1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
3 #if HAVE_VALGRIND_MEMCHECK_H
4 #include <valgrind/memcheck.h>
7 #include <linux/dm-ioctl.h>
8 #include <linux/loop.h>
10 #include <sys/mount.h>
11 #include <sys/prctl.h>
16 #include <openssl/err.h>
17 #include <openssl/pem.h>
18 #include <openssl/x509.h>
21 #include "sd-device.h"
24 #include "architecture.h"
25 #include "ask-password-api.h"
26 #include "blkid-util.h"
27 #include "blockdev-util.h"
28 #include "btrfs-util.h"
30 #include "conf-files.h"
31 #include "constants.h"
33 #include "cryptsetup-util.h"
34 #include "device-nodes.h"
35 #include "device-util.h"
36 #include "devnum-util.h"
37 #include "discover-image.h"
38 #include "dissect-image.h"
42 #include "extension-util.h"
46 #include "fsck-util.h"
48 #include "hexdecoct.h"
49 #include "hostname-setup.h"
50 #include "id128-util.h"
51 #include "import-util.h"
53 #include "missing_mount.h"
54 #include "mkdir-label.h"
55 #include "mount-util.h"
56 #include "mountpoint-util.h"
57 #include "namespace-util.h"
58 #include "nulstr-util.h"
59 #include "openssl-util.h"
61 #include "path-util.h"
62 #include "process-util.h"
63 #include "raw-clone.h"
64 #include "resize-fs.h"
65 #include "signal-util.h"
66 #include "sparse-endian.h"
67 #include "stat-util.h"
68 #include "stdio-util.h"
69 #include "string-table.h"
70 #include "string-util.h"
72 #include "tmpfile-util.h"
73 #include "udev-util.h"
74 #include "user-util.h"
75 #include "xattr-util.h"
77 /* how many times to wait for the device nodes to appear */
78 #define N_DEVICE_NODE_LIST_ATTEMPTS 10
80 int dissect_fstype_ok(const char *fstype
) {
84 /* When we automatically mount file systems, be a bit conservative by default what we are willing to
85 * mount, just as an extra safety net to not mount with badly maintained legacy file system
88 e
= secure_getenv("SYSTEMD_DISSECT_FILE_SYSTEMS");
90 _cleanup_strv_free_
char **l
= NULL
;
92 l
= strv_split(e
, ":");
96 b
= strv_contains(l
, fstype
);
98 b
= STR_IN_SET(fstype
,
108 log_debug("File system type '%s' is not allowed to be mounted as result of automatic dissection.", fstype
);
112 int probe_sector_size(int fd
, uint32_t *ret
) {
114 /* Disk images might be for 512B or for 4096 sector sizes, let's try to auto-detect that by searching
115 * for the GPT headers at the relevant byte offsets */
117 assert_cc(sizeof(GptHeader
) == 92);
119 /* We expect a sector size in the range 512…4096. The GPT header is located in the second
120 * sector. Hence it could be at byte 512 at the earliest, and at byte 4096 at the latest. And we must
121 * read with granularity of the largest sector size we care about. Which means 8K. */
122 uint8_t sectors
[2 * 4096];
129 n
= pread(fd
, sectors
, sizeof(sectors
), 0);
132 if (n
!= sizeof(sectors
)) /* too short? */
135 /* Let's see if we find the GPT partition header with various expected sector sizes */
136 for (uint32_t sz
= 512; sz
<= 4096; sz
<<= 1) {
139 assert(sizeof(sectors
) >= sz
* 2);
140 p
= (const GptHeader
*) (sectors
+ sz
);
142 if (!gpt_header_has_signature(p
))
146 return log_debug_errno(SYNTHETIC_ERRNO(ENOTUNIQ
),
147 "Detected valid partition table at offsets matching multiple sector sizes, refusing.");
153 log_debug("Determined sector size %" PRIu32
" based on discovered partition table.", found
);
155 return 1; /* indicate we *did* find it */
159 log_debug("Couldn't find any partition table to derive sector size of.");
160 *ret
= 512; /* pick the traditional default */
161 return 0; /* indicate we didn't find it */
164 int probe_sector_size_prefer_ioctl(int fd
, uint32_t *ret
) {
170 /* Just like probe_sector_size(), but if we are looking at a block device, will use the already
171 * configured sector size rather than probing by contents */
173 if (fstat(fd
, &st
) < 0)
176 if (S_ISBLK(st
.st_mode
))
177 return blockdev_get_sector_size(fd
, ret
);
179 return probe_sector_size(fd
, ret
);
182 int probe_filesystem_full(
189 /* Try to find device content type and return it in *ret_fstype. If nothing is found,
190 * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and a
191 * different error otherwise. */
194 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
195 _cleanup_free_
char *path_by_fd
= NULL
;
196 _cleanup_close_
int fd_close
= -EBADF
;
200 assert(fd
>= 0 || path
);
204 fd_close
= open(path
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
212 r
= fd_get_path(fd
, &path_by_fd
);
219 if (size
== 0) /* empty size? nothing found! */
222 b
= blkid_new_probe();
226 /* The Linux kernel maintains separate block device caches for main ("whole") and partition block
227 * devices, which means making a change to one might not be reflected immediately when reading via
228 * the other. That's massively confusing when mixing accesses to such devices. Let's address this in
229 * a limited way: when probing a file system that is not at the beginning of the block device we
230 * apparently probe a partition via the main block device, and in that case let's first flush the
231 * main block device cache, so that we get the data that the per-partition block device last
234 * This only works under the assumption that any tools that write to the partition block devices
235 * issue an syncfs()/fsync() on the device after making changes. Typically file system formatting
236 * tools that write a superblock onto a partition block device do that, however. */
238 if (ioctl(fd
, BLKFLSBUF
, 0) < 0)
239 log_debug_errno(errno
, "Failed to flush block device cache, ignoring: %m");
242 r
= blkid_probe_set_device(
246 size
== UINT64_MAX
? 0 : size
); /* when blkid sees size=0 it understands "everything". We prefer using UINT64_MAX for that */
248 return errno_or_else(ENOMEM
);
250 blkid_probe_enable_superblocks(b
, 1);
251 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
);
254 r
= blkid_do_safeprobe(b
);
255 if (r
== _BLKID_SAFEPROBE_NOT_FOUND
)
257 if (r
== _BLKID_SAFEPROBE_AMBIGUOUS
)
258 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
),
259 "Results ambiguous for partition %s", path
);
260 if (r
== _BLKID_SAFEPROBE_ERROR
)
261 return log_debug_errno(errno_or_else(EIO
), "Failed to probe partition %s: %m", path
);
263 assert(r
== _BLKID_SAFEPROBE_FOUND
);
265 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
270 log_debug("Probed fstype '%s' on partition %s.", fstype
, path
);
281 log_debug("No type detected on partition %s", path
);
290 static int image_policy_may_use(
291 const ImagePolicy
*policy
,
292 PartitionDesignator designator
) {
294 PartitionPolicyFlags f
;
296 /* For each partition we find in the partition table do a first check if it may exist at all given
297 * the policy, or if it shall be ignored. */
299 f
= image_policy_get_exhaustively(policy
, designator
);
303 if ((f
& _PARTITION_POLICY_USE_MASK
) == PARTITION_POLICY_ABSENT
)
304 /* only flag set in policy is "absent"? then this partition may not exist at all */
305 return log_debug_errno(
306 SYNTHETIC_ERRNO(ERFKILL
),
307 "Partition of designator '%s' exists, but not allowed by policy, refusing.",
308 partition_designator_to_string(designator
));
309 if ((f
& _PARTITION_POLICY_USE_MASK
& ~PARTITION_POLICY_ABSENT
) == PARTITION_POLICY_UNUSED
) {
310 /* only "unused" or "unused" + "absent" are set? then don't use it */
311 log_debug("Partition of designator '%s' exists, and policy dictates to ignore it, doing so.",
312 partition_designator_to_string(designator
));
313 return false; /* ignore! */
316 return true; /* use! */
319 static int image_policy_check_protection(
320 const ImagePolicy
*policy
,
321 PartitionDesignator designator
,
322 PartitionPolicyFlags found_flags
) {
324 PartitionPolicyFlags policy_flags
;
326 /* Checks if the flags in the policy for the designated partition overlap the flags of what we found */
331 policy_flags
= image_policy_get_exhaustively(policy
, designator
);
332 if (policy_flags
< 0)
335 if ((found_flags
& policy_flags
) == 0) {
336 _cleanup_free_
char *found_flags_string
= NULL
, *policy_flags_string
= NULL
;
338 (void) partition_policy_flags_to_string(found_flags
, /* simplify= */ true, &found_flags_string
);
339 (void) partition_policy_flags_to_string(policy_flags
, /* simplify= */ true, &policy_flags_string
);
341 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL
), "Partition %s discovered with policy '%s' but '%s' was required, refusing.",
342 partition_designator_to_string(designator
),
343 strnull(found_flags_string
), strnull(policy_flags_string
));
349 static int image_policy_check_partition_flags(
350 const ImagePolicy
*policy
,
351 PartitionDesignator designator
,
352 uint64_t gpt_flags
) {
354 PartitionPolicyFlags policy_flags
;
357 /* Checks if the partition flags in the policy match reality */
359 policy_flags
= image_policy_get_exhaustively(policy
, designator
);
360 if (policy_flags
< 0)
363 b
= FLAGS_SET(gpt_flags
, SD_GPT_FLAG_READ_ONLY
);
364 if ((policy_flags
& _PARTITION_POLICY_READ_ONLY_MASK
) == (b
? PARTITION_POLICY_READ_ONLY_OFF
: PARTITION_POLICY_READ_ONLY_ON
))
365 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL
), "Partition %s has 'read-only' flag incorrectly set (must be %s, is %s), refusing.",
366 partition_designator_to_string(designator
),
367 one_zero(!b
), one_zero(b
));
369 b
= FLAGS_SET(gpt_flags
, SD_GPT_FLAG_GROWFS
);
370 if ((policy_flags
& _PARTITION_POLICY_GROWFS_MASK
) == (b
? PARTITION_POLICY_GROWFS_OFF
: PARTITION_POLICY_GROWFS_ON
))
371 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL
), "Partition %s has 'growfs' flag incorrectly set (must be %s, is %s), refusing.",
372 partition_designator_to_string(designator
),
373 one_zero(!b
), one_zero(b
));
378 static int dissected_image_probe_filesystems(
381 const ImagePolicy
*policy
) {
387 /* Fill in file system types if we don't know them yet. */
389 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
390 DissectedPartition
*p
= m
->partitions
+ i
;
391 PartitionPolicyFlags found_flags
;
397 /* If we have an fd referring to the partition block device, use that. Otherwise go
398 * via the whole block device or backing regular file, and read via offset. */
399 if (p
->mount_node_fd
>= 0)
400 r
= probe_filesystem_full(p
->mount_node_fd
, p
->node
, 0, UINT64_MAX
, &p
->fstype
);
402 r
= probe_filesystem_full(fd
, p
->node
, p
->offset
, p
->size
, &p
->fstype
);
407 if (streq_ptr(p
->fstype
, "crypto_LUKS")) {
409 found_flags
= PARTITION_POLICY_ENCRYPTED
; /* found this one, and its definitely encrypted */
411 /* found it, but it's definitely not encrypted, hence mask the encrypted flag, but
412 * set all other ways that indicate "present". */
413 found_flags
= PARTITION_POLICY_UNPROTECTED
|PARTITION_POLICY_VERITY
|PARTITION_POLICY_SIGNED
;
415 if (p
->fstype
&& fstype_is_ro(p
->fstype
))
421 /* We might have learnt more about the file system now (i.e. whether it is encrypted or not),
422 * hence we need to validate this against policy again, to see if the policy still matches
423 * with this new information. Note that image_policy_check_protection() will check for
424 * overlap between what's allowed in the policy and what we pass as 'found_policy' here. In
425 * the unencrypted case we thus might pass an overly unspecific mask here (i.e. unprotected
426 * OR verity OR signed), but that's fine since the earlier policy check already checked more
427 * specific which of those three cases where OK. Keep in mind that this function here only
428 * looks at specific partitions (and thus can only deduce encryption or not) but not the
429 * overall partition table (and thus cannot deduce verity or not). The earlier dissection
430 * checks already did the relevant checks that look at the whole partition table, and
431 * enforced policy there as needed. */
432 r
= image_policy_check_protection(policy
, i
, found_flags
);
440 static void check_partition_flags(
442 unsigned long long pflags
,
443 unsigned long long supported
) {
447 /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
448 pflags
&= ~(supported
|
449 SD_GPT_FLAG_REQUIRED_PARTITION
|
450 SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL
|
451 SD_GPT_FLAG_LEGACY_BIOS_BOOTABLE
);
456 /* If there are other bits set, then log about it, to make things discoverable */
457 for (unsigned i
= 0; i
< sizeof(pflags
) * 8; i
++) {
458 unsigned long long bit
= 1ULL << i
;
459 if (!FLAGS_SET(pflags
, bit
))
462 log_debug("Unexpected partition flag %llu set on %s!", bit
, node
);
466 static int dissected_image_new(const char *path
, DissectedImage
**ret
) {
467 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
468 _cleanup_free_
char *name
= NULL
;
474 _cleanup_free_
char *filename
= NULL
;
476 r
= path_extract_filename(path
, &filename
);
480 r
= raw_strip_suffixes(filename
, &name
);
484 if (!image_name_is_valid(name
)) {
485 log_debug("Image name %s is not valid, ignoring.", strna(name
));
490 m
= new(DissectedImage
, 1);
494 *m
= (DissectedImage
) {
495 .has_init_system
= -1,
496 .image_name
= TAKE_PTR(name
),
499 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++)
500 m
->partitions
[i
] = DISSECTED_PARTITION_NULL
;
507 static void dissected_partition_done(DissectedPartition
*p
) {
513 free(p
->decrypted_fstype
);
514 free(p
->decrypted_node
);
515 free(p
->mount_options
);
516 safe_close(p
->mount_node_fd
);
518 *p
= DISSECTED_PARTITION_NULL
;
522 static int make_partition_devname(
523 const char *whole_devname
,
526 DissectImageFlags flags
,
529 _cleanup_free_
char *s
= NULL
;
532 assert(whole_devname
);
533 assert(nr
!= 0); /* zero is not a valid partition nr */
536 if (!FLAGS_SET(flags
, DISSECT_IMAGE_DISKSEQ_DEVNODE
) || diskseq
== 0) {
538 /* Given a whole block device node name (e.g. /dev/sda or /dev/loop7) generate a partition
539 * device name (e.g. /dev/sda7 or /dev/loop7p5). The rule the kernel uses is simple: if whole
540 * block device node name ends in a digit, then suffix a 'p', followed by the partition
541 * number. Otherwise, just suffix the partition number without any 'p'. */
543 if (nr
< 0) { /* whole disk? */
544 s
= strdup(whole_devname
);
548 size_t l
= strlen(whole_devname
);
549 if (l
< 1) /* underflow check for the subtraction below */
552 bool need_p
= ascii_isdigit(whole_devname
[l
-1]); /* Last char a digit? */
554 if (asprintf(&s
, "%s%s%i", whole_devname
, need_p
? "p" : "", nr
) < 0)
558 if (nr
< 0) /* whole disk? */
559 r
= asprintf(&s
, "/dev/disk/by-diskseq/%" PRIu64
, diskseq
);
561 r
= asprintf(&s
, "/dev/disk/by-diskseq/%" PRIu64
"-part%i", diskseq
, nr
);
570 static int open_partition(
573 const LoopDevice
*loop
) {
575 _cleanup_(sd_device_unrefp
) sd_device
*dev
= NULL
;
576 _cleanup_close_
int fd
= -EBADF
;
583 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
587 /* Check if the block device is a child of (or equivalent to) the originally provided one. */
588 r
= block_device_new_from_fd(fd
, is_partition
? BLOCK_DEVICE_LOOKUP_WHOLE_DISK
: 0, &dev
);
592 r
= sd_device_get_devnum(dev
, &devnum
);
596 if (loop
->devno
!= devnum
)
599 /* Also check diskseq. */
600 if (loop
->diskseq
!= 0) {
603 r
= fd_get_diskseq(fd
, &diskseq
);
607 if (loop
->diskseq
!= diskseq
)
611 log_debug("Opened %s (fd=%i, whole_block_devnum=" DEVNUM_FORMAT_STR
", diskseq=%" PRIu64
").",
612 node
, fd
, DEVNUM_FORMAT_VAL(loop
->devno
), loop
->diskseq
);
616 static int compare_arch(Architecture a
, Architecture b
) {
620 if (a
== native_architecture())
623 if (b
== native_architecture())
626 #ifdef ARCHITECTURE_SECONDARY
627 if (a
== ARCHITECTURE_SECONDARY
)
630 if (b
== ARCHITECTURE_SECONDARY
)
637 static int dissect_image(
641 const VeritySettings
*verity
,
642 const MountOptions
*mount_options
,
643 const ImagePolicy
*policy
,
644 DissectImageFlags flags
) {
646 sd_id128_t root_uuid
= SD_ID128_NULL
, root_verity_uuid
= SD_ID128_NULL
;
647 sd_id128_t usr_uuid
= SD_ID128_NULL
, usr_verity_uuid
= SD_ID128_NULL
;
648 bool is_gpt
, is_mbr
, multiple_generic
= false,
649 generic_rw
= false, /* initialize to appease gcc */
650 generic_growfs
= false;
651 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
652 _cleanup_free_
char *generic_node
= NULL
;
653 sd_id128_t generic_uuid
= SD_ID128_NULL
;
654 const char *pttype
= NULL
, *sptuuid
= NULL
;
656 int r
, generic_nr
= -1, n_partitions
;
661 assert(!verity
|| verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
662 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
663 assert(!verity
|| verity
->root_hash_sig
|| verity
->root_hash_sig_size
== 0);
664 assert(!verity
|| (verity
->root_hash
|| !verity
->root_hash_sig
));
665 assert(!((flags
& DISSECT_IMAGE_GPT_ONLY
) && (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)));
666 assert(m
->sector_size
> 0);
668 /* Probes a disk image, and returns information about what it found in *ret.
670 * Returns -ENOPKG if no suitable partition table or file system could be found.
671 * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found.
672 * Returns -ENXIO if we couldn't find any partition suitable as root or /usr partition
673 * Returns -ENOTUNIQ if we only found multiple generic partitions and thus don't know what to do with that
674 * Returns -ERFKILL if image doesn't match image policy
675 * 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)
676 * Returns -EPROTONOSUPPORT if DISSECT_IMAGE_ADD_PARTITION_DEVICES is set but the block device does not have partition logic enabled
677 * Returns -ENOMSG if we didn't find a single usable partition (and DISSECT_IMAGE_REFUSE_EMPTY is set) */
679 uint64_t diskseq
= m
->loop
? m
->loop
->diskseq
: 0;
681 if (verity
&& verity
->root_hash
) {
682 sd_id128_t fsuuid
, vuuid
;
684 /* If a root hash is supplied, then we use the root partition that has a UUID that match the
685 * first 128-bit of the root hash. And we use the verity partition that has a UUID that match
686 * the final 128-bit. */
688 if (verity
->root_hash_size
< sizeof(sd_id128_t
))
691 memcpy(&fsuuid
, verity
->root_hash
, sizeof(sd_id128_t
));
692 memcpy(&vuuid
, (const uint8_t*) verity
->root_hash
+ verity
->root_hash_size
- sizeof(sd_id128_t
), sizeof(sd_id128_t
));
694 if (sd_id128_is_null(fsuuid
))
696 if (sd_id128_is_null(vuuid
))
699 /* If the verity data declares it's for the /usr partition, then search for that, in all
700 * other cases assume it's for the root partition. */
701 if (verity
->designator
== PARTITION_USR
) {
703 usr_verity_uuid
= vuuid
;
706 root_verity_uuid
= vuuid
;
710 b
= blkid_new_probe();
715 r
= blkid_probe_set_device(b
, fd
, 0, 0);
717 return errno_or_else(ENOMEM
);
720 r
= blkid_probe_set_sectorsize(b
, m
->sector_size
);
722 return errno_or_else(EIO
);
724 if ((flags
& DISSECT_IMAGE_GPT_ONLY
) == 0) {
725 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
726 blkid_probe_enable_superblocks(b
, 1);
727 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
|BLKID_SUBLKS_USAGE
|BLKID_SUBLKS_UUID
);
730 blkid_probe_enable_partitions(b
, 1);
731 blkid_probe_set_partitions_flags(b
, BLKID_PARTS_ENTRY_DETAILS
);
734 r
= blkid_do_safeprobe(b
);
735 if (r
== _BLKID_SAFEPROBE_ERROR
)
736 return errno_or_else(EIO
);
737 if (IN_SET(r
, _BLKID_SAFEPROBE_AMBIGUOUS
, _BLKID_SAFEPROBE_NOT_FOUND
))
738 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG
), "Failed to identify any partition table.");
740 assert(r
== _BLKID_SAFEPROBE_FOUND
);
742 if ((!(flags
& DISSECT_IMAGE_GPT_ONLY
) &&
743 (flags
& DISSECT_IMAGE_GENERIC_ROOT
)) ||
744 (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)) {
745 const char *usage
= NULL
;
747 /* If flags permit this, also allow using non-partitioned single-filesystem images */
749 (void) blkid_probe_lookup_value(b
, "USAGE", &usage
, NULL
);
750 if (STRPTR_IN_SET(usage
, "filesystem", "crypto")) {
751 _cleanup_free_
char *t
= NULL
, *n
= NULL
, *o
= NULL
;
752 const char *fstype
= NULL
, *options
= NULL
, *suuid
= NULL
;
753 _cleanup_close_
int mount_node_fd
= -EBADF
;
754 sd_id128_t uuid
= SD_ID128_NULL
;
755 PartitionPolicyFlags found_flags
;
758 /* OK, we have found a file system, that's our root partition then. */
760 r
= image_policy_may_use(policy
, PARTITION_ROOT
);
763 if (r
== 0) /* policy says ignore this, so we ignore it */
766 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
767 (void) blkid_probe_lookup_value(b
, "UUID", &suuid
, NULL
);
769 encrypted
= streq_ptr(fstype
, "crypto_LUKS");
771 if (verity_settings_data_covers(verity
, PARTITION_ROOT
))
772 found_flags
= verity
->root_hash_sig
? PARTITION_POLICY_SIGNED
: PARTITION_POLICY_VERITY
;
774 found_flags
= encrypted
? PARTITION_POLICY_ENCRYPTED
: PARTITION_POLICY_UNPROTECTED
;
776 r
= image_policy_check_protection(policy
, PARTITION_ROOT
, found_flags
);
780 r
= image_policy_check_partition_flags(policy
, PARTITION_ROOT
, 0); /* we have no gpt partition flags, hence check against all bits off */
784 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
785 mount_node_fd
= open_partition(devname
, /* is_partition = */ false, m
->loop
);
786 if (mount_node_fd
< 0)
787 return mount_node_fd
;
797 /* blkid will return FAT's serial number as UUID, hence it is quite possible
798 * that parsing this will fail. We'll ignore the ID, since it's just too
799 * short to be useful as tru identifier. */
800 r
= sd_id128_from_string(suuid
, &uuid
);
802 log_debug_errno(r
, "Failed to parse file system UUID '%s', ignoring: %m", suuid
);
805 r
= make_partition_devname(devname
, diskseq
, -1, flags
, &n
);
809 m
->single_file_system
= true;
810 m
->encrypted
= encrypted
;
812 m
->has_verity
= verity
&& verity
->data_path
;
813 m
->verity_ready
= verity_settings_data_covers(verity
, PARTITION_ROOT
);
815 m
->has_verity_sig
= false; /* signature not embedded, must be specified */
816 m
->verity_sig_ready
= m
->verity_ready
&& verity
->root_hash_sig
;
818 m
->image_uuid
= uuid
;
820 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
827 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
829 .rw
= !m
->verity_ready
&& !fstype_is_ro(fstype
),
831 .architecture
= _ARCHITECTURE_INVALID
,
832 .fstype
= TAKE_PTR(t
),
834 .mount_options
= TAKE_PTR(o
),
835 .mount_node_fd
= TAKE_FD(mount_node_fd
),
844 (void) blkid_probe_lookup_value(b
, "PTTYPE", &pttype
, NULL
);
848 is_gpt
= streq_ptr(pttype
, "gpt");
849 is_mbr
= streq_ptr(pttype
, "dos");
851 if (!is_gpt
&& ((flags
& DISSECT_IMAGE_GPT_ONLY
) || !is_mbr
))
854 /* We support external verity data partitions only if the image has no partition table */
855 if (verity
&& verity
->data_path
)
858 if (FLAGS_SET(flags
, DISSECT_IMAGE_ADD_PARTITION_DEVICES
)) {
859 /* Safety check: refuse block devices that carry a partition table but for which the kernel doesn't
860 * do partition scanning. */
861 r
= blockdev_partscan_enabled(fd
);
865 return -EPROTONOSUPPORT
;
868 (void) blkid_probe_lookup_value(b
, "PTUUID", &sptuuid
, NULL
);
870 r
= sd_id128_from_string(sptuuid
, &m
->image_uuid
);
872 log_debug_errno(r
, "Failed to parse partition table UUID '%s', ignoring: %m", sptuuid
);
876 pl
= blkid_probe_get_partitions(b
);
878 return errno_or_else(ENOMEM
);
881 n_partitions
= blkid_partlist_numof_partitions(pl
);
882 if (n_partitions
< 0)
883 return errno_or_else(EIO
);
885 for (int i
= 0; i
< n_partitions
; i
++) {
886 _cleanup_free_
char *node
= NULL
;
887 unsigned long long pflags
;
888 blkid_loff_t start
, size
;
893 pp
= blkid_partlist_get_partition(pl
, i
);
895 return errno_or_else(EIO
);
897 pflags
= blkid_partition_get_flags(pp
);
900 nr
= blkid_partition_get_partno(pp
);
902 return errno_or_else(EIO
);
905 start
= blkid_partition_get_start(pp
);
907 return errno_or_else(EIO
);
909 assert((uint64_t) start
< UINT64_MAX
/512);
912 size
= blkid_partition_get_size(pp
);
914 return errno_or_else(EIO
);
916 assert((uint64_t) size
< UINT64_MAX
/512);
918 /* While probing we need the non-diskseq device node name to access the thing, hence mask off
919 * DISSECT_IMAGE_DISKSEQ_DEVNODE. */
920 r
= make_partition_devname(devname
, diskseq
, nr
, flags
& ~DISSECT_IMAGE_DISKSEQ_DEVNODE
, &node
);
924 /* So here's the thing: after the main ("whole") block device popped up it might take a while
925 * before the kernel fully probed the partition table. Waiting for that to finish is icky in
926 * userspace. So here's what we do instead. We issue the BLKPG_ADD_PARTITION ioctl to add the
927 * partition ourselves, racing against the kernel. Good thing is: if this call fails with
928 * EBUSY then the kernel was quicker than us, and that's totally OK, the outcome is good for
929 * us: the device node will exist. If OTOH our call was successful we won the race. Which is
930 * also good as the outcome is the same: the partition block device exists, and we can use
933 * Kernel returns EBUSY if there's already a partition by that number or an overlapping
934 * partition already existent. */
936 if (FLAGS_SET(flags
, DISSECT_IMAGE_ADD_PARTITION_DEVICES
)) {
937 r
= block_device_add_partition(fd
, node
, nr
, (uint64_t) start
* 512, (uint64_t) size
* 512);
940 return log_debug_errno(r
, "BLKPG_ADD_PARTITION failed: %m");
942 log_debug_errno(r
, "Kernel was quicker than us in adding partition %i.", nr
);
944 log_debug("We were quicker than kernel in adding partition %i.", nr
);
948 const char *fstype
= NULL
, *label
;
949 sd_id128_t type_id
, id
;
950 GptPartitionType type
;
951 bool rw
= true, growfs
= false;
953 r
= blkid_partition_get_uuid_id128(pp
, &id
);
955 log_debug_errno(r
, "Failed to read partition UUID, ignoring: %m");
959 r
= blkid_partition_get_type_id128(pp
, &type_id
);
961 log_debug_errno(r
, "Failed to read partition type UUID, ignoring: %m");
965 type
= gpt_partition_type_from_uuid(type_id
);
967 label
= blkid_partition_get_name(pp
); /* libblkid returns NULL here if empty */
969 if (IN_SET(type
.designator
,
975 check_partition_flags(node
, pflags
,
976 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
978 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
981 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
982 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
984 } else if (type
.designator
== PARTITION_ESP
) {
986 /* Note that we don't check the SD_GPT_FLAG_NO_AUTO flag for the ESP, as it is
987 * not defined there. We instead check the SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as
988 * recommended by the UEFI spec (See "12.3.3 Number and Location of System
991 if (pflags
& SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL
)
996 } else if (type
.designator
== PARTITION_ROOT
) {
998 check_partition_flags(node
, pflags
,
999 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1001 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1004 /* If a root ID is specified, ignore everything but the root id */
1005 if (!sd_id128_is_null(root_uuid
) && !sd_id128_equal(root_uuid
, id
))
1008 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1009 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1011 } else if (type
.designator
== PARTITION_ROOT_VERITY
) {
1013 check_partition_flags(node
, pflags
,
1014 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1016 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1019 m
->has_verity
= true;
1021 /* If no verity configuration is specified, then don't do verity */
1024 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
1027 /* If root hash is specified, then ignore everything but the root id */
1028 if (!sd_id128_is_null(root_verity_uuid
) && !sd_id128_equal(root_verity_uuid
, id
))
1031 fstype
= "DM_verity_hash";
1034 } else if (type
.designator
== PARTITION_ROOT_VERITY_SIG
) {
1036 check_partition_flags(node
, pflags
,
1037 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1039 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1042 m
->has_verity_sig
= true;
1046 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
1049 fstype
= "verity_hash_signature";
1052 } else if (type
.designator
== PARTITION_USR
) {
1054 check_partition_flags(node
, pflags
,
1055 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1057 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1060 /* If a usr ID is specified, ignore everything but the usr id */
1061 if (!sd_id128_is_null(usr_uuid
) && !sd_id128_equal(usr_uuid
, id
))
1064 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1065 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1067 } else if (type
.designator
== PARTITION_USR_VERITY
) {
1069 check_partition_flags(node
, pflags
,
1070 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1072 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1075 m
->has_verity
= true;
1079 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
1082 /* If usr hash is specified, then ignore everything but the usr id */
1083 if (!sd_id128_is_null(usr_verity_uuid
) && !sd_id128_equal(usr_verity_uuid
, id
))
1086 fstype
= "DM_verity_hash";
1089 } else if (type
.designator
== PARTITION_USR_VERITY_SIG
) {
1091 check_partition_flags(node
, pflags
,
1092 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1094 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1097 m
->has_verity_sig
= true;
1101 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
1104 fstype
= "verity_hash_signature";
1107 } else if (type
.designator
== PARTITION_SWAP
) {
1109 check_partition_flags(node
, pflags
, SD_GPT_FLAG_NO_AUTO
);
1111 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1114 /* Note: we don't set fstype = "swap" here, because we still need to probe if
1115 * it might be encrypted (i.e. fstype "crypt_LUKS") or unencrypted
1116 * (i.e. fstype "swap"), and the only way to figure that out is via fstype
1119 /* We don't have a designator for SD_GPT_LINUX_GENERIC so check the UUID instead. */
1120 } else if (sd_id128_equal(type
.uuid
, SD_GPT_LINUX_GENERIC
)) {
1122 check_partition_flags(node
, pflags
,
1123 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1125 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1129 multiple_generic
= true;
1132 generic_rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1133 generic_growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1135 generic_node
= TAKE_PTR(node
);
1138 } else if (type
.designator
== PARTITION_VAR
) {
1140 check_partition_flags(node
, pflags
,
1141 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1143 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1146 if (!FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_VAR_CHECK
)) {
1147 sd_id128_t var_uuid
;
1149 /* For /var we insist that the uuid of the partition matches the
1150 * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
1151 * ID. Why? Unlike the other partitions /var is inherently
1152 * installation specific, hence we need to be careful not to mount it
1153 * in the wrong installation. By hashing the partition UUID from
1154 * /etc/machine-id we can securely bind the partition to the
1157 r
= sd_id128_get_machine_app_specific(SD_GPT_VAR
, &var_uuid
);
1161 if (!sd_id128_equal(var_uuid
, id
)) {
1162 log_debug("Found a /var/ partition, but its UUID didn't match our expectations "
1163 "(found: " SD_ID128_UUID_FORMAT_STR
", expected: " SD_ID128_UUID_FORMAT_STR
"), ignoring.",
1164 SD_ID128_FORMAT_VAL(id
), SD_ID128_FORMAT_VAL(var_uuid
));
1169 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1170 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1173 if (type
.designator
!= _PARTITION_DESIGNATOR_INVALID
) {
1174 _cleanup_free_
char *t
= NULL
, *o
= NULL
, *l
= NULL
, *n
= NULL
;
1175 _cleanup_close_
int mount_node_fd
= -EBADF
;
1176 const char *options
= NULL
;
1178 r
= image_policy_may_use(policy
, type
.designator
);
1182 /* Policy says: ignore; Remember this fact, so that we later can distinguish between "found but ignored" and "not found at all" */
1184 if (!m
->partitions
[type
.designator
].found
)
1185 m
->partitions
[type
.designator
].ignored
= true;
1190 if (m
->partitions
[type
.designator
].found
) {
1193 /* For most partition types the first one we see wins. Except for the
1194 * rootfs and /usr, where we do a version compare of the label, and
1195 * let the newest version win. This permits a simple A/B versioning
1196 * scheme in OS images. */
1198 c
= compare_arch(type
.arch
, m
->partitions
[type
.designator
].architecture
);
1199 if (c
< 0) /* the arch we already found is better than the one we found now */
1201 if (c
== 0 && /* same arch? then go by version in label */
1202 (!partition_designator_is_versioned(type
.designator
) ||
1203 strverscmp_improved(label
, m
->partitions
[type
.designator
].label
) <= 0))
1206 dissected_partition_done(m
->partitions
+ type
.designator
);
1209 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
) &&
1210 type
.designator
!= PARTITION_SWAP
) {
1211 mount_node_fd
= open_partition(node
, /* is_partition = */ true, m
->loop
);
1212 if (mount_node_fd
< 0)
1213 return mount_node_fd
;
1216 r
= make_partition_devname(devname
, diskseq
, nr
, flags
, &n
);
1232 options
= mount_options_from_designator(mount_options
, type
.designator
);
1234 o
= strdup(options
);
1239 m
->partitions
[type
.designator
] = (DissectedPartition
) {
1244 .architecture
= type
.arch
,
1245 .node
= TAKE_PTR(n
),
1246 .fstype
= TAKE_PTR(t
),
1247 .label
= TAKE_PTR(l
),
1249 .mount_options
= TAKE_PTR(o
),
1250 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1251 .offset
= (uint64_t) start
* 512,
1252 .size
= (uint64_t) size
* 512,
1253 .gpt_flags
= pflags
,
1257 } else if (is_mbr
) {
1259 switch (blkid_partition_get_type(pp
)) {
1261 case 0x83: /* Linux partition */
1263 if (pflags
!= 0x80) /* Bootable flag */
1267 multiple_generic
= true;
1271 generic_growfs
= false;
1272 generic_node
= TAKE_PTR(node
);
1277 case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
1278 _cleanup_close_
int mount_node_fd
= -EBADF
;
1279 _cleanup_free_
char *o
= NULL
, *n
= NULL
;
1280 sd_id128_t id
= SD_ID128_NULL
;
1281 const char *options
= NULL
;
1283 r
= image_policy_may_use(policy
, PARTITION_XBOOTLDR
);
1286 if (r
== 0) { /* policy says: ignore */
1287 if (!m
->partitions
[PARTITION_XBOOTLDR
].found
)
1288 m
->partitions
[PARTITION_XBOOTLDR
].ignored
= true;
1293 /* First one wins */
1294 if (m
->partitions
[PARTITION_XBOOTLDR
].found
)
1297 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
1298 mount_node_fd
= open_partition(node
, /* is_partition = */ true, m
->loop
);
1299 if (mount_node_fd
< 0)
1300 return mount_node_fd
;
1303 (void) blkid_partition_get_uuid_id128(pp
, &id
);
1305 r
= make_partition_devname(devname
, diskseq
, nr
, flags
, &n
);
1309 options
= mount_options_from_designator(mount_options
, PARTITION_XBOOTLDR
);
1311 o
= strdup(options
);
1316 m
->partitions
[PARTITION_XBOOTLDR
] = (DissectedPartition
) {
1321 .architecture
= _ARCHITECTURE_INVALID
,
1322 .node
= TAKE_PTR(n
),
1324 .mount_options
= TAKE_PTR(o
),
1325 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1326 .offset
= (uint64_t) start
* 512,
1327 .size
= (uint64_t) size
* 512,
1335 if (!m
->partitions
[PARTITION_ROOT
].found
&&
1336 (m
->partitions
[PARTITION_ROOT_VERITY
].found
||
1337 m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
))
1338 return -EADDRNOTAVAIL
; /* Verity found but no matching rootfs? Something is off, refuse. */
1340 /* Hmm, we found a signature partition but no Verity data? Something is off. */
1341 if (m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_ROOT_VERITY
].found
)
1342 return -EADDRNOTAVAIL
;
1344 if (!m
->partitions
[PARTITION_USR
].found
&&
1345 (m
->partitions
[PARTITION_USR_VERITY
].found
||
1346 m
->partitions
[PARTITION_USR_VERITY_SIG
].found
))
1347 return -EADDRNOTAVAIL
; /* as above */
1350 if (m
->partitions
[PARTITION_USR_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_USR_VERITY
].found
)
1351 return -EADDRNOTAVAIL
;
1353 /* If root and /usr are combined then insist that the architecture matches */
1354 if (m
->partitions
[PARTITION_ROOT
].found
&&
1355 m
->partitions
[PARTITION_USR
].found
&&
1356 (m
->partitions
[PARTITION_ROOT
].architecture
>= 0 &&
1357 m
->partitions
[PARTITION_USR
].architecture
>= 0 &&
1358 m
->partitions
[PARTITION_ROOT
].architecture
!= m
->partitions
[PARTITION_USR
].architecture
))
1359 return -EADDRNOTAVAIL
;
1361 if (!m
->partitions
[PARTITION_ROOT
].found
&&
1362 !m
->partitions
[PARTITION_USR
].found
&&
1363 (flags
& DISSECT_IMAGE_GENERIC_ROOT
) &&
1364 (!verity
|| !verity
->root_hash
|| verity
->designator
!= PARTITION_USR
)) {
1366 /* OK, we found nothing usable, then check if there's a single generic partition, and use
1367 * that. If the root hash was set however, then we won't fall back to a generic node, because
1368 * the root hash decides. */
1370 /* If we didn't find a properly marked root partition, but we did find a single suitable
1371 * generic Linux partition, then use this as root partition, if the caller asked for it. */
1372 if (multiple_generic
)
1375 /* If we didn't find a generic node, then we can't fix this up either */
1377 r
= image_policy_may_use(policy
, PARTITION_ROOT
);
1381 /* Policy says: ignore; remember that we did */
1382 m
->partitions
[PARTITION_ROOT
].ignored
= true;
1384 _cleanup_close_
int mount_node_fd
= -EBADF
;
1385 _cleanup_free_
char *o
= NULL
, *n
= NULL
;
1386 const char *options
;
1388 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
1389 mount_node_fd
= open_partition(generic_node
, /* is_partition = */ true, m
->loop
);
1390 if (mount_node_fd
< 0)
1391 return mount_node_fd
;
1394 r
= make_partition_devname(devname
, diskseq
, generic_nr
, flags
, &n
);
1398 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
1400 o
= strdup(options
);
1405 assert(generic_nr
>= 0);
1406 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
1409 .growfs
= generic_growfs
,
1410 .partno
= generic_nr
,
1411 .architecture
= _ARCHITECTURE_INVALID
,
1412 .node
= TAKE_PTR(n
),
1413 .uuid
= generic_uuid
,
1414 .mount_options
= TAKE_PTR(o
),
1415 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1416 .offset
= UINT64_MAX
,
1423 /* 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 */
1424 if (FLAGS_SET(flags
, DISSECT_IMAGE_REQUIRE_ROOT
) &&
1425 !(m
->partitions
[PARTITION_ROOT
].found
|| (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
1428 if (m
->partitions
[PARTITION_ROOT_VERITY
].found
) {
1429 /* We only support one verity partition per image, i.e. can't do for both /usr and root fs */
1430 if (m
->partitions
[PARTITION_USR_VERITY
].found
)
1433 /* We don't support verity enabled root with a split out /usr. Neither with nor without
1434 * verity there. (Note that we do support verity-less root with verity-full /usr, though.) */
1435 if (m
->partitions
[PARTITION_USR
].found
)
1436 return -EADDRNOTAVAIL
;
1440 /* If a verity designator is specified, then insist that the matching partition exists */
1441 if (verity
->designator
>= 0 && !m
->partitions
[verity
->designator
].found
)
1442 return -EADDRNOTAVAIL
;
1444 bool have_verity_sig_partition
=
1445 m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR_VERITY_SIG
: PARTITION_ROOT_VERITY_SIG
].found
;
1447 if (verity
->root_hash
) {
1448 /* If we have an explicit root hash and found the partitions for it, then we are ready to use
1449 * Verity, set things up for it */
1451 if (verity
->designator
< 0 || verity
->designator
== PARTITION_ROOT
) {
1452 if (!m
->partitions
[PARTITION_ROOT_VERITY
].found
|| !m
->partitions
[PARTITION_ROOT
].found
)
1453 return -EADDRNOTAVAIL
;
1455 /* If we found a verity setup, then the root partition is necessarily read-only. */
1456 m
->partitions
[PARTITION_ROOT
].rw
= false;
1457 m
->verity_ready
= true;
1460 assert(verity
->designator
== PARTITION_USR
);
1462 if (!m
->partitions
[PARTITION_USR_VERITY
].found
|| !m
->partitions
[PARTITION_USR
].found
)
1463 return -EADDRNOTAVAIL
;
1465 m
->partitions
[PARTITION_USR
].rw
= false;
1466 m
->verity_ready
= true;
1469 if (m
->verity_ready
)
1470 m
->verity_sig_ready
= verity
->root_hash_sig
|| have_verity_sig_partition
;
1472 } else if (have_verity_sig_partition
) {
1474 /* If we found an embedded signature partition, we are ready, too. */
1476 m
->verity_ready
= m
->verity_sig_ready
= true;
1477 m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR
: PARTITION_ROOT
].rw
= false;
1483 /* After we discovered all partitions let's see if the verity requirements match the policy. (Note:
1484 * we don't check encryption requirements here, because we haven't probed the file system yet, hence
1485 * don't know if this is encrypted or not) */
1486 for (PartitionDesignator di
= 0; di
< _PARTITION_DESIGNATOR_MAX
; di
++) {
1487 PartitionDesignator vi
, si
;
1488 PartitionPolicyFlags found_flags
;
1490 any
= any
|| m
->partitions
[di
].found
;
1492 vi
= partition_verity_of(di
);
1493 si
= partition_verity_sig_of(di
);
1495 /* Determine the verity protection level for this partition. */
1496 found_flags
= m
->partitions
[di
].found
?
1497 (vi
>= 0 && m
->partitions
[vi
].found
?
1498 (si
>= 0 && m
->partitions
[si
].found
? PARTITION_POLICY_SIGNED
: PARTITION_POLICY_VERITY
) :
1499 PARTITION_POLICY_ENCRYPTED
|PARTITION_POLICY_UNPROTECTED
) :
1500 (m
->partitions
[di
].ignored
? PARTITION_POLICY_UNUSED
: PARTITION_POLICY_ABSENT
);
1502 r
= image_policy_check_protection(policy
, di
, found_flags
);
1506 if (m
->partitions
[di
].found
) {
1507 r
= image_policy_check_partition_flags(policy
, di
, m
->partitions
[di
].gpt_flags
);
1513 if (!any
&& !FLAGS_SET(flags
, DISSECT_IMAGE_ALLOW_EMPTY
))
1516 r
= dissected_image_probe_filesystems(m
, fd
, policy
);
1524 int dissect_image_file(
1526 const VeritySettings
*verity
,
1527 const MountOptions
*mount_options
,
1528 const ImagePolicy
*image_policy
,
1529 DissectImageFlags flags
,
1530 DissectedImage
**ret
) {
1533 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
1534 _cleanup_close_
int fd
= -EBADF
;
1539 fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
1543 r
= fd_verify_regular(fd
);
1547 r
= dissected_image_new(path
, &m
);
1551 r
= probe_sector_size(fd
, &m
->sector_size
);
1555 r
= dissect_image(m
, fd
, path
, verity
, mount_options
, image_policy
, flags
);
1567 int dissect_log_error(int log_level
, int r
, const char *name
, const VeritySettings
*verity
) {
1568 assert(log_level
>= 0 && log_level
<= LOG_DEBUG
);
1573 case 0 ... INT_MAX
: /* success! */
1577 return log_full_errno(log_level
, r
, "Dissecting images is not supported, compiled without blkid support.");
1580 return log_full_errno(log_level
, r
, "%s: Couldn't identify a suitable partition table or file system.", name
);
1583 return log_full_errno(log_level
, r
, "%s: The image does not pass os-release/extension-release validation.", name
);
1585 case -EADDRNOTAVAIL
:
1586 return log_full_errno(log_level
, r
, "%s: No root partition for specified root hash found.", name
);
1589 return log_full_errno(log_level
, r
, "%s: Multiple suitable root partitions found in image.", name
);
1592 return log_full_errno(log_level
, r
, "%s: No suitable root partition found in image.", name
);
1594 case -EPROTONOSUPPORT
:
1595 return log_full_errno(log_level
, r
, "Device '%s' is a loopback block device with partition scanning turned off, please turn it on.", name
);
1598 return log_full_errno(log_level
, r
, "%s: Image is not a block device.", name
);
1601 return log_full_errno(log_level
, r
,
1602 "Combining partitioned images (such as '%s') with external Verity data (such as '%s') not supported. "
1603 "(Consider setting $SYSTEMD_DISSECT_VERITY_SIDECAR=0 to disable automatic discovery of external Verity data.)",
1604 name
, strna(verity
? verity
->data_path
: NULL
));
1607 return log_full_errno(log_level
, r
, "%s: image does not match image policy.", name
);
1610 return log_full_errno(log_level
, r
, "%s: no suitable partitions found.", name
);
1613 return log_full_errno(log_level
, r
, "%s: cannot dissect image: %m", name
);
1617 int dissect_image_file_and_warn(
1619 const VeritySettings
*verity
,
1620 const MountOptions
*mount_options
,
1621 const ImagePolicy
*image_policy
,
1622 DissectImageFlags flags
,
1623 DissectedImage
**ret
) {
1625 return dissect_log_error(
1627 dissect_image_file(path
, verity
, mount_options
, image_policy
, flags
, ret
),
1632 DissectedImage
* dissected_image_unref(DissectedImage
*m
) {
1636 /* First, clear dissected partitions. */
1637 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++)
1638 dissected_partition_done(m
->partitions
+ i
);
1640 /* Second, free decrypted images. This must be after dissected_partition_done(), as freeing
1641 * DecryptedImage may try to deactivate partitions. */
1642 decrypted_image_unref(m
->decrypted_image
);
1644 /* Third, unref LoopDevice. This must be called after the above two, as freeing LoopDevice may try to
1645 * remove existing partitions on the loopback block device. */
1646 loop_device_unref(m
->loop
);
1648 free(m
->image_name
);
1650 strv_free(m
->machine_info
);
1651 strv_free(m
->os_release
);
1652 strv_free(m
->initrd_release
);
1653 strv_free(m
->extension_release
);
1658 static int is_loop_device(const char *path
) {
1659 char s
[SYS_BLOCK_PATH_MAX("/../loop/")];
1664 if (stat(path
, &st
) < 0)
1667 if (!S_ISBLK(st
.st_mode
))
1670 xsprintf_sys_block_path(s
, "/loop/", st
.st_dev
);
1671 if (access(s
, F_OK
) < 0) {
1672 if (errno
!= ENOENT
)
1675 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
1676 xsprintf_sys_block_path(s
, "/../loop/", st
.st_dev
);
1677 if (access(s
, F_OK
) < 0)
1678 return errno
== ENOENT
? false : -errno
;
1684 static int run_fsck(int node_fd
, const char *fstype
) {
1688 assert(node_fd
>= 0);
1691 r
= fsck_exists_for_fstype(fstype
);
1693 log_debug_errno(r
, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype
);
1697 log_debug("Not checking partition %s, as fsck for %s does not exist.", FORMAT_PROC_FD_PATH(node_fd
), fstype
);
1704 &node_fd
, 1, /* Leave the node fd open */
1705 FORK_RESET_SIGNALS
|FORK_CLOSE_ALL_FDS
|FORK_RLIMIT_NOFILE_SAFE
|FORK_DEATHSIG
|FORK_REARRANGE_STDIO
|FORK_CLOEXEC_OFF
,
1708 return log_debug_errno(r
, "Failed to fork off fsck: %m");
1711 execlp("fsck", "fsck", "-aT", FORMAT_PROC_FD_PATH(node_fd
), NULL
);
1713 log_debug_errno(errno
, "Failed to execl() fsck: %m");
1714 _exit(FSCK_OPERATIONAL_ERROR
);
1717 exit_status
= wait_for_terminate_and_check("fsck", pid
, 0);
1718 if (exit_status
< 0)
1719 return log_debug_errno(exit_status
, "Failed to fork off fsck: %m");
1721 if ((exit_status
& ~FSCK_ERROR_CORRECTED
) != FSCK_SUCCESS
) {
1722 log_debug("fsck failed with exit status %i.", exit_status
);
1724 if ((exit_status
& (FSCK_SYSTEM_SHOULD_REBOOT
|FSCK_ERRORS_LEFT_UNCORRECTED
)) != 0)
1725 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
), "File system is corrupted, refusing.");
1727 log_debug("Ignoring fsck error.");
1733 static int fs_grow(const char *node_path
, const char *mount_path
) {
1734 _cleanup_close_
int mount_fd
= -EBADF
, node_fd
= -EBADF
;
1735 uint64_t size
, newsize
;
1738 node_fd
= open(node_path
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
1740 return log_debug_errno(errno
, "Failed to open node device %s: %m", node_path
);
1742 if (ioctl(node_fd
, BLKGETSIZE64
, &size
) != 0)
1743 return log_debug_errno(errno
, "Failed to get block device size of %s: %m", node_path
);
1745 mount_fd
= open(mount_path
, O_RDONLY
|O_DIRECTORY
|O_CLOEXEC
);
1747 return log_debug_errno(errno
, "Failed to open mountd file system %s: %m", mount_path
);
1749 log_debug("Resizing \"%s\" to %"PRIu64
" bytes...", mount_path
, size
);
1750 r
= resize_fs(mount_fd
, size
, &newsize
);
1752 return log_debug_errno(r
, "Failed to resize \"%s\" to %"PRIu64
" bytes: %m", mount_path
, size
);
1754 if (newsize
== size
)
1755 log_debug("Successfully resized \"%s\" to %s bytes.",
1756 mount_path
, FORMAT_BYTES(newsize
));
1758 assert(newsize
< size
);
1759 log_debug("Successfully resized \"%s\" to %s bytes (%"PRIu64
" bytes lost due to blocksize).",
1760 mount_path
, FORMAT_BYTES(newsize
), size
- newsize
);
1766 int partition_pick_mount_options(
1767 PartitionDesignator d
,
1772 unsigned long *ret_ms_flags
) {
1774 _cleanup_free_
char *options
= NULL
;
1776 assert(ret_options
);
1778 /* Selects a baseline of bind mount flags, that should always apply.
1780 * Firstly, we set MS_NODEV universally on all mounts, since we don't want to allow device nodes outside of /dev/.
1782 * On /var/tmp/ we'll also set MS_NOSUID, same as we set for /tmp/ on the host.
1784 * On the ESP and XBOOTLDR partitions we'll also disable symlinks, and execution. These file systems
1785 * are generally untrusted (i.e. not encrypted or authenticated), and typically VFAT hence we should
1786 * be as restrictive as possible, and this shouldn't hurt, since the functionality is not available
1789 unsigned long flags
= MS_NODEV
;
1797 case PARTITION_XBOOTLDR
:
1798 flags
|= MS_NOSUID
|MS_NOEXEC
|ms_nosymfollow_supported();
1800 /* The ESP might contain a pre-boot random seed. Let's make this unaccessible to regular
1801 * userspace. ESP/XBOOTLDR is almost certainly VFAT, hence if we don't know assume it is. */
1802 if (!fstype
|| fstype_can_umask(fstype
))
1803 if (!strextend_with_separator(&options
, ",", "umask=0077"))
1815 /* So, when you request MS_RDONLY from ext4, then this means nothing. It happily still writes to the
1816 * backing storage. What's worse, the BLKRO[GS]ET flag and (in case of loopback devices)
1817 * LO_FLAGS_READ_ONLY don't mean anything, they affect userspace accesses only, and write accesses
1818 * from the upper file system still get propagated through to the underlying file system,
1819 * unrestricted. To actually get ext4/xfs/btrfs to stop writing to the device we need to specify
1820 * "norecovery" as mount option, in addition to MS_RDONLY. Yes, this sucks, since it means we need to
1821 * carry a per file system table here.
1823 * Note that this means that we might not be able to mount corrupted file systems as read-only
1824 * anymore (since in some cases the kernel implementations will refuse mounting when corrupted,
1825 * read-only and "norecovery" is specified). But I think for the case of automatically determined
1826 * mount options for loopback devices this is the right choice, since otherwise using the same
1827 * loopback file twice even in read-only mode, is going to fail badly sooner or later. The usecase of
1828 * making reuse of the immutable images "just work" is more relevant to us than having read-only
1829 * access that actually modifies stuff work on such image files. Or to say this differently: if
1830 * people want their file systems to be fixed up they should just open them in writable mode, where
1831 * all these problems don't exist. */
1832 if (!rw
&& fstype
&& fstype_can_norecovery(fstype
))
1833 if (!strextend_with_separator(&options
, ",", "norecovery"))
1836 if (discard
&& fstype
&& fstype_can_discard(fstype
))
1837 if (!strextend_with_separator(&options
, ",", "discard"))
1840 if (!ret_ms_flags
) /* Fold flags into option string if ret_flags specified as NULL */
1841 if (!strextend_with_separator(&options
, ",",
1842 FLAGS_SET(flags
, MS_RDONLY
) ? "ro" : "rw",
1843 FLAGS_SET(flags
, MS_NODEV
) ? "nodev" : "dev",
1844 FLAGS_SET(flags
, MS_NOSUID
) ? "nosuid" : "suid",
1845 FLAGS_SET(flags
, MS_NOEXEC
) ? "noexec" : "exec",
1846 FLAGS_SET(flags
, MS_NOSYMFOLLOW
) ? "nosymfollow" : NULL
))
1847 /* NB: we suppress 'symfollow' here, since it's the default, and old /bin/mount might not know it */
1851 *ret_ms_flags
= flags
;
1853 *ret_options
= TAKE_PTR(options
);
1857 static int mount_partition(
1858 PartitionDesignator d
,
1859 DissectedPartition
*m
,
1861 const char *directory
,
1864 DissectImageFlags flags
) {
1866 _cleanup_free_
char *chased
= NULL
, *options
= NULL
;
1867 bool rw
, discard
, remap_uid_gid
= false;
1868 const char *p
, *node
, *fstype
;
1869 unsigned long ms_flags
;
1875 if (m
->mount_node_fd
< 0)
1878 /* Use decrypted node and matching fstype if available, otherwise use the original device */
1879 node
= FORMAT_PROC_FD_PATH(m
->mount_node_fd
);
1880 fstype
= dissected_partition_fstype(m
);
1883 return -EAFNOSUPPORT
;
1885 /* We are looking at an encrypted partition? This either means stacked encryption, or the caller
1886 * didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error for this
1888 if (streq(fstype
, "crypto_LUKS"))
1891 r
= dissect_fstype_ok(fstype
);
1895 return -EIDRM
; /* Recognizable error */
1897 rw
= m
->rw
&& !(flags
& DISSECT_IMAGE_MOUNT_READ_ONLY
);
1899 discard
= ((flags
& DISSECT_IMAGE_DISCARD
) ||
1900 ((flags
& DISSECT_IMAGE_DISCARD_ON_LOOP
) && is_loop_device(m
->node
) > 0));
1902 if (FLAGS_SET(flags
, DISSECT_IMAGE_FSCK
) && rw
) {
1903 r
= run_fsck(m
->mount_node_fd
, fstype
);
1909 /* Automatically create missing mount points inside the image, if necessary. */
1910 r
= mkdir_p_root(where
, directory
, uid_shift
, (gid_t
) uid_shift
, 0755, NULL
);
1911 if (r
< 0 && r
!= -EROFS
)
1914 r
= chase(directory
, where
, CHASE_PREFIX_ROOT
, &chased
, NULL
);
1920 /* Create top-level mount if missing – but only if this is asked for. This won't modify the
1921 * image (as the branch above does) but the host hierarchy, and the created directory might
1922 * survive our mount in the host hierarchy hence. */
1923 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
1924 r
= mkdir_p(where
, 0755);
1932 r
= partition_pick_mount_options(d
, dissected_partition_fstype(m
), rw
, discard
, &options
, &ms_flags
);
1936 if (uid_is_valid(uid_shift
) && uid_shift
!= 0) {
1938 if (fstype_can_uid_gid(fstype
)) {
1939 _cleanup_free_
char *uid_option
= NULL
;
1941 if (asprintf(&uid_option
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
1944 if (!strextend_with_separator(&options
, ",", uid_option
))
1946 } else if (FLAGS_SET(flags
, DISSECT_IMAGE_MOUNT_IDMAPPED
))
1947 remap_uid_gid
= true;
1950 if (!isempty(m
->mount_options
))
1951 if (!strextend_with_separator(&options
, ",", m
->mount_options
))
1954 r
= mount_nofollow_verbose(LOG_DEBUG
, node
, p
, fstype
, ms_flags
, options
);
1958 if (rw
&& m
->growfs
&& FLAGS_SET(flags
, DISSECT_IMAGE_GROWFS
))
1959 (void) fs_grow(node
, p
);
1961 if (remap_uid_gid
) {
1962 r
= remount_idmap(p
, uid_shift
, uid_range
, UID_INVALID
, REMOUNT_IDMAPPING_HOST_ROOT
);
1970 static int mount_root_tmpfs(const char *where
, uid_t uid_shift
, DissectImageFlags flags
) {
1971 _cleanup_free_
char *options
= NULL
;
1976 /* For images that contain /usr/ but no rootfs, let's mount rootfs as tmpfs */
1978 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
1979 r
= mkdir_p(where
, 0755);
1984 if (uid_is_valid(uid_shift
)) {
1985 if (asprintf(&options
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
1989 r
= mount_nofollow_verbose(LOG_DEBUG
, "rootfs", where
, "tmpfs", MS_NODEV
, options
);
1996 static int mount_point_is_available(const char *where
, const char *path
, bool missing_ok
) {
1997 _cleanup_free_
char *p
= NULL
;
2000 /* Check whether <path> is suitable as a mountpoint, i.e. is an empty directory
2001 * or does not exist at all (when missing_ok). */
2003 r
= chase(path
, where
, CHASE_PREFIX_ROOT
, &p
, NULL
);
2007 return log_debug_errno(r
, "Failed to chase \"%s\": %m", path
);
2009 r
= dir_is_empty(p
, /* ignore_hidden_or_backup= */ false);
2013 return log_debug_errno(r
, "Failed to check directory \"%s\": %m", p
);
2017 int dissected_image_mount(
2022 DissectImageFlags flags
) {
2031 * -ENXIO → No root partition found
2032 * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release/extension-release file found
2033 * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet
2034 * -EUCLEAN → fsck for file system failed
2035 * -EBUSY → File system already mounted/used elsewhere (kernel)
2036 * -EAFNOSUPPORT → File system type not supported or not known
2037 * -EIDRM → File system is not among allowlisted "common" file systems
2040 if (!(m
->partitions
[PARTITION_ROOT
].found
||
2041 (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
2042 return -ENXIO
; /* Require a root fs or at least a /usr/ fs (the latter is subject to a flag of its own) */
2044 if ((flags
& DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
) == 0) {
2046 /* First mount the root fs. If there's none we use a tmpfs. */
2047 if (m
->partitions
[PARTITION_ROOT
].found
)
2048 r
= mount_partition(PARTITION_ROOT
, m
->partitions
+ PARTITION_ROOT
, where
, NULL
, uid_shift
, uid_range
, flags
);
2050 r
= mount_root_tmpfs(where
, uid_shift
, flags
);
2054 /* For us mounting root always means mounting /usr as well */
2055 r
= mount_partition(PARTITION_USR
, m
->partitions
+ PARTITION_USR
, where
, "/usr", uid_shift
, uid_range
, flags
);
2059 if ((flags
& (DISSECT_IMAGE_VALIDATE_OS
|DISSECT_IMAGE_VALIDATE_OS_EXT
)) != 0) {
2060 /* If either one of the validation flags are set, ensure that the image qualifies
2061 * as one or the other (or both). */
2064 if (FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS
)) {
2065 r
= path_is_os_tree(where
);
2071 if (!ok
&& FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS_EXT
)) {
2072 r
= extension_has_forbidden_content(where
);
2076 r
= path_is_extension_tree(IMAGE_SYSEXT
, where
, m
->image_name
, FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_EXTENSION_CHECK
));
2078 r
= path_is_extension_tree(IMAGE_CONFEXT
, where
, m
->image_name
, FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_EXTENSION_CHECK
));
2091 if (flags
& DISSECT_IMAGE_MOUNT_ROOT_ONLY
)
2094 r
= mount_partition(PARTITION_HOME
, m
->partitions
+ PARTITION_HOME
, where
, "/home", uid_shift
, uid_range
, flags
);
2098 r
= mount_partition(PARTITION_SRV
, m
->partitions
+ PARTITION_SRV
, where
, "/srv", uid_shift
, uid_range
, flags
);
2102 r
= mount_partition(PARTITION_VAR
, m
->partitions
+ PARTITION_VAR
, where
, "/var", uid_shift
, uid_range
, flags
);
2106 r
= mount_partition(PARTITION_TMP
, m
->partitions
+ PARTITION_TMP
, where
, "/var/tmp", uid_shift
, uid_range
, flags
);
2110 int slash_boot_is_available
;
2111 r
= slash_boot_is_available
= mount_point_is_available(where
, "/boot", /* missing_ok = */ true);
2115 r
= mount_partition(PARTITION_XBOOTLDR
, m
->partitions
+ PARTITION_XBOOTLDR
, where
, "/boot", uid_shift
, uid_range
, flags
);
2118 slash_boot_is_available
= !r
;
2121 if (m
->partitions
[PARTITION_ESP
].found
) {
2122 const char *esp_path
= NULL
;
2124 /* Mount the ESP to /boot/ if it exists and is empty and we didn't already mount the XBOOTLDR
2125 * partition into it. Otherwise, use /efi instead, but only if it exists and is empty. */
2127 if (slash_boot_is_available
) {
2128 r
= mount_point_is_available(where
, "/boot", /* missing_ok = */ false);
2136 r
= mount_point_is_available(where
, "/efi", /* missing_ok = */ true);
2144 /* OK, let's mount the ESP now (possibly creating the dir if missing) */
2145 r
= mount_partition(PARTITION_ESP
, m
->partitions
+ PARTITION_ESP
, where
, esp_path
, uid_shift
, uid_range
, flags
);
2154 int dissected_image_mount_and_warn(
2159 DissectImageFlags flags
) {
2166 r
= dissected_image_mount(m
, where
, uid_shift
, uid_range
, flags
);
2168 return log_error_errno(r
, "Not root file system found in image.");
2169 if (r
== -EMEDIUMTYPE
)
2170 return log_error_errno(r
, "No suitable os-release/extension-release file in image found.");
2172 return log_error_errno(r
, "Encrypted file system discovered, but decryption not requested.");
2174 return log_error_errno(r
, "File system check on image failed.");
2176 return log_error_errno(r
, "File system already mounted elsewhere.");
2177 if (r
== -EAFNOSUPPORT
)
2178 return log_error_errno(r
, "File system type not supported or not known.");
2180 return log_error_errno(r
, "File system is too uncommon, refused.");
2182 return log_error_errno(r
, "Failed to mount image: %m");
2187 #if HAVE_LIBCRYPTSETUP
2188 struct DecryptedPartition
{
2189 struct crypt_device
*device
;
2195 typedef struct DecryptedPartition DecryptedPartition
;
2197 struct DecryptedImage
{
2199 DecryptedPartition
*decrypted
;
2203 static DecryptedImage
* decrypted_image_free(DecryptedImage
*d
) {
2204 #if HAVE_LIBCRYPTSETUP
2210 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
2211 DecryptedPartition
*p
= d
->decrypted
+ i
;
2213 if (p
->device
&& p
->name
&& !p
->relinquished
) {
2214 _cleanup_free_
char *node
= NULL
;
2216 node
= path_join("/dev/mapper", p
->name
);
2218 r
= btrfs_forget_device(node
);
2219 if (r
< 0 && r
!= -ENOENT
)
2220 log_debug_errno(r
, "Failed to forget btrfs device %s, ignoring: %m", node
);
2224 /* Let's deactivate lazily, as the dm volume may be already/still used by other processes. */
2225 r
= sym_crypt_deactivate_by_name(p
->device
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
2227 log_debug_errno(r
, "Failed to deactivate encrypted partition %s", p
->name
);
2231 sym_crypt_free(p
->device
);
2241 DEFINE_TRIVIAL_REF_UNREF_FUNC(DecryptedImage
, decrypted_image
, decrypted_image_free
);
2243 #if HAVE_LIBCRYPTSETUP
2244 static int decrypted_image_new(DecryptedImage
**ret
) {
2245 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
2249 d
= new(DecryptedImage
, 1);
2253 *d
= (DecryptedImage
) {
2261 static int make_dm_name_and_node(const void *original_node
, const char *suffix
, char **ret_name
, char **ret_node
) {
2262 _cleanup_free_
char *name
= NULL
, *node
= NULL
;
2265 assert(original_node
);
2270 base
= strrchr(original_node
, '/');
2272 base
= original_node
;
2278 name
= strjoin(base
, suffix
);
2281 if (!filename_is_valid(name
))
2284 node
= path_join(sym_crypt_get_dir(), name
);
2288 *ret_name
= TAKE_PTR(name
);
2289 *ret_node
= TAKE_PTR(node
);
2294 static int decrypt_partition(
2295 DissectedPartition
*m
,
2296 const char *passphrase
,
2297 DissectImageFlags flags
,
2298 DecryptedImage
*d
) {
2300 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
2301 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2302 _cleanup_close_
int fd
= -EBADF
;
2308 if (!m
->found
|| !m
->node
|| !m
->fstype
)
2311 if (!streq(m
->fstype
, "crypto_LUKS"))
2317 r
= dlopen_cryptsetup();
2321 r
= make_dm_name_and_node(m
->node
, "-decrypted", &name
, &node
);
2325 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
2328 r
= sym_crypt_init(&cd
, m
->node
);
2330 return log_debug_errno(r
, "Failed to initialize dm-crypt: %m");
2332 cryptsetup_enable_logging(cd
);
2334 r
= sym_crypt_load(cd
, CRYPT_LUKS
, NULL
);
2336 return log_debug_errno(r
, "Failed to load LUKS metadata: %m");
2338 r
= sym_crypt_activate_by_passphrase(cd
, name
, CRYPT_ANY_SLOT
, passphrase
, strlen(passphrase
),
2339 ((flags
& DISSECT_IMAGE_DEVICE_READ_ONLY
) ? CRYPT_ACTIVATE_READONLY
: 0) |
2340 ((flags
& DISSECT_IMAGE_DISCARD_ON_CRYPTO
) ? CRYPT_ACTIVATE_ALLOW_DISCARDS
: 0));
2342 log_debug_errno(r
, "Failed to activate LUKS device: %m");
2343 return r
== -EPERM
? -EKEYREJECTED
: r
;
2346 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2348 return log_debug_errno(errno
, "Failed to open %s: %m", node
);
2350 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
2351 .name
= TAKE_PTR(name
),
2352 .device
= TAKE_PTR(cd
),
2355 m
->decrypted_node
= TAKE_PTR(node
);
2356 close_and_replace(m
->mount_node_fd
, fd
);
2361 static int verity_can_reuse(
2362 const VeritySettings
*verity
,
2364 struct crypt_device
**ret_cd
) {
2366 /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
2367 _cleanup_free_
char *root_hash_existing
= NULL
;
2368 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2369 struct crypt_params_verity crypt_params
= {};
2370 size_t root_hash_existing_size
;
2377 r
= sym_crypt_init_by_name(&cd
, name
);
2379 return log_debug_errno(r
, "Error opening verity device, crypt_init_by_name failed: %m");
2381 cryptsetup_enable_logging(cd
);
2383 r
= sym_crypt_get_verity_info(cd
, &crypt_params
);
2385 return log_debug_errno(r
, "Error opening verity device, crypt_get_verity_info failed: %m");
2387 root_hash_existing_size
= verity
->root_hash_size
;
2388 root_hash_existing
= malloc0(root_hash_existing_size
);
2389 if (!root_hash_existing
)
2392 r
= sym_crypt_volume_key_get(cd
, CRYPT_ANY_SLOT
, root_hash_existing
, &root_hash_existing_size
, NULL
, 0);
2394 return log_debug_errno(r
, "Error opening verity device, crypt_volume_key_get failed: %m");
2395 if (verity
->root_hash_size
!= root_hash_existing_size
||
2396 memcmp(root_hash_existing
, verity
->root_hash
, verity
->root_hash_size
) != 0)
2397 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but root hashes are different.");
2399 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
2400 /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the
2401 * same settings, so that a previous unsigned mount will not be reused if the user asks to use
2402 * signing for the new one, and vice versa. */
2403 if (!!verity
->root_hash_sig
!= !!(crypt_params
.flags
& CRYPT_VERITY_ROOT_HASH_SIGNATURE
))
2404 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but signature settings are not the same.");
2407 *ret_cd
= TAKE_PTR(cd
);
2411 static char* dm_deferred_remove_clean(char *name
) {
2415 (void) sym_crypt_deactivate_by_name(NULL
, name
, CRYPT_DEACTIVATE_DEFERRED
);
2418 DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean
);
2420 static int validate_signature_userspace(const VeritySettings
*verity
) {
2422 _cleanup_(sk_X509_free_allp
) STACK_OF(X509
) *sk
= NULL
;
2423 _cleanup_strv_free_
char **certs
= NULL
;
2424 _cleanup_(PKCS7_freep
) PKCS7
*p7
= NULL
;
2425 _cleanup_free_
char *s
= NULL
;
2426 _cleanup_(BIO_freep
) BIO
*bio
= NULL
; /* 'bio' must be freed first, 's' second, hence keep this order
2427 * of declaration in place, please */
2428 const unsigned char *d
;
2432 assert(verity
->root_hash
);
2433 assert(verity
->root_hash_sig
);
2435 /* Because installing a signature certificate into the kernel chain is so messy, let's optionally do
2436 * userspace validation. */
2438 r
= conf_files_list_nulstr(&certs
, ".crt", NULL
, CONF_FILES_REGULAR
|CONF_FILES_FILTER_MASKED
, CONF_PATHS_NULSTR("verity.d"));
2440 return log_debug_errno(r
, "Failed to enumerate certificates: %m");
2441 if (strv_isempty(certs
)) {
2442 log_debug("No userspace dm-verity certificates found.");
2446 d
= verity
->root_hash_sig
;
2447 p7
= d2i_PKCS7(NULL
, &d
, (long) verity
->root_hash_sig_size
);
2449 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to parse PKCS7 DER signature data.");
2451 s
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2453 return log_oom_debug();
2455 bio
= BIO_new_mem_buf(s
, strlen(s
));
2457 return log_oom_debug();
2459 sk
= sk_X509_new_null();
2461 return log_oom_debug();
2463 STRV_FOREACH(i
, certs
) {
2464 _cleanup_(X509_freep
) X509
*c
= NULL
;
2465 _cleanup_fclose_
FILE *f
= NULL
;
2467 f
= fopen(*i
, "re");
2469 log_debug_errno(errno
, "Failed to open '%s', ignoring: %m", *i
);
2473 c
= PEM_read_X509(f
, NULL
, NULL
, NULL
);
2475 log_debug("Failed to load X509 certificate '%s', ignoring.", *i
);
2479 if (sk_X509_push(sk
, c
) == 0)
2480 return log_oom_debug();
2485 r
= PKCS7_verify(p7
, sk
, NULL
, bio
, NULL
, PKCS7_NOINTERN
|PKCS7_NOVERIFY
);
2487 log_debug("Userspace PKCS#7 validation succeeded.");
2489 log_debug("Userspace PKCS#7 validation failed: %s", ERR_error_string(ERR_get_error(), NULL
));
2493 log_debug("Not doing client-side validation of dm-verity root hash signatures, OpenSSL support disabled.");
2498 static int do_crypt_activate_verity(
2499 struct crypt_device
*cd
,
2501 const VeritySettings
*verity
) {
2503 bool check_signature
;
2510 if (verity
->root_hash_sig
) {
2511 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIGNATURE");
2512 if (r
< 0 && r
!= -ENXIO
)
2513 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIGNATURE");
2515 check_signature
= r
!= 0;
2517 check_signature
= false;
2519 if (check_signature
) {
2521 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
2522 /* First, if we have support for signed keys in the kernel, then try that first. */
2523 r
= sym_crypt_activate_by_signed_key(
2527 verity
->root_hash_size
,
2528 verity
->root_hash_sig
,
2529 verity
->root_hash_sig_size
,
2530 CRYPT_ACTIVATE_READONLY
);
2534 log_debug("Validation of dm-verity signature failed via the kernel, trying userspace validation instead.");
2536 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.",
2537 program_invocation_short_name
);
2540 /* So this didn't work via the kernel, then let's try userspace validation instead. If that
2541 * works we'll try to activate without telling the kernel the signature. */
2543 r
= validate_signature_userspace(verity
);
2547 return log_debug_errno(SYNTHETIC_ERRNO(ENOKEY
),
2548 "Activation of signed Verity volume worked neither via the kernel nor in userspace, can't activate.");
2551 return sym_crypt_activate_by_volume_key(
2555 verity
->root_hash_size
,
2556 CRYPT_ACTIVATE_READONLY
);
2559 static usec_t
verity_timeout(void) {
2560 usec_t t
= 100 * USEC_PER_MSEC
;
2564 /* On slower machines, like non-KVM vm, setting up device may take a long time.
2565 * Let's make the timeout configurable. */
2567 e
= getenv("SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC");
2571 r
= parse_sec(e
, &t
);
2574 "Failed to parse timeout specified in $SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC, "
2575 "using the default timeout (%s).",
2576 FORMAT_TIMESPAN(t
, USEC_PER_MSEC
));
2581 static int verity_partition(
2582 PartitionDesignator designator
,
2583 DissectedPartition
*m
,
2584 DissectedPartition
*v
,
2585 const VeritySettings
*verity
,
2586 DissectImageFlags flags
,
2587 DecryptedImage
*d
) {
2589 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2590 _cleanup_(dm_deferred_remove_cleanp
) char *restore_deferred_remove
= NULL
;
2591 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
2592 _cleanup_close_
int mount_node_fd
= -EBADF
;
2596 assert(v
|| (verity
&& verity
->data_path
));
2598 if (!verity
|| !verity
->root_hash
)
2600 if (!((verity
->designator
< 0 && designator
== PARTITION_ROOT
) ||
2601 (verity
->designator
== designator
)))
2604 if (!m
->found
|| !m
->node
|| !m
->fstype
)
2606 if (!verity
->data_path
) {
2607 if (!v
->found
|| !v
->node
|| !v
->fstype
)
2610 if (!streq(v
->fstype
, "DM_verity_hash"))
2614 r
= dlopen_cryptsetup();
2618 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
2619 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
2620 _cleanup_free_
char *root_hash_encoded
= NULL
;
2622 root_hash_encoded
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2623 if (!root_hash_encoded
)
2626 r
= make_dm_name_and_node(root_hash_encoded
, "-verity", &name
, &node
);
2628 r
= make_dm_name_and_node(m
->node
, "-verity", &name
, &node
);
2632 r
= sym_crypt_init(&cd
, verity
->data_path
?: v
->node
);
2636 cryptsetup_enable_logging(cd
);
2638 r
= sym_crypt_load(cd
, CRYPT_VERITY
, NULL
);
2642 r
= sym_crypt_set_data_device(cd
, m
->node
);
2646 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
2649 /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
2650 * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
2651 * retry a few times before giving up. */
2652 for (unsigned i
= 0; i
< N_DEVICE_NODE_LIST_ATTEMPTS
; i
++) {
2653 _cleanup_(sym_crypt_freep
) struct crypt_device
*existing_cd
= NULL
;
2654 _cleanup_close_
int fd
= -EBADF
;
2656 /* First, check if the device already exists. */
2657 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2658 if (fd
< 0 && !ERRNO_IS_DEVICE_ABSENT(errno
))
2659 return log_debug_errno(errno
, "Failed to open verity device %s: %m", node
);
2661 goto check
; /* The device already exists. Let's check it. */
2663 /* The symlink to the device node does not exist yet. Assume not activated, and let's activate it. */
2664 r
= do_crypt_activate_verity(cd
, name
, verity
);
2666 goto try_open
; /* The device is activated. Let's open it. */
2667 /* libdevmapper can return EINVAL when the device is already in the activation stage.
2668 * There's no way to distinguish this situation from a genuine error due to invalid
2669 * parameters, so immediately fall back to activating the device with a unique name.
2670 * Improvements in libcrypsetup can ensure this never happens:
2671 * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
2672 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2674 if (r
== -ENODEV
) /* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */
2677 -EEXIST
, /* Volume has already been opened and ready to be used. */
2678 -EBUSY
/* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */))
2679 return log_debug_errno(r
, "Failed to activate verity device %s: %m", node
);
2682 if (!restore_deferred_remove
){
2683 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
2684 r
= dm_deferred_remove_cancel(name
);
2685 /* -EBUSY and -ENXIO: the device has already been removed or being removed. We cannot
2686 * use the device, try to open again. See target_message() in drivers/md/dm-ioctl.c
2687 * and dm_cancel_deferred_remove() in drivers/md/dm.c */
2688 if (IN_SET(r
, -EBUSY
, -ENXIO
))
2691 return log_debug_errno(r
, "Failed to disable automated deferred removal for verity device %s: %m", node
);
2693 restore_deferred_remove
= strdup(name
);
2694 if (!restore_deferred_remove
)
2695 return log_oom_debug();
2698 r
= verity_can_reuse(verity
, name
, &existing_cd
);
2699 /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
2700 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2703 -ENOENT
, /* Removed?? */
2704 -EBUSY
, /* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */
2705 -ENODEV
/* Volume is being opened but not ready, crypt_init_by_name() would fail, try to open again. */ ))
2708 return log_debug_errno(r
, "Failed to check if existing verity device %s can be reused: %m", node
);
2711 /* devmapper might say that the device exists, but the devlink might not yet have been
2712 * created. Check and wait for the udev event in that case. */
2713 r
= device_wait_for_devlink(node
, "block", verity_timeout(), NULL
);
2714 /* Fallback to activation with a unique device if it's taking too long */
2715 if (r
== -ETIMEDOUT
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2718 return log_debug_errno(r
, "Failed to wait device node symlink %s: %m", node
);
2723 /* Now, the device is activated and devlink is created. Let's open it. */
2724 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2726 if (!ERRNO_IS_DEVICE_ABSENT(errno
))
2727 return log_debug_errno(errno
, "Failed to open verity device %s: %m", node
);
2729 /* The device has already been removed?? */
2734 mount_node_fd
= TAKE_FD(fd
);
2736 crypt_free_and_replace(cd
, existing_cd
);
2741 /* Device is being removed by another process. Let's wait for a while. */
2742 (void) usleep_safe(2 * USEC_PER_MSEC
);
2745 /* All trials failed or a conflicting verity device exists. Let's try to activate with a unique name. */
2746 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
2747 /* Before trying to activate with unique name, we need to free crypt_device object.
2748 * Otherwise, we get error from libcryptsetup like the following:
2750 * systemd[1234]: Cannot use device /dev/loop5 which is in use (already mapped or mounted).
2755 return verity_partition(designator
, m
, v
, verity
, flags
& ~DISSECT_IMAGE_VERITY_SHARE
, d
);
2758 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
), "All attempts to activate verity device %s failed.", name
);
2761 /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
2762 restore_deferred_remove
= mfree(restore_deferred_remove
);
2764 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
2765 .name
= TAKE_PTR(name
),
2766 .device
= TAKE_PTR(cd
),
2769 m
->decrypted_node
= TAKE_PTR(node
);
2770 close_and_replace(m
->mount_node_fd
, mount_node_fd
);
2776 int dissected_image_decrypt(
2778 const char *passphrase
,
2779 const VeritySettings
*verity
,
2780 DissectImageFlags flags
) {
2782 #if HAVE_LIBCRYPTSETUP
2783 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
2788 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
2792 * = 0 → There was nothing to decrypt
2793 * > 0 → Decrypted successfully
2794 * -ENOKEY → There's something to decrypt but no key was supplied
2795 * -EKEYREJECTED → Passed key was not correct
2798 if (verity
&& verity
->root_hash
&& verity
->root_hash_size
< sizeof(sd_id128_t
))
2801 if (!m
->encrypted
&& !m
->verity_ready
)
2804 #if HAVE_LIBCRYPTSETUP
2805 r
= decrypted_image_new(&d
);
2809 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
2810 DissectedPartition
*p
= m
->partitions
+ i
;
2811 PartitionDesignator k
;
2816 r
= decrypt_partition(p
, passphrase
, flags
, d
);
2820 k
= partition_verity_of(i
);
2822 r
= verity_partition(i
, p
, m
->partitions
+ k
, verity
, flags
| DISSECT_IMAGE_VERITY_SHARE
, d
);
2827 if (!p
->decrypted_fstype
&& p
->mount_node_fd
>= 0 && p
->decrypted_node
) {
2828 r
= probe_filesystem_full(p
->mount_node_fd
, p
->decrypted_node
, 0, UINT64_MAX
, &p
->decrypted_fstype
);
2829 if (r
< 0 && r
!= -EUCLEAN
)
2834 m
->decrypted_image
= TAKE_PTR(d
);
2842 int dissected_image_decrypt_interactively(
2844 const char *passphrase
,
2845 const VeritySettings
*verity
,
2846 DissectImageFlags flags
) {
2848 _cleanup_strv_free_erase_
char **z
= NULL
;
2855 r
= dissected_image_decrypt(m
, passphrase
, verity
, flags
);
2858 if (r
== -EKEYREJECTED
)
2859 log_error_errno(r
, "Incorrect passphrase, try again!");
2860 else if (r
!= -ENOKEY
)
2861 return log_error_errno(r
, "Failed to decrypt image: %m");
2864 return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED
),
2865 "Too many retries.");
2869 r
= ask_password_auto("Please enter image passphrase:", NULL
, "dissect", "dissect", "dissect.passphrase", USEC_INFINITY
, 0, &z
);
2871 return log_error_errno(r
, "Failed to query for passphrase: %m");
2877 static int decrypted_image_relinquish(DecryptedImage
*d
) {
2880 /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a
2881 * boolean so that we don't clean it up ourselves either anymore */
2883 #if HAVE_LIBCRYPTSETUP
2886 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
2887 DecryptedPartition
*p
= d
->decrypted
+ i
;
2889 if (p
->relinquished
)
2892 r
= sym_crypt_deactivate_by_name(NULL
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
2894 return log_debug_errno(r
, "Failed to mark %s for auto-removal: %m", p
->name
);
2896 p
->relinquished
= true;
2903 int dissected_image_relinquish(DissectedImage
*m
) {
2908 if (m
->decrypted_image
) {
2909 r
= decrypted_image_relinquish(m
->decrypted_image
);
2915 loop_device_relinquish(m
->loop
);
2920 static char *build_auxiliary_path(const char *image
, const char *suffix
) {
2927 e
= endswith(image
, ".raw");
2929 return strjoin(e
, suffix
);
2931 n
= new(char, e
- image
+ strlen(suffix
) + 1);
2935 strcpy(mempcpy(n
, image
, e
- image
), suffix
);
2939 void verity_settings_done(VeritySettings
*v
) {
2942 v
->root_hash
= mfree(v
->root_hash
);
2943 v
->root_hash_size
= 0;
2945 v
->root_hash_sig
= mfree(v
->root_hash_sig
);
2946 v
->root_hash_sig_size
= 0;
2948 v
->data_path
= mfree(v
->data_path
);
2951 int verity_settings_load(
2952 VeritySettings
*verity
,
2954 const char *root_hash_path
,
2955 const char *root_hash_sig_path
) {
2957 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
2958 size_t root_hash_size
= 0, root_hash_sig_size
= 0;
2959 _cleanup_free_
char *verity_data_path
= NULL
;
2960 PartitionDesignator designator
;
2965 assert(verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
2967 /* If we are asked to load the root hash for a device node, exit early */
2968 if (is_device_path(image
))
2971 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIDECAR");
2972 if (r
< 0 && r
!= -ENXIO
)
2973 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIDECAR, ignoring: %m");
2977 designator
= verity
->designator
;
2979 /* We only fill in what isn't already filled in */
2981 if (!verity
->root_hash
) {
2982 _cleanup_free_
char *text
= NULL
;
2984 if (root_hash_path
) {
2985 /* If explicitly specified it takes precedence */
2986 r
= read_one_line_file(root_hash_path
, &text
);
2991 designator
= PARTITION_ROOT
;
2993 /* Otherwise look for xattr and separate file, and first for the data for root and if
2994 * that doesn't exist for /usr */
2996 if (designator
< 0 || designator
== PARTITION_ROOT
) {
2997 r
= getxattr_malloc(image
, "user.verity.roothash", &text
);
2999 _cleanup_free_
char *p
= NULL
;
3001 if (r
!= -ENOENT
&& !ERRNO_IS_XATTR_ABSENT(r
))
3004 p
= build_auxiliary_path(image
, ".roothash");
3008 r
= read_one_line_file(p
, &text
);
3009 if (r
< 0 && r
!= -ENOENT
)
3014 designator
= PARTITION_ROOT
;
3017 if (!text
&& (designator
< 0 || designator
== PARTITION_USR
)) {
3018 /* So in the "roothash" xattr/file name above the "root" of course primarily
3019 * refers to the root of the Verity Merkle tree. But coincidentally it also
3020 * is the hash for the *root* file system, i.e. the "root" neatly refers to
3021 * two distinct concepts called "root". Taking benefit of this happy
3022 * coincidence we call the file with the root hash for the /usr/ file system
3023 * `usrhash`, because `usrroothash` or `rootusrhash` would just be too
3024 * confusing. We thus drop the reference to the root of the Merkle tree, and
3025 * just indicate which file system it's about. */
3026 r
= getxattr_malloc(image
, "user.verity.usrhash", &text
);
3028 _cleanup_free_
char *p
= NULL
;
3030 if (r
!= -ENOENT
&& !ERRNO_IS_XATTR_ABSENT(r
))
3033 p
= build_auxiliary_path(image
, ".usrhash");
3037 r
= read_one_line_file(p
, &text
);
3038 if (r
< 0 && r
!= -ENOENT
)
3043 designator
= PARTITION_USR
;
3048 r
= unhexmem(text
, strlen(text
), &root_hash
, &root_hash_size
);
3051 if (root_hash_size
< sizeof(sd_id128_t
))
3056 if ((root_hash
|| verity
->root_hash
) && !verity
->root_hash_sig
) {
3057 if (root_hash_sig_path
) {
3058 r
= read_full_file(root_hash_sig_path
, (char**) &root_hash_sig
, &root_hash_sig_size
);
3059 if (r
< 0 && r
!= -ENOENT
)
3063 designator
= PARTITION_ROOT
;
3065 if (designator
< 0 || designator
== PARTITION_ROOT
) {
3066 _cleanup_free_
char *p
= NULL
;
3068 /* Follow naming convention recommended by the relevant RFC:
3069 * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
3070 p
= build_auxiliary_path(image
, ".roothash.p7s");
3074 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
3075 if (r
< 0 && r
!= -ENOENT
)
3078 designator
= PARTITION_ROOT
;
3081 if (!root_hash_sig
&& (designator
< 0 || designator
== PARTITION_USR
)) {
3082 _cleanup_free_
char *p
= NULL
;
3084 p
= build_auxiliary_path(image
, ".usrhash.p7s");
3088 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
3089 if (r
< 0 && r
!= -ENOENT
)
3092 designator
= PARTITION_USR
;
3096 if (root_hash_sig
&& root_hash_sig_size
== 0) /* refuse empty size signatures */
3100 if (!verity
->data_path
) {
3101 _cleanup_free_
char *p
= NULL
;
3103 p
= build_auxiliary_path(image
, ".verity");
3107 if (access(p
, F_OK
) < 0) {
3108 if (errno
!= ENOENT
)
3111 verity_data_path
= TAKE_PTR(p
);
3115 verity
->root_hash
= TAKE_PTR(root_hash
);
3116 verity
->root_hash_size
= root_hash_size
;
3119 if (root_hash_sig
) {
3120 verity
->root_hash_sig
= TAKE_PTR(root_hash_sig
);
3121 verity
->root_hash_sig_size
= root_hash_sig_size
;
3124 if (verity_data_path
)
3125 verity
->data_path
= TAKE_PTR(verity_data_path
);
3127 if (verity
->designator
< 0)
3128 verity
->designator
= designator
;
3133 int dissected_image_load_verity_sig_partition(
3136 VeritySettings
*verity
) {
3138 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
3139 _cleanup_(json_variant_unrefp
) JsonVariant
*v
= NULL
;
3140 size_t root_hash_size
, root_hash_sig_size
;
3141 _cleanup_free_
char *buf
= NULL
;
3142 PartitionDesignator d
;
3143 DissectedPartition
*p
;
3144 JsonVariant
*rh
, *sig
;
3153 if (verity
->root_hash
&& verity
->root_hash_sig
) /* Already loaded? */
3156 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_EMBEDDED");
3157 if (r
< 0 && r
!= -ENXIO
)
3158 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_EMBEDDED, ignoring: %m");
3162 d
= partition_verity_sig_of(verity
->designator
< 0 ? PARTITION_ROOT
: verity
->designator
);
3165 p
= m
->partitions
+ d
;
3168 if (p
->offset
== UINT64_MAX
|| p
->size
== UINT64_MAX
)
3171 if (p
->size
> 4*1024*1024) /* Signature data cannot possible be larger than 4M, refuse that */
3172 return log_debug_errno(SYNTHETIC_ERRNO(EFBIG
), "Verity signature partition is larger than 4M, refusing.");
3174 buf
= new(char, p
->size
+1);
3178 n
= pread(fd
, buf
, p
->size
, p
->offset
);
3181 if ((uint64_t) n
!= p
->size
)
3184 e
= memchr(buf
, 0, p
->size
);
3186 /* If we found a NUL byte then the rest of the data must be NUL too */
3187 if (!memeqzero(e
, p
->size
- (e
- buf
)))
3188 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature data contains embedded NUL byte.");
3192 r
= json_parse(buf
, 0, &v
, NULL
, NULL
);
3194 return log_debug_errno(r
, "Failed to parse signature JSON data: %m");
3196 rh
= json_variant_by_key(v
, "rootHash");
3198 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'rootHash' field.");
3199 if (!json_variant_is_string(rh
))
3200 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'rootHash' field of signature JSON object is not a string.");
3202 r
= unhexmem(json_variant_string(rh
), SIZE_MAX
, &root_hash
, &root_hash_size
);
3204 return log_debug_errno(r
, "Failed to parse root hash field: %m");
3206 /* Check if specified root hash matches if it is specified */
3207 if (verity
->root_hash
&&
3208 memcmp_nn(verity
->root_hash
, verity
->root_hash_size
, root_hash
, root_hash_size
) != 0) {
3209 _cleanup_free_
char *a
= NULL
, *b
= NULL
;
3211 a
= hexmem(root_hash
, root_hash_size
);
3212 b
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
3214 return log_debug_errno(r
, "Root hash in signature JSON data (%s) doesn't match configured hash (%s).", strna(a
), strna(b
));
3217 sig
= json_variant_by_key(v
, "signature");
3219 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'signature' field.");
3220 if (!json_variant_is_string(sig
))
3221 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'signature' field of signature JSON object is not a string.");
3223 r
= unbase64mem(json_variant_string(sig
), SIZE_MAX
, &root_hash_sig
, &root_hash_sig_size
);
3225 return log_debug_errno(r
, "Failed to parse signature field: %m");
3227 free_and_replace(verity
->root_hash
, root_hash
);
3228 verity
->root_hash_size
= root_hash_size
;
3230 free_and_replace(verity
->root_hash_sig
, root_hash_sig
);
3231 verity
->root_hash_sig_size
= root_hash_sig_size
;
3236 int dissected_image_acquire_metadata(DissectedImage
*m
, DissectImageFlags extra_flags
) {
3243 META_INITRD_RELEASE
,
3244 META_EXTENSION_RELEASE
,
3245 META_HAS_INIT_SYSTEM
,
3249 static const char *const paths
[_META_MAX
] = {
3250 [META_HOSTNAME
] = "/etc/hostname\0",
3251 [META_MACHINE_ID
] = "/etc/machine-id\0",
3252 [META_MACHINE_INFO
] = "/etc/machine-info\0",
3253 [META_OS_RELEASE
] = ("/etc/os-release\0"
3254 "/usr/lib/os-release\0"),
3255 [META_INITRD_RELEASE
] = ("/etc/initrd-release\0"
3256 "/usr/lib/initrd-release\0"),
3257 [META_EXTENSION_RELEASE
] = "extension-release\0", /* Used only for logging. */
3258 [META_HAS_INIT_SYSTEM
] = "has-init-system\0", /* ditto */
3261 _cleanup_strv_free_
char **machine_info
= NULL
, **os_release
= NULL
, **initrd_release
= NULL
, **extension_release
= NULL
;
3262 _cleanup_close_pair_
int error_pipe
[2] = PIPE_EBADF
;
3263 _cleanup_(rmdir_and_freep
) char *t
= NULL
;
3264 _cleanup_(sigkill_waitp
) pid_t child
= 0;
3265 sd_id128_t machine_id
= SD_ID128_NULL
;
3266 _cleanup_free_
char *hostname
= NULL
;
3267 unsigned n_meta_initialized
= 0;
3268 int fds
[2 * _META_MAX
], r
, v
;
3269 int has_init_system
= -1;
3271 ImageClass image_class
= IMAGE_SYSEXT
;
3273 BLOCK_SIGNALS(SIGCHLD
);
3276 assert(image_class
);
3278 for (; n_meta_initialized
< _META_MAX
; n_meta_initialized
++) {
3279 if (!paths
[n_meta_initialized
]) {
3280 fds
[2*n_meta_initialized
] = fds
[2*n_meta_initialized
+1] = -EBADF
;
3284 if (pipe2(fds
+ 2*n_meta_initialized
, O_CLOEXEC
) < 0) {
3290 r
= mkdtemp_malloc("/tmp/dissect-XXXXXX", &t
);
3294 if (pipe2(error_pipe
, O_CLOEXEC
) < 0) {
3299 r
= safe_fork("(sd-dissect)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_NEW_MOUNTNS
|FORK_MOUNTNS_SLAVE
, &child
);
3303 /* Child in a new mount namespace */
3304 error_pipe
[0] = safe_close(error_pipe
[0]);
3306 r
= dissected_image_mount(
3312 DISSECT_IMAGE_READ_ONLY
|
3313 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
3314 DISSECT_IMAGE_USR_NO_ROOT
);
3316 log_debug_errno(r
, "Failed to mount dissected image: %m");
3320 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
3321 _cleanup_close_
int fd
= -ENOENT
;
3326 fds
[2*k
] = safe_close(fds
[2*k
]);
3330 case META_EXTENSION_RELEASE
: {
3331 /* As per the os-release spec, if the image is an extension it will have a file
3332 * named after the image name in extension-release.d/ - we use the image name
3333 * and try to resolve it with the extension-release helpers, as sometimes
3334 * the image names are mangled on deployment and do not match anymore.
3335 * Unlike other paths this is not fixed, and the image name
3336 * can be mangled on deployment, so by calling into the helper
3337 * we allow a fallback that matches on the first extension-release
3338 * file found in the directory, if one named after the image cannot
3339 * be found first. */
3340 ImageClass
class = IMAGE_SYSEXT
;
3341 r
= open_extension_release(t
, IMAGE_SYSEXT
, m
->image_name
, /* relax_extension_release_check= */ false, NULL
, &fd
);
3343 r
= open_extension_release(t
, IMAGE_CONFEXT
, m
->image_name
, /* relax_extension_release_check= */ false, NULL
, &fd
);
3345 class = IMAGE_CONFEXT
;
3350 r
= loop_write(fds
[2*k
+1], &class, sizeof(class), false);
3352 goto inner_fail
; /* Propagate the error to the parent */
3358 case META_HAS_INIT_SYSTEM
: {
3361 FOREACH_STRING(init
,
3362 "/usr/lib/systemd/systemd", /* systemd on /usr merged system */
3363 "/lib/systemd/systemd", /* systemd on /usr non-merged systems */
3364 "/sbin/init") { /* traditional path the Linux kernel invokes */
3366 r
= chase(init
, t
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
3369 log_debug_errno(r
, "Failed to resolve %s, ignoring: %m", init
);
3376 r
= loop_write(fds
[2*k
+1], &found
, sizeof(found
), false);
3384 NULSTR_FOREACH(p
, paths
[k
]) {
3385 fd
= chase_and_open(p
, t
, CHASE_PREFIX_ROOT
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
, NULL
);
3392 log_debug_errno(fd
, "Failed to read %s file of image, ignoring: %m", paths
[k
]);
3393 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3397 r
= copy_bytes(fd
, fds
[2*k
+1], UINT64_MAX
, 0);
3401 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3404 _exit(EXIT_SUCCESS
);
3407 /* Let parent know the error */
3408 (void) write(error_pipe
[1], &r
, sizeof(r
));
3409 _exit(EXIT_FAILURE
);
3412 error_pipe
[1] = safe_close(error_pipe
[1]);
3414 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
3415 _cleanup_fclose_
FILE *f
= NULL
;
3420 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3422 f
= take_fdopen(&fds
[2*k
], "r");
3431 r
= read_etc_hostname_stream(f
, &hostname
);
3433 log_debug_errno(r
, "Failed to read /etc/hostname of image: %m");
3437 case META_MACHINE_ID
: {
3438 _cleanup_free_
char *line
= NULL
;
3440 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3442 log_debug_errno(r
, "Failed to read /etc/machine-id of image: %m");
3444 r
= sd_id128_from_string(line
, &machine_id
);
3446 log_debug_errno(r
, "Image contains invalid /etc/machine-id: %s", line
);
3448 log_debug("/etc/machine-id file of image is empty.");
3449 else if (streq(line
, "uninitialized"))
3450 log_debug("/etc/machine-id file of image is uninitialized (likely aborted first boot).");
3452 log_debug("/etc/machine-id file of image has unexpected length %i.", r
);
3457 case META_MACHINE_INFO
:
3458 r
= load_env_file_pairs(f
, "machine-info", &machine_info
);
3460 log_debug_errno(r
, "Failed to read /etc/machine-info of image: %m");
3464 case META_OS_RELEASE
:
3465 r
= load_env_file_pairs(f
, "os-release", &os_release
);
3467 log_debug_errno(r
, "Failed to read OS release file of image: %m");
3471 case META_INITRD_RELEASE
:
3472 r
= load_env_file_pairs(f
, "initrd-release", &initrd_release
);
3474 log_debug_errno(r
, "Failed to read initrd release file of image: %m");
3478 case META_EXTENSION_RELEASE
: {
3479 ImageClass cl
= IMAGE_SYSEXT
;
3483 nr
= fread(&cl
, 1, sizeof(cl
), f
);
3484 if (nr
!= sizeof(cl
))
3485 log_debug_errno(errno_or_else(EIO
), "Failed to read class of extension image: %m");
3488 r
= load_env_file_pairs(f
, "extension-release", &extension_release
);
3490 log_debug_errno(r
, "Failed to read extension release file of image: %m");
3496 case META_HAS_INIT_SYSTEM
: {
3501 nr
= fread(&b
, 1, sizeof(b
), f
);
3502 if (nr
!= sizeof(b
))
3503 log_debug_errno(errno_or_else(EIO
), "Failed to read has-init-system boolean: %m");
3505 has_init_system
= b
;
3511 r
= wait_for_terminate_and_check("(sd-dissect)", child
, 0);
3516 n
= read(error_pipe
[0], &v
, sizeof(v
));
3520 return v
; /* propagate error sent to us from child */
3524 if (r
!= EXIT_SUCCESS
)
3527 free_and_replace(m
->hostname
, hostname
);
3528 m
->machine_id
= machine_id
;
3529 strv_free_and_replace(m
->machine_info
, machine_info
);
3530 strv_free_and_replace(m
->os_release
, os_release
);
3531 strv_free_and_replace(m
->initrd_release
, initrd_release
);
3532 strv_free_and_replace(m
->extension_release
, extension_release
);
3533 m
->has_init_system
= has_init_system
;
3534 m
->image_class
= image_class
;
3537 for (unsigned k
= 0; k
< n_meta_initialized
; k
++)
3538 safe_close_pair(fds
+ 2*k
);
3543 Architecture
dissected_image_architecture(DissectedImage
*img
) {
3546 if (img
->partitions
[PARTITION_ROOT
].found
&&
3547 img
->partitions
[PARTITION_ROOT
].architecture
>= 0)
3548 return img
->partitions
[PARTITION_ROOT
].architecture
;
3550 if (img
->partitions
[PARTITION_USR
].found
&&
3551 img
->partitions
[PARTITION_USR
].architecture
>= 0)
3552 return img
->partitions
[PARTITION_USR
].architecture
;
3554 return _ARCHITECTURE_INVALID
;
3557 int dissect_loop_device(
3559 const VeritySettings
*verity
,
3560 const MountOptions
*mount_options
,
3561 const ImagePolicy
*image_policy
,
3562 DissectImageFlags flags
,
3563 DissectedImage
**ret
) {
3566 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
3571 r
= dissected_image_new(loop
->backing_file
?: loop
->node
, &m
);
3575 m
->loop
= loop_device_ref(loop
);
3576 m
->sector_size
= m
->loop
->sector_size
;
3578 r
= dissect_image(m
, loop
->fd
, loop
->node
, verity
, mount_options
, image_policy
, flags
);
3591 int dissect_loop_device_and_warn(
3593 const VeritySettings
*verity
,
3594 const MountOptions
*mount_options
,
3595 const ImagePolicy
*image_policy
,
3596 DissectImageFlags flags
,
3597 DissectedImage
**ret
) {
3601 return dissect_log_error(
3603 dissect_loop_device(loop
, verity
, mount_options
, image_policy
, flags
, ret
),
3604 loop
->backing_file
?: loop
->node
,
3609 bool dissected_image_verity_candidate(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3612 /* Checks if this partition could theoretically do Verity. For non-partitioned images this only works
3613 * if there's an external verity file supplied, for which we can consult .has_verity. For partitioned
3614 * images we only check the partition type.
3616 * This call is used to decide whether to suppress or show a verity column in tabular output of the
3619 if (image
->single_file_system
)
3620 return partition_designator
== PARTITION_ROOT
&& image
->has_verity
;
3622 return partition_verity_of(partition_designator
) >= 0;
3625 bool dissected_image_verity_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3626 PartitionDesignator k
;
3630 /* Checks if this partition has verity data available that we can activate. For non-partitioned this
3631 * works for the root partition, for others only if the associated verity partition was found. */
3633 if (!image
->verity_ready
)
3636 if (image
->single_file_system
)
3637 return partition_designator
== PARTITION_ROOT
;
3639 k
= partition_verity_of(partition_designator
);
3640 return k
>= 0 && image
->partitions
[k
].found
;
3643 bool dissected_image_verity_sig_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3644 PartitionDesignator k
;
3648 /* Checks if this partition has verity signature data available that we can use. */
3650 if (!image
->verity_sig_ready
)
3653 if (image
->single_file_system
)
3654 return partition_designator
== PARTITION_ROOT
;
3656 k
= partition_verity_sig_of(partition_designator
);
3657 return k
>= 0 && image
->partitions
[k
].found
;
3660 MountOptions
* mount_options_free_all(MountOptions
*options
) {
3663 while ((m
= LIST_POP(mount_options
, options
))) {
3671 const char* mount_options_from_designator(const MountOptions
*options
, PartitionDesignator designator
) {
3672 LIST_FOREACH(mount_options
, m
, options
)
3673 if (designator
== m
->partition_designator
&& !isempty(m
->options
))
3679 int mount_image_privately_interactively(
3681 const ImagePolicy
*image_policy
,
3682 DissectImageFlags flags
,
3683 char **ret_directory
,
3685 LoopDevice
**ret_loop_device
) {
3687 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3688 _cleanup_(loop_device_unrefp
) LoopDevice
*d
= NULL
;
3689 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3690 _cleanup_free_
char *dir
= NULL
;
3693 /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
3694 * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
3698 assert(ret_loop_device
);
3700 /* We intend to mount this right-away, hence add the partitions if needed and pin them. */
3701 flags
|= DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
3702 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
3704 r
= verity_settings_load(&verity
, image
, NULL
, NULL
);
3706 return log_error_errno(r
, "Failed to load root hash data: %m");
3708 r
= loop_device_make_by_path(
3710 FLAGS_SET(flags
, DISSECT_IMAGE_DEVICE_READ_ONLY
) ? O_RDONLY
: O_RDWR
,
3711 /* sector_size= */ UINT32_MAX
,
3712 FLAGS_SET(flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
3716 return log_error_errno(r
, "Failed to set up loopback device for %s: %m", image
);
3718 r
= dissect_loop_device_and_warn(
3721 /* mount_options= */ NULL
,
3728 r
= dissected_image_load_verity_sig_partition(dissected_image
, d
->fd
, &verity
);
3732 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, &verity
, flags
);
3736 r
= detach_mount_namespace();
3738 return log_error_errno(r
, "Failed to detach mount namespace: %m");
3740 r
= mkdir_p("/run/systemd/mount-rootfs", 0555);
3742 return log_error_errno(r
, "Failed to create mount point: %m");
3744 r
= dissected_image_mount_and_warn(
3746 "/run/systemd/mount-rootfs",
3747 /* uid_shift= */ UID_INVALID
,
3748 /* uid_range= */ UID_INVALID
,
3753 r
= loop_device_flock(d
, LOCK_UN
);
3757 r
= dissected_image_relinquish(dissected_image
);
3759 return log_error_errno(r
, "Failed to relinquish DM and loopback block devices: %m");
3761 if (ret_directory
) {
3762 dir
= strdup("/run/systemd/mount-rootfs");
3768 _cleanup_close_
int dir_fd
= -EBADF
;
3770 dir_fd
= open("/run/systemd/mount-rootfs", O_CLOEXEC
|O_DIRECTORY
);
3772 return log_error_errno(errno
, "Failed to open mount point directory: %m");
3774 *ret_dir_fd
= TAKE_FD(dir_fd
);
3778 *ret_directory
= TAKE_PTR(dir
);
3780 *ret_loop_device
= TAKE_PTR(d
);
3784 static bool mount_options_relax_extension_release_checks(const MountOptions
*options
) {
3788 return string_contains_word(mount_options_from_designator(options
, PARTITION_ROOT
), ",", "x-systemd.relax-extension-release-check") ||
3789 string_contains_word(mount_options_from_designator(options
, PARTITION_USR
), ",", "x-systemd.relax-extension-release-check") ||
3790 string_contains_word(options
->options
, ",", "x-systemd.relax-extension-release-check");
3793 int verity_dissect_and_mount(
3797 const MountOptions
*options
,
3798 const ImagePolicy
*image_policy
,
3799 const char *required_host_os_release_id
,
3800 const char *required_host_os_release_version_id
,
3801 const char *required_host_os_release_sysext_level
,
3802 const char *required_sysext_scope
) {
3804 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
3805 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3806 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3807 DissectImageFlags dissect_image_flags
;
3808 bool relax_extension_release_check
;
3814 relax_extension_release_check
= mount_options_relax_extension_release_checks(options
);
3816 /* We might get an FD for the image, but we use the original path to look for the dm-verity files */
3817 r
= verity_settings_load(&verity
, src
, NULL
, NULL
);
3819 return log_debug_errno(r
, "Failed to load root hash: %m");
3821 dissect_image_flags
= (verity
.data_path
? DISSECT_IMAGE_NO_PARTITION_TABLE
: 0) |
3822 (relax_extension_release_check
? DISSECT_IMAGE_RELAX_EXTENSION_CHECK
: 0) |
3823 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
3824 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
3826 /* Note that we don't use loop_device_make here, as the FD is most likely O_PATH which would not be
3827 * accepted by LOOP_CONFIGURE, so just let loop_device_make_by_path reopen it as a regular FD. */
3828 r
= loop_device_make_by_path(
3829 src_fd
>= 0 ? FORMAT_PROC_FD_PATH(src_fd
) : src
,
3830 /* open_flags= */ -1,
3831 /* sector_size= */ UINT32_MAX
,
3832 verity
.data_path
? 0 : LO_FLAGS_PARTSCAN
,
3836 return log_debug_errno(r
, "Failed to create loop device for image: %m");
3838 r
= dissect_loop_device(
3843 dissect_image_flags
,
3845 /* No partition table? Might be a single-filesystem image, try again */
3846 if (!verity
.data_path
&& r
== -ENOPKG
)
3847 r
= dissect_loop_device(
3852 dissect_image_flags
| DISSECT_IMAGE_NO_PARTITION_TABLE
,
3855 return log_debug_errno(r
, "Failed to dissect image: %m");
3857 r
= dissected_image_load_verity_sig_partition(dissected_image
, loop_device
->fd
, &verity
);
3861 r
= dissected_image_decrypt(
3865 dissect_image_flags
);
3867 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
3869 r
= mkdir_p_label(dest
, 0755);
3871 return log_debug_errno(r
, "Failed to create destination directory %s: %m", dest
);
3872 r
= umount_recursive(dest
, 0);
3874 return log_debug_errno(r
, "Failed to umount under destination directory %s: %m", dest
);
3876 r
= dissected_image_mount(dissected_image
, dest
, UID_INVALID
, UID_INVALID
, dissect_image_flags
);
3878 return log_debug_errno(r
, "Failed to mount image: %m");
3880 r
= loop_device_flock(loop_device
, LOCK_UN
);
3882 return log_debug_errno(r
, "Failed to unlock loopback device: %m");
3884 /* If we got os-release values from the caller, then we need to match them with the image's
3885 * extension-release.d/ content. Return -EINVAL if there's any mismatch.
3886 * First, check the distro ID. If that matches, then check the new SYSEXT_LEVEL value if
3887 * available, or else fallback to VERSION_ID. If neither is present (eg: rolling release),
3888 * then a simple match on the ID will be performed. */
3889 if (required_host_os_release_id
) {
3890 _cleanup_strv_free_
char **extension_release
= NULL
;
3891 ImageClass
class = IMAGE_SYSEXT
;
3893 assert(!isempty(required_host_os_release_id
));
3895 r
= load_extension_release_pairs(dest
, IMAGE_SYSEXT
, dissected_image
->image_name
, relax_extension_release_check
, &extension_release
);
3897 r
= load_extension_release_pairs(dest
, IMAGE_CONFEXT
, dissected_image
->image_name
, relax_extension_release_check
, &extension_release
);
3899 class = IMAGE_CONFEXT
;
3902 return log_debug_errno(r
, "Failed to parse image %s extension-release metadata: %m", dissected_image
->image_name
);
3904 r
= extension_release_validate(
3905 dissected_image
->image_name
,
3906 required_host_os_release_id
,
3907 required_host_os_release_version_id
,
3908 required_host_os_release_sysext_level
,
3909 required_sysext_scope
,
3913 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE
), "Image %s extension-release metadata does not match the root's", dissected_image
->image_name
);
3915 return log_debug_errno(r
, "Failed to compare image %s extension-release metadata with the root's os-release: %m", dissected_image
->image_name
);
3918 r
= dissected_image_relinquish(dissected_image
);
3920 return log_debug_errno(r
, "Failed to relinquish dissected image: %m");