1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
3 #if HAVE_VALGRIND_MEMCHECK_H
4 #include <valgrind/memcheck.h>
7 #include <linux/dm-ioctl.h>
8 #include <linux/loop.h>
10 #include <sys/mount.h>
11 #include <sys/prctl.h>
16 #include <openssl/err.h>
17 #include <openssl/pem.h>
18 #include <openssl/x509.h>
21 #include "sd-device.h"
24 #include "architecture.h"
25 #include "ask-password-api.h"
26 #include "blkid-util.h"
27 #include "blockdev-util.h"
28 #include "btrfs-util.h"
30 #include "conf-files.h"
31 #include "constants.h"
33 #include "cryptsetup-util.h"
34 #include "device-nodes.h"
35 #include "device-util.h"
36 #include "devnum-util.h"
37 #include "discover-image.h"
38 #include "dissect-image.h"
42 #include "extension-util.h"
46 #include "fsck-util.h"
48 #include "hexdecoct.h"
49 #include "hostname-setup.h"
50 #include "id128-util.h"
51 #include "import-util.h"
53 #include "missing_mount.h"
54 #include "mkdir-label.h"
55 #include "mount-util.h"
56 #include "mountpoint-util.h"
57 #include "namespace-util.h"
58 #include "nulstr-util.h"
59 #include "openssl-util.h"
61 #include "path-util.h"
62 #include "process-util.h"
63 #include "raw-clone.h"
64 #include "resize-fs.h"
65 #include "signal-util.h"
66 #include "sparse-endian.h"
67 #include "stat-util.h"
68 #include "stdio-util.h"
69 #include "string-table.h"
70 #include "string-util.h"
72 #include "tmpfile-util.h"
73 #include "udev-util.h"
74 #include "user-util.h"
75 #include "xattr-util.h"
77 /* how many times to wait for the device nodes to appear */
78 #define N_DEVICE_NODE_LIST_ATTEMPTS 10
80 int dissect_fstype_ok(const char *fstype
) {
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
88 e
= secure_getenv("SYSTEMD_DISSECT_FILE_SYSTEMS");
90 _cleanup_strv_free_
char **l
= NULL
;
92 l
= strv_split(e
, ":");
96 b
= strv_contains(l
, fstype
);
98 b
= STR_IN_SET(fstype
,
108 log_debug("File system type '%s' is not allowed to be mounted as result of automatic dissection.", fstype
);
112 int probe_sector_size(int fd
, uint32_t *ret
) {
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
;
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 */
134 assert_cc(sizeof(struct gpt_header
) == 92);
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];
146 n
= pread(fd
, sectors
, sizeof(sectors
), 0);
149 if (n
!= sizeof(sectors
)) /* too short? */
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
;
156 assert(sizeof(sectors
) >= sz
* 2);
157 p
= (struct gpt_header
*) (sectors
+ sz
);
159 if (memcmp(p
->signature
, (const char[8]) { 'E', 'F', 'I', ' ', 'P', 'A', 'R', 'T' }, 8) != 0)
162 if (le32toh(p
->revision
) != UINT32_C(0x00010000)) /* the only known revision of the spec: 1.0 */
165 if (le32toh(p
->header_size
) < sizeof(struct gpt_header
))
168 if (le32toh(p
->header_size
) > 4096) /* larger than a sector? something is off… */
171 if (le64toh(p
->my_lba
) != 1) /* this sector must claim to be at sector offset 1 */
175 return log_debug_errno(SYNTHETIC_ERRNO(ENOTUNIQ
),
176 "Detected valid partition table at offsets matching multiple sector sizes, refusing.");
182 log_debug("Determined sector size %" PRIu32
" based on discovered partition table.", found
);
184 return 1; /* indicate we *did* find it */
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 */
193 int probe_sector_size_prefer_ioctl(int fd
, uint32_t *ret
) {
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 */
202 if (fstat(fd
, &st
) < 0)
205 if (S_ISBLK(st
.st_mode
))
206 return blockdev_get_sector_size(fd
, ret
);
208 return probe_sector_size(fd
, ret
);
211 int probe_filesystem_full(
218 /* Try to find device content type and return it in *ret_fstype. If nothing is found,
219 * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and a
220 * different error otherwise. */
223 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
224 _cleanup_free_
char *path_by_fd
= NULL
;
225 _cleanup_close_
int fd_close
= -EBADF
;
229 assert(fd
>= 0 || path
);
233 fd_close
= open(path
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
241 r
= fd_get_path(fd
, &path_by_fd
);
248 if (size
== 0) /* empty size? nothing found! */
251 b
= blkid_new_probe();
255 /* The Linux kernel maintains separate block device caches for main ("whole") and partition block
256 * devices, which means making a change to one might not be reflected immediately when reading via
257 * the other. That's massively confusing when mixing accesses to such devices. Let's address this in
258 * a limited way: when probing a file system that is not at the beginning of the block device we
259 * apparently probe a partition via the main block device, and in that case let's first flush the
260 * main block device cache, so that we get the data that the per-partition block device last
263 * This only works under the assumption that any tools that write to the partition block devices
264 * issue an syncfs()/fsync() on the device after making changes. Typically file system formatting
265 * tools that write a superblock onto a partition block device do that, however. */
267 if (ioctl(fd
, BLKFLSBUF
, 0) < 0)
268 log_debug_errno(errno
, "Failed to flush block device cache, ignoring: %m");
271 r
= blkid_probe_set_device(
275 size
== UINT64_MAX
? 0 : size
); /* when blkid sees size=0 it understands "everything". We prefer using UINT64_MAX for that */
277 return errno_or_else(ENOMEM
);
279 blkid_probe_enable_superblocks(b
, 1);
280 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
);
283 r
= blkid_do_safeprobe(b
);
284 if (r
== _BLKID_SAFEPROBE_NOT_FOUND
)
286 if (r
== _BLKID_SAFEPROBE_AMBIGUOUS
)
287 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
),
288 "Results ambiguous for partition %s", path
);
289 if (r
== _BLKID_SAFEPROBE_ERROR
)
290 return log_debug_errno(errno_or_else(EIO
), "Failed to probe partition %s: %m", path
);
292 assert(r
== _BLKID_SAFEPROBE_FOUND
);
294 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
299 log_debug("Probed fstype '%s' on partition %s.", fstype
, path
);
310 log_debug("No type detected on partition %s", path
);
319 static int image_policy_may_use(
320 const ImagePolicy
*policy
,
321 PartitionDesignator designator
) {
323 PartitionPolicyFlags f
;
325 /* For each partition we find in the partition table do a first check if it may exist at all given
326 * the policy, or if it shall be ignored. */
328 f
= image_policy_get_exhaustively(policy
, designator
);
332 if ((f
& _PARTITION_POLICY_USE_MASK
) == PARTITION_POLICY_ABSENT
)
333 /* only flag set in policy is "absent"? then this partition may not exist at all */
334 return log_debug_errno(
335 SYNTHETIC_ERRNO(ERFKILL
),
336 "Partition of designator '%s' exists, but not allowed by policy, refusing.",
337 partition_designator_to_string(designator
));
338 if ((f
& _PARTITION_POLICY_USE_MASK
& ~PARTITION_POLICY_ABSENT
) == PARTITION_POLICY_UNUSED
) {
339 /* only "unused" or "unused" + "absent" are set? then don't use it */
340 log_debug("Partition of designator '%s' exists, and policy dictates to ignore it, doing so.",
341 partition_designator_to_string(designator
));
342 return false; /* ignore! */
345 return true; /* use! */
348 static int image_policy_check_protection(
349 const ImagePolicy
*policy
,
350 PartitionDesignator designator
,
351 PartitionPolicyFlags found_flags
) {
353 PartitionPolicyFlags policy_flags
;
355 /* Checks if the flags in the policy for the designated partition overlap the flags of what we found */
360 policy_flags
= image_policy_get_exhaustively(policy
, designator
);
361 if (policy_flags
< 0)
364 if ((found_flags
& policy_flags
) == 0) {
365 _cleanup_free_
char *found_flags_string
= NULL
, *policy_flags_string
= NULL
;
367 (void) partition_policy_flags_to_string(found_flags
, /* simplify= */ true, &found_flags_string
);
368 (void) partition_policy_flags_to_string(policy_flags
, /* simplify= */ true, &policy_flags_string
);
370 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL
), "Partition %s discovered with policy '%s' but '%s' was required, refusing.",
371 partition_designator_to_string(designator
),
372 strnull(found_flags_string
), strnull(policy_flags_string
));
378 static int image_policy_check_partition_flags(
379 const ImagePolicy
*policy
,
380 PartitionDesignator designator
,
381 uint64_t gpt_flags
) {
383 PartitionPolicyFlags policy_flags
;
386 /* Checks if the partition flags in the policy match reality */
388 policy_flags
= image_policy_get_exhaustively(policy
, designator
);
389 if (policy_flags
< 0)
392 b
= FLAGS_SET(gpt_flags
, SD_GPT_FLAG_READ_ONLY
);
393 if ((policy_flags
& _PARTITION_POLICY_READ_ONLY_MASK
) == (b
? PARTITION_POLICY_READ_ONLY_OFF
: PARTITION_POLICY_READ_ONLY_ON
))
394 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL
), "Partition %s has 'read-only' flag incorrectly set (must be %s, is %s), refusing.",
395 partition_designator_to_string(designator
),
396 one_zero(!b
), one_zero(b
));
398 b
= FLAGS_SET(gpt_flags
, SD_GPT_FLAG_GROWFS
);
399 if ((policy_flags
& _PARTITION_POLICY_GROWFS_MASK
) == (b
? PARTITION_POLICY_GROWFS_OFF
: PARTITION_POLICY_GROWFS_ON
))
400 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL
), "Partition %s has 'growfs' flag incorrectly set (must be %s, is %s), refusing.",
401 partition_designator_to_string(designator
),
402 one_zero(!b
), one_zero(b
));
407 static int dissected_image_probe_filesystems(
410 const ImagePolicy
*policy
) {
416 /* Fill in file system types if we don't know them yet. */
418 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
419 DissectedPartition
*p
= m
->partitions
+ i
;
420 PartitionPolicyFlags found_flags
;
426 /* If we have an fd referring to the partition block device, use that. Otherwise go
427 * via the whole block device or backing regular file, and read via offset. */
428 if (p
->mount_node_fd
>= 0)
429 r
= probe_filesystem_full(p
->mount_node_fd
, p
->node
, 0, UINT64_MAX
, &p
->fstype
);
431 r
= probe_filesystem_full(fd
, p
->node
, p
->offset
, p
->size
, &p
->fstype
);
436 if (streq_ptr(p
->fstype
, "crypto_LUKS")) {
438 found_flags
= PARTITION_POLICY_ENCRYPTED
; /* found this one, and its definitely encrypted */
440 /* found it, but it's definitely not encrypted, hence mask the encrypted flag, but
441 * set all other ways that indicate "present". */
442 found_flags
= PARTITION_POLICY_UNPROTECTED
|PARTITION_POLICY_VERITY
|PARTITION_POLICY_SIGNED
;
444 if (p
->fstype
&& fstype_is_ro(p
->fstype
))
450 /* We might have learnt more about the file system now (i.e. whether it is encrypted or not),
451 * hence we need to validate this against policy again, to see if the policy still matches
452 * with this new information. Note that image_policy_check_protection() will check for
453 * overlap between what's allowed in the policy and what we pass as 'found_policy' here. In
454 * the unencrypted case we thus might pass an overly unspecific mask here (i.e. unprotected
455 * OR verity OR signed), but that's fine since the earlier policy check already checked more
456 * specific which of those three cases where OK. Keep in mind that this function here only
457 * looks at specific partitions (and thus can only deduce encryption or not) but not the
458 * overall partition table (and thus cannot deduce verity or not). The earlier dissection
459 * checks already did the relevant checks that look at the whole partition table, and
460 * enforced policy there as needed. */
461 r
= image_policy_check_protection(policy
, i
, found_flags
);
469 static void check_partition_flags(
471 unsigned long long pflags
,
472 unsigned long long supported
) {
476 /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
477 pflags
&= ~(supported
|
478 SD_GPT_FLAG_REQUIRED_PARTITION
|
479 SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL
|
480 SD_GPT_FLAG_LEGACY_BIOS_BOOTABLE
);
485 /* If there are other bits set, then log about it, to make things discoverable */
486 for (unsigned i
= 0; i
< sizeof(pflags
) * 8; i
++) {
487 unsigned long long bit
= 1ULL << i
;
488 if (!FLAGS_SET(pflags
, bit
))
491 log_debug("Unexpected partition flag %llu set on %s!", bit
, node
);
495 static int dissected_image_new(const char *path
, DissectedImage
**ret
) {
496 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
497 _cleanup_free_
char *name
= NULL
;
503 _cleanup_free_
char *filename
= NULL
;
505 r
= path_extract_filename(path
, &filename
);
509 r
= raw_strip_suffixes(filename
, &name
);
513 if (!image_name_is_valid(name
)) {
514 log_debug("Image name %s is not valid, ignoring.", strna(name
));
519 m
= new(DissectedImage
, 1);
523 *m
= (DissectedImage
) {
524 .has_init_system
= -1,
525 .image_name
= TAKE_PTR(name
),
528 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++)
529 m
->partitions
[i
] = DISSECTED_PARTITION_NULL
;
536 static void dissected_partition_done(DissectedPartition
*p
) {
542 free(p
->decrypted_fstype
);
543 free(p
->decrypted_node
);
544 free(p
->mount_options
);
545 safe_close(p
->mount_node_fd
);
547 *p
= DISSECTED_PARTITION_NULL
;
551 static int make_partition_devname(
552 const char *whole_devname
,
555 DissectImageFlags flags
,
558 _cleanup_free_
char *s
= NULL
;
561 assert(whole_devname
);
562 assert(nr
!= 0); /* zero is not a valid partition nr */
565 if (!FLAGS_SET(flags
, DISSECT_IMAGE_DISKSEQ_DEVNODE
) || diskseq
== 0) {
567 /* Given a whole block device node name (e.g. /dev/sda or /dev/loop7) generate a partition
568 * device name (e.g. /dev/sda7 or /dev/loop7p5). The rule the kernel uses is simple: if whole
569 * block device node name ends in a digit, then suffix a 'p', followed by the partition
570 * number. Otherwise, just suffix the partition number without any 'p'. */
572 if (nr
< 0) { /* whole disk? */
573 s
= strdup(whole_devname
);
577 size_t l
= strlen(whole_devname
);
578 if (l
< 1) /* underflow check for the subtraction below */
581 bool need_p
= ascii_isdigit(whole_devname
[l
-1]); /* Last char a digit? */
583 if (asprintf(&s
, "%s%s%i", whole_devname
, need_p
? "p" : "", nr
) < 0)
587 if (nr
< 0) /* whole disk? */
588 r
= asprintf(&s
, "/dev/disk/by-diskseq/%" PRIu64
, diskseq
);
590 r
= asprintf(&s
, "/dev/disk/by-diskseq/%" PRIu64
"-part%i", diskseq
, nr
);
599 static int open_partition(
602 const LoopDevice
*loop
) {
604 _cleanup_(sd_device_unrefp
) sd_device
*dev
= NULL
;
605 _cleanup_close_
int fd
= -EBADF
;
612 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
616 /* Check if the block device is a child of (or equivalent to) the originally provided one. */
617 r
= block_device_new_from_fd(fd
, is_partition
? BLOCK_DEVICE_LOOKUP_WHOLE_DISK
: 0, &dev
);
621 r
= sd_device_get_devnum(dev
, &devnum
);
625 if (loop
->devno
!= devnum
)
628 /* Also check diskseq. */
629 if (loop
->diskseq
!= 0) {
632 r
= fd_get_diskseq(fd
, &diskseq
);
636 if (loop
->diskseq
!= diskseq
)
640 log_debug("Opened %s (fd=%i, whole_block_devnum=" DEVNUM_FORMAT_STR
", diskseq=%" PRIu64
").",
641 node
, fd
, DEVNUM_FORMAT_VAL(loop
->devno
), loop
->diskseq
);
645 static int compare_arch(Architecture a
, Architecture b
) {
649 if (a
== native_architecture())
652 if (b
== native_architecture())
655 #ifdef ARCHITECTURE_SECONDARY
656 if (a
== ARCHITECTURE_SECONDARY
)
659 if (b
== ARCHITECTURE_SECONDARY
)
666 static int dissect_image(
670 const VeritySettings
*verity
,
671 const MountOptions
*mount_options
,
672 const ImagePolicy
*policy
,
673 DissectImageFlags flags
) {
675 sd_id128_t root_uuid
= SD_ID128_NULL
, root_verity_uuid
= SD_ID128_NULL
;
676 sd_id128_t usr_uuid
= SD_ID128_NULL
, usr_verity_uuid
= SD_ID128_NULL
;
677 bool is_gpt
, is_mbr
, multiple_generic
= false,
678 generic_rw
= false, /* initialize to appease gcc */
679 generic_growfs
= false;
680 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
681 _cleanup_free_
char *generic_node
= NULL
;
682 sd_id128_t generic_uuid
= SD_ID128_NULL
;
683 const char *pttype
= NULL
, *sptuuid
= NULL
;
685 int r
, generic_nr
= -1, n_partitions
;
690 assert(!verity
|| verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
691 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
692 assert(!verity
|| verity
->root_hash_sig
|| verity
->root_hash_sig_size
== 0);
693 assert(!verity
|| (verity
->root_hash
|| !verity
->root_hash_sig
));
694 assert(!((flags
& DISSECT_IMAGE_GPT_ONLY
) && (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)));
695 assert(m
->sector_size
> 0);
697 /* Probes a disk image, and returns information about what it found in *ret.
699 * Returns -ENOPKG if no suitable partition table or file system could be found.
700 * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found.
701 * Returns -ENXIO if we couldn't find any partition suitable as root or /usr partition
702 * Returns -ENOTUNIQ if we only found multiple generic partitions and thus don't know what to do with that
703 * Returns -ERFKILL if image doesn't match image policy
704 * Returns -EBADR if verity data was provided externally for an image that has a GPT partition table (i.e. is not just a naked fs)
705 * Returns -EPROTONOSUPPORT if DISSECT_IMAGE_ADD_PARTITION_DEVICES is set but the block device does not have partition logic enabled
706 * Returns -ENOMSG if we didn't find a single usable partition (and DISSECT_IMAGE_REFUSE_EMPTY is set) */
708 uint64_t diskseq
= m
->loop
? m
->loop
->diskseq
: 0;
710 if (verity
&& verity
->root_hash
) {
711 sd_id128_t fsuuid
, vuuid
;
713 /* If a root hash is supplied, then we use the root partition that has a UUID that match the
714 * first 128-bit of the root hash. And we use the verity partition that has a UUID that match
715 * the final 128-bit. */
717 if (verity
->root_hash_size
< sizeof(sd_id128_t
))
720 memcpy(&fsuuid
, verity
->root_hash
, sizeof(sd_id128_t
));
721 memcpy(&vuuid
, (const uint8_t*) verity
->root_hash
+ verity
->root_hash_size
- sizeof(sd_id128_t
), sizeof(sd_id128_t
));
723 if (sd_id128_is_null(fsuuid
))
725 if (sd_id128_is_null(vuuid
))
728 /* If the verity data declares it's for the /usr partition, then search for that, in all
729 * other cases assume it's for the root partition. */
730 if (verity
->designator
== PARTITION_USR
) {
732 usr_verity_uuid
= vuuid
;
735 root_verity_uuid
= vuuid
;
739 b
= blkid_new_probe();
744 r
= blkid_probe_set_device(b
, fd
, 0, 0);
746 return errno_or_else(ENOMEM
);
749 r
= blkid_probe_set_sectorsize(b
, m
->sector_size
);
751 return errno_or_else(EIO
);
753 if ((flags
& DISSECT_IMAGE_GPT_ONLY
) == 0) {
754 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
755 blkid_probe_enable_superblocks(b
, 1);
756 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
|BLKID_SUBLKS_USAGE
|BLKID_SUBLKS_UUID
);
759 blkid_probe_enable_partitions(b
, 1);
760 blkid_probe_set_partitions_flags(b
, BLKID_PARTS_ENTRY_DETAILS
);
763 r
= blkid_do_safeprobe(b
);
764 if (r
== _BLKID_SAFEPROBE_ERROR
)
765 return errno_or_else(EIO
);
766 if (IN_SET(r
, _BLKID_SAFEPROBE_AMBIGUOUS
, _BLKID_SAFEPROBE_NOT_FOUND
))
767 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG
), "Failed to identify any partition table.");
769 assert(r
== _BLKID_SAFEPROBE_FOUND
);
771 if ((!(flags
& DISSECT_IMAGE_GPT_ONLY
) &&
772 (flags
& DISSECT_IMAGE_GENERIC_ROOT
)) ||
773 (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)) {
774 const char *usage
= NULL
;
776 /* If flags permit this, also allow using non-partitioned single-filesystem images */
778 (void) blkid_probe_lookup_value(b
, "USAGE", &usage
, NULL
);
779 if (STRPTR_IN_SET(usage
, "filesystem", "crypto")) {
780 _cleanup_free_
char *t
= NULL
, *n
= NULL
, *o
= NULL
;
781 const char *fstype
= NULL
, *options
= NULL
, *suuid
= NULL
;
782 _cleanup_close_
int mount_node_fd
= -EBADF
;
783 sd_id128_t uuid
= SD_ID128_NULL
;
784 PartitionPolicyFlags found_flags
;
787 /* OK, we have found a file system, that's our root partition then. */
789 r
= image_policy_may_use(policy
, PARTITION_ROOT
);
792 if (r
== 0) /* policy says ignore this, so we ignore it */
795 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
796 (void) blkid_probe_lookup_value(b
, "UUID", &suuid
, NULL
);
798 encrypted
= streq_ptr(fstype
, "crypto_LUKS");
800 if (verity_settings_data_covers(verity
, PARTITION_ROOT
))
801 found_flags
= verity
->root_hash_sig
? PARTITION_POLICY_SIGNED
: PARTITION_POLICY_VERITY
;
803 found_flags
= encrypted
? PARTITION_POLICY_ENCRYPTED
: PARTITION_POLICY_UNPROTECTED
;
805 r
= image_policy_check_protection(policy
, PARTITION_ROOT
, found_flags
);
809 r
= image_policy_check_partition_flags(policy
, PARTITION_ROOT
, 0); /* we have no gpt partition flags, hence check against all bits off */
813 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
814 mount_node_fd
= open_partition(devname
, /* is_partition = */ false, m
->loop
);
815 if (mount_node_fd
< 0)
816 return mount_node_fd
;
826 /* blkid will return FAT's serial number as UUID, hence it is quite possible
827 * that parsing this will fail. We'll ignore the ID, since it's just too
828 * short to be useful as tru identifier. */
829 r
= sd_id128_from_string(suuid
, &uuid
);
831 log_debug_errno(r
, "Failed to parse file system UUID '%s', ignoring: %m", suuid
);
834 r
= make_partition_devname(devname
, diskseq
, -1, flags
, &n
);
838 m
->single_file_system
= true;
839 m
->encrypted
= encrypted
;
841 m
->has_verity
= verity
&& verity
->data_path
;
842 m
->verity_ready
= verity_settings_data_covers(verity
, PARTITION_ROOT
);
844 m
->has_verity_sig
= false; /* signature not embedded, must be specified */
845 m
->verity_sig_ready
= m
->verity_ready
&& verity
->root_hash_sig
;
847 m
->image_uuid
= uuid
;
849 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
856 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
858 .rw
= !m
->verity_ready
&& !fstype_is_ro(fstype
),
860 .architecture
= _ARCHITECTURE_INVALID
,
861 .fstype
= TAKE_PTR(t
),
863 .mount_options
= TAKE_PTR(o
),
864 .mount_node_fd
= TAKE_FD(mount_node_fd
),
873 (void) blkid_probe_lookup_value(b
, "PTTYPE", &pttype
, NULL
);
877 is_gpt
= streq_ptr(pttype
, "gpt");
878 is_mbr
= streq_ptr(pttype
, "dos");
880 if (!is_gpt
&& ((flags
& DISSECT_IMAGE_GPT_ONLY
) || !is_mbr
))
883 /* We support external verity data partitions only if the image has no partition table */
884 if (verity
&& verity
->data_path
)
887 if (FLAGS_SET(flags
, DISSECT_IMAGE_ADD_PARTITION_DEVICES
)) {
888 /* Safety check: refuse block devices that carry a partition table but for which the kernel doesn't
889 * do partition scanning. */
890 r
= blockdev_partscan_enabled(fd
);
894 return -EPROTONOSUPPORT
;
897 (void) blkid_probe_lookup_value(b
, "PTUUID", &sptuuid
, NULL
);
899 r
= sd_id128_from_string(sptuuid
, &m
->image_uuid
);
901 log_debug_errno(r
, "Failed to parse partition table UUID '%s', ignoring: %m", sptuuid
);
905 pl
= blkid_probe_get_partitions(b
);
907 return errno_or_else(ENOMEM
);
910 n_partitions
= blkid_partlist_numof_partitions(pl
);
911 if (n_partitions
< 0)
912 return errno_or_else(EIO
);
914 for (int i
= 0; i
< n_partitions
; i
++) {
915 _cleanup_free_
char *node
= NULL
;
916 unsigned long long pflags
;
917 blkid_loff_t start
, size
;
922 pp
= blkid_partlist_get_partition(pl
, i
);
924 return errno_or_else(EIO
);
926 pflags
= blkid_partition_get_flags(pp
);
929 nr
= blkid_partition_get_partno(pp
);
931 return errno_or_else(EIO
);
934 start
= blkid_partition_get_start(pp
);
936 return errno_or_else(EIO
);
938 assert((uint64_t) start
< UINT64_MAX
/512);
941 size
= blkid_partition_get_size(pp
);
943 return errno_or_else(EIO
);
945 assert((uint64_t) size
< UINT64_MAX
/512);
947 /* While probing we need the non-diskseq device node name to access the thing, hence mask off
948 * DISSECT_IMAGE_DISKSEQ_DEVNODE. */
949 r
= make_partition_devname(devname
, diskseq
, nr
, flags
& ~DISSECT_IMAGE_DISKSEQ_DEVNODE
, &node
);
953 /* So here's the thing: after the main ("whole") block device popped up it might take a while
954 * before the kernel fully probed the partition table. Waiting for that to finish is icky in
955 * userspace. So here's what we do instead. We issue the BLKPG_ADD_PARTITION ioctl to add the
956 * partition ourselves, racing against the kernel. Good thing is: if this call fails with
957 * EBUSY then the kernel was quicker than us, and that's totally OK, the outcome is good for
958 * us: the device node will exist. If OTOH our call was successful we won the race. Which is
959 * also good as the outcome is the same: the partition block device exists, and we can use
962 * Kernel returns EBUSY if there's already a partition by that number or an overlapping
963 * partition already existent. */
965 if (FLAGS_SET(flags
, DISSECT_IMAGE_ADD_PARTITION_DEVICES
)) {
966 r
= block_device_add_partition(fd
, node
, nr
, (uint64_t) start
* 512, (uint64_t) size
* 512);
969 return log_debug_errno(r
, "BLKPG_ADD_PARTITION failed: %m");
971 log_debug_errno(r
, "Kernel was quicker than us in adding partition %i.", nr
);
973 log_debug("We were quicker than kernel in adding partition %i.", nr
);
977 const char *fstype
= NULL
, *label
;
978 sd_id128_t type_id
, id
;
979 GptPartitionType type
;
980 bool rw
= true, growfs
= false;
982 r
= blkid_partition_get_uuid_id128(pp
, &id
);
984 log_debug_errno(r
, "Failed to read partition UUID, ignoring: %m");
988 r
= blkid_partition_get_type_id128(pp
, &type_id
);
990 log_debug_errno(r
, "Failed to read partition type UUID, ignoring: %m");
994 type
= gpt_partition_type_from_uuid(type_id
);
996 label
= blkid_partition_get_name(pp
); /* libblkid returns NULL here if empty */
998 if (IN_SET(type
.designator
,
1004 check_partition_flags(node
, pflags
,
1005 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1007 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1010 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1011 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1013 } else if (type
.designator
== PARTITION_ESP
) {
1015 /* Note that we don't check the SD_GPT_FLAG_NO_AUTO flag for the ESP, as it is
1016 * not defined there. We instead check the SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as
1017 * recommended by the UEFI spec (See "12.3.3 Number and Location of System
1020 if (pflags
& SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL
)
1025 } else if (type
.designator
== PARTITION_ROOT
) {
1027 check_partition_flags(node
, pflags
,
1028 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1030 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1033 /* If a root ID is specified, ignore everything but the root id */
1034 if (!sd_id128_is_null(root_uuid
) && !sd_id128_equal(root_uuid
, id
))
1037 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1038 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1040 } else if (type
.designator
== PARTITION_ROOT_VERITY
) {
1042 check_partition_flags(node
, pflags
,
1043 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1045 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1048 m
->has_verity
= true;
1050 /* If no verity configuration is specified, then don't do verity */
1053 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
1056 /* If root hash is specified, then ignore everything but the root id */
1057 if (!sd_id128_is_null(root_verity_uuid
) && !sd_id128_equal(root_verity_uuid
, id
))
1060 fstype
= "DM_verity_hash";
1063 } else if (type
.designator
== PARTITION_ROOT_VERITY_SIG
) {
1065 check_partition_flags(node
, pflags
,
1066 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1068 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1071 m
->has_verity_sig
= true;
1075 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
1078 fstype
= "verity_hash_signature";
1081 } else if (type
.designator
== PARTITION_USR
) {
1083 check_partition_flags(node
, pflags
,
1084 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1086 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1089 /* If a usr ID is specified, ignore everything but the usr id */
1090 if (!sd_id128_is_null(usr_uuid
) && !sd_id128_equal(usr_uuid
, id
))
1093 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1094 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1096 } else if (type
.designator
== PARTITION_USR_VERITY
) {
1098 check_partition_flags(node
, pflags
,
1099 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1101 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1104 m
->has_verity
= true;
1108 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
1111 /* If usr hash is specified, then ignore everything but the usr id */
1112 if (!sd_id128_is_null(usr_verity_uuid
) && !sd_id128_equal(usr_verity_uuid
, id
))
1115 fstype
= "DM_verity_hash";
1118 } else if (type
.designator
== PARTITION_USR_VERITY_SIG
) {
1120 check_partition_flags(node
, pflags
,
1121 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1123 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1126 m
->has_verity_sig
= true;
1130 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
1133 fstype
= "verity_hash_signature";
1136 } else if (type
.designator
== PARTITION_SWAP
) {
1138 check_partition_flags(node
, pflags
, SD_GPT_FLAG_NO_AUTO
);
1140 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1143 /* Note: we don't set fstype = "swap" here, because we still need to probe if
1144 * it might be encrypted (i.e. fstype "crypt_LUKS") or unencrypted
1145 * (i.e. fstype "swap"), and the only way to figure that out is via fstype
1148 /* We don't have a designator for SD_GPT_LINUX_GENERIC so check the UUID instead. */
1149 } else if (sd_id128_equal(type
.uuid
, SD_GPT_LINUX_GENERIC
)) {
1151 check_partition_flags(node
, pflags
,
1152 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1154 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1158 multiple_generic
= true;
1161 generic_rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1162 generic_growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1164 generic_node
= TAKE_PTR(node
);
1167 } else if (type
.designator
== PARTITION_VAR
) {
1169 check_partition_flags(node
, pflags
,
1170 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1172 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1175 if (!FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_VAR_CHECK
)) {
1176 sd_id128_t var_uuid
;
1178 /* For /var we insist that the uuid of the partition matches the
1179 * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
1180 * ID. Why? Unlike the other partitions /var is inherently
1181 * installation specific, hence we need to be careful not to mount it
1182 * in the wrong installation. By hashing the partition UUID from
1183 * /etc/machine-id we can securely bind the partition to the
1186 r
= sd_id128_get_machine_app_specific(SD_GPT_VAR
, &var_uuid
);
1190 if (!sd_id128_equal(var_uuid
, id
)) {
1191 log_debug("Found a /var/ partition, but its UUID didn't match our expectations "
1192 "(found: " SD_ID128_UUID_FORMAT_STR
", expected: " SD_ID128_UUID_FORMAT_STR
"), ignoring.",
1193 SD_ID128_FORMAT_VAL(id
), SD_ID128_FORMAT_VAL(var_uuid
));
1198 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1199 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1202 if (type
.designator
!= _PARTITION_DESIGNATOR_INVALID
) {
1203 _cleanup_free_
char *t
= NULL
, *o
= NULL
, *l
= NULL
, *n
= NULL
;
1204 _cleanup_close_
int mount_node_fd
= -EBADF
;
1205 const char *options
= NULL
;
1207 r
= image_policy_may_use(policy
, type
.designator
);
1211 /* Policy says: ignore; Remember this fact, so that we later can distinguish between "found but ignored" and "not found at all" */
1213 if (!m
->partitions
[type
.designator
].found
)
1214 m
->partitions
[type
.designator
].ignored
= true;
1219 if (m
->partitions
[type
.designator
].found
) {
1222 /* For most partition types the first one we see wins. Except for the
1223 * rootfs and /usr, where we do a version compare of the label, and
1224 * let the newest version win. This permits a simple A/B versioning
1225 * scheme in OS images. */
1227 c
= compare_arch(type
.arch
, m
->partitions
[type
.designator
].architecture
);
1228 if (c
< 0) /* the arch we already found is better than the one we found now */
1230 if (c
== 0 && /* same arch? then go by version in label */
1231 (!partition_designator_is_versioned(type
.designator
) ||
1232 strverscmp_improved(label
, m
->partitions
[type
.designator
].label
) <= 0))
1235 dissected_partition_done(m
->partitions
+ type
.designator
);
1238 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
) &&
1239 type
.designator
!= PARTITION_SWAP
) {
1240 mount_node_fd
= open_partition(node
, /* is_partition = */ true, m
->loop
);
1241 if (mount_node_fd
< 0)
1242 return mount_node_fd
;
1245 r
= make_partition_devname(devname
, diskseq
, nr
, flags
, &n
);
1261 options
= mount_options_from_designator(mount_options
, type
.designator
);
1263 o
= strdup(options
);
1268 m
->partitions
[type
.designator
] = (DissectedPartition
) {
1273 .architecture
= type
.arch
,
1274 .node
= TAKE_PTR(n
),
1275 .fstype
= TAKE_PTR(t
),
1276 .label
= TAKE_PTR(l
),
1278 .mount_options
= TAKE_PTR(o
),
1279 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1280 .offset
= (uint64_t) start
* 512,
1281 .size
= (uint64_t) size
* 512,
1282 .gpt_flags
= pflags
,
1286 } else if (is_mbr
) {
1288 switch (blkid_partition_get_type(pp
)) {
1290 case 0x83: /* Linux partition */
1292 if (pflags
!= 0x80) /* Bootable flag */
1296 multiple_generic
= true;
1300 generic_growfs
= false;
1301 generic_node
= TAKE_PTR(node
);
1306 case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
1307 _cleanup_close_
int mount_node_fd
= -EBADF
;
1308 _cleanup_free_
char *o
= NULL
, *n
= NULL
;
1309 sd_id128_t id
= SD_ID128_NULL
;
1310 const char *options
= NULL
;
1312 r
= image_policy_may_use(policy
, PARTITION_XBOOTLDR
);
1315 if (r
== 0) { /* policy says: ignore */
1316 if (!m
->partitions
[PARTITION_XBOOTLDR
].found
)
1317 m
->partitions
[PARTITION_XBOOTLDR
].ignored
= true;
1322 /* First one wins */
1323 if (m
->partitions
[PARTITION_XBOOTLDR
].found
)
1326 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
1327 mount_node_fd
= open_partition(node
, /* is_partition = */ true, m
->loop
);
1328 if (mount_node_fd
< 0)
1329 return mount_node_fd
;
1332 (void) blkid_partition_get_uuid_id128(pp
, &id
);
1334 r
= make_partition_devname(devname
, diskseq
, nr
, flags
, &n
);
1338 options
= mount_options_from_designator(mount_options
, PARTITION_XBOOTLDR
);
1340 o
= strdup(options
);
1345 m
->partitions
[PARTITION_XBOOTLDR
] = (DissectedPartition
) {
1350 .architecture
= _ARCHITECTURE_INVALID
,
1351 .node
= TAKE_PTR(n
),
1353 .mount_options
= TAKE_PTR(o
),
1354 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1355 .offset
= (uint64_t) start
* 512,
1356 .size
= (uint64_t) size
* 512,
1364 if (!m
->partitions
[PARTITION_ROOT
].found
&&
1365 (m
->partitions
[PARTITION_ROOT_VERITY
].found
||
1366 m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
))
1367 return -EADDRNOTAVAIL
; /* Verity found but no matching rootfs? Something is off, refuse. */
1369 /* Hmm, we found a signature partition but no Verity data? Something is off. */
1370 if (m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_ROOT_VERITY
].found
)
1371 return -EADDRNOTAVAIL
;
1373 if (!m
->partitions
[PARTITION_USR
].found
&&
1374 (m
->partitions
[PARTITION_USR_VERITY
].found
||
1375 m
->partitions
[PARTITION_USR_VERITY_SIG
].found
))
1376 return -EADDRNOTAVAIL
; /* as above */
1379 if (m
->partitions
[PARTITION_USR_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_USR_VERITY
].found
)
1380 return -EADDRNOTAVAIL
;
1382 /* If root and /usr are combined then insist that the architecture matches */
1383 if (m
->partitions
[PARTITION_ROOT
].found
&&
1384 m
->partitions
[PARTITION_USR
].found
&&
1385 (m
->partitions
[PARTITION_ROOT
].architecture
>= 0 &&
1386 m
->partitions
[PARTITION_USR
].architecture
>= 0 &&
1387 m
->partitions
[PARTITION_ROOT
].architecture
!= m
->partitions
[PARTITION_USR
].architecture
))
1388 return -EADDRNOTAVAIL
;
1390 if (!m
->partitions
[PARTITION_ROOT
].found
&&
1391 !m
->partitions
[PARTITION_USR
].found
&&
1392 (flags
& DISSECT_IMAGE_GENERIC_ROOT
) &&
1393 (!verity
|| !verity
->root_hash
|| verity
->designator
!= PARTITION_USR
)) {
1395 /* OK, we found nothing usable, then check if there's a single generic partition, and use
1396 * that. If the root hash was set however, then we won't fall back to a generic node, because
1397 * the root hash decides. */
1399 /* If we didn't find a properly marked root partition, but we did find a single suitable
1400 * generic Linux partition, then use this as root partition, if the caller asked for it. */
1401 if (multiple_generic
)
1404 /* If we didn't find a generic node, then we can't fix this up either */
1406 r
= image_policy_may_use(policy
, PARTITION_ROOT
);
1410 /* Policy says: ignore; remember that we did */
1411 m
->partitions
[PARTITION_ROOT
].ignored
= true;
1413 _cleanup_close_
int mount_node_fd
= -EBADF
;
1414 _cleanup_free_
char *o
= NULL
, *n
= NULL
;
1415 const char *options
;
1417 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
1418 mount_node_fd
= open_partition(generic_node
, /* is_partition = */ true, m
->loop
);
1419 if (mount_node_fd
< 0)
1420 return mount_node_fd
;
1423 r
= make_partition_devname(devname
, diskseq
, generic_nr
, flags
, &n
);
1427 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
1429 o
= strdup(options
);
1434 assert(generic_nr
>= 0);
1435 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
1438 .growfs
= generic_growfs
,
1439 .partno
= generic_nr
,
1440 .architecture
= _ARCHITECTURE_INVALID
,
1441 .node
= TAKE_PTR(n
),
1442 .uuid
= generic_uuid
,
1443 .mount_options
= TAKE_PTR(o
),
1444 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1445 .offset
= UINT64_MAX
,
1452 /* 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 */
1453 if (FLAGS_SET(flags
, DISSECT_IMAGE_REQUIRE_ROOT
) &&
1454 !(m
->partitions
[PARTITION_ROOT
].found
|| (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
1457 if (m
->partitions
[PARTITION_ROOT_VERITY
].found
) {
1458 /* We only support one verity partition per image, i.e. can't do for both /usr and root fs */
1459 if (m
->partitions
[PARTITION_USR_VERITY
].found
)
1462 /* We don't support verity enabled root with a split out /usr. Neither with nor without
1463 * verity there. (Note that we do support verity-less root with verity-full /usr, though.) */
1464 if (m
->partitions
[PARTITION_USR
].found
)
1465 return -EADDRNOTAVAIL
;
1469 /* If a verity designator is specified, then insist that the matching partition exists */
1470 if (verity
->designator
>= 0 && !m
->partitions
[verity
->designator
].found
)
1471 return -EADDRNOTAVAIL
;
1473 bool have_verity_sig_partition
=
1474 m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR_VERITY_SIG
: PARTITION_ROOT_VERITY_SIG
].found
;
1476 if (verity
->root_hash
) {
1477 /* If we have an explicit root hash and found the partitions for it, then we are ready to use
1478 * Verity, set things up for it */
1480 if (verity
->designator
< 0 || verity
->designator
== PARTITION_ROOT
) {
1481 if (!m
->partitions
[PARTITION_ROOT_VERITY
].found
|| !m
->partitions
[PARTITION_ROOT
].found
)
1482 return -EADDRNOTAVAIL
;
1484 /* If we found a verity setup, then the root partition is necessarily read-only. */
1485 m
->partitions
[PARTITION_ROOT
].rw
= false;
1486 m
->verity_ready
= true;
1489 assert(verity
->designator
== PARTITION_USR
);
1491 if (!m
->partitions
[PARTITION_USR_VERITY
].found
|| !m
->partitions
[PARTITION_USR
].found
)
1492 return -EADDRNOTAVAIL
;
1494 m
->partitions
[PARTITION_USR
].rw
= false;
1495 m
->verity_ready
= true;
1498 if (m
->verity_ready
)
1499 m
->verity_sig_ready
= verity
->root_hash_sig
|| have_verity_sig_partition
;
1501 } else if (have_verity_sig_partition
) {
1503 /* If we found an embedded signature partition, we are ready, too. */
1505 m
->verity_ready
= m
->verity_sig_ready
= true;
1506 m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR
: PARTITION_ROOT
].rw
= false;
1512 /* After we discovered all partitions let's see if the verity requirements match the policy. (Note:
1513 * we don't check encryption requirements here, because we haven't probed the file system yet, hence
1514 * don't know if this is encrypted or not) */
1515 for (PartitionDesignator di
= 0; di
< _PARTITION_DESIGNATOR_MAX
; di
++) {
1516 PartitionDesignator vi
, si
;
1517 PartitionPolicyFlags found_flags
;
1519 any
= any
|| m
->partitions
[di
].found
;
1521 vi
= partition_verity_of(di
);
1522 si
= partition_verity_sig_of(di
);
1524 /* Determine the verity protection level for this partition. */
1525 found_flags
= m
->partitions
[di
].found
?
1526 (vi
>= 0 && m
->partitions
[vi
].found
?
1527 (si
>= 0 && m
->partitions
[si
].found
? PARTITION_POLICY_SIGNED
: PARTITION_POLICY_VERITY
) :
1528 PARTITION_POLICY_ENCRYPTED
|PARTITION_POLICY_UNPROTECTED
) :
1529 (m
->partitions
[di
].ignored
? PARTITION_POLICY_UNUSED
: PARTITION_POLICY_ABSENT
);
1531 r
= image_policy_check_protection(policy
, di
, found_flags
);
1535 if (m
->partitions
[di
].found
) {
1536 r
= image_policy_check_partition_flags(policy
, di
, m
->partitions
[di
].gpt_flags
);
1542 if (!any
&& !FLAGS_SET(flags
, DISSECT_IMAGE_ALLOW_EMPTY
))
1545 r
= dissected_image_probe_filesystems(m
, fd
, policy
);
1553 int dissect_image_file(
1555 const VeritySettings
*verity
,
1556 const MountOptions
*mount_options
,
1557 const ImagePolicy
*image_policy
,
1558 DissectImageFlags flags
,
1559 DissectedImage
**ret
) {
1562 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
1563 _cleanup_close_
int fd
= -EBADF
;
1568 fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
1572 r
= fd_verify_regular(fd
);
1576 r
= dissected_image_new(path
, &m
);
1580 r
= probe_sector_size(fd
, &m
->sector_size
);
1584 r
= dissect_image(m
, fd
, path
, verity
, mount_options
, image_policy
, flags
);
1596 int dissect_log_error(int log_level
, int r
, const char *name
, const VeritySettings
*verity
) {
1597 assert(log_level
>= 0 && log_level
<= LOG_DEBUG
);
1602 case 0 ... INT_MAX
: /* success! */
1606 return log_full_errno(log_level
, r
, "Dissecting images is not supported, compiled without blkid support.");
1609 return log_full_errno(log_level
, r
, "%s: Couldn't identify a suitable partition table or file system.", name
);
1612 return log_full_errno(log_level
, r
, "%s: The image does not pass os-release/extension-release validation.", name
);
1614 case -EADDRNOTAVAIL
:
1615 return log_full_errno(log_level
, r
, "%s: No root partition for specified root hash found.", name
);
1618 return log_full_errno(log_level
, r
, "%s: Multiple suitable root partitions found in image.", name
);
1621 return log_full_errno(log_level
, r
, "%s: No suitable root partition found in image.", name
);
1623 case -EPROTONOSUPPORT
:
1624 return log_full_errno(log_level
, r
, "Device '%s' is a loopback block device with partition scanning turned off, please turn it on.", name
);
1627 return log_full_errno(log_level
, r
, "%s: Image is not a block device.", name
);
1630 return log_full_errno(log_level
, r
,
1631 "Combining partitioned images (such as '%s') with external Verity data (such as '%s') not supported. "
1632 "(Consider setting $SYSTEMD_DISSECT_VERITY_SIDECAR=0 to disable automatic discovery of external Verity data.)",
1633 name
, strna(verity
? verity
->data_path
: NULL
));
1636 return log_full_errno(log_level
, r
, "%s: image does not match image policy.", name
);
1639 return log_full_errno(log_level
, r
, "%s: no suitable partitions found.", name
);
1642 return log_full_errno(log_level
, r
, "%s: cannot dissect image: %m", name
);
1646 int dissect_image_file_and_warn(
1648 const VeritySettings
*verity
,
1649 const MountOptions
*mount_options
,
1650 const ImagePolicy
*image_policy
,
1651 DissectImageFlags flags
,
1652 DissectedImage
**ret
) {
1654 return dissect_log_error(
1656 dissect_image_file(path
, verity
, mount_options
, image_policy
, flags
, ret
),
1661 DissectedImage
* dissected_image_unref(DissectedImage
*m
) {
1665 /* First, clear dissected partitions. */
1666 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++)
1667 dissected_partition_done(m
->partitions
+ i
);
1669 /* Second, free decrypted images. This must be after dissected_partition_done(), as freeing
1670 * DecryptedImage may try to deactivate partitions. */
1671 decrypted_image_unref(m
->decrypted_image
);
1673 /* Third, unref LoopDevice. This must be called after the above two, as freeing LoopDevice may try to
1674 * remove existing partitions on the loopback block device. */
1675 loop_device_unref(m
->loop
);
1677 free(m
->image_name
);
1679 strv_free(m
->machine_info
);
1680 strv_free(m
->os_release
);
1681 strv_free(m
->initrd_release
);
1682 strv_free(m
->extension_release
);
1687 static int is_loop_device(const char *path
) {
1688 char s
[SYS_BLOCK_PATH_MAX("/../loop/")];
1693 if (stat(path
, &st
) < 0)
1696 if (!S_ISBLK(st
.st_mode
))
1699 xsprintf_sys_block_path(s
, "/loop/", st
.st_dev
);
1700 if (access(s
, F_OK
) < 0) {
1701 if (errno
!= ENOENT
)
1704 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
1705 xsprintf_sys_block_path(s
, "/../loop/", st
.st_dev
);
1706 if (access(s
, F_OK
) < 0)
1707 return errno
== ENOENT
? false : -errno
;
1713 static int run_fsck(int node_fd
, const char *fstype
) {
1717 assert(node_fd
>= 0);
1720 r
= fsck_exists_for_fstype(fstype
);
1722 log_debug_errno(r
, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype
);
1726 log_debug("Not checking partition %s, as fsck for %s does not exist.", FORMAT_PROC_FD_PATH(node_fd
), fstype
);
1733 &node_fd
, 1, /* Leave the node fd open */
1734 FORK_RESET_SIGNALS
|FORK_CLOSE_ALL_FDS
|FORK_RLIMIT_NOFILE_SAFE
|FORK_DEATHSIG
|FORK_REARRANGE_STDIO
|FORK_CLOEXEC_OFF
,
1737 return log_debug_errno(r
, "Failed to fork off fsck: %m");
1740 execlp("fsck", "fsck", "-aT", FORMAT_PROC_FD_PATH(node_fd
), NULL
);
1742 log_debug_errno(errno
, "Failed to execl() fsck: %m");
1743 _exit(FSCK_OPERATIONAL_ERROR
);
1746 exit_status
= wait_for_terminate_and_check("fsck", pid
, 0);
1747 if (exit_status
< 0)
1748 return log_debug_errno(exit_status
, "Failed to fork off fsck: %m");
1750 if ((exit_status
& ~FSCK_ERROR_CORRECTED
) != FSCK_SUCCESS
) {
1751 log_debug("fsck failed with exit status %i.", exit_status
);
1753 if ((exit_status
& (FSCK_SYSTEM_SHOULD_REBOOT
|FSCK_ERRORS_LEFT_UNCORRECTED
)) != 0)
1754 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
), "File system is corrupted, refusing.");
1756 log_debug("Ignoring fsck error.");
1762 static int fs_grow(const char *node_path
, const char *mount_path
) {
1763 _cleanup_close_
int mount_fd
= -EBADF
, node_fd
= -EBADF
;
1764 uint64_t size
, newsize
;
1767 node_fd
= open(node_path
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
1769 return log_debug_errno(errno
, "Failed to open node device %s: %m", node_path
);
1771 if (ioctl(node_fd
, BLKGETSIZE64
, &size
) != 0)
1772 return log_debug_errno(errno
, "Failed to get block device size of %s: %m", node_path
);
1774 mount_fd
= open(mount_path
, O_RDONLY
|O_DIRECTORY
|O_CLOEXEC
);
1776 return log_debug_errno(errno
, "Failed to open mountd file system %s: %m", mount_path
);
1778 log_debug("Resizing \"%s\" to %"PRIu64
" bytes...", mount_path
, size
);
1779 r
= resize_fs(mount_fd
, size
, &newsize
);
1781 return log_debug_errno(r
, "Failed to resize \"%s\" to %"PRIu64
" bytes: %m", mount_path
, size
);
1783 if (newsize
== size
)
1784 log_debug("Successfully resized \"%s\" to %s bytes.",
1785 mount_path
, FORMAT_BYTES(newsize
));
1787 assert(newsize
< size
);
1788 log_debug("Successfully resized \"%s\" to %s bytes (%"PRIu64
" bytes lost due to blocksize).",
1789 mount_path
, FORMAT_BYTES(newsize
), size
- newsize
);
1795 int partition_pick_mount_options(
1796 PartitionDesignator d
,
1801 unsigned long *ret_ms_flags
) {
1803 _cleanup_free_
char *options
= NULL
;
1805 assert(ret_options
);
1807 /* Selects a baseline of bind mount flags, that should always apply.
1809 * Firstly, we set MS_NODEV universally on all mounts, since we don't want to allow device nodes outside of /dev/.
1811 * On /var/tmp/ we'll also set MS_NOSUID, same as we set for /tmp/ on the host.
1813 * On the ESP and XBOOTLDR partitions we'll also disable symlinks, and execution. These file systems
1814 * are generally untrusted (i.e. not encrypted or authenticated), and typically VFAT hence we should
1815 * be as restrictive as possible, and this shouldn't hurt, since the functionality is not available
1818 unsigned long flags
= MS_NODEV
;
1826 case PARTITION_XBOOTLDR
:
1827 flags
|= MS_NOSUID
|MS_NOEXEC
|ms_nosymfollow_supported();
1829 /* The ESP might contain a pre-boot random seed. Let's make this unaccessible to regular
1830 * userspace. ESP/XBOOTLDR is almost certainly VFAT, hence if we don't know assume it is. */
1831 if (!fstype
|| fstype_can_umask(fstype
))
1832 if (!strextend_with_separator(&options
, ",", "umask=0077"))
1844 /* So, when you request MS_RDONLY from ext4, then this means nothing. It happily still writes to the
1845 * backing storage. What's worse, the BLKRO[GS]ET flag and (in case of loopback devices)
1846 * LO_FLAGS_READ_ONLY don't mean anything, they affect userspace accesses only, and write accesses
1847 * from the upper file system still get propagated through to the underlying file system,
1848 * unrestricted. To actually get ext4/xfs/btrfs to stop writing to the device we need to specify
1849 * "norecovery" as mount option, in addition to MS_RDONLY. Yes, this sucks, since it means we need to
1850 * carry a per file system table here.
1852 * Note that this means that we might not be able to mount corrupted file systems as read-only
1853 * anymore (since in some cases the kernel implementations will refuse mounting when corrupted,
1854 * read-only and "norecovery" is specified). But I think for the case of automatically determined
1855 * mount options for loopback devices this is the right choice, since otherwise using the same
1856 * loopback file twice even in read-only mode, is going to fail badly sooner or later. The usecase of
1857 * making reuse of the immutable images "just work" is more relevant to us than having read-only
1858 * access that actually modifies stuff work on such image files. Or to say this differently: if
1859 * people want their file systems to be fixed up they should just open them in writable mode, where
1860 * all these problems don't exist. */
1861 if (!rw
&& fstype
&& fstype_can_norecovery(fstype
))
1862 if (!strextend_with_separator(&options
, ",", "norecovery"))
1865 if (discard
&& fstype
&& fstype_can_discard(fstype
))
1866 if (!strextend_with_separator(&options
, ",", "discard"))
1869 if (!ret_ms_flags
) /* Fold flags into option string if ret_flags specified as NULL */
1870 if (!strextend_with_separator(&options
, ",",
1871 FLAGS_SET(flags
, MS_RDONLY
) ? "ro" : "rw",
1872 FLAGS_SET(flags
, MS_NODEV
) ? "nodev" : "dev",
1873 FLAGS_SET(flags
, MS_NOSUID
) ? "nosuid" : "suid",
1874 FLAGS_SET(flags
, MS_NOEXEC
) ? "noexec" : "exec",
1875 FLAGS_SET(flags
, MS_NOSYMFOLLOW
) ? "nosymfollow" : NULL
))
1876 /* NB: we suppress 'symfollow' here, since it's the default, and old /bin/mount might not know it */
1880 *ret_ms_flags
= flags
;
1882 *ret_options
= TAKE_PTR(options
);
1886 static int mount_partition(
1887 PartitionDesignator d
,
1888 DissectedPartition
*m
,
1890 const char *directory
,
1893 DissectImageFlags flags
) {
1895 _cleanup_free_
char *chased
= NULL
, *options
= NULL
;
1896 bool rw
, discard
, remap_uid_gid
= false;
1897 const char *p
, *node
, *fstype
;
1898 unsigned long ms_flags
;
1904 if (m
->mount_node_fd
< 0)
1907 /* Use decrypted node and matching fstype if available, otherwise use the original device */
1908 node
= FORMAT_PROC_FD_PATH(m
->mount_node_fd
);
1909 fstype
= dissected_partition_fstype(m
);
1912 return -EAFNOSUPPORT
;
1914 /* We are looking at an encrypted partition? This either means stacked encryption, or the caller
1915 * didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error for this
1917 if (streq(fstype
, "crypto_LUKS"))
1920 r
= dissect_fstype_ok(fstype
);
1924 return -EIDRM
; /* Recognizable error */
1926 rw
= m
->rw
&& !(flags
& DISSECT_IMAGE_MOUNT_READ_ONLY
);
1928 discard
= ((flags
& DISSECT_IMAGE_DISCARD
) ||
1929 ((flags
& DISSECT_IMAGE_DISCARD_ON_LOOP
) && is_loop_device(m
->node
) > 0));
1931 if (FLAGS_SET(flags
, DISSECT_IMAGE_FSCK
) && rw
) {
1932 r
= run_fsck(m
->mount_node_fd
, fstype
);
1938 /* Automatically create missing mount points inside the image, if necessary. */
1939 r
= mkdir_p_root(where
, directory
, uid_shift
, (gid_t
) uid_shift
, 0755, NULL
);
1940 if (r
< 0 && r
!= -EROFS
)
1943 r
= chase(directory
, where
, CHASE_PREFIX_ROOT
, &chased
, NULL
);
1949 /* Create top-level mount if missing – but only if this is asked for. This won't modify the
1950 * image (as the branch above does) but the host hierarchy, and the created directory might
1951 * survive our mount in the host hierarchy hence. */
1952 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
1953 r
= mkdir_p(where
, 0755);
1961 r
= partition_pick_mount_options(d
, dissected_partition_fstype(m
), rw
, discard
, &options
, &ms_flags
);
1965 if (uid_is_valid(uid_shift
) && uid_shift
!= 0) {
1967 if (fstype_can_uid_gid(fstype
)) {
1968 _cleanup_free_
char *uid_option
= NULL
;
1970 if (asprintf(&uid_option
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
1973 if (!strextend_with_separator(&options
, ",", uid_option
))
1975 } else if (FLAGS_SET(flags
, DISSECT_IMAGE_MOUNT_IDMAPPED
))
1976 remap_uid_gid
= true;
1979 if (!isempty(m
->mount_options
))
1980 if (!strextend_with_separator(&options
, ",", m
->mount_options
))
1983 r
= mount_nofollow_verbose(LOG_DEBUG
, node
, p
, fstype
, ms_flags
, options
);
1987 if (rw
&& m
->growfs
&& FLAGS_SET(flags
, DISSECT_IMAGE_GROWFS
))
1988 (void) fs_grow(node
, p
);
1990 if (remap_uid_gid
) {
1991 r
= remount_idmap(p
, uid_shift
, uid_range
, UID_INVALID
, REMOUNT_IDMAPPING_HOST_ROOT
);
1999 static int mount_root_tmpfs(const char *where
, uid_t uid_shift
, DissectImageFlags flags
) {
2000 _cleanup_free_
char *options
= NULL
;
2005 /* For images that contain /usr/ but no rootfs, let's mount rootfs as tmpfs */
2007 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
2008 r
= mkdir_p(where
, 0755);
2013 if (uid_is_valid(uid_shift
)) {
2014 if (asprintf(&options
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
2018 r
= mount_nofollow_verbose(LOG_DEBUG
, "rootfs", where
, "tmpfs", MS_NODEV
, options
);
2025 static int mount_point_is_available(const char *where
, const char *path
, bool missing_ok
) {
2026 _cleanup_free_
char *p
= NULL
;
2029 /* Check whether <path> is suitable as a mountpoint, i.e. is an empty directory
2030 * or does not exist at all (when missing_ok). */
2032 r
= chase(path
, where
, CHASE_PREFIX_ROOT
, &p
, NULL
);
2036 return log_debug_errno(r
, "Failed to chase \"%s\": %m", path
);
2038 r
= dir_is_empty(p
, /* ignore_hidden_or_backup= */ false);
2042 return log_debug_errno(r
, "Failed to check directory \"%s\": %m", p
);
2046 int dissected_image_mount(
2051 DissectImageFlags flags
) {
2060 * -ENXIO → No root partition found
2061 * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release/extension-release file found
2062 * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet
2063 * -EUCLEAN → fsck for file system failed
2064 * -EBUSY → File system already mounted/used elsewhere (kernel)
2065 * -EAFNOSUPPORT → File system type not supported or not known
2066 * -EIDRM → File system is not among allowlisted "common" file systems
2069 if (!(m
->partitions
[PARTITION_ROOT
].found
||
2070 (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
2071 return -ENXIO
; /* Require a root fs or at least a /usr/ fs (the latter is subject to a flag of its own) */
2073 if ((flags
& DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
) == 0) {
2075 /* First mount the root fs. If there's none we use a tmpfs. */
2076 if (m
->partitions
[PARTITION_ROOT
].found
)
2077 r
= mount_partition(PARTITION_ROOT
, m
->partitions
+ PARTITION_ROOT
, where
, NULL
, uid_shift
, uid_range
, flags
);
2079 r
= mount_root_tmpfs(where
, uid_shift
, flags
);
2083 /* For us mounting root always means mounting /usr as well */
2084 r
= mount_partition(PARTITION_USR
, m
->partitions
+ PARTITION_USR
, where
, "/usr", uid_shift
, uid_range
, flags
);
2088 if ((flags
& (DISSECT_IMAGE_VALIDATE_OS
|DISSECT_IMAGE_VALIDATE_OS_EXT
)) != 0) {
2089 /* If either one of the validation flags are set, ensure that the image qualifies
2090 * as one or the other (or both). */
2093 if (FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS
)) {
2094 r
= path_is_os_tree(where
);
2100 if (!ok
&& FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS_EXT
)) {
2101 r
= extension_has_forbidden_content(where
);
2105 r
= path_is_extension_tree(IMAGE_SYSEXT
, where
, m
->image_name
, FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_EXTENSION_CHECK
));
2107 r
= path_is_extension_tree(IMAGE_CONFEXT
, where
, m
->image_name
, FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_EXTENSION_CHECK
));
2120 if (flags
& DISSECT_IMAGE_MOUNT_ROOT_ONLY
)
2123 r
= mount_partition(PARTITION_HOME
, m
->partitions
+ PARTITION_HOME
, where
, "/home", uid_shift
, uid_range
, flags
);
2127 r
= mount_partition(PARTITION_SRV
, m
->partitions
+ PARTITION_SRV
, where
, "/srv", uid_shift
, uid_range
, flags
);
2131 r
= mount_partition(PARTITION_VAR
, m
->partitions
+ PARTITION_VAR
, where
, "/var", uid_shift
, uid_range
, flags
);
2135 r
= mount_partition(PARTITION_TMP
, m
->partitions
+ PARTITION_TMP
, where
, "/var/tmp", uid_shift
, uid_range
, flags
);
2139 int slash_boot_is_available
;
2140 r
= slash_boot_is_available
= mount_point_is_available(where
, "/boot", /* missing_ok = */ true);
2144 r
= mount_partition(PARTITION_XBOOTLDR
, m
->partitions
+ PARTITION_XBOOTLDR
, where
, "/boot", uid_shift
, uid_range
, flags
);
2147 slash_boot_is_available
= !r
;
2150 if (m
->partitions
[PARTITION_ESP
].found
) {
2151 const char *esp_path
= NULL
;
2153 /* Mount the ESP to /boot/ if it exists and is empty and we didn't already mount the XBOOTLDR
2154 * partition into it. Otherwise, use /efi instead, but only if it exists and is empty. */
2156 if (slash_boot_is_available
) {
2157 r
= mount_point_is_available(where
, "/boot", /* missing_ok = */ false);
2165 r
= mount_point_is_available(where
, "/efi", /* missing_ok = */ true);
2173 /* OK, let's mount the ESP now (possibly creating the dir if missing) */
2174 r
= mount_partition(PARTITION_ESP
, m
->partitions
+ PARTITION_ESP
, where
, esp_path
, uid_shift
, uid_range
, flags
);
2183 int dissected_image_mount_and_warn(
2188 DissectImageFlags flags
) {
2195 r
= dissected_image_mount(m
, where
, uid_shift
, uid_range
, flags
);
2197 return log_error_errno(r
, "Not root file system found in image.");
2198 if (r
== -EMEDIUMTYPE
)
2199 return log_error_errno(r
, "No suitable os-release/extension-release file in image found.");
2201 return log_error_errno(r
, "Encrypted file system discovered, but decryption not requested.");
2203 return log_error_errno(r
, "File system check on image failed.");
2205 return log_error_errno(r
, "File system already mounted elsewhere.");
2206 if (r
== -EAFNOSUPPORT
)
2207 return log_error_errno(r
, "File system type not supported or not known.");
2209 return log_error_errno(r
, "File system is too uncommon, refused.");
2211 return log_error_errno(r
, "Failed to mount image: %m");
2216 #if HAVE_LIBCRYPTSETUP
2217 struct DecryptedPartition
{
2218 struct crypt_device
*device
;
2224 typedef struct DecryptedPartition DecryptedPartition
;
2226 struct DecryptedImage
{
2228 DecryptedPartition
*decrypted
;
2232 static DecryptedImage
* decrypted_image_free(DecryptedImage
*d
) {
2233 #if HAVE_LIBCRYPTSETUP
2239 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
2240 DecryptedPartition
*p
= d
->decrypted
+ i
;
2242 if (p
->device
&& p
->name
&& !p
->relinquished
) {
2243 _cleanup_free_
char *node
= NULL
;
2245 node
= path_join("/dev/mapper", p
->name
);
2247 r
= btrfs_forget_device(node
);
2248 if (r
< 0 && r
!= -ENOENT
)
2249 log_debug_errno(r
, "Failed to forget btrfs device %s, ignoring: %m", node
);
2253 /* Let's deactivate lazily, as the dm volume may be already/still used by other processes. */
2254 r
= sym_crypt_deactivate_by_name(p
->device
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
2256 log_debug_errno(r
, "Failed to deactivate encrypted partition %s", p
->name
);
2260 sym_crypt_free(p
->device
);
2270 DEFINE_TRIVIAL_REF_UNREF_FUNC(DecryptedImage
, decrypted_image
, decrypted_image_free
);
2272 #if HAVE_LIBCRYPTSETUP
2273 static int decrypted_image_new(DecryptedImage
**ret
) {
2274 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
2278 d
= new(DecryptedImage
, 1);
2282 *d
= (DecryptedImage
) {
2290 static int make_dm_name_and_node(const void *original_node
, const char *suffix
, char **ret_name
, char **ret_node
) {
2291 _cleanup_free_
char *name
= NULL
, *node
= NULL
;
2294 assert(original_node
);
2299 base
= strrchr(original_node
, '/');
2301 base
= original_node
;
2307 name
= strjoin(base
, suffix
);
2310 if (!filename_is_valid(name
))
2313 node
= path_join(sym_crypt_get_dir(), name
);
2317 *ret_name
= TAKE_PTR(name
);
2318 *ret_node
= TAKE_PTR(node
);
2323 static int decrypt_partition(
2324 DissectedPartition
*m
,
2325 const char *passphrase
,
2326 DissectImageFlags flags
,
2327 DecryptedImage
*d
) {
2329 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
2330 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2331 _cleanup_close_
int fd
= -EBADF
;
2337 if (!m
->found
|| !m
->node
|| !m
->fstype
)
2340 if (!streq(m
->fstype
, "crypto_LUKS"))
2346 r
= dlopen_cryptsetup();
2350 r
= make_dm_name_and_node(m
->node
, "-decrypted", &name
, &node
);
2354 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
2357 r
= sym_crypt_init(&cd
, m
->node
);
2359 return log_debug_errno(r
, "Failed to initialize dm-crypt: %m");
2361 cryptsetup_enable_logging(cd
);
2363 r
= sym_crypt_load(cd
, CRYPT_LUKS
, NULL
);
2365 return log_debug_errno(r
, "Failed to load LUKS metadata: %m");
2367 r
= sym_crypt_activate_by_passphrase(cd
, name
, CRYPT_ANY_SLOT
, passphrase
, strlen(passphrase
),
2368 ((flags
& DISSECT_IMAGE_DEVICE_READ_ONLY
) ? CRYPT_ACTIVATE_READONLY
: 0) |
2369 ((flags
& DISSECT_IMAGE_DISCARD_ON_CRYPTO
) ? CRYPT_ACTIVATE_ALLOW_DISCARDS
: 0));
2371 log_debug_errno(r
, "Failed to activate LUKS device: %m");
2372 return r
== -EPERM
? -EKEYREJECTED
: r
;
2375 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2377 return log_debug_errno(errno
, "Failed to open %s: %m", node
);
2379 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
2380 .name
= TAKE_PTR(name
),
2381 .device
= TAKE_PTR(cd
),
2384 m
->decrypted_node
= TAKE_PTR(node
);
2385 close_and_replace(m
->mount_node_fd
, fd
);
2390 static int verity_can_reuse(
2391 const VeritySettings
*verity
,
2393 struct crypt_device
**ret_cd
) {
2395 /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
2396 _cleanup_free_
char *root_hash_existing
= NULL
;
2397 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2398 struct crypt_params_verity crypt_params
= {};
2399 size_t root_hash_existing_size
;
2406 r
= sym_crypt_init_by_name(&cd
, name
);
2408 return log_debug_errno(r
, "Error opening verity device, crypt_init_by_name failed: %m");
2410 cryptsetup_enable_logging(cd
);
2412 r
= sym_crypt_get_verity_info(cd
, &crypt_params
);
2414 return log_debug_errno(r
, "Error opening verity device, crypt_get_verity_info failed: %m");
2416 root_hash_existing_size
= verity
->root_hash_size
;
2417 root_hash_existing
= malloc0(root_hash_existing_size
);
2418 if (!root_hash_existing
)
2421 r
= sym_crypt_volume_key_get(cd
, CRYPT_ANY_SLOT
, root_hash_existing
, &root_hash_existing_size
, NULL
, 0);
2423 return log_debug_errno(r
, "Error opening verity device, crypt_volume_key_get failed: %m");
2424 if (verity
->root_hash_size
!= root_hash_existing_size
||
2425 memcmp(root_hash_existing
, verity
->root_hash
, verity
->root_hash_size
) != 0)
2426 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but root hashes are different.");
2428 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
2429 /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the
2430 * same settings, so that a previous unsigned mount will not be reused if the user asks to use
2431 * signing for the new one, and vice versa. */
2432 if (!!verity
->root_hash_sig
!= !!(crypt_params
.flags
& CRYPT_VERITY_ROOT_HASH_SIGNATURE
))
2433 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but signature settings are not the same.");
2436 *ret_cd
= TAKE_PTR(cd
);
2440 static char* dm_deferred_remove_clean(char *name
) {
2444 (void) sym_crypt_deactivate_by_name(NULL
, name
, CRYPT_DEACTIVATE_DEFERRED
);
2447 DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean
);
2449 static int validate_signature_userspace(const VeritySettings
*verity
) {
2451 _cleanup_(sk_X509_free_allp
) STACK_OF(X509
) *sk
= NULL
;
2452 _cleanup_strv_free_
char **certs
= NULL
;
2453 _cleanup_(PKCS7_freep
) PKCS7
*p7
= NULL
;
2454 _cleanup_free_
char *s
= NULL
;
2455 _cleanup_(BIO_freep
) BIO
*bio
= NULL
; /* 'bio' must be freed first, 's' second, hence keep this order
2456 * of declaration in place, please */
2457 const unsigned char *d
;
2461 assert(verity
->root_hash
);
2462 assert(verity
->root_hash_sig
);
2464 /* Because installing a signature certificate into the kernel chain is so messy, let's optionally do
2465 * userspace validation. */
2467 r
= conf_files_list_nulstr(&certs
, ".crt", NULL
, CONF_FILES_REGULAR
|CONF_FILES_FILTER_MASKED
, CONF_PATHS_NULSTR("verity.d"));
2469 return log_debug_errno(r
, "Failed to enumerate certificates: %m");
2470 if (strv_isempty(certs
)) {
2471 log_debug("No userspace dm-verity certificates found.");
2475 d
= verity
->root_hash_sig
;
2476 p7
= d2i_PKCS7(NULL
, &d
, (long) verity
->root_hash_sig_size
);
2478 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to parse PKCS7 DER signature data.");
2480 s
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2482 return log_oom_debug();
2484 bio
= BIO_new_mem_buf(s
, strlen(s
));
2486 return log_oom_debug();
2488 sk
= sk_X509_new_null();
2490 return log_oom_debug();
2492 STRV_FOREACH(i
, certs
) {
2493 _cleanup_(X509_freep
) X509
*c
= NULL
;
2494 _cleanup_fclose_
FILE *f
= NULL
;
2496 f
= fopen(*i
, "re");
2498 log_debug_errno(errno
, "Failed to open '%s', ignoring: %m", *i
);
2502 c
= PEM_read_X509(f
, NULL
, NULL
, NULL
);
2504 log_debug("Failed to load X509 certificate '%s', ignoring.", *i
);
2508 if (sk_X509_push(sk
, c
) == 0)
2509 return log_oom_debug();
2514 r
= PKCS7_verify(p7
, sk
, NULL
, bio
, NULL
, PKCS7_NOINTERN
|PKCS7_NOVERIFY
);
2516 log_debug("Userspace PKCS#7 validation succeeded.");
2518 log_debug("Userspace PKCS#7 validation failed: %s", ERR_error_string(ERR_get_error(), NULL
));
2522 log_debug("Not doing client-side validation of dm-verity root hash signatures, OpenSSL support disabled.");
2527 static int do_crypt_activate_verity(
2528 struct crypt_device
*cd
,
2530 const VeritySettings
*verity
) {
2532 bool check_signature
;
2539 if (verity
->root_hash_sig
) {
2540 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIGNATURE");
2541 if (r
< 0 && r
!= -ENXIO
)
2542 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIGNATURE");
2544 check_signature
= r
!= 0;
2546 check_signature
= false;
2548 if (check_signature
) {
2550 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
2551 /* First, if we have support for signed keys in the kernel, then try that first. */
2552 r
= sym_crypt_activate_by_signed_key(
2556 verity
->root_hash_size
,
2557 verity
->root_hash_sig
,
2558 verity
->root_hash_sig_size
,
2559 CRYPT_ACTIVATE_READONLY
);
2563 log_debug("Validation of dm-verity signature failed via the kernel, trying userspace validation instead.");
2565 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.",
2566 program_invocation_short_name
);
2569 /* So this didn't work via the kernel, then let's try userspace validation instead. If that
2570 * works we'll try to activate without telling the kernel the signature. */
2572 r
= validate_signature_userspace(verity
);
2576 return log_debug_errno(SYNTHETIC_ERRNO(ENOKEY
),
2577 "Activation of signed Verity volume worked neither via the kernel nor in userspace, can't activate.");
2580 return sym_crypt_activate_by_volume_key(
2584 verity
->root_hash_size
,
2585 CRYPT_ACTIVATE_READONLY
);
2588 static usec_t
verity_timeout(void) {
2589 usec_t t
= 100 * USEC_PER_MSEC
;
2593 /* On slower machines, like non-KVM vm, setting up device may take a long time.
2594 * Let's make the timeout configurable. */
2596 e
= getenv("SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC");
2600 r
= parse_sec(e
, &t
);
2603 "Failed to parse timeout specified in $SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC, "
2604 "using the default timeout (%s).",
2605 FORMAT_TIMESPAN(t
, USEC_PER_MSEC
));
2610 static int verity_partition(
2611 PartitionDesignator designator
,
2612 DissectedPartition
*m
,
2613 DissectedPartition
*v
,
2614 const VeritySettings
*verity
,
2615 DissectImageFlags flags
,
2616 DecryptedImage
*d
) {
2618 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2619 _cleanup_(dm_deferred_remove_cleanp
) char *restore_deferred_remove
= NULL
;
2620 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
2621 _cleanup_close_
int mount_node_fd
= -EBADF
;
2625 assert(v
|| (verity
&& verity
->data_path
));
2627 if (!verity
|| !verity
->root_hash
)
2629 if (!((verity
->designator
< 0 && designator
== PARTITION_ROOT
) ||
2630 (verity
->designator
== designator
)))
2633 if (!m
->found
|| !m
->node
|| !m
->fstype
)
2635 if (!verity
->data_path
) {
2636 if (!v
->found
|| !v
->node
|| !v
->fstype
)
2639 if (!streq(v
->fstype
, "DM_verity_hash"))
2643 r
= dlopen_cryptsetup();
2647 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
2648 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
2649 _cleanup_free_
char *root_hash_encoded
= NULL
;
2651 root_hash_encoded
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2652 if (!root_hash_encoded
)
2655 r
= make_dm_name_and_node(root_hash_encoded
, "-verity", &name
, &node
);
2657 r
= make_dm_name_and_node(m
->node
, "-verity", &name
, &node
);
2661 r
= sym_crypt_init(&cd
, verity
->data_path
?: v
->node
);
2665 cryptsetup_enable_logging(cd
);
2667 r
= sym_crypt_load(cd
, CRYPT_VERITY
, NULL
);
2671 r
= sym_crypt_set_data_device(cd
, m
->node
);
2675 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
2678 /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
2679 * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
2680 * retry a few times before giving up. */
2681 for (unsigned i
= 0; i
< N_DEVICE_NODE_LIST_ATTEMPTS
; i
++) {
2682 _cleanup_(sym_crypt_freep
) struct crypt_device
*existing_cd
= NULL
;
2683 _cleanup_close_
int fd
= -EBADF
;
2685 /* First, check if the device already exists. */
2686 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2687 if (fd
< 0 && !ERRNO_IS_DEVICE_ABSENT(errno
))
2688 return log_debug_errno(errno
, "Failed to open verity device %s: %m", node
);
2690 goto check
; /* The device already exists. Let's check it. */
2692 /* The symlink to the device node does not exist yet. Assume not activated, and let's activate it. */
2693 r
= do_crypt_activate_verity(cd
, name
, verity
);
2695 goto try_open
; /* The device is activated. Let's open it. */
2696 /* libdevmapper can return EINVAL when the device is already in the activation stage.
2697 * There's no way to distinguish this situation from a genuine error due to invalid
2698 * parameters, so immediately fall back to activating the device with a unique name.
2699 * Improvements in libcrypsetup can ensure this never happens:
2700 * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
2701 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2703 if (r
== -ENODEV
) /* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */
2706 -EEXIST
, /* Volume has already been opened and ready to be used. */
2707 -EBUSY
/* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */))
2708 return log_debug_errno(r
, "Failed to activate verity device %s: %m", node
);
2711 if (!restore_deferred_remove
){
2712 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
2713 r
= dm_deferred_remove_cancel(name
);
2714 /* -EBUSY and -ENXIO: the device has already been removed or being removed. We cannot
2715 * use the device, try to open again. See target_message() in drivers/md/dm-ioctl.c
2716 * and dm_cancel_deferred_remove() in drivers/md/dm.c */
2717 if (IN_SET(r
, -EBUSY
, -ENXIO
))
2720 return log_debug_errno(r
, "Failed to disable automated deferred removal for verity device %s: %m", node
);
2722 restore_deferred_remove
= strdup(name
);
2723 if (!restore_deferred_remove
)
2724 return log_oom_debug();
2727 r
= verity_can_reuse(verity
, name
, &existing_cd
);
2728 /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
2729 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2732 -ENOENT
, /* Removed?? */
2733 -EBUSY
, /* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */
2734 -ENODEV
/* Volume is being opened but not ready, crypt_init_by_name() would fail, try to open again. */ ))
2737 return log_debug_errno(r
, "Failed to check if existing verity device %s can be reused: %m", node
);
2740 /* devmapper might say that the device exists, but the devlink might not yet have been
2741 * created. Check and wait for the udev event in that case. */
2742 r
= device_wait_for_devlink(node
, "block", verity_timeout(), NULL
);
2743 /* Fallback to activation with a unique device if it's taking too long */
2744 if (r
== -ETIMEDOUT
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2747 return log_debug_errno(r
, "Failed to wait device node symlink %s: %m", node
);
2752 /* Now, the device is activated and devlink is created. Let's open it. */
2753 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2755 if (!ERRNO_IS_DEVICE_ABSENT(errno
))
2756 return log_debug_errno(errno
, "Failed to open verity device %s: %m", node
);
2758 /* The device has already been removed?? */
2763 mount_node_fd
= TAKE_FD(fd
);
2765 crypt_free_and_replace(cd
, existing_cd
);
2770 /* Device is being removed by another process. Let's wait for a while. */
2771 (void) usleep_safe(2 * USEC_PER_MSEC
);
2774 /* All trials failed or a conflicting verity device exists. Let's try to activate with a unique name. */
2775 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
2776 /* Before trying to activate with unique name, we need to free crypt_device object.
2777 * Otherwise, we get error from libcryptsetup like the following:
2779 * systemd[1234]: Cannot use device /dev/loop5 which is in use (already mapped or mounted).
2784 return verity_partition(designator
, m
, v
, verity
, flags
& ~DISSECT_IMAGE_VERITY_SHARE
, d
);
2787 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
), "All attempts to activate verity device %s failed.", name
);
2790 /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
2791 restore_deferred_remove
= mfree(restore_deferred_remove
);
2793 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
2794 .name
= TAKE_PTR(name
),
2795 .device
= TAKE_PTR(cd
),
2798 m
->decrypted_node
= TAKE_PTR(node
);
2799 close_and_replace(m
->mount_node_fd
, mount_node_fd
);
2805 int dissected_image_decrypt(
2807 const char *passphrase
,
2808 const VeritySettings
*verity
,
2809 DissectImageFlags flags
) {
2811 #if HAVE_LIBCRYPTSETUP
2812 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
2817 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
2821 * = 0 → There was nothing to decrypt
2822 * > 0 → Decrypted successfully
2823 * -ENOKEY → There's something to decrypt but no key was supplied
2824 * -EKEYREJECTED → Passed key was not correct
2827 if (verity
&& verity
->root_hash
&& verity
->root_hash_size
< sizeof(sd_id128_t
))
2830 if (!m
->encrypted
&& !m
->verity_ready
)
2833 #if HAVE_LIBCRYPTSETUP
2834 r
= decrypted_image_new(&d
);
2838 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
2839 DissectedPartition
*p
= m
->partitions
+ i
;
2840 PartitionDesignator k
;
2845 r
= decrypt_partition(p
, passphrase
, flags
, d
);
2849 k
= partition_verity_of(i
);
2851 r
= verity_partition(i
, p
, m
->partitions
+ k
, verity
, flags
| DISSECT_IMAGE_VERITY_SHARE
, d
);
2856 if (!p
->decrypted_fstype
&& p
->mount_node_fd
>= 0 && p
->decrypted_node
) {
2857 r
= probe_filesystem_full(p
->mount_node_fd
, p
->decrypted_node
, 0, UINT64_MAX
, &p
->decrypted_fstype
);
2858 if (r
< 0 && r
!= -EUCLEAN
)
2863 m
->decrypted_image
= TAKE_PTR(d
);
2871 int dissected_image_decrypt_interactively(
2873 const char *passphrase
,
2874 const VeritySettings
*verity
,
2875 DissectImageFlags flags
) {
2877 _cleanup_strv_free_erase_
char **z
= NULL
;
2884 r
= dissected_image_decrypt(m
, passphrase
, verity
, flags
);
2887 if (r
== -EKEYREJECTED
)
2888 log_error_errno(r
, "Incorrect passphrase, try again!");
2889 else if (r
!= -ENOKEY
)
2890 return log_error_errno(r
, "Failed to decrypt image: %m");
2893 return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED
),
2894 "Too many retries.");
2898 r
= ask_password_auto("Please enter image passphrase:", NULL
, "dissect", "dissect", "dissect.passphrase", USEC_INFINITY
, 0, &z
);
2900 return log_error_errno(r
, "Failed to query for passphrase: %m");
2906 static int decrypted_image_relinquish(DecryptedImage
*d
) {
2909 /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a
2910 * boolean so that we don't clean it up ourselves either anymore */
2912 #if HAVE_LIBCRYPTSETUP
2915 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
2916 DecryptedPartition
*p
= d
->decrypted
+ i
;
2918 if (p
->relinquished
)
2921 r
= sym_crypt_deactivate_by_name(NULL
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
2923 return log_debug_errno(r
, "Failed to mark %s for auto-removal: %m", p
->name
);
2925 p
->relinquished
= true;
2932 int dissected_image_relinquish(DissectedImage
*m
) {
2937 if (m
->decrypted_image
) {
2938 r
= decrypted_image_relinquish(m
->decrypted_image
);
2944 loop_device_relinquish(m
->loop
);
2949 static char *build_auxiliary_path(const char *image
, const char *suffix
) {
2956 e
= endswith(image
, ".raw");
2958 return strjoin(e
, suffix
);
2960 n
= new(char, e
- image
+ strlen(suffix
) + 1);
2964 strcpy(mempcpy(n
, image
, e
- image
), suffix
);
2968 void verity_settings_done(VeritySettings
*v
) {
2971 v
->root_hash
= mfree(v
->root_hash
);
2972 v
->root_hash_size
= 0;
2974 v
->root_hash_sig
= mfree(v
->root_hash_sig
);
2975 v
->root_hash_sig_size
= 0;
2977 v
->data_path
= mfree(v
->data_path
);
2980 int verity_settings_load(
2981 VeritySettings
*verity
,
2983 const char *root_hash_path
,
2984 const char *root_hash_sig_path
) {
2986 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
2987 size_t root_hash_size
= 0, root_hash_sig_size
= 0;
2988 _cleanup_free_
char *verity_data_path
= NULL
;
2989 PartitionDesignator designator
;
2994 assert(verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
2996 /* If we are asked to load the root hash for a device node, exit early */
2997 if (is_device_path(image
))
3000 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIDECAR");
3001 if (r
< 0 && r
!= -ENXIO
)
3002 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIDECAR, ignoring: %m");
3006 designator
= verity
->designator
;
3008 /* We only fill in what isn't already filled in */
3010 if (!verity
->root_hash
) {
3011 _cleanup_free_
char *text
= NULL
;
3013 if (root_hash_path
) {
3014 /* If explicitly specified it takes precedence */
3015 r
= read_one_line_file(root_hash_path
, &text
);
3020 designator
= PARTITION_ROOT
;
3022 /* Otherwise look for xattr and separate file, and first for the data for root and if
3023 * that doesn't exist for /usr */
3025 if (designator
< 0 || designator
== PARTITION_ROOT
) {
3026 r
= getxattr_malloc(image
, "user.verity.roothash", &text
);
3028 _cleanup_free_
char *p
= NULL
;
3030 if (r
!= -ENOENT
&& !ERRNO_IS_XATTR_ABSENT(r
))
3033 p
= build_auxiliary_path(image
, ".roothash");
3037 r
= read_one_line_file(p
, &text
);
3038 if (r
< 0 && r
!= -ENOENT
)
3043 designator
= PARTITION_ROOT
;
3046 if (!text
&& (designator
< 0 || designator
== PARTITION_USR
)) {
3047 /* So in the "roothash" xattr/file name above the "root" of course primarily
3048 * refers to the root of the Verity Merkle tree. But coincidentally it also
3049 * is the hash for the *root* file system, i.e. the "root" neatly refers to
3050 * two distinct concepts called "root". Taking benefit of this happy
3051 * coincidence we call the file with the root hash for the /usr/ file system
3052 * `usrhash`, because `usrroothash` or `rootusrhash` would just be too
3053 * confusing. We thus drop the reference to the root of the Merkle tree, and
3054 * just indicate which file system it's about. */
3055 r
= getxattr_malloc(image
, "user.verity.usrhash", &text
);
3057 _cleanup_free_
char *p
= NULL
;
3059 if (r
!= -ENOENT
&& !ERRNO_IS_XATTR_ABSENT(r
))
3062 p
= build_auxiliary_path(image
, ".usrhash");
3066 r
= read_one_line_file(p
, &text
);
3067 if (r
< 0 && r
!= -ENOENT
)
3072 designator
= PARTITION_USR
;
3077 r
= unhexmem(text
, strlen(text
), &root_hash
, &root_hash_size
);
3080 if (root_hash_size
< sizeof(sd_id128_t
))
3085 if ((root_hash
|| verity
->root_hash
) && !verity
->root_hash_sig
) {
3086 if (root_hash_sig_path
) {
3087 r
= read_full_file(root_hash_sig_path
, (char**) &root_hash_sig
, &root_hash_sig_size
);
3088 if (r
< 0 && r
!= -ENOENT
)
3092 designator
= PARTITION_ROOT
;
3094 if (designator
< 0 || designator
== PARTITION_ROOT
) {
3095 _cleanup_free_
char *p
= NULL
;
3097 /* Follow naming convention recommended by the relevant RFC:
3098 * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
3099 p
= build_auxiliary_path(image
, ".roothash.p7s");
3103 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
3104 if (r
< 0 && r
!= -ENOENT
)
3107 designator
= PARTITION_ROOT
;
3110 if (!root_hash_sig
&& (designator
< 0 || designator
== PARTITION_USR
)) {
3111 _cleanup_free_
char *p
= NULL
;
3113 p
= build_auxiliary_path(image
, ".usrhash.p7s");
3117 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
3118 if (r
< 0 && r
!= -ENOENT
)
3121 designator
= PARTITION_USR
;
3125 if (root_hash_sig
&& root_hash_sig_size
== 0) /* refuse empty size signatures */
3129 if (!verity
->data_path
) {
3130 _cleanup_free_
char *p
= NULL
;
3132 p
= build_auxiliary_path(image
, ".verity");
3136 if (access(p
, F_OK
) < 0) {
3137 if (errno
!= ENOENT
)
3140 verity_data_path
= TAKE_PTR(p
);
3144 verity
->root_hash
= TAKE_PTR(root_hash
);
3145 verity
->root_hash_size
= root_hash_size
;
3148 if (root_hash_sig
) {
3149 verity
->root_hash_sig
= TAKE_PTR(root_hash_sig
);
3150 verity
->root_hash_sig_size
= root_hash_sig_size
;
3153 if (verity_data_path
)
3154 verity
->data_path
= TAKE_PTR(verity_data_path
);
3156 if (verity
->designator
< 0)
3157 verity
->designator
= designator
;
3162 int dissected_image_load_verity_sig_partition(
3165 VeritySettings
*verity
) {
3167 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
3168 _cleanup_(json_variant_unrefp
) JsonVariant
*v
= NULL
;
3169 size_t root_hash_size
, root_hash_sig_size
;
3170 _cleanup_free_
char *buf
= NULL
;
3171 PartitionDesignator d
;
3172 DissectedPartition
*p
;
3173 JsonVariant
*rh
, *sig
;
3182 if (verity
->root_hash
&& verity
->root_hash_sig
) /* Already loaded? */
3185 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_EMBEDDED");
3186 if (r
< 0 && r
!= -ENXIO
)
3187 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_EMBEDDED, ignoring: %m");
3191 d
= partition_verity_sig_of(verity
->designator
< 0 ? PARTITION_ROOT
: verity
->designator
);
3194 p
= m
->partitions
+ d
;
3197 if (p
->offset
== UINT64_MAX
|| p
->size
== UINT64_MAX
)
3200 if (p
->size
> 4*1024*1024) /* Signature data cannot possible be larger than 4M, refuse that */
3201 return log_debug_errno(SYNTHETIC_ERRNO(EFBIG
), "Verity signature partition is larger than 4M, refusing.");
3203 buf
= new(char, p
->size
+1);
3207 n
= pread(fd
, buf
, p
->size
, p
->offset
);
3210 if ((uint64_t) n
!= p
->size
)
3213 e
= memchr(buf
, 0, p
->size
);
3215 /* If we found a NUL byte then the rest of the data must be NUL too */
3216 if (!memeqzero(e
, p
->size
- (e
- buf
)))
3217 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature data contains embedded NUL byte.");
3221 r
= json_parse(buf
, 0, &v
, NULL
, NULL
);
3223 return log_debug_errno(r
, "Failed to parse signature JSON data: %m");
3225 rh
= json_variant_by_key(v
, "rootHash");
3227 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'rootHash' field.");
3228 if (!json_variant_is_string(rh
))
3229 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'rootHash' field of signature JSON object is not a string.");
3231 r
= unhexmem(json_variant_string(rh
), SIZE_MAX
, &root_hash
, &root_hash_size
);
3233 return log_debug_errno(r
, "Failed to parse root hash field: %m");
3235 /* Check if specified root hash matches if it is specified */
3236 if (verity
->root_hash
&&
3237 memcmp_nn(verity
->root_hash
, verity
->root_hash_size
, root_hash
, root_hash_size
) != 0) {
3238 _cleanup_free_
char *a
= NULL
, *b
= NULL
;
3240 a
= hexmem(root_hash
, root_hash_size
);
3241 b
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
3243 return log_debug_errno(r
, "Root hash in signature JSON data (%s) doesn't match configured hash (%s).", strna(a
), strna(b
));
3246 sig
= json_variant_by_key(v
, "signature");
3248 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'signature' field.");
3249 if (!json_variant_is_string(sig
))
3250 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'signature' field of signature JSON object is not a string.");
3252 r
= unbase64mem(json_variant_string(sig
), SIZE_MAX
, &root_hash_sig
, &root_hash_sig_size
);
3254 return log_debug_errno(r
, "Failed to parse signature field: %m");
3256 free_and_replace(verity
->root_hash
, root_hash
);
3257 verity
->root_hash_size
= root_hash_size
;
3259 free_and_replace(verity
->root_hash_sig
, root_hash_sig
);
3260 verity
->root_hash_sig_size
= root_hash_sig_size
;
3265 int dissected_image_acquire_metadata(DissectedImage
*m
, DissectImageFlags extra_flags
) {
3272 META_INITRD_RELEASE
,
3273 META_EXTENSION_RELEASE
,
3274 META_HAS_INIT_SYSTEM
,
3278 static const char *const paths
[_META_MAX
] = {
3279 [META_HOSTNAME
] = "/etc/hostname\0",
3280 [META_MACHINE_ID
] = "/etc/machine-id\0",
3281 [META_MACHINE_INFO
] = "/etc/machine-info\0",
3282 [META_OS_RELEASE
] = ("/etc/os-release\0"
3283 "/usr/lib/os-release\0"),
3284 [META_INITRD_RELEASE
] = ("/etc/initrd-release\0"
3285 "/usr/lib/initrd-release\0"),
3286 [META_EXTENSION_RELEASE
] = "extension-release\0", /* Used only for logging. */
3287 [META_HAS_INIT_SYSTEM
] = "has-init-system\0", /* ditto */
3290 _cleanup_strv_free_
char **machine_info
= NULL
, **os_release
= NULL
, **initrd_release
= NULL
, **extension_release
= NULL
;
3291 _cleanup_close_pair_
int error_pipe
[2] = PIPE_EBADF
;
3292 _cleanup_(rmdir_and_freep
) char *t
= NULL
;
3293 _cleanup_(sigkill_waitp
) pid_t child
= 0;
3294 sd_id128_t machine_id
= SD_ID128_NULL
;
3295 _cleanup_free_
char *hostname
= NULL
;
3296 unsigned n_meta_initialized
= 0;
3297 int fds
[2 * _META_MAX
], r
, v
;
3298 int has_init_system
= -1;
3300 ImageClass image_class
= IMAGE_SYSEXT
;
3302 BLOCK_SIGNALS(SIGCHLD
);
3305 assert(image_class
);
3307 for (; n_meta_initialized
< _META_MAX
; n_meta_initialized
++) {
3308 if (!paths
[n_meta_initialized
]) {
3309 fds
[2*n_meta_initialized
] = fds
[2*n_meta_initialized
+1] = -EBADF
;
3313 if (pipe2(fds
+ 2*n_meta_initialized
, O_CLOEXEC
) < 0) {
3319 r
= mkdtemp_malloc("/tmp/dissect-XXXXXX", &t
);
3323 if (pipe2(error_pipe
, O_CLOEXEC
) < 0) {
3328 r
= safe_fork("(sd-dissect)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_NEW_MOUNTNS
|FORK_MOUNTNS_SLAVE
, &child
);
3332 /* Child in a new mount namespace */
3333 error_pipe
[0] = safe_close(error_pipe
[0]);
3335 r
= dissected_image_mount(
3341 DISSECT_IMAGE_READ_ONLY
|
3342 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
3343 DISSECT_IMAGE_USR_NO_ROOT
);
3345 log_debug_errno(r
, "Failed to mount dissected image: %m");
3349 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
3350 _cleanup_close_
int fd
= -ENOENT
;
3355 fds
[2*k
] = safe_close(fds
[2*k
]);
3359 case META_EXTENSION_RELEASE
: {
3360 /* As per the os-release spec, if the image is an extension it will have a file
3361 * named after the image name in extension-release.d/ - we use the image name
3362 * and try to resolve it with the extension-release helpers, as sometimes
3363 * the image names are mangled on deployment and do not match anymore.
3364 * Unlike other paths this is not fixed, and the image name
3365 * can be mangled on deployment, so by calling into the helper
3366 * we allow a fallback that matches on the first extension-release
3367 * file found in the directory, if one named after the image cannot
3368 * be found first. */
3369 ImageClass
class = IMAGE_SYSEXT
;
3370 r
= open_extension_release(t
, IMAGE_SYSEXT
, m
->image_name
, /* relax_extension_release_check= */ false, NULL
, &fd
);
3372 r
= open_extension_release(t
, IMAGE_CONFEXT
, m
->image_name
, /* relax_extension_release_check= */ false, NULL
, &fd
);
3374 class = IMAGE_CONFEXT
;
3379 r
= loop_write(fds
[2*k
+1], &class, sizeof(class), false);
3381 goto inner_fail
; /* Propagate the error to the parent */
3387 case META_HAS_INIT_SYSTEM
: {
3390 FOREACH_STRING(init
,
3391 "/usr/lib/systemd/systemd", /* systemd on /usr merged system */
3392 "/lib/systemd/systemd", /* systemd on /usr non-merged systems */
3393 "/sbin/init") { /* traditional path the Linux kernel invokes */
3395 r
= chase(init
, t
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
3398 log_debug_errno(r
, "Failed to resolve %s, ignoring: %m", init
);
3405 r
= loop_write(fds
[2*k
+1], &found
, sizeof(found
), false);
3413 NULSTR_FOREACH(p
, paths
[k
]) {
3414 fd
= chase_and_open(p
, t
, CHASE_PREFIX_ROOT
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
, NULL
);
3421 log_debug_errno(fd
, "Failed to read %s file of image, ignoring: %m", paths
[k
]);
3422 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3426 r
= copy_bytes(fd
, fds
[2*k
+1], UINT64_MAX
, 0);
3430 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3433 _exit(EXIT_SUCCESS
);
3436 /* Let parent know the error */
3437 (void) write(error_pipe
[1], &r
, sizeof(r
));
3438 _exit(EXIT_FAILURE
);
3441 error_pipe
[1] = safe_close(error_pipe
[1]);
3443 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
3444 _cleanup_fclose_
FILE *f
= NULL
;
3449 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3451 f
= take_fdopen(&fds
[2*k
], "r");
3460 r
= read_etc_hostname_stream(f
, &hostname
);
3462 log_debug_errno(r
, "Failed to read /etc/hostname of image: %m");
3466 case META_MACHINE_ID
: {
3467 _cleanup_free_
char *line
= NULL
;
3469 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3471 log_debug_errno(r
, "Failed to read /etc/machine-id of image: %m");
3473 r
= sd_id128_from_string(line
, &machine_id
);
3475 log_debug_errno(r
, "Image contains invalid /etc/machine-id: %s", line
);
3477 log_debug("/etc/machine-id file of image is empty.");
3478 else if (streq(line
, "uninitialized"))
3479 log_debug("/etc/machine-id file of image is uninitialized (likely aborted first boot).");
3481 log_debug("/etc/machine-id file of image has unexpected length %i.", r
);
3486 case META_MACHINE_INFO
:
3487 r
= load_env_file_pairs(f
, "machine-info", &machine_info
);
3489 log_debug_errno(r
, "Failed to read /etc/machine-info of image: %m");
3493 case META_OS_RELEASE
:
3494 r
= load_env_file_pairs(f
, "os-release", &os_release
);
3496 log_debug_errno(r
, "Failed to read OS release file of image: %m");
3500 case META_INITRD_RELEASE
:
3501 r
= load_env_file_pairs(f
, "initrd-release", &initrd_release
);
3503 log_debug_errno(r
, "Failed to read initrd release file of image: %m");
3507 case META_EXTENSION_RELEASE
: {
3508 ImageClass cl
= IMAGE_SYSEXT
;
3512 nr
= fread(&cl
, 1, sizeof(cl
), f
);
3513 if (nr
!= sizeof(cl
))
3514 log_debug_errno(errno_or_else(EIO
), "Failed to read class of extension image: %m");
3517 r
= load_env_file_pairs(f
, "extension-release", &extension_release
);
3519 log_debug_errno(r
, "Failed to read extension release file of image: %m");
3525 case META_HAS_INIT_SYSTEM
: {
3530 nr
= fread(&b
, 1, sizeof(b
), f
);
3531 if (nr
!= sizeof(b
))
3532 log_debug_errno(errno_or_else(EIO
), "Failed to read has-init-system boolean: %m");
3534 has_init_system
= b
;
3540 r
= wait_for_terminate_and_check("(sd-dissect)", child
, 0);
3545 n
= read(error_pipe
[0], &v
, sizeof(v
));
3549 return v
; /* propagate error sent to us from child */
3553 if (r
!= EXIT_SUCCESS
)
3556 free_and_replace(m
->hostname
, hostname
);
3557 m
->machine_id
= machine_id
;
3558 strv_free_and_replace(m
->machine_info
, machine_info
);
3559 strv_free_and_replace(m
->os_release
, os_release
);
3560 strv_free_and_replace(m
->initrd_release
, initrd_release
);
3561 strv_free_and_replace(m
->extension_release
, extension_release
);
3562 m
->has_init_system
= has_init_system
;
3563 m
->image_class
= image_class
;
3566 for (unsigned k
= 0; k
< n_meta_initialized
; k
++)
3567 safe_close_pair(fds
+ 2*k
);
3572 Architecture
dissected_image_architecture(DissectedImage
*img
) {
3575 if (img
->partitions
[PARTITION_ROOT
].found
&&
3576 img
->partitions
[PARTITION_ROOT
].architecture
>= 0)
3577 return img
->partitions
[PARTITION_ROOT
].architecture
;
3579 if (img
->partitions
[PARTITION_USR
].found
&&
3580 img
->partitions
[PARTITION_USR
].architecture
>= 0)
3581 return img
->partitions
[PARTITION_USR
].architecture
;
3583 return _ARCHITECTURE_INVALID
;
3586 int dissect_loop_device(
3588 const VeritySettings
*verity
,
3589 const MountOptions
*mount_options
,
3590 const ImagePolicy
*image_policy
,
3591 DissectImageFlags flags
,
3592 DissectedImage
**ret
) {
3595 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
3600 r
= dissected_image_new(loop
->backing_file
?: loop
->node
, &m
);
3604 m
->loop
= loop_device_ref(loop
);
3605 m
->sector_size
= m
->loop
->sector_size
;
3607 r
= dissect_image(m
, loop
->fd
, loop
->node
, verity
, mount_options
, image_policy
, flags
);
3620 int dissect_loop_device_and_warn(
3622 const VeritySettings
*verity
,
3623 const MountOptions
*mount_options
,
3624 const ImagePolicy
*image_policy
,
3625 DissectImageFlags flags
,
3626 DissectedImage
**ret
) {
3630 return dissect_log_error(
3632 dissect_loop_device(loop
, verity
, mount_options
, image_policy
, flags
, ret
),
3633 loop
->backing_file
?: loop
->node
,
3638 bool dissected_image_verity_candidate(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3641 /* Checks if this partition could theoretically do Verity. For non-partitioned images this only works
3642 * if there's an external verity file supplied, for which we can consult .has_verity. For partitioned
3643 * images we only check the partition type.
3645 * This call is used to decide whether to suppress or show a verity column in tabular output of the
3648 if (image
->single_file_system
)
3649 return partition_designator
== PARTITION_ROOT
&& image
->has_verity
;
3651 return partition_verity_of(partition_designator
) >= 0;
3654 bool dissected_image_verity_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3655 PartitionDesignator k
;
3659 /* Checks if this partition has verity data available that we can activate. For non-partitioned this
3660 * works for the root partition, for others only if the associated verity partition was found. */
3662 if (!image
->verity_ready
)
3665 if (image
->single_file_system
)
3666 return partition_designator
== PARTITION_ROOT
;
3668 k
= partition_verity_of(partition_designator
);
3669 return k
>= 0 && image
->partitions
[k
].found
;
3672 bool dissected_image_verity_sig_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3673 PartitionDesignator k
;
3677 /* Checks if this partition has verity signature data available that we can use. */
3679 if (!image
->verity_sig_ready
)
3682 if (image
->single_file_system
)
3683 return partition_designator
== PARTITION_ROOT
;
3685 k
= partition_verity_sig_of(partition_designator
);
3686 return k
>= 0 && image
->partitions
[k
].found
;
3689 MountOptions
* mount_options_free_all(MountOptions
*options
) {
3692 while ((m
= LIST_POP(mount_options
, options
))) {
3700 const char* mount_options_from_designator(const MountOptions
*options
, PartitionDesignator designator
) {
3701 LIST_FOREACH(mount_options
, m
, options
)
3702 if (designator
== m
->partition_designator
&& !isempty(m
->options
))
3708 int mount_image_privately_interactively(
3710 const ImagePolicy
*image_policy
,
3711 DissectImageFlags flags
,
3712 char **ret_directory
,
3714 LoopDevice
**ret_loop_device
) {
3716 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3717 _cleanup_(loop_device_unrefp
) LoopDevice
*d
= NULL
;
3718 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3719 _cleanup_free_
char *dir
= NULL
;
3722 /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
3723 * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
3727 assert(ret_loop_device
);
3729 /* We intend to mount this right-away, hence add the partitions if needed and pin them. */
3730 flags
|= DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
3731 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
3733 r
= verity_settings_load(&verity
, image
, NULL
, NULL
);
3735 return log_error_errno(r
, "Failed to load root hash data: %m");
3737 r
= loop_device_make_by_path(
3739 FLAGS_SET(flags
, DISSECT_IMAGE_DEVICE_READ_ONLY
) ? O_RDONLY
: O_RDWR
,
3740 /* sector_size= */ UINT32_MAX
,
3741 FLAGS_SET(flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
3745 return log_error_errno(r
, "Failed to set up loopback device for %s: %m", image
);
3747 r
= dissect_loop_device_and_warn(
3750 /* mount_options= */ NULL
,
3757 r
= dissected_image_load_verity_sig_partition(dissected_image
, d
->fd
, &verity
);
3761 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, &verity
, flags
);
3765 r
= detach_mount_namespace();
3767 return log_error_errno(r
, "Failed to detach mount namespace: %m");
3769 r
= mkdir_p("/run/systemd/mount-rootfs", 0555);
3771 return log_error_errno(r
, "Failed to create mount point: %m");
3773 r
= dissected_image_mount_and_warn(
3775 "/run/systemd/mount-rootfs",
3776 /* uid_shift= */ UID_INVALID
,
3777 /* uid_range= */ UID_INVALID
,
3782 r
= loop_device_flock(d
, LOCK_UN
);
3786 r
= dissected_image_relinquish(dissected_image
);
3788 return log_error_errno(r
, "Failed to relinquish DM and loopback block devices: %m");
3790 if (ret_directory
) {
3791 dir
= strdup("/run/systemd/mount-rootfs");
3797 _cleanup_close_
int dir_fd
= -EBADF
;
3799 dir_fd
= open("/run/systemd/mount-rootfs", O_CLOEXEC
|O_DIRECTORY
);
3801 return log_error_errno(errno
, "Failed to open mount point directory: %m");
3803 *ret_dir_fd
= TAKE_FD(dir_fd
);
3807 *ret_directory
= TAKE_PTR(dir
);
3809 *ret_loop_device
= TAKE_PTR(d
);
3813 static bool mount_options_relax_extension_release_checks(const MountOptions
*options
) {
3817 return string_contains_word(mount_options_from_designator(options
, PARTITION_ROOT
), ",", "x-systemd.relax-extension-release-check") ||
3818 string_contains_word(mount_options_from_designator(options
, PARTITION_USR
), ",", "x-systemd.relax-extension-release-check") ||
3819 string_contains_word(options
->options
, ",", "x-systemd.relax-extension-release-check");
3822 int verity_dissect_and_mount(
3826 const MountOptions
*options
,
3827 const ImagePolicy
*image_policy
,
3828 const char *required_host_os_release_id
,
3829 const char *required_host_os_release_version_id
,
3830 const char *required_host_os_release_sysext_level
,
3831 const char *required_sysext_scope
) {
3833 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
3834 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3835 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3836 DissectImageFlags dissect_image_flags
;
3837 bool relax_extension_release_check
;
3843 relax_extension_release_check
= mount_options_relax_extension_release_checks(options
);
3845 /* We might get an FD for the image, but we use the original path to look for the dm-verity files */
3846 r
= verity_settings_load(&verity
, src
, NULL
, NULL
);
3848 return log_debug_errno(r
, "Failed to load root hash: %m");
3850 dissect_image_flags
= (verity
.data_path
? DISSECT_IMAGE_NO_PARTITION_TABLE
: 0) |
3851 (relax_extension_release_check
? DISSECT_IMAGE_RELAX_EXTENSION_CHECK
: 0) |
3852 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
3853 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
3855 /* Note that we don't use loop_device_make here, as the FD is most likely O_PATH which would not be
3856 * accepted by LOOP_CONFIGURE, so just let loop_device_make_by_path reopen it as a regular FD. */
3857 r
= loop_device_make_by_path(
3858 src_fd
>= 0 ? FORMAT_PROC_FD_PATH(src_fd
) : src
,
3859 /* open_flags= */ -1,
3860 /* sector_size= */ UINT32_MAX
,
3861 verity
.data_path
? 0 : LO_FLAGS_PARTSCAN
,
3865 return log_debug_errno(r
, "Failed to create loop device for image: %m");
3867 r
= dissect_loop_device(
3872 dissect_image_flags
,
3874 /* No partition table? Might be a single-filesystem image, try again */
3875 if (!verity
.data_path
&& r
== -ENOPKG
)
3876 r
= dissect_loop_device(
3881 dissect_image_flags
| DISSECT_IMAGE_NO_PARTITION_TABLE
,
3884 return log_debug_errno(r
, "Failed to dissect image: %m");
3886 r
= dissected_image_load_verity_sig_partition(dissected_image
, loop_device
->fd
, &verity
);
3890 r
= dissected_image_decrypt(
3894 dissect_image_flags
);
3896 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
3898 r
= mkdir_p_label(dest
, 0755);
3900 return log_debug_errno(r
, "Failed to create destination directory %s: %m", dest
);
3901 r
= umount_recursive(dest
, 0);
3903 return log_debug_errno(r
, "Failed to umount under destination directory %s: %m", dest
);
3905 r
= dissected_image_mount(dissected_image
, dest
, UID_INVALID
, UID_INVALID
, dissect_image_flags
);
3907 return log_debug_errno(r
, "Failed to mount image: %m");
3909 r
= loop_device_flock(loop_device
, LOCK_UN
);
3911 return log_debug_errno(r
, "Failed to unlock loopback device: %m");
3913 /* If we got os-release values from the caller, then we need to match them with the image's
3914 * extension-release.d/ content. Return -EINVAL if there's any mismatch.
3915 * First, check the distro ID. If that matches, then check the new SYSEXT_LEVEL value if
3916 * available, or else fallback to VERSION_ID. If neither is present (eg: rolling release),
3917 * then a simple match on the ID will be performed. */
3918 if (required_host_os_release_id
) {
3919 _cleanup_strv_free_
char **extension_release
= NULL
;
3920 ImageClass
class = IMAGE_SYSEXT
;
3922 assert(!isempty(required_host_os_release_id
));
3924 r
= load_extension_release_pairs(dest
, IMAGE_SYSEXT
, dissected_image
->image_name
, relax_extension_release_check
, &extension_release
);
3926 r
= load_extension_release_pairs(dest
, IMAGE_CONFEXT
, dissected_image
->image_name
, relax_extension_release_check
, &extension_release
);
3928 class = IMAGE_CONFEXT
;
3931 return log_debug_errno(r
, "Failed to parse image %s extension-release metadata: %m", dissected_image
->image_name
);
3933 r
= extension_release_validate(
3934 dissected_image
->image_name
,
3935 required_host_os_release_id
,
3936 required_host_os_release_version_id
,
3937 required_host_os_release_sysext_level
,
3938 required_sysext_scope
,
3942 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE
), "Image %s extension-release metadata does not match the root's", dissected_image
->image_name
);
3944 return log_debug_errno(r
, "Failed to compare image %s extension-release metadata with the root's os-release: %m", dissected_image
->image_name
);
3947 r
= dissected_image_relinquish(dissected_image
);
3949 return log_debug_errno(r
, "Failed to relinquish dissected image: %m");