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
,
109 log_debug("File system type '%s' is not allowed to be mounted as result of automatic dissection.", fstype
);
113 int probe_sector_size(int fd
, uint32_t *ret
) {
115 /* Disk images might be for 512B or for 4096 sector sizes, let's try to auto-detect that by searching
116 * for the GPT headers at the relevant byte offsets */
118 assert_cc(sizeof(GptHeader
) == 92);
120 /* We expect a sector size in the range 512…4096. The GPT header is located in the second
121 * sector. Hence it could be at byte 512 at the earliest, and at byte 4096 at the latest. And we must
122 * read with granularity of the largest sector size we care about. Which means 8K. */
123 uint8_t sectors
[2 * 4096];
130 n
= pread(fd
, sectors
, sizeof(sectors
), 0);
133 if (n
!= sizeof(sectors
)) /* too short? */
136 /* Let's see if we find the GPT partition header with various expected sector sizes */
137 for (uint32_t sz
= 512; sz
<= 4096; sz
<<= 1) {
140 assert(sizeof(sectors
) >= sz
* 2);
141 p
= (const GptHeader
*) (sectors
+ sz
);
143 if (!gpt_header_has_signature(p
))
147 return log_debug_errno(SYNTHETIC_ERRNO(ENOTUNIQ
),
148 "Detected valid partition table at offsets matching multiple sector sizes, refusing.");
154 log_debug("Determined sector size %" PRIu32
" based on discovered partition table.", found
);
156 return 1; /* indicate we *did* find it */
160 log_debug("Couldn't find any partition table to derive sector size of.");
161 *ret
= 512; /* pick the traditional default */
162 return 0; /* indicate we didn't find it */
165 int probe_sector_size_prefer_ioctl(int fd
, uint32_t *ret
) {
171 /* Just like probe_sector_size(), but if we are looking at a block device, will use the already
172 * configured sector size rather than probing by contents */
174 if (fstat(fd
, &st
) < 0)
177 if (S_ISBLK(st
.st_mode
))
178 return blockdev_get_sector_size(fd
, ret
);
180 return probe_sector_size(fd
, ret
);
183 int probe_filesystem_full(
190 /* Try to find device content type and return it in *ret_fstype. If nothing is found,
191 * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and a
192 * different error otherwise. */
195 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
196 _cleanup_free_
char *path_by_fd
= NULL
;
197 _cleanup_close_
int fd_close
= -EBADF
;
201 assert(fd
>= 0 || path
);
205 fd_close
= open(path
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
213 r
= fd_get_path(fd
, &path_by_fd
);
220 if (size
== 0) /* empty size? nothing found! */
223 b
= blkid_new_probe();
227 /* The Linux kernel maintains separate block device caches for main ("whole") and partition block
228 * devices, which means making a change to one might not be reflected immediately when reading via
229 * the other. That's massively confusing when mixing accesses to such devices. Let's address this in
230 * a limited way: when probing a file system that is not at the beginning of the block device we
231 * apparently probe a partition via the main block device, and in that case let's first flush the
232 * main block device cache, so that we get the data that the per-partition block device last
235 * This only works under the assumption that any tools that write to the partition block devices
236 * issue an syncfs()/fsync() on the device after making changes. Typically file system formatting
237 * tools that write a superblock onto a partition block device do that, however. */
239 if (ioctl(fd
, BLKFLSBUF
, 0) < 0)
240 log_debug_errno(errno
, "Failed to flush block device cache, ignoring: %m");
243 r
= blkid_probe_set_device(
247 size
== UINT64_MAX
? 0 : size
); /* when blkid sees size=0 it understands "everything". We prefer using UINT64_MAX for that */
249 return errno_or_else(ENOMEM
);
251 blkid_probe_enable_superblocks(b
, 1);
252 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
);
255 r
= blkid_do_safeprobe(b
);
256 if (r
== _BLKID_SAFEPROBE_NOT_FOUND
)
258 if (r
== _BLKID_SAFEPROBE_AMBIGUOUS
)
259 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
),
260 "Results ambiguous for partition %s", path
);
261 if (r
== _BLKID_SAFEPROBE_ERROR
)
262 return log_debug_errno(errno_or_else(EIO
), "Failed to probe partition %s: %m", path
);
264 assert(r
== _BLKID_SAFEPROBE_FOUND
);
266 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
271 log_debug("Probed fstype '%s' on partition %s.", fstype
, path
);
282 log_debug("No type detected on partition %s", path
);
291 static int image_policy_may_use(
292 const ImagePolicy
*policy
,
293 PartitionDesignator designator
) {
295 PartitionPolicyFlags f
;
297 /* For each partition we find in the partition table do a first check if it may exist at all given
298 * the policy, or if it shall be ignored. */
300 f
= image_policy_get_exhaustively(policy
, designator
);
304 if ((f
& _PARTITION_POLICY_USE_MASK
) == PARTITION_POLICY_ABSENT
)
305 /* only flag set in policy is "absent"? then this partition may not exist at all */
306 return log_debug_errno(
307 SYNTHETIC_ERRNO(ERFKILL
),
308 "Partition of designator '%s' exists, but not allowed by policy, refusing.",
309 partition_designator_to_string(designator
));
310 if ((f
& _PARTITION_POLICY_USE_MASK
& ~PARTITION_POLICY_ABSENT
) == PARTITION_POLICY_UNUSED
) {
311 /* only "unused" or "unused" + "absent" are set? then don't use it */
312 log_debug("Partition of designator '%s' exists, and policy dictates to ignore it, doing so.",
313 partition_designator_to_string(designator
));
314 return false; /* ignore! */
317 return true; /* use! */
320 static int image_policy_check_protection(
321 const ImagePolicy
*policy
,
322 PartitionDesignator designator
,
323 PartitionPolicyFlags found_flags
) {
325 PartitionPolicyFlags policy_flags
;
327 /* Checks if the flags in the policy for the designated partition overlap the flags of what we found */
332 policy_flags
= image_policy_get_exhaustively(policy
, designator
);
333 if (policy_flags
< 0)
336 if ((found_flags
& policy_flags
) == 0) {
337 _cleanup_free_
char *found_flags_string
= NULL
, *policy_flags_string
= NULL
;
339 (void) partition_policy_flags_to_string(found_flags
, /* simplify= */ true, &found_flags_string
);
340 (void) partition_policy_flags_to_string(policy_flags
, /* simplify= */ true, &policy_flags_string
);
342 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL
), "Partition %s discovered with policy '%s' but '%s' was required, refusing.",
343 partition_designator_to_string(designator
),
344 strnull(found_flags_string
), strnull(policy_flags_string
));
350 static int image_policy_check_partition_flags(
351 const ImagePolicy
*policy
,
352 PartitionDesignator designator
,
353 uint64_t gpt_flags
) {
355 PartitionPolicyFlags policy_flags
;
358 /* Checks if the partition flags in the policy match reality */
360 policy_flags
= image_policy_get_exhaustively(policy
, designator
);
361 if (policy_flags
< 0)
364 b
= FLAGS_SET(gpt_flags
, SD_GPT_FLAG_READ_ONLY
);
365 if ((policy_flags
& _PARTITION_POLICY_READ_ONLY_MASK
) == (b
? PARTITION_POLICY_READ_ONLY_OFF
: PARTITION_POLICY_READ_ONLY_ON
))
366 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL
), "Partition %s has 'read-only' flag incorrectly set (must be %s, is %s), refusing.",
367 partition_designator_to_string(designator
),
368 one_zero(!b
), one_zero(b
));
370 b
= FLAGS_SET(gpt_flags
, SD_GPT_FLAG_GROWFS
);
371 if ((policy_flags
& _PARTITION_POLICY_GROWFS_MASK
) == (b
? PARTITION_POLICY_GROWFS_OFF
: PARTITION_POLICY_GROWFS_ON
))
372 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL
), "Partition %s has 'growfs' flag incorrectly set (must be %s, is %s), refusing.",
373 partition_designator_to_string(designator
),
374 one_zero(!b
), one_zero(b
));
379 static int dissected_image_probe_filesystems(
382 const ImagePolicy
*policy
) {
388 /* Fill in file system types if we don't know them yet. */
390 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
391 DissectedPartition
*p
= m
->partitions
+ i
;
392 PartitionPolicyFlags found_flags
;
398 /* If we have an fd referring to the partition block device, use that. Otherwise go
399 * via the whole block device or backing regular file, and read via offset. */
400 if (p
->mount_node_fd
>= 0)
401 r
= probe_filesystem_full(p
->mount_node_fd
, p
->node
, 0, UINT64_MAX
, &p
->fstype
);
403 r
= probe_filesystem_full(fd
, p
->node
, p
->offset
, p
->size
, &p
->fstype
);
408 if (streq_ptr(p
->fstype
, "crypto_LUKS")) {
410 found_flags
= PARTITION_POLICY_ENCRYPTED
; /* found this one, and its definitely encrypted */
412 /* found it, but it's definitely not encrypted, hence mask the encrypted flag, but
413 * set all other ways that indicate "present". */
414 found_flags
= PARTITION_POLICY_UNPROTECTED
|PARTITION_POLICY_VERITY
|PARTITION_POLICY_SIGNED
;
416 if (p
->fstype
&& fstype_is_ro(p
->fstype
))
422 /* We might have learnt more about the file system now (i.e. whether it is encrypted or not),
423 * hence we need to validate this against policy again, to see if the policy still matches
424 * with this new information. Note that image_policy_check_protection() will check for
425 * overlap between what's allowed in the policy and what we pass as 'found_policy' here. In
426 * the unencrypted case we thus might pass an overly unspecific mask here (i.e. unprotected
427 * OR verity OR signed), but that's fine since the earlier policy check already checked more
428 * specific which of those three cases where OK. Keep in mind that this function here only
429 * looks at specific partitions (and thus can only deduce encryption or not) but not the
430 * overall partition table (and thus cannot deduce verity or not). The earlier dissection
431 * checks already did the relevant checks that look at the whole partition table, and
432 * enforced policy there as needed. */
433 r
= image_policy_check_protection(policy
, i
, found_flags
);
441 static void check_partition_flags(
443 unsigned long long pflags
,
444 unsigned long long supported
) {
448 /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
449 pflags
&= ~(supported
|
450 SD_GPT_FLAG_REQUIRED_PARTITION
|
451 SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL
|
452 SD_GPT_FLAG_LEGACY_BIOS_BOOTABLE
);
457 /* If there are other bits set, then log about it, to make things discoverable */
458 for (unsigned i
= 0; i
< sizeof(pflags
) * 8; i
++) {
459 unsigned long long bit
= 1ULL << i
;
460 if (!FLAGS_SET(pflags
, bit
))
463 log_debug("Unexpected partition flag %llu set on %s!", bit
, node
);
467 static int dissected_image_new(const char *path
, DissectedImage
**ret
) {
468 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
469 _cleanup_free_
char *name
= NULL
;
475 _cleanup_free_
char *filename
= NULL
;
477 r
= path_extract_filename(path
, &filename
);
481 r
= raw_strip_suffixes(filename
, &name
);
485 if (!image_name_is_valid(name
)) {
486 log_debug("Image name %s is not valid, ignoring.", strna(name
));
491 m
= new(DissectedImage
, 1);
495 *m
= (DissectedImage
) {
496 .has_init_system
= -1,
497 .image_name
= TAKE_PTR(name
),
500 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++)
501 m
->partitions
[i
] = DISSECTED_PARTITION_NULL
;
508 static void dissected_partition_done(DissectedPartition
*p
) {
514 free(p
->decrypted_fstype
);
515 free(p
->decrypted_node
);
516 free(p
->mount_options
);
517 safe_close(p
->mount_node_fd
);
519 *p
= DISSECTED_PARTITION_NULL
;
523 static int make_partition_devname(
524 const char *whole_devname
,
527 DissectImageFlags flags
,
530 _cleanup_free_
char *s
= NULL
;
533 assert(whole_devname
);
534 assert(nr
!= 0); /* zero is not a valid partition nr */
537 if (!FLAGS_SET(flags
, DISSECT_IMAGE_DISKSEQ_DEVNODE
) || diskseq
== 0) {
539 /* Given a whole block device node name (e.g. /dev/sda or /dev/loop7) generate a partition
540 * device name (e.g. /dev/sda7 or /dev/loop7p5). The rule the kernel uses is simple: if whole
541 * block device node name ends in a digit, then suffix a 'p', followed by the partition
542 * number. Otherwise, just suffix the partition number without any 'p'. */
544 if (nr
< 0) { /* whole disk? */
545 s
= strdup(whole_devname
);
549 size_t l
= strlen(whole_devname
);
550 if (l
< 1) /* underflow check for the subtraction below */
553 bool need_p
= ascii_isdigit(whole_devname
[l
-1]); /* Last char a digit? */
555 if (asprintf(&s
, "%s%s%i", whole_devname
, need_p
? "p" : "", nr
) < 0)
559 if (nr
< 0) /* whole disk? */
560 r
= asprintf(&s
, "/dev/disk/by-diskseq/%" PRIu64
, diskseq
);
562 r
= asprintf(&s
, "/dev/disk/by-diskseq/%" PRIu64
"-part%i", diskseq
, nr
);
571 static int open_partition(
574 const LoopDevice
*loop
) {
576 _cleanup_(sd_device_unrefp
) sd_device
*dev
= NULL
;
577 _cleanup_close_
int fd
= -EBADF
;
584 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
588 /* Check if the block device is a child of (or equivalent to) the originally provided one. */
589 r
= block_device_new_from_fd(fd
, is_partition
? BLOCK_DEVICE_LOOKUP_WHOLE_DISK
: 0, &dev
);
593 r
= sd_device_get_devnum(dev
, &devnum
);
597 if (loop
->devno
!= devnum
)
600 /* Also check diskseq. */
601 if (loop
->diskseq
!= 0) {
604 r
= fd_get_diskseq(fd
, &diskseq
);
608 if (loop
->diskseq
!= diskseq
)
612 log_debug("Opened %s (fd=%i, whole_block_devnum=" DEVNUM_FORMAT_STR
", diskseq=%" PRIu64
").",
613 node
, fd
, DEVNUM_FORMAT_VAL(loop
->devno
), loop
->diskseq
);
617 static int compare_arch(Architecture a
, Architecture b
) {
621 if (a
== native_architecture())
624 if (b
== native_architecture())
627 #ifdef ARCHITECTURE_SECONDARY
628 if (a
== ARCHITECTURE_SECONDARY
)
631 if (b
== ARCHITECTURE_SECONDARY
)
638 static int dissect_image(
642 const VeritySettings
*verity
,
643 const MountOptions
*mount_options
,
644 const ImagePolicy
*policy
,
645 DissectImageFlags flags
) {
647 sd_id128_t root_uuid
= SD_ID128_NULL
, root_verity_uuid
= SD_ID128_NULL
;
648 sd_id128_t usr_uuid
= SD_ID128_NULL
, usr_verity_uuid
= SD_ID128_NULL
;
649 bool is_gpt
, is_mbr
, multiple_generic
= false,
650 generic_rw
= false, /* initialize to appease gcc */
651 generic_growfs
= false;
652 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
653 _cleanup_free_
char *generic_node
= NULL
;
654 sd_id128_t generic_uuid
= SD_ID128_NULL
;
655 const char *pttype
= NULL
, *sptuuid
= NULL
;
657 int r
, generic_nr
= -1, n_partitions
;
662 assert(!verity
|| verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
663 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
664 assert(!verity
|| verity
->root_hash_sig
|| verity
->root_hash_sig_size
== 0);
665 assert(!verity
|| (verity
->root_hash
|| !verity
->root_hash_sig
));
666 assert(!((flags
& DISSECT_IMAGE_GPT_ONLY
) && (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)));
667 assert(m
->sector_size
> 0);
669 /* Probes a disk image, and returns information about what it found in *ret.
671 * Returns -ENOPKG if no suitable partition table or file system could be found.
672 * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found.
673 * Returns -ENXIO if we couldn't find any partition suitable as root or /usr partition
674 * Returns -ENOTUNIQ if we only found multiple generic partitions and thus don't know what to do with that
675 * Returns -ERFKILL if image doesn't match image policy
676 * 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)
677 * Returns -EPROTONOSUPPORT if DISSECT_IMAGE_ADD_PARTITION_DEVICES is set but the block device does not have partition logic enabled
678 * Returns -ENOMSG if we didn't find a single usable partition (and DISSECT_IMAGE_REFUSE_EMPTY is set) */
680 uint64_t diskseq
= m
->loop
? m
->loop
->diskseq
: 0;
682 if (verity
&& verity
->root_hash
) {
683 sd_id128_t fsuuid
, vuuid
;
685 /* If a root hash is supplied, then we use the root partition that has a UUID that match the
686 * first 128-bit of the root hash. And we use the verity partition that has a UUID that match
687 * the final 128-bit. */
689 if (verity
->root_hash_size
< sizeof(sd_id128_t
))
692 memcpy(&fsuuid
, verity
->root_hash
, sizeof(sd_id128_t
));
693 memcpy(&vuuid
, (const uint8_t*) verity
->root_hash
+ verity
->root_hash_size
- sizeof(sd_id128_t
), sizeof(sd_id128_t
));
695 if (sd_id128_is_null(fsuuid
))
697 if (sd_id128_is_null(vuuid
))
700 /* If the verity data declares it's for the /usr partition, then search for that, in all
701 * other cases assume it's for the root partition. */
702 if (verity
->designator
== PARTITION_USR
) {
704 usr_verity_uuid
= vuuid
;
707 root_verity_uuid
= vuuid
;
711 b
= blkid_new_probe();
716 r
= blkid_probe_set_device(b
, fd
, 0, 0);
718 return errno_or_else(ENOMEM
);
721 r
= blkid_probe_set_sectorsize(b
, m
->sector_size
);
723 return errno_or_else(EIO
);
725 if ((flags
& DISSECT_IMAGE_GPT_ONLY
) == 0) {
726 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
727 blkid_probe_enable_superblocks(b
, 1);
728 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
|BLKID_SUBLKS_USAGE
|BLKID_SUBLKS_UUID
);
731 blkid_probe_enable_partitions(b
, 1);
732 blkid_probe_set_partitions_flags(b
, BLKID_PARTS_ENTRY_DETAILS
);
735 r
= blkid_do_safeprobe(b
);
736 if (r
== _BLKID_SAFEPROBE_ERROR
)
737 return errno_or_else(EIO
);
738 if (IN_SET(r
, _BLKID_SAFEPROBE_AMBIGUOUS
, _BLKID_SAFEPROBE_NOT_FOUND
))
739 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG
), "Failed to identify any partition table.");
741 assert(r
== _BLKID_SAFEPROBE_FOUND
);
743 if ((!(flags
& DISSECT_IMAGE_GPT_ONLY
) &&
744 (flags
& DISSECT_IMAGE_GENERIC_ROOT
)) ||
745 (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)) {
746 const char *usage
= NULL
;
748 /* If flags permit this, also allow using non-partitioned single-filesystem images */
750 (void) blkid_probe_lookup_value(b
, "USAGE", &usage
, NULL
);
751 if (STRPTR_IN_SET(usage
, "filesystem", "crypto")) {
752 _cleanup_free_
char *t
= NULL
, *n
= NULL
, *o
= NULL
;
753 const char *fstype
= NULL
, *options
= NULL
, *suuid
= NULL
;
754 _cleanup_close_
int mount_node_fd
= -EBADF
;
755 sd_id128_t uuid
= SD_ID128_NULL
;
756 PartitionPolicyFlags found_flags
;
759 /* OK, we have found a file system, that's our root partition then. */
761 r
= image_policy_may_use(policy
, PARTITION_ROOT
);
764 if (r
== 0) /* policy says ignore this, so we ignore it */
767 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
768 (void) blkid_probe_lookup_value(b
, "UUID", &suuid
, NULL
);
770 encrypted
= streq_ptr(fstype
, "crypto_LUKS");
772 if (verity_settings_data_covers(verity
, PARTITION_ROOT
))
773 found_flags
= verity
->root_hash_sig
? PARTITION_POLICY_SIGNED
: PARTITION_POLICY_VERITY
;
775 found_flags
= encrypted
? PARTITION_POLICY_ENCRYPTED
: PARTITION_POLICY_UNPROTECTED
;
777 r
= image_policy_check_protection(policy
, PARTITION_ROOT
, found_flags
);
781 r
= image_policy_check_partition_flags(policy
, PARTITION_ROOT
, 0); /* we have no gpt partition flags, hence check against all bits off */
785 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
786 mount_node_fd
= open_partition(devname
, /* is_partition = */ false, m
->loop
);
787 if (mount_node_fd
< 0)
788 return mount_node_fd
;
798 /* blkid will return FAT's serial number as UUID, hence it is quite possible
799 * that parsing this will fail. We'll ignore the ID, since it's just too
800 * short to be useful as tru identifier. */
801 r
= sd_id128_from_string(suuid
, &uuid
);
803 log_debug_errno(r
, "Failed to parse file system UUID '%s', ignoring: %m", suuid
);
806 r
= make_partition_devname(devname
, diskseq
, -1, flags
, &n
);
810 m
->single_file_system
= true;
811 m
->encrypted
= encrypted
;
813 m
->has_verity
= verity
&& verity
->data_path
;
814 m
->verity_ready
= verity_settings_data_covers(verity
, PARTITION_ROOT
);
816 m
->has_verity_sig
= false; /* signature not embedded, must be specified */
817 m
->verity_sig_ready
= m
->verity_ready
&& verity
->root_hash_sig
;
819 m
->image_uuid
= uuid
;
821 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
828 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
830 .rw
= !m
->verity_ready
&& !fstype_is_ro(fstype
),
832 .architecture
= _ARCHITECTURE_INVALID
,
833 .fstype
= TAKE_PTR(t
),
835 .mount_options
= TAKE_PTR(o
),
836 .mount_node_fd
= TAKE_FD(mount_node_fd
),
845 (void) blkid_probe_lookup_value(b
, "PTTYPE", &pttype
, NULL
);
849 is_gpt
= streq_ptr(pttype
, "gpt");
850 is_mbr
= streq_ptr(pttype
, "dos");
852 if (!is_gpt
&& ((flags
& DISSECT_IMAGE_GPT_ONLY
) || !is_mbr
))
855 /* We support external verity data partitions only if the image has no partition table */
856 if (verity
&& verity
->data_path
)
859 if (FLAGS_SET(flags
, DISSECT_IMAGE_ADD_PARTITION_DEVICES
)) {
860 /* Safety check: refuse block devices that carry a partition table but for which the kernel doesn't
861 * do partition scanning. */
862 r
= blockdev_partscan_enabled(fd
);
866 return -EPROTONOSUPPORT
;
869 (void) blkid_probe_lookup_value(b
, "PTUUID", &sptuuid
, NULL
);
871 r
= sd_id128_from_string(sptuuid
, &m
->image_uuid
);
873 log_debug_errno(r
, "Failed to parse partition table UUID '%s', ignoring: %m", sptuuid
);
877 pl
= blkid_probe_get_partitions(b
);
879 return errno_or_else(ENOMEM
);
882 n_partitions
= blkid_partlist_numof_partitions(pl
);
883 if (n_partitions
< 0)
884 return errno_or_else(EIO
);
886 for (int i
= 0; i
< n_partitions
; i
++) {
887 _cleanup_free_
char *node
= NULL
;
888 unsigned long long pflags
;
889 blkid_loff_t start
, size
;
894 pp
= blkid_partlist_get_partition(pl
, i
);
896 return errno_or_else(EIO
);
898 pflags
= blkid_partition_get_flags(pp
);
901 nr
= blkid_partition_get_partno(pp
);
903 return errno_or_else(EIO
);
906 start
= blkid_partition_get_start(pp
);
908 return errno_or_else(EIO
);
910 assert((uint64_t) start
< UINT64_MAX
/512);
913 size
= blkid_partition_get_size(pp
);
915 return errno_or_else(EIO
);
917 assert((uint64_t) size
< UINT64_MAX
/512);
919 /* While probing we need the non-diskseq device node name to access the thing, hence mask off
920 * DISSECT_IMAGE_DISKSEQ_DEVNODE. */
921 r
= make_partition_devname(devname
, diskseq
, nr
, flags
& ~DISSECT_IMAGE_DISKSEQ_DEVNODE
, &node
);
925 /* So here's the thing: after the main ("whole") block device popped up it might take a while
926 * before the kernel fully probed the partition table. Waiting for that to finish is icky in
927 * userspace. So here's what we do instead. We issue the BLKPG_ADD_PARTITION ioctl to add the
928 * partition ourselves, racing against the kernel. Good thing is: if this call fails with
929 * EBUSY then the kernel was quicker than us, and that's totally OK, the outcome is good for
930 * us: the device node will exist. If OTOH our call was successful we won the race. Which is
931 * also good as the outcome is the same: the partition block device exists, and we can use
934 * Kernel returns EBUSY if there's already a partition by that number or an overlapping
935 * partition already existent. */
937 if (FLAGS_SET(flags
, DISSECT_IMAGE_ADD_PARTITION_DEVICES
)) {
938 r
= block_device_add_partition(fd
, node
, nr
, (uint64_t) start
* 512, (uint64_t) size
* 512);
941 return log_debug_errno(r
, "BLKPG_ADD_PARTITION failed: %m");
943 log_debug_errno(r
, "Kernel was quicker than us in adding partition %i.", nr
);
945 log_debug("We were quicker than kernel in adding partition %i.", nr
);
949 const char *fstype
= NULL
, *label
;
950 sd_id128_t type_id
, id
;
951 GptPartitionType type
;
952 bool rw
= true, growfs
= false;
954 r
= blkid_partition_get_uuid_id128(pp
, &id
);
956 log_debug_errno(r
, "Failed to read partition UUID, ignoring: %m");
960 r
= blkid_partition_get_type_id128(pp
, &type_id
);
962 log_debug_errno(r
, "Failed to read partition type UUID, ignoring: %m");
966 type
= gpt_partition_type_from_uuid(type_id
);
968 label
= blkid_partition_get_name(pp
); /* libblkid returns NULL here if empty */
970 if (IN_SET(type
.designator
,
976 check_partition_flags(node
, pflags
,
977 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
979 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
982 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
983 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
985 } else if (type
.designator
== PARTITION_ESP
) {
987 /* Note that we don't check the SD_GPT_FLAG_NO_AUTO flag for the ESP, as it is
988 * not defined there. We instead check the SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as
989 * recommended by the UEFI spec (See "12.3.3 Number and Location of System
992 if (pflags
& SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL
)
997 } else if (type
.designator
== PARTITION_ROOT
) {
999 check_partition_flags(node
, pflags
,
1000 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1002 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1005 /* If a root ID is specified, ignore everything but the root id */
1006 if (!sd_id128_is_null(root_uuid
) && !sd_id128_equal(root_uuid
, id
))
1009 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1010 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1012 } else if (type
.designator
== PARTITION_ROOT_VERITY
) {
1014 check_partition_flags(node
, pflags
,
1015 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1017 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1020 m
->has_verity
= true;
1022 /* If no verity configuration is specified, then don't do verity */
1025 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
1028 /* If root hash is specified, then ignore everything but the root id */
1029 if (!sd_id128_is_null(root_verity_uuid
) && !sd_id128_equal(root_verity_uuid
, id
))
1032 fstype
= "DM_verity_hash";
1035 } else if (type
.designator
== PARTITION_ROOT_VERITY_SIG
) {
1037 check_partition_flags(node
, pflags
,
1038 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1040 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1043 m
->has_verity_sig
= true;
1047 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
1050 fstype
= "verity_hash_signature";
1053 } else if (type
.designator
== PARTITION_USR
) {
1055 check_partition_flags(node
, pflags
,
1056 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1058 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1061 /* If a usr ID is specified, ignore everything but the usr id */
1062 if (!sd_id128_is_null(usr_uuid
) && !sd_id128_equal(usr_uuid
, id
))
1065 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1066 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1068 } else if (type
.designator
== PARTITION_USR_VERITY
) {
1070 check_partition_flags(node
, pflags
,
1071 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1073 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1076 m
->has_verity
= true;
1080 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
1083 /* If usr hash is specified, then ignore everything but the usr id */
1084 if (!sd_id128_is_null(usr_verity_uuid
) && !sd_id128_equal(usr_verity_uuid
, id
))
1087 fstype
= "DM_verity_hash";
1090 } else if (type
.designator
== PARTITION_USR_VERITY_SIG
) {
1092 check_partition_flags(node
, pflags
,
1093 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1095 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1098 m
->has_verity_sig
= true;
1102 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
1105 fstype
= "verity_hash_signature";
1108 } else if (type
.designator
== PARTITION_SWAP
) {
1110 check_partition_flags(node
, pflags
, SD_GPT_FLAG_NO_AUTO
);
1112 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1115 /* Note: we don't set fstype = "swap" here, because we still need to probe if
1116 * it might be encrypted (i.e. fstype "crypt_LUKS") or unencrypted
1117 * (i.e. fstype "swap"), and the only way to figure that out is via fstype
1120 /* We don't have a designator for SD_GPT_LINUX_GENERIC so check the UUID instead. */
1121 } else if (sd_id128_equal(type
.uuid
, SD_GPT_LINUX_GENERIC
)) {
1123 check_partition_flags(node
, pflags
,
1124 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1126 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1130 multiple_generic
= true;
1133 generic_rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1134 generic_growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1136 generic_node
= TAKE_PTR(node
);
1139 } else if (type
.designator
== PARTITION_VAR
) {
1141 check_partition_flags(node
, pflags
,
1142 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1144 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1147 if (!FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_VAR_CHECK
)) {
1148 sd_id128_t var_uuid
;
1150 /* For /var we insist that the uuid of the partition matches the
1151 * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
1152 * ID. Why? Unlike the other partitions /var is inherently
1153 * installation specific, hence we need to be careful not to mount it
1154 * in the wrong installation. By hashing the partition UUID from
1155 * /etc/machine-id we can securely bind the partition to the
1158 r
= sd_id128_get_machine_app_specific(SD_GPT_VAR
, &var_uuid
);
1162 if (!sd_id128_equal(var_uuid
, id
)) {
1163 log_debug("Found a /var/ partition, but its UUID didn't match our expectations "
1164 "(found: " SD_ID128_UUID_FORMAT_STR
", expected: " SD_ID128_UUID_FORMAT_STR
"), ignoring.",
1165 SD_ID128_FORMAT_VAL(id
), SD_ID128_FORMAT_VAL(var_uuid
));
1170 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1171 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1174 if (type
.designator
!= _PARTITION_DESIGNATOR_INVALID
) {
1175 _cleanup_free_
char *t
= NULL
, *o
= NULL
, *l
= NULL
, *n
= NULL
;
1176 _cleanup_close_
int mount_node_fd
= -EBADF
;
1177 const char *options
= NULL
;
1179 r
= image_policy_may_use(policy
, type
.designator
);
1183 /* Policy says: ignore; Remember this fact, so that we later can distinguish between "found but ignored" and "not found at all" */
1185 if (!m
->partitions
[type
.designator
].found
)
1186 m
->partitions
[type
.designator
].ignored
= true;
1191 if (m
->partitions
[type
.designator
].found
) {
1194 /* For most partition types the first one we see wins. Except for the
1195 * rootfs and /usr, where we do a version compare of the label, and
1196 * let the newest version win. This permits a simple A/B versioning
1197 * scheme in OS images. */
1199 c
= compare_arch(type
.arch
, m
->partitions
[type
.designator
].architecture
);
1200 if (c
< 0) /* the arch we already found is better than the one we found now */
1202 if (c
== 0 && /* same arch? then go by version in label */
1203 (!partition_designator_is_versioned(type
.designator
) ||
1204 strverscmp_improved(label
, m
->partitions
[type
.designator
].label
) <= 0))
1207 dissected_partition_done(m
->partitions
+ type
.designator
);
1210 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
) &&
1211 type
.designator
!= PARTITION_SWAP
) {
1212 mount_node_fd
= open_partition(node
, /* is_partition = */ true, m
->loop
);
1213 if (mount_node_fd
< 0)
1214 return mount_node_fd
;
1217 r
= make_partition_devname(devname
, diskseq
, nr
, flags
, &n
);
1233 options
= mount_options_from_designator(mount_options
, type
.designator
);
1235 o
= strdup(options
);
1240 m
->partitions
[type
.designator
] = (DissectedPartition
) {
1245 .architecture
= type
.arch
,
1246 .node
= TAKE_PTR(n
),
1247 .fstype
= TAKE_PTR(t
),
1248 .label
= TAKE_PTR(l
),
1250 .mount_options
= TAKE_PTR(o
),
1251 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1252 .offset
= (uint64_t) start
* 512,
1253 .size
= (uint64_t) size
* 512,
1254 .gpt_flags
= pflags
,
1258 } else if (is_mbr
) {
1260 switch (blkid_partition_get_type(pp
)) {
1262 case 0x83: /* Linux partition */
1264 if (pflags
!= 0x80) /* Bootable flag */
1268 multiple_generic
= true;
1272 generic_growfs
= false;
1273 generic_node
= TAKE_PTR(node
);
1278 case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
1279 _cleanup_close_
int mount_node_fd
= -EBADF
;
1280 _cleanup_free_
char *o
= NULL
, *n
= NULL
;
1281 sd_id128_t id
= SD_ID128_NULL
;
1282 const char *options
= NULL
;
1284 r
= image_policy_may_use(policy
, PARTITION_XBOOTLDR
);
1287 if (r
== 0) { /* policy says: ignore */
1288 if (!m
->partitions
[PARTITION_XBOOTLDR
].found
)
1289 m
->partitions
[PARTITION_XBOOTLDR
].ignored
= true;
1294 /* First one wins */
1295 if (m
->partitions
[PARTITION_XBOOTLDR
].found
)
1298 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
1299 mount_node_fd
= open_partition(node
, /* is_partition = */ true, m
->loop
);
1300 if (mount_node_fd
< 0)
1301 return mount_node_fd
;
1304 (void) blkid_partition_get_uuid_id128(pp
, &id
);
1306 r
= make_partition_devname(devname
, diskseq
, nr
, flags
, &n
);
1310 options
= mount_options_from_designator(mount_options
, PARTITION_XBOOTLDR
);
1312 o
= strdup(options
);
1317 m
->partitions
[PARTITION_XBOOTLDR
] = (DissectedPartition
) {
1322 .architecture
= _ARCHITECTURE_INVALID
,
1323 .node
= TAKE_PTR(n
),
1325 .mount_options
= TAKE_PTR(o
),
1326 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1327 .offset
= (uint64_t) start
* 512,
1328 .size
= (uint64_t) size
* 512,
1336 if (!m
->partitions
[PARTITION_ROOT
].found
&&
1337 (m
->partitions
[PARTITION_ROOT_VERITY
].found
||
1338 m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
))
1339 return -EADDRNOTAVAIL
; /* Verity found but no matching rootfs? Something is off, refuse. */
1341 /* Hmm, we found a signature partition but no Verity data? Something is off. */
1342 if (m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_ROOT_VERITY
].found
)
1343 return -EADDRNOTAVAIL
;
1345 if (!m
->partitions
[PARTITION_USR
].found
&&
1346 (m
->partitions
[PARTITION_USR_VERITY
].found
||
1347 m
->partitions
[PARTITION_USR_VERITY_SIG
].found
))
1348 return -EADDRNOTAVAIL
; /* as above */
1351 if (m
->partitions
[PARTITION_USR_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_USR_VERITY
].found
)
1352 return -EADDRNOTAVAIL
;
1354 /* If root and /usr are combined then insist that the architecture matches */
1355 if (m
->partitions
[PARTITION_ROOT
].found
&&
1356 m
->partitions
[PARTITION_USR
].found
&&
1357 (m
->partitions
[PARTITION_ROOT
].architecture
>= 0 &&
1358 m
->partitions
[PARTITION_USR
].architecture
>= 0 &&
1359 m
->partitions
[PARTITION_ROOT
].architecture
!= m
->partitions
[PARTITION_USR
].architecture
))
1360 return -EADDRNOTAVAIL
;
1362 if (!m
->partitions
[PARTITION_ROOT
].found
&&
1363 !m
->partitions
[PARTITION_USR
].found
&&
1364 (flags
& DISSECT_IMAGE_GENERIC_ROOT
) &&
1365 (!verity
|| !verity
->root_hash
|| verity
->designator
!= PARTITION_USR
)) {
1367 /* OK, we found nothing usable, then check if there's a single generic partition, and use
1368 * that. If the root hash was set however, then we won't fall back to a generic node, because
1369 * the root hash decides. */
1371 /* If we didn't find a properly marked root partition, but we did find a single suitable
1372 * generic Linux partition, then use this as root partition, if the caller asked for it. */
1373 if (multiple_generic
)
1376 /* If we didn't find a generic node, then we can't fix this up either */
1378 r
= image_policy_may_use(policy
, PARTITION_ROOT
);
1382 /* Policy says: ignore; remember that we did */
1383 m
->partitions
[PARTITION_ROOT
].ignored
= true;
1385 _cleanup_close_
int mount_node_fd
= -EBADF
;
1386 _cleanup_free_
char *o
= NULL
, *n
= NULL
;
1387 const char *options
;
1389 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
1390 mount_node_fd
= open_partition(generic_node
, /* is_partition = */ true, m
->loop
);
1391 if (mount_node_fd
< 0)
1392 return mount_node_fd
;
1395 r
= make_partition_devname(devname
, diskseq
, generic_nr
, flags
, &n
);
1399 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
1401 o
= strdup(options
);
1406 assert(generic_nr
>= 0);
1407 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
1410 .growfs
= generic_growfs
,
1411 .partno
= generic_nr
,
1412 .architecture
= _ARCHITECTURE_INVALID
,
1413 .node
= TAKE_PTR(n
),
1414 .uuid
= generic_uuid
,
1415 .mount_options
= TAKE_PTR(o
),
1416 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1417 .offset
= UINT64_MAX
,
1424 /* 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 */
1425 if (FLAGS_SET(flags
, DISSECT_IMAGE_REQUIRE_ROOT
) &&
1426 !(m
->partitions
[PARTITION_ROOT
].found
|| (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
1429 if (m
->partitions
[PARTITION_ROOT_VERITY
].found
) {
1430 /* We only support one verity partition per image, i.e. can't do for both /usr and root fs */
1431 if (m
->partitions
[PARTITION_USR_VERITY
].found
)
1434 /* We don't support verity enabled root with a split out /usr. Neither with nor without
1435 * verity there. (Note that we do support verity-less root with verity-full /usr, though.) */
1436 if (m
->partitions
[PARTITION_USR
].found
)
1437 return -EADDRNOTAVAIL
;
1441 /* If a verity designator is specified, then insist that the matching partition exists */
1442 if (verity
->designator
>= 0 && !m
->partitions
[verity
->designator
].found
)
1443 return -EADDRNOTAVAIL
;
1445 bool have_verity_sig_partition
=
1446 m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR_VERITY_SIG
: PARTITION_ROOT_VERITY_SIG
].found
;
1448 if (verity
->root_hash
) {
1449 /* If we have an explicit root hash and found the partitions for it, then we are ready to use
1450 * Verity, set things up for it */
1452 if (verity
->designator
< 0 || verity
->designator
== PARTITION_ROOT
) {
1453 if (!m
->partitions
[PARTITION_ROOT_VERITY
].found
|| !m
->partitions
[PARTITION_ROOT
].found
)
1454 return -EADDRNOTAVAIL
;
1456 /* If we found a verity setup, then the root partition is necessarily read-only. */
1457 m
->partitions
[PARTITION_ROOT
].rw
= false;
1458 m
->verity_ready
= true;
1461 assert(verity
->designator
== PARTITION_USR
);
1463 if (!m
->partitions
[PARTITION_USR_VERITY
].found
|| !m
->partitions
[PARTITION_USR
].found
)
1464 return -EADDRNOTAVAIL
;
1466 m
->partitions
[PARTITION_USR
].rw
= false;
1467 m
->verity_ready
= true;
1470 if (m
->verity_ready
)
1471 m
->verity_sig_ready
= verity
->root_hash_sig
|| have_verity_sig_partition
;
1473 } else if (have_verity_sig_partition
) {
1475 /* If we found an embedded signature partition, we are ready, too. */
1477 m
->verity_ready
= m
->verity_sig_ready
= true;
1478 m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR
: PARTITION_ROOT
].rw
= false;
1484 /* After we discovered all partitions let's see if the verity requirements match the policy. (Note:
1485 * we don't check encryption requirements here, because we haven't probed the file system yet, hence
1486 * don't know if this is encrypted or not) */
1487 for (PartitionDesignator di
= 0; di
< _PARTITION_DESIGNATOR_MAX
; di
++) {
1488 PartitionDesignator vi
, si
;
1489 PartitionPolicyFlags found_flags
;
1491 any
= any
|| m
->partitions
[di
].found
;
1493 vi
= partition_verity_of(di
);
1494 si
= partition_verity_sig_of(di
);
1496 /* Determine the verity protection level for this partition. */
1497 found_flags
= m
->partitions
[di
].found
?
1498 (vi
>= 0 && m
->partitions
[vi
].found
?
1499 (si
>= 0 && m
->partitions
[si
].found
? PARTITION_POLICY_SIGNED
: PARTITION_POLICY_VERITY
) :
1500 PARTITION_POLICY_ENCRYPTED
|PARTITION_POLICY_UNPROTECTED
) :
1501 (m
->partitions
[di
].ignored
? PARTITION_POLICY_UNUSED
: PARTITION_POLICY_ABSENT
);
1503 r
= image_policy_check_protection(policy
, di
, found_flags
);
1507 if (m
->partitions
[di
].found
) {
1508 r
= image_policy_check_partition_flags(policy
, di
, m
->partitions
[di
].gpt_flags
);
1514 if (!any
&& !FLAGS_SET(flags
, DISSECT_IMAGE_ALLOW_EMPTY
))
1517 r
= dissected_image_probe_filesystems(m
, fd
, policy
);
1525 int dissect_image_file(
1527 const VeritySettings
*verity
,
1528 const MountOptions
*mount_options
,
1529 const ImagePolicy
*image_policy
,
1530 DissectImageFlags flags
,
1531 DissectedImage
**ret
) {
1534 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
1535 _cleanup_close_
int fd
= -EBADF
;
1540 fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
1544 r
= fd_verify_regular(fd
);
1548 r
= dissected_image_new(path
, &m
);
1552 r
= probe_sector_size(fd
, &m
->sector_size
);
1556 r
= dissect_image(m
, fd
, path
, verity
, mount_options
, image_policy
, flags
);
1568 int dissect_log_error(int log_level
, int r
, const char *name
, const VeritySettings
*verity
) {
1569 assert(log_level
>= 0 && log_level
<= LOG_DEBUG
);
1574 case 0 ... INT_MAX
: /* success! */
1578 return log_full_errno(log_level
, r
, "Dissecting images is not supported, compiled without blkid support.");
1581 return log_full_errno(log_level
, r
, "%s: Couldn't identify a suitable partition table or file system.", name
);
1584 return log_full_errno(log_level
, r
, "%s: The image does not pass os-release/extension-release validation.", name
);
1586 case -EADDRNOTAVAIL
:
1587 return log_full_errno(log_level
, r
, "%s: No root partition for specified root hash found.", name
);
1590 return log_full_errno(log_level
, r
, "%s: Multiple suitable root partitions found in image.", name
);
1593 return log_full_errno(log_level
, r
, "%s: No suitable root partition found in image.", name
);
1595 case -EPROTONOSUPPORT
:
1596 return log_full_errno(log_level
, r
, "Device '%s' is a loopback block device with partition scanning turned off, please turn it on.", name
);
1599 return log_full_errno(log_level
, r
, "%s: Image is not a block device.", name
);
1602 return log_full_errno(log_level
, r
,
1603 "Combining partitioned images (such as '%s') with external Verity data (such as '%s') not supported. "
1604 "(Consider setting $SYSTEMD_DISSECT_VERITY_SIDECAR=0 to disable automatic discovery of external Verity data.)",
1605 name
, strna(verity
? verity
->data_path
: NULL
));
1608 return log_full_errno(log_level
, r
, "%s: image does not match image policy.", name
);
1611 return log_full_errno(log_level
, r
, "%s: no suitable partitions found.", name
);
1614 return log_full_errno(log_level
, r
, "%s: cannot dissect image: %m", name
);
1618 int dissect_image_file_and_warn(
1620 const VeritySettings
*verity
,
1621 const MountOptions
*mount_options
,
1622 const ImagePolicy
*image_policy
,
1623 DissectImageFlags flags
,
1624 DissectedImage
**ret
) {
1626 return dissect_log_error(
1628 dissect_image_file(path
, verity
, mount_options
, image_policy
, flags
, ret
),
1633 DissectedImage
* dissected_image_unref(DissectedImage
*m
) {
1637 /* First, clear dissected partitions. */
1638 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++)
1639 dissected_partition_done(m
->partitions
+ i
);
1641 /* Second, free decrypted images. This must be after dissected_partition_done(), as freeing
1642 * DecryptedImage may try to deactivate partitions. */
1643 decrypted_image_unref(m
->decrypted_image
);
1645 /* Third, unref LoopDevice. This must be called after the above two, as freeing LoopDevice may try to
1646 * remove existing partitions on the loopback block device. */
1647 loop_device_unref(m
->loop
);
1649 free(m
->image_name
);
1651 strv_free(m
->machine_info
);
1652 strv_free(m
->os_release
);
1653 strv_free(m
->initrd_release
);
1654 strv_free(m
->extension_release
);
1659 static int is_loop_device(const char *path
) {
1660 char s
[SYS_BLOCK_PATH_MAX("/../loop/")];
1665 if (stat(path
, &st
) < 0)
1668 if (!S_ISBLK(st
.st_mode
))
1671 xsprintf_sys_block_path(s
, "/loop/", st
.st_dev
);
1672 if (access(s
, F_OK
) < 0) {
1673 if (errno
!= ENOENT
)
1676 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
1677 xsprintf_sys_block_path(s
, "/../loop/", st
.st_dev
);
1678 if (access(s
, F_OK
) < 0)
1679 return errno
== ENOENT
? false : -errno
;
1685 static int run_fsck(int node_fd
, const char *fstype
) {
1689 assert(node_fd
>= 0);
1692 r
= fsck_exists_for_fstype(fstype
);
1694 log_debug_errno(r
, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype
);
1698 log_debug("Not checking partition %s, as fsck for %s does not exist.", FORMAT_PROC_FD_PATH(node_fd
), fstype
);
1705 &node_fd
, 1, /* Leave the node fd open */
1706 FORK_RESET_SIGNALS
|FORK_CLOSE_ALL_FDS
|FORK_RLIMIT_NOFILE_SAFE
|FORK_DEATHSIG
|FORK_REARRANGE_STDIO
|FORK_CLOEXEC_OFF
,
1709 return log_debug_errno(r
, "Failed to fork off fsck: %m");
1712 execlp("fsck", "fsck", "-aT", FORMAT_PROC_FD_PATH(node_fd
), NULL
);
1714 log_debug_errno(errno
, "Failed to execl() fsck: %m");
1715 _exit(FSCK_OPERATIONAL_ERROR
);
1718 exit_status
= wait_for_terminate_and_check("fsck", pid
, 0);
1719 if (exit_status
< 0)
1720 return log_debug_errno(exit_status
, "Failed to fork off fsck: %m");
1722 if ((exit_status
& ~FSCK_ERROR_CORRECTED
) != FSCK_SUCCESS
) {
1723 log_debug("fsck failed with exit status %i.", exit_status
);
1725 if ((exit_status
& (FSCK_SYSTEM_SHOULD_REBOOT
|FSCK_ERRORS_LEFT_UNCORRECTED
)) != 0)
1726 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
), "File system is corrupted, refusing.");
1728 log_debug("Ignoring fsck error.");
1734 static int fs_grow(const char *node_path
, const char *mount_path
) {
1735 _cleanup_close_
int mount_fd
= -EBADF
, node_fd
= -EBADF
;
1736 uint64_t size
, newsize
;
1739 node_fd
= open(node_path
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
1741 return log_debug_errno(errno
, "Failed to open node device %s: %m", node_path
);
1743 if (ioctl(node_fd
, BLKGETSIZE64
, &size
) != 0)
1744 return log_debug_errno(errno
, "Failed to get block device size of %s: %m", node_path
);
1746 mount_fd
= open(mount_path
, O_RDONLY
|O_DIRECTORY
|O_CLOEXEC
);
1748 return log_debug_errno(errno
, "Failed to open mountd file system %s: %m", mount_path
);
1750 log_debug("Resizing \"%s\" to %"PRIu64
" bytes...", mount_path
, size
);
1751 r
= resize_fs(mount_fd
, size
, &newsize
);
1753 return log_debug_errno(r
, "Failed to resize \"%s\" to %"PRIu64
" bytes: %m", mount_path
, size
);
1755 if (newsize
== size
)
1756 log_debug("Successfully resized \"%s\" to %s bytes.",
1757 mount_path
, FORMAT_BYTES(newsize
));
1759 assert(newsize
< size
);
1760 log_debug("Successfully resized \"%s\" to %s bytes (%"PRIu64
" bytes lost due to blocksize).",
1761 mount_path
, FORMAT_BYTES(newsize
), size
- newsize
);
1767 int partition_pick_mount_options(
1768 PartitionDesignator d
,
1773 unsigned long *ret_ms_flags
) {
1775 _cleanup_free_
char *options
= NULL
;
1777 assert(ret_options
);
1779 /* Selects a baseline of bind mount flags, that should always apply.
1781 * Firstly, we set MS_NODEV universally on all mounts, since we don't want to allow device nodes outside of /dev/.
1783 * On /var/tmp/ we'll also set MS_NOSUID, same as we set for /tmp/ on the host.
1785 * On the ESP and XBOOTLDR partitions we'll also disable symlinks, and execution. These file systems
1786 * are generally untrusted (i.e. not encrypted or authenticated), and typically VFAT hence we should
1787 * be as restrictive as possible, and this shouldn't hurt, since the functionality is not available
1790 unsigned long flags
= MS_NODEV
;
1798 case PARTITION_XBOOTLDR
:
1799 flags
|= MS_NOSUID
|MS_NOEXEC
|ms_nosymfollow_supported();
1801 /* The ESP might contain a pre-boot random seed. Let's make this unaccessible to regular
1802 * userspace. ESP/XBOOTLDR is almost certainly VFAT, hence if we don't know assume it is. */
1803 if (!fstype
|| fstype_can_umask(fstype
))
1804 if (!strextend_with_separator(&options
, ",", "umask=0077"))
1816 /* So, when you request MS_RDONLY from ext4, then this means nothing. It happily still writes to the
1817 * backing storage. What's worse, the BLKRO[GS]ET flag and (in case of loopback devices)
1818 * LO_FLAGS_READ_ONLY don't mean anything, they affect userspace accesses only, and write accesses
1819 * from the upper file system still get propagated through to the underlying file system,
1820 * unrestricted. To actually get ext4/xfs/btrfs to stop writing to the device we need to specify
1821 * "norecovery" as mount option, in addition to MS_RDONLY. Yes, this sucks, since it means we need to
1822 * carry a per file system table here.
1824 * Note that this means that we might not be able to mount corrupted file systems as read-only
1825 * anymore (since in some cases the kernel implementations will refuse mounting when corrupted,
1826 * read-only and "norecovery" is specified). But I think for the case of automatically determined
1827 * mount options for loopback devices this is the right choice, since otherwise using the same
1828 * loopback file twice even in read-only mode, is going to fail badly sooner or later. The use case of
1829 * making reuse of the immutable images "just work" is more relevant to us than having read-only
1830 * access that actually modifies stuff work on such image files. Or to say this differently: if
1831 * people want their file systems to be fixed up they should just open them in writable mode, where
1832 * all these problems don't exist. */
1833 if (!rw
&& fstype
&& fstype_can_norecovery(fstype
))
1834 if (!strextend_with_separator(&options
, ",", "norecovery"))
1837 if (discard
&& fstype
&& fstype_can_discard(fstype
))
1838 if (!strextend_with_separator(&options
, ",", "discard"))
1841 if (!ret_ms_flags
) /* Fold flags into option string if ret_flags specified as NULL */
1842 if (!strextend_with_separator(&options
, ",",
1843 FLAGS_SET(flags
, MS_RDONLY
) ? "ro" : "rw",
1844 FLAGS_SET(flags
, MS_NODEV
) ? "nodev" : "dev",
1845 FLAGS_SET(flags
, MS_NOSUID
) ? "nosuid" : "suid",
1846 FLAGS_SET(flags
, MS_NOEXEC
) ? "noexec" : "exec",
1847 FLAGS_SET(flags
, MS_NOSYMFOLLOW
) ? "nosymfollow" : NULL
))
1848 /* NB: we suppress 'symfollow' here, since it's the default, and old /bin/mount might not know it */
1852 *ret_ms_flags
= flags
;
1854 *ret_options
= TAKE_PTR(options
);
1858 static int mount_partition(
1859 PartitionDesignator d
,
1860 DissectedPartition
*m
,
1862 const char *directory
,
1865 DissectImageFlags flags
) {
1867 _cleanup_free_
char *chased
= NULL
, *options
= NULL
;
1868 bool rw
, discard
, remap_uid_gid
= false;
1869 const char *p
, *node
, *fstype
;
1870 unsigned long ms_flags
;
1876 if (m
->mount_node_fd
< 0)
1879 /* Use decrypted node and matching fstype if available, otherwise use the original device */
1880 node
= FORMAT_PROC_FD_PATH(m
->mount_node_fd
);
1881 fstype
= dissected_partition_fstype(m
);
1884 return -EAFNOSUPPORT
;
1886 /* We are looking at an encrypted partition? This either means stacked encryption, or the caller
1887 * didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error for this
1889 if (streq(fstype
, "crypto_LUKS"))
1892 r
= dissect_fstype_ok(fstype
);
1896 return -EIDRM
; /* Recognizable error */
1898 rw
= m
->rw
&& !(flags
& DISSECT_IMAGE_MOUNT_READ_ONLY
);
1900 discard
= ((flags
& DISSECT_IMAGE_DISCARD
) ||
1901 ((flags
& DISSECT_IMAGE_DISCARD_ON_LOOP
) && is_loop_device(m
->node
) > 0));
1903 if (FLAGS_SET(flags
, DISSECT_IMAGE_FSCK
) && rw
) {
1904 r
= run_fsck(m
->mount_node_fd
, fstype
);
1910 /* Automatically create missing mount points inside the image, if necessary. */
1911 r
= mkdir_p_root(where
, directory
, uid_shift
, (gid_t
) uid_shift
, 0755, NULL
);
1912 if (r
< 0 && r
!= -EROFS
)
1915 r
= chase(directory
, where
, CHASE_PREFIX_ROOT
, &chased
, NULL
);
1921 /* Create top-level mount if missing – but only if this is asked for. This won't modify the
1922 * image (as the branch above does) but the host hierarchy, and the created directory might
1923 * survive our mount in the host hierarchy hence. */
1924 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
1925 r
= mkdir_p(where
, 0755);
1933 r
= partition_pick_mount_options(d
, dissected_partition_fstype(m
), rw
, discard
, &options
, &ms_flags
);
1937 if (uid_is_valid(uid_shift
) && uid_shift
!= 0) {
1939 if (fstype_can_uid_gid(fstype
)) {
1940 _cleanup_free_
char *uid_option
= NULL
;
1942 if (asprintf(&uid_option
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
1945 if (!strextend_with_separator(&options
, ",", uid_option
))
1947 } else if (FLAGS_SET(flags
, DISSECT_IMAGE_MOUNT_IDMAPPED
))
1948 remap_uid_gid
= true;
1951 if (!isempty(m
->mount_options
))
1952 if (!strextend_with_separator(&options
, ",", m
->mount_options
))
1955 r
= mount_nofollow_verbose(LOG_DEBUG
, node
, p
, fstype
, ms_flags
, options
);
1959 if (rw
&& m
->growfs
&& FLAGS_SET(flags
, DISSECT_IMAGE_GROWFS
))
1960 (void) fs_grow(node
, p
);
1962 if (remap_uid_gid
) {
1963 r
= remount_idmap(p
, uid_shift
, uid_range
, UID_INVALID
, REMOUNT_IDMAPPING_HOST_ROOT
);
1971 static int mount_root_tmpfs(const char *where
, uid_t uid_shift
, DissectImageFlags flags
) {
1972 _cleanup_free_
char *options
= NULL
;
1977 /* For images that contain /usr/ but no rootfs, let's mount rootfs as tmpfs */
1979 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
1980 r
= mkdir_p(where
, 0755);
1985 if (uid_is_valid(uid_shift
)) {
1986 if (asprintf(&options
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
1990 r
= mount_nofollow_verbose(LOG_DEBUG
, "rootfs", where
, "tmpfs", MS_NODEV
, options
);
1997 static int mount_point_is_available(const char *where
, const char *path
, bool missing_ok
) {
1998 _cleanup_free_
char *p
= NULL
;
2001 /* Check whether <path> is suitable as a mountpoint, i.e. is an empty directory
2002 * or does not exist at all (when missing_ok). */
2004 r
= chase(path
, where
, CHASE_PREFIX_ROOT
, &p
, NULL
);
2008 return log_debug_errno(r
, "Failed to chase \"%s\": %m", path
);
2010 r
= dir_is_empty(p
, /* ignore_hidden_or_backup= */ false);
2014 return log_debug_errno(r
, "Failed to check directory \"%s\": %m", p
);
2018 int dissected_image_mount(
2023 DissectImageFlags flags
) {
2032 * -ENXIO → No root partition found
2033 * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release/extension-release file found
2034 * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet
2035 * -EUCLEAN → fsck for file system failed
2036 * -EBUSY → File system already mounted/used elsewhere (kernel)
2037 * -EAFNOSUPPORT → File system type not supported or not known
2038 * -EIDRM → File system is not among allowlisted "common" file systems
2041 if (!(m
->partitions
[PARTITION_ROOT
].found
||
2042 (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
2043 return -ENXIO
; /* Require a root fs or at least a /usr/ fs (the latter is subject to a flag of its own) */
2045 if ((flags
& DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
) == 0) {
2047 /* First mount the root fs. If there's none we use a tmpfs. */
2048 if (m
->partitions
[PARTITION_ROOT
].found
)
2049 r
= mount_partition(PARTITION_ROOT
, m
->partitions
+ PARTITION_ROOT
, where
, NULL
, uid_shift
, uid_range
, flags
);
2051 r
= mount_root_tmpfs(where
, uid_shift
, flags
);
2055 /* For us mounting root always means mounting /usr as well */
2056 r
= mount_partition(PARTITION_USR
, m
->partitions
+ PARTITION_USR
, where
, "/usr", uid_shift
, uid_range
, flags
);
2060 if ((flags
& (DISSECT_IMAGE_VALIDATE_OS
|DISSECT_IMAGE_VALIDATE_OS_EXT
)) != 0) {
2061 /* If either one of the validation flags are set, ensure that the image qualifies
2062 * as one or the other (or both). */
2065 if (FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS
)) {
2066 r
= path_is_os_tree(where
);
2072 if (!ok
&& FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS_EXT
)) {
2073 r
= extension_has_forbidden_content(where
);
2077 r
= path_is_extension_tree(IMAGE_SYSEXT
, where
, m
->image_name
, FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_EXTENSION_CHECK
));
2079 r
= path_is_extension_tree(IMAGE_CONFEXT
, where
, m
->image_name
, FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_EXTENSION_CHECK
));
2092 if (flags
& DISSECT_IMAGE_MOUNT_ROOT_ONLY
)
2095 r
= mount_partition(PARTITION_HOME
, m
->partitions
+ PARTITION_HOME
, where
, "/home", uid_shift
, uid_range
, flags
);
2099 r
= mount_partition(PARTITION_SRV
, m
->partitions
+ PARTITION_SRV
, where
, "/srv", uid_shift
, uid_range
, flags
);
2103 r
= mount_partition(PARTITION_VAR
, m
->partitions
+ PARTITION_VAR
, where
, "/var", uid_shift
, uid_range
, flags
);
2107 r
= mount_partition(PARTITION_TMP
, m
->partitions
+ PARTITION_TMP
, where
, "/var/tmp", uid_shift
, uid_range
, flags
);
2111 int slash_boot_is_available
;
2112 r
= slash_boot_is_available
= mount_point_is_available(where
, "/boot", /* missing_ok = */ true);
2116 r
= mount_partition(PARTITION_XBOOTLDR
, m
->partitions
+ PARTITION_XBOOTLDR
, where
, "/boot", uid_shift
, uid_range
, flags
);
2119 slash_boot_is_available
= !r
;
2122 if (m
->partitions
[PARTITION_ESP
].found
) {
2123 const char *esp_path
= NULL
;
2125 /* Mount the ESP to /boot/ if it exists and is empty and we didn't already mount the XBOOTLDR
2126 * partition into it. Otherwise, use /efi instead, but only if it exists and is empty. */
2128 if (slash_boot_is_available
) {
2129 r
= mount_point_is_available(where
, "/boot", /* missing_ok = */ false);
2137 r
= mount_point_is_available(where
, "/efi", /* missing_ok = */ true);
2145 /* OK, let's mount the ESP now (possibly creating the dir if missing) */
2146 r
= mount_partition(PARTITION_ESP
, m
->partitions
+ PARTITION_ESP
, where
, esp_path
, uid_shift
, uid_range
, flags
);
2155 int dissected_image_mount_and_warn(
2160 DissectImageFlags flags
) {
2167 r
= dissected_image_mount(m
, where
, uid_shift
, uid_range
, flags
);
2169 return log_error_errno(r
, "Not root file system found in image.");
2170 if (r
== -EMEDIUMTYPE
)
2171 return log_error_errno(r
, "No suitable os-release/extension-release file in image found.");
2173 return log_error_errno(r
, "Encrypted file system discovered, but decryption not requested.");
2175 return log_error_errno(r
, "File system check on image failed.");
2177 return log_error_errno(r
, "File system already mounted elsewhere.");
2178 if (r
== -EAFNOSUPPORT
)
2179 return log_error_errno(r
, "File system type not supported or not known.");
2181 return log_error_errno(r
, "File system is too uncommon, refused.");
2183 return log_error_errno(r
, "Failed to mount image: %m");
2188 #if HAVE_LIBCRYPTSETUP
2189 struct DecryptedPartition
{
2190 struct crypt_device
*device
;
2196 typedef struct DecryptedPartition DecryptedPartition
;
2198 struct DecryptedImage
{
2200 DecryptedPartition
*decrypted
;
2204 static DecryptedImage
* decrypted_image_free(DecryptedImage
*d
) {
2205 #if HAVE_LIBCRYPTSETUP
2211 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
2212 DecryptedPartition
*p
= d
->decrypted
+ i
;
2214 if (p
->device
&& p
->name
&& !p
->relinquished
) {
2215 _cleanup_free_
char *node
= NULL
;
2217 node
= path_join("/dev/mapper", p
->name
);
2219 r
= btrfs_forget_device(node
);
2220 if (r
< 0 && r
!= -ENOENT
)
2221 log_debug_errno(r
, "Failed to forget btrfs device %s, ignoring: %m", node
);
2225 /* Let's deactivate lazily, as the dm volume may be already/still used by other processes. */
2226 r
= sym_crypt_deactivate_by_name(p
->device
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
2228 log_debug_errno(r
, "Failed to deactivate encrypted partition %s", p
->name
);
2232 sym_crypt_free(p
->device
);
2242 DEFINE_TRIVIAL_REF_UNREF_FUNC(DecryptedImage
, decrypted_image
, decrypted_image_free
);
2244 #if HAVE_LIBCRYPTSETUP
2245 static int decrypted_image_new(DecryptedImage
**ret
) {
2246 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
2250 d
= new(DecryptedImage
, 1);
2254 *d
= (DecryptedImage
) {
2262 static int make_dm_name_and_node(const void *original_node
, const char *suffix
, char **ret_name
, char **ret_node
) {
2263 _cleanup_free_
char *name
= NULL
, *node
= NULL
;
2266 assert(original_node
);
2271 base
= strrchr(original_node
, '/');
2273 base
= original_node
;
2279 name
= strjoin(base
, suffix
);
2282 if (!filename_is_valid(name
))
2285 node
= path_join(sym_crypt_get_dir(), name
);
2289 *ret_name
= TAKE_PTR(name
);
2290 *ret_node
= TAKE_PTR(node
);
2295 static int decrypt_partition(
2296 DissectedPartition
*m
,
2297 const char *passphrase
,
2298 DissectImageFlags flags
,
2299 DecryptedImage
*d
) {
2301 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
2302 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2303 _cleanup_close_
int fd
= -EBADF
;
2309 if (!m
->found
|| !m
->node
|| !m
->fstype
)
2312 if (!streq(m
->fstype
, "crypto_LUKS"))
2318 r
= dlopen_cryptsetup();
2322 r
= make_dm_name_and_node(m
->node
, "-decrypted", &name
, &node
);
2326 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
2329 r
= sym_crypt_init(&cd
, m
->node
);
2331 return log_debug_errno(r
, "Failed to initialize dm-crypt: %m");
2333 cryptsetup_enable_logging(cd
);
2335 r
= sym_crypt_load(cd
, CRYPT_LUKS
, NULL
);
2337 return log_debug_errno(r
, "Failed to load LUKS metadata: %m");
2339 r
= sym_crypt_activate_by_passphrase(cd
, name
, CRYPT_ANY_SLOT
, passphrase
, strlen(passphrase
),
2340 ((flags
& DISSECT_IMAGE_DEVICE_READ_ONLY
) ? CRYPT_ACTIVATE_READONLY
: 0) |
2341 ((flags
& DISSECT_IMAGE_DISCARD_ON_CRYPTO
) ? CRYPT_ACTIVATE_ALLOW_DISCARDS
: 0));
2343 log_debug_errno(r
, "Failed to activate LUKS device: %m");
2344 return r
== -EPERM
? -EKEYREJECTED
: r
;
2347 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2349 return log_debug_errno(errno
, "Failed to open %s: %m", node
);
2351 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
2352 .name
= TAKE_PTR(name
),
2353 .device
= TAKE_PTR(cd
),
2356 m
->decrypted_node
= TAKE_PTR(node
);
2357 close_and_replace(m
->mount_node_fd
, fd
);
2362 static int verity_can_reuse(
2363 const VeritySettings
*verity
,
2365 struct crypt_device
**ret_cd
) {
2367 /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
2368 _cleanup_free_
char *root_hash_existing
= NULL
;
2369 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2370 struct crypt_params_verity crypt_params
= {};
2371 size_t root_hash_existing_size
;
2378 r
= sym_crypt_init_by_name(&cd
, name
);
2380 return log_debug_errno(r
, "Error opening verity device, crypt_init_by_name failed: %m");
2382 cryptsetup_enable_logging(cd
);
2384 r
= sym_crypt_get_verity_info(cd
, &crypt_params
);
2386 return log_debug_errno(r
, "Error opening verity device, crypt_get_verity_info failed: %m");
2388 root_hash_existing_size
= verity
->root_hash_size
;
2389 root_hash_existing
= malloc0(root_hash_existing_size
);
2390 if (!root_hash_existing
)
2393 r
= sym_crypt_volume_key_get(cd
, CRYPT_ANY_SLOT
, root_hash_existing
, &root_hash_existing_size
, NULL
, 0);
2395 return log_debug_errno(r
, "Error opening verity device, crypt_volume_key_get failed: %m");
2396 if (verity
->root_hash_size
!= root_hash_existing_size
||
2397 memcmp(root_hash_existing
, verity
->root_hash
, verity
->root_hash_size
) != 0)
2398 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but root hashes are different.");
2400 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
2401 /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the
2402 * same settings, so that a previous unsigned mount will not be reused if the user asks to use
2403 * signing for the new one, and vice versa. */
2404 if (!!verity
->root_hash_sig
!= !!(crypt_params
.flags
& CRYPT_VERITY_ROOT_HASH_SIGNATURE
))
2405 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but signature settings are not the same.");
2408 *ret_cd
= TAKE_PTR(cd
);
2412 static char* dm_deferred_remove_clean(char *name
) {
2416 (void) sym_crypt_deactivate_by_name(NULL
, name
, CRYPT_DEACTIVATE_DEFERRED
);
2419 DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean
);
2421 static int validate_signature_userspace(const VeritySettings
*verity
) {
2423 _cleanup_(sk_X509_free_allp
) STACK_OF(X509
) *sk
= NULL
;
2424 _cleanup_strv_free_
char **certs
= NULL
;
2425 _cleanup_(PKCS7_freep
) PKCS7
*p7
= NULL
;
2426 _cleanup_free_
char *s
= NULL
;
2427 _cleanup_(BIO_freep
) BIO
*bio
= NULL
; /* 'bio' must be freed first, 's' second, hence keep this order
2428 * of declaration in place, please */
2429 const unsigned char *d
;
2433 assert(verity
->root_hash
);
2434 assert(verity
->root_hash_sig
);
2436 /* Because installing a signature certificate into the kernel chain is so messy, let's optionally do
2437 * userspace validation. */
2439 r
= conf_files_list_nulstr(&certs
, ".crt", NULL
, CONF_FILES_REGULAR
|CONF_FILES_FILTER_MASKED
, CONF_PATHS_NULSTR("verity.d"));
2441 return log_debug_errno(r
, "Failed to enumerate certificates: %m");
2442 if (strv_isempty(certs
)) {
2443 log_debug("No userspace dm-verity certificates found.");
2447 d
= verity
->root_hash_sig
;
2448 p7
= d2i_PKCS7(NULL
, &d
, (long) verity
->root_hash_sig_size
);
2450 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to parse PKCS7 DER signature data.");
2452 s
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2454 return log_oom_debug();
2456 bio
= BIO_new_mem_buf(s
, strlen(s
));
2458 return log_oom_debug();
2460 sk
= sk_X509_new_null();
2462 return log_oom_debug();
2464 STRV_FOREACH(i
, certs
) {
2465 _cleanup_(X509_freep
) X509
*c
= NULL
;
2466 _cleanup_fclose_
FILE *f
= NULL
;
2468 f
= fopen(*i
, "re");
2470 log_debug_errno(errno
, "Failed to open '%s', ignoring: %m", *i
);
2474 c
= PEM_read_X509(f
, NULL
, NULL
, NULL
);
2476 log_debug("Failed to load X509 certificate '%s', ignoring.", *i
);
2480 if (sk_X509_push(sk
, c
) == 0)
2481 return log_oom_debug();
2486 r
= PKCS7_verify(p7
, sk
, NULL
, bio
, NULL
, PKCS7_NOINTERN
|PKCS7_NOVERIFY
);
2488 log_debug("Userspace PKCS#7 validation succeeded.");
2490 log_debug("Userspace PKCS#7 validation failed: %s", ERR_error_string(ERR_get_error(), NULL
));
2494 log_debug("Not doing client-side validation of dm-verity root hash signatures, OpenSSL support disabled.");
2499 static int do_crypt_activate_verity(
2500 struct crypt_device
*cd
,
2502 const VeritySettings
*verity
) {
2504 bool check_signature
;
2511 if (verity
->root_hash_sig
) {
2512 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIGNATURE");
2513 if (r
< 0 && r
!= -ENXIO
)
2514 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIGNATURE");
2516 check_signature
= r
!= 0;
2518 check_signature
= false;
2520 if (check_signature
) {
2522 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
2523 /* First, if we have support for signed keys in the kernel, then try that first. */
2524 r
= sym_crypt_activate_by_signed_key(
2528 verity
->root_hash_size
,
2529 verity
->root_hash_sig
,
2530 verity
->root_hash_sig_size
,
2531 CRYPT_ACTIVATE_READONLY
);
2535 log_debug("Validation of dm-verity signature failed via the kernel, trying userspace validation instead.");
2537 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.",
2538 program_invocation_short_name
);
2541 /* So this didn't work via the kernel, then let's try userspace validation instead. If that
2542 * works we'll try to activate without telling the kernel the signature. */
2544 r
= validate_signature_userspace(verity
);
2548 return log_debug_errno(SYNTHETIC_ERRNO(ENOKEY
),
2549 "Activation of signed Verity volume worked neither via the kernel nor in userspace, can't activate.");
2552 return sym_crypt_activate_by_volume_key(
2556 verity
->root_hash_size
,
2557 CRYPT_ACTIVATE_READONLY
);
2560 static usec_t
verity_timeout(void) {
2561 usec_t t
= 100 * USEC_PER_MSEC
;
2565 /* On slower machines, like non-KVM vm, setting up device may take a long time.
2566 * Let's make the timeout configurable. */
2568 e
= getenv("SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC");
2572 r
= parse_sec(e
, &t
);
2575 "Failed to parse timeout specified in $SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC, "
2576 "using the default timeout (%s).",
2577 FORMAT_TIMESPAN(t
, USEC_PER_MSEC
));
2582 static int verity_partition(
2583 PartitionDesignator designator
,
2584 DissectedPartition
*m
,
2585 DissectedPartition
*v
,
2586 const VeritySettings
*verity
,
2587 DissectImageFlags flags
,
2588 DecryptedImage
*d
) {
2590 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2591 _cleanup_(dm_deferred_remove_cleanp
) char *restore_deferred_remove
= NULL
;
2592 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
2593 _cleanup_close_
int mount_node_fd
= -EBADF
;
2597 assert(v
|| (verity
&& verity
->data_path
));
2599 if (!verity
|| !verity
->root_hash
)
2601 if (!((verity
->designator
< 0 && designator
== PARTITION_ROOT
) ||
2602 (verity
->designator
== designator
)))
2605 if (!m
->found
|| !m
->node
|| !m
->fstype
)
2607 if (!verity
->data_path
) {
2608 if (!v
->found
|| !v
->node
|| !v
->fstype
)
2611 if (!streq(v
->fstype
, "DM_verity_hash"))
2615 r
= dlopen_cryptsetup();
2619 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
2620 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
2621 _cleanup_free_
char *root_hash_encoded
= NULL
;
2623 root_hash_encoded
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2624 if (!root_hash_encoded
)
2627 r
= make_dm_name_and_node(root_hash_encoded
, "-verity", &name
, &node
);
2629 r
= make_dm_name_and_node(m
->node
, "-verity", &name
, &node
);
2633 r
= sym_crypt_init(&cd
, verity
->data_path
?: v
->node
);
2637 cryptsetup_enable_logging(cd
);
2639 r
= sym_crypt_load(cd
, CRYPT_VERITY
, NULL
);
2643 r
= sym_crypt_set_data_device(cd
, m
->node
);
2647 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
2650 /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
2651 * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
2652 * retry a few times before giving up. */
2653 for (unsigned i
= 0; i
< N_DEVICE_NODE_LIST_ATTEMPTS
; i
++) {
2654 _cleanup_(sym_crypt_freep
) struct crypt_device
*existing_cd
= NULL
;
2655 _cleanup_close_
int fd
= -EBADF
;
2657 /* First, check if the device already exists. */
2658 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2659 if (fd
< 0 && !ERRNO_IS_DEVICE_ABSENT(errno
))
2660 return log_debug_errno(errno
, "Failed to open verity device %s: %m", node
);
2662 goto check
; /* The device already exists. Let's check it. */
2664 /* The symlink to the device node does not exist yet. Assume not activated, and let's activate it. */
2665 r
= do_crypt_activate_verity(cd
, name
, verity
);
2667 goto try_open
; /* The device is activated. Let's open it. */
2668 /* libdevmapper can return EINVAL when the device is already in the activation stage.
2669 * There's no way to distinguish this situation from a genuine error due to invalid
2670 * parameters, so immediately fall back to activating the device with a unique name.
2671 * Improvements in libcrypsetup can ensure this never happens:
2672 * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
2673 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2675 if (r
== -ENODEV
) /* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */
2678 -EEXIST
, /* Volume has already been opened and ready to be used. */
2679 -EBUSY
/* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */))
2680 return log_debug_errno(r
, "Failed to activate verity device %s: %m", node
);
2683 if (!restore_deferred_remove
){
2684 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
2685 r
= dm_deferred_remove_cancel(name
);
2686 /* -EBUSY and -ENXIO: the device has already been removed or being removed. We cannot
2687 * use the device, try to open again. See target_message() in drivers/md/dm-ioctl.c
2688 * and dm_cancel_deferred_remove() in drivers/md/dm.c */
2689 if (IN_SET(r
, -EBUSY
, -ENXIO
))
2692 return log_debug_errno(r
, "Failed to disable automated deferred removal for verity device %s: %m", node
);
2694 restore_deferred_remove
= strdup(name
);
2695 if (!restore_deferred_remove
)
2696 return log_oom_debug();
2699 r
= verity_can_reuse(verity
, name
, &existing_cd
);
2700 /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
2701 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2704 -ENOENT
, /* Removed?? */
2705 -EBUSY
, /* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */
2706 -ENODEV
/* Volume is being opened but not ready, crypt_init_by_name() would fail, try to open again. */ ))
2709 return log_debug_errno(r
, "Failed to check if existing verity device %s can be reused: %m", node
);
2712 /* devmapper might say that the device exists, but the devlink might not yet have been
2713 * created. Check and wait for the udev event in that case. */
2714 r
= device_wait_for_devlink(node
, "block", verity_timeout(), NULL
);
2715 /* Fallback to activation with a unique device if it's taking too long */
2716 if (r
== -ETIMEDOUT
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2719 return log_debug_errno(r
, "Failed to wait device node symlink %s: %m", node
);
2724 /* Now, the device is activated and devlink is created. Let's open it. */
2725 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2727 if (!ERRNO_IS_DEVICE_ABSENT(errno
))
2728 return log_debug_errno(errno
, "Failed to open verity device %s: %m", node
);
2730 /* The device has already been removed?? */
2735 mount_node_fd
= TAKE_FD(fd
);
2737 crypt_free_and_replace(cd
, existing_cd
);
2742 /* Device is being removed by another process. Let's wait for a while. */
2743 (void) usleep_safe(2 * USEC_PER_MSEC
);
2746 /* All trials failed or a conflicting verity device exists. Let's try to activate with a unique name. */
2747 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
2748 /* Before trying to activate with unique name, we need to free crypt_device object.
2749 * Otherwise, we get error from libcryptsetup like the following:
2751 * systemd[1234]: Cannot use device /dev/loop5 which is in use (already mapped or mounted).
2756 return verity_partition(designator
, m
, v
, verity
, flags
& ~DISSECT_IMAGE_VERITY_SHARE
, d
);
2759 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
), "All attempts to activate verity device %s failed.", name
);
2762 /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
2763 restore_deferred_remove
= mfree(restore_deferred_remove
);
2765 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
2766 .name
= TAKE_PTR(name
),
2767 .device
= TAKE_PTR(cd
),
2770 m
->decrypted_node
= TAKE_PTR(node
);
2771 close_and_replace(m
->mount_node_fd
, mount_node_fd
);
2777 int dissected_image_decrypt(
2779 const char *passphrase
,
2780 const VeritySettings
*verity
,
2781 DissectImageFlags flags
) {
2783 #if HAVE_LIBCRYPTSETUP
2784 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
2789 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
2793 * = 0 → There was nothing to decrypt
2794 * > 0 → Decrypted successfully
2795 * -ENOKEY → There's something to decrypt but no key was supplied
2796 * -EKEYREJECTED → Passed key was not correct
2799 if (verity
&& verity
->root_hash
&& verity
->root_hash_size
< sizeof(sd_id128_t
))
2802 if (!m
->encrypted
&& !m
->verity_ready
)
2805 #if HAVE_LIBCRYPTSETUP
2806 r
= decrypted_image_new(&d
);
2810 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
2811 DissectedPartition
*p
= m
->partitions
+ i
;
2812 PartitionDesignator k
;
2817 r
= decrypt_partition(p
, passphrase
, flags
, d
);
2821 k
= partition_verity_of(i
);
2823 r
= verity_partition(i
, p
, m
->partitions
+ k
, verity
, flags
| DISSECT_IMAGE_VERITY_SHARE
, d
);
2828 if (!p
->decrypted_fstype
&& p
->mount_node_fd
>= 0 && p
->decrypted_node
) {
2829 r
= probe_filesystem_full(p
->mount_node_fd
, p
->decrypted_node
, 0, UINT64_MAX
, &p
->decrypted_fstype
);
2830 if (r
< 0 && r
!= -EUCLEAN
)
2835 m
->decrypted_image
= TAKE_PTR(d
);
2843 int dissected_image_decrypt_interactively(
2845 const char *passphrase
,
2846 const VeritySettings
*verity
,
2847 DissectImageFlags flags
) {
2849 _cleanup_strv_free_erase_
char **z
= NULL
;
2856 r
= dissected_image_decrypt(m
, passphrase
, verity
, flags
);
2859 if (r
== -EKEYREJECTED
)
2860 log_error_errno(r
, "Incorrect passphrase, try again!");
2861 else if (r
!= -ENOKEY
)
2862 return log_error_errno(r
, "Failed to decrypt image: %m");
2865 return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED
),
2866 "Too many retries.");
2870 r
= ask_password_auto("Please enter image passphrase:", NULL
, "dissect", "dissect", "dissect.passphrase", USEC_INFINITY
, 0, &z
);
2872 return log_error_errno(r
, "Failed to query for passphrase: %m");
2878 static int decrypted_image_relinquish(DecryptedImage
*d
) {
2881 /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a
2882 * boolean so that we don't clean it up ourselves either anymore */
2884 #if HAVE_LIBCRYPTSETUP
2887 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
2888 DecryptedPartition
*p
= d
->decrypted
+ i
;
2890 if (p
->relinquished
)
2893 r
= sym_crypt_deactivate_by_name(NULL
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
2895 return log_debug_errno(r
, "Failed to mark %s for auto-removal: %m", p
->name
);
2897 p
->relinquished
= true;
2904 int dissected_image_relinquish(DissectedImage
*m
) {
2909 if (m
->decrypted_image
) {
2910 r
= decrypted_image_relinquish(m
->decrypted_image
);
2916 loop_device_relinquish(m
->loop
);
2921 static char *build_auxiliary_path(const char *image
, const char *suffix
) {
2928 e
= endswith(image
, ".raw");
2930 return strjoin(e
, suffix
);
2932 n
= new(char, e
- image
+ strlen(suffix
) + 1);
2936 strcpy(mempcpy(n
, image
, e
- image
), suffix
);
2940 void verity_settings_done(VeritySettings
*v
) {
2943 v
->root_hash
= mfree(v
->root_hash
);
2944 v
->root_hash_size
= 0;
2946 v
->root_hash_sig
= mfree(v
->root_hash_sig
);
2947 v
->root_hash_sig_size
= 0;
2949 v
->data_path
= mfree(v
->data_path
);
2952 int verity_settings_load(
2953 VeritySettings
*verity
,
2955 const char *root_hash_path
,
2956 const char *root_hash_sig_path
) {
2958 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
2959 size_t root_hash_size
= 0, root_hash_sig_size
= 0;
2960 _cleanup_free_
char *verity_data_path
= NULL
;
2961 PartitionDesignator designator
;
2966 assert(verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
2968 /* If we are asked to load the root hash for a device node, exit early */
2969 if (is_device_path(image
))
2972 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIDECAR");
2973 if (r
< 0 && r
!= -ENXIO
)
2974 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIDECAR, ignoring: %m");
2978 designator
= verity
->designator
;
2980 /* We only fill in what isn't already filled in */
2982 if (!verity
->root_hash
) {
2983 _cleanup_free_
char *text
= NULL
;
2985 if (root_hash_path
) {
2986 /* If explicitly specified it takes precedence */
2987 r
= read_one_line_file(root_hash_path
, &text
);
2992 designator
= PARTITION_ROOT
;
2994 /* Otherwise look for xattr and separate file, and first for the data for root and if
2995 * that doesn't exist for /usr */
2997 if (designator
< 0 || designator
== PARTITION_ROOT
) {
2998 r
= getxattr_malloc(image
, "user.verity.roothash", &text
);
3000 _cleanup_free_
char *p
= NULL
;
3002 if (r
!= -ENOENT
&& !ERRNO_IS_XATTR_ABSENT(r
))
3005 p
= build_auxiliary_path(image
, ".roothash");
3009 r
= read_one_line_file(p
, &text
);
3010 if (r
< 0 && r
!= -ENOENT
)
3015 designator
= PARTITION_ROOT
;
3018 if (!text
&& (designator
< 0 || designator
== PARTITION_USR
)) {
3019 /* So in the "roothash" xattr/file name above the "root" of course primarily
3020 * refers to the root of the Verity Merkle tree. But coincidentally it also
3021 * is the hash for the *root* file system, i.e. the "root" neatly refers to
3022 * two distinct concepts called "root". Taking benefit of this happy
3023 * coincidence we call the file with the root hash for the /usr/ file system
3024 * `usrhash`, because `usrroothash` or `rootusrhash` would just be too
3025 * confusing. We thus drop the reference to the root of the Merkle tree, and
3026 * just indicate which file system it's about. */
3027 r
= getxattr_malloc(image
, "user.verity.usrhash", &text
);
3029 _cleanup_free_
char *p
= NULL
;
3031 if (r
!= -ENOENT
&& !ERRNO_IS_XATTR_ABSENT(r
))
3034 p
= build_auxiliary_path(image
, ".usrhash");
3038 r
= read_one_line_file(p
, &text
);
3039 if (r
< 0 && r
!= -ENOENT
)
3044 designator
= PARTITION_USR
;
3049 r
= unhexmem(text
, strlen(text
), &root_hash
, &root_hash_size
);
3052 if (root_hash_size
< sizeof(sd_id128_t
))
3057 if ((root_hash
|| verity
->root_hash
) && !verity
->root_hash_sig
) {
3058 if (root_hash_sig_path
) {
3059 r
= read_full_file(root_hash_sig_path
, (char**) &root_hash_sig
, &root_hash_sig_size
);
3060 if (r
< 0 && r
!= -ENOENT
)
3064 designator
= PARTITION_ROOT
;
3066 if (designator
< 0 || designator
== PARTITION_ROOT
) {
3067 _cleanup_free_
char *p
= NULL
;
3069 /* Follow naming convention recommended by the relevant RFC:
3070 * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
3071 p
= build_auxiliary_path(image
, ".roothash.p7s");
3075 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
3076 if (r
< 0 && r
!= -ENOENT
)
3079 designator
= PARTITION_ROOT
;
3082 if (!root_hash_sig
&& (designator
< 0 || designator
== PARTITION_USR
)) {
3083 _cleanup_free_
char *p
= NULL
;
3085 p
= build_auxiliary_path(image
, ".usrhash.p7s");
3089 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
3090 if (r
< 0 && r
!= -ENOENT
)
3093 designator
= PARTITION_USR
;
3097 if (root_hash_sig
&& root_hash_sig_size
== 0) /* refuse empty size signatures */
3101 if (!verity
->data_path
) {
3102 _cleanup_free_
char *p
= NULL
;
3104 p
= build_auxiliary_path(image
, ".verity");
3108 if (access(p
, F_OK
) < 0) {
3109 if (errno
!= ENOENT
)
3112 verity_data_path
= TAKE_PTR(p
);
3116 verity
->root_hash
= TAKE_PTR(root_hash
);
3117 verity
->root_hash_size
= root_hash_size
;
3120 if (root_hash_sig
) {
3121 verity
->root_hash_sig
= TAKE_PTR(root_hash_sig
);
3122 verity
->root_hash_sig_size
= root_hash_sig_size
;
3125 if (verity_data_path
)
3126 verity
->data_path
= TAKE_PTR(verity_data_path
);
3128 if (verity
->designator
< 0)
3129 verity
->designator
= designator
;
3134 int dissected_image_load_verity_sig_partition(
3137 VeritySettings
*verity
) {
3139 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
3140 _cleanup_(json_variant_unrefp
) JsonVariant
*v
= NULL
;
3141 size_t root_hash_size
, root_hash_sig_size
;
3142 _cleanup_free_
char *buf
= NULL
;
3143 PartitionDesignator d
;
3144 DissectedPartition
*p
;
3145 JsonVariant
*rh
, *sig
;
3154 if (verity
->root_hash
&& verity
->root_hash_sig
) /* Already loaded? */
3157 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_EMBEDDED");
3158 if (r
< 0 && r
!= -ENXIO
)
3159 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_EMBEDDED, ignoring: %m");
3163 d
= partition_verity_sig_of(verity
->designator
< 0 ? PARTITION_ROOT
: verity
->designator
);
3166 p
= m
->partitions
+ d
;
3169 if (p
->offset
== UINT64_MAX
|| p
->size
== UINT64_MAX
)
3172 if (p
->size
> 4*1024*1024) /* Signature data cannot possible be larger than 4M, refuse that */
3173 return log_debug_errno(SYNTHETIC_ERRNO(EFBIG
), "Verity signature partition is larger than 4M, refusing.");
3175 buf
= new(char, p
->size
+1);
3179 n
= pread(fd
, buf
, p
->size
, p
->offset
);
3182 if ((uint64_t) n
!= p
->size
)
3185 e
= memchr(buf
, 0, p
->size
);
3187 /* If we found a NUL byte then the rest of the data must be NUL too */
3188 if (!memeqzero(e
, p
->size
- (e
- buf
)))
3189 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature data contains embedded NUL byte.");
3193 r
= json_parse(buf
, 0, &v
, NULL
, NULL
);
3195 return log_debug_errno(r
, "Failed to parse signature JSON data: %m");
3197 rh
= json_variant_by_key(v
, "rootHash");
3199 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'rootHash' field.");
3200 if (!json_variant_is_string(rh
))
3201 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'rootHash' field of signature JSON object is not a string.");
3203 r
= unhexmem(json_variant_string(rh
), SIZE_MAX
, &root_hash
, &root_hash_size
);
3205 return log_debug_errno(r
, "Failed to parse root hash field: %m");
3207 /* Check if specified root hash matches if it is specified */
3208 if (verity
->root_hash
&&
3209 memcmp_nn(verity
->root_hash
, verity
->root_hash_size
, root_hash
, root_hash_size
) != 0) {
3210 _cleanup_free_
char *a
= NULL
, *b
= NULL
;
3212 a
= hexmem(root_hash
, root_hash_size
);
3213 b
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
3215 return log_debug_errno(r
, "Root hash in signature JSON data (%s) doesn't match configured hash (%s).", strna(a
), strna(b
));
3218 sig
= json_variant_by_key(v
, "signature");
3220 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'signature' field.");
3221 if (!json_variant_is_string(sig
))
3222 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'signature' field of signature JSON object is not a string.");
3224 r
= unbase64mem(json_variant_string(sig
), SIZE_MAX
, &root_hash_sig
, &root_hash_sig_size
);
3226 return log_debug_errno(r
, "Failed to parse signature field: %m");
3228 free_and_replace(verity
->root_hash
, root_hash
);
3229 verity
->root_hash_size
= root_hash_size
;
3231 free_and_replace(verity
->root_hash_sig
, root_hash_sig
);
3232 verity
->root_hash_sig_size
= root_hash_sig_size
;
3237 int dissected_image_acquire_metadata(DissectedImage
*m
, DissectImageFlags extra_flags
) {
3244 META_INITRD_RELEASE
,
3245 META_EXTENSION_RELEASE
,
3246 META_HAS_INIT_SYSTEM
,
3250 static const char *const paths
[_META_MAX
] = {
3251 [META_HOSTNAME
] = "/etc/hostname\0",
3252 [META_MACHINE_ID
] = "/etc/machine-id\0",
3253 [META_MACHINE_INFO
] = "/etc/machine-info\0",
3254 [META_OS_RELEASE
] = ("/etc/os-release\0"
3255 "/usr/lib/os-release\0"),
3256 [META_INITRD_RELEASE
] = ("/etc/initrd-release\0"
3257 "/usr/lib/initrd-release\0"),
3258 [META_EXTENSION_RELEASE
] = "extension-release\0", /* Used only for logging. */
3259 [META_HAS_INIT_SYSTEM
] = "has-init-system\0", /* ditto */
3262 _cleanup_strv_free_
char **machine_info
= NULL
, **os_release
= NULL
, **initrd_release
= NULL
, **extension_release
= NULL
;
3263 _cleanup_close_pair_
int error_pipe
[2] = PIPE_EBADF
;
3264 _cleanup_(rmdir_and_freep
) char *t
= NULL
;
3265 _cleanup_(sigkill_waitp
) pid_t child
= 0;
3266 sd_id128_t machine_id
= SD_ID128_NULL
;
3267 _cleanup_free_
char *hostname
= NULL
;
3268 unsigned n_meta_initialized
= 0;
3269 int fds
[2 * _META_MAX
], r
, v
;
3270 int has_init_system
= -1;
3272 ImageClass image_class
= IMAGE_SYSEXT
;
3274 BLOCK_SIGNALS(SIGCHLD
);
3277 assert(image_class
);
3279 for (; n_meta_initialized
< _META_MAX
; n_meta_initialized
++) {
3280 if (!paths
[n_meta_initialized
]) {
3281 fds
[2*n_meta_initialized
] = fds
[2*n_meta_initialized
+1] = -EBADF
;
3285 if (pipe2(fds
+ 2*n_meta_initialized
, O_CLOEXEC
) < 0) {
3291 r
= mkdtemp_malloc("/tmp/dissect-XXXXXX", &t
);
3295 if (pipe2(error_pipe
, O_CLOEXEC
) < 0) {
3300 r
= safe_fork("(sd-dissect)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_NEW_MOUNTNS
|FORK_MOUNTNS_SLAVE
, &child
);
3304 /* Child in a new mount namespace */
3305 error_pipe
[0] = safe_close(error_pipe
[0]);
3307 r
= dissected_image_mount(
3313 DISSECT_IMAGE_READ_ONLY
|
3314 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
3315 DISSECT_IMAGE_USR_NO_ROOT
);
3317 log_debug_errno(r
, "Failed to mount dissected image: %m");
3321 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
3322 _cleanup_close_
int fd
= -ENOENT
;
3327 fds
[2*k
] = safe_close(fds
[2*k
]);
3331 case META_EXTENSION_RELEASE
: {
3332 /* As per the os-release spec, if the image is an extension it will have a file
3333 * named after the image name in extension-release.d/ - we use the image name
3334 * and try to resolve it with the extension-release helpers, as sometimes
3335 * the image names are mangled on deployment and do not match anymore.
3336 * Unlike other paths this is not fixed, and the image name
3337 * can be mangled on deployment, so by calling into the helper
3338 * we allow a fallback that matches on the first extension-release
3339 * file found in the directory, if one named after the image cannot
3340 * be found first. */
3341 ImageClass
class = IMAGE_SYSEXT
;
3342 r
= open_extension_release(t
, IMAGE_SYSEXT
, m
->image_name
, /* relax_extension_release_check= */ false, NULL
, &fd
);
3344 r
= open_extension_release(t
, IMAGE_CONFEXT
, m
->image_name
, /* relax_extension_release_check= */ false, NULL
, &fd
);
3346 class = IMAGE_CONFEXT
;
3351 r
= loop_write(fds
[2*k
+1], &class, sizeof(class));
3353 goto inner_fail
; /* Propagate the error to the parent */
3359 case META_HAS_INIT_SYSTEM
: {
3362 FOREACH_STRING(init
,
3363 "/usr/lib/systemd/systemd", /* systemd on /usr merged system */
3364 "/lib/systemd/systemd", /* systemd on /usr non-merged systems */
3365 "/sbin/init") { /* traditional path the Linux kernel invokes */
3367 r
= chase(init
, t
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
3370 log_debug_errno(r
, "Failed to resolve %s, ignoring: %m", init
);
3377 r
= loop_write(fds
[2*k
+1], &found
, sizeof(found
));
3385 NULSTR_FOREACH(p
, paths
[k
]) {
3386 fd
= chase_and_open(p
, t
, CHASE_PREFIX_ROOT
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
, NULL
);
3393 log_debug_errno(fd
, "Failed to read %s file of image, ignoring: %m", paths
[k
]);
3394 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3398 r
= copy_bytes(fd
, fds
[2*k
+1], UINT64_MAX
, 0);
3402 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3405 _exit(EXIT_SUCCESS
);
3408 /* Let parent know the error */
3409 (void) write(error_pipe
[1], &r
, sizeof(r
));
3410 _exit(EXIT_FAILURE
);
3413 error_pipe
[1] = safe_close(error_pipe
[1]);
3415 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
3416 _cleanup_fclose_
FILE *f
= NULL
;
3421 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3423 f
= take_fdopen(&fds
[2*k
], "r");
3432 r
= read_etc_hostname_stream(f
, &hostname
);
3434 log_debug_errno(r
, "Failed to read /etc/hostname of image: %m");
3438 case META_MACHINE_ID
: {
3439 _cleanup_free_
char *line
= NULL
;
3441 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3443 log_debug_errno(r
, "Failed to read /etc/machine-id of image: %m");
3445 r
= sd_id128_from_string(line
, &machine_id
);
3447 log_debug_errno(r
, "Image contains invalid /etc/machine-id: %s", line
);
3449 log_debug("/etc/machine-id file of image is empty.");
3450 else if (streq(line
, "uninitialized"))
3451 log_debug("/etc/machine-id file of image is uninitialized (likely aborted first boot).");
3453 log_debug("/etc/machine-id file of image has unexpected length %i.", r
);
3458 case META_MACHINE_INFO
:
3459 r
= load_env_file_pairs(f
, "machine-info", &machine_info
);
3461 log_debug_errno(r
, "Failed to read /etc/machine-info of image: %m");
3465 case META_OS_RELEASE
:
3466 r
= load_env_file_pairs(f
, "os-release", &os_release
);
3468 log_debug_errno(r
, "Failed to read OS release file of image: %m");
3472 case META_INITRD_RELEASE
:
3473 r
= load_env_file_pairs(f
, "initrd-release", &initrd_release
);
3475 log_debug_errno(r
, "Failed to read initrd release file of image: %m");
3479 case META_EXTENSION_RELEASE
: {
3480 ImageClass cl
= IMAGE_SYSEXT
;
3484 nr
= fread(&cl
, 1, sizeof(cl
), f
);
3485 if (nr
!= sizeof(cl
))
3486 log_debug_errno(errno_or_else(EIO
), "Failed to read class of extension image: %m");
3489 r
= load_env_file_pairs(f
, "extension-release", &extension_release
);
3491 log_debug_errno(r
, "Failed to read extension release file of image: %m");
3497 case META_HAS_INIT_SYSTEM
: {
3502 nr
= fread(&b
, 1, sizeof(b
), f
);
3503 if (nr
!= sizeof(b
))
3504 log_debug_errno(errno_or_else(EIO
), "Failed to read has-init-system boolean: %m");
3506 has_init_system
= b
;
3512 r
= wait_for_terminate_and_check("(sd-dissect)", child
, 0);
3517 n
= read(error_pipe
[0], &v
, sizeof(v
));
3521 return v
; /* propagate error sent to us from child */
3525 if (r
!= EXIT_SUCCESS
)
3528 free_and_replace(m
->hostname
, hostname
);
3529 m
->machine_id
= machine_id
;
3530 strv_free_and_replace(m
->machine_info
, machine_info
);
3531 strv_free_and_replace(m
->os_release
, os_release
);
3532 strv_free_and_replace(m
->initrd_release
, initrd_release
);
3533 strv_free_and_replace(m
->extension_release
, extension_release
);
3534 m
->has_init_system
= has_init_system
;
3535 m
->image_class
= image_class
;
3538 for (unsigned k
= 0; k
< n_meta_initialized
; k
++)
3539 safe_close_pair(fds
+ 2*k
);
3544 Architecture
dissected_image_architecture(DissectedImage
*img
) {
3547 if (img
->partitions
[PARTITION_ROOT
].found
&&
3548 img
->partitions
[PARTITION_ROOT
].architecture
>= 0)
3549 return img
->partitions
[PARTITION_ROOT
].architecture
;
3551 if (img
->partitions
[PARTITION_USR
].found
&&
3552 img
->partitions
[PARTITION_USR
].architecture
>= 0)
3553 return img
->partitions
[PARTITION_USR
].architecture
;
3555 return _ARCHITECTURE_INVALID
;
3558 int dissect_loop_device(
3560 const VeritySettings
*verity
,
3561 const MountOptions
*mount_options
,
3562 const ImagePolicy
*image_policy
,
3563 DissectImageFlags flags
,
3564 DissectedImage
**ret
) {
3567 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
3572 r
= dissected_image_new(loop
->backing_file
?: loop
->node
, &m
);
3576 m
->loop
= loop_device_ref(loop
);
3577 m
->sector_size
= m
->loop
->sector_size
;
3579 r
= dissect_image(m
, loop
->fd
, loop
->node
, verity
, mount_options
, image_policy
, flags
);
3592 int dissect_loop_device_and_warn(
3594 const VeritySettings
*verity
,
3595 const MountOptions
*mount_options
,
3596 const ImagePolicy
*image_policy
,
3597 DissectImageFlags flags
,
3598 DissectedImage
**ret
) {
3602 return dissect_log_error(
3604 dissect_loop_device(loop
, verity
, mount_options
, image_policy
, flags
, ret
),
3605 loop
->backing_file
?: loop
->node
,
3610 bool dissected_image_verity_candidate(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3613 /* Checks if this partition could theoretically do Verity. For non-partitioned images this only works
3614 * if there's an external verity file supplied, for which we can consult .has_verity. For partitioned
3615 * images we only check the partition type.
3617 * This call is used to decide whether to suppress or show a verity column in tabular output of the
3620 if (image
->single_file_system
)
3621 return partition_designator
== PARTITION_ROOT
&& image
->has_verity
;
3623 return partition_verity_of(partition_designator
) >= 0;
3626 bool dissected_image_verity_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3627 PartitionDesignator k
;
3631 /* Checks if this partition has verity data available that we can activate. For non-partitioned this
3632 * works for the root partition, for others only if the associated verity partition was found. */
3634 if (!image
->verity_ready
)
3637 if (image
->single_file_system
)
3638 return partition_designator
== PARTITION_ROOT
;
3640 k
= partition_verity_of(partition_designator
);
3641 return k
>= 0 && image
->partitions
[k
].found
;
3644 bool dissected_image_verity_sig_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3645 PartitionDesignator k
;
3649 /* Checks if this partition has verity signature data available that we can use. */
3651 if (!image
->verity_sig_ready
)
3654 if (image
->single_file_system
)
3655 return partition_designator
== PARTITION_ROOT
;
3657 k
= partition_verity_sig_of(partition_designator
);
3658 return k
>= 0 && image
->partitions
[k
].found
;
3661 MountOptions
* mount_options_free_all(MountOptions
*options
) {
3664 while ((m
= LIST_POP(mount_options
, options
))) {
3672 const char* mount_options_from_designator(const MountOptions
*options
, PartitionDesignator designator
) {
3673 LIST_FOREACH(mount_options
, m
, options
)
3674 if (designator
== m
->partition_designator
&& !isempty(m
->options
))
3680 int mount_image_privately_interactively(
3682 const ImagePolicy
*image_policy
,
3683 DissectImageFlags flags
,
3684 char **ret_directory
,
3686 LoopDevice
**ret_loop_device
) {
3688 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3689 _cleanup_(loop_device_unrefp
) LoopDevice
*d
= NULL
;
3690 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3691 _cleanup_free_
char *dir
= NULL
;
3694 /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
3695 * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
3699 assert(ret_loop_device
);
3701 /* We intend to mount this right-away, hence add the partitions if needed and pin them. */
3702 flags
|= DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
3703 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
3705 r
= verity_settings_load(&verity
, image
, NULL
, NULL
);
3707 return log_error_errno(r
, "Failed to load root hash data: %m");
3709 r
= loop_device_make_by_path(
3711 FLAGS_SET(flags
, DISSECT_IMAGE_DEVICE_READ_ONLY
) ? O_RDONLY
: O_RDWR
,
3712 /* sector_size= */ UINT32_MAX
,
3713 FLAGS_SET(flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
3717 return log_error_errno(r
, "Failed to set up loopback device for %s: %m", image
);
3719 r
= dissect_loop_device_and_warn(
3722 /* mount_options= */ NULL
,
3729 r
= dissected_image_load_verity_sig_partition(dissected_image
, d
->fd
, &verity
);
3733 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, &verity
, flags
);
3737 r
= detach_mount_namespace();
3739 return log_error_errno(r
, "Failed to detach mount namespace: %m");
3741 r
= mkdir_p("/run/systemd/mount-rootfs", 0555);
3743 return log_error_errno(r
, "Failed to create mount point: %m");
3745 r
= dissected_image_mount_and_warn(
3747 "/run/systemd/mount-rootfs",
3748 /* uid_shift= */ UID_INVALID
,
3749 /* uid_range= */ UID_INVALID
,
3754 r
= loop_device_flock(d
, LOCK_UN
);
3758 r
= dissected_image_relinquish(dissected_image
);
3760 return log_error_errno(r
, "Failed to relinquish DM and loopback block devices: %m");
3762 if (ret_directory
) {
3763 dir
= strdup("/run/systemd/mount-rootfs");
3769 _cleanup_close_
int dir_fd
= -EBADF
;
3771 dir_fd
= open("/run/systemd/mount-rootfs", O_CLOEXEC
|O_DIRECTORY
);
3773 return log_error_errno(errno
, "Failed to open mount point directory: %m");
3775 *ret_dir_fd
= TAKE_FD(dir_fd
);
3779 *ret_directory
= TAKE_PTR(dir
);
3781 *ret_loop_device
= TAKE_PTR(d
);
3785 static bool mount_options_relax_extension_release_checks(const MountOptions
*options
) {
3789 return string_contains_word(mount_options_from_designator(options
, PARTITION_ROOT
), ",", "x-systemd.relax-extension-release-check") ||
3790 string_contains_word(mount_options_from_designator(options
, PARTITION_USR
), ",", "x-systemd.relax-extension-release-check") ||
3791 string_contains_word(options
->options
, ",", "x-systemd.relax-extension-release-check");
3794 int verity_dissect_and_mount(
3798 const MountOptions
*options
,
3799 const ImagePolicy
*image_policy
,
3800 const char *required_host_os_release_id
,
3801 const char *required_host_os_release_version_id
,
3802 const char *required_host_os_release_sysext_level
,
3803 const char *required_sysext_scope
) {
3805 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
3806 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3807 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3808 DissectImageFlags dissect_image_flags
;
3809 bool relax_extension_release_check
;
3815 relax_extension_release_check
= mount_options_relax_extension_release_checks(options
);
3817 /* We might get an FD for the image, but we use the original path to look for the dm-verity files */
3818 r
= verity_settings_load(&verity
, src
, NULL
, NULL
);
3820 return log_debug_errno(r
, "Failed to load root hash: %m");
3822 dissect_image_flags
= (verity
.data_path
? DISSECT_IMAGE_NO_PARTITION_TABLE
: 0) |
3823 (relax_extension_release_check
? DISSECT_IMAGE_RELAX_EXTENSION_CHECK
: 0) |
3824 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
3825 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
3827 /* Note that we don't use loop_device_make here, as the FD is most likely O_PATH which would not be
3828 * accepted by LOOP_CONFIGURE, so just let loop_device_make_by_path reopen it as a regular FD. */
3829 r
= loop_device_make_by_path(
3830 src_fd
>= 0 ? FORMAT_PROC_FD_PATH(src_fd
) : src
,
3831 /* open_flags= */ -1,
3832 /* sector_size= */ UINT32_MAX
,
3833 verity
.data_path
? 0 : LO_FLAGS_PARTSCAN
,
3837 return log_debug_errno(r
, "Failed to create loop device for image: %m");
3839 r
= dissect_loop_device(
3844 dissect_image_flags
,
3846 /* No partition table? Might be a single-filesystem image, try again */
3847 if (!verity
.data_path
&& r
== -ENOPKG
)
3848 r
= dissect_loop_device(
3853 dissect_image_flags
| DISSECT_IMAGE_NO_PARTITION_TABLE
,
3856 return log_debug_errno(r
, "Failed to dissect image: %m");
3858 r
= dissected_image_load_verity_sig_partition(dissected_image
, loop_device
->fd
, &verity
);
3862 r
= dissected_image_decrypt(
3866 dissect_image_flags
);
3868 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
3870 r
= mkdir_p_label(dest
, 0755);
3872 return log_debug_errno(r
, "Failed to create destination directory %s: %m", dest
);
3873 r
= umount_recursive(dest
, 0);
3875 return log_debug_errno(r
, "Failed to umount under destination directory %s: %m", dest
);
3877 r
= dissected_image_mount(dissected_image
, dest
, UID_INVALID
, UID_INVALID
, dissect_image_flags
);
3879 return log_debug_errno(r
, "Failed to mount image: %m");
3881 r
= loop_device_flock(loop_device
, LOCK_UN
);
3883 return log_debug_errno(r
, "Failed to unlock loopback device: %m");
3885 /* If we got os-release values from the caller, then we need to match them with the image's
3886 * extension-release.d/ content. Return -EINVAL if there's any mismatch.
3887 * First, check the distro ID. If that matches, then check the new SYSEXT_LEVEL value if
3888 * available, or else fallback to VERSION_ID. If neither is present (eg: rolling release),
3889 * then a simple match on the ID will be performed. */
3890 if (required_host_os_release_id
) {
3891 _cleanup_strv_free_
char **extension_release
= NULL
;
3892 ImageClass
class = IMAGE_SYSEXT
;
3894 assert(!isempty(required_host_os_release_id
));
3896 r
= load_extension_release_pairs(dest
, IMAGE_SYSEXT
, dissected_image
->image_name
, relax_extension_release_check
, &extension_release
);
3898 r
= load_extension_release_pairs(dest
, IMAGE_CONFEXT
, dissected_image
->image_name
, relax_extension_release_check
, &extension_release
);
3900 class = IMAGE_CONFEXT
;
3903 return log_debug_errno(r
, "Failed to parse image %s extension-release metadata: %m", dissected_image
->image_name
);
3905 r
= extension_release_validate(
3906 dissected_image
->image_name
,
3907 required_host_os_release_id
,
3908 required_host_os_release_version_id
,
3909 required_host_os_release_sysext_level
,
3910 required_sysext_scope
,
3914 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE
), "Image %s extension-release metadata does not match the root's", dissected_image
->image_name
);
3916 return log_debug_errno(r
, "Failed to compare image %s extension-release metadata with the root's os-release: %m", dissected_image
->image_name
);
3919 r
= dissected_image_relinquish(dissected_image
);
3921 return log_debug_errno(r
, "Failed to relinquish dissected image: %m");