boot)."); |
3b925504 | 3150 | else |
f6048e5e | 3151 | log_debug("/etc/machine-id file of image has unexpected length %i.", r); |
3b925504 LP |
3152 | |
3153 | break; | |
3154 | } | |
3155 | ||
3156 | case META_MACHINE_INFO: | |
aa8fbc74 | 3157 | r = load_env_file_pairs(f, "machine-info", &machine_info); |
3b925504 | 3158 | if (r < 0) |
f6048e5e | 3159 | log_debug_errno(r, "Failed to read /etc/machine-info of image: %m"); |
3b925504 LP |
3160 | |
3161 | break; | |
3162 | ||
3163 | case META_OS_RELEASE: | |
aa8fbc74 | 3164 | r = load_env_file_pairs(f, "os-release", &os_release); |
3b925504 | 3165 | if (r < 0) |
f6048e5e | 3166 | log_debug_errno(r, "Failed to read OS release file of image: %m"); |
3b925504 LP |
3167 | |
3168 | break; | |
7718ac97 | 3169 | |
fab22946 LP |
3170 | case META_INITRD_RELEASE: |
3171 | r = load_env_file_pairs(f, "initrd-release", &initrd_release); | |
3172 | if (r < 0) | |
3173 | log_debug_errno(r, "Failed to read initrd release file of image: %m"); | |
3174 | ||
3175 | break; | |
3176 | ||
7718ac97 LB |
3177 | case META_EXTENSION_RELEASE: |
3178 | r = load_env_file_pairs(f, "extension-release", &extension_release); | |
3179 | if (r < 0) | |
f6048e5e | 3180 | log_debug_errno(r, "Failed to read extension release file of image: %m"); |
7718ac97 LB |
3181 | |
3182 | break; | |
a4e0d617 LP |
3183 | |
3184 | case META_HAS_INIT_SYSTEM: { | |
3185 | bool b = false; | |
3186 | size_t nr; | |
3187 | ||
3188 | errno = 0; | |
3189 | nr = fread(&b, 1, sizeof(b), f); | |
3190 | if (nr != sizeof(b)) | |
3191 | log_debug_errno(errno_or_else(EIO), "Failed to read has-init-system boolean: %m"); | |
3192 | else | |
3193 | has_init_system = b; | |
3194 | ||
3195 | break; | |
3196 | }} | |
3b925504 LP |
3197 | } |
3198 | ||
2e87a1fd | 3199 | r = wait_for_terminate_and_check("(sd-dissect)", child, 0); |
3b925504 | 3200 | child = 0; |
2e87a1fd | 3201 | if (r < 0) |
af8219d5 LP |
3202 | return r; |
3203 | ||
3204 | n = read(error_pipe[0], &v, sizeof(v)); | |
3205 | if (n < 0) | |
3206 | return -errno; | |
3207 | if (n == sizeof(v)) | |
3208 | return v; /* propagate error sent to us from child */ | |
3209 | if (n != 0) | |
3210 | return -EIO; | |
3211 | ||
2e87a1fd LP |
3212 | if (r != EXIT_SUCCESS) |
3213 | return -EPROTO; | |
3b925504 LP |
3214 | |
3215 | free_and_replace(m->hostname, hostname); | |
3216 | m->machine_id = machine_id; | |
3217 | strv_free_and_replace(m->machine_info, machine_info); | |
3218 | strv_free_and_replace(m->os_release, os_release); | |
fab22946 | 3219 | strv_free_and_replace(m->initrd_release, initrd_release); |
7718ac97 | 3220 | strv_free_and_replace(m->extension_release, extension_release); |
a4e0d617 | 3221 | m->has_init_system = has_init_system; |
3b925504 LP |
3222 | |
3223 | finish: | |
67f63ee5 | 3224 | for (unsigned k = 0; k < n_meta_initialized; k++) |
3b925504 LP |
3225 | safe_close_pair(fds + 2*k); |
3226 | ||
3227 | return r; | |
3228 | } | |
3229 | ||
2348043f LP |
3230 | Architecture dissected_image_architecture(DissectedImage *img) { |
3231 | assert(img); | |
3232 | ||
3233 | if (img->partitions[PARTITION_ROOT].found && | |
3234 | img->partitions[PARTITION_ROOT].architecture >= 0) | |
3235 | return img->partitions[PARTITION_ROOT].architecture; | |
3236 | ||
3237 | if (img->partitions[PARTITION_USR].found && | |
3238 | img->partitions[PARTITION_USR].architecture >= 0) | |
3239 | return img->partitions[PARTITION_USR].architecture; | |
3240 | ||
3241 | return _ARCHITECTURE_INVALID; | |
3242 | } | |
3243 | ||
1e63dc4f YW |
3244 | int dissect_loop_device( |
3245 | LoopDevice *loop, | |
3246 | const VeritySettings *verity, | |
3247 | const MountOptions *mount_options, | |
3248 | DissectImageFlags flags, | |
3249 | DissectedImage **ret) { | |
3250 | ||
08f14be4 | 3251 | #if HAVE_BLKID |
1e63dc4f YW |
3252 | _cleanup_(dissected_image_unrefp) DissectedImage *m = NULL; |
3253 | int r; | |
3254 | ||
3255 | assert(loop); | |
3256 | assert(ret); | |
3257 | ||
08f14be4 | 3258 | r = dissected_image_new(loop->backing_file ?: loop->node, &m); |
1e63dc4f YW |
3259 | if (r < 0) |
3260 | return r; | |
3261 | ||
3262 | m->loop = loop_device_ref(loop); | |
22ee78a8 | 3263 | m->sector_size = m->loop->sector_size; |
1e63dc4f | 3264 | |
73d88b80 | 3265 | r = dissect_image(m, loop->fd, loop->node, verity, mount_options, flags); |
08f14be4 | 3266 | if (r < 0) |
08f14be4 YW |
3267 | return r; |
3268 | ||
1e63dc4f YW |
3269 | *ret = TAKE_PTR(m); |
3270 | return 0; | |
08f14be4 YW |
3271 | #else |
3272 | return -EOPNOTSUPP; | |
3273 | #endif | |
1e63dc4f YW |
3274 | } |
3275 | ||
bad31660 | 3276 | int dissect_loop_device_and_warn( |
1e63dc4f | 3277 | LoopDevice *loop, |
89e62e0b | 3278 | const VeritySettings *verity, |
18d73705 | 3279 | const MountOptions *mount_options, |
4526113f LP |
3280 | DissectImageFlags flags, |
3281 | DissectedImage **ret) { | |
3282 | ||
64dd3a24 | 3283 | const char *name; |
4526113f LP |
3284 | int r; |
3285 | ||
bad31660 YW |
3286 | assert(loop); |
3287 | assert(loop->fd >= 0); | |
4526113f | 3288 | |
64dd3a24 | 3289 | name = ASSERT_PTR(loop->backing_file ?: loop->node); |
4526113f | 3290 | |
369de26f | 3291 | r = dissect_loop_device(loop, verity, mount_options, flags, ret); |
4526113f LP |
3292 | switch (r) { |
3293 | ||
3294 | case -EOPNOTSUPP: | |
3295 | return log_error_errno(r, "Dissecting images is not supported, compiled without blkid support."); | |
3296 | ||
3297 | case -ENOPKG: | |
48084df6 ZJS |
3298 | return log_error_errno(r, "%s: Couldn't identify a suitable partition table or file system.", name); |
3299 | ||
3300 | case -ENOMEDIUM: | |
3301 | return log_error_errno(r, "%s: The image does not pass validation.", name); | |
4526113f LP |
3302 | |
3303 | case -EADDRNOTAVAIL: | |
48084df6 | 3304 | return log_error_errno(r, "%s: No root partition for specified root hash found.", name); |
4526113f LP |
3305 | |
3306 | case -ENOTUNIQ: | |
48084df6 | 3307 | return log_error_errno(r, "%s: Multiple suitable root partitions found in image.", name); |
4526113f LP |
3308 | |
3309 | case -ENXIO: | |
48084df6 | 3310 | return log_error_errno(r, "%s: No suitable root partition found in image.", name); |
4526113f LP |
3311 | |
3312 | case -EPROTONOSUPPORT: | |
3313 | return log_error_errno(r, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name); | |
3314 | ||
48084df6 ZJS |
3315 | case -ENOTBLK: |
3316 | return log_error_errno(r, "%s: Image is not a block device.", name); | |
3317 | ||
a94aa2b9 LP |
3318 | case -EBADR: |
3319 | return log_error_errno(r, | |
3320 | "Combining partitioned images (such as '%s') with external Verity data (such as '%s') not supported. " | |
3321 | "(Consider setting $SYSTEMD_DISSECT_VERITY_SIDECAR=0 to disable automatic discovery of external Verity data.)", | |
3322 | name, strna(verity ? verity->data_path : NULL)); | |
3323 | ||
4526113f LP |
3324 | default: |
3325 | if (r < 0) | |
3326 | return log_error_errno(r, "Failed to dissect image '%s': %m", name); | |
3327 | ||
3328 | return r; | |
3329 | } | |
3330 | } | |
3331 | ||
49536766 LP |
3332 | bool dissected_image_verity_candidate(const DissectedImage *image, PartitionDesignator partition_designator) { |
3333 | assert(image); | |
3334 | ||
3335 | /* Checks if this partition could theoretically do Verity. For non-partitioned images this only works | |
3336 | * if there's an external verity file supplied, for which we can consult .has_verity. For partitioned | |
3337 | * images we only check the partition type. | |
3338 | * | |
3339 | * This call is used to decide whether to suppress or show a verity column in tabular output of the | |
3340 | * image. */ | |
3341 | ||
e7cbe5cb | 3342 | if (image->single_file_system) |
c3c88d67 | 3343 | return partition_designator == PARTITION_ROOT && image->has_verity; |
e7cbe5cb | 3344 | |
dd894023 | 3345 | return partition_verity_of(partition_designator) >= 0; |
e7cbe5cb LB |
3346 | } |
3347 | ||
49536766 LP |
3348 | bool dissected_image_verity_ready(const DissectedImage *image, PartitionDesignator partition_designator) { |
3349 | PartitionDesignator k; | |
3350 | ||
3351 | assert(image); | |
3352 | ||
3353 | /* Checks if this partition has verity data available that we can activate. For non-partitioned this | |
3354 | * works for the root partition, for others only if the associated verity partition was found. */ | |
3355 | ||
3356 | if (!image->verity_ready) | |
3357 | return false; | |
e7cbe5cb LB |
3358 | |
3359 | if (image->single_file_system) | |
49536766 | 3360 | return partition_designator == PARTITION_ROOT; |
e7cbe5cb | 3361 | |
dd894023 | 3362 | k = partition_verity_of(partition_designator); |
e7cbe5cb LB |
3363 | return k >= 0 && image->partitions[k].found; |
3364 | } | |
3365 | ||
8ee9615e LP |
3366 | bool dissected_image_verity_sig_ready(const DissectedImage *image, PartitionDesignator partition_designator) { |
3367 | PartitionDesignator k; | |
3368 | ||
3369 | assert(image); | |
3370 | ||
3371 | /* Checks if this partition has verity signature data available that we can use. */ | |
3372 | ||
3373 | if (!image->verity_sig_ready) | |
3374 | return false; | |
3375 | ||
3376 | if (image->single_file_system) | |
3377 | return partition_designator == PARTITION_ROOT; | |
3378 | ||
dd894023 | 3379 | k = partition_verity_sig_of(partition_designator); |
8ee9615e LP |
3380 | return k >= 0 && image->partitions[k].found; |
3381 | } | |
3382 | ||
18d73705 LB |
3383 | MountOptions* mount_options_free_all(MountOptions *options) { |
3384 | MountOptions *m; | |
3385 | ||
3386 | while ((m = options)) { | |
3387 | LIST_REMOVE(mount_options, options, m); | |
3388 | free(m->options); | |
3389 | free(m); | |
3390 | } | |
3391 | ||
3392 | return NULL; | |
3393 | } | |
3394 | ||
569a0e42 | 3395 | const char* mount_options_from_designator(const MountOptions *options, PartitionDesignator designator) { |
f5215bc8 | 3396 | LIST_FOREACH(mount_options, m, options) |
9ece6444 | 3397 | if (designator == m->partition_designator && !isempty(m->options)) |
18d73705 | 3398 | return m->options; |
6aa05ebd | 3399 | |
18d73705 LB |
3400 | return NULL; |
3401 | } | |
3402 | ||
6aa05ebd LP |
3403 | int mount_image_privately_interactively( |
3404 | const char *image, | |
3405 | DissectImageFlags flags, | |
3406 | char **ret_directory, | |
a133d2c3 | 3407 | int *ret_dir_fd, |
e330f97a | 3408 | LoopDevice **ret_loop_device) { |
6aa05ebd | 3409 | |
27ec815e | 3410 | _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT; |
6aa05ebd | 3411 | _cleanup_(loop_device_unrefp) LoopDevice *d = NULL; |
6aa05ebd LP |
3412 | _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL; |
3413 | _cleanup_(rmdir_and_freep) char *created_dir = NULL; | |
3414 | _cleanup_free_ char *temp = NULL; | |
3415 | int r; | |
3416 | ||
3417 | /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This | |
3418 | * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image | |
3419 | * easily. */ | |
3420 | ||
3421 | assert(image); | |
3422 | assert(ret_directory); | |
3423 | assert(ret_loop_device); | |
6aa05ebd | 3424 | |
0b75493d | 3425 | /* We intend to mount this right-away, hence add the partitions if needed and pin them. */ |
73d88b80 LP |
3426 | flags |= DISSECT_IMAGE_ADD_PARTITION_DEVICES | |
3427 | DISSECT_IMAGE_PIN_PARTITION_DEVICES; | |
3428 | ||
27ec815e LP |
3429 | r = verity_settings_load(&verity, image, NULL, NULL); |
3430 | if (r < 0) | |
3431 | return log_error_errno(r, "Failed to load root hash data: %m"); | |
3432 | ||
6aa05ebd LP |
3433 | r = tempfn_random_child(NULL, program_invocation_short_name, &temp); |
3434 | if (r < 0) | |
3435 | return log_error_errno(r, "Failed to generate temporary mount directory: %m"); | |
3436 | ||
3437 | r = loop_device_make_by_path( | |
3438 | image, | |
ef9c184d | 3439 | FLAGS_SET(flags, DISSECT_IMAGE_DEVICE_READ_ONLY) ? O_RDONLY : O_RDWR, |
22ee78a8 | 3440 | /* sector_size= */ UINT32_MAX, |
6aa05ebd | 3441 | FLAGS_SET(flags, DISSECT_IMAGE_NO_PARTITION_TABLE) ? 0 : LO_FLAGS_PARTSCAN, |
7f52206a | 3442 | LOCK_SH, |
6aa05ebd LP |
3443 | &d); |
3444 | if (r < 0) | |
7b87fe4c | 3445 | return log_error_errno(r, "Failed to set up loopback device for %s: %m", image); |
6aa05ebd | 3446 | |
64dd3a24 | 3447 | r = dissect_loop_device_and_warn(d, &verity, NULL, flags, &dissected_image); |
6aa05ebd LP |
3448 | if (r < 0) |
3449 | return r; | |
3450 | ||
88b3300f LP |
3451 | r = dissected_image_load_verity_sig_partition(dissected_image, d->fd, &verity); |
3452 | if (r < 0) | |
3453 | return r; | |
3454 | ||
e330f97a | 3455 | r = dissected_image_decrypt_interactively(dissected_image, NULL, &verity, flags); |
6aa05ebd LP |
3456 | if (r < 0) |
3457 | return r; | |
3458 | ||
3459 | r = detach_mount_namespace(); | |
3460 | if (r < 0) | |
3461 | return log_error_errno(r, "Failed to detach mount namespace: %m"); | |
3462 | ||
3463 | r = mkdir_p(temp, 0700); | |
3464 | if (r < 0) | |
3465 | return log_error_errno(r, "Failed to create mount point: %m"); | |
3466 | ||
3467 | created_dir = TAKE_PTR(temp); | |
3468 | ||
21b61b1d | 3469 | r = dissected_image_mount_and_warn(dissected_image, created_dir, UID_INVALID, UID_INVALID, flags); |
6aa05ebd | 3470 | if (r < 0) |
af187ab2 | 3471 | return r; |
6aa05ebd | 3472 | |
41bc4849 LP |
3473 | r = loop_device_flock(d, LOCK_UN); |
3474 | if (r < 0) | |
3475 | return r; | |
3476 | ||
3044d343 YW |
3477 | r = dissected_image_relinquish(dissected_image); |
3478 | if (r < 0) | |
3479 | return log_error_errno(r, "Failed to relinquish DM and loopback block devices: %m"); | |
6aa05ebd | 3480 | |
a133d2c3 DDM |
3481 | if (ret_dir_fd) { |
3482 | _cleanup_close_ int dir_fd = -EBADF; | |
3483 | ||
3484 | dir_fd = open(created_dir, O_CLOEXEC|O_DIRECTORY); | |
3485 | if (dir_fd < 0) | |
3486 | return log_error_errno(errno, "Failed to open mount point directory: %m"); | |
3487 | ||
3488 | *ret_dir_fd = TAKE_FD(dir_fd); | |
3489 | } | |
3490 | ||
6aa05ebd LP |
3491 | *ret_directory = TAKE_PTR(created_dir); |
3492 | *ret_loop_device = TAKE_PTR(d); | |
6aa05ebd LP |
3493 | |
3494 | return 0; | |
3495 | } | |
3496 | ||
06768b90 LB |
3497 | static bool mount_options_relax_extension_release_checks(const MountOptions *options) { |
3498 | if (!options) | |
3499 | return false; | |
3500 | ||
3501 | return string_contains_word(mount_options_from_designator(options, PARTITION_ROOT), ",", "x-systemd.relax-extension-release-check") || | |
3502 | string_contains_word(mount_options_from_designator(options, PARTITION_USR), ",", "x-systemd.relax-extension-release-check") || | |
3503 | string_contains_word(options->options, ",", "x-systemd.relax-extension-release-check"); | |
3504 | } | |
3505 | ||
93f59701 | 3506 | int verity_dissect_and_mount( |
cedf5b1a | 3507 | int src_fd, |
93f59701 LB |
3508 | const char *src, |
3509 | const char *dest, | |
3510 | const MountOptions *options, | |
3511 | const char *required_host_os_release_id, | |
3512 | const char *required_host_os_release_version_id, | |
60c5f700 LP |
3513 | const char *required_host_os_release_sysext_level, |
3514 | const char *required_sysext_scope) { | |
93f59701 | 3515 | |
4beda316 | 3516 | _cleanup_(loop_device_unrefp) LoopDevice *loop_device = NULL; |
4beda316 LB |
3517 | _cleanup_(dissected_image_unrefp) DissectedImage *dissected_image = NULL; |
3518 | _cleanup_(verity_settings_done) VeritySettings verity = VERITY_SETTINGS_DEFAULT; | |
3519 | DissectImageFlags dissect_image_flags; | |
06768b90 | 3520 | bool relax_extension_release_check; |
4beda316 LB |
3521 | int r; |
3522 | ||
3523 | assert(src); | |
3524 | assert(dest); | |
3525 | ||
06768b90 LB |
3526 | relax_extension_release_check = mount_options_relax_extension_release_checks(options); |
3527 | ||
cedf5b1a | 3528 | /* We might get an FD for the image, but we use the original path to look for the dm-verity files */ |
4beda316 LB |
3529 | r = verity_settings_load(&verity, src, NULL, NULL); |
3530 | if (r < 0) | |
3531 | return log_debug_errno(r, "Failed to load root hash: %m"); | |
3532 | ||
06768b90 | 3533 | dissect_image_flags = (verity.data_path ? DISSECT_IMAGE_NO_PARTITION_TABLE : 0) | |
73d88b80 LP |
3534 | (relax_extension_release_check ? DISSECT_IMAGE_RELAX_SYSEXT_CHECK : 0) | |
3535 | DISSECT_IMAGE_ADD_PARTITION_DEVICES | | |
3536 | DISSECT_IMAGE_PIN_PARTITION_DEVICES; | |
4beda316 | 3537 | |
cedf5b1a LB |
3538 | /* Note that we don't use loop_device_make here, as the FD is most likely O_PATH which would not be |
3539 | * accepted by LOOP_CONFIGURE, so just let loop_device_make_by_path reopen it as a regular FD. */ | |
4beda316 | 3540 | r = loop_device_make_by_path( |
cedf5b1a | 3541 | src_fd >= 0 ? FORMAT_PROC_FD_PATH(src_fd) : src, |
22ee78a8 LP |
3542 | /* open_flags= */ -1, |
3543 | /* sector_size= */ UINT32_MAX, | |
4beda316 | 3544 | verity.data_path ? 0 : LO_FLAGS_PARTSCAN, |
7f52206a | 3545 | LOCK_SH, |
4beda316 LB |
3546 | &loop_device); |
3547 | if (r < 0) | |
3548 | return log_debug_errno(r, "Failed to create loop device for image: %m"); | |
3549 | ||
bad31660 YW |
3550 | r = dissect_loop_device( |
3551 | loop_device, | |
4beda316 LB |
3552 | &verity, |
3553 | options, | |
3554 | dissect_image_flags, | |
3555 | &dissected_image); | |
3556 | /* No partition table? Might be a single-filesystem image, try again */ | |
3557 | if (!verity.data_path && r == -ENOPKG) | |
bad31660 YW |
3558 | r = dissect_loop_device( |
3559 | loop_device, | |
4beda316 LB |
3560 | &verity, |
3561 | options, | |
75dc190d | 3562 | dissect_image_flags | DISSECT_IMAGE_NO_PARTITION_TABLE, |
4beda316 LB |
3563 | &dissected_image); |
3564 | if (r < 0) | |
3565 | return log_debug_errno(r, "Failed to dissect image: %m"); | |
3566 | ||
88b3300f LP |
3567 | r = dissected_image_load_verity_sig_partition(dissected_image, loop_device->fd, &verity); |
3568 | if (r < 0) | |
3569 | return r; | |
3570 | ||
4beda316 LB |
3571 | r = dissected_image_decrypt( |
3572 | dissected_image, | |
3573 | NULL, | |
3574 | &verity, | |
e330f97a | 3575 | dissect_image_flags); |
4beda316 LB |
3576 | if (r < 0) |
3577 | return log_debug_errno(r, "Failed to decrypt dissected image: %m"); | |
3578 | ||
3579 | r = mkdir_p_label(dest, 0755); | |
3580 | if (r < 0) | |
3581 | return log_debug_errno(r, "Failed to create destination directory %s: %m", dest); | |
3582 | r = umount_recursive(dest, 0); | |
3583 | if (r < 0) | |
3584 | return log_debug_errno(r, "Failed to umount under destination directory %s: %m", dest); | |
3585 | ||
21b61b1d | 3586 | r = dissected_image_mount(dissected_image, dest, UID_INVALID, UID_INVALID, dissect_image_flags); |
4beda316 LB |
3587 | if (r < 0) |
3588 | return log_debug_errno(r, "Failed to mount image: %m"); | |
3589 | ||
41bc4849 LP |
3590 | r = loop_device_flock(loop_device, LOCK_UN); |
3591 | if (r < 0) | |
3592 | return log_debug_errno(r, "Failed to unlock loopback device: %m"); | |
3593 | ||
93f59701 LB |
3594 | /* If we got os-release values from the caller, then we need to match them with the image's |
3595 | * extension-release.d/ content. Return -EINVAL if there's any mismatch. | |
3596 | * First, check the distro ID. If that matches, then check the new SYSEXT_LEVEL value if | |
37361f46 LB |
3597 | * available, or else fallback to VERSION_ID. If neither is present (eg: rolling release), |
3598 | * then a simple match on the ID will be performed. */ | |
8b2dcbbd | 3599 | if (required_host_os_release_id) { |
93f59701 LB |
3600 | _cleanup_strv_free_ char **extension_release = NULL; |
3601 | ||
d30d86b7 YW |
3602 | assert(!isempty(required_host_os_release_id)); |
3603 | ||
06768b90 | 3604 | r = load_extension_release_pairs(dest, dissected_image->image_name, relax_extension_release_check, &extension_release); |
93f59701 LB |
3605 | if (r < 0) |
3606 | return log_debug_errno(r, "Failed to parse image %s extension-release metadata: %m", dissected_image->image_name); | |
3607 | ||
3608 | r = extension_release_validate( | |
60c5f700 LP |
3609 | dissected_image->image_name, |
3610 | required_host_os_release_id, | |
3611 | required_host_os_release_version_id, | |
3612 | required_host_os_release_sysext_level, | |
3613 | required_sysext_scope, | |
3614 | extension_release); | |
93f59701 LB |
3615 | if (r == 0) |
3616 | return log_debug_errno(SYNTHETIC_ERRNO(ESTALE), "Image %s extension-release metadata does not match the root's", dissected_image->image_name); | |
3617 | if (r < 0) | |
3618 | return log_debug_errno(r, "Failed to compare image %s extension-release metadata with the root's os-release: %m", dissected_image->image_name); | |
3619 | } | |
3620 | ||
3044d343 YW |
3621 | r = dissected_image_relinquish(dissected_image); |
3622 | if (r < 0) | |
3623 | return log_debug_errno(r, "Failed to relinquish dissected image: %m"); | |
4beda316 LB |
3624 | |
3625 | return 0; | |
3626 | } |