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 "proc-cmdline.h"
64 #include "process-util.h"
65 #include "raw-clone.h"
66 #include "resize-fs.h"
67 #include "signal-util.h"
68 #include "sparse-endian.h"
69 #include "stat-util.h"
70 #include "stdio-util.h"
71 #include "string-table.h"
72 #include "string-util.h"
74 #include "tmpfile-util.h"
75 #include "udev-util.h"
76 #include "user-util.h"
78 #include "xattr-util.h"
80 /* how many times to wait for the device nodes to appear */
81 #define N_DEVICE_NODE_LIST_ATTEMPTS 10
83 int dissect_fstype_ok(const char *fstype
) {
87 /* When we automatically mount file systems, be a bit conservative by default what we are willing to
88 * mount, just as an extra safety net to not mount with badly maintained legacy file system
91 e
= secure_getenv("SYSTEMD_DISSECT_FILE_SYSTEMS");
93 _cleanup_strv_free_
char **l
= NULL
;
95 l
= strv_split(e
, ":");
99 b
= strv_contains(l
, fstype
);
101 b
= STR_IN_SET(fstype
,
112 log_debug("File system type '%s' is not allowed to be mounted as result of automatic dissection.", fstype
);
116 int probe_sector_size(int fd
, uint32_t *ret
) {
118 /* Disk images might be for 512B or for 4096 sector sizes, let's try to auto-detect that by searching
119 * for the GPT headers at the relevant byte offsets */
121 assert_cc(sizeof(GptHeader
) == 92);
123 /* We expect a sector size in the range 512…4096. The GPT header is located in the second
124 * sector. Hence it could be at byte 512 at the earliest, and at byte 4096 at the latest. And we must
125 * read with granularity of the largest sector size we care about. Which means 8K. */
126 uint8_t sectors
[2 * 4096];
133 n
= pread(fd
, sectors
, sizeof(sectors
), 0);
136 if (n
!= sizeof(sectors
)) /* too short? */
139 /* Let's see if we find the GPT partition header with various expected sector sizes */
140 for (uint32_t sz
= 512; sz
<= 4096; sz
<<= 1) {
143 assert(sizeof(sectors
) >= sz
* 2);
144 p
= (const GptHeader
*) (sectors
+ sz
);
146 if (!gpt_header_has_signature(p
))
150 return log_debug_errno(SYNTHETIC_ERRNO(ENOTUNIQ
),
151 "Detected valid partition table at offsets matching multiple sector sizes, refusing.");
157 log_debug("Determined sector size %" PRIu32
" based on discovered partition table.", found
);
159 return 1; /* indicate we *did* find it */
163 log_debug("Couldn't find any partition table to derive sector size of.");
164 *ret
= 512; /* pick the traditional default */
165 return 0; /* indicate we didn't find it */
168 int probe_sector_size_prefer_ioctl(int fd
, uint32_t *ret
) {
174 /* Just like probe_sector_size(), but if we are looking at a block device, will use the already
175 * configured sector size rather than probing by contents */
177 if (fstat(fd
, &st
) < 0)
180 if (S_ISBLK(st
.st_mode
))
181 return blockdev_get_sector_size(fd
, ret
);
183 return probe_sector_size(fd
, ret
);
186 int probe_filesystem_full(
193 /* Try to find device content type and return it in *ret_fstype. If nothing is found,
194 * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and a
195 * different error otherwise. */
198 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
199 _cleanup_free_
char *path_by_fd
= NULL
;
200 _cleanup_close_
int fd_close
= -EBADF
;
204 assert(fd
>= 0 || path
);
208 fd_close
= open(path
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
216 r
= fd_get_path(fd
, &path_by_fd
);
223 if (size
== 0) /* empty size? nothing found! */
226 b
= blkid_new_probe();
230 /* The Linux kernel maintains separate block device caches for main ("whole") and partition block
231 * devices, which means making a change to one might not be reflected immediately when reading via
232 * the other. That's massively confusing when mixing accesses to such devices. Let's address this in
233 * a limited way: when probing a file system that is not at the beginning of the block device we
234 * apparently probe a partition via the main block device, and in that case let's first flush the
235 * main block device cache, so that we get the data that the per-partition block device last
238 * This only works under the assumption that any tools that write to the partition block devices
239 * issue an syncfs()/fsync() on the device after making changes. Typically file system formatting
240 * tools that write a superblock onto a partition block device do that, however. */
242 if (ioctl(fd
, BLKFLSBUF
, 0) < 0)
243 log_debug_errno(errno
, "Failed to flush block device cache, ignoring: %m");
246 r
= blkid_probe_set_device(
250 size
== UINT64_MAX
? 0 : size
); /* when blkid sees size=0 it understands "everything". We prefer using UINT64_MAX for that */
252 return errno_or_else(ENOMEM
);
254 blkid_probe_enable_superblocks(b
, 1);
255 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
);
258 r
= blkid_do_safeprobe(b
);
259 if (r
== _BLKID_SAFEPROBE_NOT_FOUND
)
261 if (r
== _BLKID_SAFEPROBE_AMBIGUOUS
)
262 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
),
263 "Results ambiguous for partition %s", path
);
264 if (r
== _BLKID_SAFEPROBE_ERROR
)
265 return log_debug_errno(errno_or_else(EIO
), "Failed to probe partition %s: %m", path
);
267 assert(r
== _BLKID_SAFEPROBE_FOUND
);
269 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
272 log_debug("Probed fstype '%s' on partition %s.", fstype
, path
);
273 return strdup_to_full(ret_fstype
, fstype
);
277 log_debug("No type detected on partition %s", path
);
286 static int image_policy_may_use(
287 const ImagePolicy
*policy
,
288 PartitionDesignator designator
) {
290 PartitionPolicyFlags f
;
292 /* For each partition we find in the partition table do a first check if it may exist at all given
293 * the policy, or if it shall be ignored. */
295 f
= image_policy_get_exhaustively(policy
, designator
);
299 if ((f
& _PARTITION_POLICY_USE_MASK
) == PARTITION_POLICY_ABSENT
)
300 /* only flag set in policy is "absent"? then this partition may not exist at all */
301 return log_debug_errno(
302 SYNTHETIC_ERRNO(ERFKILL
),
303 "Partition of designator '%s' exists, but not allowed by policy, refusing.",
304 partition_designator_to_string(designator
));
305 if ((f
& _PARTITION_POLICY_USE_MASK
& ~PARTITION_POLICY_ABSENT
) == PARTITION_POLICY_UNUSED
) {
306 /* only "unused" or "unused" + "absent" are set? then don't use it */
307 log_debug("Partition of designator '%s' exists, and policy dictates to ignore it, doing so.",
308 partition_designator_to_string(designator
));
309 return false; /* ignore! */
312 return true; /* use! */
315 static int image_policy_check_protection(
316 const ImagePolicy
*policy
,
317 PartitionDesignator designator
,
318 PartitionPolicyFlags found_flags
) {
320 PartitionPolicyFlags policy_flags
;
322 /* Checks if the flags in the policy for the designated partition overlap the flags of what we found */
327 policy_flags
= image_policy_get_exhaustively(policy
, designator
);
328 if (policy_flags
< 0)
331 if ((found_flags
& policy_flags
) == 0) {
332 _cleanup_free_
char *found_flags_string
= NULL
, *policy_flags_string
= NULL
;
334 (void) partition_policy_flags_to_string(found_flags
, /* simplify= */ true, &found_flags_string
);
335 (void) partition_policy_flags_to_string(policy_flags
, /* simplify= */ true, &policy_flags_string
);
337 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL
), "Partition %s discovered with policy '%s' but '%s' was required, refusing.",
338 partition_designator_to_string(designator
),
339 strnull(found_flags_string
), strnull(policy_flags_string
));
345 static int image_policy_check_partition_flags(
346 const ImagePolicy
*policy
,
347 PartitionDesignator designator
,
348 uint64_t gpt_flags
) {
350 PartitionPolicyFlags policy_flags
;
353 /* Checks if the partition flags in the policy match reality */
355 policy_flags
= image_policy_get_exhaustively(policy
, designator
);
356 if (policy_flags
< 0)
359 b
= FLAGS_SET(gpt_flags
, SD_GPT_FLAG_READ_ONLY
);
360 if ((policy_flags
& _PARTITION_POLICY_READ_ONLY_MASK
) == (b
? PARTITION_POLICY_READ_ONLY_OFF
: PARTITION_POLICY_READ_ONLY_ON
))
361 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL
), "Partition %s has 'read-only' flag incorrectly set (must be %s, is %s), refusing.",
362 partition_designator_to_string(designator
),
363 one_zero(!b
), one_zero(b
));
365 b
= FLAGS_SET(gpt_flags
, SD_GPT_FLAG_GROWFS
);
366 if ((policy_flags
& _PARTITION_POLICY_GROWFS_MASK
) == (b
? PARTITION_POLICY_GROWFS_OFF
: PARTITION_POLICY_GROWFS_ON
))
367 return log_debug_errno(SYNTHETIC_ERRNO(ERFKILL
), "Partition %s has 'growfs' flag incorrectly set (must be %s, is %s), refusing.",
368 partition_designator_to_string(designator
),
369 one_zero(!b
), one_zero(b
));
374 static int dissected_image_probe_filesystems(
377 const ImagePolicy
*policy
) {
383 /* Fill in file system types if we don't know them yet. */
385 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
386 DissectedPartition
*p
= m
->partitions
+ i
;
387 PartitionPolicyFlags found_flags
;
393 /* If we have an fd referring to the partition block device, use that. Otherwise go
394 * via the whole block device or backing regular file, and read via offset. */
395 if (p
->mount_node_fd
>= 0)
396 r
= probe_filesystem_full(p
->mount_node_fd
, p
->node
, 0, UINT64_MAX
, &p
->fstype
);
398 r
= probe_filesystem_full(fd
, p
->node
, p
->offset
, p
->size
, &p
->fstype
);
403 if (streq_ptr(p
->fstype
, "crypto_LUKS")) {
405 found_flags
= PARTITION_POLICY_ENCRYPTED
; /* found this one, and its definitely encrypted */
407 /* found it, but it's definitely not encrypted, hence mask the encrypted flag, but
408 * set all other ways that indicate "present". */
409 found_flags
= PARTITION_POLICY_UNPROTECTED
|PARTITION_POLICY_VERITY
|PARTITION_POLICY_SIGNED
;
411 if (p
->fstype
&& fstype_is_ro(p
->fstype
))
417 /* We might have learnt more about the file system now (i.e. whether it is encrypted or not),
418 * hence we need to validate this against policy again, to see if the policy still matches
419 * with this new information. Note that image_policy_check_protection() will check for
420 * overlap between what's allowed in the policy and what we pass as 'found_policy' here. In
421 * the unencrypted case we thus might pass an overly unspecific mask here (i.e. unprotected
422 * OR verity OR signed), but that's fine since the earlier policy check already checked more
423 * specific which of those three cases where OK. Keep in mind that this function here only
424 * looks at specific partitions (and thus can only deduce encryption or not) but not the
425 * overall partition table (and thus cannot deduce verity or not). The earlier dissection
426 * checks already did the relevant checks that look at the whole partition table, and
427 * enforced policy there as needed. */
428 r
= image_policy_check_protection(policy
, i
, found_flags
);
436 static void check_partition_flags(
438 unsigned long long pflags
,
439 unsigned long long supported
) {
443 /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
444 pflags
&= ~(supported
|
445 SD_GPT_FLAG_REQUIRED_PARTITION
|
446 SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL
|
447 SD_GPT_FLAG_LEGACY_BIOS_BOOTABLE
);
452 /* If there are other bits set, then log about it, to make things discoverable */
453 for (unsigned i
= 0; i
< sizeof(pflags
) * 8; i
++) {
454 unsigned long long bit
= 1ULL << i
;
455 if (!FLAGS_SET(pflags
, bit
))
458 log_debug("Unexpected partition flag %llu set on %s!", bit
, node
);
462 static int dissected_image_new(const char *path
, DissectedImage
**ret
) {
463 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
464 _cleanup_free_
char *name
= NULL
;
470 _cleanup_free_
char *filename
= NULL
;
472 r
= path_extract_filename(path
, &filename
);
476 r
= raw_strip_suffixes(filename
, &name
);
480 if (!image_name_is_valid(name
)) {
481 log_debug("Image name %s is not valid, ignoring.", strna(name
));
486 m
= new(DissectedImage
, 1);
490 *m
= (DissectedImage
) {
491 .has_init_system
= -1,
492 .image_name
= TAKE_PTR(name
),
495 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++)
496 m
->partitions
[i
] = DISSECTED_PARTITION_NULL
;
503 static void dissected_partition_done(DissectedPartition
*p
) {
509 free(p
->decrypted_fstype
);
510 free(p
->decrypted_node
);
511 free(p
->mount_options
);
512 safe_close(p
->mount_node_fd
);
513 safe_close(p
->fsmount_fd
);
515 *p
= DISSECTED_PARTITION_NULL
;
519 static int make_partition_devname(
520 const char *whole_devname
,
523 DissectImageFlags flags
,
526 _cleanup_free_
char *s
= NULL
;
529 assert(whole_devname
);
530 assert(nr
!= 0); /* zero is not a valid partition nr */
533 if (!FLAGS_SET(flags
, DISSECT_IMAGE_DISKSEQ_DEVNODE
) || diskseq
== 0) {
535 /* Given a whole block device node name (e.g. /dev/sda or /dev/loop7) generate a partition
536 * device name (e.g. /dev/sda7 or /dev/loop7p5). The rule the kernel uses is simple: if whole
537 * block device node name ends in a digit, then suffix a 'p', followed by the partition
538 * number. Otherwise, just suffix the partition number without any 'p'. */
540 if (nr
< 0) { /* whole disk? */
541 s
= strdup(whole_devname
);
545 size_t l
= strlen(whole_devname
);
546 if (l
< 1) /* underflow check for the subtraction below */
549 bool need_p
= ascii_isdigit(whole_devname
[l
-1]); /* Last char a digit? */
551 if (asprintf(&s
, "%s%s%i", whole_devname
, need_p
? "p" : "", nr
) < 0)
555 if (nr
< 0) /* whole disk? */
556 r
= asprintf(&s
, "/dev/disk/by-diskseq/%" PRIu64
, diskseq
);
558 r
= asprintf(&s
, "/dev/disk/by-diskseq/%" PRIu64
"-part%i", diskseq
, nr
);
567 static int open_partition(
570 const LoopDevice
*loop
) {
572 _cleanup_(sd_device_unrefp
) sd_device
*dev
= NULL
;
573 _cleanup_close_
int fd
= -EBADF
;
580 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
584 /* Check if the block device is a child of (or equivalent to) the originally provided one. */
585 r
= block_device_new_from_fd(fd
, is_partition
? BLOCK_DEVICE_LOOKUP_WHOLE_DISK
: 0, &dev
);
589 r
= sd_device_get_devnum(dev
, &devnum
);
593 if (loop
->devno
!= devnum
)
596 /* Also check diskseq. */
597 if (loop
->diskseq
!= 0) {
600 r
= fd_get_diskseq(fd
, &diskseq
);
604 if (loop
->diskseq
!= diskseq
)
608 log_debug("Opened %s (fd=%i, whole_block_devnum=" DEVNUM_FORMAT_STR
", diskseq=%" PRIu64
").",
609 node
, fd
, DEVNUM_FORMAT_VAL(loop
->devno
), loop
->diskseq
);
613 static int compare_arch(Architecture a
, Architecture b
) {
617 if (a
== native_architecture())
620 if (b
== native_architecture())
623 #ifdef ARCHITECTURE_SECONDARY
624 if (a
== ARCHITECTURE_SECONDARY
)
627 if (b
== ARCHITECTURE_SECONDARY
)
634 static int dissect_image(
638 const VeritySettings
*verity
,
639 const MountOptions
*mount_options
,
640 const ImagePolicy
*policy
,
641 DissectImageFlags flags
) {
643 sd_id128_t root_uuid
= SD_ID128_NULL
, root_verity_uuid
= SD_ID128_NULL
;
644 sd_id128_t usr_uuid
= SD_ID128_NULL
, usr_verity_uuid
= SD_ID128_NULL
;
645 bool is_gpt
, is_mbr
, multiple_generic
= false,
646 generic_rw
= false, /* initialize to appease gcc */
647 generic_growfs
= false;
648 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
649 _cleanup_free_
char *generic_node
= NULL
;
650 sd_id128_t generic_uuid
= SD_ID128_NULL
;
651 const char *pttype
= NULL
, *sptuuid
= NULL
;
653 int r
, generic_nr
= -1, n_partitions
;
658 assert(!verity
|| verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
659 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
660 assert(!verity
|| verity
->root_hash_sig
|| verity
->root_hash_sig_size
== 0);
661 assert(!verity
|| (verity
->root_hash
|| !verity
->root_hash_sig
));
662 assert(!((flags
& DISSECT_IMAGE_GPT_ONLY
) && (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)));
663 assert(m
->sector_size
> 0);
665 /* Probes a disk image, and returns information about what it found in *ret.
667 * Returns -ENOPKG if no suitable partition table or file system could be found.
668 * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found.
669 * Returns -ENXIO if we couldn't find any partition suitable as root or /usr partition
670 * Returns -ENOTUNIQ if we only found multiple generic partitions and thus don't know what to do with that
671 * Returns -ERFKILL if image doesn't match image policy
672 * 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)
673 * Returns -EPROTONOSUPPORT if DISSECT_IMAGE_ADD_PARTITION_DEVICES is set but the block device does not have partition logic enabled
674 * Returns -ENOMSG if we didn't find a single usable partition (and DISSECT_IMAGE_REFUSE_EMPTY is set) */
676 uint64_t diskseq
= m
->loop
? m
->loop
->diskseq
: 0;
678 if (verity
&& verity
->root_hash
) {
679 sd_id128_t fsuuid
, vuuid
;
681 /* If a root hash is supplied, then we use the root partition that has a UUID that match the
682 * first 128-bit of the root hash. And we use the verity partition that has a UUID that match
683 * the final 128-bit. */
685 if (verity
->root_hash_size
< sizeof(sd_id128_t
))
688 memcpy(&fsuuid
, verity
->root_hash
, sizeof(sd_id128_t
));
689 memcpy(&vuuid
, (const uint8_t*) verity
->root_hash
+ verity
->root_hash_size
- sizeof(sd_id128_t
), sizeof(sd_id128_t
));
691 if (sd_id128_is_null(fsuuid
))
693 if (sd_id128_is_null(vuuid
))
696 /* If the verity data declares it's for the /usr partition, then search for that, in all
697 * other cases assume it's for the root partition. */
698 if (verity
->designator
== PARTITION_USR
) {
700 usr_verity_uuid
= vuuid
;
703 root_verity_uuid
= vuuid
;
707 b
= blkid_new_probe();
712 r
= blkid_probe_set_device(b
, fd
, 0, 0);
714 return errno_or_else(ENOMEM
);
717 r
= blkid_probe_set_sectorsize(b
, m
->sector_size
);
719 return errno_or_else(EIO
);
721 if ((flags
& DISSECT_IMAGE_GPT_ONLY
) == 0) {
722 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
723 blkid_probe_enable_superblocks(b
, 1);
724 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
|BLKID_SUBLKS_USAGE
|BLKID_SUBLKS_UUID
);
727 blkid_probe_enable_partitions(b
, 1);
728 blkid_probe_set_partitions_flags(b
, BLKID_PARTS_ENTRY_DETAILS
);
731 r
= blkid_do_safeprobe(b
);
732 if (r
== _BLKID_SAFEPROBE_ERROR
)
733 return errno_or_else(EIO
);
734 if (IN_SET(r
, _BLKID_SAFEPROBE_AMBIGUOUS
, _BLKID_SAFEPROBE_NOT_FOUND
))
735 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG
), "Failed to identify any partition table.");
737 assert(r
== _BLKID_SAFEPROBE_FOUND
);
739 if ((!(flags
& DISSECT_IMAGE_GPT_ONLY
) &&
740 (flags
& DISSECT_IMAGE_GENERIC_ROOT
)) ||
741 (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)) {
742 const char *usage
= NULL
;
744 /* If flags permit this, also allow using non-partitioned single-filesystem images */
746 (void) blkid_probe_lookup_value(b
, "USAGE", &usage
, NULL
);
747 if (STRPTR_IN_SET(usage
, "filesystem", "crypto")) {
748 _cleanup_free_
char *t
= NULL
, *n
= NULL
, *o
= NULL
;
749 const char *fstype
= NULL
, *options
= NULL
, *suuid
= NULL
;
750 _cleanup_close_
int mount_node_fd
= -EBADF
;
751 sd_id128_t uuid
= SD_ID128_NULL
;
752 PartitionPolicyFlags found_flags
;
755 /* OK, we have found a file system, that's our root partition then. */
757 r
= image_policy_may_use(policy
, PARTITION_ROOT
);
760 if (r
== 0) /* policy says ignore this, so we ignore it */
763 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
764 (void) blkid_probe_lookup_value(b
, "UUID", &suuid
, NULL
);
766 encrypted
= streq_ptr(fstype
, "crypto_LUKS");
768 if (verity_settings_data_covers(verity
, PARTITION_ROOT
))
769 found_flags
= verity
->root_hash_sig
? PARTITION_POLICY_SIGNED
: PARTITION_POLICY_VERITY
;
771 found_flags
= encrypted
? PARTITION_POLICY_ENCRYPTED
: PARTITION_POLICY_UNPROTECTED
;
773 r
= image_policy_check_protection(policy
, PARTITION_ROOT
, found_flags
);
777 r
= image_policy_check_partition_flags(policy
, PARTITION_ROOT
, 0); /* we have no gpt partition flags, hence check against all bits off */
781 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
782 mount_node_fd
= open_partition(devname
, /* is_partition = */ false, m
->loop
);
783 if (mount_node_fd
< 0)
784 return mount_node_fd
;
794 /* blkid will return FAT's serial number as UUID, hence it is quite possible
795 * that parsing this will fail. We'll ignore the ID, since it's just too
796 * short to be useful as true identifier. */
797 r
= sd_id128_from_string(suuid
, &uuid
);
799 log_debug_errno(r
, "Failed to parse file system UUID '%s', ignoring: %m", suuid
);
802 r
= make_partition_devname(devname
, diskseq
, -1, flags
, &n
);
806 m
->single_file_system
= true;
807 m
->encrypted
= encrypted
;
809 m
->has_verity
= verity
&& verity
->data_path
;
810 m
->verity_ready
= verity_settings_data_covers(verity
, PARTITION_ROOT
);
812 m
->has_verity_sig
= false; /* signature not embedded, must be specified */
813 m
->verity_sig_ready
= m
->verity_ready
&& verity
->root_hash_sig
;
815 m
->image_uuid
= uuid
;
817 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
824 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
826 .rw
= !m
->verity_ready
&& !fstype_is_ro(fstype
),
828 .architecture
= _ARCHITECTURE_INVALID
,
829 .fstype
= TAKE_PTR(t
),
831 .mount_options
= TAKE_PTR(o
),
832 .mount_node_fd
= TAKE_FD(mount_node_fd
),
835 .fsmount_fd
= -EBADF
,
842 (void) blkid_probe_lookup_value(b
, "PTTYPE", &pttype
, NULL
);
846 is_gpt
= streq_ptr(pttype
, "gpt");
847 is_mbr
= streq_ptr(pttype
, "dos");
849 if (!is_gpt
&& ((flags
& DISSECT_IMAGE_GPT_ONLY
) || !is_mbr
))
852 /* We support external verity data partitions only if the image has no partition table */
853 if (verity
&& verity
->data_path
)
856 if (FLAGS_SET(flags
, DISSECT_IMAGE_ADD_PARTITION_DEVICES
)) {
857 /* Safety check: refuse block devices that carry a partition table but for which the kernel doesn't
858 * do partition scanning. */
859 r
= blockdev_partscan_enabled(fd
);
863 return -EPROTONOSUPPORT
;
866 (void) blkid_probe_lookup_value(b
, "PTUUID", &sptuuid
, NULL
);
868 r
= sd_id128_from_string(sptuuid
, &m
->image_uuid
);
870 log_debug_errno(r
, "Failed to parse partition table UUID '%s', ignoring: %m", sptuuid
);
874 pl
= blkid_probe_get_partitions(b
);
876 return errno_or_else(ENOMEM
);
879 n_partitions
= blkid_partlist_numof_partitions(pl
);
880 if (n_partitions
< 0)
881 return errno_or_else(EIO
);
883 for (int i
= 0; i
< n_partitions
; i
++) {
884 _cleanup_free_
char *node
= NULL
;
885 unsigned long long pflags
;
886 blkid_loff_t start
, size
;
891 pp
= blkid_partlist_get_partition(pl
, i
);
893 return errno_or_else(EIO
);
895 pflags
= blkid_partition_get_flags(pp
);
898 nr
= blkid_partition_get_partno(pp
);
900 return errno_or_else(EIO
);
903 start
= blkid_partition_get_start(pp
);
905 return errno_or_else(EIO
);
907 assert((uint64_t) start
< UINT64_MAX
/512);
910 size
= blkid_partition_get_size(pp
);
912 return errno_or_else(EIO
);
914 assert((uint64_t) size
< UINT64_MAX
/512);
916 /* While probing we need the non-diskseq device node name to access the thing, hence mask off
917 * DISSECT_IMAGE_DISKSEQ_DEVNODE. */
918 r
= make_partition_devname(devname
, diskseq
, nr
, flags
& ~DISSECT_IMAGE_DISKSEQ_DEVNODE
, &node
);
922 /* So here's the thing: after the main ("whole") block device popped up it might take a while
923 * before the kernel fully probed the partition table. Waiting for that to finish is icky in
924 * userspace. So here's what we do instead. We issue the BLKPG_ADD_PARTITION ioctl to add the
925 * partition ourselves, racing against the kernel. Good thing is: if this call fails with
926 * EBUSY then the kernel was quicker than us, and that's totally OK, the outcome is good for
927 * us: the device node will exist. If OTOH our call was successful we won the race. Which is
928 * also good as the outcome is the same: the partition block device exists, and we can use
931 * Kernel returns EBUSY if there's already a partition by that number or an overlapping
932 * partition already existent. */
934 if (FLAGS_SET(flags
, DISSECT_IMAGE_ADD_PARTITION_DEVICES
)) {
935 r
= block_device_add_partition(fd
, node
, nr
, (uint64_t) start
* 512, (uint64_t) size
* 512);
938 return log_debug_errno(r
, "BLKPG_ADD_PARTITION failed: %m");
940 log_debug_errno(r
, "Kernel was quicker than us in adding partition %i.", nr
);
942 log_debug("We were quicker than kernel in adding partition %i.", nr
);
946 const char *fstype
= NULL
, *label
;
947 sd_id128_t type_id
, id
;
948 GptPartitionType type
;
949 bool rw
= true, growfs
= false;
951 r
= blkid_partition_get_uuid_id128(pp
, &id
);
953 log_debug_errno(r
, "Failed to read partition UUID, ignoring: %m");
957 r
= blkid_partition_get_type_id128(pp
, &type_id
);
959 log_debug_errno(r
, "Failed to read partition type UUID, ignoring: %m");
963 type
= gpt_partition_type_from_uuid(type_id
);
965 label
= blkid_partition_get_name(pp
); /* libblkid returns NULL here if empty */
967 if (IN_SET(type
.designator
,
973 check_partition_flags(node
, pflags
,
974 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
976 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
979 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
980 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
982 } else if (type
.designator
== PARTITION_ESP
) {
984 /* Note that we don't check the SD_GPT_FLAG_NO_AUTO flag for the ESP, as it is
985 * not defined there. We instead check the SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as
986 * recommended by the UEFI spec (See "12.3.3 Number and Location of System
989 if (pflags
& SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL
)
994 } else if (type
.designator
== PARTITION_ROOT
) {
996 check_partition_flags(node
, pflags
,
997 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
999 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1002 /* If a root ID is specified, ignore everything but the root id */
1003 if (!sd_id128_is_null(root_uuid
) && !sd_id128_equal(root_uuid
, id
))
1006 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1007 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1009 } else if (type
.designator
== PARTITION_ROOT_VERITY
) {
1011 check_partition_flags(node
, pflags
,
1012 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1014 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1017 m
->has_verity
= true;
1019 /* If no verity configuration is specified, then don't do verity */
1022 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
1025 /* If root hash is specified, then ignore everything but the root id */
1026 if (!sd_id128_is_null(root_verity_uuid
) && !sd_id128_equal(root_verity_uuid
, id
))
1029 fstype
= "DM_verity_hash";
1032 } else if (type
.designator
== PARTITION_ROOT_VERITY_SIG
) {
1034 check_partition_flags(node
, pflags
,
1035 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1037 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1040 m
->has_verity_sig
= true;
1044 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
1047 fstype
= "verity_hash_signature";
1050 } else if (type
.designator
== PARTITION_USR
) {
1052 check_partition_flags(node
, pflags
,
1053 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1055 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1058 /* If a usr ID is specified, ignore everything but the usr id */
1059 if (!sd_id128_is_null(usr_uuid
) && !sd_id128_equal(usr_uuid
, id
))
1062 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1063 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1065 } else if (type
.designator
== PARTITION_USR_VERITY
) {
1067 check_partition_flags(node
, pflags
,
1068 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1070 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1073 m
->has_verity
= true;
1077 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
1080 /* If usr hash is specified, then ignore everything but the usr id */
1081 if (!sd_id128_is_null(usr_verity_uuid
) && !sd_id128_equal(usr_verity_uuid
, id
))
1084 fstype
= "DM_verity_hash";
1087 } else if (type
.designator
== PARTITION_USR_VERITY_SIG
) {
1089 check_partition_flags(node
, pflags
,
1090 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
1092 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1095 m
->has_verity_sig
= true;
1099 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
1102 fstype
= "verity_hash_signature";
1105 } else if (type
.designator
== PARTITION_SWAP
) {
1107 check_partition_flags(node
, pflags
, SD_GPT_FLAG_NO_AUTO
);
1109 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1112 /* Note: we don't set fstype = "swap" here, because we still need to probe if
1113 * it might be encrypted (i.e. fstype "crypt_LUKS") or unencrypted
1114 * (i.e. fstype "swap"), and the only way to figure that out is via fstype
1117 /* We don't have a designator for SD_GPT_LINUX_GENERIC so check the UUID instead. */
1118 } else if (sd_id128_equal(type
.uuid
, SD_GPT_LINUX_GENERIC
)) {
1120 check_partition_flags(node
, pflags
,
1121 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1123 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1127 multiple_generic
= true;
1130 generic_rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1131 generic_growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1133 generic_node
= TAKE_PTR(node
);
1136 } else if (type
.designator
== PARTITION_VAR
) {
1138 check_partition_flags(node
, pflags
,
1139 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1141 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1144 if (!FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_VAR_CHECK
)) {
1145 sd_id128_t var_uuid
;
1147 /* For /var we insist that the uuid of the partition matches the
1148 * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
1149 * ID. Why? Unlike the other partitions /var is inherently
1150 * installation specific, hence we need to be careful not to mount it
1151 * in the wrong installation. By hashing the partition UUID from
1152 * /etc/machine-id we can securely bind the partition to the
1155 r
= sd_id128_get_machine_app_specific(SD_GPT_VAR
, &var_uuid
);
1159 if (!sd_id128_equal(var_uuid
, id
)) {
1160 log_debug("Found a /var/ partition, but its UUID didn't match our expectations "
1161 "(found: " SD_ID128_UUID_FORMAT_STR
", expected: " SD_ID128_UUID_FORMAT_STR
"), ignoring.",
1162 SD_ID128_FORMAT_VAL(id
), SD_ID128_FORMAT_VAL(var_uuid
));
1167 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1168 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1171 if (type
.designator
!= _PARTITION_DESIGNATOR_INVALID
) {
1172 _cleanup_free_
char *t
= NULL
, *o
= NULL
, *l
= NULL
, *n
= NULL
;
1173 _cleanup_close_
int mount_node_fd
= -EBADF
;
1174 const char *options
= NULL
;
1176 r
= image_policy_may_use(policy
, type
.designator
);
1180 /* Policy says: ignore; Remember this fact, so that we later can distinguish between "found but ignored" and "not found at all" */
1182 if (!m
->partitions
[type
.designator
].found
)
1183 m
->partitions
[type
.designator
].ignored
= true;
1188 if (m
->partitions
[type
.designator
].found
) {
1191 /* For most partition types the first one we see wins. Except for the
1192 * rootfs and /usr, where we do a version compare of the label, and
1193 * let the newest version win. This permits a simple A/B versioning
1194 * scheme in OS images. */
1196 c
= compare_arch(type
.arch
, m
->partitions
[type
.designator
].architecture
);
1197 if (c
< 0) /* the arch we already found is better than the one we found now */
1199 if (c
== 0 && /* same arch? then go by version in label */
1200 (!partition_designator_is_versioned(type
.designator
) ||
1201 strverscmp_improved(label
, m
->partitions
[type
.designator
].label
) <= 0))
1204 dissected_partition_done(m
->partitions
+ type
.designator
);
1207 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
) &&
1208 type
.designator
!= PARTITION_SWAP
) {
1209 mount_node_fd
= open_partition(node
, /* is_partition = */ true, m
->loop
);
1210 if (mount_node_fd
< 0)
1211 return mount_node_fd
;
1214 r
= make_partition_devname(devname
, diskseq
, nr
, flags
, &n
);
1230 options
= mount_options_from_designator(mount_options
, type
.designator
);
1232 o
= strdup(options
);
1237 m
->partitions
[type
.designator
] = (DissectedPartition
) {
1242 .architecture
= type
.arch
,
1243 .node
= TAKE_PTR(n
),
1244 .fstype
= TAKE_PTR(t
),
1245 .label
= TAKE_PTR(l
),
1247 .mount_options
= TAKE_PTR(o
),
1248 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1249 .offset
= (uint64_t) start
* 512,
1250 .size
= (uint64_t) size
* 512,
1251 .gpt_flags
= pflags
,
1252 .fsmount_fd
= -EBADF
,
1256 } else if (is_mbr
) {
1258 switch (blkid_partition_get_type(pp
)) {
1260 case 0x83: /* Linux partition */
1262 if (pflags
!= 0x80) /* Bootable flag */
1266 multiple_generic
= true;
1270 generic_growfs
= false;
1271 generic_node
= TAKE_PTR(node
);
1276 case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
1277 _cleanup_close_
int mount_node_fd
= -EBADF
;
1278 _cleanup_free_
char *o
= NULL
, *n
= NULL
;
1279 sd_id128_t id
= SD_ID128_NULL
;
1280 const char *options
= NULL
;
1282 r
= image_policy_may_use(policy
, PARTITION_XBOOTLDR
);
1285 if (r
== 0) { /* policy says: ignore */
1286 if (!m
->partitions
[PARTITION_XBOOTLDR
].found
)
1287 m
->partitions
[PARTITION_XBOOTLDR
].ignored
= true;
1292 /* First one wins */
1293 if (m
->partitions
[PARTITION_XBOOTLDR
].found
)
1296 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
1297 mount_node_fd
= open_partition(node
, /* is_partition = */ true, m
->loop
);
1298 if (mount_node_fd
< 0)
1299 return mount_node_fd
;
1302 (void) blkid_partition_get_uuid_id128(pp
, &id
);
1304 r
= make_partition_devname(devname
, diskseq
, nr
, flags
, &n
);
1308 options
= mount_options_from_designator(mount_options
, PARTITION_XBOOTLDR
);
1310 o
= strdup(options
);
1315 m
->partitions
[PARTITION_XBOOTLDR
] = (DissectedPartition
) {
1320 .architecture
= _ARCHITECTURE_INVALID
,
1321 .node
= TAKE_PTR(n
),
1323 .mount_options
= TAKE_PTR(o
),
1324 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1325 .offset
= (uint64_t) start
* 512,
1326 .size
= (uint64_t) size
* 512,
1327 .fsmount_fd
= -EBADF
,
1335 if (!m
->partitions
[PARTITION_ROOT
].found
&&
1336 (m
->partitions
[PARTITION_ROOT_VERITY
].found
||
1337 m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
))
1338 return -EADDRNOTAVAIL
; /* Verity found but no matching rootfs? Something is off, refuse. */
1340 /* Hmm, we found a signature partition but no Verity data? Something is off. */
1341 if (m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_ROOT_VERITY
].found
)
1342 return -EADDRNOTAVAIL
;
1344 if (!m
->partitions
[PARTITION_USR
].found
&&
1345 (m
->partitions
[PARTITION_USR_VERITY
].found
||
1346 m
->partitions
[PARTITION_USR_VERITY_SIG
].found
))
1347 return -EADDRNOTAVAIL
; /* as above */
1350 if (m
->partitions
[PARTITION_USR_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_USR_VERITY
].found
)
1351 return -EADDRNOTAVAIL
;
1353 /* If root and /usr are combined then insist that the architecture matches */
1354 if (m
->partitions
[PARTITION_ROOT
].found
&&
1355 m
->partitions
[PARTITION_USR
].found
&&
1356 (m
->partitions
[PARTITION_ROOT
].architecture
>= 0 &&
1357 m
->partitions
[PARTITION_USR
].architecture
>= 0 &&
1358 m
->partitions
[PARTITION_ROOT
].architecture
!= m
->partitions
[PARTITION_USR
].architecture
))
1359 return -EADDRNOTAVAIL
;
1361 if (!m
->partitions
[PARTITION_ROOT
].found
&&
1362 !m
->partitions
[PARTITION_USR
].found
&&
1363 (flags
& DISSECT_IMAGE_GENERIC_ROOT
) &&
1364 (!verity
|| !verity
->root_hash
|| verity
->designator
!= PARTITION_USR
)) {
1366 /* OK, we found nothing usable, then check if there's a single generic partition, and use
1367 * that. If the root hash was set however, then we won't fall back to a generic node, because
1368 * the root hash decides. */
1370 /* If we didn't find a properly marked root partition, but we did find a single suitable
1371 * generic Linux partition, then use this as root partition, if the caller asked for it. */
1372 if (multiple_generic
)
1375 /* If we didn't find a generic node, then we can't fix this up either */
1377 r
= image_policy_may_use(policy
, PARTITION_ROOT
);
1381 /* Policy says: ignore; remember that we did */
1382 m
->partitions
[PARTITION_ROOT
].ignored
= true;
1384 _cleanup_close_
int mount_node_fd
= -EBADF
;
1385 _cleanup_free_
char *o
= NULL
, *n
= NULL
;
1386 const char *options
;
1388 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
1389 mount_node_fd
= open_partition(generic_node
, /* is_partition = */ true, m
->loop
);
1390 if (mount_node_fd
< 0)
1391 return mount_node_fd
;
1394 r
= make_partition_devname(devname
, diskseq
, generic_nr
, flags
, &n
);
1398 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
1400 o
= strdup(options
);
1405 assert(generic_nr
>= 0);
1406 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
1409 .growfs
= generic_growfs
,
1410 .partno
= generic_nr
,
1411 .architecture
= _ARCHITECTURE_INVALID
,
1412 .node
= TAKE_PTR(n
),
1413 .uuid
= generic_uuid
,
1414 .mount_options
= TAKE_PTR(o
),
1415 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1416 .offset
= UINT64_MAX
,
1418 .fsmount_fd
= -EBADF
,
1424 /* Check if we have a root fs if we are told to do check. /usr alone is fine too, but only if appropriate flag for that is set too */
1425 if (FLAGS_SET(flags
, DISSECT_IMAGE_REQUIRE_ROOT
) &&
1426 !(m
->partitions
[PARTITION_ROOT
].found
|| (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
1429 if (m
->partitions
[PARTITION_ROOT_VERITY
].found
) {
1430 /* We only support one verity partition per image, i.e. can't do for both /usr and root fs */
1431 if (m
->partitions
[PARTITION_USR_VERITY
].found
)
1434 /* We don't support verity enabled root with a split out /usr. Neither with nor without
1435 * verity there. (Note that we do support verity-less root with verity-full /usr, though.) */
1436 if (m
->partitions
[PARTITION_USR
].found
)
1437 return -EADDRNOTAVAIL
;
1441 /* If a verity designator is specified, then insist that the matching partition exists */
1442 if (verity
->designator
>= 0 && !m
->partitions
[verity
->designator
].found
)
1443 return -EADDRNOTAVAIL
;
1445 bool have_verity_sig_partition
;
1446 if (verity
->designator
>= 0)
1447 have_verity_sig_partition
= m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR_VERITY_SIG
: PARTITION_ROOT_VERITY_SIG
].found
;
1449 have_verity_sig_partition
= m
->partitions
[PARTITION_USR_VERITY_SIG
].found
|| m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
;
1451 if (verity
->root_hash
) {
1452 /* If we have an explicit root hash and found the partitions for it, then we are ready to use
1453 * Verity, set things up for it */
1455 if (verity
->designator
< 0 || verity
->designator
== PARTITION_ROOT
) {
1456 if (!m
->partitions
[PARTITION_ROOT_VERITY
].found
|| !m
->partitions
[PARTITION_ROOT
].found
)
1457 return -EADDRNOTAVAIL
;
1459 /* If we found a verity setup, then the root partition is necessarily read-only. */
1460 m
->partitions
[PARTITION_ROOT
].rw
= false;
1461 m
->verity_ready
= true;
1464 assert(verity
->designator
== PARTITION_USR
);
1466 if (!m
->partitions
[PARTITION_USR_VERITY
].found
|| !m
->partitions
[PARTITION_USR
].found
)
1467 return -EADDRNOTAVAIL
;
1469 m
->partitions
[PARTITION_USR
].rw
= false;
1470 m
->verity_ready
= true;
1473 if (m
->verity_ready
)
1474 m
->verity_sig_ready
= verity
->root_hash_sig
|| have_verity_sig_partition
;
1476 } else if (have_verity_sig_partition
) {
1478 /* If we found an embedded signature partition, we are ready, too. */
1480 m
->verity_ready
= m
->verity_sig_ready
= true;
1481 if (verity
->designator
>= 0)
1482 m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR
: PARTITION_ROOT
].rw
= false;
1483 else if (m
->partitions
[PARTITION_USR_VERITY_SIG
].found
)
1484 m
->partitions
[PARTITION_USR
].rw
= false;
1485 else if (m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
)
1486 m
->partitions
[PARTITION_ROOT
].rw
= false;
1492 /* After we discovered all partitions let's see if the verity requirements match the policy. (Note:
1493 * we don't check encryption requirements here, because we haven't probed the file system yet, hence
1494 * don't know if this is encrypted or not) */
1495 for (PartitionDesignator di
= 0; di
< _PARTITION_DESIGNATOR_MAX
; di
++) {
1496 PartitionDesignator vi
, si
;
1497 PartitionPolicyFlags found_flags
;
1499 any
= any
|| m
->partitions
[di
].found
;
1501 vi
= partition_verity_of(di
);
1502 si
= partition_verity_sig_of(di
);
1504 /* Determine the verity protection level for this partition. */
1505 found_flags
= m
->partitions
[di
].found
?
1506 (vi
>= 0 && m
->partitions
[vi
].found
?
1507 (si
>= 0 && m
->partitions
[si
].found
? PARTITION_POLICY_SIGNED
: PARTITION_POLICY_VERITY
) :
1508 PARTITION_POLICY_ENCRYPTED
|PARTITION_POLICY_UNPROTECTED
) :
1509 (m
->partitions
[di
].ignored
? PARTITION_POLICY_UNUSED
: PARTITION_POLICY_ABSENT
);
1511 r
= image_policy_check_protection(policy
, di
, found_flags
);
1515 if (m
->partitions
[di
].found
) {
1516 r
= image_policy_check_partition_flags(policy
, di
, m
->partitions
[di
].gpt_flags
);
1522 if (!any
&& !FLAGS_SET(flags
, DISSECT_IMAGE_ALLOW_EMPTY
))
1525 r
= dissected_image_probe_filesystems(m
, fd
, policy
);
1533 int dissect_image_file(
1535 const VeritySettings
*verity
,
1536 const MountOptions
*mount_options
,
1537 const ImagePolicy
*image_policy
,
1538 DissectImageFlags flags
,
1539 DissectedImage
**ret
) {
1542 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
1543 _cleanup_close_
int fd
= -EBADF
;
1549 fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
1553 if (fstat(fd
, &st
) < 0)
1556 r
= stat_verify_regular(&st
);
1560 r
= dissected_image_new(path
, &m
);
1564 m
->image_size
= st
.st_size
;
1566 r
= probe_sector_size(fd
, &m
->sector_size
);
1570 r
= dissect_image(m
, fd
, path
, verity
, mount_options
, image_policy
, flags
);
1582 int dissect_log_error(int log_level
, int r
, const char *name
, const VeritySettings
*verity
) {
1583 assert(log_level
>= 0 && log_level
<= LOG_DEBUG
);
1588 case 0 ... INT_MAX
: /* success! */
1592 return log_full_errno(log_level
, r
, "Dissecting images is not supported, compiled without blkid support.");
1595 return log_full_errno(log_level
, r
, "%s: Couldn't identify a suitable partition table or file system.", name
);
1598 return log_full_errno(log_level
, r
, "%s: The image does not pass os-release/extension-release validation.", name
);
1600 case -EADDRNOTAVAIL
:
1601 return log_full_errno(log_level
, r
, "%s: No root partition for specified root hash found.", name
);
1604 return log_full_errno(log_level
, r
, "%s: Multiple suitable root partitions found in image.", name
);
1607 return log_full_errno(log_level
, r
, "%s: No suitable root partition found in image.", name
);
1609 case -EPROTONOSUPPORT
:
1610 return log_full_errno(log_level
, r
, "Device '%s' is a loopback block device with partition scanning turned off, please turn it on.", name
);
1613 return log_full_errno(log_level
, r
, "%s: Image is not a block device.", name
);
1616 return log_full_errno(log_level
, r
,
1617 "Combining partitioned images (such as '%s') with external Verity data (such as '%s') not supported. "
1618 "(Consider setting $SYSTEMD_DISSECT_VERITY_SIDECAR=0 to disable automatic discovery of external Verity data.)",
1619 name
, strna(verity
? verity
->data_path
: NULL
));
1622 return log_full_errno(log_level
, r
, "%s: image does not match image policy.", name
);
1625 return log_full_errno(log_level
, r
, "%s: no suitable partitions found.", name
);
1628 return log_full_errno(log_level
, r
, "%s: cannot dissect image: %m", name
);
1632 int dissect_image_file_and_warn(
1634 const VeritySettings
*verity
,
1635 const MountOptions
*mount_options
,
1636 const ImagePolicy
*image_policy
,
1637 DissectImageFlags flags
,
1638 DissectedImage
**ret
) {
1640 return dissect_log_error(
1642 dissect_image_file(path
, verity
, mount_options
, image_policy
, flags
, ret
),
1647 void dissected_image_close(DissectedImage
*m
) {
1651 /* Closes all fds we keep open associated with this, but nothing else */
1653 FOREACH_ARRAY(p
, m
->partitions
, _PARTITION_DESIGNATOR_MAX
) {
1654 p
->mount_node_fd
= safe_close(p
->mount_node_fd
);
1655 p
->fsmount_fd
= safe_close(p
->fsmount_fd
);
1658 m
->loop
= loop_device_unref(m
->loop
);
1661 DissectedImage
* dissected_image_unref(DissectedImage
*m
) {
1665 /* First, clear dissected partitions. */
1666 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++)
1667 dissected_partition_done(m
->partitions
+ i
);
1669 /* Second, free decrypted images. This must be after dissected_partition_done(), as freeing
1670 * DecryptedImage may try to deactivate partitions. */
1671 decrypted_image_unref(m
->decrypted_image
);
1673 /* Third, unref LoopDevice. This must be called after the above two, as freeing LoopDevice may try to
1674 * remove existing partitions on the loopback block device. */
1675 loop_device_unref(m
->loop
);
1677 free(m
->image_name
);
1679 strv_free(m
->machine_info
);
1680 strv_free(m
->os_release
);
1681 strv_free(m
->initrd_release
);
1682 strv_free(m
->confext_release
);
1683 strv_free(m
->sysext_release
);
1688 static int is_loop_device(const char *path
) {
1689 char s
[SYS_BLOCK_PATH_MAX("/../loop/")];
1694 if (stat(path
, &st
) < 0)
1697 if (!S_ISBLK(st
.st_mode
))
1700 xsprintf_sys_block_path(s
, "/loop/", st
.st_dev
);
1701 if (access(s
, F_OK
) < 0) {
1702 if (errno
!= ENOENT
)
1705 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
1706 xsprintf_sys_block_path(s
, "/../loop/", st
.st_dev
);
1707 if (access(s
, F_OK
) < 0)
1708 return errno
== ENOENT
? false : -errno
;
1714 static int run_fsck(int node_fd
, const char *fstype
) {
1718 assert(node_fd
>= 0);
1721 r
= fsck_exists_for_fstype(fstype
);
1723 log_debug_errno(r
, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype
);
1727 log_debug("Not checking partition %s, as fsck for %s does not exist.", FORMAT_PROC_FD_PATH(node_fd
), fstype
);
1734 &node_fd
, 1, /* Leave the node fd open */
1735 FORK_RESET_SIGNALS
|FORK_CLOSE_ALL_FDS
|FORK_RLIMIT_NOFILE_SAFE
|FORK_DEATHSIG_SIGTERM
|FORK_REARRANGE_STDIO
|FORK_CLOEXEC_OFF
,
1738 return log_debug_errno(r
, "Failed to fork off fsck: %m");
1741 execlp("fsck", "fsck", "-aT", FORMAT_PROC_FD_PATH(node_fd
), NULL
);
1743 log_debug_errno(errno
, "Failed to execl() fsck: %m");
1744 _exit(FSCK_OPERATIONAL_ERROR
);
1747 exit_status
= wait_for_terminate_and_check("fsck", pid
, 0);
1748 if (exit_status
< 0)
1749 return log_debug_errno(exit_status
, "Failed to fork off fsck: %m");
1751 if ((exit_status
& ~FSCK_ERROR_CORRECTED
) != FSCK_SUCCESS
) {
1752 log_debug("fsck failed with exit status %i.", exit_status
);
1754 if ((exit_status
& (FSCK_SYSTEM_SHOULD_REBOOT
|FSCK_ERRORS_LEFT_UNCORRECTED
)) != 0)
1755 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
), "File system is corrupted, refusing.");
1757 log_debug("Ignoring fsck error.");
1763 static int fs_grow(const char *node_path
, int mount_fd
, const char *mount_path
) {
1764 _cleanup_close_
int _mount_fd
= -EBADF
, node_fd
= -EBADF
;
1765 uint64_t size
, newsize
;
1770 assert(mount_fd
>= 0 || mount_path
);
1772 node_fd
= open(node_path
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
1774 return log_debug_errno(errno
, "Failed to open node device %s: %m", node_path
);
1776 r
= blockdev_get_device_size(node_fd
, &size
);
1778 return log_debug_errno(r
, "Failed to get block device size of %s: %m", node_path
);
1783 _mount_fd
= open(mount_path
, O_RDONLY
|O_DIRECTORY
|O_CLOEXEC
);
1785 return log_debug_errno(errno
, "Failed to open mounted file system %s: %m", mount_path
);
1787 mount_fd
= _mount_fd
;
1789 mount_fd
= fd_reopen_condition(mount_fd
, O_RDONLY
|O_DIRECTORY
|O_CLOEXEC
, O_RDONLY
|O_DIRECTORY
|O_CLOEXEC
, &_mount_fd
);
1791 return log_debug_errno(errno
, "Failed to reopen mount node: %m");
1794 id
= mount_path
?: node_path
;
1796 log_debug("Resizing \"%s\" to %"PRIu64
" bytes...", id
, size
);
1797 r
= resize_fs(mount_fd
, size
, &newsize
);
1799 return log_debug_errno(r
, "Failed to resize \"%s\" to %"PRIu64
" bytes: %m", id
, size
);
1801 if (newsize
== size
)
1802 log_debug("Successfully resized \"%s\" to %s bytes.",
1803 id
, FORMAT_BYTES(newsize
));
1805 assert(newsize
< size
);
1806 log_debug("Successfully resized \"%s\" to %s bytes (%"PRIu64
" bytes lost due to blocksize).",
1807 id
, FORMAT_BYTES(newsize
), size
- newsize
);
1813 int partition_pick_mount_options(
1814 PartitionDesignator d
,
1819 unsigned long *ret_ms_flags
) {
1821 _cleanup_free_
char *options
= NULL
;
1823 assert(ret_options
);
1825 /* Selects a baseline of bind mount flags, that should always apply.
1827 * Firstly, we set MS_NODEV universally on all mounts, since we don't want to allow device nodes outside of /dev/.
1829 * On /var/tmp/ we'll also set MS_NOSUID, same as we set for /tmp/ on the host.
1831 * On the ESP and XBOOTLDR partitions we'll also disable symlinks, and execution. These file systems
1832 * are generally untrusted (i.e. not encrypted or authenticated), and typically VFAT hence we should
1833 * be as restrictive as possible, and this shouldn't hurt, since the functionality is not available
1836 unsigned long flags
= MS_NODEV
;
1844 case PARTITION_XBOOTLDR
:
1845 flags
|= MS_NOSUID
|MS_NOEXEC
|ms_nosymfollow_supported();
1847 /* The ESP might contain a pre-boot random seed. Let's make this unaccessible to regular
1848 * userspace. ESP/XBOOTLDR is almost certainly VFAT, hence if we don't know assume it is. */
1849 if (!fstype
|| fstype_can_umask(fstype
))
1850 if (!strextend_with_separator(&options
, ",", "umask=0077"))
1862 /* So, when you request MS_RDONLY from ext4, then this means nothing. It happily still writes to the
1863 * backing storage. What's worse, the BLKRO[GS]ET flag and (in case of loopback devices)
1864 * LO_FLAGS_READ_ONLY don't mean anything, they affect userspace accesses only, and write accesses
1865 * from the upper file system still get propagated through to the underlying file system,
1866 * unrestricted. To actually get ext4/xfs/btrfs to stop writing to the device we need to specify
1867 * "norecovery" as mount option, in addition to MS_RDONLY. Yes, this sucks, since it means we need to
1868 * carry a per file system table here.
1870 * Note that this means that we might not be able to mount corrupted file systems as read-only
1871 * anymore (since in some cases the kernel implementations will refuse mounting when corrupted,
1872 * read-only and "norecovery" is specified). But I think for the case of automatically determined
1873 * mount options for loopback devices this is the right choice, since otherwise using the same
1874 * loopback file twice even in read-only mode, is going to fail badly sooner or later. The use case of
1875 * making reuse of the immutable images "just work" is more relevant to us than having read-only
1876 * access that actually modifies stuff work on such image files. Or to say this differently: if
1877 * people want their file systems to be fixed up they should just open them in writable mode, where
1878 * all these problems don't exist. */
1879 if (!rw
&& fstype
&& fstype_can_norecovery(fstype
))
1880 if (!strextend_with_separator(&options
, ",", "norecovery"))
1883 if (discard
&& fstype
&& fstype_can_discard(fstype
))
1884 if (!strextend_with_separator(&options
, ",", "discard"))
1887 if (!ret_ms_flags
) /* Fold flags into option string if ret_flags specified as NULL */
1888 if (!strextend_with_separator(&options
, ",",
1889 FLAGS_SET(flags
, MS_RDONLY
) ? "ro" : "rw",
1890 FLAGS_SET(flags
, MS_NODEV
) ? "nodev" : "dev",
1891 FLAGS_SET(flags
, MS_NOSUID
) ? "nosuid" : "suid",
1892 FLAGS_SET(flags
, MS_NOEXEC
) ? "noexec" : "exec",
1893 FLAGS_SET(flags
, MS_NOSYMFOLLOW
) ? "nosymfollow" : NULL
))
1894 /* NB: we suppress 'symfollow' here, since it's the default, and old /bin/mount might not know it */
1898 *ret_ms_flags
= flags
;
1900 *ret_options
= TAKE_PTR(options
);
1904 static bool need_user_mapping(uid_t uid_shift
, uid_t uid_range
) {
1906 if (!uid_is_valid(uid_shift
))
1909 return uid_shift
!= 0 || uid_range
!= UINT32_MAX
;
1912 static int mount_partition(
1913 PartitionDesignator d
,
1914 DissectedPartition
*m
,
1916 const char *directory
,
1920 DissectImageFlags flags
) {
1922 _cleanup_free_
char *chased
= NULL
, *options
= NULL
;
1923 const char *p
= NULL
, *node
, *fstype
= NULL
;
1924 bool rw
, discard
, grow
;
1925 unsigned long ms_flags
;
1933 /* Check the various combinations when we can't do anything anymore */
1934 if (m
->fsmount_fd
< 0 && m
->mount_node_fd
< 0)
1936 if (m
->fsmount_fd
>= 0 && !where
)
1938 if (!where
&& m
->mount_node_fd
< 0)
1941 if (m
->fsmount_fd
< 0) {
1942 fstype
= dissected_partition_fstype(m
);
1944 return -EAFNOSUPPORT
;
1946 /* We are looking at an encrypted partition? This either means stacked encryption, or the
1947 * caller didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error
1949 if (streq(fstype
, "crypto_LUKS"))
1952 r
= dissect_fstype_ok(fstype
);
1956 return -EIDRM
; /* Recognizable error */
1959 node
= m
->mount_node_fd
< 0 ? NULL
: FORMAT_PROC_FD_PATH(m
->mount_node_fd
);
1960 rw
= m
->rw
&& !(flags
& DISSECT_IMAGE_MOUNT_READ_ONLY
);
1962 discard
= ((flags
& DISSECT_IMAGE_DISCARD
) ||
1963 ((flags
& DISSECT_IMAGE_DISCARD_ON_LOOP
) && (m
->node
&& is_loop_device(m
->node
) > 0)));
1965 grow
= rw
&& m
->growfs
&& FLAGS_SET(flags
, DISSECT_IMAGE_GROWFS
);
1967 if (FLAGS_SET(flags
, DISSECT_IMAGE_FSCK
) && rw
&& m
->mount_node_fd
>= 0 && m
->fsmount_fd
< 0) {
1968 r
= run_fsck(m
->mount_node_fd
, fstype
);
1975 /* Automatically create missing mount points inside the image, if necessary. */
1976 r
= mkdir_p_root(where
, directory
, uid_shift
, (gid_t
) uid_shift
, 0755, NULL
);
1977 if (r
< 0 && r
!= -EROFS
)
1980 r
= chase(directory
, where
, CHASE_PREFIX_ROOT
, &chased
, NULL
);
1986 /* Create top-level mount if missing – but only if this is asked for. This won't modify the
1987 * image (as the branch above does) but the host hierarchy, and the created directory might
1988 * survive our mount in the host hierarchy hence. */
1989 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
1990 r
= mkdir_p(where
, 0755);
1999 if (m
->fsmount_fd
< 0) {
2000 r
= partition_pick_mount_options(d
, fstype
, rw
, discard
, &options
, &ms_flags
);
2004 if (need_user_mapping(uid_shift
, uid_range
) && fstype_can_uid_gid(fstype
)) {
2005 _cleanup_free_
char *uid_option
= NULL
;
2007 if (asprintf(&uid_option
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
2010 if (!strextend_with_separator(&options
, ",", uid_option
))
2013 userns_fd
= -EBADF
; /* Not needed */
2016 if (!isempty(m
->mount_options
))
2017 if (!strextend_with_separator(&options
, ",", m
->mount_options
))
2022 if (m
->fsmount_fd
>= 0) {
2023 /* Case #1: Attach existing fsmount fd to the file system */
2025 r
= mount_exchange_graceful(
2028 FLAGS_SET(flags
, DISSECT_IMAGE_TRY_ATOMIC_MOUNT_EXCHANGE
));
2030 return log_debug_errno(r
, "Failed to mount image on '%s': %m", p
);
2035 /* Case #2: Mount directly into place */
2036 r
= mount_nofollow_verbose(LOG_DEBUG
, node
, p
, fstype
, ms_flags
, options
);
2041 (void) fs_grow(node
, -EBADF
, p
);
2043 if (userns_fd
>= 0) {
2044 r
= remount_idmap_fd(STRV_MAKE(p
), userns_fd
);
2052 /* Case #3: Create fsmount fd */
2054 m
->fsmount_fd
= make_fsmount(LOG_DEBUG
, node
, fstype
, ms_flags
, options
, userns_fd
);
2055 if (m
->fsmount_fd
< 0)
2056 return m
->fsmount_fd
;
2059 (void) fs_grow(node
, m
->fsmount_fd
, NULL
);
2065 static int mount_root_tmpfs(const char *where
, uid_t uid_shift
, uid_t uid_range
, DissectImageFlags flags
) {
2066 _cleanup_free_
char *options
= NULL
;
2071 /* For images that contain /usr/ but no rootfs, let's mount rootfs as tmpfs */
2073 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
2074 r
= mkdir_p(where
, 0755);
2079 if (need_user_mapping(uid_shift
, uid_range
)) {
2080 if (asprintf(&options
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
2084 r
= mount_nofollow_verbose(LOG_DEBUG
, "rootfs", where
, "tmpfs", MS_NODEV
, options
);
2091 static int mount_point_is_available(const char *where
, const char *path
, bool missing_ok
) {
2092 _cleanup_free_
char *p
= NULL
;
2095 /* Check whether <path> is suitable as a mountpoint, i.e. is an empty directory
2096 * or does not exist at all (when missing_ok). */
2098 r
= chase(path
, where
, CHASE_PREFIX_ROOT
, &p
, NULL
);
2102 return log_debug_errno(r
, "Failed to chase \"%s\": %m", path
);
2104 r
= dir_is_empty(p
, /* ignore_hidden_or_backup= */ false);
2108 return log_debug_errno(r
, "Failed to check directory \"%s\": %m", p
);
2112 int dissected_image_mount(
2118 DissectImageFlags flags
) {
2120 _cleanup_close_
int my_userns_fd
= -EBADF
;
2125 /* If 'where' is NULL then we'll use the new mount API to create fsmount() fds for the mounts and
2126 * store them in DissectedPartition.fsmount_fd.
2128 * If 'where' is not NULL then we'll either mount the partitions to the right places ourselves,
2129 * or use DissectedPartition.fsmount_fd and bind it to the right places.
2131 * This allows splitting the setting up up the superblocks and the binding to file systems paths into
2132 * two distinct and differently privileged components: one that gets the fsmount fds, and the other
2133 * that then applies them.
2137 * -ENXIO → No root partition found
2138 * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release/extension-release file found
2139 * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet
2140 * -EUCLEAN → fsck for file system failed
2141 * -EBUSY → File system already mounted/used elsewhere (kernel)
2142 * -EAFNOSUPPORT → File system type not supported or not known
2143 * -EIDRM → File system is not among allowlisted "common" file systems
2146 if (!where
&& (flags
& (DISSECT_IMAGE_VALIDATE_OS
|DISSECT_IMAGE_VALIDATE_OS_EXT
)) != 0)
2147 return -EOPNOTSUPP
; /* for now, not supported */
2149 if (!(m
->partitions
[PARTITION_ROOT
].found
||
2150 (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
2151 return -ENXIO
; /* Require a root fs or at least a /usr/ fs (the latter is subject to a flag of its own) */
2153 if (userns_fd
< 0 && need_user_mapping(uid_shift
, uid_range
) && FLAGS_SET(flags
, DISSECT_IMAGE_MOUNT_IDMAPPED
)) {
2155 my_userns_fd
= make_userns(uid_shift
, uid_range
, UID_INVALID
, UID_INVALID
, REMOUNT_IDMAPPING_HOST_ROOT
);
2156 if (my_userns_fd
< 0)
2157 return my_userns_fd
;
2159 userns_fd
= my_userns_fd
;
2162 if ((flags
& DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
) == 0) {
2164 /* First mount the root fs. If there's none we use a tmpfs. */
2165 if (m
->partitions
[PARTITION_ROOT
].found
) {
2166 r
= mount_partition(PARTITION_ROOT
, m
->partitions
+ PARTITION_ROOT
, where
, NULL
, uid_shift
, uid_range
, userns_fd
, flags
);
2171 r
= mount_root_tmpfs(where
, uid_shift
, uid_range
, flags
);
2176 /* For us mounting root always means mounting /usr as well */
2177 r
= mount_partition(PARTITION_USR
, m
->partitions
+ PARTITION_USR
, where
, "/usr", uid_shift
, uid_range
, userns_fd
, flags
);
2182 if ((flags
& DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
) == 0 &&
2183 (flags
& (DISSECT_IMAGE_VALIDATE_OS
|DISSECT_IMAGE_VALIDATE_OS_EXT
)) != 0) {
2184 /* If either one of the validation flags are set, ensure that the image qualifies as
2185 * one or the other (or both). */
2190 if (FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS
)) {
2191 r
= path_is_os_tree(where
);
2197 if (!ok
&& FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS_EXT
) && m
->image_name
) {
2198 r
= extension_has_forbidden_content(where
);
2202 r
= path_is_extension_tree(IMAGE_SYSEXT
, where
, m
->image_name
, FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_EXTENSION_CHECK
));
2204 r
= path_is_extension_tree(IMAGE_CONFEXT
, where
, m
->image_name
, FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_EXTENSION_CHECK
));
2216 if (flags
& DISSECT_IMAGE_MOUNT_ROOT_ONLY
)
2219 r
= mount_partition(PARTITION_HOME
, m
->partitions
+ PARTITION_HOME
, where
, "/home", uid_shift
, uid_range
, userns_fd
, flags
);
2223 r
= mount_partition(PARTITION_SRV
, m
->partitions
+ PARTITION_SRV
, where
, "/srv", uid_shift
, uid_range
, userns_fd
, flags
);
2227 r
= mount_partition(PARTITION_VAR
, m
->partitions
+ PARTITION_VAR
, where
, "/var", uid_shift
, uid_range
, userns_fd
, flags
);
2231 r
= mount_partition(PARTITION_TMP
, m
->partitions
+ PARTITION_TMP
, where
, "/var/tmp", uid_shift
, uid_range
, userns_fd
, flags
);
2235 int slash_boot_is_available
= 0;
2237 r
= slash_boot_is_available
= mount_point_is_available(where
, "/boot", /* missing_ok = */ true);
2241 if (!where
|| slash_boot_is_available
) {
2242 r
= mount_partition(PARTITION_XBOOTLDR
, m
->partitions
+ PARTITION_XBOOTLDR
, where
, "/boot", uid_shift
, uid_range
, userns_fd
, flags
);
2245 slash_boot_is_available
= !r
;
2248 if (m
->partitions
[PARTITION_ESP
].found
) {
2249 const char *esp_path
= NULL
;
2252 /* Mount the ESP to /boot/ if it exists and is empty and we didn't already mount the
2253 * XBOOTLDR partition into it. Otherwise, use /efi instead, but only if it exists
2256 if (slash_boot_is_available
) {
2257 r
= mount_point_is_available(where
, "/boot", /* missing_ok = */ false);
2265 r
= mount_point_is_available(where
, "/efi", /* missing_ok = */ true);
2273 /* OK, let's mount the ESP now (possibly creating the dir if missing) */
2274 r
= mount_partition(PARTITION_ESP
, m
->partitions
+ PARTITION_ESP
, where
, esp_path
, uid_shift
, uid_range
, userns_fd
, flags
);
2282 int dissected_image_mount_and_warn(
2288 DissectImageFlags flags
) {
2294 r
= dissected_image_mount(m
, where
, uid_shift
, uid_range
, userns_fd
, flags
);
2296 return log_error_errno(r
, "Failed to mount image: No root file system found in image.");
2297 if (r
== -EMEDIUMTYPE
)
2298 return log_error_errno(r
, "Failed to mount image: No suitable os-release/extension-release file in image found.");
2300 return log_error_errno(r
, "Failed to mount image: Encrypted file system discovered, but decryption not requested.");
2302 return log_error_errno(r
, "Failed to mount image: File system check on image failed.");
2304 return log_error_errno(r
, "Failed to mount image: File system already mounted elsewhere.");
2305 if (r
== -EAFNOSUPPORT
)
2306 return log_error_errno(r
, "Failed to mount image: File system type not supported or not known.");
2308 return log_error_errno(r
, "Failed to mount image: File system is too uncommon, refused.");
2310 return log_error_errno(r
, "Failed to mount image: %m");
2315 #if HAVE_LIBCRYPTSETUP
2316 struct DecryptedPartition
{
2317 struct crypt_device
*device
;
2323 typedef struct DecryptedPartition DecryptedPartition
;
2325 struct DecryptedImage
{
2327 DecryptedPartition
*decrypted
;
2331 static DecryptedImage
* decrypted_image_free(DecryptedImage
*d
) {
2332 #if HAVE_LIBCRYPTSETUP
2338 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
2339 DecryptedPartition
*p
= d
->decrypted
+ i
;
2341 if (p
->device
&& p
->name
&& !p
->relinquished
) {
2342 _cleanup_free_
char *node
= NULL
;
2344 node
= path_join("/dev/mapper", p
->name
);
2346 r
= btrfs_forget_device(node
);
2347 if (r
< 0 && r
!= -ENOENT
)
2348 log_debug_errno(r
, "Failed to forget btrfs device %s, ignoring: %m", node
);
2352 /* Let's deactivate lazily, as the dm volume may be already/still used by other processes. */
2353 r
= sym_crypt_deactivate_by_name(p
->device
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
2355 log_debug_errno(r
, "Failed to deactivate encrypted partition %s", p
->name
);
2359 sym_crypt_free(p
->device
);
2369 DEFINE_TRIVIAL_REF_UNREF_FUNC(DecryptedImage
, decrypted_image
, decrypted_image_free
);
2371 #if HAVE_LIBCRYPTSETUP
2372 static int decrypted_image_new(DecryptedImage
**ret
) {
2373 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
2377 d
= new(DecryptedImage
, 1);
2381 *d
= (DecryptedImage
) {
2389 static int make_dm_name_and_node(const void *original_node
, const char *suffix
, char **ret_name
, char **ret_node
) {
2390 _cleanup_free_
char *name
= NULL
, *node
= NULL
;
2393 assert(original_node
);
2398 base
= strrchr(original_node
, '/');
2400 base
= original_node
;
2406 name
= strjoin(base
, suffix
);
2409 if (!filename_is_valid(name
))
2412 node
= path_join(sym_crypt_get_dir(), name
);
2416 *ret_name
= TAKE_PTR(name
);
2417 *ret_node
= TAKE_PTR(node
);
2422 static int decrypt_partition(
2423 DissectedPartition
*m
,
2424 const char *passphrase
,
2425 DissectImageFlags flags
,
2426 DecryptedImage
*d
) {
2428 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
2429 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2430 _cleanup_close_
int fd
= -EBADF
;
2436 if (!m
->found
|| !m
->node
|| !m
->fstype
)
2439 if (!streq(m
->fstype
, "crypto_LUKS"))
2445 r
= dlopen_cryptsetup();
2449 r
= make_dm_name_and_node(m
->node
, "-decrypted", &name
, &node
);
2453 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
2456 r
= sym_crypt_init(&cd
, m
->node
);
2458 return log_debug_errno(r
, "Failed to initialize dm-crypt: %m");
2460 cryptsetup_enable_logging(cd
);
2462 r
= sym_crypt_load(cd
, CRYPT_LUKS
, NULL
);
2464 return log_debug_errno(r
, "Failed to load LUKS metadata: %m");
2466 r
= sym_crypt_activate_by_passphrase(cd
, name
, CRYPT_ANY_SLOT
, passphrase
, strlen(passphrase
),
2467 ((flags
& DISSECT_IMAGE_DEVICE_READ_ONLY
) ? CRYPT_ACTIVATE_READONLY
: 0) |
2468 ((flags
& DISSECT_IMAGE_DISCARD_ON_CRYPTO
) ? CRYPT_ACTIVATE_ALLOW_DISCARDS
: 0));
2470 log_debug_errno(r
, "Failed to activate LUKS device: %m");
2471 return r
== -EPERM
? -EKEYREJECTED
: r
;
2474 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2476 return log_debug_errno(errno
, "Failed to open %s: %m", node
);
2478 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
2479 .name
= TAKE_PTR(name
),
2480 .device
= TAKE_PTR(cd
),
2483 m
->decrypted_node
= TAKE_PTR(node
);
2484 close_and_replace(m
->mount_node_fd
, fd
);
2489 static int verity_can_reuse(
2490 const VeritySettings
*verity
,
2492 struct crypt_device
**ret_cd
) {
2494 /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
2495 _cleanup_free_
char *root_hash_existing
= NULL
;
2496 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2497 struct crypt_params_verity crypt_params
= {};
2498 size_t root_hash_existing_size
;
2505 r
= sym_crypt_init_by_name(&cd
, name
);
2507 return log_debug_errno(r
, "Error opening verity device, crypt_init_by_name failed: %m");
2509 cryptsetup_enable_logging(cd
);
2511 r
= sym_crypt_get_verity_info(cd
, &crypt_params
);
2513 return log_debug_errno(r
, "Error opening verity device, crypt_get_verity_info failed: %m");
2515 root_hash_existing_size
= verity
->root_hash_size
;
2516 root_hash_existing
= malloc0(root_hash_existing_size
);
2517 if (!root_hash_existing
)
2520 r
= sym_crypt_volume_key_get(cd
, CRYPT_ANY_SLOT
, root_hash_existing
, &root_hash_existing_size
, NULL
, 0);
2522 return log_debug_errno(r
, "Error opening verity device, crypt_volume_key_get failed: %m");
2523 if (verity
->root_hash_size
!= root_hash_existing_size
||
2524 memcmp(root_hash_existing
, verity
->root_hash
, verity
->root_hash_size
) != 0)
2525 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but root hashes are different.");
2527 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
2528 /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the
2529 * same settings, so that a previous unsigned mount will not be reused if the user asks to use
2530 * signing for the new one, and vice versa. */
2531 if (!!verity
->root_hash_sig
!= !!(crypt_params
.flags
& CRYPT_VERITY_ROOT_HASH_SIGNATURE
))
2532 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but signature settings are not the same.");
2535 *ret_cd
= TAKE_PTR(cd
);
2539 static char* dm_deferred_remove_clean(char *name
) {
2543 (void) sym_crypt_deactivate_by_name(NULL
, name
, CRYPT_DEACTIVATE_DEFERRED
);
2546 DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean
);
2548 static int validate_signature_userspace(const VeritySettings
*verity
, DissectImageFlags flags
) {
2551 if (!FLAGS_SET(flags
, DISSECT_IMAGE_ALLOW_USERSPACE_VERITY
)) {
2552 log_debug("Userspace dm-verity signature authentication disabled via flag.");
2556 r
= secure_getenv_bool("SYSTEMD_ALLOW_USERSPACE_VERITY");
2557 if (r
< 0 && r
!= -ENXIO
) {
2558 log_debug_errno(r
, "Failed to parse $SYSTEMD_ALLOW_USERSPACE_VERITY environment variable, refusing userspace dm-verity signature authentication.");
2562 log_debug("Userspace dm-verity signature authentication disabled via $SYSTEMD_ALLOW_USERSPACE_VERITY environment variable.");
2567 r
= proc_cmdline_get_bool("systemd.allow_userspace_verity", PROC_CMDLINE_TRUE_WHEN_MISSING
, &b
);
2569 log_debug_errno(r
, "Failed to parse systemd.allow_userspace_verity= kernel command line option, refusing userspace dm-verity signature authentication.");
2573 log_debug("Userspace dm-verity signature authentication disabled via systemd.allow_userspace_verity= kernel command line variable.");
2578 _cleanup_(sk_X509_free_allp
) STACK_OF(X509
) *sk
= NULL
;
2579 _cleanup_strv_free_
char **certs
= NULL
;
2580 _cleanup_(PKCS7_freep
) PKCS7
*p7
= NULL
;
2581 _cleanup_free_
char *s
= NULL
;
2582 _cleanup_(BIO_freep
) BIO
*bio
= NULL
; /* 'bio' must be freed first, 's' second, hence keep this order
2583 * of declaration in place, please */
2584 const unsigned char *d
;
2587 assert(verity
->root_hash
);
2588 assert(verity
->root_hash_sig
);
2590 /* Because installing a signature certificate into the kernel chain is so messy, let's optionally do
2591 * userspace validation. */
2593 r
= conf_files_list_nulstr(&certs
, ".crt", NULL
, CONF_FILES_REGULAR
|CONF_FILES_FILTER_MASKED
, CONF_PATHS_NULSTR("verity.d"));
2595 return log_debug_errno(r
, "Failed to enumerate certificates: %m");
2596 if (strv_isempty(certs
)) {
2597 log_debug("No userspace dm-verity certificates found.");
2601 d
= verity
->root_hash_sig
;
2602 p7
= d2i_PKCS7(NULL
, &d
, (long) verity
->root_hash_sig_size
);
2604 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to parse PKCS7 DER signature data.");
2606 s
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2608 return log_oom_debug();
2610 bio
= BIO_new_mem_buf(s
, strlen(s
));
2612 return log_oom_debug();
2614 sk
= sk_X509_new_null();
2616 return log_oom_debug();
2618 STRV_FOREACH(i
, certs
) {
2619 _cleanup_(X509_freep
) X509
*c
= NULL
;
2620 _cleanup_fclose_
FILE *f
= NULL
;
2622 f
= fopen(*i
, "re");
2624 log_debug_errno(errno
, "Failed to open '%s', ignoring: %m", *i
);
2628 c
= PEM_read_X509(f
, NULL
, NULL
, NULL
);
2630 log_debug("Failed to load X509 certificate '%s', ignoring.", *i
);
2634 if (sk_X509_push(sk
, c
) == 0)
2635 return log_oom_debug();
2640 r
= PKCS7_verify(p7
, sk
, NULL
, bio
, NULL
, PKCS7_NOINTERN
|PKCS7_NOVERIFY
);
2642 log_debug("Userspace PKCS#7 validation succeeded.");
2644 log_debug("Userspace PKCS#7 validation failed: %s", ERR_error_string(ERR_get_error(), NULL
));
2648 log_debug("Not doing client-side validation of dm-verity root hash signatures, OpenSSL support disabled.");
2653 static int do_crypt_activate_verity(
2654 struct crypt_device
*cd
,
2656 const VeritySettings
*verity
,
2657 DissectImageFlags flags
) {
2659 bool check_signature
;
2666 if (verity
->root_hash_sig
) {
2667 r
= secure_getenv_bool("SYSTEMD_DISSECT_VERITY_SIGNATURE");
2668 if (r
< 0 && r
!= -ENXIO
)
2669 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIGNATURE");
2671 check_signature
= r
!= 0;
2673 check_signature
= false;
2675 if (check_signature
) {
2677 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
2678 /* First, if we have support for signed keys in the kernel, then try that first. */
2679 r
= sym_crypt_activate_by_signed_key(
2683 verity
->root_hash_size
,
2684 verity
->root_hash_sig
,
2685 verity
->root_hash_sig_size
,
2686 CRYPT_ACTIVATE_READONLY
);
2690 log_debug_errno(r
, "Validation of dm-verity signature failed via the kernel, trying userspace validation instead: %m");
2692 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.",
2693 program_invocation_short_name
);
2694 r
= 0; /* Set for the propagation below */
2697 /* So this didn't work via the kernel, then let's try userspace validation instead. If that
2698 * works we'll try to activate without telling the kernel the signature. */
2700 /* Preferably propagate the original kernel error, so that the fallback logic can work,
2701 * as the device-mapper is finicky around concurrent activations of the same volume */
2702 k
= validate_signature_userspace(verity
, flags
);
2704 return r
< 0 ? r
: k
;
2706 return log_debug_errno(r
< 0 ? r
: SYNTHETIC_ERRNO(ENOKEY
),
2707 "Activation of signed Verity volume worked neither via the kernel nor in userspace, can't activate.");
2710 return sym_crypt_activate_by_volume_key(
2714 verity
->root_hash_size
,
2715 CRYPT_ACTIVATE_READONLY
);
2718 static usec_t
verity_timeout(void) {
2719 usec_t t
= 100 * USEC_PER_MSEC
;
2723 /* On slower machines, like non-KVM vm, setting up device may take a long time.
2724 * Let's make the timeout configurable. */
2726 e
= getenv("SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC");
2730 r
= parse_sec(e
, &t
);
2733 "Failed to parse timeout specified in $SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC, "
2734 "using the default timeout (%s).",
2735 FORMAT_TIMESPAN(t
, USEC_PER_MSEC
));
2740 static int verity_partition(
2741 PartitionDesignator designator
,
2742 DissectedPartition
*m
,
2743 DissectedPartition
*v
,
2744 const VeritySettings
*verity
,
2745 DissectImageFlags flags
,
2746 DecryptedImage
*d
) {
2748 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2749 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
2750 _cleanup_close_
int mount_node_fd
= -EBADF
;
2754 assert(v
|| (verity
&& verity
->data_path
));
2756 if (!verity
|| !verity
->root_hash
)
2758 if (!((verity
->designator
< 0 && designator
== PARTITION_ROOT
) ||
2759 (verity
->designator
== designator
)))
2762 if (!m
->found
|| !m
->node
|| !m
->fstype
)
2764 if (!verity
->data_path
) {
2765 if (!v
->found
|| !v
->node
|| !v
->fstype
)
2768 if (!streq(v
->fstype
, "DM_verity_hash"))
2772 r
= dlopen_cryptsetup();
2776 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
2777 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
2778 _cleanup_free_
char *root_hash_encoded
= NULL
;
2780 root_hash_encoded
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2781 if (!root_hash_encoded
)
2784 r
= make_dm_name_and_node(root_hash_encoded
, "-verity", &name
, &node
);
2786 r
= make_dm_name_and_node(m
->node
, "-verity", &name
, &node
);
2790 r
= sym_crypt_init(&cd
, verity
->data_path
?: v
->node
);
2794 cryptsetup_enable_logging(cd
);
2796 r
= sym_crypt_load(cd
, CRYPT_VERITY
, NULL
);
2800 r
= sym_crypt_set_data_device(cd
, m
->node
);
2804 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
2807 /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
2808 * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
2809 * retry a few times before giving up. */
2810 for (unsigned i
= 0; i
< N_DEVICE_NODE_LIST_ATTEMPTS
; i
++) {
2811 _cleanup_(dm_deferred_remove_cleanp
) char *restore_deferred_remove
= NULL
;
2812 _cleanup_(sym_crypt_freep
) struct crypt_device
*existing_cd
= NULL
;
2813 _cleanup_close_
int fd
= -EBADF
;
2815 /* First, check if the device already exists. */
2816 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2817 if (fd
< 0 && !ERRNO_IS_DEVICE_ABSENT(errno
))
2818 return log_debug_errno(errno
, "Failed to open verity device %s: %m", node
);
2820 goto check
; /* The device already exists. Let's check it. */
2822 /* The symlink to the device node does not exist yet. Assume not activated, and let's activate it. */
2823 r
= do_crypt_activate_verity(cd
, name
, verity
, flags
);
2825 goto try_open
; /* The device is activated. Let's open it. */
2826 /* libdevmapper can return EINVAL when the device is already in the activation stage.
2827 * There's no way to distinguish this situation from a genuine error due to invalid
2828 * parameters, so immediately fall back to activating the device with a unique name.
2829 * Improvements in libcrypsetup can ensure this never happens:
2830 * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
2831 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2833 /* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again if
2834 * sharing is enabled. */
2835 if (r
== -ENODEV
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2838 -EEXIST
, /* Volume has already been opened and ready to be used. */
2839 -EBUSY
/* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */))
2840 return log_debug_errno(r
, "Failed to activate verity device %s: %m", node
);
2843 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
2844 r
= dm_deferred_remove_cancel(name
);
2845 /* -EBUSY and -ENXIO: the device has already been removed or being removed. We cannot
2846 * use the device, try to open again. See target_message() in drivers/md/dm-ioctl.c
2847 * and dm_cancel_deferred_remove() in drivers/md/dm.c */
2848 if (IN_SET(r
, -EBUSY
, -ENXIO
))
2851 return log_debug_errno(r
, "Failed to disable automated deferred removal for verity device %s: %m", node
);
2853 restore_deferred_remove
= strdup(name
);
2854 if (!restore_deferred_remove
)
2855 return log_oom_debug();
2857 r
= verity_can_reuse(verity
, name
, &existing_cd
);
2858 /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
2859 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2862 -ENOENT
, /* Removed?? */
2863 -EBUSY
, /* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */
2864 -ENODEV
/* Volume is being opened but not ready, crypt_init_by_name() would fail, try to open again. */ ))
2867 return log_debug_errno(r
, "Failed to check if existing verity device %s can be reused: %m", node
);
2870 /* devmapper might say that the device exists, but the devlink might not yet have been
2871 * created. Check and wait for the udev event in that case. */
2872 r
= device_wait_for_devlink(node
, "block", verity_timeout(), NULL
);
2873 /* Fallback to activation with a unique device if it's taking too long */
2874 if (r
== -ETIMEDOUT
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2877 return log_debug_errno(r
, "Failed to wait device node symlink %s: %m", node
);
2882 /* Now, the device is activated and devlink is created. Let's open it. */
2883 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2885 if (!ERRNO_IS_DEVICE_ABSENT(errno
))
2886 return log_debug_errno(errno
, "Failed to open verity device %s: %m", node
);
2888 /* The device has already been removed?? */
2893 /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
2894 restore_deferred_remove
= mfree(restore_deferred_remove
);
2896 mount_node_fd
= TAKE_FD(fd
);
2898 crypt_free_and_replace(cd
, existing_cd
);
2903 /* Device is being removed by another process. Let's wait for a while. */
2904 (void) usleep_safe(2 * USEC_PER_MSEC
);
2907 /* All trials failed or a conflicting verity device exists. Let's try to activate with a unique name. */
2908 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
2909 /* Before trying to activate with unique name, we need to free crypt_device object.
2910 * Otherwise, we get error from libcryptsetup like the following:
2912 * systemd[1234]: Cannot use device /dev/loop5 which is in use (already mapped or mounted).
2917 return verity_partition(designator
, m
, v
, verity
, flags
& ~DISSECT_IMAGE_VERITY_SHARE
, d
);
2920 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
), "All attempts to activate verity device %s failed.", name
);
2923 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
2924 .name
= TAKE_PTR(name
),
2925 .device
= TAKE_PTR(cd
),
2928 m
->decrypted_node
= TAKE_PTR(node
);
2929 close_and_replace(m
->mount_node_fd
, mount_node_fd
);
2935 int dissected_image_decrypt(
2937 const char *passphrase
,
2938 const VeritySettings
*verity
,
2939 DissectImageFlags flags
) {
2941 #if HAVE_LIBCRYPTSETUP
2942 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
2947 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
2951 * = 0 → There was nothing to decrypt
2952 * > 0 → Decrypted successfully
2953 * -ENOKEY → There's something to decrypt but no key was supplied
2954 * -EKEYREJECTED → Passed key was not correct
2955 * -EBUSY → Generic Verity error (kernel is not very explanatory)
2958 if (verity
&& verity
->root_hash
&& verity
->root_hash_size
< sizeof(sd_id128_t
))
2961 if (!m
->encrypted
&& !m
->verity_ready
)
2964 #if HAVE_LIBCRYPTSETUP
2965 r
= decrypted_image_new(&d
);
2969 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
2970 DissectedPartition
*p
= m
->partitions
+ i
;
2971 PartitionDesignator k
;
2976 r
= decrypt_partition(p
, passphrase
, flags
, d
);
2980 k
= partition_verity_of(i
);
2982 flags
|= getenv_bool("SYSTEMD_VERITY_SHARING") != 0 ? DISSECT_IMAGE_VERITY_SHARE
: 0;
2984 r
= verity_partition(i
, p
, m
->partitions
+ k
, verity
, flags
, d
);
2989 if (!p
->decrypted_fstype
&& p
->mount_node_fd
>= 0 && p
->decrypted_node
) {
2990 r
= probe_filesystem_full(p
->mount_node_fd
, p
->decrypted_node
, 0, UINT64_MAX
, &p
->decrypted_fstype
);
2991 if (r
< 0 && r
!= -EUCLEAN
)
2996 m
->decrypted_image
= TAKE_PTR(d
);
3004 int dissected_image_decrypt_interactively(
3006 const char *passphrase
,
3007 const VeritySettings
*verity
,
3008 DissectImageFlags flags
) {
3010 _cleanup_strv_free_erase_
char **z
= NULL
;
3017 r
= dissected_image_decrypt(m
, passphrase
, verity
, flags
);
3020 if (r
== -EKEYREJECTED
)
3021 log_error_errno(r
, "Incorrect passphrase, try again!");
3022 else if (r
!= -ENOKEY
)
3023 return log_error_errno(r
, "Failed to decrypt image: %m");
3026 return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED
),
3027 "Too many retries.");
3029 z
= strv_free_erase(z
);
3031 static const AskPasswordRequest req
= {
3032 .message
= "Please enter image passphrase:",
3034 .keyring
= "dissect",
3035 .credential
= "dissect.passphrase",
3038 r
= ask_password_auto(&req
, USEC_INFINITY
, /* flags= */ 0, &z
);
3040 return log_error_errno(r
, "Failed to query for passphrase: %m");
3046 static int decrypted_image_relinquish(DecryptedImage
*d
) {
3049 /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a
3050 * boolean so that we don't clean it up ourselves either anymore */
3052 #if HAVE_LIBCRYPTSETUP
3055 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
3056 DecryptedPartition
*p
= d
->decrypted
+ i
;
3058 if (p
->relinquished
)
3061 r
= sym_crypt_deactivate_by_name(NULL
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
3063 return log_debug_errno(r
, "Failed to mark %s for auto-removal: %m", p
->name
);
3065 p
->relinquished
= true;
3072 int dissected_image_relinquish(DissectedImage
*m
) {
3077 if (m
->decrypted_image
) {
3078 r
= decrypted_image_relinquish(m
->decrypted_image
);
3084 loop_device_relinquish(m
->loop
);
3089 static char *build_auxiliary_path(const char *image
, const char *suffix
) {
3096 e
= endswith(image
, ".raw");
3098 return strjoin(e
, suffix
);
3100 n
= new(char, e
- image
+ strlen(suffix
) + 1);
3104 strcpy(mempcpy(n
, image
, e
- image
), suffix
);
3108 void verity_settings_done(VeritySettings
*v
) {
3111 v
->root_hash
= mfree(v
->root_hash
);
3112 v
->root_hash_size
= 0;
3114 v
->root_hash_sig
= mfree(v
->root_hash_sig
);
3115 v
->root_hash_sig_size
= 0;
3117 v
->data_path
= mfree(v
->data_path
);
3120 int verity_settings_load(
3121 VeritySettings
*verity
,
3123 const char *root_hash_path
,
3124 const char *root_hash_sig_path
) {
3126 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
3127 size_t root_hash_size
= 0, root_hash_sig_size
= 0;
3128 _cleanup_free_
char *verity_data_path
= NULL
;
3129 PartitionDesignator designator
;
3134 assert(verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
3136 /* If we are asked to load the root hash for a device node, exit early */
3137 if (is_device_path(image
))
3140 r
= secure_getenv_bool("SYSTEMD_DISSECT_VERITY_SIDECAR");
3141 if (r
< 0 && r
!= -ENXIO
)
3142 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIDECAR, ignoring: %m");
3146 designator
= verity
->designator
;
3148 /* We only fill in what isn't already filled in */
3150 if (!verity
->root_hash
) {
3151 _cleanup_free_
char *text
= NULL
;
3153 if (root_hash_path
) {
3154 /* If explicitly specified it takes precedence */
3155 r
= read_one_line_file(root_hash_path
, &text
);
3160 designator
= PARTITION_ROOT
;
3162 /* Otherwise look for xattr and separate file, and first for the data for root and if
3163 * that doesn't exist for /usr */
3165 if (designator
< 0 || designator
== PARTITION_ROOT
) {
3166 r
= getxattr_malloc(image
, "user.verity.roothash", &text
);
3168 _cleanup_free_
char *p
= NULL
;
3170 if (r
!= -ENOENT
&& !ERRNO_IS_XATTR_ABSENT(r
))
3173 p
= build_auxiliary_path(image
, ".roothash");
3177 r
= read_one_line_file(p
, &text
);
3178 if (r
< 0 && r
!= -ENOENT
)
3183 designator
= PARTITION_ROOT
;
3186 if (!text
&& (designator
< 0 || designator
== PARTITION_USR
)) {
3187 /* So in the "roothash" xattr/file name above the "root" of course primarily
3188 * refers to the root of the Verity Merkle tree. But coincidentally it also
3189 * is the hash for the *root* file system, i.e. the "root" neatly refers to
3190 * two distinct concepts called "root". Taking benefit of this happy
3191 * coincidence we call the file with the root hash for the /usr/ file system
3192 * `usrhash`, because `usrroothash` or `rootusrhash` would just be too
3193 * confusing. We thus drop the reference to the root of the Merkle tree, and
3194 * just indicate which file system it's about. */
3195 r
= getxattr_malloc(image
, "user.verity.usrhash", &text
);
3197 _cleanup_free_
char *p
= NULL
;
3199 if (r
!= -ENOENT
&& !ERRNO_IS_XATTR_ABSENT(r
))
3202 p
= build_auxiliary_path(image
, ".usrhash");
3206 r
= read_one_line_file(p
, &text
);
3207 if (r
< 0 && r
!= -ENOENT
)
3212 designator
= PARTITION_USR
;
3217 r
= unhexmem(text
, &root_hash
, &root_hash_size
);
3220 if (root_hash_size
< sizeof(sd_id128_t
))
3225 if ((root_hash
|| verity
->root_hash
) && !verity
->root_hash_sig
) {
3226 if (root_hash_sig_path
) {
3227 r
= read_full_file(root_hash_sig_path
, (char**) &root_hash_sig
, &root_hash_sig_size
);
3228 if (r
< 0 && r
!= -ENOENT
)
3232 designator
= PARTITION_ROOT
;
3234 if (designator
< 0 || designator
== PARTITION_ROOT
) {
3235 _cleanup_free_
char *p
= NULL
;
3237 /* Follow naming convention recommended by the relevant RFC:
3238 * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
3239 p
= build_auxiliary_path(image
, ".roothash.p7s");
3243 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
3244 if (r
< 0 && r
!= -ENOENT
)
3247 designator
= PARTITION_ROOT
;
3250 if (!root_hash_sig
&& (designator
< 0 || designator
== PARTITION_USR
)) {
3251 _cleanup_free_
char *p
= NULL
;
3253 p
= build_auxiliary_path(image
, ".usrhash.p7s");
3257 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
3258 if (r
< 0 && r
!= -ENOENT
)
3261 designator
= PARTITION_USR
;
3265 if (root_hash_sig
&& root_hash_sig_size
== 0) /* refuse empty size signatures */
3269 if (!verity
->data_path
) {
3270 _cleanup_free_
char *p
= NULL
;
3272 p
= build_auxiliary_path(image
, ".verity");
3276 if (access(p
, F_OK
) < 0) {
3277 if (errno
!= ENOENT
)
3280 verity_data_path
= TAKE_PTR(p
);
3284 verity
->root_hash
= TAKE_PTR(root_hash
);
3285 verity
->root_hash_size
= root_hash_size
;
3288 if (root_hash_sig
) {
3289 verity
->root_hash_sig
= TAKE_PTR(root_hash_sig
);
3290 verity
->root_hash_sig_size
= root_hash_sig_size
;
3293 if (verity_data_path
)
3294 verity
->data_path
= TAKE_PTR(verity_data_path
);
3296 if (verity
->designator
< 0)
3297 verity
->designator
= designator
;
3302 int dissected_image_load_verity_sig_partition(
3305 VeritySettings
*verity
) {
3307 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
3308 _cleanup_(json_variant_unrefp
) JsonVariant
*v
= NULL
;
3309 size_t root_hash_size
, root_hash_sig_size
;
3310 _cleanup_free_
char *buf
= NULL
;
3311 PartitionDesignator d
;
3312 DissectedPartition
*p
;
3313 JsonVariant
*rh
, *sig
;
3322 if (verity
->root_hash
&& verity
->root_hash_sig
) /* Already loaded? */
3325 r
= secure_getenv_bool("SYSTEMD_DISSECT_VERITY_EMBEDDED");
3326 if (r
< 0 && r
!= -ENXIO
)
3327 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_EMBEDDED, ignoring: %m");
3331 d
= partition_verity_sig_of(verity
->designator
< 0 ? PARTITION_ROOT
: verity
->designator
);
3334 p
= m
->partitions
+ d
;
3337 if (p
->offset
== UINT64_MAX
|| p
->size
== UINT64_MAX
)
3340 if (p
->size
> 4*1024*1024) /* Signature data cannot possible be larger than 4M, refuse that */
3341 return log_debug_errno(SYNTHETIC_ERRNO(EFBIG
), "Verity signature partition is larger than 4M, refusing.");
3343 buf
= new(char, p
->size
+1);
3347 n
= pread(fd
, buf
, p
->size
, p
->offset
);
3350 if ((uint64_t) n
!= p
->size
)
3353 e
= memchr(buf
, 0, p
->size
);
3355 /* If we found a NUL byte then the rest of the data must be NUL too */
3356 if (!memeqzero(e
, p
->size
- (e
- buf
)))
3357 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature data contains embedded NUL byte.");
3361 r
= json_parse(buf
, 0, &v
, NULL
, NULL
);
3363 return log_debug_errno(r
, "Failed to parse signature JSON data: %m");
3365 rh
= json_variant_by_key(v
, "rootHash");
3367 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'rootHash' field.");
3368 if (!json_variant_is_string(rh
))
3369 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'rootHash' field of signature JSON object is not a string.");
3371 r
= unhexmem(json_variant_string(rh
), &root_hash
, &root_hash_size
);
3373 return log_debug_errno(r
, "Failed to parse root hash field: %m");
3375 /* Check if specified root hash matches if it is specified */
3376 if (verity
->root_hash
&&
3377 memcmp_nn(verity
->root_hash
, verity
->root_hash_size
, root_hash
, root_hash_size
) != 0) {
3378 _cleanup_free_
char *a
= NULL
, *b
= NULL
;
3380 a
= hexmem(root_hash
, root_hash_size
);
3381 b
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
3383 return log_debug_errno(r
, "Root hash in signature JSON data (%s) doesn't match configured hash (%s).", strna(a
), strna(b
));
3386 sig
= json_variant_by_key(v
, "signature");
3388 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'signature' field.");
3389 if (!json_variant_is_string(sig
))
3390 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'signature' field of signature JSON object is not a string.");
3392 r
= unbase64mem(json_variant_string(sig
), &root_hash_sig
, &root_hash_sig_size
);
3394 return log_debug_errno(r
, "Failed to parse signature field: %m");
3396 free_and_replace(verity
->root_hash
, root_hash
);
3397 verity
->root_hash_size
= root_hash_size
;
3399 free_and_replace(verity
->root_hash_sig
, root_hash_sig
);
3400 verity
->root_hash_sig_size
= root_hash_sig_size
;
3405 int dissected_image_acquire_metadata(
3408 DissectImageFlags extra_flags
) {
3415 META_INITRD_RELEASE
,
3416 META_SYSEXT_RELEASE
,
3417 META_CONFEXT_RELEASE
,
3418 META_HAS_INIT_SYSTEM
,
3422 static const char *const paths
[_META_MAX
] = {
3423 [META_HOSTNAME
] = "/etc/hostname\0",
3424 [META_MACHINE_ID
] = "/etc/machine-id\0",
3425 [META_MACHINE_INFO
] = "/etc/machine-info\0",
3426 [META_OS_RELEASE
] = "/etc/os-release\0"
3427 "/usr/lib/os-release\0",
3428 [META_INITRD_RELEASE
] = "/etc/initrd-release\0"
3429 "/usr/lib/initrd-release\0",
3430 [META_SYSEXT_RELEASE
] = "sysext-release\0", /* String used only for logging. */
3431 [META_CONFEXT_RELEASE
] = "confext-release\0", /* ditto */
3432 [META_HAS_INIT_SYSTEM
] = "has-init-system\0", /* ditto */
3435 _cleanup_strv_free_
char **machine_info
= NULL
, **os_release
= NULL
, **initrd_release
= NULL
, **sysext_release
= NULL
, **confext_release
= NULL
;
3436 _cleanup_free_
char *hostname
= NULL
, *t
= NULL
;
3437 _cleanup_close_pair_
int error_pipe
[2] = EBADF_PAIR
;
3438 _cleanup_(sigkill_waitp
) pid_t child
= 0;
3439 sd_id128_t machine_id
= SD_ID128_NULL
;
3440 unsigned n_meta_initialized
= 0;
3441 int fds
[2 * _META_MAX
], r
, v
;
3442 int has_init_system
= -1;
3445 BLOCK_SIGNALS(SIGCHLD
);
3449 for (; n_meta_initialized
< _META_MAX
; n_meta_initialized
++) {
3450 assert(paths
[n_meta_initialized
]);
3452 if (pipe2(fds
+ 2*n_meta_initialized
, O_CLOEXEC
) < 0) {
3458 r
= get_common_dissect_directory(&t
);
3462 if (pipe2(error_pipe
, O_CLOEXEC
) < 0) {
3467 r
= safe_fork("(sd-dissect)", FORK_RESET_SIGNALS
|FORK_DEATHSIG_SIGTERM
, &child
);
3472 error_pipe
[0] = safe_close(error_pipe
[0]);
3475 r
= detach_mount_namespace_harder(0, 0);
3477 r
= detach_mount_namespace_userns(userns_fd
);
3479 log_debug_errno(r
, "Failed to detach mount namespace: %m");
3483 r
= dissected_image_mount(
3486 /* uid_shift= */ UID_INVALID
,
3487 /* uid_range= */ UID_INVALID
,
3488 /* userns_fd= */ -EBADF
,
3490 DISSECT_IMAGE_READ_ONLY
|
3491 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
3492 DISSECT_IMAGE_USR_NO_ROOT
);
3494 log_debug_errno(r
, "Failed to mount dissected image: %m");
3498 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
3499 _cleanup_close_
int fd
= -ENOENT
;
3503 fds
[2*k
] = safe_close(fds
[2*k
]);
3507 case META_SYSEXT_RELEASE
:
3511 /* As per the os-release spec, if the image is an extension it will have a
3512 * file named after the image name in extension-release.d/ - we use the image
3513 * name and try to resolve it with the extension-release helpers, as
3514 * sometimes the image names are mangled on deployment and do not match
3515 * anymore. Unlike other paths this is not fixed, and the image name can be
3516 * mangled on deployment, so by calling into the helper we allow a fallback
3517 * that matches on the first extension-release file found in the directory,
3518 * if one named after the image cannot be found first. */
3519 r
= open_extension_release(
3523 /* relax_extension_release_check= */ false,
3524 /* ret_path= */ NULL
,
3530 case META_CONFEXT_RELEASE
:
3535 r
= open_extension_release(
3539 /* relax_extension_release_check= */ false,
3540 /* ret_path= */ NULL
,
3547 case META_HAS_INIT_SYSTEM
: {
3550 FOREACH_STRING(init
,
3551 "/usr/lib/systemd/systemd", /* systemd on /usr/ merged system */
3552 "/lib/systemd/systemd", /* systemd on /usr/ non-merged systems */
3553 "/sbin/init") { /* traditional path the Linux kernel invokes */
3555 r
= chase(init
, t
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
3558 log_debug_errno(r
, "Failed to resolve %s, ignoring: %m", init
);
3565 r
= loop_write(fds
[2*k
+1], &found
, sizeof(found
));
3573 NULSTR_FOREACH(p
, paths
[k
]) {
3574 fd
= chase_and_open(p
, t
, CHASE_PREFIX_ROOT
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
, NULL
);
3581 log_debug_errno(fd
, "Failed to read %s file of image, ignoring: %m", paths
[k
]);
3585 r
= copy_bytes(fd
, fds
[2*k
+1], UINT64_MAX
, 0);
3590 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3593 _exit(EXIT_SUCCESS
);
3596 /* Let parent know the error */
3597 (void) write(error_pipe
[1], &r
, sizeof(r
));
3598 _exit(EXIT_FAILURE
);
3601 error_pipe
[1] = safe_close(error_pipe
[1]);
3603 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
3604 _cleanup_fclose_
FILE *f
= NULL
;
3608 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3610 f
= take_fdopen(&fds
[2*k
], "r");
3619 r
= read_etc_hostname_stream(f
, &hostname
);
3621 log_debug_errno(r
, "Failed to read /etc/hostname of image: %m");
3625 case META_MACHINE_ID
: {
3626 _cleanup_free_
char *line
= NULL
;
3628 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3630 log_debug_errno(r
, "Failed to read /etc/machine-id of image: %m");
3632 r
= sd_id128_from_string(line
, &machine_id
);
3634 log_debug_errno(r
, "Image contains invalid /etc/machine-id: %s", line
);
3636 log_debug("/etc/machine-id file of image is empty.");
3637 else if (streq(line
, "uninitialized"))
3638 log_debug("/etc/machine-id file of image is uninitialized (likely aborted first boot).");
3640 log_debug("/etc/machine-id file of image has unexpected length %i.", r
);
3645 case META_MACHINE_INFO
:
3646 r
= load_env_file_pairs(f
, "machine-info", &machine_info
);
3648 log_debug_errno(r
, "Failed to read /etc/machine-info of image: %m");
3652 case META_OS_RELEASE
:
3653 r
= load_env_file_pairs(f
, "os-release", &os_release
);
3655 log_debug_errno(r
, "Failed to read OS release file of image: %m");
3659 case META_INITRD_RELEASE
:
3660 r
= load_env_file_pairs(f
, "initrd-release", &initrd_release
);
3662 log_debug_errno(r
, "Failed to read initrd release file of image: %m");
3666 case META_SYSEXT_RELEASE
:
3667 r
= load_env_file_pairs(f
, "sysext-release", &sysext_release
);
3669 log_debug_errno(r
, "Failed to read sysext release file of image: %m");
3673 case META_CONFEXT_RELEASE
:
3674 r
= load_env_file_pairs(f
, "confext-release", &confext_release
);
3676 log_debug_errno(r
, "Failed to read confext release file of image: %m");
3680 case META_HAS_INIT_SYSTEM
: {
3685 nr
= fread(&b
, 1, sizeof(b
), f
);
3686 if (nr
!= sizeof(b
))
3687 log_debug_errno(errno_or_else(EIO
), "Failed to read has-init-system boolean: %m");
3689 has_init_system
= b
;
3695 r
= wait_for_terminate_and_check("(sd-dissect)", child
, 0);
3700 n
= read(error_pipe
[0], &v
, sizeof(v
));
3705 if (n
== sizeof(v
)) {
3706 r
= v
; /* propagate error sent to us from child */
3713 if (r
!= EXIT_SUCCESS
) {
3718 free_and_replace(m
->hostname
, hostname
);
3719 m
->machine_id
= machine_id
;
3720 strv_free_and_replace(m
->machine_info
, machine_info
);
3721 strv_free_and_replace(m
->os_release
, os_release
);
3722 strv_free_and_replace(m
->initrd_release
, initrd_release
);
3723 strv_free_and_replace(m
->sysext_release
, sysext_release
);
3724 strv_free_and_replace(m
->confext_release
, confext_release
);
3725 m
->has_init_system
= has_init_system
;
3728 for (unsigned k
= 0; k
< n_meta_initialized
; k
++)
3729 safe_close_pair(fds
+ 2*k
);
3734 Architecture
dissected_image_architecture(DissectedImage
*img
) {
3737 if (img
->partitions
[PARTITION_ROOT
].found
&&
3738 img
->partitions
[PARTITION_ROOT
].architecture
>= 0)
3739 return img
->partitions
[PARTITION_ROOT
].architecture
;
3741 if (img
->partitions
[PARTITION_USR
].found
&&
3742 img
->partitions
[PARTITION_USR
].architecture
>= 0)
3743 return img
->partitions
[PARTITION_USR
].architecture
;
3745 return _ARCHITECTURE_INVALID
;
3748 int dissect_loop_device(
3750 const VeritySettings
*verity
,
3751 const MountOptions
*mount_options
,
3752 const ImagePolicy
*image_policy
,
3753 DissectImageFlags flags
,
3754 DissectedImage
**ret
) {
3757 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
3762 r
= dissected_image_new(loop
->backing_file
?: loop
->node
, &m
);
3766 m
->loop
= loop_device_ref(loop
);
3767 m
->image_size
= m
->loop
->device_size
;
3768 m
->sector_size
= m
->loop
->sector_size
;
3770 r
= dissect_image(m
, loop
->fd
, loop
->node
, verity
, mount_options
, image_policy
, flags
);
3783 int dissect_loop_device_and_warn(
3785 const VeritySettings
*verity
,
3786 const MountOptions
*mount_options
,
3787 const ImagePolicy
*image_policy
,
3788 DissectImageFlags flags
,
3789 DissectedImage
**ret
) {
3793 return dissect_log_error(
3795 dissect_loop_device(loop
, verity
, mount_options
, image_policy
, flags
, ret
),
3796 loop
->backing_file
?: loop
->node
,
3801 bool dissected_image_verity_candidate(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3804 /* Checks if this partition could theoretically do Verity. For non-partitioned images this only works
3805 * if there's an external verity file supplied, for which we can consult .has_verity. For partitioned
3806 * images we only check the partition type.
3808 * This call is used to decide whether to suppress or show a verity column in tabular output of the
3811 if (image
->single_file_system
)
3812 return partition_designator
== PARTITION_ROOT
&& image
->has_verity
;
3814 return partition_verity_of(partition_designator
) >= 0;
3817 bool dissected_image_verity_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3818 PartitionDesignator k
;
3822 /* Checks if this partition has verity data available that we can activate. For non-partitioned this
3823 * works for the root partition, for others only if the associated verity partition was found. */
3825 if (!image
->verity_ready
)
3828 if (image
->single_file_system
)
3829 return partition_designator
== PARTITION_ROOT
;
3831 k
= partition_verity_of(partition_designator
);
3832 return k
>= 0 && image
->partitions
[k
].found
;
3835 bool dissected_image_verity_sig_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3836 PartitionDesignator k
;
3840 /* Checks if this partition has verity signature data available that we can use. */
3842 if (!image
->verity_sig_ready
)
3845 if (image
->single_file_system
)
3846 return partition_designator
== PARTITION_ROOT
;
3848 k
= partition_verity_sig_of(partition_designator
);
3849 return k
>= 0 && image
->partitions
[k
].found
;
3852 MountOptions
* mount_options_free_all(MountOptions
*options
) {
3855 while ((m
= LIST_POP(mount_options
, options
))) {
3863 const char* mount_options_from_designator(const MountOptions
*options
, PartitionDesignator designator
) {
3864 LIST_FOREACH(mount_options
, m
, options
)
3865 if (designator
== m
->partition_designator
&& !isempty(m
->options
))
3871 int mount_image_privately_interactively(
3873 const ImagePolicy
*image_policy
,
3874 DissectImageFlags flags
,
3875 char **ret_directory
,
3877 LoopDevice
**ret_loop_device
) {
3879 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3880 _cleanup_(loop_device_unrefp
) LoopDevice
*d
= NULL
;
3881 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3882 _cleanup_free_
char *dir
= NULL
;
3885 /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
3886 * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
3890 assert(ret_loop_device
);
3892 /* We intend to mount this right-away, hence add the partitions if needed and pin them. */
3893 flags
|= DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
3894 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
3896 r
= verity_settings_load(&verity
, image
, NULL
, NULL
);
3898 return log_error_errno(r
, "Failed to load root hash data: %m");
3900 r
= loop_device_make_by_path(
3902 FLAGS_SET(flags
, DISSECT_IMAGE_DEVICE_READ_ONLY
) ? O_RDONLY
: O_RDWR
,
3903 /* sector_size= */ UINT32_MAX
,
3904 FLAGS_SET(flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
3908 return log_error_errno(r
, "Failed to set up loopback device for %s: %m", image
);
3910 r
= dissect_loop_device_and_warn(
3913 /* mount_options= */ NULL
,
3920 r
= dissected_image_load_verity_sig_partition(dissected_image
, d
->fd
, &verity
);
3924 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, &verity
, flags
);
3928 r
= detach_mount_namespace();
3930 return log_error_errno(r
, "Failed to detach mount namespace: %m");
3932 r
= mkdir_p("/run/systemd/mount-rootfs", 0555);
3934 return log_error_errno(r
, "Failed to create mount point: %m");
3936 r
= dissected_image_mount_and_warn(
3938 "/run/systemd/mount-rootfs",
3939 /* uid_shift= */ UID_INVALID
,
3940 /* uid_range= */ UID_INVALID
,
3941 /* userns_fd= */ -EBADF
,
3946 r
= loop_device_flock(d
, LOCK_UN
);
3950 r
= dissected_image_relinquish(dissected_image
);
3952 return log_error_errno(r
, "Failed to relinquish DM and loopback block devices: %m");
3954 if (ret_directory
) {
3955 dir
= strdup("/run/systemd/mount-rootfs");
3961 _cleanup_close_
int dir_fd
= -EBADF
;
3963 dir_fd
= open("/run/systemd/mount-rootfs", O_CLOEXEC
|O_DIRECTORY
);
3965 return log_error_errno(errno
, "Failed to open mount point directory: %m");
3967 *ret_dir_fd
= TAKE_FD(dir_fd
);
3971 *ret_directory
= TAKE_PTR(dir
);
3973 *ret_loop_device
= TAKE_PTR(d
);
3977 static bool mount_options_relax_extension_release_checks(const MountOptions
*options
) {
3981 return string_contains_word(mount_options_from_designator(options
, PARTITION_ROOT
), ",", "x-systemd.relax-extension-release-check") ||
3982 string_contains_word(mount_options_from_designator(options
, PARTITION_USR
), ",", "x-systemd.relax-extension-release-check") ||
3983 string_contains_word(options
->options
, ",", "x-systemd.relax-extension-release-check");
3986 int verity_dissect_and_mount(
3990 const MountOptions
*options
,
3991 const ImagePolicy
*image_policy
,
3992 const char *required_host_os_release_id
,
3993 const char *required_host_os_release_version_id
,
3994 const char *required_host_os_release_sysext_level
,
3995 const char *required_host_os_release_confext_level
,
3996 const char *required_sysext_scope
,
3997 DissectedImage
**ret_image
) {
3999 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
4000 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
4001 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
4002 DissectImageFlags dissect_image_flags
;
4003 bool relax_extension_release_check
;
4007 /* Verifying release metadata requires mounted image for now, so ensure the check is skipped when
4008 * opening an image without mounting it immediately (i.e.: 'dest' is NULL). */
4009 assert(!required_host_os_release_id
|| dest
);
4011 relax_extension_release_check
= mount_options_relax_extension_release_checks(options
);
4013 /* We might get an FD for the image, but we use the original path to look for the dm-verity files */
4014 r
= verity_settings_load(&verity
, src
, NULL
, NULL
);
4016 return log_debug_errno(r
, "Failed to load root hash: %m");
4018 dissect_image_flags
=
4019 (verity
.data_path
? DISSECT_IMAGE_NO_PARTITION_TABLE
: 0) |
4020 (relax_extension_release_check
? DISSECT_IMAGE_RELAX_EXTENSION_CHECK
: 0) |
4021 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
4022 DISSECT_IMAGE_PIN_PARTITION_DEVICES
|
4023 DISSECT_IMAGE_ALLOW_USERSPACE_VERITY
;
4025 /* Note that we don't use loop_device_make here, as the FD is most likely O_PATH which would not be
4026 * accepted by LOOP_CONFIGURE, so just let loop_device_make_by_path reopen it as a regular FD. */
4027 r
= loop_device_make_by_path(
4028 src_fd
>= 0 ? FORMAT_PROC_FD_PATH(src_fd
) : src
,
4029 /* open_flags= */ -1,
4030 /* sector_size= */ UINT32_MAX
,
4031 verity
.data_path
? 0 : LO_FLAGS_PARTSCAN
,
4035 return log_debug_errno(r
, "Failed to create loop device for image: %m");
4037 r
= dissect_loop_device(
4042 dissect_image_flags
,
4044 /* No partition table? Might be a single-filesystem image, try again */
4045 if (!verity
.data_path
&& r
== -ENOPKG
)
4046 r
= dissect_loop_device(
4051 dissect_image_flags
| DISSECT_IMAGE_NO_PARTITION_TABLE
,
4054 return log_debug_errno(r
, "Failed to dissect image: %m");
4056 r
= dissected_image_load_verity_sig_partition(dissected_image
, loop_device
->fd
, &verity
);
4060 r
= dissected_image_decrypt(
4064 dissect_image_flags
);
4066 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
4069 r
= mkdir_p_label(dest
, 0755);
4071 return log_debug_errno(r
, "Failed to create destination directory %s: %m", dest
);
4072 r
= umount_recursive(dest
, 0);
4074 return log_debug_errno(r
, "Failed to umount under destination directory %s: %m", dest
);
4077 r
= dissected_image_mount(
4080 /* uid_shift= */ UID_INVALID
,
4081 /* uid_range= */ UID_INVALID
,
4082 /* userns_fd= */ -EBADF
,
4083 dissect_image_flags
);
4085 return log_debug_errno(r
, "Failed to mount image: %m");
4087 r
= loop_device_flock(loop_device
, LOCK_UN
);
4089 return log_debug_errno(r
, "Failed to unlock loopback device: %m");
4091 /* If we got os-release values from the caller, then we need to match them with the image's
4092 * extension-release.d/ content. Return -EINVAL if there's any mismatch.
4093 * First, check the distro ID. If that matches, then check the new SYSEXT_LEVEL value if
4094 * available, or else fallback to VERSION_ID. If neither is present (eg: rolling release),
4095 * then a simple match on the ID will be performed. */
4096 if (required_host_os_release_id
) {
4097 _cleanup_strv_free_
char **extension_release
= NULL
;
4098 ImageClass
class = IMAGE_SYSEXT
;
4100 assert(!isempty(required_host_os_release_id
));
4102 r
= load_extension_release_pairs(dest
, IMAGE_SYSEXT
, dissected_image
->image_name
, relax_extension_release_check
, &extension_release
);
4104 r
= load_extension_release_pairs(dest
, IMAGE_CONFEXT
, dissected_image
->image_name
, relax_extension_release_check
, &extension_release
);
4106 class = IMAGE_CONFEXT
;
4109 return log_debug_errno(r
, "Failed to parse image %s extension-release metadata: %m", dissected_image
->image_name
);
4111 r
= extension_release_validate(
4112 dissected_image
->image_name
,
4113 required_host_os_release_id
,
4114 required_host_os_release_version_id
,
4115 class == IMAGE_SYSEXT
? required_host_os_release_sysext_level
: required_host_os_release_confext_level
,
4116 required_sysext_scope
,
4120 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE
), "Image %s extension-release metadata does not match the root's", dissected_image
->image_name
);
4122 return log_debug_errno(r
, "Failed to compare image %s extension-release metadata with the root's os-release: %m", dissected_image
->image_name
);
4125 r
= dissected_image_relinquish(dissected_image
);
4127 return log_debug_errno(r
, "Failed to relinquish dissected image: %m");
4130 *ret_image
= TAKE_PTR(dissected_image
);
4135 int get_common_dissect_directory(char **ret
) {
4136 _cleanup_free_
char *t
= NULL
;
4139 /* A common location we mount dissected images to. The assumption is that everyone who uses this
4140 * function runs in their own private mount namespace (with mount propagation off on /run/systemd/,
4141 * and thus can mount something here without affecting anyone else). */
4143 t
= strdup("/run/systemd/dissect-root");
4145 return log_oom_debug();
4147 r
= mkdir_parents(t
, 0755);
4149 return log_debug_errno(r
, "Failed to create parent dirs of mount point '%s': %m", t
);
4151 r
= RET_NERRNO(mkdir(t
, 0000)); /* It's supposed to be overmounted, hence let's make this inaccessible */
4152 if (r
< 0 && r
!= -EEXIST
)
4153 return log_debug_errno(r
, "Failed to create mount point '%s': %m", t
);
4163 static JSON_DISPATCH_ENUM_DEFINE(dispatch_architecture
, Architecture
, architecture_from_string
);
4164 static JSON_DISPATCH_ENUM_DEFINE(dispatch_partition_designator
, PartitionDesignator
, partition_designator_from_string
);
4166 typedef struct PartitionFields
{
4167 PartitionDesignator designator
;
4171 Architecture architecture
;
4177 unsigned fsmount_fd_idx
;
4180 static void partition_fields_done(PartitionFields
*f
) {
4183 f
->fstype
= mfree(f
->fstype
);
4184 f
->label
= mfree(f
->label
);
4187 typedef struct ReplyParameters
{
4188 JsonVariant
*partitions
;
4190 uint64_t image_size
;
4191 uint32_t sector_size
;
4192 sd_id128_t image_uuid
;
4195 static void reply_parameters_done(ReplyParameters
*p
) {
4198 p
->image_policy
= mfree(p
->image_policy
);
4199 p
->partitions
= json_variant_unref(p
->partitions
);
4204 int mountfsd_mount_image(
4207 const ImagePolicy
*image_policy
,
4208 DissectImageFlags flags
,
4209 DissectedImage
**ret
) {
4212 _cleanup_(reply_parameters_done
) ReplyParameters p
= {};
4214 static const JsonDispatch dispatch_table
[] = {
4215 { "partitions", JSON_VARIANT_ARRAY
, json_dispatch_variant
, offsetof(struct ReplyParameters
, partitions
), JSON_MANDATORY
},
4216 { "imagePolicy", JSON_VARIANT_STRING
, json_dispatch_string
, offsetof(struct ReplyParameters
, image_policy
), 0 },
4217 { "imageSize", _JSON_VARIANT_TYPE_INVALID
, json_dispatch_uint64
, offsetof(struct ReplyParameters
, image_size
), JSON_MANDATORY
},
4218 { "sectorSize", _JSON_VARIANT_TYPE_INVALID
, json_dispatch_uint32
, offsetof(struct ReplyParameters
, sector_size
), JSON_MANDATORY
},
4219 { "imageUuid", JSON_VARIANT_STRING
, json_dispatch_id128
, offsetof(struct ReplyParameters
, image_uuid
), 0 },
4223 _cleanup_(dissected_image_unrefp
) DissectedImage
*di
= NULL
;
4224 _cleanup_close_
int image_fd
= -EBADF
;
4225 _cleanup_(varlink_unrefp
) Varlink
*vl
= NULL
;
4226 _cleanup_free_
char *ps
= NULL
;
4227 unsigned max_fd
= UINT_MAX
;
4228 const char *error_id
;
4234 r
= varlink_connect_address(&vl
, "/run/systemd/io.systemd.MountFileSystem");
4236 return log_error_errno(r
, "Failed to connect to mountfsd: %m");
4238 r
= varlink_set_allow_fd_passing_input(vl
, true);
4240 return log_error_errno(r
, "Failed to enable varlink fd passing for read: %m");
4242 r
= varlink_set_allow_fd_passing_output(vl
, true);
4244 return log_error_errno(r
, "Failed to enable varlink fd passing for write: %m");
4246 image_fd
= open(path
, O_RDONLY
|O_CLOEXEC
);
4248 return log_error_errno(errno
, "Failed to open '%s': %m", path
);
4250 r
= varlink_push_dup_fd(vl
, image_fd
);
4252 return log_error_errno(r
, "Failed to push image fd into varlink connection: %m");
4254 if (userns_fd
>= 0) {
4255 r
= varlink_push_dup_fd(vl
, userns_fd
);
4257 return log_error_errno(r
, "Failed to push image fd into varlink connection: %m");
4261 r
= image_policy_to_string(image_policy
, /* simplify= */ false, &ps
);
4263 return log_error_errno(r
, "Failed format image policy to string: %m");
4266 JsonVariant
*reply
= NULL
;
4269 "io.systemd.MountFileSystem.MountImage",
4273 JSON_BUILD_PAIR("imageFileDescriptor", JSON_BUILD_UNSIGNED(0)),
4274 JSON_BUILD_PAIR_CONDITION(userns_fd
>= 0, "userNamespaceFileDescriptor", JSON_BUILD_UNSIGNED(1)),
4275 JSON_BUILD_PAIR("readOnly", JSON_BUILD_BOOLEAN(FLAGS_SET(flags
, DISSECT_IMAGE_MOUNT_READ_ONLY
))),
4276 JSON_BUILD_PAIR("growFileSystems", JSON_BUILD_BOOLEAN(FLAGS_SET(flags
, DISSECT_IMAGE_GROWFS
))),
4277 JSON_BUILD_PAIR_CONDITION(ps
, "imagePolicy", JSON_BUILD_STRING(ps
)),
4278 JSON_BUILD_PAIR("allowInteractiveAuthentication", JSON_BUILD_BOOLEAN(FLAGS_SET(flags
, DISSECT_IMAGE_ALLOW_INTERACTIVE_AUTH
)))));
4280 return log_error_errno(r
, "Failed to call MountImage() varlink call: %m");
4281 if (!isempty(error_id
))
4282 return log_error_errno(varlink_error_to_errno(error_id
, reply
), "Failed to call MountImage() varlink call: %s", error_id
);
4284 r
= json_dispatch(reply
, dispatch_table
, JSON_ALLOW_EXTENSIONS
, &p
);
4286 return log_error_errno(r
, "Failed to parse MountImage() reply: %m");
4288 log_debug("Effective image policy: %s", p
.image_policy
);
4291 JSON_VARIANT_ARRAY_FOREACH(i
, p
.partitions
) {
4292 _cleanup_close_
int fsmount_fd
= -EBADF
;
4294 _cleanup_(partition_fields_done
) PartitionFields pp
= {
4295 .designator
= _PARTITION_DESIGNATOR_INVALID
,
4296 .architecture
= _ARCHITECTURE_INVALID
,
4298 .offset
= UINT64_MAX
,
4299 .fsmount_fd_idx
= UINT_MAX
,
4302 static const JsonDispatch partition_dispatch_table
[] = {
4303 { "designator", JSON_VARIANT_STRING
, dispatch_partition_designator
, offsetof(struct PartitionFields
, designator
), JSON_MANDATORY
},
4304 { "writable", JSON_VARIANT_BOOLEAN
, json_dispatch_boolean
, offsetof(struct PartitionFields
, rw
), JSON_MANDATORY
},
4305 { "growFileSystem", JSON_VARIANT_BOOLEAN
, json_dispatch_boolean
, offsetof(struct PartitionFields
, growfs
), JSON_MANDATORY
},
4306 { "partitionNumber", _JSON_VARIANT_TYPE_INVALID
, json_dispatch_uint
, offsetof(struct PartitionFields
, partno
), 0 },
4307 { "architecture", JSON_VARIANT_STRING
, dispatch_architecture
, offsetof(struct PartitionFields
, architecture
), 0 },
4308 { "partitionUuid", JSON_VARIANT_STRING
, json_dispatch_id128
, offsetof(struct PartitionFields
, uuid
), 0 },
4309 { "fileSystemType", JSON_VARIANT_STRING
, json_dispatch_string
, offsetof(struct PartitionFields
, fstype
), JSON_MANDATORY
},
4310 { "partitionLabel", JSON_VARIANT_STRING
, json_dispatch_string
, offsetof(struct PartitionFields
, label
), 0 },
4311 { "size", _JSON_VARIANT_TYPE_INVALID
, json_dispatch_uint64
, offsetof(struct PartitionFields
, size
), JSON_MANDATORY
},
4312 { "offset", _JSON_VARIANT_TYPE_INVALID
, json_dispatch_uint64
, offsetof(struct PartitionFields
, offset
), JSON_MANDATORY
},
4313 { "mountFileDescriptor", _JSON_VARIANT_TYPE_INVALID
, json_dispatch_uint
, offsetof(struct PartitionFields
, fsmount_fd_idx
), JSON_MANDATORY
},
4317 r
= json_dispatch(i
, partition_dispatch_table
, JSON_ALLOW_EXTENSIONS
, &pp
);
4319 return log_error_errno(r
, "Failed to parse partition data: %m");
4321 if (pp
.fsmount_fd_idx
!= UINT_MAX
) {
4322 if (max_fd
== UINT_MAX
|| pp
.fsmount_fd_idx
> max_fd
)
4323 max_fd
= pp
.fsmount_fd_idx
;
4325 fsmount_fd
= varlink_take_fd(vl
, pp
.fsmount_fd_idx
);
4330 assert(pp
.designator
>= 0);
4333 r
= dissected_image_new(path
, &di
);
4335 return log_error_errno(r
, "Failed to allocated new dissected image structure: %m");
4338 if (di
->partitions
[pp
.designator
].found
)
4339 return log_error_errno(SYNTHETIC_ERRNO(EBADMSG
), "Duplicate partition data for '%s'.", partition_designator_to_string(pp
.designator
));
4341 di
->partitions
[pp
.designator
] = (DissectedPartition
) {
4344 .growfs
= pp
.growfs
,
4345 .partno
= pp
.partno
,
4346 .architecture
= pp
.architecture
,
4348 .fstype
= TAKE_PTR(pp
.fstype
),
4349 .label
= TAKE_PTR(pp
.label
),
4350 .mount_node_fd
= -EBADF
,
4352 .offset
= pp
.offset
,
4353 .fsmount_fd
= TAKE_FD(fsmount_fd
),
4357 di
->image_size
= p
.image_size
;
4358 di
->sector_size
= p
.sector_size
;
4359 di
->image_uuid
= p
.image_uuid
;
4361 *ret
= TAKE_PTR(di
);