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 "missing_syscall.h"
55 #include "mkdir-label.h"
56 #include "mount-util.h"
57 #include "mountpoint-util.h"
58 #include "namespace-util.h"
59 #include "nulstr-util.h"
60 #include "openssl-util.h"
62 #include "path-util.h"
63 #include "process-util.h"
64 #include "raw-clone.h"
65 #include "resize-fs.h"
66 #include "signal-util.h"
67 #include "sparse-endian.h"
68 #include "stat-util.h"
69 #include "stdio-util.h"
70 #include "string-table.h"
71 #include "string-util.h"
73 #include "tmpfile-util.h"
74 #include "udev-util.h"
75 #include "user-util.h"
76 #include "xattr-util.h"
78 /* how many times to wait for the device nodes to appear */
79 #define N_DEVICE_NODE_LIST_ATTEMPTS 10
81 int dissect_fstype_ok(const char *fstype
) {
85 /* When we automatically mount file systems, be a bit conservative by default what we are willing to
86 * mount, just as an extra safety net to not mount with badly maintained legacy file system
89 e
= secure_getenv("SYSTEMD_DISSECT_FILE_SYSTEMS");
91 _cleanup_strv_free_
char **l
= NULL
;
93 l
= strv_split(e
, ":");
97 b
= strv_contains(l
, fstype
);
99 b
= STR_IN_SET(fstype
,
110 log_debug("File system type '%s' is not allowed to be mounted as result of automatic dissection.", fstype
);
114 int probe_sector_size(int fd
, uint32_t *ret
) {
116 /* Disk images might be for 512B or for 4096 sector sizes, let's try to auto-detect that by searching
117 * for the GPT headers at the relevant byte offsets */
119 assert_cc(sizeof(GptHeader
) == 92);
121 /* We expect a sector size in the range 512…4096. The GPT header is located in the second
122 * sector. Hence it could be at byte 512 at the earliest, and at byte 4096 at the latest. And we must
123 * read with granularity of the largest sector size we care about. Which means 8K. */
124 uint8_t sectors
[2 * 4096];
131 n
= pread(fd
, sectors
, sizeof(sectors
), 0);
134 if (n
!= sizeof(sectors
)) /* too short? */
137 /* Let's see if we find the GPT partition header with various expected sector sizes */
138 for (uint32_t sz
= 512; sz
<= 4096; sz
<<= 1) {
141 assert(sizeof(sectors
) >= sz
* 2);
142 p
= (const GptHeader
*) (sectors
+ sz
);
144 if (!gpt_header_has_signature(p
))
148 return log_debug_errno(SYNTHETIC_ERRNO(ENOTUNIQ
),
149 "Detected valid partition table at offsets matching multiple sector sizes, refusing.");
155 log_debug("Determined sector size %" PRIu32
" based on discovered partition table.", found
);
157 return 1; /* indicate we *did* find it */
161 log_debug("Couldn't find any partition table to derive sector size of.");
162 *ret
= 512; /* pick the traditional default */
163 return 0; /* indicate we didn't find it */
166 int probe_sector_size_prefer_ioctl(int fd
, uint32_t *ret
) {
172 /* Just like probe_sector_size(), but if we are looking at a block device, will use the already
173 * configured sector size rather than probing by contents */
175 if (fstat(fd
, &st
) < 0)
178 if (S_ISBLK(st
.st_mode
))
179 return blockdev_get_sector_size(fd
, ret
);
181 return probe_sector_size(fd
, ret
);
184 int probe_filesystem_full(
191 /* Try to find device content type and return it in *ret_fstype. If nothing is found,
192 * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and a
193 * different error otherwise. */
196 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
197 _cleanup_free_
char *path_by_fd
= NULL
;
198 _cleanup_close_
int fd_close
= -EBADF
;
202 assert(fd
>= 0 || path
);
206 fd_close
= open(path
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
214 r
= fd_get_path(fd
, &path_by_fd
);
221 if (size
== 0) /* empty size? nothing found! */
224 b
= blkid_new_probe();
228 /* The Linux kernel maintains separate block device caches for main ("whole") and partition block
229 * devices, which means making a change to one might not be reflected immediately when reading via
230 * the other. That's massively confusing when mixing accesses to such devices. Let's address this in
231 * a limited way: when probing a file system that is not at the beginning of the block device we
232 * apparently probe a partition via the main block device, and in that case let's first flush the
233 * main block device cache, so that we get the data that the per-partition block device last
236 * This only works under the assumption that any tools that write to the partition block devices
237 * issue an syncfs()/fsync() on the device after making changes. Typically file system formatting
238 * tools that write a superblock onto a partition block device do that, however. */
240 if (ioctl(fd
, BLKFLSBUF
, 0) < 0)
241 log_debug_errno(errno
, "Failed to flush block device cache, ignoring: %m");
244 r
= blkid_probe_set_device(
248 size
== UINT64_MAX
? 0 : size
); /* when blkid sees size=0 it understands "everything". We prefer using UINT64_MAX for that */
250 return errno_or_else(ENOMEM
);
252 blkid_probe_enable_superblocks(b
, 1);
253 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
);
256 r
= blkid_do_safeprobe(b
);
257 if (r
== _BLKID_SAFEPROBE_NOT_FOUND
)
259 if (r
== _BLKID_SAFEPROBE_AMBIGUOUS
)
260 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
),
261 "Results ambiguous for partition %s", path
);
262 if (r
== _BLKID_SAFEPROBE_ERROR
)
263 return log_debug_errno(errno_or_else(EIO
), "Failed to probe partition %s: %m", path
);
265 assert(r
== _BLKID_SAFEPROBE_FOUND
);
267 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
272 log_debug("Probed fstype '%s' on partition %s.", fstype
, path
);
283 log_debug("No type detected on partition %s", path
);
292 static int image_policy_may_use(
293 const ImagePolicy
*policy
,
294 PartitionDesignator designator
) {
296 PartitionPolicyFlags f
;
298 /* For each partition we find in the partition table do a first check if it may exist at all given
299 * the policy, or if it shall be ignored. */
301 f
= image_policy_get_exhaustively(policy
, designator
);
305 if ((f
& _PARTITION_POLICY_USE_MASK
) == PARTITION_POLICY_ABSENT
)
306 /* only flag set in policy is "absent"? then this partition may not exist at all */
307 return log_debug_errno(
308 SYNTHETIC_ERRNO(ERFKILL
),
309 "Partition of designator '%s' exists, but not allowed by policy, refusing.",
310 partition_designator_to_string(designator
));
311 if ((f
& _PARTITION_POLICY_USE_MASK
& ~PARTITION_POLICY_ABSENT
) == PARTITION_POLICY_UNUSED
) {
312 /* only "unused" or "unused" + "absent" are set? then don't use it */
313 log_debug("Partition of designator '%s' exists, and policy dictates to ignore it, doing so.",
314 partition_designator_to_string(designator
));
315 return false; /* ignore! */
318 return true; /* use! */
321 static int image_policy_check_protection(
322 const ImagePolicy
*policy
,
323 PartitionDesignator designator
,
324 PartitionPolicyFlags found_flags
) {
326 PartitionPolicyFlags policy_flags
;
328 /* Checks if the flags in the policy for the designated partition overlap the flags of what we found */
333 policy_flags
= image_policy_get_exhaustively(policy
, designator
);
334 if (policy_flags
< 0)
337 if ((found_flags
& policy_flags
) == 0) {
338 _cleanup_free_
char *found_flags_string
= NULL
, *policy_flags_string
= NULL
;
340 (void) partition_policy_flags_to_string(found_flags
, /* simplify= */ true, &found_flags_string
);
341 (void) partition_policy_flags_to_string(policy_flags
, /* simplify= */ true, &policy_flags_string
);
343 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL
), "Partition %s discovered with policy '%s' but '%s' was required, refusing.",
344 partition_designator_to_string(designator
),
345 strnull(found_flags_string
), strnull(policy_flags_string
));
351 static int image_policy_check_partition_flags(
352 const ImagePolicy
*policy
,
353 PartitionDesignator designator
,
354 uint64_t gpt_flags
) {
356 PartitionPolicyFlags policy_flags
;
359 /* Checks if the partition flags in the policy match reality */
361 policy_flags
= image_policy_get_exhaustively(policy
, designator
);
362 if (policy_flags
< 0)
365 b
= FLAGS_SET(gpt_flags
, SD_GPT_FLAG_READ_ONLY
);
366 if ((policy_flags
& _PARTITION_POLICY_READ_ONLY_MASK
) == (b
? PARTITION_POLICY_READ_ONLY_OFF
: PARTITION_POLICY_READ_ONLY_ON
))
367 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL
), "Partition %s has 'read-only' flag incorrectly set (must be %s, is %s), refusing.",
368 partition_designator_to_string(designator
),
369 one_zero(!b
), one_zero(b
));
371 b
= FLAGS_SET(gpt_flags
, SD_GPT_FLAG_GROWFS
);
372 if ((policy_flags
& _PARTITION_POLICY_GROWFS_MASK
) == (b
? PARTITION_POLICY_GROWFS_OFF
: PARTITION_POLICY_GROWFS_ON
))
373 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL
), "Partition %s has 'growfs' flag incorrectly set (must be %s, is %s), refusing.",
374 partition_designator_to_string(designator
),
375 one_zero(!b
), one_zero(b
));
380 static int dissected_image_probe_filesystems(
383 const ImagePolicy
*policy
) {
389 /* Fill in file system types if we don't know them yet. */
391 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
392 DissectedPartition
*p
= m
->partitions
+ i
;
393 PartitionPolicyFlags found_flags
;
399 /* If we have an fd referring to the partition block device, use that. Otherwise go
400 * via the whole block device or backing regular file, and read via offset. */
401 if (p
->mount_node_fd
>= 0)
402 r
= probe_filesystem_full(p
->mount_node_fd
, p
->node
, 0, UINT64_MAX
, &p
->fstype
);
404 r
= probe_filesystem_full(fd
, p
->node
, p
->offset
, p
->size
, &p
->fstype
);
409 if (streq_ptr(p
->fstype
, "crypto_LUKS")) {
411 found_flags
= PARTITION_POLICY_ENCRYPTED
; /* found this one, and its definitely encrypted */
413 /* found it, but it's definitely not encrypted, hence mask the encrypted flag, but
414 * set all other ways that indicate "present". */
415 found_flags
= PARTITION_POLICY_UNPROTECTED
|PARTITION_POLICY_VERITY
|PARTITION_POLICY_SIGNED
;
417 if (p
->fstype
&& fstype_is_ro(p
->fstype
))
423 /* We might have learnt more about the file system now (i.e. whether it is encrypted or not),
424 * hence we need to validate this against policy again, to see if the policy still matches
425 * with this new information. Note that image_policy_check_protection() will check for
426 * overlap between what's allowed in the policy and what we pass as 'found_policy' here. In
427 * the unencrypted case we thus might pass an overly unspecific mask here (i.e. unprotected
428 * OR verity OR signed), but that's fine since the earlier policy check already checked more
429 * specific which of those three cases where OK. Keep in mind that this function here only
430 * looks at specific partitions (and thus can only deduce encryption or not) but not the
431 * overall partition table (and thus cannot deduce verity or not). The earlier dissection
432 * checks already did the relevant checks that look at the whole partition table, and
433 * enforced policy there as needed. */
434 r
= image_policy_check_protection(policy
, i
, found_flags
);
442 static void check_partition_flags(
444 unsigned long long pflags
,
445 unsigned long long supported
) {
449 /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
450 pflags
&= ~(supported
|
451 SD_GPT_FLAG_REQUIRED_PARTITION
|
452 SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL
|
453 SD_GPT_FLAG_LEGACY_BIOS_BOOTABLE
);
458 /* If there are other bits set, then log about it, to make things discoverable */
459 for (unsigned i
= 0; i
< sizeof(pflags
) * 8; i
++) {
460 unsigned long long bit
= 1ULL << i
;
461 if (!FLAGS_SET(pflags
, bit
))
464 log_debug("Unexpected partition flag %llu set on %s!", bit
, node
);
468 static int dissected_image_new(const char *path
, DissectedImage
**ret
) {
469 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
470 _cleanup_free_
char *name
= NULL
;
476 _cleanup_free_
char *filename
= NULL
;
478 r
= path_extract_filename(path
, &filename
);
482 r
= raw_strip_suffixes(filename
, &name
);
486 if (!image_name_is_valid(name
)) {
487 log_debug("Image name %s is not valid, ignoring.", strna(name
));
492 m
= new(DissectedImage
, 1);
496 *m
= (DissectedImage
) {
497 .has_init_system
= -1,
498 .image_name
= TAKE_PTR(name
),
501 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++)
502 m
->partitions
[i
] = DISSECTED_PARTITION_NULL
;
509 static void dissected_partition_done(DissectedPartition
*p
) {
515 free(p
->decrypted_fstype
);
516 free(p
->decrypted_node
);
517 free(p
->mount_options
);
518 safe_close(p
->mount_node_fd
);
519 safe_close(p
->fsmount_fd
);
521 *p
= DISSECTED_PARTITION_NULL
;
525 static int make_partition_devname(
526 const char *whole_devname
,
529 DissectImageFlags flags
,
532 _cleanup_free_
char *s
= NULL
;
535 assert(whole_devname
);
536 assert(nr
!= 0); /* zero is not a valid partition nr */
539 if (!FLAGS_SET(flags
, DISSECT_IMAGE_DISKSEQ_DEVNODE
) || diskseq
== 0) {
541 /* Given a whole block device node name (e.g. /dev/sda or /dev/loop7) generate a partition
542 * device name (e.g. /dev/sda7 or /dev/loop7p5). The rule the kernel uses is simple: if whole
543 * block device node name ends in a digit, then suffix a 'p', followed by the partition
544 * number. Otherwise, just suffix the partition number without any 'p'. */
546 if (nr
< 0) { /* whole disk? */
547 s
= strdup(whole_devname
);
551 size_t l
= strlen(whole_devname
);
552 if (l
< 1) /* underflow check for the subtraction below */
555 bool need_p
= ascii_isdigit(whole_devname
[l
-1]); /* Last char a digit? */
557 if (asprintf(&s
, "%s%s%i", whole_devname
, need_p
? "p" : "", nr
) < 0)
561 if (nr
< 0) /* whole disk? */
562 r
= asprintf(&s
, "/dev/disk/by-diskseq/%" PRIu64
, diskseq
);
564 r
= asprintf(&s
, "/dev/disk/by-diskseq/%" PRIu64
"-part%i", diskseq
, nr
);
573 static int open_partition(
576 const LoopDevice
*loop
) {
578 _cleanup_(sd_device_unrefp
) sd_device
*dev
= NULL
;
579 _cleanup_close_
int fd
= -EBADF
;
586 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
590 /* Check if the block device is a child of (or equivalent to) the originally provided one. */
591 r
= block_device_new_from_fd(fd
, is_partition
? BLOCK_DEVICE_LOOKUP_WHOLE_DISK
: 0, &dev
);
595 r
= sd_device_get_devnum(dev
, &devnum
);
599 if (loop
->devno
!= devnum
)
602 /* Also check diskseq. */
603 if (loop
->diskseq
!= 0) {
606 r
= fd_get_diskseq(fd
, &diskseq
);
610 if (loop
->diskseq
!= diskseq
)
614 log_debug("Opened %s (fd=%i, whole_block_devnum=" DEVNUM_FORMAT_STR
", diskseq=%" PRIu64
").",
615 node
, fd
, DEVNUM_FORMAT_VAL(loop
->devno
), loop
->diskseq
);
619 static int compare_arch(Architecture a
, Architecture b
) {
623 if (a
== native_architecture())
626 if (b
== native_architecture())
629 #ifdef ARCHITECTURE_SECONDARY
630 if (a
== ARCHITECTURE_SECONDARY
)
633 if (b
== ARCHITECTURE_SECONDARY
)
640 static int dissect_image(
644 const VeritySettings
*verity
,
645 const MountOptions
*mount_options
,
646 const ImagePolicy
*policy
,
647 DissectImageFlags flags
) {
649 sd_id128_t root_uuid
= SD_ID128_NULL
, root_verity_uuid
= SD_ID128_NULL
;
650 sd_id128_t usr_uuid
= SD_ID128_NULL
, usr_verity_uuid
= SD_ID128_NULL
;
651 bool is_gpt
, is_mbr
, multiple_generic
= false,
652 generic_rw
= false, /* initialize to appease gcc */
653 generic_growfs
= false;
654 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
655 _cleanup_free_
char *generic_node
= NULL
;
656 sd_id128_t generic_uuid
= SD_ID128_NULL
;
657 const char *pttype
= NULL
, *sptuuid
= NULL
;
659 int r
, generic_nr
= -1, n_partitions
;
664 assert(!verity
|| verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
665 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
666 assert(!verity
|| verity
->root_hash_sig
|| verity
->root_hash_sig_size
== 0);
667 assert(!verity
|| (verity
->root_hash
|| !verity
->root_hash_sig
));
668 assert(!((flags
& DISSECT_IMAGE_GPT_ONLY
) && (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)));
669 assert(m
->sector_size
> 0);
671 /* Probes a disk image, and returns information about what it found in *ret.
673 * Returns -ENOPKG if no suitable partition table or file system could be found.
674 * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found.
675 * Returns -ENXIO if we couldn't find any partition suitable as root or /usr partition
676 * Returns -ENOTUNIQ if we only found multiple generic partitions and thus don't know what to do with that
677 * Returns -ERFKILL if image doesn't match image policy
678 * Returns -EBADR if verity data was provided externally for an image that has a GPT partition table (i.e. is not just a naked fs)
679 * Returns -EPROTONOSUPPORT if DISSECT_IMAGE_ADD_PARTITION_DEVICES is set but the block device does not have partition logic enabled
680 * Returns -ENOMSG if we didn't find a single usable partition (and DISSECT_IMAGE_REFUSE_EMPTY is set) */
682 uint64_t diskseq
= m
->loop
? m
->loop
->diskseq
: 0;
684 if (verity
&& verity
->root_hash
) {
685 sd_id128_t fsuuid
, vuuid
;
687 /* If a root hash is supplied, then we use the root partition that has a UUID that match the
688 * first 128-bit of the root hash. And we use the verity partition that has a UUID that match
689 * the final 128-bit. */
691 if (verity
->root_hash_size
< sizeof(sd_id128_t
))
694 memcpy(&fsuuid
, verity
->root_hash
, sizeof(sd_id128_t
));
695 memcpy(&vuuid
, (const uint8_t*) verity
->root_hash
+ verity
->root_hash_size
- sizeof(sd_id128_t
), sizeof(sd_id128_t
));
697 if (sd_id128_is_null(fsuuid
))
699 if (sd_id128_is_null(vuuid
))
702 /* If the verity data declares it's for the /usr partition, then search for that, in all
703 * other cases assume it's for the root partition. */
704 if (verity
->designator
== PARTITION_USR
) {
706 usr_verity_uuid
= vuuid
;
709 root_verity_uuid
= vuuid
;
713 b
= blkid_new_probe();
718 r
= blkid_probe_set_device(b
, fd
, 0, 0);
720 return errno_or_else(ENOMEM
);
723 r
= blkid_probe_set_sectorsize(b
, m
->sector_size
);
725 return errno_or_else(EIO
);
727 if ((flags
& DISSECT_IMAGE_GPT_ONLY
) == 0) {
728 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
729 blkid_probe_enable_superblocks(b
, 1);
730 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
|BLKID_SUBLKS_USAGE
|BLKID_SUBLKS_UUID
);
733 blkid_probe_enable_partitions(b
, 1);
734 blkid_probe_set_partitions_flags(b
, BLKID_PARTS_ENTRY_DETAILS
);
737 r
= blkid_do_safeprobe(b
);
738 if (r
== _BLKID_SAFEPROBE_ERROR
)
739 return errno_or_else(EIO
);
740 if (IN_SET(r
, _BLKID_SAFEPROBE_AMBIGUOUS
, _BLKID_SAFEPROBE_NOT_FOUND
))
741 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG
), "Failed to identify any partition table.");
743 assert(r
== _BLKID_SAFEPROBE_FOUND
);
745 if ((!(flags
& DISSECT_IMAGE_GPT_ONLY
) &&
746 (flags
& DISSECT_IMAGE_GENERIC_ROOT
)) ||
747 (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)) {
748 const char *usage
= NULL
;
750 /* If flags permit this, also allow using non-partitioned single-filesystem images */
752 (void) blkid_probe_lookup_value(b
, "USAGE", &usage
, NULL
);
753 if (STRPTR_IN_SET(usage
, "filesystem", "crypto")) {
754 _cleanup_free_
char *t
= NULL
, *n
= NULL
, *o
= NULL
;
755 const char *fstype
= NULL
, *options
= NULL
, *suuid
= NULL
;
756 _cleanup_close_
int mount_node_fd
= -EBADF
;
757 sd_id128_t uuid
= SD_ID128_NULL
;
758 PartitionPolicyFlags found_flags
;
761 /* OK, we have found a file system, that's our root partition then. */
763 r
= image_policy_may_use(policy
, PARTITION_ROOT
);
766 if (r
== 0) /* policy says ignore this, so we ignore it */
769 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
770 (void) blkid_probe_lookup_value(b
, "UUID", &suuid
, NULL
);
772 encrypted
= streq_ptr(fstype
, "crypto_LUKS");
774 if (verity_settings_data_covers(verity
, PARTITION_ROOT
))
775 found_flags
= verity
->root_hash_sig
? PARTITION_POLICY_SIGNED
: PARTITION_POLICY_VERITY
;
777 found_flags
= encrypted
? PARTITION_POLICY_ENCRYPTED
: PARTITION_POLICY_UNPROTECTED
;
779 r
= image_policy_check_protection(policy
, PARTITION_ROOT
, found_flags
);
783 r
= image_policy_check_partition_flags(policy
, PARTITION_ROOT
, 0); /* we have no gpt partition flags, hence check against all bits off */
787 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
788 mount_node_fd
= open_partition(devname
, /* is_partition = */ false, m
->loop
);
789 if (mount_node_fd
< 0)
790 return mount_node_fd
;
800 /* blkid will return FAT's serial number as UUID, hence it is quite possible
801 * that parsing this will fail. We'll ignore the ID, since it's just too
802 * short to be useful as tru identifier. */
803 r
= sd_id128_from_string(suuid
, &uuid
);
805 log_debug_errno(r
, "Failed to parse file system UUID '%s', ignoring: %m", suuid
);
808 r
= make_partition_devname(devname
, diskseq
, -1, flags
, &n
);
812 m
->single_file_system
= true;
813 m
->encrypted
= encrypted
;
815 m
->has_verity
= verity
&& verity
->data_path
;
816 m
->verity_ready
= verity_settings_data_covers(verity
, PARTITION_ROOT
);
818 m
->has_verity_sig
= false; /* signature not embedded, must be specified */
819 m
->verity_sig_ready
= m
->verity_ready
&& verity
->root_hash_sig
;
821 m
->image_uuid
= uuid
;
823 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
830 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
832 .rw
= !m
->verity_ready
&& !fstype_is_ro(fstype
),
834 .architecture
= _ARCHITECTURE_INVALID
,
835 .fstype
= TAKE_PTR(t
),
837 .mount_options
= TAKE_PTR(o
),
838 .mount_node_fd
= TAKE_FD(mount_node_fd
),
841 .fsmount_fd
= -EBADF
,
848 (void) blkid_probe_lookup_value(b
, "PTTYPE", &pttype
, NULL
);
852 is_gpt
= streq_ptr(pttype
, "gpt");
853 is_mbr
= streq_ptr(pttype
, "dos");
855 if (!is_gpt
&& ((flags
& DISSECT_IMAGE_GPT_ONLY
) || !is_mbr
))
858 /* We support external verity data partitions only if the image has no partition table */
859 if (verity
&& verity
->data_path
)
862 if (FLAGS_SET(flags
, DISSECT_IMAGE_ADD_PARTITION_DEVICES
)) {
863 /* Safety check: refuse block devices that carry a partition table but for which the kernel doesn't
864 * do partition scanning. */
865 r
= blockdev_partscan_enabled(fd
);
869 return -EPROTONOSUPPORT
;
872 (void) blkid_probe_lookup_value(b
, "PTUUID", &sptuuid
, NULL
);
874 r
= sd_id128_from_string(sptuuid
, &m
->image_uuid
);
876 log_debug_errno(r
, "Failed to parse partition table UUID '%s', ignoring: %m", sptuuid
);
880 pl
= blkid_probe_get_partitions(b
);
882 return errno_or_else(ENOMEM
);
885 n_partitions
= blkid_partlist_numof_partitions(pl
);
886 if (n_partitions
< 0)
887 return errno_or_else(EIO
);
889 for (int i
= 0; i
< n_partitions
; i
++) {
890 _cleanup_free_
char *node
= NULL
;
891 unsigned long long pflags
;
892 blkid_loff_t start
, size
;
897 pp
= blkid_partlist_get_partition(pl
, i
);
899 return errno_or_else(EIO
);
901 pflags
= blkid_partition_get_flags(pp
);
904 nr
= blkid_partition_get_partno(pp
);
906 return errno_or_else(EIO
);
909 start
= blkid_partition_get_start(pp
);
911 return errno_or_else(EIO
);
913 assert((uint64_t) start
< UINT64_MAX
/512);
916 size
= blkid_partition_get_size(pp
);
918 return errno_or_else(EIO
);
920 assert((uint64_t) size
< UINT64_MAX
/512);
922 /* While probing we need the non-diskseq device node name to access the thing, hence mask off
923 * DISSECT_IMAGE_DISKSEQ_DEVNODE. */
924 r
= make_partition_devname(devname
, diskseq
, nr
, flags
& ~DISSECT_IMAGE_DISKSEQ_DEVNODE
, &node
);
928 /* So here's the thing: after the main ("whole") block device popped up it might take a while
929 * before the kernel fully probed the partition table. Waiting for that to finish is icky in
930 * userspace. So here's what we do instead. We issue the BLKPG_ADD_PARTITION ioctl to add the
931 * partition ourselves, racing against the kernel. Good thing is: if this call fails with
932 * EBUSY then the kernel was quicker than us, and that's totally OK, the outcome is good for
933 * us: the device node will exist. If OTOH our call was successful we won the race. Which is
934 * also good as the outcome is the same: the partition block device exists, and we can use
937 * Kernel returns EBUSY if there's already a partition by that number or an overlapping
938 * partition already existent. */
940 if (FLAGS_SET(flags
, DISSECT_IMAGE_ADD_PARTITION_DEVICES
)) {
941 r
= block_device_add_partition(fd
, node
, nr
, (uint64_t) start
* 512, (uint64_t) size
* 512);
944 return log_debug_errno(r
, "BLKPG_ADD_PARTITION failed: %m");
946 log_debug_errno(r
, "Kernel was quicker than us in adding partition %i.", nr
);
948 log_debug("We were quicker than kernel in adding partition %i.", nr
);
952 const char *fstype
= NULL
, *label
;
953 sd_id128_t type_id
, id
;
954 GptPartitionType type
;
955 bool rw
= true, growfs
= false;
957 r
= blkid_partition_get_uuid_id128(pp
, &id
);
959 log_debug_errno(r
, "Failed to read partition UUID, ignoring: %m");
963 r
= blkid_partition_get_type_id128(pp
, &type_id
);
965 log_debug_errno(r
, "Failed to read partition type UUID, ignoring: %m");
969 type
= gpt_partition_type_from_uuid(type_id
);
971 label
= blkid_partition_get_name(pp
); /* libblkid returns NULL here if empty */
973 if (IN_SET(type
.designator
,
979 check_partition_flags(node
, pflags
,
980 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
982 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
985 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
986 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
988 } else if (type
.designator
== PARTITION_ESP
) {
990 /* Note that we don't check the SD_GPT_FLAG_NO_AUTO flag for the ESP, as it is
991 * not defined there. We instead check the SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as
992 * recommended by the UEFI spec (See "12.3.3 Number and Location of System
995 if (pflags
& SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL
)
1000 } else if (type
.designator
== PARTITION_ROOT
) {
1002 check_partition_flags(node
, pflags
,
1003 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1005 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1008 /* If a root ID is specified, ignore everything but the root id */
1009 if (!sd_id128_is_null(root_uuid
) && !sd_id128_equal(root_uuid
, id
))
1012 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1013 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1015 } else if (type
.designator
== PARTITION_ROOT_VERITY
) {
1017 check_partition_flags(node
, pflags
,
1018 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1020 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1023 m
->has_verity
= true;
1025 /* If no verity configuration is specified, then don't do verity */
1028 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
1031 /* If root hash is specified, then ignore everything but the root id */
1032 if (!sd_id128_is_null(root_verity_uuid
) && !sd_id128_equal(root_verity_uuid
, id
))
1035 fstype
= "DM_verity_hash";
1038 } else if (type
.designator
== PARTITION_ROOT_VERITY_SIG
) {
1040 check_partition_flags(node
, pflags
,
1041 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1043 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1046 m
->has_verity_sig
= true;
1050 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
1053 fstype
= "verity_hash_signature";
1056 } else if (type
.designator
== PARTITION_USR
) {
1058 check_partition_flags(node
, pflags
,
1059 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1061 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1064 /* If a usr ID is specified, ignore everything but the usr id */
1065 if (!sd_id128_is_null(usr_uuid
) && !sd_id128_equal(usr_uuid
, id
))
1068 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1069 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1071 } else if (type
.designator
== PARTITION_USR_VERITY
) {
1073 check_partition_flags(node
, pflags
,
1074 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1076 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1079 m
->has_verity
= true;
1083 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
1086 /* If usr hash is specified, then ignore everything but the usr id */
1087 if (!sd_id128_is_null(usr_verity_uuid
) && !sd_id128_equal(usr_verity_uuid
, id
))
1090 fstype
= "DM_verity_hash";
1093 } else if (type
.designator
== PARTITION_USR_VERITY_SIG
) {
1095 check_partition_flags(node
, pflags
,
1096 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1098 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1101 m
->has_verity_sig
= true;
1105 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
1108 fstype
= "verity_hash_signature";
1111 } else if (type
.designator
== PARTITION_SWAP
) {
1113 check_partition_flags(node
, pflags
, SD_GPT_FLAG_NO_AUTO
);
1115 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1118 /* Note: we don't set fstype = "swap" here, because we still need to probe if
1119 * it might be encrypted (i.e. fstype "crypt_LUKS") or unencrypted
1120 * (i.e. fstype "swap"), and the only way to figure that out is via fstype
1123 /* We don't have a designator for SD_GPT_LINUX_GENERIC so check the UUID instead. */
1124 } else if (sd_id128_equal(type
.uuid
, SD_GPT_LINUX_GENERIC
)) {
1126 check_partition_flags(node
, pflags
,
1127 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1129 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1133 multiple_generic
= true;
1136 generic_rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1137 generic_growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1139 generic_node
= TAKE_PTR(node
);
1142 } else if (type
.designator
== PARTITION_VAR
) {
1144 check_partition_flags(node
, pflags
,
1145 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1147 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1150 if (!FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_VAR_CHECK
)) {
1151 sd_id128_t var_uuid
;
1153 /* For /var we insist that the uuid of the partition matches the
1154 * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
1155 * ID. Why? Unlike the other partitions /var is inherently
1156 * installation specific, hence we need to be careful not to mount it
1157 * in the wrong installation. By hashing the partition UUID from
1158 * /etc/machine-id we can securely bind the partition to the
1161 r
= sd_id128_get_machine_app_specific(SD_GPT_VAR
, &var_uuid
);
1165 if (!sd_id128_equal(var_uuid
, id
)) {
1166 log_debug("Found a /var/ partition, but its UUID didn't match our expectations "
1167 "(found: " SD_ID128_UUID_FORMAT_STR
", expected: " SD_ID128_UUID_FORMAT_STR
"), ignoring.",
1168 SD_ID128_FORMAT_VAL(id
), SD_ID128_FORMAT_VAL(var_uuid
));
1173 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1174 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1177 if (type
.designator
!= _PARTITION_DESIGNATOR_INVALID
) {
1178 _cleanup_free_
char *t
= NULL
, *o
= NULL
, *l
= NULL
, *n
= NULL
;
1179 _cleanup_close_
int mount_node_fd
= -EBADF
;
1180 const char *options
= NULL
;
1182 r
= image_policy_may_use(policy
, type
.designator
);
1186 /* Policy says: ignore; Remember this fact, so that we later can distinguish between "found but ignored" and "not found at all" */
1188 if (!m
->partitions
[type
.designator
].found
)
1189 m
->partitions
[type
.designator
].ignored
= true;
1194 if (m
->partitions
[type
.designator
].found
) {
1197 /* For most partition types the first one we see wins. Except for the
1198 * rootfs and /usr, where we do a version compare of the label, and
1199 * let the newest version win. This permits a simple A/B versioning
1200 * scheme in OS images. */
1202 c
= compare_arch(type
.arch
, m
->partitions
[type
.designator
].architecture
);
1203 if (c
< 0) /* the arch we already found is better than the one we found now */
1205 if (c
== 0 && /* same arch? then go by version in label */
1206 (!partition_designator_is_versioned(type
.designator
) ||
1207 strverscmp_improved(label
, m
->partitions
[type
.designator
].label
) <= 0))
1210 dissected_partition_done(m
->partitions
+ type
.designator
);
1213 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
) &&
1214 type
.designator
!= PARTITION_SWAP
) {
1215 mount_node_fd
= open_partition(node
, /* is_partition = */ true, m
->loop
);
1216 if (mount_node_fd
< 0)
1217 return mount_node_fd
;
1220 r
= make_partition_devname(devname
, diskseq
, nr
, flags
, &n
);
1236 options
= mount_options_from_designator(mount_options
, type
.designator
);
1238 o
= strdup(options
);
1243 m
->partitions
[type
.designator
] = (DissectedPartition
) {
1248 .architecture
= type
.arch
,
1249 .node
= TAKE_PTR(n
),
1250 .fstype
= TAKE_PTR(t
),
1251 .label
= TAKE_PTR(l
),
1253 .mount_options
= TAKE_PTR(o
),
1254 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1255 .offset
= (uint64_t) start
* 512,
1256 .size
= (uint64_t) size
* 512,
1257 .gpt_flags
= pflags
,
1258 .fsmount_fd
= -EBADF
,
1262 } else if (is_mbr
) {
1264 switch (blkid_partition_get_type(pp
)) {
1266 case 0x83: /* Linux partition */
1268 if (pflags
!= 0x80) /* Bootable flag */
1272 multiple_generic
= true;
1276 generic_growfs
= false;
1277 generic_node
= TAKE_PTR(node
);
1282 case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
1283 _cleanup_close_
int mount_node_fd
= -EBADF
;
1284 _cleanup_free_
char *o
= NULL
, *n
= NULL
;
1285 sd_id128_t id
= SD_ID128_NULL
;
1286 const char *options
= NULL
;
1288 r
= image_policy_may_use(policy
, PARTITION_XBOOTLDR
);
1291 if (r
== 0) { /* policy says: ignore */
1292 if (!m
->partitions
[PARTITION_XBOOTLDR
].found
)
1293 m
->partitions
[PARTITION_XBOOTLDR
].ignored
= true;
1298 /* First one wins */
1299 if (m
->partitions
[PARTITION_XBOOTLDR
].found
)
1302 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
1303 mount_node_fd
= open_partition(node
, /* is_partition = */ true, m
->loop
);
1304 if (mount_node_fd
< 0)
1305 return mount_node_fd
;
1308 (void) blkid_partition_get_uuid_id128(pp
, &id
);
1310 r
= make_partition_devname(devname
, diskseq
, nr
, flags
, &n
);
1314 options
= mount_options_from_designator(mount_options
, PARTITION_XBOOTLDR
);
1316 o
= strdup(options
);
1321 m
->partitions
[PARTITION_XBOOTLDR
] = (DissectedPartition
) {
1326 .architecture
= _ARCHITECTURE_INVALID
,
1327 .node
= TAKE_PTR(n
),
1329 .mount_options
= TAKE_PTR(o
),
1330 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1331 .offset
= (uint64_t) start
* 512,
1332 .size
= (uint64_t) size
* 512,
1333 .fsmount_fd
= -EBADF
,
1341 if (!m
->partitions
[PARTITION_ROOT
].found
&&
1342 (m
->partitions
[PARTITION_ROOT_VERITY
].found
||
1343 m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
))
1344 return -EADDRNOTAVAIL
; /* Verity found but no matching rootfs? Something is off, refuse. */
1346 /* Hmm, we found a signature partition but no Verity data? Something is off. */
1347 if (m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_ROOT_VERITY
].found
)
1348 return -EADDRNOTAVAIL
;
1350 if (!m
->partitions
[PARTITION_USR
].found
&&
1351 (m
->partitions
[PARTITION_USR_VERITY
].found
||
1352 m
->partitions
[PARTITION_USR_VERITY_SIG
].found
))
1353 return -EADDRNOTAVAIL
; /* as above */
1356 if (m
->partitions
[PARTITION_USR_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_USR_VERITY
].found
)
1357 return -EADDRNOTAVAIL
;
1359 /* If root and /usr are combined then insist that the architecture matches */
1360 if (m
->partitions
[PARTITION_ROOT
].found
&&
1361 m
->partitions
[PARTITION_USR
].found
&&
1362 (m
->partitions
[PARTITION_ROOT
].architecture
>= 0 &&
1363 m
->partitions
[PARTITION_USR
].architecture
>= 0 &&
1364 m
->partitions
[PARTITION_ROOT
].architecture
!= m
->partitions
[PARTITION_USR
].architecture
))
1365 return -EADDRNOTAVAIL
;
1367 if (!m
->partitions
[PARTITION_ROOT
].found
&&
1368 !m
->partitions
[PARTITION_USR
].found
&&
1369 (flags
& DISSECT_IMAGE_GENERIC_ROOT
) &&
1370 (!verity
|| !verity
->root_hash
|| verity
->designator
!= PARTITION_USR
)) {
1372 /* OK, we found nothing usable, then check if there's a single generic partition, and use
1373 * that. If the root hash was set however, then we won't fall back to a generic node, because
1374 * the root hash decides. */
1376 /* If we didn't find a properly marked root partition, but we did find a single suitable
1377 * generic Linux partition, then use this as root partition, if the caller asked for it. */
1378 if (multiple_generic
)
1381 /* If we didn't find a generic node, then we can't fix this up either */
1383 r
= image_policy_may_use(policy
, PARTITION_ROOT
);
1387 /* Policy says: ignore; remember that we did */
1388 m
->partitions
[PARTITION_ROOT
].ignored
= true;
1390 _cleanup_close_
int mount_node_fd
= -EBADF
;
1391 _cleanup_free_
char *o
= NULL
, *n
= NULL
;
1392 const char *options
;
1394 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
1395 mount_node_fd
= open_partition(generic_node
, /* is_partition = */ true, m
->loop
);
1396 if (mount_node_fd
< 0)
1397 return mount_node_fd
;
1400 r
= make_partition_devname(devname
, diskseq
, generic_nr
, flags
, &n
);
1404 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
1406 o
= strdup(options
);
1411 assert(generic_nr
>= 0);
1412 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
1415 .growfs
= generic_growfs
,
1416 .partno
= generic_nr
,
1417 .architecture
= _ARCHITECTURE_INVALID
,
1418 .node
= TAKE_PTR(n
),
1419 .uuid
= generic_uuid
,
1420 .mount_options
= TAKE_PTR(o
),
1421 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1422 .offset
= UINT64_MAX
,
1424 .fsmount_fd
= -EBADF
,
1430 /* Check if we have a root fs if we are told to do check. /usr alone is fine too, but only if appropriate flag for that is set too */
1431 if (FLAGS_SET(flags
, DISSECT_IMAGE_REQUIRE_ROOT
) &&
1432 !(m
->partitions
[PARTITION_ROOT
].found
|| (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
1435 if (m
->partitions
[PARTITION_ROOT_VERITY
].found
) {
1436 /* We only support one verity partition per image, i.e. can't do for both /usr and root fs */
1437 if (m
->partitions
[PARTITION_USR_VERITY
].found
)
1440 /* We don't support verity enabled root with a split out /usr. Neither with nor without
1441 * verity there. (Note that we do support verity-less root with verity-full /usr, though.) */
1442 if (m
->partitions
[PARTITION_USR
].found
)
1443 return -EADDRNOTAVAIL
;
1447 /* If a verity designator is specified, then insist that the matching partition exists */
1448 if (verity
->designator
>= 0 && !m
->partitions
[verity
->designator
].found
)
1449 return -EADDRNOTAVAIL
;
1451 bool have_verity_sig_partition
;
1452 if (verity
->designator
>= 0)
1453 have_verity_sig_partition
= m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR_VERITY_SIG
: PARTITION_ROOT_VERITY_SIG
].found
;
1455 have_verity_sig_partition
= m
->partitions
[PARTITION_USR_VERITY_SIG
].found
|| m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
;
1457 if (verity
->root_hash
) {
1458 /* If we have an explicit root hash and found the partitions for it, then we are ready to use
1459 * Verity, set things up for it */
1461 if (verity
->designator
< 0 || verity
->designator
== PARTITION_ROOT
) {
1462 if (!m
->partitions
[PARTITION_ROOT_VERITY
].found
|| !m
->partitions
[PARTITION_ROOT
].found
)
1463 return -EADDRNOTAVAIL
;
1465 /* If we found a verity setup, then the root partition is necessarily read-only. */
1466 m
->partitions
[PARTITION_ROOT
].rw
= false;
1467 m
->verity_ready
= true;
1470 assert(verity
->designator
== PARTITION_USR
);
1472 if (!m
->partitions
[PARTITION_USR_VERITY
].found
|| !m
->partitions
[PARTITION_USR
].found
)
1473 return -EADDRNOTAVAIL
;
1475 m
->partitions
[PARTITION_USR
].rw
= false;
1476 m
->verity_ready
= true;
1479 if (m
->verity_ready
)
1480 m
->verity_sig_ready
= verity
->root_hash_sig
|| have_verity_sig_partition
;
1482 } else if (have_verity_sig_partition
) {
1484 /* If we found an embedded signature partition, we are ready, too. */
1486 m
->verity_ready
= m
->verity_sig_ready
= true;
1487 if (verity
->designator
>= 0)
1488 m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR
: PARTITION_ROOT
].rw
= false;
1489 else if (m
->partitions
[PARTITION_USR_VERITY_SIG
].found
)
1490 m
->partitions
[PARTITION_USR
].rw
= false;
1491 else if (m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
)
1492 m
->partitions
[PARTITION_ROOT
].rw
= false;
1498 /* After we discovered all partitions let's see if the verity requirements match the policy. (Note:
1499 * we don't check encryption requirements here, because we haven't probed the file system yet, hence
1500 * don't know if this is encrypted or not) */
1501 for (PartitionDesignator di
= 0; di
< _PARTITION_DESIGNATOR_MAX
; di
++) {
1502 PartitionDesignator vi
, si
;
1503 PartitionPolicyFlags found_flags
;
1505 any
= any
|| m
->partitions
[di
].found
;
1507 vi
= partition_verity_of(di
);
1508 si
= partition_verity_sig_of(di
);
1510 /* Determine the verity protection level for this partition. */
1511 found_flags
= m
->partitions
[di
].found
?
1512 (vi
>= 0 && m
->partitions
[vi
].found
?
1513 (si
>= 0 && m
->partitions
[si
].found
? PARTITION_POLICY_SIGNED
: PARTITION_POLICY_VERITY
) :
1514 PARTITION_POLICY_ENCRYPTED
|PARTITION_POLICY_UNPROTECTED
) :
1515 (m
->partitions
[di
].ignored
? PARTITION_POLICY_UNUSED
: PARTITION_POLICY_ABSENT
);
1517 r
= image_policy_check_protection(policy
, di
, found_flags
);
1521 if (m
->partitions
[di
].found
) {
1522 r
= image_policy_check_partition_flags(policy
, di
, m
->partitions
[di
].gpt_flags
);
1528 if (!any
&& !FLAGS_SET(flags
, DISSECT_IMAGE_ALLOW_EMPTY
))
1531 r
= dissected_image_probe_filesystems(m
, fd
, policy
);
1539 int dissect_image_file(
1541 const VeritySettings
*verity
,
1542 const MountOptions
*mount_options
,
1543 const ImagePolicy
*image_policy
,
1544 DissectImageFlags flags
,
1545 DissectedImage
**ret
) {
1548 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
1549 _cleanup_close_
int fd
= -EBADF
;
1555 fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
1559 if (fstat(fd
, &st
) < 0)
1562 r
= stat_verify_regular(&st
);
1566 r
= dissected_image_new(path
, &m
);
1570 m
->image_size
= st
.st_size
;
1572 r
= probe_sector_size(fd
, &m
->sector_size
);
1576 r
= dissect_image(m
, fd
, path
, verity
, mount_options
, image_policy
, flags
);
1588 int dissect_log_error(int log_level
, int r
, const char *name
, const VeritySettings
*verity
) {
1589 assert(log_level
>= 0 && log_level
<= LOG_DEBUG
);
1594 case 0 ... INT_MAX
: /* success! */
1598 return log_full_errno(log_level
, r
, "Dissecting images is not supported, compiled without blkid support.");
1601 return log_full_errno(log_level
, r
, "%s: Couldn't identify a suitable partition table or file system.", name
);
1604 return log_full_errno(log_level
, r
, "%s: The image does not pass os-release/extension-release validation.", name
);
1606 case -EADDRNOTAVAIL
:
1607 return log_full_errno(log_level
, r
, "%s: No root partition for specified root hash found.", name
);
1610 return log_full_errno(log_level
, r
, "%s: Multiple suitable root partitions found in image.", name
);
1613 return log_full_errno(log_level
, r
, "%s: No suitable root partition found in image.", name
);
1615 case -EPROTONOSUPPORT
:
1616 return log_full_errno(log_level
, r
, "Device '%s' is a loopback block device with partition scanning turned off, please turn it on.", name
);
1619 return log_full_errno(log_level
, r
, "%s: Image is not a block device.", name
);
1622 return log_full_errno(log_level
, r
,
1623 "Combining partitioned images (such as '%s') with external Verity data (such as '%s') not supported. "
1624 "(Consider setting $SYSTEMD_DISSECT_VERITY_SIDECAR=0 to disable automatic discovery of external Verity data.)",
1625 name
, strna(verity
? verity
->data_path
: NULL
));
1628 return log_full_errno(log_level
, r
, "%s: image does not match image policy.", name
);
1631 return log_full_errno(log_level
, r
, "%s: no suitable partitions found.", name
);
1634 return log_full_errno(log_level
, r
, "%s: cannot dissect image: %m", name
);
1638 int dissect_image_file_and_warn(
1640 const VeritySettings
*verity
,
1641 const MountOptions
*mount_options
,
1642 const ImagePolicy
*image_policy
,
1643 DissectImageFlags flags
,
1644 DissectedImage
**ret
) {
1646 return dissect_log_error(
1648 dissect_image_file(path
, verity
, mount_options
, image_policy
, flags
, ret
),
1653 DissectedImage
* dissected_image_unref(DissectedImage
*m
) {
1657 /* First, clear dissected partitions. */
1658 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++)
1659 dissected_partition_done(m
->partitions
+ i
);
1661 /* Second, free decrypted images. This must be after dissected_partition_done(), as freeing
1662 * DecryptedImage may try to deactivate partitions. */
1663 decrypted_image_unref(m
->decrypted_image
);
1665 /* Third, unref LoopDevice. This must be called after the above two, as freeing LoopDevice may try to
1666 * remove existing partitions on the loopback block device. */
1667 loop_device_unref(m
->loop
);
1669 free(m
->image_name
);
1671 strv_free(m
->machine_info
);
1672 strv_free(m
->os_release
);
1673 strv_free(m
->initrd_release
);
1674 strv_free(m
->confext_release
);
1675 strv_free(m
->sysext_release
);
1680 static int is_loop_device(const char *path
) {
1681 char s
[SYS_BLOCK_PATH_MAX("/../loop/")];
1686 if (stat(path
, &st
) < 0)
1689 if (!S_ISBLK(st
.st_mode
))
1692 xsprintf_sys_block_path(s
, "/loop/", st
.st_dev
);
1693 if (access(s
, F_OK
) < 0) {
1694 if (errno
!= ENOENT
)
1697 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
1698 xsprintf_sys_block_path(s
, "/../loop/", st
.st_dev
);
1699 if (access(s
, F_OK
) < 0)
1700 return errno
== ENOENT
? false : -errno
;
1706 static int run_fsck(int node_fd
, const char *fstype
) {
1710 assert(node_fd
>= 0);
1713 r
= fsck_exists_for_fstype(fstype
);
1715 log_debug_errno(r
, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype
);
1719 log_debug("Not checking partition %s, as fsck for %s does not exist.", FORMAT_PROC_FD_PATH(node_fd
), fstype
);
1726 &node_fd
, 1, /* Leave the node fd open */
1727 FORK_RESET_SIGNALS
|FORK_CLOSE_ALL_FDS
|FORK_RLIMIT_NOFILE_SAFE
|FORK_DEATHSIG_SIGTERM
|FORK_REARRANGE_STDIO
|FORK_CLOEXEC_OFF
,
1730 return log_debug_errno(r
, "Failed to fork off fsck: %m");
1733 execlp("fsck", "fsck", "-aT", FORMAT_PROC_FD_PATH(node_fd
), NULL
);
1735 log_debug_errno(errno
, "Failed to execl() fsck: %m");
1736 _exit(FSCK_OPERATIONAL_ERROR
);
1739 exit_status
= wait_for_terminate_and_check("fsck", pid
, 0);
1740 if (exit_status
< 0)
1741 return log_debug_errno(exit_status
, "Failed to fork off fsck: %m");
1743 if ((exit_status
& ~FSCK_ERROR_CORRECTED
) != FSCK_SUCCESS
) {
1744 log_debug("fsck failed with exit status %i.", exit_status
);
1746 if ((exit_status
& (FSCK_SYSTEM_SHOULD_REBOOT
|FSCK_ERRORS_LEFT_UNCORRECTED
)) != 0)
1747 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
), "File system is corrupted, refusing.");
1749 log_debug("Ignoring fsck error.");
1755 static int fs_grow(const char *node_path
, int mount_fd
, const char *mount_path
) {
1756 _cleanup_close_
int _mount_fd
= -EBADF
, node_fd
= -EBADF
;
1757 uint64_t size
, newsize
;
1762 assert(mount_fd
>= 0 || mount_path
);
1764 node_fd
= open(node_path
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
1766 return log_debug_errno(errno
, "Failed to open node device %s: %m", node_path
);
1768 r
= blockdev_get_device_size(node_fd
, &size
);
1770 return log_debug_errno(r
, "Failed to get block device size of %s: %m", node_path
);
1775 _mount_fd
= open(mount_path
, O_RDONLY
|O_DIRECTORY
|O_CLOEXEC
);
1777 return log_debug_errno(errno
, "Failed to open mounted file system %s: %m", mount_path
);
1779 mount_fd
= _mount_fd
;
1781 mount_fd
= fd_reopen_condition(mount_fd
, O_RDONLY
|O_DIRECTORY
|O_CLOEXEC
, O_RDONLY
|O_DIRECTORY
|O_CLOEXEC
, &_mount_fd
);
1783 return log_debug_errno(errno
, "Failed to reopen mount node: %m");
1786 id
= mount_path
?: node_path
;
1788 log_debug("Resizing \"%s\" to %"PRIu64
" bytes...", id
, size
);
1789 r
= resize_fs(mount_fd
, size
, &newsize
);
1791 return log_debug_errno(r
, "Failed to resize \"%s\" to %"PRIu64
" bytes: %m", id
, size
);
1793 if (newsize
== size
)
1794 log_debug("Successfully resized \"%s\" to %s bytes.",
1795 id
, FORMAT_BYTES(newsize
));
1797 assert(newsize
< size
);
1798 log_debug("Successfully resized \"%s\" to %s bytes (%"PRIu64
" bytes lost due to blocksize).",
1799 id
, FORMAT_BYTES(newsize
), size
- newsize
);
1805 int partition_pick_mount_options(
1806 PartitionDesignator d
,
1811 unsigned long *ret_ms_flags
) {
1813 _cleanup_free_
char *options
= NULL
;
1815 assert(ret_options
);
1817 /* Selects a baseline of bind mount flags, that should always apply.
1819 * Firstly, we set MS_NODEV universally on all mounts, since we don't want to allow device nodes outside of /dev/.
1821 * On /var/tmp/ we'll also set MS_NOSUID, same as we set for /tmp/ on the host.
1823 * On the ESP and XBOOTLDR partitions we'll also disable symlinks, and execution. These file systems
1824 * are generally untrusted (i.e. not encrypted or authenticated), and typically VFAT hence we should
1825 * be as restrictive as possible, and this shouldn't hurt, since the functionality is not available
1828 unsigned long flags
= MS_NODEV
;
1836 case PARTITION_XBOOTLDR
:
1837 flags
|= MS_NOSUID
|MS_NOEXEC
|ms_nosymfollow_supported();
1839 /* The ESP might contain a pre-boot random seed. Let's make this unaccessible to regular
1840 * userspace. ESP/XBOOTLDR is almost certainly VFAT, hence if we don't know assume it is. */
1841 if (!fstype
|| fstype_can_umask(fstype
))
1842 if (!strextend_with_separator(&options
, ",", "umask=0077"))
1854 /* So, when you request MS_RDONLY from ext4, then this means nothing. It happily still writes to the
1855 * backing storage. What's worse, the BLKRO[GS]ET flag and (in case of loopback devices)
1856 * LO_FLAGS_READ_ONLY don't mean anything, they affect userspace accesses only, and write accesses
1857 * from the upper file system still get propagated through to the underlying file system,
1858 * unrestricted. To actually get ext4/xfs/btrfs to stop writing to the device we need to specify
1859 * "norecovery" as mount option, in addition to MS_RDONLY. Yes, this sucks, since it means we need to
1860 * carry a per file system table here.
1862 * Note that this means that we might not be able to mount corrupted file systems as read-only
1863 * anymore (since in some cases the kernel implementations will refuse mounting when corrupted,
1864 * read-only and "norecovery" is specified). But I think for the case of automatically determined
1865 * mount options for loopback devices this is the right choice, since otherwise using the same
1866 * loopback file twice even in read-only mode, is going to fail badly sooner or later. The use case of
1867 * making reuse of the immutable images "just work" is more relevant to us than having read-only
1868 * access that actually modifies stuff work on such image files. Or to say this differently: if
1869 * people want their file systems to be fixed up they should just open them in writable mode, where
1870 * all these problems don't exist. */
1871 if (!rw
&& fstype
&& fstype_can_norecovery(fstype
))
1872 if (!strextend_with_separator(&options
, ",", "norecovery"))
1875 if (discard
&& fstype
&& fstype_can_discard(fstype
))
1876 if (!strextend_with_separator(&options
, ",", "discard"))
1879 if (!ret_ms_flags
) /* Fold flags into option string if ret_flags specified as NULL */
1880 if (!strextend_with_separator(&options
, ",",
1881 FLAGS_SET(flags
, MS_RDONLY
) ? "ro" : "rw",
1882 FLAGS_SET(flags
, MS_NODEV
) ? "nodev" : "dev",
1883 FLAGS_SET(flags
, MS_NOSUID
) ? "nosuid" : "suid",
1884 FLAGS_SET(flags
, MS_NOEXEC
) ? "noexec" : "exec",
1885 FLAGS_SET(flags
, MS_NOSYMFOLLOW
) ? "nosymfollow" : NULL
))
1886 /* NB: we suppress 'symfollow' here, since it's the default, and old /bin/mount might not know it */
1890 *ret_ms_flags
= flags
;
1892 *ret_options
= TAKE_PTR(options
);
1896 static bool need_user_mapping(uid_t uid_shift
, uid_t uid_range
) {
1898 if (!uid_is_valid(uid_shift
))
1901 return uid_shift
!= 0 || uid_range
!= UINT32_MAX
;
1904 static int mount_partition(
1905 PartitionDesignator d
,
1906 DissectedPartition
*m
,
1908 const char *directory
,
1912 DissectImageFlags flags
) {
1914 _cleanup_free_
char *chased
= NULL
, *options
= NULL
;
1915 const char *p
= NULL
, *node
, *fstype
= NULL
;
1916 bool rw
, discard
, grow
;
1917 unsigned long ms_flags
;
1925 /* Check the various combinations when we can't do anything anymore */
1926 if (m
->fsmount_fd
< 0 && m
->mount_node_fd
< 0)
1928 if (m
->fsmount_fd
>= 0 && !where
)
1930 if (!where
&& m
->mount_node_fd
< 0)
1933 if (m
->fsmount_fd
< 0) {
1934 fstype
= dissected_partition_fstype(m
);
1936 return -EAFNOSUPPORT
;
1938 /* We are looking at an encrypted partition? This either means stacked encryption, or the
1939 * caller didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error
1941 if (streq(fstype
, "crypto_LUKS"))
1944 r
= dissect_fstype_ok(fstype
);
1948 return -EIDRM
; /* Recognizable error */
1951 node
= m
->mount_node_fd
< 0 ? NULL
: FORMAT_PROC_FD_PATH(m
->mount_node_fd
);
1952 rw
= m
->rw
&& !(flags
& DISSECT_IMAGE_MOUNT_READ_ONLY
);
1954 discard
= ((flags
& DISSECT_IMAGE_DISCARD
) ||
1955 ((flags
& DISSECT_IMAGE_DISCARD_ON_LOOP
) && (m
->node
&& is_loop_device(m
->node
) > 0)));
1957 grow
= rw
&& m
->growfs
&& FLAGS_SET(flags
, DISSECT_IMAGE_GROWFS
);
1959 if (FLAGS_SET(flags
, DISSECT_IMAGE_FSCK
) && rw
&& m
->mount_node_fd
>= 0 && m
->fsmount_fd
< 0) {
1960 r
= run_fsck(m
->mount_node_fd
, fstype
);
1967 /* Automatically create missing mount points inside the image, if necessary. */
1968 r
= mkdir_p_root(where
, directory
, uid_shift
, (gid_t
) uid_shift
, 0755, NULL
);
1969 if (r
< 0 && r
!= -EROFS
)
1972 r
= chase(directory
, where
, CHASE_PREFIX_ROOT
, &chased
, NULL
);
1978 /* Create top-level mount if missing – but only if this is asked for. This won't modify the
1979 * image (as the branch above does) but the host hierarchy, and the created directory might
1980 * survive our mount in the host hierarchy hence. */
1981 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
1982 r
= mkdir_p(where
, 0755);
1991 if (m
->fsmount_fd
< 0) {
1992 r
= partition_pick_mount_options(d
, fstype
, rw
, discard
, &options
, &ms_flags
);
1996 if (need_user_mapping(uid_shift
, uid_range
) && fstype_can_uid_gid(fstype
)) {
1997 _cleanup_free_
char *uid_option
= NULL
;
1999 if (asprintf(&uid_option
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
2002 if (!strextend_with_separator(&options
, ",", uid_option
))
2005 userns_fd
= -EBADF
; /* Not needed */
2008 if (!isempty(m
->mount_options
))
2009 if (!strextend_with_separator(&options
, ",", m
->mount_options
))
2014 if (m
->fsmount_fd
>= 0) {
2015 /* Case #1: Attach existing fsmount fd to the file system */
2017 r
= mount_exchange_graceful(
2020 FLAGS_SET(flags
, DISSECT_IMAGE_TRY_ATOMIC_MOUNT_EXCHANGE
));
2022 return log_debug_errno(r
, "Failed to mount image on '%s': %m", p
);
2027 /* Case #2: Mount directly into place */
2028 r
= mount_nofollow_verbose(LOG_DEBUG
, node
, p
, fstype
, ms_flags
, options
);
2033 (void) fs_grow(node
, -EBADF
, p
);
2035 if (userns_fd
>= 0) {
2036 r
= remount_idmap_fd(STRV_MAKE(p
), userns_fd
);
2044 /* Case #3: Create fsmount fd */
2046 m
->fsmount_fd
= make_fsmount(LOG_DEBUG
, node
, fstype
, ms_flags
, options
, userns_fd
);
2047 if (m
->fsmount_fd
< 0)
2048 return m
->fsmount_fd
;
2051 (void) fs_grow(node
, m
->fsmount_fd
, NULL
);
2057 static int mount_root_tmpfs(const char *where
, uid_t uid_shift
, uid_t uid_range
, DissectImageFlags flags
) {
2058 _cleanup_free_
char *options
= NULL
;
2063 /* For images that contain /usr/ but no rootfs, let's mount rootfs as tmpfs */
2065 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
2066 r
= mkdir_p(where
, 0755);
2071 if (need_user_mapping(uid_shift
, uid_range
)) {
2072 if (asprintf(&options
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
2076 r
= mount_nofollow_verbose(LOG_DEBUG
, "rootfs", where
, "tmpfs", MS_NODEV
, options
);
2083 static int mount_point_is_available(const char *where
, const char *path
, bool missing_ok
) {
2084 _cleanup_free_
char *p
= NULL
;
2087 /* Check whether <path> is suitable as a mountpoint, i.e. is an empty directory
2088 * or does not exist at all (when missing_ok). */
2090 r
= chase(path
, where
, CHASE_PREFIX_ROOT
, &p
, NULL
);
2094 return log_debug_errno(r
, "Failed to chase \"%s\": %m", path
);
2096 r
= dir_is_empty(p
, /* ignore_hidden_or_backup= */ false);
2100 return log_debug_errno(r
, "Failed to check directory \"%s\": %m", p
);
2104 int dissected_image_mount(
2110 DissectImageFlags flags
) {
2112 _cleanup_close_
int my_userns_fd
= -EBADF
;
2117 /* If 'where' is NULL then we'll use the new mount API to create fsmount() fds for the mounts and
2118 * store them in DissectedPartition.fsmount_fd.
2120 * If 'where' is not NULL then we'll either mount the partitions to the right places ourselves,
2121 * or use DissectedPartition.fsmount_fd and bind it to the right places.
2123 * This allows splitting the setting up up the superblocks and the binding to file systems paths into
2124 * two distinct and differently privileged components: one that gets the fsmount fds, and the other
2125 * that then applies them.
2129 * -ENXIO → No root partition found
2130 * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release/extension-release file found
2131 * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet
2132 * -EUCLEAN → fsck for file system failed
2133 * -EBUSY → File system already mounted/used elsewhere (kernel)
2134 * -EAFNOSUPPORT → File system type not supported or not known
2135 * -EIDRM → File system is not among allowlisted "common" file systems
2138 if (!where
&& (flags
& (DISSECT_IMAGE_VALIDATE_OS
|DISSECT_IMAGE_VALIDATE_OS_EXT
)) != 0)
2139 return -EOPNOTSUPP
; /* for now, not supported */
2141 if (!(m
->partitions
[PARTITION_ROOT
].found
||
2142 (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
2143 return -ENXIO
; /* Require a root fs or at least a /usr/ fs (the latter is subject to a flag of its own) */
2145 if (userns_fd
< 0 && need_user_mapping(uid_shift
, uid_range
) && FLAGS_SET(flags
, DISSECT_IMAGE_MOUNT_IDMAPPED
)) {
2147 my_userns_fd
= make_userns(uid_shift
, uid_range
, UID_INVALID
, REMOUNT_IDMAPPING_HOST_ROOT
);
2148 if (my_userns_fd
< 0)
2149 return my_userns_fd
;
2151 userns_fd
= my_userns_fd
;
2154 if ((flags
& DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
) == 0) {
2156 /* First mount the root fs. If there's none we use a tmpfs. */
2157 if (m
->partitions
[PARTITION_ROOT
].found
) {
2158 r
= mount_partition(PARTITION_ROOT
, m
->partitions
+ PARTITION_ROOT
, where
, NULL
, uid_shift
, uid_range
, userns_fd
, flags
);
2163 r
= mount_root_tmpfs(where
, uid_shift
, uid_range
, flags
);
2168 /* For us mounting root always means mounting /usr as well */
2169 r
= mount_partition(PARTITION_USR
, m
->partitions
+ PARTITION_USR
, where
, "/usr", uid_shift
, uid_range
, userns_fd
, flags
);
2174 if ((flags
& DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
) == 0 &&
2175 (flags
& (DISSECT_IMAGE_VALIDATE_OS
|DISSECT_IMAGE_VALIDATE_OS_EXT
)) != 0) {
2176 /* If either one of the validation flags are set, ensure that the image qualifies as
2177 * one or the other (or both). */
2182 if (FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS
)) {
2183 r
= path_is_os_tree(where
);
2189 if (!ok
&& FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS_EXT
) && m
->image_name
) {
2190 r
= extension_has_forbidden_content(where
);
2194 r
= path_is_extension_tree(IMAGE_SYSEXT
, where
, m
->image_name
, FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_EXTENSION_CHECK
));
2196 r
= path_is_extension_tree(IMAGE_CONFEXT
, where
, m
->image_name
, FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_EXTENSION_CHECK
));
2208 if (flags
& DISSECT_IMAGE_MOUNT_ROOT_ONLY
)
2211 r
= mount_partition(PARTITION_HOME
, m
->partitions
+ PARTITION_HOME
, where
, "/home", uid_shift
, uid_range
, userns_fd
, flags
);
2215 r
= mount_partition(PARTITION_SRV
, m
->partitions
+ PARTITION_SRV
, where
, "/srv", uid_shift
, uid_range
, userns_fd
, flags
);
2219 r
= mount_partition(PARTITION_VAR
, m
->partitions
+ PARTITION_VAR
, where
, "/var", uid_shift
, uid_range
, userns_fd
, flags
);
2223 r
= mount_partition(PARTITION_TMP
, m
->partitions
+ PARTITION_TMP
, where
, "/var/tmp", uid_shift
, uid_range
, userns_fd
, flags
);
2227 int slash_boot_is_available
= 0;
2229 r
= slash_boot_is_available
= mount_point_is_available(where
, "/boot", /* missing_ok = */ true);
2233 if (!where
|| slash_boot_is_available
) {
2234 r
= mount_partition(PARTITION_XBOOTLDR
, m
->partitions
+ PARTITION_XBOOTLDR
, where
, "/boot", uid_shift
, uid_range
, userns_fd
, flags
);
2237 slash_boot_is_available
= !r
;
2240 if (m
->partitions
[PARTITION_ESP
].found
) {
2241 const char *esp_path
= NULL
;
2244 /* Mount the ESP to /boot/ if it exists and is empty and we didn't already mount the
2245 * XBOOTLDR partition into it. Otherwise, use /efi instead, but only if it exists
2248 if (slash_boot_is_available
) {
2249 r
= mount_point_is_available(where
, "/boot", /* missing_ok = */ false);
2257 r
= mount_point_is_available(where
, "/efi", /* missing_ok = */ true);
2265 /* OK, let's mount the ESP now (possibly creating the dir if missing) */
2266 r
= mount_partition(PARTITION_ESP
, m
->partitions
+ PARTITION_ESP
, where
, esp_path
, uid_shift
, uid_range
, userns_fd
, flags
);
2274 int dissected_image_mount_and_warn(
2280 DissectImageFlags flags
) {
2286 r
= dissected_image_mount(m
, where
, uid_shift
, uid_range
, userns_fd
, flags
);
2288 return log_error_errno(r
, "Not root file system found in image.");
2289 if (r
== -EMEDIUMTYPE
)
2290 return log_error_errno(r
, "No suitable os-release/extension-release file in image found.");
2292 return log_error_errno(r
, "Encrypted file system discovered, but decryption not requested.");
2294 return log_error_errno(r
, "File system check on image failed.");
2296 return log_error_errno(r
, "File system already mounted elsewhere.");
2297 if (r
== -EAFNOSUPPORT
)
2298 return log_error_errno(r
, "File system type not supported or not known.");
2300 return log_error_errno(r
, "File system is too uncommon, refused.");
2302 return log_error_errno(r
, "Failed to mount image: %m");
2307 #if HAVE_LIBCRYPTSETUP
2308 struct DecryptedPartition
{
2309 struct crypt_device
*device
;
2315 typedef struct DecryptedPartition DecryptedPartition
;
2317 struct DecryptedImage
{
2319 DecryptedPartition
*decrypted
;
2323 static DecryptedImage
* decrypted_image_free(DecryptedImage
*d
) {
2324 #if HAVE_LIBCRYPTSETUP
2330 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
2331 DecryptedPartition
*p
= d
->decrypted
+ i
;
2333 if (p
->device
&& p
->name
&& !p
->relinquished
) {
2334 _cleanup_free_
char *node
= NULL
;
2336 node
= path_join("/dev/mapper", p
->name
);
2338 r
= btrfs_forget_device(node
);
2339 if (r
< 0 && r
!= -ENOENT
)
2340 log_debug_errno(r
, "Failed to forget btrfs device %s, ignoring: %m", node
);
2344 /* Let's deactivate lazily, as the dm volume may be already/still used by other processes. */
2345 r
= sym_crypt_deactivate_by_name(p
->device
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
2347 log_debug_errno(r
, "Failed to deactivate encrypted partition %s", p
->name
);
2351 sym_crypt_free(p
->device
);
2361 DEFINE_TRIVIAL_REF_UNREF_FUNC(DecryptedImage
, decrypted_image
, decrypted_image_free
);
2363 #if HAVE_LIBCRYPTSETUP
2364 static int decrypted_image_new(DecryptedImage
**ret
) {
2365 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
2369 d
= new(DecryptedImage
, 1);
2373 *d
= (DecryptedImage
) {
2381 static int make_dm_name_and_node(const void *original_node
, const char *suffix
, char **ret_name
, char **ret_node
) {
2382 _cleanup_free_
char *name
= NULL
, *node
= NULL
;
2385 assert(original_node
);
2390 base
= strrchr(original_node
, '/');
2392 base
= original_node
;
2398 name
= strjoin(base
, suffix
);
2401 if (!filename_is_valid(name
))
2404 node
= path_join(sym_crypt_get_dir(), name
);
2408 *ret_name
= TAKE_PTR(name
);
2409 *ret_node
= TAKE_PTR(node
);
2414 static int decrypt_partition(
2415 DissectedPartition
*m
,
2416 const char *passphrase
,
2417 DissectImageFlags flags
,
2418 DecryptedImage
*d
) {
2420 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
2421 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2422 _cleanup_close_
int fd
= -EBADF
;
2428 if (!m
->found
|| !m
->node
|| !m
->fstype
)
2431 if (!streq(m
->fstype
, "crypto_LUKS"))
2437 r
= dlopen_cryptsetup();
2441 r
= make_dm_name_and_node(m
->node
, "-decrypted", &name
, &node
);
2445 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
2448 r
= sym_crypt_init(&cd
, m
->node
);
2450 return log_debug_errno(r
, "Failed to initialize dm-crypt: %m");
2452 cryptsetup_enable_logging(cd
);
2454 r
= sym_crypt_load(cd
, CRYPT_LUKS
, NULL
);
2456 return log_debug_errno(r
, "Failed to load LUKS metadata: %m");
2458 r
= sym_crypt_activate_by_passphrase(cd
, name
, CRYPT_ANY_SLOT
, passphrase
, strlen(passphrase
),
2459 ((flags
& DISSECT_IMAGE_DEVICE_READ_ONLY
) ? CRYPT_ACTIVATE_READONLY
: 0) |
2460 ((flags
& DISSECT_IMAGE_DISCARD_ON_CRYPTO
) ? CRYPT_ACTIVATE_ALLOW_DISCARDS
: 0));
2462 log_debug_errno(r
, "Failed to activate LUKS device: %m");
2463 return r
== -EPERM
? -EKEYREJECTED
: r
;
2466 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2468 return log_debug_errno(errno
, "Failed to open %s: %m", node
);
2470 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
2471 .name
= TAKE_PTR(name
),
2472 .device
= TAKE_PTR(cd
),
2475 m
->decrypted_node
= TAKE_PTR(node
);
2476 close_and_replace(m
->mount_node_fd
, fd
);
2481 static int verity_can_reuse(
2482 const VeritySettings
*verity
,
2484 struct crypt_device
**ret_cd
) {
2486 /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
2487 _cleanup_free_
char *root_hash_existing
= NULL
;
2488 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2489 struct crypt_params_verity crypt_params
= {};
2490 size_t root_hash_existing_size
;
2497 r
= sym_crypt_init_by_name(&cd
, name
);
2499 return log_debug_errno(r
, "Error opening verity device, crypt_init_by_name failed: %m");
2501 cryptsetup_enable_logging(cd
);
2503 r
= sym_crypt_get_verity_info(cd
, &crypt_params
);
2505 return log_debug_errno(r
, "Error opening verity device, crypt_get_verity_info failed: %m");
2507 root_hash_existing_size
= verity
->root_hash_size
;
2508 root_hash_existing
= malloc0(root_hash_existing_size
);
2509 if (!root_hash_existing
)
2512 r
= sym_crypt_volume_key_get(cd
, CRYPT_ANY_SLOT
, root_hash_existing
, &root_hash_existing_size
, NULL
, 0);
2514 return log_debug_errno(r
, "Error opening verity device, crypt_volume_key_get failed: %m");
2515 if (verity
->root_hash_size
!= root_hash_existing_size
||
2516 memcmp(root_hash_existing
, verity
->root_hash
, verity
->root_hash_size
) != 0)
2517 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but root hashes are different.");
2519 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
2520 /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the
2521 * same settings, so that a previous unsigned mount will not be reused if the user asks to use
2522 * signing for the new one, and vice versa. */
2523 if (!!verity
->root_hash_sig
!= !!(crypt_params
.flags
& CRYPT_VERITY_ROOT_HASH_SIGNATURE
))
2524 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but signature settings are not the same.");
2527 *ret_cd
= TAKE_PTR(cd
);
2531 static char* dm_deferred_remove_clean(char *name
) {
2535 (void) sym_crypt_deactivate_by_name(NULL
, name
, CRYPT_DEACTIVATE_DEFERRED
);
2538 DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean
);
2540 static int validate_signature_userspace(const VeritySettings
*verity
) {
2542 _cleanup_(sk_X509_free_allp
) STACK_OF(X509
) *sk
= NULL
;
2543 _cleanup_strv_free_
char **certs
= NULL
;
2544 _cleanup_(PKCS7_freep
) PKCS7
*p7
= NULL
;
2545 _cleanup_free_
char *s
= NULL
;
2546 _cleanup_(BIO_freep
) BIO
*bio
= NULL
; /* 'bio' must be freed first, 's' second, hence keep this order
2547 * of declaration in place, please */
2548 const unsigned char *d
;
2552 assert(verity
->root_hash
);
2553 assert(verity
->root_hash_sig
);
2555 /* Because installing a signature certificate into the kernel chain is so messy, let's optionally do
2556 * userspace validation. */
2558 r
= conf_files_list_nulstr(&certs
, ".crt", NULL
, CONF_FILES_REGULAR
|CONF_FILES_FILTER_MASKED
, CONF_PATHS_NULSTR("verity.d"));
2560 return log_debug_errno(r
, "Failed to enumerate certificates: %m");
2561 if (strv_isempty(certs
)) {
2562 log_debug("No userspace dm-verity certificates found.");
2566 d
= verity
->root_hash_sig
;
2567 p7
= d2i_PKCS7(NULL
, &d
, (long) verity
->root_hash_sig_size
);
2569 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to parse PKCS7 DER signature data.");
2571 s
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2573 return log_oom_debug();
2575 bio
= BIO_new_mem_buf(s
, strlen(s
));
2577 return log_oom_debug();
2579 sk
= sk_X509_new_null();
2581 return log_oom_debug();
2583 STRV_FOREACH(i
, certs
) {
2584 _cleanup_(X509_freep
) X509
*c
= NULL
;
2585 _cleanup_fclose_
FILE *f
= NULL
;
2587 f
= fopen(*i
, "re");
2589 log_debug_errno(errno
, "Failed to open '%s', ignoring: %m", *i
);
2593 c
= PEM_read_X509(f
, NULL
, NULL
, NULL
);
2595 log_debug("Failed to load X509 certificate '%s', ignoring.", *i
);
2599 if (sk_X509_push(sk
, c
) == 0)
2600 return log_oom_debug();
2605 r
= PKCS7_verify(p7
, sk
, NULL
, bio
, NULL
, PKCS7_NOINTERN
|PKCS7_NOVERIFY
);
2607 log_debug("Userspace PKCS#7 validation succeeded.");
2609 log_debug("Userspace PKCS#7 validation failed: %s", ERR_error_string(ERR_get_error(), NULL
));
2613 log_debug("Not doing client-side validation of dm-verity root hash signatures, OpenSSL support disabled.");
2618 static int do_crypt_activate_verity(
2619 struct crypt_device
*cd
,
2621 const VeritySettings
*verity
) {
2623 bool check_signature
;
2630 if (verity
->root_hash_sig
) {
2631 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIGNATURE");
2632 if (r
< 0 && r
!= -ENXIO
)
2633 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIGNATURE");
2635 check_signature
= r
!= 0;
2637 check_signature
= false;
2639 if (check_signature
) {
2641 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
2642 /* First, if we have support for signed keys in the kernel, then try that first. */
2643 r
= sym_crypt_activate_by_signed_key(
2647 verity
->root_hash_size
,
2648 verity
->root_hash_sig
,
2649 verity
->root_hash_sig_size
,
2650 CRYPT_ACTIVATE_READONLY
);
2654 log_debug_errno(r
, "Validation of dm-verity signature failed via the kernel, trying userspace validation instead: %m");
2656 log_debug("Activation of verity device with signature requested, but not supported via the kernel by %s due to missing crypt_activate_by_signed_key(), trying userspace validation instead.",
2657 program_invocation_short_name
);
2658 r
= 0; /* Set for the propagation below */
2661 /* So this didn't work via the kernel, then let's try userspace validation instead. If that
2662 * works we'll try to activate without telling the kernel the signature. */
2664 /* Preferably propagate the original kernel error, so that the fallback logic can work,
2665 * as the device-mapper is finicky around concurrent activations of the same volume */
2666 k
= validate_signature_userspace(verity
);
2668 return r
< 0 ? r
: k
;
2670 return log_debug_errno(r
< 0 ? r
: SYNTHETIC_ERRNO(ENOKEY
),
2671 "Activation of signed Verity volume worked neither via the kernel nor in userspace, can't activate.");
2674 return sym_crypt_activate_by_volume_key(
2678 verity
->root_hash_size
,
2679 CRYPT_ACTIVATE_READONLY
);
2682 static usec_t
verity_timeout(void) {
2683 usec_t t
= 100 * USEC_PER_MSEC
;
2687 /* On slower machines, like non-KVM vm, setting up device may take a long time.
2688 * Let's make the timeout configurable. */
2690 e
= getenv("SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC");
2694 r
= parse_sec(e
, &t
);
2697 "Failed to parse timeout specified in $SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC, "
2698 "using the default timeout (%s).",
2699 FORMAT_TIMESPAN(t
, USEC_PER_MSEC
));
2704 static int verity_partition(
2705 PartitionDesignator designator
,
2706 DissectedPartition
*m
,
2707 DissectedPartition
*v
,
2708 const VeritySettings
*verity
,
2709 DissectImageFlags flags
,
2710 DecryptedImage
*d
) {
2712 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2713 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
2714 _cleanup_close_
int mount_node_fd
= -EBADF
;
2718 assert(v
|| (verity
&& verity
->data_path
));
2720 if (!verity
|| !verity
->root_hash
)
2722 if (!((verity
->designator
< 0 && designator
== PARTITION_ROOT
) ||
2723 (verity
->designator
== designator
)))
2726 if (!m
->found
|| !m
->node
|| !m
->fstype
)
2728 if (!verity
->data_path
) {
2729 if (!v
->found
|| !v
->node
|| !v
->fstype
)
2732 if (!streq(v
->fstype
, "DM_verity_hash"))
2736 r
= dlopen_cryptsetup();
2740 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
2741 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
2742 _cleanup_free_
char *root_hash_encoded
= NULL
;
2744 root_hash_encoded
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2745 if (!root_hash_encoded
)
2748 r
= make_dm_name_and_node(root_hash_encoded
, "-verity", &name
, &node
);
2750 r
= make_dm_name_and_node(m
->node
, "-verity", &name
, &node
);
2754 r
= sym_crypt_init(&cd
, verity
->data_path
?: v
->node
);
2758 cryptsetup_enable_logging(cd
);
2760 r
= sym_crypt_load(cd
, CRYPT_VERITY
, NULL
);
2764 r
= sym_crypt_set_data_device(cd
, m
->node
);
2768 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
2771 /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
2772 * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
2773 * retry a few times before giving up. */
2774 for (unsigned i
= 0; i
< N_DEVICE_NODE_LIST_ATTEMPTS
; i
++) {
2775 _cleanup_(dm_deferred_remove_cleanp
) char *restore_deferred_remove
= NULL
;
2776 _cleanup_(sym_crypt_freep
) struct crypt_device
*existing_cd
= NULL
;
2777 _cleanup_close_
int fd
= -EBADF
;
2779 /* First, check if the device already exists. */
2780 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2781 if (fd
< 0 && !ERRNO_IS_DEVICE_ABSENT(errno
))
2782 return log_debug_errno(errno
, "Failed to open verity device %s: %m", node
);
2784 goto check
; /* The device already exists. Let's check it. */
2786 /* The symlink to the device node does not exist yet. Assume not activated, and let's activate it. */
2787 r
= do_crypt_activate_verity(cd
, name
, verity
);
2789 goto try_open
; /* The device is activated. Let's open it. */
2790 /* libdevmapper can return EINVAL when the device is already in the activation stage.
2791 * There's no way to distinguish this situation from a genuine error due to invalid
2792 * parameters, so immediately fall back to activating the device with a unique name.
2793 * Improvements in libcrypsetup can ensure this never happens:
2794 * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
2795 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2797 if (r
== -ENODEV
) /* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */
2800 -EEXIST
, /* Volume has already been opened and ready to be used. */
2801 -EBUSY
/* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */))
2802 return log_debug_errno(r
, "Failed to activate verity device %s: %m", node
);
2805 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
2806 r
= dm_deferred_remove_cancel(name
);
2807 /* -EBUSY and -ENXIO: the device has already been removed or being removed. We cannot
2808 * use the device, try to open again. See target_message() in drivers/md/dm-ioctl.c
2809 * and dm_cancel_deferred_remove() in drivers/md/dm.c */
2810 if (IN_SET(r
, -EBUSY
, -ENXIO
))
2813 return log_debug_errno(r
, "Failed to disable automated deferred removal for verity device %s: %m", node
);
2815 restore_deferred_remove
= strdup(name
);
2816 if (!restore_deferred_remove
)
2817 return log_oom_debug();
2819 r
= verity_can_reuse(verity
, name
, &existing_cd
);
2820 /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
2821 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2824 -ENOENT
, /* Removed?? */
2825 -EBUSY
, /* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */
2826 -ENODEV
/* Volume is being opened but not ready, crypt_init_by_name() would fail, try to open again. */ ))
2829 return log_debug_errno(r
, "Failed to check if existing verity device %s can be reused: %m", node
);
2832 /* devmapper might say that the device exists, but the devlink might not yet have been
2833 * created. Check and wait for the udev event in that case. */
2834 r
= device_wait_for_devlink(node
, "block", verity_timeout(), NULL
);
2835 /* Fallback to activation with a unique device if it's taking too long */
2836 if (r
== -ETIMEDOUT
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2839 return log_debug_errno(r
, "Failed to wait device node symlink %s: %m", node
);
2844 /* Now, the device is activated and devlink is created. Let's open it. */
2845 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2847 if (!ERRNO_IS_DEVICE_ABSENT(errno
))
2848 return log_debug_errno(errno
, "Failed to open verity device %s: %m", node
);
2850 /* The device has already been removed?? */
2855 /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
2856 restore_deferred_remove
= mfree(restore_deferred_remove
);
2858 mount_node_fd
= TAKE_FD(fd
);
2860 crypt_free_and_replace(cd
, existing_cd
);
2865 /* Device is being removed by another process. Let's wait for a while. */
2866 (void) usleep_safe(2 * USEC_PER_MSEC
);
2869 /* All trials failed or a conflicting verity device exists. Let's try to activate with a unique name. */
2870 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
2871 /* Before trying to activate with unique name, we need to free crypt_device object.
2872 * Otherwise, we get error from libcryptsetup like the following:
2874 * systemd[1234]: Cannot use device /dev/loop5 which is in use (already mapped or mounted).
2879 return verity_partition(designator
, m
, v
, verity
, flags
& ~DISSECT_IMAGE_VERITY_SHARE
, d
);
2882 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
), "All attempts to activate verity device %s failed.", name
);
2885 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
2886 .name
= TAKE_PTR(name
),
2887 .device
= TAKE_PTR(cd
),
2890 m
->decrypted_node
= TAKE_PTR(node
);
2891 close_and_replace(m
->mount_node_fd
, mount_node_fd
);
2897 int dissected_image_decrypt(
2899 const char *passphrase
,
2900 const VeritySettings
*verity
,
2901 DissectImageFlags flags
) {
2903 #if HAVE_LIBCRYPTSETUP
2904 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
2909 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
2913 * = 0 → There was nothing to decrypt
2914 * > 0 → Decrypted successfully
2915 * -ENOKEY → There's something to decrypt but no key was supplied
2916 * -EKEYREJECTED → Passed key was not correct
2919 if (verity
&& verity
->root_hash
&& verity
->root_hash_size
< sizeof(sd_id128_t
))
2922 if (!m
->encrypted
&& !m
->verity_ready
)
2925 #if HAVE_LIBCRYPTSETUP
2926 r
= decrypted_image_new(&d
);
2930 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
2931 DissectedPartition
*p
= m
->partitions
+ i
;
2932 PartitionDesignator k
;
2937 r
= decrypt_partition(p
, passphrase
, flags
, d
);
2941 k
= partition_verity_of(i
);
2943 r
= verity_partition(i
, p
, m
->partitions
+ k
, verity
, flags
| DISSECT_IMAGE_VERITY_SHARE
, d
);
2948 if (!p
->decrypted_fstype
&& p
->mount_node_fd
>= 0 && p
->decrypted_node
) {
2949 r
= probe_filesystem_full(p
->mount_node_fd
, p
->decrypted_node
, 0, UINT64_MAX
, &p
->decrypted_fstype
);
2950 if (r
< 0 && r
!= -EUCLEAN
)
2955 m
->decrypted_image
= TAKE_PTR(d
);
2963 int dissected_image_decrypt_interactively(
2965 const char *passphrase
,
2966 const VeritySettings
*verity
,
2967 DissectImageFlags flags
) {
2969 _cleanup_strv_free_erase_
char **z
= NULL
;
2976 r
= dissected_image_decrypt(m
, passphrase
, verity
, flags
);
2979 if (r
== -EKEYREJECTED
)
2980 log_error_errno(r
, "Incorrect passphrase, try again!");
2981 else if (r
!= -ENOKEY
)
2982 return log_error_errno(r
, "Failed to decrypt image: %m");
2985 return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED
),
2986 "Too many retries.");
2990 r
= ask_password_auto("Please enter image passphrase:", NULL
, "dissect", "dissect", "dissect.passphrase", USEC_INFINITY
, 0, &z
);
2992 return log_error_errno(r
, "Failed to query for passphrase: %m");
2998 static int decrypted_image_relinquish(DecryptedImage
*d
) {
3001 /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a
3002 * boolean so that we don't clean it up ourselves either anymore */
3004 #if HAVE_LIBCRYPTSETUP
3007 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
3008 DecryptedPartition
*p
= d
->decrypted
+ i
;
3010 if (p
->relinquished
)
3013 r
= sym_crypt_deactivate_by_name(NULL
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
3015 return log_debug_errno(r
, "Failed to mark %s for auto-removal: %m", p
->name
);
3017 p
->relinquished
= true;
3024 int dissected_image_relinquish(DissectedImage
*m
) {
3029 if (m
->decrypted_image
) {
3030 r
= decrypted_image_relinquish(m
->decrypted_image
);
3036 loop_device_relinquish(m
->loop
);
3041 static char *build_auxiliary_path(const char *image
, const char *suffix
) {
3048 e
= endswith(image
, ".raw");
3050 return strjoin(e
, suffix
);
3052 n
= new(char, e
- image
+ strlen(suffix
) + 1);
3056 strcpy(mempcpy(n
, image
, e
- image
), suffix
);
3060 void verity_settings_done(VeritySettings
*v
) {
3063 v
->root_hash
= mfree(v
->root_hash
);
3064 v
->root_hash_size
= 0;
3066 v
->root_hash_sig
= mfree(v
->root_hash_sig
);
3067 v
->root_hash_sig_size
= 0;
3069 v
->data_path
= mfree(v
->data_path
);
3072 int verity_settings_load(
3073 VeritySettings
*verity
,
3075 const char *root_hash_path
,
3076 const char *root_hash_sig_path
) {
3078 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
3079 size_t root_hash_size
= 0, root_hash_sig_size
= 0;
3080 _cleanup_free_
char *verity_data_path
= NULL
;
3081 PartitionDesignator designator
;
3086 assert(verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
3088 /* If we are asked to load the root hash for a device node, exit early */
3089 if (is_device_path(image
))
3092 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIDECAR");
3093 if (r
< 0 && r
!= -ENXIO
)
3094 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIDECAR, ignoring: %m");
3098 designator
= verity
->designator
;
3100 /* We only fill in what isn't already filled in */
3102 if (!verity
->root_hash
) {
3103 _cleanup_free_
char *text
= NULL
;
3105 if (root_hash_path
) {
3106 /* If explicitly specified it takes precedence */
3107 r
= read_one_line_file(root_hash_path
, &text
);
3112 designator
= PARTITION_ROOT
;
3114 /* Otherwise look for xattr and separate file, and first for the data for root and if
3115 * that doesn't exist for /usr */
3117 if (designator
< 0 || designator
== PARTITION_ROOT
) {
3118 r
= getxattr_malloc(image
, "user.verity.roothash", &text
);
3120 _cleanup_free_
char *p
= NULL
;
3122 if (r
!= -ENOENT
&& !ERRNO_IS_XATTR_ABSENT(r
))
3125 p
= build_auxiliary_path(image
, ".roothash");
3129 r
= read_one_line_file(p
, &text
);
3130 if (r
< 0 && r
!= -ENOENT
)
3135 designator
= PARTITION_ROOT
;
3138 if (!text
&& (designator
< 0 || designator
== PARTITION_USR
)) {
3139 /* So in the "roothash" xattr/file name above the "root" of course primarily
3140 * refers to the root of the Verity Merkle tree. But coincidentally it also
3141 * is the hash for the *root* file system, i.e. the "root" neatly refers to
3142 * two distinct concepts called "root". Taking benefit of this happy
3143 * coincidence we call the file with the root hash for the /usr/ file system
3144 * `usrhash`, because `usrroothash` or `rootusrhash` would just be too
3145 * confusing. We thus drop the reference to the root of the Merkle tree, and
3146 * just indicate which file system it's about. */
3147 r
= getxattr_malloc(image
, "user.verity.usrhash", &text
);
3149 _cleanup_free_
char *p
= NULL
;
3151 if (r
!= -ENOENT
&& !ERRNO_IS_XATTR_ABSENT(r
))
3154 p
= build_auxiliary_path(image
, ".usrhash");
3158 r
= read_one_line_file(p
, &text
);
3159 if (r
< 0 && r
!= -ENOENT
)
3164 designator
= PARTITION_USR
;
3169 r
= unhexmem(text
, strlen(text
), &root_hash
, &root_hash_size
);
3172 if (root_hash_size
< sizeof(sd_id128_t
))
3177 if ((root_hash
|| verity
->root_hash
) && !verity
->root_hash_sig
) {
3178 if (root_hash_sig_path
) {
3179 r
= read_full_file(root_hash_sig_path
, (char**) &root_hash_sig
, &root_hash_sig_size
);
3180 if (r
< 0 && r
!= -ENOENT
)
3184 designator
= PARTITION_ROOT
;
3186 if (designator
< 0 || designator
== PARTITION_ROOT
) {
3187 _cleanup_free_
char *p
= NULL
;
3189 /* Follow naming convention recommended by the relevant RFC:
3190 * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
3191 p
= build_auxiliary_path(image
, ".roothash.p7s");
3195 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
3196 if (r
< 0 && r
!= -ENOENT
)
3199 designator
= PARTITION_ROOT
;
3202 if (!root_hash_sig
&& (designator
< 0 || designator
== PARTITION_USR
)) {
3203 _cleanup_free_
char *p
= NULL
;
3205 p
= build_auxiliary_path(image
, ".usrhash.p7s");
3209 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
3210 if (r
< 0 && r
!= -ENOENT
)
3213 designator
= PARTITION_USR
;
3217 if (root_hash_sig
&& root_hash_sig_size
== 0) /* refuse empty size signatures */
3221 if (!verity
->data_path
) {
3222 _cleanup_free_
char *p
= NULL
;
3224 p
= build_auxiliary_path(image
, ".verity");
3228 if (access(p
, F_OK
) < 0) {
3229 if (errno
!= ENOENT
)
3232 verity_data_path
= TAKE_PTR(p
);
3236 verity
->root_hash
= TAKE_PTR(root_hash
);
3237 verity
->root_hash_size
= root_hash_size
;
3240 if (root_hash_sig
) {
3241 verity
->root_hash_sig
= TAKE_PTR(root_hash_sig
);
3242 verity
->root_hash_sig_size
= root_hash_sig_size
;
3245 if (verity_data_path
)
3246 verity
->data_path
= TAKE_PTR(verity_data_path
);
3248 if (verity
->designator
< 0)
3249 verity
->designator
= designator
;
3254 int dissected_image_load_verity_sig_partition(
3257 VeritySettings
*verity
) {
3259 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
3260 _cleanup_(json_variant_unrefp
) JsonVariant
*v
= NULL
;
3261 size_t root_hash_size
, root_hash_sig_size
;
3262 _cleanup_free_
char *buf
= NULL
;
3263 PartitionDesignator d
;
3264 DissectedPartition
*p
;
3265 JsonVariant
*rh
, *sig
;
3274 if (verity
->root_hash
&& verity
->root_hash_sig
) /* Already loaded? */
3277 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_EMBEDDED");
3278 if (r
< 0 && r
!= -ENXIO
)
3279 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_EMBEDDED, ignoring: %m");
3283 d
= partition_verity_sig_of(verity
->designator
< 0 ? PARTITION_ROOT
: verity
->designator
);
3286 p
= m
->partitions
+ d
;
3289 if (p
->offset
== UINT64_MAX
|| p
->size
== UINT64_MAX
)
3292 if (p
->size
> 4*1024*1024) /* Signature data cannot possible be larger than 4M, refuse that */
3293 return log_debug_errno(SYNTHETIC_ERRNO(EFBIG
), "Verity signature partition is larger than 4M, refusing.");
3295 buf
= new(char, p
->size
+1);
3299 n
= pread(fd
, buf
, p
->size
, p
->offset
);
3302 if ((uint64_t) n
!= p
->size
)
3305 e
= memchr(buf
, 0, p
->size
);
3307 /* If we found a NUL byte then the rest of the data must be NUL too */
3308 if (!memeqzero(e
, p
->size
- (e
- buf
)))
3309 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature data contains embedded NUL byte.");
3313 r
= json_parse(buf
, 0, &v
, NULL
, NULL
);
3315 return log_debug_errno(r
, "Failed to parse signature JSON data: %m");
3317 rh
= json_variant_by_key(v
, "rootHash");
3319 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'rootHash' field.");
3320 if (!json_variant_is_string(rh
))
3321 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'rootHash' field of signature JSON object is not a string.");
3323 r
= unhexmem(json_variant_string(rh
), SIZE_MAX
, &root_hash
, &root_hash_size
);
3325 return log_debug_errno(r
, "Failed to parse root hash field: %m");
3327 /* Check if specified root hash matches if it is specified */
3328 if (verity
->root_hash
&&
3329 memcmp_nn(verity
->root_hash
, verity
->root_hash_size
, root_hash
, root_hash_size
) != 0) {
3330 _cleanup_free_
char *a
= NULL
, *b
= NULL
;
3332 a
= hexmem(root_hash
, root_hash_size
);
3333 b
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
3335 return log_debug_errno(r
, "Root hash in signature JSON data (%s) doesn't match configured hash (%s).", strna(a
), strna(b
));
3338 sig
= json_variant_by_key(v
, "signature");
3340 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'signature' field.");
3341 if (!json_variant_is_string(sig
))
3342 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'signature' field of signature JSON object is not a string.");
3344 r
= unbase64mem(json_variant_string(sig
), SIZE_MAX
, &root_hash_sig
, &root_hash_sig_size
);
3346 return log_debug_errno(r
, "Failed to parse signature field: %m");
3348 free_and_replace(verity
->root_hash
, root_hash
);
3349 verity
->root_hash_size
= root_hash_size
;
3351 free_and_replace(verity
->root_hash_sig
, root_hash_sig
);
3352 verity
->root_hash_sig_size
= root_hash_sig_size
;
3357 int dissected_image_acquire_metadata(DissectedImage
*m
, DissectImageFlags extra_flags
) {
3364 META_INITRD_RELEASE
,
3365 META_SYSEXT_RELEASE
,
3366 META_CONFEXT_RELEASE
,
3367 META_HAS_INIT_SYSTEM
,
3371 static const char *const paths
[_META_MAX
] = {
3372 [META_HOSTNAME
] = "/etc/hostname\0",
3373 [META_MACHINE_ID
] = "/etc/machine-id\0",
3374 [META_MACHINE_INFO
] = "/etc/machine-info\0",
3375 [META_OS_RELEASE
] = "/etc/os-release\0"
3376 "/usr/lib/os-release\0",
3377 [META_INITRD_RELEASE
] = "/etc/initrd-release\0"
3378 "/usr/lib/initrd-release\0",
3379 [META_SYSEXT_RELEASE
] = "sysext-release\0", /* String used only for logging. */
3380 [META_CONFEXT_RELEASE
] = "confext-release\0", /* ditto */
3381 [META_HAS_INIT_SYSTEM
] = "has-init-system\0", /* ditto */
3384 _cleanup_strv_free_
char **machine_info
= NULL
, **os_release
= NULL
, **initrd_release
= NULL
, **sysext_release
= NULL
, **confext_release
= NULL
;
3385 _cleanup_close_pair_
int error_pipe
[2] = EBADF_PAIR
;
3386 _cleanup_(rmdir_and_freep
) char *t
= NULL
;
3387 _cleanup_(sigkill_waitp
) pid_t child
= 0;
3388 sd_id128_t machine_id
= SD_ID128_NULL
;
3389 _cleanup_free_
char *hostname
= NULL
;
3390 unsigned n_meta_initialized
= 0;
3391 int fds
[2 * _META_MAX
], r
, v
;
3392 int has_init_system
= -1;
3395 BLOCK_SIGNALS(SIGCHLD
);
3399 for (; n_meta_initialized
< _META_MAX
; n_meta_initialized
++) {
3400 assert(paths
[n_meta_initialized
]);
3402 if (pipe2(fds
+ 2*n_meta_initialized
, O_CLOEXEC
) < 0) {
3408 r
= mkdtemp_malloc("/tmp/dissect-XXXXXX", &t
);
3412 if (pipe2(error_pipe
, O_CLOEXEC
) < 0) {
3417 r
= safe_fork("(sd-dissect)", FORK_RESET_SIGNALS
|FORK_DEATHSIG_SIGTERM
|FORK_NEW_MOUNTNS
|FORK_MOUNTNS_SLAVE
, &child
);
3421 /* Child in a new mount namespace */
3422 error_pipe
[0] = safe_close(error_pipe
[0]);
3424 r
= dissected_image_mount(
3427 /* uid_shift= */ UID_INVALID
,
3428 /* uid_range= */ UID_INVALID
,
3429 /* userns_fd= */ -EBADF
,
3431 DISSECT_IMAGE_READ_ONLY
|
3432 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
3433 DISSECT_IMAGE_USR_NO_ROOT
);
3435 log_debug_errno(r
, "Failed to mount dissected image: %m");
3439 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
3440 _cleanup_close_
int fd
= -ENOENT
;
3444 fds
[2*k
] = safe_close(fds
[2*k
]);
3448 case META_SYSEXT_RELEASE
:
3452 /* As per the os-release spec, if the image is an extension it will have a
3453 * file named after the image name in extension-release.d/ - we use the image
3454 * name and try to resolve it with the extension-release helpers, as
3455 * sometimes the image names are mangled on deployment and do not match
3456 * anymore. Unlike other paths this is not fixed, and the image name can be
3457 * mangled on deployment, so by calling into the helper we allow a fallback
3458 * that matches on the first extension-release file found in the directory,
3459 * if one named after the image cannot be found first. */
3460 r
= open_extension_release(
3464 /* relax_extension_release_check= */ false,
3465 /* ret_path= */ NULL
,
3471 case META_CONFEXT_RELEASE
:
3476 r
= open_extension_release(
3480 /* relax_extension_release_check= */ false,
3481 /* ret_path= */ NULL
,
3488 case META_HAS_INIT_SYSTEM
: {
3491 FOREACH_STRING(init
,
3492 "/usr/lib/systemd/systemd", /* systemd on /usr/ merged system */
3493 "/lib/systemd/systemd", /* systemd on /usr/ non-merged systems */
3494 "/sbin/init") { /* traditional path the Linux kernel invokes */
3496 r
= chase(init
, t
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
3499 log_debug_errno(r
, "Failed to resolve %s, ignoring: %m", init
);
3506 r
= loop_write(fds
[2*k
+1], &found
, sizeof(found
));
3514 NULSTR_FOREACH(p
, paths
[k
]) {
3515 fd
= chase_and_open(p
, t
, CHASE_PREFIX_ROOT
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
, NULL
);
3522 log_debug_errno(fd
, "Failed to read %s file of image, ignoring: %m", paths
[k
]);
3526 r
= copy_bytes(fd
, fds
[2*k
+1], UINT64_MAX
, 0);
3531 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3534 _exit(EXIT_SUCCESS
);
3537 /* Let parent know the error */
3538 (void) write(error_pipe
[1], &r
, sizeof(r
));
3539 _exit(EXIT_FAILURE
);
3542 error_pipe
[1] = safe_close(error_pipe
[1]);
3544 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
3545 _cleanup_fclose_
FILE *f
= NULL
;
3549 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3551 f
= take_fdopen(&fds
[2*k
], "r");
3560 r
= read_etc_hostname_stream(f
, &hostname
);
3562 log_debug_errno(r
, "Failed to read /etc/hostname of image: %m");
3566 case META_MACHINE_ID
: {
3567 _cleanup_free_
char *line
= NULL
;
3569 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3571 log_debug_errno(r
, "Failed to read /etc/machine-id of image: %m");
3573 r
= sd_id128_from_string(line
, &machine_id
);
3575 log_debug_errno(r
, "Image contains invalid /etc/machine-id: %s", line
);
3577 log_debug("/etc/machine-id file of image is empty.");
3578 else if (streq(line
, "uninitialized"))
3579 log_debug("/etc/machine-id file of image is uninitialized (likely aborted first boot).");
3581 log_debug("/etc/machine-id file of image has unexpected length %i.", r
);
3586 case META_MACHINE_INFO
:
3587 r
= load_env_file_pairs(f
, "machine-info", &machine_info
);
3589 log_debug_errno(r
, "Failed to read /etc/machine-info of image: %m");
3593 case META_OS_RELEASE
:
3594 r
= load_env_file_pairs(f
, "os-release", &os_release
);
3596 log_debug_errno(r
, "Failed to read OS release file of image: %m");
3600 case META_INITRD_RELEASE
:
3601 r
= load_env_file_pairs(f
, "initrd-release", &initrd_release
);
3603 log_debug_errno(r
, "Failed to read initrd release file of image: %m");
3607 case META_SYSEXT_RELEASE
:
3608 r
= load_env_file_pairs(f
, "sysext-release", &sysext_release
);
3610 log_debug_errno(r
, "Failed to read sysext release file of image: %m");
3614 case META_CONFEXT_RELEASE
:
3615 r
= load_env_file_pairs(f
, "confext-release", &confext_release
);
3617 log_debug_errno(r
, "Failed to read confext release file of image: %m");
3621 case META_HAS_INIT_SYSTEM
: {
3626 nr
= fread(&b
, 1, sizeof(b
), f
);
3627 if (nr
!= sizeof(b
))
3628 log_debug_errno(errno_or_else(EIO
), "Failed to read has-init-system boolean: %m");
3630 has_init_system
= b
;
3636 r
= wait_for_terminate_and_check("(sd-dissect)", child
, 0);
3641 n
= read(error_pipe
[0], &v
, sizeof(v
));
3646 if (n
== sizeof(v
)) {
3647 r
= v
; /* propagate error sent to us from child */
3654 if (r
!= EXIT_SUCCESS
) {
3659 free_and_replace(m
->hostname
, hostname
);
3660 m
->machine_id
= machine_id
;
3661 strv_free_and_replace(m
->machine_info
, machine_info
);
3662 strv_free_and_replace(m
->os_release
, os_release
);
3663 strv_free_and_replace(m
->initrd_release
, initrd_release
);
3664 strv_free_and_replace(m
->sysext_release
, sysext_release
);
3665 strv_free_and_replace(m
->confext_release
, confext_release
);
3666 m
->has_init_system
= has_init_system
;
3669 for (unsigned k
= 0; k
< n_meta_initialized
; k
++)
3670 safe_close_pair(fds
+ 2*k
);
3675 Architecture
dissected_image_architecture(DissectedImage
*img
) {
3678 if (img
->partitions
[PARTITION_ROOT
].found
&&
3679 img
->partitions
[PARTITION_ROOT
].architecture
>= 0)
3680 return img
->partitions
[PARTITION_ROOT
].architecture
;
3682 if (img
->partitions
[PARTITION_USR
].found
&&
3683 img
->partitions
[PARTITION_USR
].architecture
>= 0)
3684 return img
->partitions
[PARTITION_USR
].architecture
;
3686 return _ARCHITECTURE_INVALID
;
3689 int dissect_loop_device(
3691 const VeritySettings
*verity
,
3692 const MountOptions
*mount_options
,
3693 const ImagePolicy
*image_policy
,
3694 DissectImageFlags flags
,
3695 DissectedImage
**ret
) {
3698 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
3703 r
= dissected_image_new(loop
->backing_file
?: loop
->node
, &m
);
3707 m
->loop
= loop_device_ref(loop
);
3708 m
->image_size
= m
->loop
->device_size
;
3709 m
->sector_size
= m
->loop
->sector_size
;
3711 r
= dissect_image(m
, loop
->fd
, loop
->node
, verity
, mount_options
, image_policy
, flags
);
3724 int dissect_loop_device_and_warn(
3726 const VeritySettings
*verity
,
3727 const MountOptions
*mount_options
,
3728 const ImagePolicy
*image_policy
,
3729 DissectImageFlags flags
,
3730 DissectedImage
**ret
) {
3734 return dissect_log_error(
3736 dissect_loop_device(loop
, verity
, mount_options
, image_policy
, flags
, ret
),
3737 loop
->backing_file
?: loop
->node
,
3742 bool dissected_image_verity_candidate(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3745 /* Checks if this partition could theoretically do Verity. For non-partitioned images this only works
3746 * if there's an external verity file supplied, for which we can consult .has_verity. For partitioned
3747 * images we only check the partition type.
3749 * This call is used to decide whether to suppress or show a verity column in tabular output of the
3752 if (image
->single_file_system
)
3753 return partition_designator
== PARTITION_ROOT
&& image
->has_verity
;
3755 return partition_verity_of(partition_designator
) >= 0;
3758 bool dissected_image_verity_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3759 PartitionDesignator k
;
3763 /* Checks if this partition has verity data available that we can activate. For non-partitioned this
3764 * works for the root partition, for others only if the associated verity partition was found. */
3766 if (!image
->verity_ready
)
3769 if (image
->single_file_system
)
3770 return partition_designator
== PARTITION_ROOT
;
3772 k
= partition_verity_of(partition_designator
);
3773 return k
>= 0 && image
->partitions
[k
].found
;
3776 bool dissected_image_verity_sig_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3777 PartitionDesignator k
;
3781 /* Checks if this partition has verity signature data available that we can use. */
3783 if (!image
->verity_sig_ready
)
3786 if (image
->single_file_system
)
3787 return partition_designator
== PARTITION_ROOT
;
3789 k
= partition_verity_sig_of(partition_designator
);
3790 return k
>= 0 && image
->partitions
[k
].found
;
3793 MountOptions
* mount_options_free_all(MountOptions
*options
) {
3796 while ((m
= LIST_POP(mount_options
, options
))) {
3804 const char* mount_options_from_designator(const MountOptions
*options
, PartitionDesignator designator
) {
3805 LIST_FOREACH(mount_options
, m
, options
)
3806 if (designator
== m
->partition_designator
&& !isempty(m
->options
))
3812 int mount_image_privately_interactively(
3814 const ImagePolicy
*image_policy
,
3815 DissectImageFlags flags
,
3816 char **ret_directory
,
3818 LoopDevice
**ret_loop_device
) {
3820 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3821 _cleanup_(loop_device_unrefp
) LoopDevice
*d
= NULL
;
3822 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3823 _cleanup_free_
char *dir
= NULL
;
3826 /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
3827 * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
3831 assert(ret_loop_device
);
3833 /* We intend to mount this right-away, hence add the partitions if needed and pin them. */
3834 flags
|= DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
3835 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
3837 r
= verity_settings_load(&verity
, image
, NULL
, NULL
);
3839 return log_error_errno(r
, "Failed to load root hash data: %m");
3841 r
= loop_device_make_by_path(
3843 FLAGS_SET(flags
, DISSECT_IMAGE_DEVICE_READ_ONLY
) ? O_RDONLY
: O_RDWR
,
3844 /* sector_size= */ UINT32_MAX
,
3845 FLAGS_SET(flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
3849 return log_error_errno(r
, "Failed to set up loopback device for %s: %m", image
);
3851 r
= dissect_loop_device_and_warn(
3854 /* mount_options= */ NULL
,
3861 r
= dissected_image_load_verity_sig_partition(dissected_image
, d
->fd
, &verity
);
3865 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, &verity
, flags
);
3869 r
= detach_mount_namespace();
3871 return log_error_errno(r
, "Failed to detach mount namespace: %m");
3873 r
= mkdir_p("/run/systemd/mount-rootfs", 0555);
3875 return log_error_errno(r
, "Failed to create mount point: %m");
3877 r
= dissected_image_mount_and_warn(
3879 "/run/systemd/mount-rootfs",
3880 /* uid_shift= */ UID_INVALID
,
3881 /* uid_range= */ UID_INVALID
,
3882 /* userns_fd= */ -EBADF
,
3887 r
= loop_device_flock(d
, LOCK_UN
);
3891 r
= dissected_image_relinquish(dissected_image
);
3893 return log_error_errno(r
, "Failed to relinquish DM and loopback block devices: %m");
3895 if (ret_directory
) {
3896 dir
= strdup("/run/systemd/mount-rootfs");
3902 _cleanup_close_
int dir_fd
= -EBADF
;
3904 dir_fd
= open("/run/systemd/mount-rootfs", O_CLOEXEC
|O_DIRECTORY
);
3906 return log_error_errno(errno
, "Failed to open mount point directory: %m");
3908 *ret_dir_fd
= TAKE_FD(dir_fd
);
3912 *ret_directory
= TAKE_PTR(dir
);
3914 *ret_loop_device
= TAKE_PTR(d
);
3918 static bool mount_options_relax_extension_release_checks(const MountOptions
*options
) {
3922 return string_contains_word(mount_options_from_designator(options
, PARTITION_ROOT
), ",", "x-systemd.relax-extension-release-check") ||
3923 string_contains_word(mount_options_from_designator(options
, PARTITION_USR
), ",", "x-systemd.relax-extension-release-check") ||
3924 string_contains_word(options
->options
, ",", "x-systemd.relax-extension-release-check");
3927 int verity_dissect_and_mount(
3931 const MountOptions
*options
,
3932 const ImagePolicy
*image_policy
,
3933 const char *required_host_os_release_id
,
3934 const char *required_host_os_release_version_id
,
3935 const char *required_host_os_release_sysext_level
,
3936 const char *required_host_os_release_confext_level
,
3937 const char *required_sysext_scope
,
3938 DissectedImage
**ret_image
) {
3940 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
3941 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3942 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3943 DissectImageFlags dissect_image_flags
;
3944 bool relax_extension_release_check
;
3948 /* Verifying release metadata requires mounted image for now, so ensure the check is skipped when
3949 * opening an image without mounting it immediately (i.e.: 'dest' is NULL). */
3950 assert(!required_host_os_release_id
|| dest
);
3952 relax_extension_release_check
= mount_options_relax_extension_release_checks(options
);
3954 /* We might get an FD for the image, but we use the original path to look for the dm-verity files */
3955 r
= verity_settings_load(&verity
, src
, NULL
, NULL
);
3957 return log_debug_errno(r
, "Failed to load root hash: %m");
3959 dissect_image_flags
= (verity
.data_path
? DISSECT_IMAGE_NO_PARTITION_TABLE
: 0) |
3960 (relax_extension_release_check
? DISSECT_IMAGE_RELAX_EXTENSION_CHECK
: 0) |
3961 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
3962 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
3964 /* Note that we don't use loop_device_make here, as the FD is most likely O_PATH which would not be
3965 * accepted by LOOP_CONFIGURE, so just let loop_device_make_by_path reopen it as a regular FD. */
3966 r
= loop_device_make_by_path(
3967 src_fd
>= 0 ? FORMAT_PROC_FD_PATH(src_fd
) : src
,
3968 /* open_flags= */ -1,
3969 /* sector_size= */ UINT32_MAX
,
3970 verity
.data_path
? 0 : LO_FLAGS_PARTSCAN
,
3974 return log_debug_errno(r
, "Failed to create loop device for image: %m");
3976 r
= dissect_loop_device(
3981 dissect_image_flags
,
3983 /* No partition table? Might be a single-filesystem image, try again */
3984 if (!verity
.data_path
&& r
== -ENOPKG
)
3985 r
= dissect_loop_device(
3990 dissect_image_flags
| DISSECT_IMAGE_NO_PARTITION_TABLE
,
3993 return log_debug_errno(r
, "Failed to dissect image: %m");
3995 r
= dissected_image_load_verity_sig_partition(dissected_image
, loop_device
->fd
, &verity
);
3999 r
= dissected_image_decrypt(
4003 dissect_image_flags
);
4005 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
4008 r
= mkdir_p_label(dest
, 0755);
4010 return log_debug_errno(r
, "Failed to create destination directory %s: %m", dest
);
4011 r
= umount_recursive(dest
, 0);
4013 return log_debug_errno(r
, "Failed to umount under destination directory %s: %m", dest
);
4016 r
= dissected_image_mount(
4019 /* uid_shift= */ UID_INVALID
,
4020 /* uid_range= */ UID_INVALID
,
4021 /* userns_fd= */ -EBADF
,
4022 dissect_image_flags
);
4024 return log_debug_errno(r
, "Failed to mount image: %m");
4026 r
= loop_device_flock(loop_device
, LOCK_UN
);
4028 return log_debug_errno(r
, "Failed to unlock loopback device: %m");
4030 /* If we got os-release values from the caller, then we need to match them with the image's
4031 * extension-release.d/ content. Return -EINVAL if there's any mismatch.
4032 * First, check the distro ID. If that matches, then check the new SYSEXT_LEVEL value if
4033 * available, or else fallback to VERSION_ID. If neither is present (eg: rolling release),
4034 * then a simple match on the ID will be performed. */
4035 if (required_host_os_release_id
) {
4036 _cleanup_strv_free_
char **extension_release
= NULL
;
4037 ImageClass
class = IMAGE_SYSEXT
;
4039 assert(!isempty(required_host_os_release_id
));
4041 r
= load_extension_release_pairs(dest
, IMAGE_SYSEXT
, dissected_image
->image_name
, relax_extension_release_check
, &extension_release
);
4043 r
= load_extension_release_pairs(dest
, IMAGE_CONFEXT
, dissected_image
->image_name
, relax_extension_release_check
, &extension_release
);
4045 class = IMAGE_CONFEXT
;
4048 return log_debug_errno(r
, "Failed to parse image %s extension-release metadata: %m", dissected_image
->image_name
);
4050 r
= extension_release_validate(
4051 dissected_image
->image_name
,
4052 required_host_os_release_id
,
4053 required_host_os_release_version_id
,
4054 class == IMAGE_SYSEXT
? required_host_os_release_sysext_level
: required_host_os_release_confext_level
,
4055 required_sysext_scope
,
4059 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE
), "Image %s extension-release metadata does not match the root's", dissected_image
->image_name
);
4061 return log_debug_errno(r
, "Failed to compare image %s extension-release metadata with the root's os-release: %m", dissected_image
->image_name
);
4064 r
= dissected_image_relinquish(dissected_image
);
4066 return log_debug_errno(r
, "Failed to relinquish dissected image: %m");
4069 *ret_image
= TAKE_PTR(dissected_image
);