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"
29 #include "chase-symlinks.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-release.h"
46 #include "fsck-util.h"
48 #include "hexdecoct.h"
49 #include "hostname-setup.h"
50 #include "id128-util.h"
51 #include "import-util.h"
53 #include "missing_mount.h"
54 #include "mkdir-label.h"
55 #include "mount-util.h"
56 #include "mountpoint-util.h"
57 #include "namespace-util.h"
58 #include "nulstr-util.h"
59 #include "openssl-util.h"
61 #include "path-util.h"
62 #include "process-util.h"
63 #include "raw-clone.h"
64 #include "resize-fs.h"
65 #include "signal-util.h"
66 #include "sparse-endian.h"
67 #include "stat-util.h"
68 #include "stdio-util.h"
69 #include "string-table.h"
70 #include "string-util.h"
72 #include "tmpfile-util.h"
73 #include "udev-util.h"
74 #include "user-util.h"
75 #include "xattr-util.h"
77 /* how many times to wait for the device nodes to appear */
78 #define N_DEVICE_NODE_LIST_ATTEMPTS 10
80 int dissect_fstype_ok(const char *fstype
) {
84 /* When we automatically mount file systems, be a bit conservative by default what we are willing to
85 * mount, just as an extra safety net to not mount with badly maintained legacy file system
88 e
= secure_getenv("SYSTEMD_DISSECT_FILE_SYSTEMS");
90 _cleanup_strv_free_
char **l
= NULL
;
92 l
= strv_split(e
, ":");
96 b
= strv_contains(l
, fstype
);
98 b
= STR_IN_SET(fstype
,
108 log_debug("File system type '%s' is not allowed to be mounted as result of automatic dissection.", fstype
);
112 int probe_sector_size(int fd
, uint32_t *ret
) {
121 le64_t alternate_lba
;
122 le64_t first_usable_lba
;
123 le64_t last_usable_lba
;
124 sd_id128_t disk_guid
;
125 le64_t partition_entry_lba
;
126 le32_t number_of_partition_entries
;
127 le32_t size_of_partition_entry
;
128 le32_t partition_entry_array_crc32
;
131 /* Disk images might be for 512B or for 4096 sector sizes, let's try to auto-detect that by searching
132 * for the GPT headers at the relevant byte offsets */
134 assert_cc(sizeof(struct gpt_header
) == 92);
136 /* We expect a sector size in the range 512…4096. The GPT header is located in the second
137 * sector. Hence it could be at byte 512 at the earliest, and at byte 4096 at the latest. And we must
138 * read with granularity of the largest sector size we care about. Which means 8K. */
139 uint8_t sectors
[2 * 4096];
146 n
= pread(fd
, sectors
, sizeof(sectors
), 0);
149 if (n
!= sizeof(sectors
)) /* too short? */
152 /* Let's see if we find the GPT partition header with various expected sector sizes */
153 for (uint32_t sz
= 512; sz
<= 4096; sz
<<= 1) {
154 struct gpt_header
*p
;
156 assert(sizeof(sectors
) >= sz
* 2);
157 p
= (struct gpt_header
*) (sectors
+ sz
);
159 if (memcmp(p
->signature
, (const char[8]) { 'E', 'F', 'I', ' ', 'P', 'A', 'R', 'T' }, 8) != 0)
162 if (le32toh(p
->revision
) != UINT32_C(0x00010000)) /* the only known revision of the spec: 1.0 */
165 if (le32toh(p
->header_size
) < sizeof(struct gpt_header
))
168 if (le32toh(p
->header_size
) > 4096) /* larger than a sector? something is off… */
171 if (le64toh(p
->my_lba
) != 1) /* this sector must claim to be at sector offset 1 */
175 return log_debug_errno(SYNTHETIC_ERRNO(ENOTUNIQ
),
176 "Detected valid partition table at offsets matching multiple sector sizes, refusing.");
182 log_debug("Determined sector size %" PRIu32
" based on discovered partition table.", found
);
184 return 1; /* indicate we *did* find it */
188 log_debug("Couldn't find any partition table to derive sector size of.");
189 *ret
= 512; /* pick the traditional default */
190 return 0; /* indicate we didn't find it */
193 int probe_sector_size_prefer_ioctl(int fd
, uint32_t *ret
) {
199 /* Just like probe_sector_size(), but if we are looking at a block device, will use the already
200 * configured sector size rather than probing by contents */
202 if (fstat(fd
, &st
) < 0)
205 if (S_ISBLK(st
.st_mode
))
206 return blockdev_get_sector_size(fd
, ret
);
208 return probe_sector_size(fd
, ret
);
211 int probe_filesystem_full(
218 /* Try to find device content type and return it in *ret_fstype. If nothing is found,
219 * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and a
220 * different error otherwise. */
223 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
224 _cleanup_free_
char *path_by_fd
= NULL
;
225 _cleanup_close_
int fd_close
= -EBADF
;
229 assert(fd
>= 0 || path
);
233 fd_close
= open(path
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
241 r
= fd_get_path(fd
, &path_by_fd
);
248 if (size
== 0) /* empty size? nothing found! */
251 b
= blkid_new_probe();
256 r
= blkid_probe_set_device(
260 size
== UINT64_MAX
? 0 : size
); /* when blkid sees size=0 it understands "everything". We prefer using UINT64_MAX for that */
262 return errno_or_else(ENOMEM
);
264 blkid_probe_enable_superblocks(b
, 1);
265 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
);
268 r
= blkid_do_safeprobe(b
);
269 if (r
== _BLKID_SAFEPROBE_NOT_FOUND
)
271 if (r
== _BLKID_SAFEPROBE_AMBIGUOUS
)
272 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
),
273 "Results ambiguous for partition %s", path
);
274 if (r
== _BLKID_SAFEPROBE_ERROR
)
275 return log_debug_errno(errno_or_else(EIO
), "Failed to probe partition %s: %m", path
);
277 assert(r
== _BLKID_SAFEPROBE_FOUND
);
279 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
284 log_debug("Probed fstype '%s' on partition %s.", fstype
, path
);
295 log_debug("No type detected on partition %s", path
);
304 static int dissected_image_probe_filesystems(DissectedImage
*m
, int fd
) {
309 /* Fill in file system types if we don't know them yet. */
311 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
312 DissectedPartition
*p
= m
->partitions
+ i
;
318 /* If we have an fd referring to the partition block device, use that. Otherwise go
319 * via the whole block device or backing regular file, and read via offset. */
320 if (p
->mount_node_fd
>= 0)
321 r
= probe_filesystem_full(p
->mount_node_fd
, p
->node
, 0, UINT64_MAX
, &p
->fstype
);
323 r
= probe_filesystem_full(fd
, p
->node
, p
->offset
, p
->size
, &p
->fstype
);
328 if (streq_ptr(p
->fstype
, "crypto_LUKS"))
331 if (p
->fstype
&& fstype_is_ro(p
->fstype
))
341 static void check_partition_flags(
343 unsigned long long pflags
,
344 unsigned long long supported
) {
348 /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
349 pflags
&= ~(supported
|
350 SD_GPT_FLAG_REQUIRED_PARTITION
|
351 SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL
|
352 SD_GPT_FLAG_LEGACY_BIOS_BOOTABLE
);
357 /* If there are other bits set, then log about it, to make things discoverable */
358 for (unsigned i
= 0; i
< sizeof(pflags
) * 8; i
++) {
359 unsigned long long bit
= 1ULL << i
;
360 if (!FLAGS_SET(pflags
, bit
))
363 log_debug("Unexpected partition flag %llu set on %s!", bit
, node
);
369 static int dissected_image_new(const char *path
, DissectedImage
**ret
) {
370 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
371 _cleanup_free_
char *name
= NULL
;
377 _cleanup_free_
char *filename
= NULL
;
379 r
= path_extract_filename(path
, &filename
);
383 r
= raw_strip_suffixes(filename
, &name
);
387 if (!image_name_is_valid(name
)) {
388 log_debug("Image name %s is not valid, ignoring.", strna(name
));
393 m
= new(DissectedImage
, 1);
397 *m
= (DissectedImage
) {
398 .has_init_system
= -1,
399 .image_name
= TAKE_PTR(name
),
402 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++)
403 m
->partitions
[i
] = DISSECTED_PARTITION_NULL
;
410 static void dissected_partition_done(DissectedPartition
*p
) {
416 free(p
->decrypted_fstype
);
417 free(p
->decrypted_node
);
418 free(p
->mount_options
);
419 safe_close(p
->mount_node_fd
);
421 *p
= DISSECTED_PARTITION_NULL
;
425 static int make_partition_devname(
426 const char *whole_devname
,
429 DissectImageFlags flags
,
432 _cleanup_free_
char *s
= NULL
;
435 assert(whole_devname
);
436 assert(nr
!= 0); /* zero is not a valid partition nr */
439 if (!FLAGS_SET(flags
, DISSECT_IMAGE_DISKSEQ_DEVNODE
) || diskseq
== 0) {
441 /* Given a whole block device node name (e.g. /dev/sda or /dev/loop7) generate a partition
442 * device name (e.g. /dev/sda7 or /dev/loop7p5). The rule the kernel uses is simple: if whole
443 * block device node name ends in a digit, then suffix a 'p', followed by the partition
444 * number. Otherwise, just suffix the partition number without any 'p'. */
446 if (nr
< 0) { /* whole disk? */
447 s
= strdup(whole_devname
);
451 size_t l
= strlen(whole_devname
);
452 if (l
< 1) /* underflow check for the subtraction below */
455 bool need_p
= ascii_isdigit(whole_devname
[l
-1]); /* Last char a digit? */
457 if (asprintf(&s
, "%s%s%i", whole_devname
, need_p
? "p" : "", nr
) < 0)
461 if (nr
< 0) /* whole disk? */
462 r
= asprintf(&s
, "/dev/disk/by-diskseq/%" PRIu64
, diskseq
);
464 r
= asprintf(&s
, "/dev/disk/by-diskseq/%" PRIu64
"-part%i", diskseq
, nr
);
473 static int open_partition(
476 const LoopDevice
*loop
) {
478 _cleanup_(sd_device_unrefp
) sd_device
*dev
= NULL
;
479 _cleanup_close_
int fd
= -EBADF
;
486 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
490 /* Check if the block device is a child of (or equivalent to) the originally provided one. */
491 r
= block_device_new_from_fd(fd
, is_partition
? BLOCK_DEVICE_LOOKUP_WHOLE_DISK
: 0, &dev
);
495 r
= sd_device_get_devnum(dev
, &devnum
);
499 if (loop
->devno
!= devnum
)
502 /* Also check diskseq. */
503 if (loop
->diskseq
!= 0) {
506 r
= fd_get_diskseq(fd
, &diskseq
);
510 if (loop
->diskseq
!= diskseq
)
514 log_debug("Opened %s (fd=%i, whole_block_devnum=" DEVNUM_FORMAT_STR
", diskseq=%" PRIu64
").",
515 node
, fd
, DEVNUM_FORMAT_VAL(loop
->devno
), loop
->diskseq
);
519 static int compare_arch(Architecture a
, Architecture b
) {
523 if (a
== native_architecture())
526 if (b
== native_architecture())
529 #ifdef ARCHITECTURE_SECONDARY
530 if (a
== ARCHITECTURE_SECONDARY
)
533 if (b
== ARCHITECTURE_SECONDARY
)
540 static int dissect_image(
544 const VeritySettings
*verity
,
545 const MountOptions
*mount_options
,
546 DissectImageFlags flags
) {
548 sd_id128_t root_uuid
= SD_ID128_NULL
, root_verity_uuid
= SD_ID128_NULL
;
549 sd_id128_t usr_uuid
= SD_ID128_NULL
, usr_verity_uuid
= SD_ID128_NULL
;
550 bool is_gpt
, is_mbr
, multiple_generic
= false,
551 generic_rw
= false, /* initialize to appease gcc */
552 generic_growfs
= false;
553 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
554 _cleanup_free_
char *generic_node
= NULL
;
555 sd_id128_t generic_uuid
= SD_ID128_NULL
;
556 const char *pttype
= NULL
, *sptuuid
= NULL
;
558 int r
, generic_nr
= -1, n_partitions
;
563 assert(!verity
|| verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
564 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
565 assert(!verity
|| verity
->root_hash_sig
|| verity
->root_hash_sig_size
== 0);
566 assert(!verity
|| (verity
->root_hash
|| !verity
->root_hash_sig
));
567 assert(!((flags
& DISSECT_IMAGE_GPT_ONLY
) && (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)));
568 assert(m
->sector_size
> 0);
570 /* Probes a disk image, and returns information about what it found in *ret.
572 * Returns -ENOPKG if no suitable partition table or file system could be found.
573 * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found.
574 * Returns -ENXIO if we couldn't find any partition suitable as root or /usr partition
575 * Returns -ENOTUNIQ if we only found multiple generic partitions and thus don't know what to do with that */
577 uint64_t diskseq
= m
->loop
? m
->loop
->diskseq
: 0;
579 if (verity
&& verity
->root_hash
) {
580 sd_id128_t fsuuid
, vuuid
;
582 /* If a root hash is supplied, then we use the root partition that has a UUID that match the
583 * first 128bit of the root hash. And we use the verity partition that has a UUID that match
584 * the final 128bit. */
586 if (verity
->root_hash_size
< sizeof(sd_id128_t
))
589 memcpy(&fsuuid
, verity
->root_hash
, sizeof(sd_id128_t
));
590 memcpy(&vuuid
, (const uint8_t*) verity
->root_hash
+ verity
->root_hash_size
- sizeof(sd_id128_t
), sizeof(sd_id128_t
));
592 if (sd_id128_is_null(fsuuid
))
594 if (sd_id128_is_null(vuuid
))
597 /* If the verity data declares it's for the /usr partition, then search for that, in all
598 * other cases assume it's for the root partition. */
599 if (verity
->designator
== PARTITION_USR
) {
601 usr_verity_uuid
= vuuid
;
604 root_verity_uuid
= vuuid
;
608 b
= blkid_new_probe();
613 r
= blkid_probe_set_device(b
, fd
, 0, 0);
615 return errno_or_else(ENOMEM
);
618 r
= blkid_probe_set_sectorsize(b
, m
->sector_size
);
620 return errno_or_else(EIO
);
622 if ((flags
& DISSECT_IMAGE_GPT_ONLY
) == 0) {
623 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
624 blkid_probe_enable_superblocks(b
, 1);
625 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
|BLKID_SUBLKS_USAGE
|BLKID_SUBLKS_UUID
);
628 blkid_probe_enable_partitions(b
, 1);
629 blkid_probe_set_partitions_flags(b
, BLKID_PARTS_ENTRY_DETAILS
);
632 r
= blkid_do_safeprobe(b
);
633 if (r
== _BLKID_SAFEPROBE_ERROR
)
634 return errno_or_else(EIO
);
635 if (IN_SET(r
, _BLKID_SAFEPROBE_AMBIGUOUS
, _BLKID_SAFEPROBE_NOT_FOUND
))
636 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG
), "Failed to identify any partition table.");
638 assert(r
== _BLKID_SAFEPROBE_FOUND
);
640 if ((!(flags
& DISSECT_IMAGE_GPT_ONLY
) &&
641 (flags
& DISSECT_IMAGE_GENERIC_ROOT
)) ||
642 (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)) {
643 const char *usage
= NULL
;
645 /* If flags permit this, also allow using non-partitioned single-filesystem images */
647 (void) blkid_probe_lookup_value(b
, "USAGE", &usage
, NULL
);
648 if (STRPTR_IN_SET(usage
, "filesystem", "crypto")) {
649 _cleanup_free_
char *t
= NULL
, *n
= NULL
, *o
= NULL
;
650 const char *fstype
= NULL
, *options
= NULL
, *suuid
= NULL
;
651 _cleanup_close_
int mount_node_fd
= -EBADF
;
652 sd_id128_t uuid
= SD_ID128_NULL
;
654 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
655 mount_node_fd
= open_partition(devname
, /* is_partition = */ false, m
->loop
);
656 if (mount_node_fd
< 0)
657 return mount_node_fd
;
660 /* OK, we have found a file system, that's our root partition then. */
661 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
662 (void) blkid_probe_lookup_value(b
, "UUID", &suuid
, NULL
);
671 /* blkid will return FAT's serial number as UUID, hence it is quite possible
672 * that parsing this will fail. We'll ignore the ID, since it's just too
673 * short to be useful as tru identifier. */
674 r
= sd_id128_from_string(suuid
, &uuid
);
676 log_debug_errno(r
, "Failed to parse file system UUID '%s', ignoring: %m", suuid
);
679 r
= make_partition_devname(devname
, diskseq
, -1, flags
, &n
);
683 m
->single_file_system
= true;
684 m
->encrypted
= streq_ptr(fstype
, "crypto_LUKS");
686 m
->has_verity
= verity
&& verity
->data_path
;
687 m
->verity_ready
= verity_settings_data_covers(verity
, PARTITION_ROOT
);
689 m
->has_verity_sig
= false; /* signature not embedded, must be specified */
690 m
->verity_sig_ready
= m
->verity_ready
&& verity
->root_hash_sig
;
692 m
->image_uuid
= uuid
;
694 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
701 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
703 .rw
= !m
->verity_ready
&& !fstype_is_ro(fstype
),
705 .architecture
= _ARCHITECTURE_INVALID
,
706 .fstype
= TAKE_PTR(t
),
708 .mount_options
= TAKE_PTR(o
),
709 .mount_node_fd
= TAKE_FD(mount_node_fd
),
718 (void) blkid_probe_lookup_value(b
, "PTTYPE", &pttype
, NULL
);
722 is_gpt
= streq_ptr(pttype
, "gpt");
723 is_mbr
= streq_ptr(pttype
, "dos");
725 if (!is_gpt
&& ((flags
& DISSECT_IMAGE_GPT_ONLY
) || !is_mbr
))
728 /* We support external verity data partitions only if the image has no partition table */
729 if (verity
&& verity
->data_path
)
732 if (FLAGS_SET(flags
, DISSECT_IMAGE_ADD_PARTITION_DEVICES
)) {
733 /* Safety check: refuse block devices that carry a partition table but for which the kernel doesn't
734 * do partition scanning. */
735 r
= blockdev_partscan_enabled(fd
);
739 return -EPROTONOSUPPORT
;
742 (void) blkid_probe_lookup_value(b
, "PTUUID", &sptuuid
, NULL
);
744 r
= sd_id128_from_string(sptuuid
, &m
->image_uuid
);
746 log_debug_errno(r
, "Failed to parse partition table UUID '%s', ignoring: %m", sptuuid
);
750 pl
= blkid_probe_get_partitions(b
);
752 return errno_or_else(ENOMEM
);
755 n_partitions
= blkid_partlist_numof_partitions(pl
);
756 if (n_partitions
< 0)
757 return errno_or_else(EIO
);
759 for (int i
= 0; i
< n_partitions
; i
++) {
760 _cleanup_free_
char *node
= NULL
;
761 unsigned long long pflags
;
762 blkid_loff_t start
, size
;
767 pp
= blkid_partlist_get_partition(pl
, i
);
769 return errno_or_else(EIO
);
771 pflags
= blkid_partition_get_flags(pp
);
774 nr
= blkid_partition_get_partno(pp
);
776 return errno_or_else(EIO
);
779 start
= blkid_partition_get_start(pp
);
781 return errno_or_else(EIO
);
783 assert((uint64_t) start
< UINT64_MAX
/512);
786 size
= blkid_partition_get_size(pp
);
788 return errno_or_else(EIO
);
790 assert((uint64_t) size
< UINT64_MAX
/512);
792 /* While probing we need the non-diskseq device node name to access the thing, hence mask off
793 * DISSECT_IMAGE_DISKSEQ_DEVNODE. */
794 r
= make_partition_devname(devname
, diskseq
, nr
, flags
& ~DISSECT_IMAGE_DISKSEQ_DEVNODE
, &node
);
798 /* So here's the thing: after the main ("whole") block device popped up it might take a while
799 * before the kernel fully probed the partition table. Waiting for that to finish is icky in
800 * userspace. So here's what we do instead. We issue the BLKPG_ADD_PARTITION ioctl to add the
801 * partition ourselves, racing against the kernel. Good thing is: if this call fails with
802 * EBUSY then the kernel was quicker than us, and that's totally OK, the outcome is good for
803 * us: the device node will exist. If OTOH our call was successful we won the race. Which is
804 * also good as the outcome is the same: the partition block device exists, and we can use
807 * Kernel returns EBUSY if there's already a partition by that number or an overlapping
808 * partition already existent. */
810 if (FLAGS_SET(flags
, DISSECT_IMAGE_ADD_PARTITION_DEVICES
)) {
811 r
= block_device_add_partition(fd
, node
, nr
, (uint64_t) start
* 512, (uint64_t) size
* 512);
814 return log_debug_errno(r
, "BLKPG_ADD_PARTITION failed: %m");
816 log_debug_errno(r
, "Kernel was quicker than us in adding partition %i.", nr
);
818 log_debug("We were quicker than kernel in adding partition %i.", nr
);
822 const char *fstype
= NULL
, *label
;
823 sd_id128_t type_id
, id
;
824 GptPartitionType type
;
825 bool rw
= true, growfs
= false;
827 r
= blkid_partition_get_uuid_id128(pp
, &id
);
829 log_debug_errno(r
, "Failed to read partition UUID, ignoring: %m");
833 r
= blkid_partition_get_type_id128(pp
, &type_id
);
835 log_debug_errno(r
, "Failed to read partition type UUID, ignoring: %m");
839 type
= gpt_partition_type_from_uuid(type_id
);
841 label
= blkid_partition_get_name(pp
); /* libblkid returns NULL here if empty */
843 if (IN_SET(type
.designator
,
849 check_partition_flags(node
, pflags
,
850 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
852 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
855 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
856 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
858 } else if (type
.designator
== PARTITION_ESP
) {
860 /* Note that we don't check the SD_GPT_FLAG_NO_AUTO flag for the ESP, as it is
861 * not defined there. We instead check the SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as
862 * recommended by the UEFI spec (See "12.3.3 Number and Location of System
865 if (pflags
& SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL
)
870 } else if (type
.designator
== PARTITION_ROOT
) {
872 check_partition_flags(node
, pflags
,
873 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
875 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
878 /* If a root ID is specified, ignore everything but the root id */
879 if (!sd_id128_is_null(root_uuid
) && !sd_id128_equal(root_uuid
, id
))
882 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
883 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
885 } else if (type
.designator
== PARTITION_ROOT_VERITY
) {
887 check_partition_flags(node
, pflags
,
888 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
890 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
893 m
->has_verity
= true;
895 /* If no verity configuration is specified, then don't do verity */
898 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
901 /* If root hash is specified, then ignore everything but the root id */
902 if (!sd_id128_is_null(root_verity_uuid
) && !sd_id128_equal(root_verity_uuid
, id
))
905 fstype
= "DM_verity_hash";
908 } else if (type
.designator
== PARTITION_ROOT_VERITY_SIG
) {
910 check_partition_flags(node
, pflags
,
911 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
913 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
916 m
->has_verity_sig
= true;
920 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
923 fstype
= "verity_hash_signature";
926 } else if (type
.designator
== PARTITION_USR
) {
928 check_partition_flags(node
, pflags
,
929 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
931 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
934 /* If a usr ID is specified, ignore everything but the usr id */
935 if (!sd_id128_is_null(usr_uuid
) && !sd_id128_equal(usr_uuid
, id
))
938 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
939 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
941 } else if (type
.designator
== PARTITION_USR_VERITY
) {
943 check_partition_flags(node
, pflags
,
944 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
946 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
949 m
->has_verity
= true;
953 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
956 /* If usr hash is specified, then ignore everything but the usr id */
957 if (!sd_id128_is_null(usr_verity_uuid
) && !sd_id128_equal(usr_verity_uuid
, id
))
960 fstype
= "DM_verity_hash";
963 } else if (type
.designator
== PARTITION_USR_VERITY_SIG
) {
965 check_partition_flags(node
, pflags
,
966 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
968 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
971 m
->has_verity_sig
= true;
975 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
978 fstype
= "verity_hash_signature";
981 } else if (type
.designator
== PARTITION_SWAP
) {
983 check_partition_flags(node
, pflags
, SD_GPT_FLAG_NO_AUTO
);
985 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
988 /* Note: we don't set fstype = "swap" here, because we still need to probe if
989 * it might be encrypted (i.e. fstype "crypt_LUKS") or unencrypted
990 * (i.e. fstype "swap"), and the only way to figure that out is via fstype
993 /* We don't have a designator for SD_GPT_LINUX_GENERIC so check the UUID instead. */
994 } else if (sd_id128_equal(type
.uuid
, SD_GPT_LINUX_GENERIC
)) {
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
)
1003 multiple_generic
= true;
1006 generic_rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1007 generic_growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1009 generic_node
= TAKE_PTR(node
);
1012 } else if (type
.designator
== PARTITION_VAR
) {
1014 check_partition_flags(node
, pflags
,
1015 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1017 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1020 if (!FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_VAR_CHECK
)) {
1021 sd_id128_t var_uuid
;
1023 /* For /var we insist that the uuid of the partition matches the
1024 * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
1025 * ID. Why? Unlike the other partitions /var is inherently
1026 * installation specific, hence we need to be careful not to mount it
1027 * in the wrong installation. By hashing the partition UUID from
1028 * /etc/machine-id we can securely bind the partition to the
1031 r
= sd_id128_get_machine_app_specific(SD_GPT_VAR
, &var_uuid
);
1035 if (!sd_id128_equal(var_uuid
, id
)) {
1036 log_debug("Found a /var/ partition, but its UUID didn't match our expectations "
1037 "(found: " SD_ID128_UUID_FORMAT_STR
", expected: " SD_ID128_UUID_FORMAT_STR
"), ignoring.",
1038 SD_ID128_FORMAT_VAL(id
), SD_ID128_FORMAT_VAL(var_uuid
));
1043 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1044 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1047 if (type
.designator
!= _PARTITION_DESIGNATOR_INVALID
) {
1048 _cleanup_free_
char *t
= NULL
, *o
= NULL
, *l
= NULL
, *n
= NULL
;
1049 _cleanup_close_
int mount_node_fd
= -EBADF
;
1050 const char *options
= NULL
;
1052 if (m
->partitions
[type
.designator
].found
) {
1053 /* For most partition types the first one we see wins. Except for the
1054 * rootfs and /usr, where we do a version compare of the label, and
1055 * let the newest version win. This permits a simple A/B versioning
1056 * scheme in OS images. */
1058 if (compare_arch(type
.arch
, m
->partitions
[type
.designator
].architecture
) <= 0)
1061 if (!partition_designator_is_versioned(type
.designator
) ||
1062 strverscmp_improved(m
->partitions
[type
.designator
].label
, label
) >= 0)
1065 dissected_partition_done(m
->partitions
+ type
.designator
);
1068 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
) &&
1069 type
.designator
!= PARTITION_SWAP
) {
1070 mount_node_fd
= open_partition(node
, /* is_partition = */ true, m
->loop
);
1071 if (mount_node_fd
< 0)
1072 return mount_node_fd
;
1075 r
= make_partition_devname(devname
, diskseq
, nr
, flags
, &n
);
1091 options
= mount_options_from_designator(mount_options
, type
.designator
);
1093 o
= strdup(options
);
1098 m
->partitions
[type
.designator
] = (DissectedPartition
) {
1103 .architecture
= type
.arch
,
1104 .node
= TAKE_PTR(n
),
1105 .fstype
= TAKE_PTR(t
),
1106 .label
= TAKE_PTR(l
),
1108 .mount_options
= TAKE_PTR(o
),
1109 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1110 .offset
= (uint64_t) start
* 512,
1111 .size
= (uint64_t) size
* 512,
1112 .gpt_flags
= pflags
,
1116 } else if (is_mbr
) {
1118 switch (blkid_partition_get_type(pp
)) {
1120 case 0x83: /* Linux partition */
1122 if (pflags
!= 0x80) /* Bootable flag */
1126 multiple_generic
= true;
1130 generic_growfs
= false;
1131 generic_node
= TAKE_PTR(node
);
1136 case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
1137 _cleanup_close_
int mount_node_fd
= -EBADF
;
1138 _cleanup_free_
char *o
= NULL
, *n
= NULL
;
1139 sd_id128_t id
= SD_ID128_NULL
;
1140 const char *options
= NULL
;
1142 /* First one wins */
1143 if (m
->partitions
[PARTITION_XBOOTLDR
].found
)
1146 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
1147 mount_node_fd
= open_partition(node
, /* is_partition = */ true, m
->loop
);
1148 if (mount_node_fd
< 0)
1149 return mount_node_fd
;
1152 (void) blkid_partition_get_uuid_id128(pp
, &id
);
1154 r
= make_partition_devname(devname
, diskseq
, nr
, flags
, &n
);
1158 options
= mount_options_from_designator(mount_options
, PARTITION_XBOOTLDR
);
1160 o
= strdup(options
);
1165 m
->partitions
[PARTITION_XBOOTLDR
] = (DissectedPartition
) {
1170 .architecture
= _ARCHITECTURE_INVALID
,
1171 .node
= TAKE_PTR(n
),
1173 .mount_options
= TAKE_PTR(o
),
1174 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1175 .offset
= (uint64_t) start
* 512,
1176 .size
= (uint64_t) size
* 512,
1184 if (!m
->partitions
[PARTITION_ROOT
].found
&&
1185 (m
->partitions
[PARTITION_ROOT_VERITY
].found
||
1186 m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
))
1187 return -EADDRNOTAVAIL
; /* Verity found but no matching rootfs? Something is off, refuse. */
1189 /* Hmm, we found a signature partition but no Verity data? Something is off. */
1190 if (m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_ROOT_VERITY
].found
)
1191 return -EADDRNOTAVAIL
;
1193 if (!m
->partitions
[PARTITION_USR
].found
&&
1194 (m
->partitions
[PARTITION_USR_VERITY
].found
||
1195 m
->partitions
[PARTITION_USR_VERITY_SIG
].found
))
1196 return -EADDRNOTAVAIL
; /* as above */
1199 if (m
->partitions
[PARTITION_USR_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_USR_VERITY
].found
)
1200 return -EADDRNOTAVAIL
;
1202 /* If root and /usr are combined then insist that the architecture matches */
1203 if (m
->partitions
[PARTITION_ROOT
].found
&&
1204 m
->partitions
[PARTITION_USR
].found
&&
1205 (m
->partitions
[PARTITION_ROOT
].architecture
>= 0 &&
1206 m
->partitions
[PARTITION_USR
].architecture
>= 0 &&
1207 m
->partitions
[PARTITION_ROOT
].architecture
!= m
->partitions
[PARTITION_USR
].architecture
))
1208 return -EADDRNOTAVAIL
;
1210 if (!m
->partitions
[PARTITION_ROOT
].found
&&
1211 !m
->partitions
[PARTITION_USR
].found
&&
1212 (flags
& DISSECT_IMAGE_GENERIC_ROOT
) &&
1213 (!verity
|| !verity
->root_hash
|| verity
->designator
!= PARTITION_USR
)) {
1215 /* OK, we found nothing usable, then check if there's a single generic partition, and use
1216 * that. If the root hash was set however, then we won't fall back to a generic node, because
1217 * the root hash decides. */
1219 /* If we didn't find a properly marked root partition, but we did find a single suitable
1220 * generic Linux partition, then use this as root partition, if the caller asked for it. */
1221 if (multiple_generic
)
1224 /* If we didn't find a generic node, then we can't fix this up either */
1226 _cleanup_close_
int mount_node_fd
= -EBADF
;
1227 _cleanup_free_
char *o
= NULL
, *n
= NULL
;
1228 const char *options
;
1230 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
1231 mount_node_fd
= open_partition(generic_node
, /* is_partition = */ true, m
->loop
);
1232 if (mount_node_fd
< 0)
1233 return mount_node_fd
;
1236 r
= make_partition_devname(devname
, diskseq
, generic_nr
, flags
, &n
);
1240 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
1242 o
= strdup(options
);
1247 assert(generic_nr
>= 0);
1248 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
1251 .growfs
= generic_growfs
,
1252 .partno
= generic_nr
,
1253 .architecture
= _ARCHITECTURE_INVALID
,
1254 .node
= TAKE_PTR(n
),
1255 .uuid
= generic_uuid
,
1256 .mount_options
= TAKE_PTR(o
),
1257 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1258 .offset
= UINT64_MAX
,
1264 /* 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 */
1265 if (FLAGS_SET(flags
, DISSECT_IMAGE_REQUIRE_ROOT
) &&
1266 !(m
->partitions
[PARTITION_ROOT
].found
|| (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
1269 if (m
->partitions
[PARTITION_ROOT_VERITY
].found
) {
1270 /* We only support one verity partition per image, i.e. can't do for both /usr and root fs */
1271 if (m
->partitions
[PARTITION_USR_VERITY
].found
)
1274 /* We don't support verity enabled root with a split out /usr. Neither with nor without
1275 * verity there. (Note that we do support verity-less root with verity-full /usr, though.) */
1276 if (m
->partitions
[PARTITION_USR
].found
)
1277 return -EADDRNOTAVAIL
;
1281 /* If a verity designator is specified, then insist that the matching partition exists */
1282 if (verity
->designator
>= 0 && !m
->partitions
[verity
->designator
].found
)
1283 return -EADDRNOTAVAIL
;
1285 bool have_verity_sig_partition
=
1286 m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR_VERITY_SIG
: PARTITION_ROOT_VERITY_SIG
].found
;
1288 if (verity
->root_hash
) {
1289 /* If we have an explicit root hash and found the partitions for it, then we are ready to use
1290 * Verity, set things up for it */
1292 if (verity
->designator
< 0 || verity
->designator
== PARTITION_ROOT
) {
1293 if (!m
->partitions
[PARTITION_ROOT_VERITY
].found
|| !m
->partitions
[PARTITION_ROOT
].found
)
1294 return -EADDRNOTAVAIL
;
1296 /* If we found a verity setup, then the root partition is necessarily read-only. */
1297 m
->partitions
[PARTITION_ROOT
].rw
= false;
1298 m
->verity_ready
= true;
1301 assert(verity
->designator
== PARTITION_USR
);
1303 if (!m
->partitions
[PARTITION_USR_VERITY
].found
|| !m
->partitions
[PARTITION_USR
].found
)
1304 return -EADDRNOTAVAIL
;
1306 m
->partitions
[PARTITION_USR
].rw
= false;
1307 m
->verity_ready
= true;
1310 if (m
->verity_ready
)
1311 m
->verity_sig_ready
= verity
->root_hash_sig
|| have_verity_sig_partition
;
1313 } else if (have_verity_sig_partition
) {
1315 /* If we found an embedded signature partition, we are ready, too. */
1317 m
->verity_ready
= m
->verity_sig_ready
= true;
1318 m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR
: PARTITION_ROOT
].rw
= false;
1322 r
= dissected_image_probe_filesystems(m
, fd
);
1330 int dissect_image_file(
1332 const VeritySettings
*verity
,
1333 const MountOptions
*mount_options
,
1334 DissectImageFlags flags
,
1335 DissectedImage
**ret
) {
1338 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
1339 _cleanup_close_
int fd
= -EBADF
;
1345 fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
1349 r
= fd_verify_regular(fd
);
1353 r
= dissected_image_new(path
, &m
);
1357 r
= probe_sector_size(fd
, &m
->sector_size
);
1361 r
= dissect_image(m
, fd
, path
, verity
, mount_options
, flags
);
1372 DissectedImage
* dissected_image_unref(DissectedImage
*m
) {
1376 /* First, clear dissected partitions. */
1377 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++)
1378 dissected_partition_done(m
->partitions
+ i
);
1380 /* Second, free decrypted images. This must be after dissected_partition_done(), as freeing
1381 * DecryptedImage may try to deactivate partitions. */
1382 decrypted_image_unref(m
->decrypted_image
);
1384 /* Third, unref LoopDevice. This must be called after the above two, as freeing LoopDevice may try to
1385 * remove existing partitions on the loopback block device. */
1386 loop_device_unref(m
->loop
);
1388 free(m
->image_name
);
1390 strv_free(m
->machine_info
);
1391 strv_free(m
->os_release
);
1392 strv_free(m
->initrd_release
);
1393 strv_free(m
->extension_release
);
1398 static int is_loop_device(const char *path
) {
1399 char s
[SYS_BLOCK_PATH_MAX("/../loop/")];
1404 if (stat(path
, &st
) < 0)
1407 if (!S_ISBLK(st
.st_mode
))
1410 xsprintf_sys_block_path(s
, "/loop/", st
.st_dev
);
1411 if (access(s
, F_OK
) < 0) {
1412 if (errno
!= ENOENT
)
1415 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
1416 xsprintf_sys_block_path(s
, "/../loop/", st
.st_dev
);
1417 if (access(s
, F_OK
) < 0)
1418 return errno
== ENOENT
? false : -errno
;
1424 static int run_fsck(int node_fd
, const char *fstype
) {
1428 assert(node_fd
>= 0);
1431 r
= fsck_exists_for_fstype(fstype
);
1433 log_debug_errno(r
, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype
);
1437 log_debug("Not checking partition %s, as fsck for %s does not exist.", FORMAT_PROC_FD_PATH(node_fd
), fstype
);
1444 &node_fd
, 1, /* Leave the node fd open */
1445 FORK_RESET_SIGNALS
|FORK_CLOSE_ALL_FDS
|FORK_RLIMIT_NOFILE_SAFE
|FORK_DEATHSIG
|FORK_REARRANGE_STDIO
|FORK_CLOEXEC_OFF
,
1448 return log_debug_errno(r
, "Failed to fork off fsck: %m");
1451 execl("/sbin/fsck", "/sbin/fsck", "-aT", FORMAT_PROC_FD_PATH(node_fd
), NULL
);
1453 log_debug_errno(errno
, "Failed to execl() fsck: %m");
1454 _exit(FSCK_OPERATIONAL_ERROR
);
1457 exit_status
= wait_for_terminate_and_check("fsck", pid
, 0);
1458 if (exit_status
< 0)
1459 return log_debug_errno(exit_status
, "Failed to fork off /sbin/fsck: %m");
1461 if ((exit_status
& ~FSCK_ERROR_CORRECTED
) != FSCK_SUCCESS
) {
1462 log_debug("fsck failed with exit status %i.", exit_status
);
1464 if ((exit_status
& (FSCK_SYSTEM_SHOULD_REBOOT
|FSCK_ERRORS_LEFT_UNCORRECTED
)) != 0)
1465 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
), "File system is corrupted, refusing.");
1467 log_debug("Ignoring fsck error.");
1473 static int fs_grow(const char *node_path
, const char *mount_path
) {
1474 _cleanup_close_
int mount_fd
= -EBADF
, node_fd
= -EBADF
;
1475 uint64_t size
, newsize
;
1478 node_fd
= open(node_path
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
1480 return log_debug_errno(errno
, "Failed to open node device %s: %m", node_path
);
1482 if (ioctl(node_fd
, BLKGETSIZE64
, &size
) != 0)
1483 return log_debug_errno(errno
, "Failed to get block device size of %s: %m", node_path
);
1485 mount_fd
= open(mount_path
, O_RDONLY
|O_DIRECTORY
|O_CLOEXEC
);
1487 return log_debug_errno(errno
, "Failed to open mountd file system %s: %m", mount_path
);
1489 log_debug("Resizing \"%s\" to %"PRIu64
" bytes...", mount_path
, size
);
1490 r
= resize_fs(mount_fd
, size
, &newsize
);
1492 return log_debug_errno(r
, "Failed to resize \"%s\" to %"PRIu64
" bytes: %m", mount_path
, size
);
1494 if (newsize
== size
)
1495 log_debug("Successfully resized \"%s\" to %s bytes.",
1496 mount_path
, FORMAT_BYTES(newsize
));
1498 assert(newsize
< size
);
1499 log_debug("Successfully resized \"%s\" to %s bytes (%"PRIu64
" bytes lost due to blocksize).",
1500 mount_path
, FORMAT_BYTES(newsize
), size
- newsize
);
1506 int partition_pick_mount_options(
1507 PartitionDesignator d
,
1512 unsigned long *ret_ms_flags
) {
1514 _cleanup_free_
char *options
= NULL
;
1516 assert(ret_options
);
1518 /* Selects a baseline of bind mount flags, that should always apply.
1520 * Firstly, we set MS_NODEV universally on all mounts, since we don't want to allow device nodes outside of /dev/.
1522 * On /var/tmp/ we'll also set MS_NOSUID, same as we set for /tmp/ on the host.
1524 * On the ESP and XBOOTLDR partitions we'll also disable symlinks, and execution. These file systems
1525 * are generally untrusted (i.e. not encrypted or authenticated), and typically VFAT hence we should
1526 * be as restrictive as possible, and this shouldn't hurt, since the functionality is not available
1529 unsigned long flags
= MS_NODEV
;
1537 case PARTITION_XBOOTLDR
:
1538 flags
|= MS_NOSUID
|MS_NOEXEC
|ms_nosymfollow_supported();
1540 if (!fstype
|| streq(fstype
, "vfat"))
1541 if (!strextend_with_separator(&options
, ",", "umask=0077"))
1553 /* So, when you request MS_RDONLY from ext4, then this means nothing. It happily still writes to the
1554 * backing storage. What's worse, the BLKRO[GS]ET flag and (in case of loopback devices)
1555 * LO_FLAGS_READ_ONLY don't mean anything, they affect userspace accesses only, and write accesses
1556 * from the upper file system still get propagated through to the underlying file system,
1557 * unrestricted. To actually get ext4/xfs/btrfs to stop writing to the device we need to specify
1558 * "norecovery" as mount option, in addition to MS_RDONLY. Yes, this sucks, since it means we need to
1559 * carry a per file system table here.
1561 * Note that this means that we might not be able to mount corrupted file systems as read-only
1562 * anymore (since in some cases the kernel implementations will refuse mounting when corrupted,
1563 * read-only and "norecovery" is specified). But I think for the case of automatically determined
1564 * mount options for loopback devices this is the right choice, since otherwise using the same
1565 * loopback file twice even in read-only mode, is going to fail badly sooner or later. The usecase of
1566 * making reuse of the immutable images "just work" is more relevant to us than having read-only
1567 * access that actually modifies stuff work on such image files. Or to say this differently: if
1568 * people want their file systems to be fixed up they should just open them in writable mode, where
1569 * all these problems don't exist. */
1570 if (!rw
&& STRPTR_IN_SET(fstype
, "ext3", "ext4", "xfs", "btrfs"))
1571 if (!strextend_with_separator(&options
, ",", "norecovery"))
1574 if (discard
&& fstype
&& fstype_can_discard(fstype
))
1575 if (!strextend_with_separator(&options
, ",", "discard"))
1578 if (!ret_ms_flags
) /* Fold flags into option string if ret_flags specified as NULL */
1579 if (!strextend_with_separator(&options
, ",",
1580 FLAGS_SET(flags
, MS_RDONLY
) ? "ro" : "rw",
1581 FLAGS_SET(flags
, MS_NODEV
) ? "nodev" : "dev",
1582 FLAGS_SET(flags
, MS_NOSUID
) ? "nosuid" : "suid",
1583 FLAGS_SET(flags
, MS_NOEXEC
) ? "noexec" : "exec",
1584 FLAGS_SET(flags
, MS_NOSYMFOLLOW
) ? "nosymfollow" : NULL
))
1585 /* NB: we suppress 'symfollow' here, since it's the default, and old /bin/mount might not know it */
1589 *ret_ms_flags
= flags
;
1591 *ret_options
= TAKE_PTR(options
);
1595 static int mount_partition(
1596 PartitionDesignator d
,
1597 DissectedPartition
*m
,
1599 const char *directory
,
1602 DissectImageFlags flags
) {
1604 _cleanup_free_
char *chased
= NULL
, *options
= NULL
;
1605 bool rw
, discard
, remap_uid_gid
= false;
1606 const char *p
, *node
, *fstype
;
1607 unsigned long ms_flags
;
1613 if (m
->mount_node_fd
< 0)
1616 /* Use decrypted node and matching fstype if available, otherwise use the original device */
1617 node
= FORMAT_PROC_FD_PATH(m
->mount_node_fd
);
1618 fstype
= dissected_partition_fstype(m
);
1621 return -EAFNOSUPPORT
;
1622 r
= dissect_fstype_ok(fstype
);
1626 return -EIDRM
; /* Recognizable error */
1628 /* We are looking at an encrypted partition? This either means stacked encryption, or the caller
1629 * didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error for this
1631 if (streq(fstype
, "crypto_LUKS"))
1634 rw
= m
->rw
&& !(flags
& DISSECT_IMAGE_MOUNT_READ_ONLY
);
1636 discard
= ((flags
& DISSECT_IMAGE_DISCARD
) ||
1637 ((flags
& DISSECT_IMAGE_DISCARD_ON_LOOP
) && is_loop_device(m
->node
) > 0));
1639 if (FLAGS_SET(flags
, DISSECT_IMAGE_FSCK
) && rw
) {
1640 r
= run_fsck(m
->mount_node_fd
, fstype
);
1646 /* Automatically create missing mount points inside the image, if necessary. */
1647 r
= mkdir_p_root(where
, directory
, uid_shift
, (gid_t
) uid_shift
, 0755);
1648 if (r
< 0 && r
!= -EROFS
)
1651 r
= chase_symlinks(directory
, where
, CHASE_PREFIX_ROOT
, &chased
, NULL
);
1657 /* Create top-level mount if missing – but only if this is asked for. This won't modify the
1658 * image (as the branch above does) but the host hierarchy, and the created directory might
1659 * survive our mount in the host hierarchy hence. */
1660 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
1661 r
= mkdir_p(where
, 0755);
1669 r
= partition_pick_mount_options(d
, dissected_partition_fstype(m
), rw
, discard
, &options
, &ms_flags
);
1673 if (uid_is_valid(uid_shift
) && uid_shift
!= 0) {
1675 if (fstype_can_uid_gid(fstype
)) {
1676 _cleanup_free_
char *uid_option
= NULL
;
1678 if (asprintf(&uid_option
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
1681 if (!strextend_with_separator(&options
, ",", uid_option
))
1683 } else if (FLAGS_SET(flags
, DISSECT_IMAGE_MOUNT_IDMAPPED
))
1684 remap_uid_gid
= true;
1687 if (!isempty(m
->mount_options
))
1688 if (!strextend_with_separator(&options
, ",", m
->mount_options
))
1691 r
= mount_nofollow_verbose(LOG_DEBUG
, node
, p
, fstype
, ms_flags
, options
);
1695 if (rw
&& m
->growfs
&& FLAGS_SET(flags
, DISSECT_IMAGE_GROWFS
))
1696 (void) fs_grow(node
, p
);
1698 if (remap_uid_gid
) {
1699 r
= remount_idmap(p
, uid_shift
, uid_range
, UID_INVALID
, REMOUNT_IDMAPPING_HOST_ROOT
);
1707 static int mount_root_tmpfs(const char *where
, uid_t uid_shift
, DissectImageFlags flags
) {
1708 _cleanup_free_
char *options
= NULL
;
1713 /* For images that contain /usr/ but no rootfs, let's mount rootfs as tmpfs */
1715 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
1716 r
= mkdir_p(where
, 0755);
1721 if (uid_is_valid(uid_shift
)) {
1722 if (asprintf(&options
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
1726 r
= mount_nofollow_verbose(LOG_DEBUG
, "rootfs", where
, "tmpfs", MS_NODEV
, options
);
1733 int dissected_image_mount(
1738 DissectImageFlags flags
) {
1740 int r
, xbootldr_mounted
;
1747 * -ENXIO → No root partition found
1748 * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release/extension-release file found
1749 * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet
1750 * -EUCLEAN → fsck for file system failed
1751 * -EBUSY → File system already mounted/used elsewhere (kernel)
1752 * -EAFNOSUPPORT → File system type not supported or not known
1753 * -EIDRM → File system is not among allowlisted "common" file systems
1756 if (!(m
->partitions
[PARTITION_ROOT
].found
||
1757 (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
1758 return -ENXIO
; /* Require a root fs or at least a /usr/ fs (the latter is subject to a flag of its own) */
1760 if ((flags
& DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
) == 0) {
1762 /* First mount the root fs. If there's none we use a tmpfs. */
1763 if (m
->partitions
[PARTITION_ROOT
].found
)
1764 r
= mount_partition(PARTITION_ROOT
, m
->partitions
+ PARTITION_ROOT
, where
, NULL
, uid_shift
, uid_range
, flags
);
1766 r
= mount_root_tmpfs(where
, uid_shift
, flags
);
1770 /* For us mounting root always means mounting /usr as well */
1771 r
= mount_partition(PARTITION_USR
, m
->partitions
+ PARTITION_USR
, where
, "/usr", uid_shift
, uid_range
, flags
);
1775 if ((flags
& (DISSECT_IMAGE_VALIDATE_OS
|DISSECT_IMAGE_VALIDATE_OS_EXT
)) != 0) {
1776 /* If either one of the validation flags are set, ensure that the image qualifies
1777 * as one or the other (or both). */
1780 if (FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS
)) {
1781 r
= path_is_os_tree(where
);
1787 if (!ok
&& FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS_EXT
)) {
1788 r
= path_is_extension_tree(where
, m
->image_name
, FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_SYSEXT_CHECK
));
1800 if (flags
& DISSECT_IMAGE_MOUNT_ROOT_ONLY
)
1803 r
= mount_partition(PARTITION_HOME
, m
->partitions
+ PARTITION_HOME
, where
, "/home", uid_shift
, uid_range
, flags
);
1807 r
= mount_partition(PARTITION_SRV
, m
->partitions
+ PARTITION_SRV
, where
, "/srv", uid_shift
, uid_range
, flags
);
1811 r
= mount_partition(PARTITION_VAR
, m
->partitions
+ PARTITION_VAR
, where
, "/var", uid_shift
, uid_range
, flags
);
1815 r
= mount_partition(PARTITION_TMP
, m
->partitions
+ PARTITION_TMP
, where
, "/var/tmp", uid_shift
, uid_range
, flags
);
1819 xbootldr_mounted
= mount_partition(PARTITION_XBOOTLDR
, m
->partitions
+ PARTITION_XBOOTLDR
, where
, "/boot", uid_shift
, uid_range
, flags
);
1820 if (xbootldr_mounted
< 0)
1821 return xbootldr_mounted
;
1823 if (m
->partitions
[PARTITION_ESP
].found
) {
1824 int esp_done
= false;
1826 /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it
1827 * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */
1829 r
= chase_symlinks("/efi", where
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
1834 /* /efi doesn't exist. Let's see if /boot is suitable then */
1836 if (!xbootldr_mounted
) {
1837 _cleanup_free_
char *p
= NULL
;
1839 r
= chase_symlinks("/boot", where
, CHASE_PREFIX_ROOT
, &p
, NULL
);
1843 } else if (dir_is_empty(p
, /* ignore_hidden_or_backup= */ false) > 0) {
1844 /* It exists and is an empty directory. Let's mount the ESP there. */
1845 r
= mount_partition(PARTITION_ESP
, m
->partitions
+ PARTITION_ESP
, where
, "/boot", uid_shift
, uid_range
, flags
);
1855 /* OK, let's mount the ESP now to /efi (possibly creating the dir if missing) */
1857 r
= mount_partition(PARTITION_ESP
, m
->partitions
+ PARTITION_ESP
, where
, "/efi", uid_shift
, uid_range
, flags
);
1866 int dissected_image_mount_and_warn(
1871 DissectImageFlags flags
) {
1878 r
= dissected_image_mount(m
, where
, uid_shift
, uid_range
, flags
);
1880 return log_error_errno(r
, "Not root file system found in image.");
1881 if (r
== -EMEDIUMTYPE
)
1882 return log_error_errno(r
, "No suitable os-release/extension-release file in image found.");
1884 return log_error_errno(r
, "Encrypted file system discovered, but decryption not requested.");
1886 return log_error_errno(r
, "File system check on image failed.");
1888 return log_error_errno(r
, "File system already mounted elsewhere.");
1889 if (r
== -EAFNOSUPPORT
)
1890 return log_error_errno(r
, "File system type not supported or not known.");
1892 return log_error_errno(r
, "File system is too uncommon, refused.");
1894 return log_error_errno(r
, "Failed to mount image: %m");
1899 #if HAVE_LIBCRYPTSETUP
1900 struct DecryptedPartition
{
1901 struct crypt_device
*device
;
1907 typedef struct DecryptedPartition DecryptedPartition
;
1909 struct DecryptedImage
{
1911 DecryptedPartition
*decrypted
;
1915 static DecryptedImage
* decrypted_image_free(DecryptedImage
*d
) {
1916 #if HAVE_LIBCRYPTSETUP
1922 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
1923 DecryptedPartition
*p
= d
->decrypted
+ i
;
1925 if (p
->device
&& p
->name
&& !p
->relinquished
) {
1926 _cleanup_free_
char *node
= NULL
;
1928 node
= path_join("/dev/mapper", p
->name
);
1930 r
= btrfs_forget_device(node
);
1931 if (r
< 0 && r
!= -ENOENT
)
1932 log_debug_errno(r
, "Failed to forget btrfs device %s, ignoring: %m", node
);
1936 /* Let's deactivate lazily, as the dm volume may be already/still used by other processes. */
1937 r
= sym_crypt_deactivate_by_name(p
->device
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
1939 log_debug_errno(r
, "Failed to deactivate encrypted partition %s", p
->name
);
1943 sym_crypt_free(p
->device
);
1953 DEFINE_TRIVIAL_REF_UNREF_FUNC(DecryptedImage
, decrypted_image
, decrypted_image_free
);
1955 #if HAVE_LIBCRYPTSETUP
1956 static int decrypted_image_new(DecryptedImage
**ret
) {
1957 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
1961 d
= new(DecryptedImage
, 1);
1965 *d
= (DecryptedImage
) {
1973 static int make_dm_name_and_node(const void *original_node
, const char *suffix
, char **ret_name
, char **ret_node
) {
1974 _cleanup_free_
char *name
= NULL
, *node
= NULL
;
1977 assert(original_node
);
1982 base
= strrchr(original_node
, '/');
1984 base
= original_node
;
1990 name
= strjoin(base
, suffix
);
1993 if (!filename_is_valid(name
))
1996 node
= path_join(sym_crypt_get_dir(), name
);
2000 *ret_name
= TAKE_PTR(name
);
2001 *ret_node
= TAKE_PTR(node
);
2006 static int decrypt_partition(
2007 DissectedPartition
*m
,
2008 const char *passphrase
,
2009 DissectImageFlags flags
,
2010 DecryptedImage
*d
) {
2012 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
2013 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2014 _cleanup_close_
int fd
= -EBADF
;
2020 if (!m
->found
|| !m
->node
|| !m
->fstype
)
2023 if (!streq(m
->fstype
, "crypto_LUKS"))
2029 r
= dlopen_cryptsetup();
2033 r
= make_dm_name_and_node(m
->node
, "-decrypted", &name
, &node
);
2037 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
2040 r
= sym_crypt_init(&cd
, m
->node
);
2042 return log_debug_errno(r
, "Failed to initialize dm-crypt: %m");
2044 cryptsetup_enable_logging(cd
);
2046 r
= sym_crypt_load(cd
, CRYPT_LUKS
, NULL
);
2048 return log_debug_errno(r
, "Failed to load LUKS metadata: %m");
2050 r
= sym_crypt_activate_by_passphrase(cd
, name
, CRYPT_ANY_SLOT
, passphrase
, strlen(passphrase
),
2051 ((flags
& DISSECT_IMAGE_DEVICE_READ_ONLY
) ? CRYPT_ACTIVATE_READONLY
: 0) |
2052 ((flags
& DISSECT_IMAGE_DISCARD_ON_CRYPTO
) ? CRYPT_ACTIVATE_ALLOW_DISCARDS
: 0));
2054 log_debug_errno(r
, "Failed to activate LUKS device: %m");
2055 return r
== -EPERM
? -EKEYREJECTED
: r
;
2058 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2060 return log_debug_errno(errno
, "Failed to open %s: %m", node
);
2062 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
2063 .name
= TAKE_PTR(name
),
2064 .device
= TAKE_PTR(cd
),
2067 m
->decrypted_node
= TAKE_PTR(node
);
2068 close_and_replace(m
->mount_node_fd
, fd
);
2073 static int verity_can_reuse(
2074 const VeritySettings
*verity
,
2076 struct crypt_device
**ret_cd
) {
2078 /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
2079 _cleanup_free_
char *root_hash_existing
= NULL
;
2080 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2081 struct crypt_params_verity crypt_params
= {};
2082 size_t root_hash_existing_size
;
2089 r
= sym_crypt_init_by_name(&cd
, name
);
2091 return log_debug_errno(r
, "Error opening verity device, crypt_init_by_name failed: %m");
2093 cryptsetup_enable_logging(cd
);
2095 r
= sym_crypt_get_verity_info(cd
, &crypt_params
);
2097 return log_debug_errno(r
, "Error opening verity device, crypt_get_verity_info failed: %m");
2099 root_hash_existing_size
= verity
->root_hash_size
;
2100 root_hash_existing
= malloc0(root_hash_existing_size
);
2101 if (!root_hash_existing
)
2104 r
= sym_crypt_volume_key_get(cd
, CRYPT_ANY_SLOT
, root_hash_existing
, &root_hash_existing_size
, NULL
, 0);
2106 return log_debug_errno(r
, "Error opening verity device, crypt_volume_key_get failed: %m");
2107 if (verity
->root_hash_size
!= root_hash_existing_size
||
2108 memcmp(root_hash_existing
, verity
->root_hash
, verity
->root_hash_size
) != 0)
2109 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but root hashes are different.");
2111 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
2112 /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the
2113 * same settings, so that a previous unsigned mount will not be reused if the user asks to use
2114 * signing for the new one, and vice versa. */
2115 if (!!verity
->root_hash_sig
!= !!(crypt_params
.flags
& CRYPT_VERITY_ROOT_HASH_SIGNATURE
))
2116 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but signature settings are not the same.");
2119 *ret_cd
= TAKE_PTR(cd
);
2123 static inline char* dm_deferred_remove_clean(char *name
) {
2127 (void) sym_crypt_deactivate_by_name(NULL
, name
, CRYPT_DEACTIVATE_DEFERRED
);
2130 DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean
);
2132 static int validate_signature_userspace(const VeritySettings
*verity
) {
2134 _cleanup_(sk_X509_free_allp
) STACK_OF(X509
) *sk
= NULL
;
2135 _cleanup_strv_free_
char **certs
= NULL
;
2136 _cleanup_(PKCS7_freep
) PKCS7
*p7
= NULL
;
2137 _cleanup_free_
char *s
= NULL
;
2138 _cleanup_(BIO_freep
) BIO
*bio
= NULL
; /* 'bio' must be freed first, 's' second, hence keep this order
2139 * of declaration in place, please */
2140 const unsigned char *d
;
2144 assert(verity
->root_hash
);
2145 assert(verity
->root_hash_sig
);
2147 /* Because installing a signature certificate into the kernel chain is so messy, let's optionally do
2148 * userspace validation. */
2150 r
= conf_files_list_nulstr(&certs
, ".crt", NULL
, CONF_FILES_REGULAR
|CONF_FILES_FILTER_MASKED
, CONF_PATHS_NULSTR("verity.d"));
2152 return log_debug_errno(r
, "Failed to enumerate certificates: %m");
2153 if (strv_isempty(certs
)) {
2154 log_debug("No userspace dm-verity certificates found.");
2158 d
= verity
->root_hash_sig
;
2159 p7
= d2i_PKCS7(NULL
, &d
, (long) verity
->root_hash_sig_size
);
2161 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to parse PKCS7 DER signature data.");
2163 s
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2165 return log_oom_debug();
2167 bio
= BIO_new_mem_buf(s
, strlen(s
));
2169 return log_oom_debug();
2171 sk
= sk_X509_new_null();
2173 return log_oom_debug();
2175 STRV_FOREACH(i
, certs
) {
2176 _cleanup_(X509_freep
) X509
*c
= NULL
;
2177 _cleanup_fclose_
FILE *f
= NULL
;
2179 f
= fopen(*i
, "re");
2181 log_debug_errno(errno
, "Failed to open '%s', ignoring: %m", *i
);
2185 c
= PEM_read_X509(f
, NULL
, NULL
, NULL
);
2187 log_debug("Failed to load X509 certificate '%s', ignoring.", *i
);
2191 if (sk_X509_push(sk
, c
) == 0)
2192 return log_oom_debug();
2197 r
= PKCS7_verify(p7
, sk
, NULL
, bio
, NULL
, PKCS7_NOINTERN
|PKCS7_NOVERIFY
);
2199 log_debug("Userspace PKCS#7 validation succeeded.");
2201 log_debug("Userspace PKCS#7 validation failed: %s", ERR_error_string(ERR_get_error(), NULL
));
2205 log_debug("Not doing client-side validation of dm-verity root hash signatures, OpenSSL support disabled.");
2210 static int do_crypt_activate_verity(
2211 struct crypt_device
*cd
,
2213 const VeritySettings
*verity
) {
2215 bool check_signature
;
2222 if (verity
->root_hash_sig
) {
2223 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIGNATURE");
2224 if (r
< 0 && r
!= -ENXIO
)
2225 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIGNATURE");
2227 check_signature
= r
!= 0;
2229 check_signature
= false;
2231 if (check_signature
) {
2233 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
2234 /* First, if we have support for signed keys in the kernel, then try that first. */
2235 r
= sym_crypt_activate_by_signed_key(
2239 verity
->root_hash_size
,
2240 verity
->root_hash_sig
,
2241 verity
->root_hash_sig_size
,
2242 CRYPT_ACTIVATE_READONLY
);
2246 log_debug("Validation of dm-verity signature failed via the kernel, trying userspace validation instead.");
2248 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.",
2249 program_invocation_short_name
);
2252 /* So this didn't work via the kernel, then let's try userspace validation instead. If that
2253 * works we'll try to activate without telling the kernel the signature. */
2255 r
= validate_signature_userspace(verity
);
2259 return log_debug_errno(SYNTHETIC_ERRNO(ENOKEY
),
2260 "Activation of signed Verity volume worked neither via the kernel nor in userspace, can't activate.");
2263 return sym_crypt_activate_by_volume_key(
2267 verity
->root_hash_size
,
2268 CRYPT_ACTIVATE_READONLY
);
2271 static usec_t
verity_timeout(void) {
2272 usec_t t
= 100 * USEC_PER_MSEC
;
2276 /* On slower machines, like non-KVM vm, setting up device may take a long time.
2277 * Let's make the timeout configurable. */
2279 e
= getenv("SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC");
2283 r
= parse_sec(e
, &t
);
2286 "Failed to parse timeout specified in $SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC, "
2287 "using the default timeout (%s).",
2288 FORMAT_TIMESPAN(t
, USEC_PER_MSEC
));
2293 static int verity_partition(
2294 PartitionDesignator designator
,
2295 DissectedPartition
*m
,
2296 DissectedPartition
*v
,
2297 const VeritySettings
*verity
,
2298 DissectImageFlags flags
,
2299 DecryptedImage
*d
) {
2301 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2302 _cleanup_(dm_deferred_remove_cleanp
) char *restore_deferred_remove
= NULL
;
2303 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
2304 _cleanup_close_
int mount_node_fd
= -EBADF
;
2308 assert(v
|| (verity
&& verity
->data_path
));
2310 if (!verity
|| !verity
->root_hash
)
2312 if (!((verity
->designator
< 0 && designator
== PARTITION_ROOT
) ||
2313 (verity
->designator
== designator
)))
2316 if (!m
->found
|| !m
->node
|| !m
->fstype
)
2318 if (!verity
->data_path
) {
2319 if (!v
->found
|| !v
->node
|| !v
->fstype
)
2322 if (!streq(v
->fstype
, "DM_verity_hash"))
2326 r
= dlopen_cryptsetup();
2330 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
2331 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
2332 _cleanup_free_
char *root_hash_encoded
= NULL
;
2334 root_hash_encoded
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2335 if (!root_hash_encoded
)
2338 r
= make_dm_name_and_node(root_hash_encoded
, "-verity", &name
, &node
);
2340 r
= make_dm_name_and_node(m
->node
, "-verity", &name
, &node
);
2344 r
= sym_crypt_init(&cd
, verity
->data_path
?: v
->node
);
2348 cryptsetup_enable_logging(cd
);
2350 r
= sym_crypt_load(cd
, CRYPT_VERITY
, NULL
);
2354 r
= sym_crypt_set_data_device(cd
, m
->node
);
2358 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
2361 /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
2362 * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
2363 * retry a few times before giving up. */
2364 for (unsigned i
= 0; i
< N_DEVICE_NODE_LIST_ATTEMPTS
; i
++) {
2365 _cleanup_(sym_crypt_freep
) struct crypt_device
*existing_cd
= NULL
;
2366 _cleanup_close_
int fd
= -EBADF
;
2368 /* First, check if the device already exists. */
2369 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2370 if (fd
< 0 && !ERRNO_IS_DEVICE_ABSENT(errno
))
2371 return log_debug_errno(errno
, "Failed to open verity device %s: %m", node
);
2373 goto check
; /* The device already exists. Let's check it. */
2375 /* The symlink to the device node does not exist yet. Assume not activated, and let's activate it. */
2376 r
= do_crypt_activate_verity(cd
, name
, verity
);
2378 goto try_open
; /* The device is activated. Let's open it. */
2379 /* libdevmapper can return EINVAL when the device is already in the activation stage.
2380 * There's no way to distinguish this situation from a genuine error due to invalid
2381 * parameters, so immediately fall back to activating the device with a unique name.
2382 * Improvements in libcrypsetup can ensure this never happens:
2383 * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
2384 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2386 if (r
== -ENODEV
) /* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */
2389 -EEXIST
, /* Volume has already been opened and ready to be used. */
2390 -EBUSY
/* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */))
2391 return log_debug_errno(r
, "Failed to activate verity device %s: %m", node
);
2394 if (!restore_deferred_remove
){
2395 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
2396 r
= dm_deferred_remove_cancel(name
);
2397 /* -EBUSY and -ENXIO: the device has already been removed or being removed. We cannot
2398 * use the device, try to open again. See target_message() in drivers/md/dm-ioctl.c
2399 * and dm_cancel_deferred_remove() in drivers/md/dm.c */
2400 if (IN_SET(r
, -EBUSY
, -ENXIO
))
2403 return log_debug_errno(r
, "Failed to disable automated deferred removal for verity device %s: %m", node
);
2405 restore_deferred_remove
= strdup(name
);
2406 if (!restore_deferred_remove
)
2407 return log_oom_debug();
2410 r
= verity_can_reuse(verity
, name
, &existing_cd
);
2411 /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
2412 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2415 -ENOENT
, /* Removed?? */
2416 -EBUSY
, /* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */
2417 -ENODEV
/* Volume is being opened but not ready, crypt_init_by_name() would fail, try to open again. */ ))
2420 return log_debug_errno(r
, "Failed to check if existing verity device %s can be reused: %m", node
);
2423 /* devmapper might say that the device exists, but the devlink might not yet have been
2424 * created. Check and wait for the udev event in that case. */
2425 r
= device_wait_for_devlink(node
, "block", verity_timeout(), NULL
);
2426 /* Fallback to activation with a unique device if it's taking too long */
2427 if (r
== -ETIMEDOUT
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2430 return log_debug_errno(r
, "Failed to wait device node symlink %s: %m", node
);
2435 /* Now, the device is activated and devlink is created. Let's open it. */
2436 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2438 if (!ERRNO_IS_DEVICE_ABSENT(errno
))
2439 return log_debug_errno(errno
, "Failed to open verity device %s: %m", node
);
2441 /* The device has already been removed?? */
2446 mount_node_fd
= TAKE_FD(fd
);
2448 crypt_free_and_replace(cd
, existing_cd
);
2453 /* Device is being removed by another process. Let's wait for a while. */
2454 (void) usleep(2 * USEC_PER_MSEC
);
2457 /* All trials failed or a conflicting verity device exists. Let's try to activate with a unique name. */
2458 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
2459 /* Before trying to activate with unique name, we need to free crypt_device object.
2460 * Otherwise, we get error from libcryptsetup like the following:
2462 * systemd[1234]: Cannot use device /dev/loop5 which is in use (already mapped or mounted).
2467 return verity_partition(designator
, m
, v
, verity
, flags
& ~DISSECT_IMAGE_VERITY_SHARE
, d
);
2470 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
), "All attempts to activate verity device %s failed.", name
);
2473 /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
2474 restore_deferred_remove
= mfree(restore_deferred_remove
);
2476 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
2477 .name
= TAKE_PTR(name
),
2478 .device
= TAKE_PTR(cd
),
2481 m
->decrypted_node
= TAKE_PTR(node
);
2482 close_and_replace(m
->mount_node_fd
, mount_node_fd
);
2488 int dissected_image_decrypt(
2490 const char *passphrase
,
2491 const VeritySettings
*verity
,
2492 DissectImageFlags flags
) {
2494 #if HAVE_LIBCRYPTSETUP
2495 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
2500 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
2504 * = 0 → There was nothing to decrypt
2505 * > 0 → Decrypted successfully
2506 * -ENOKEY → There's something to decrypt but no key was supplied
2507 * -EKEYREJECTED → Passed key was not correct
2510 if (verity
&& verity
->root_hash
&& verity
->root_hash_size
< sizeof(sd_id128_t
))
2513 if (!m
->encrypted
&& !m
->verity_ready
)
2516 #if HAVE_LIBCRYPTSETUP
2517 r
= decrypted_image_new(&d
);
2521 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
2522 DissectedPartition
*p
= m
->partitions
+ i
;
2523 PartitionDesignator k
;
2528 r
= decrypt_partition(p
, passphrase
, flags
, d
);
2532 k
= partition_verity_of(i
);
2534 r
= verity_partition(i
, p
, m
->partitions
+ k
, verity
, flags
| DISSECT_IMAGE_VERITY_SHARE
, d
);
2539 if (!p
->decrypted_fstype
&& p
->mount_node_fd
>= 0 && p
->decrypted_node
) {
2540 r
= probe_filesystem_full(p
->mount_node_fd
, p
->decrypted_node
, 0, UINT64_MAX
, &p
->decrypted_fstype
);
2541 if (r
< 0 && r
!= -EUCLEAN
)
2546 m
->decrypted_image
= TAKE_PTR(d
);
2554 int dissected_image_decrypt_interactively(
2556 const char *passphrase
,
2557 const VeritySettings
*verity
,
2558 DissectImageFlags flags
) {
2560 _cleanup_strv_free_erase_
char **z
= NULL
;
2567 r
= dissected_image_decrypt(m
, passphrase
, verity
, flags
);
2570 if (r
== -EKEYREJECTED
)
2571 log_error_errno(r
, "Incorrect passphrase, try again!");
2572 else if (r
!= -ENOKEY
)
2573 return log_error_errno(r
, "Failed to decrypt image: %m");
2576 return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED
),
2577 "Too many retries.");
2581 r
= ask_password_auto("Please enter image passphrase:", NULL
, "dissect", "dissect", "dissect.passphrase", USEC_INFINITY
, 0, &z
);
2583 return log_error_errno(r
, "Failed to query for passphrase: %m");
2589 static int decrypted_image_relinquish(DecryptedImage
*d
) {
2592 /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a
2593 * boolean so that we don't clean it up ourselves either anymore */
2595 #if HAVE_LIBCRYPTSETUP
2598 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
2599 DecryptedPartition
*p
= d
->decrypted
+ i
;
2601 if (p
->relinquished
)
2604 r
= sym_crypt_deactivate_by_name(NULL
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
2606 return log_debug_errno(r
, "Failed to mark %s for auto-removal: %m", p
->name
);
2608 p
->relinquished
= true;
2615 int dissected_image_relinquish(DissectedImage
*m
) {
2620 if (m
->decrypted_image
) {
2621 r
= decrypted_image_relinquish(m
->decrypted_image
);
2627 loop_device_relinquish(m
->loop
);
2632 static char *build_auxiliary_path(const char *image
, const char *suffix
) {
2639 e
= endswith(image
, ".raw");
2641 return strjoin(e
, suffix
);
2643 n
= new(char, e
- image
+ strlen(suffix
) + 1);
2647 strcpy(mempcpy(n
, image
, e
- image
), suffix
);
2651 void verity_settings_done(VeritySettings
*v
) {
2654 v
->root_hash
= mfree(v
->root_hash
);
2655 v
->root_hash_size
= 0;
2657 v
->root_hash_sig
= mfree(v
->root_hash_sig
);
2658 v
->root_hash_sig_size
= 0;
2660 v
->data_path
= mfree(v
->data_path
);
2663 int verity_settings_load(
2664 VeritySettings
*verity
,
2666 const char *root_hash_path
,
2667 const char *root_hash_sig_path
) {
2669 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
2670 size_t root_hash_size
= 0, root_hash_sig_size
= 0;
2671 _cleanup_free_
char *verity_data_path
= NULL
;
2672 PartitionDesignator designator
;
2677 assert(verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
2679 /* If we are asked to load the root hash for a device node, exit early */
2680 if (is_device_path(image
))
2683 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIDECAR");
2684 if (r
< 0 && r
!= -ENXIO
)
2685 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIDECAR, ignoring: %m");
2689 designator
= verity
->designator
;
2691 /* We only fill in what isn't already filled in */
2693 if (!verity
->root_hash
) {
2694 _cleanup_free_
char *text
= NULL
;
2696 if (root_hash_path
) {
2697 /* If explicitly specified it takes precedence */
2698 r
= read_one_line_file(root_hash_path
, &text
);
2703 designator
= PARTITION_ROOT
;
2705 /* Otherwise look for xattr and separate file, and first for the data for root and if
2706 * that doesn't exist for /usr */
2708 if (designator
< 0 || designator
== PARTITION_ROOT
) {
2709 r
= getxattr_malloc(image
, "user.verity.roothash", &text
);
2711 _cleanup_free_
char *p
= NULL
;
2713 if (r
!= -ENOENT
&& !ERRNO_IS_XATTR_ABSENT(r
))
2716 p
= build_auxiliary_path(image
, ".roothash");
2720 r
= read_one_line_file(p
, &text
);
2721 if (r
< 0 && r
!= -ENOENT
)
2726 designator
= PARTITION_ROOT
;
2729 if (!text
&& (designator
< 0 || designator
== PARTITION_USR
)) {
2730 /* So in the "roothash" xattr/file name above the "root" of course primarily
2731 * refers to the root of the Verity Merkle tree. But coincidentally it also
2732 * is the hash for the *root* file system, i.e. the "root" neatly refers to
2733 * two distinct concepts called "root". Taking benefit of this happy
2734 * coincidence we call the file with the root hash for the /usr/ file system
2735 * `usrhash`, because `usrroothash` or `rootusrhash` would just be too
2736 * confusing. We thus drop the reference to the root of the Merkle tree, and
2737 * just indicate which file system it's about. */
2738 r
= getxattr_malloc(image
, "user.verity.usrhash", &text
);
2740 _cleanup_free_
char *p
= NULL
;
2742 if (r
!= -ENOENT
&& !ERRNO_IS_XATTR_ABSENT(r
))
2745 p
= build_auxiliary_path(image
, ".usrhash");
2749 r
= read_one_line_file(p
, &text
);
2750 if (r
< 0 && r
!= -ENOENT
)
2755 designator
= PARTITION_USR
;
2760 r
= unhexmem(text
, strlen(text
), &root_hash
, &root_hash_size
);
2763 if (root_hash_size
< sizeof(sd_id128_t
))
2768 if ((root_hash
|| verity
->root_hash
) && !verity
->root_hash_sig
) {
2769 if (root_hash_sig_path
) {
2770 r
= read_full_file(root_hash_sig_path
, (char**) &root_hash_sig
, &root_hash_sig_size
);
2771 if (r
< 0 && r
!= -ENOENT
)
2775 designator
= PARTITION_ROOT
;
2777 if (designator
< 0 || designator
== PARTITION_ROOT
) {
2778 _cleanup_free_
char *p
= NULL
;
2780 /* Follow naming convention recommended by the relevant RFC:
2781 * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
2782 p
= build_auxiliary_path(image
, ".roothash.p7s");
2786 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
2787 if (r
< 0 && r
!= -ENOENT
)
2790 designator
= PARTITION_ROOT
;
2793 if (!root_hash_sig
&& (designator
< 0 || designator
== PARTITION_USR
)) {
2794 _cleanup_free_
char *p
= NULL
;
2796 p
= build_auxiliary_path(image
, ".usrhash.p7s");
2800 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
2801 if (r
< 0 && r
!= -ENOENT
)
2804 designator
= PARTITION_USR
;
2808 if (root_hash_sig
&& root_hash_sig_size
== 0) /* refuse empty size signatures */
2812 if (!verity
->data_path
) {
2813 _cleanup_free_
char *p
= NULL
;
2815 p
= build_auxiliary_path(image
, ".verity");
2819 if (access(p
, F_OK
) < 0) {
2820 if (errno
!= ENOENT
)
2823 verity_data_path
= TAKE_PTR(p
);
2827 verity
->root_hash
= TAKE_PTR(root_hash
);
2828 verity
->root_hash_size
= root_hash_size
;
2831 if (root_hash_sig
) {
2832 verity
->root_hash_sig
= TAKE_PTR(root_hash_sig
);
2833 verity
->root_hash_sig_size
= root_hash_sig_size
;
2836 if (verity_data_path
)
2837 verity
->data_path
= TAKE_PTR(verity_data_path
);
2839 if (verity
->designator
< 0)
2840 verity
->designator
= designator
;
2845 int dissected_image_load_verity_sig_partition(
2848 VeritySettings
*verity
) {
2850 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
2851 _cleanup_(json_variant_unrefp
) JsonVariant
*v
= NULL
;
2852 size_t root_hash_size
, root_hash_sig_size
;
2853 _cleanup_free_
char *buf
= NULL
;
2854 PartitionDesignator d
;
2855 DissectedPartition
*p
;
2856 JsonVariant
*rh
, *sig
;
2865 if (verity
->root_hash
&& verity
->root_hash_sig
) /* Already loaded? */
2868 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_EMBEDDED");
2869 if (r
< 0 && r
!= -ENXIO
)
2870 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_EMBEDDED, ignoring: %m");
2874 d
= partition_verity_sig_of(verity
->designator
< 0 ? PARTITION_ROOT
: verity
->designator
);
2877 p
= m
->partitions
+ d
;
2880 if (p
->offset
== UINT64_MAX
|| p
->size
== UINT64_MAX
)
2883 if (p
->size
> 4*1024*1024) /* Signature data cannot possible be larger than 4M, refuse that */
2886 buf
= new(char, p
->size
+1);
2890 n
= pread(fd
, buf
, p
->size
, p
->offset
);
2893 if ((uint64_t) n
!= p
->size
)
2896 e
= memchr(buf
, 0, p
->size
);
2898 /* If we found a NUL byte then the rest of the data must be NUL too */
2899 if (!memeqzero(e
, p
->size
- (e
- buf
)))
2900 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature data contains embedded NUL byte.");
2904 r
= json_parse(buf
, 0, &v
, NULL
, NULL
);
2906 return log_debug_errno(r
, "Failed to parse signature JSON data: %m");
2908 rh
= json_variant_by_key(v
, "rootHash");
2910 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'rootHash' field.");
2911 if (!json_variant_is_string(rh
))
2912 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'rootHash' field of signature JSON object is not a string.");
2914 r
= unhexmem(json_variant_string(rh
), SIZE_MAX
, &root_hash
, &root_hash_size
);
2916 return log_debug_errno(r
, "Failed to parse root hash field: %m");
2918 /* Check if specified root hash matches if it is specified */
2919 if (verity
->root_hash
&&
2920 memcmp_nn(verity
->root_hash
, verity
->root_hash_size
, root_hash
, root_hash_size
) != 0) {
2921 _cleanup_free_
char *a
= NULL
, *b
= NULL
;
2923 a
= hexmem(root_hash
, root_hash_size
);
2924 b
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2926 return log_debug_errno(r
, "Root hash in signature JSON data (%s) doesn't match configured hash (%s).", strna(a
), strna(b
));
2929 sig
= json_variant_by_key(v
, "signature");
2931 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'signature' field.");
2932 if (!json_variant_is_string(sig
))
2933 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'signature' field of signature JSON object is not a string.");
2935 r
= unbase64mem(json_variant_string(sig
), SIZE_MAX
, &root_hash_sig
, &root_hash_sig_size
);
2937 return log_debug_errno(r
, "Failed to parse signature field: %m");
2939 free_and_replace(verity
->root_hash
, root_hash
);
2940 verity
->root_hash_size
= root_hash_size
;
2942 free_and_replace(verity
->root_hash_sig
, root_hash_sig
);
2943 verity
->root_hash_sig_size
= root_hash_sig_size
;
2948 int dissected_image_acquire_metadata(DissectedImage
*m
, DissectImageFlags extra_flags
) {
2955 META_INITRD_RELEASE
,
2956 META_EXTENSION_RELEASE
,
2957 META_HAS_INIT_SYSTEM
,
2961 static const char *const paths
[_META_MAX
] = {
2962 [META_HOSTNAME
] = "/etc/hostname\0",
2963 [META_MACHINE_ID
] = "/etc/machine-id\0",
2964 [META_MACHINE_INFO
] = "/etc/machine-info\0",
2965 [META_OS_RELEASE
] = ("/etc/os-release\0"
2966 "/usr/lib/os-release\0"),
2967 [META_INITRD_RELEASE
] = ("/etc/initrd-release\0"
2968 "/usr/lib/initrd-release\0"),
2969 [META_EXTENSION_RELEASE
] = "extension-release\0", /* Used only for logging. */
2970 [META_HAS_INIT_SYSTEM
] = "has-init-system\0", /* ditto */
2973 _cleanup_strv_free_
char **machine_info
= NULL
, **os_release
= NULL
, **initrd_release
= NULL
, **extension_release
= NULL
;
2974 _cleanup_close_pair_
int error_pipe
[2] = PIPE_EBADF
;
2975 _cleanup_(rmdir_and_freep
) char *t
= NULL
;
2976 _cleanup_(sigkill_waitp
) pid_t child
= 0;
2977 sd_id128_t machine_id
= SD_ID128_NULL
;
2978 _cleanup_free_
char *hostname
= NULL
;
2979 unsigned n_meta_initialized
= 0;
2980 int fds
[2 * _META_MAX
], r
, v
;
2981 int has_init_system
= -1;
2984 BLOCK_SIGNALS(SIGCHLD
);
2988 for (; n_meta_initialized
< _META_MAX
; n_meta_initialized
++) {
2989 if (!paths
[n_meta_initialized
]) {
2990 fds
[2*n_meta_initialized
] = fds
[2*n_meta_initialized
+1] = -EBADF
;
2994 if (pipe2(fds
+ 2*n_meta_initialized
, O_CLOEXEC
) < 0) {
3000 r
= mkdtemp_malloc("/tmp/dissect-XXXXXX", &t
);
3004 if (pipe2(error_pipe
, O_CLOEXEC
) < 0) {
3009 r
= safe_fork("(sd-dissect)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_NEW_MOUNTNS
|FORK_MOUNTNS_SLAVE
, &child
);
3013 /* Child in a new mount namespace */
3014 error_pipe
[0] = safe_close(error_pipe
[0]);
3016 r
= dissected_image_mount(
3022 DISSECT_IMAGE_READ_ONLY
|
3023 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
3024 DISSECT_IMAGE_USR_NO_ROOT
);
3026 log_debug_errno(r
, "Failed to mount dissected image: %m");
3030 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
3031 _cleanup_close_
int fd
= -ENOENT
;
3036 fds
[2*k
] = safe_close(fds
[2*k
]);
3040 case META_EXTENSION_RELEASE
:
3041 /* As per the os-release spec, if the image is an extension it will have a file
3042 * named after the image name in extension-release.d/ - we use the image name
3043 * and try to resolve it with the extension-release helpers, as sometimes
3044 * the image names are mangled on deployment and do not match anymore.
3045 * Unlike other paths this is not fixed, and the image name
3046 * can be mangled on deployment, so by calling into the helper
3047 * we allow a fallback that matches on the first extension-release
3048 * file found in the directory, if one named after the image cannot
3049 * be found first. */
3050 r
= open_extension_release(t
, m
->image_name
, /* relax_extension_release_check= */ false, NULL
, &fd
);
3052 fd
= r
; /* Propagate the error. */
3055 case META_HAS_INIT_SYSTEM
: {
3058 FOREACH_STRING(init
,
3059 "/usr/lib/systemd/systemd", /* systemd on /usr merged system */
3060 "/lib/systemd/systemd", /* systemd on /usr non-merged systems */
3061 "/sbin/init") { /* traditional path the Linux kernel invokes */
3063 r
= chase_symlinks(init
, t
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
3066 log_debug_errno(r
, "Failed to resolve %s, ignoring: %m", init
);
3073 r
= loop_write(fds
[2*k
+1], &found
, sizeof(found
), false);
3081 NULSTR_FOREACH(p
, paths
[k
]) {
3082 fd
= chase_symlinks_and_open(p
, t
, CHASE_PREFIX_ROOT
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
, NULL
);
3089 log_debug_errno(fd
, "Failed to read %s file of image, ignoring: %m", paths
[k
]);
3090 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3094 r
= copy_bytes(fd
, fds
[2*k
+1], UINT64_MAX
, 0);
3098 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3101 _exit(EXIT_SUCCESS
);
3104 /* Let parent know the error */
3105 (void) write(error_pipe
[1], &r
, sizeof(r
));
3106 _exit(EXIT_FAILURE
);
3109 error_pipe
[1] = safe_close(error_pipe
[1]);
3111 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
3112 _cleanup_fclose_
FILE *f
= NULL
;
3117 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3119 f
= take_fdopen(&fds
[2*k
], "r");
3128 r
= read_etc_hostname_stream(f
, &hostname
);
3130 log_debug_errno(r
, "Failed to read /etc/hostname of image: %m");
3134 case META_MACHINE_ID
: {
3135 _cleanup_free_
char *line
= NULL
;
3137 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3139 log_debug_errno(r
, "Failed to read /etc/machine-id of image: %m");
3141 r
= sd_id128_from_string(line
, &machine_id
);
3143 log_debug_errno(r
, "Image contains invalid /etc/machine-id: %s", line
);
3145 log_debug("/etc/machine-id file of image is empty.");
3146 else if (streq(line
, "uninitialized"))
3147 log_debug("/etc/machine-id file of image is uninitialized (likely aborted first boot).");
3149 log_debug("/etc/machine-id file of image has unexpected length %i.", r
);
3154 case META_MACHINE_INFO
:
3155 r
= load_env_file_pairs(f
, "machine-info", &machine_info
);
3157 log_debug_errno(r
, "Failed to read /etc/machine-info of image: %m");
3161 case META_OS_RELEASE
:
3162 r
= load_env_file_pairs(f
, "os-release", &os_release
);
3164 log_debug_errno(r
, "Failed to read OS release file of image: %m");
3168 case META_INITRD_RELEASE
:
3169 r
= load_env_file_pairs(f
, "initrd-release", &initrd_release
);
3171 log_debug_errno(r
, "Failed to read initrd release file of image: %m");
3175 case META_EXTENSION_RELEASE
:
3176 r
= load_env_file_pairs(f
, "extension-release", &extension_release
);
3178 log_debug_errno(r
, "Failed to read extension release file of image: %m");
3182 case META_HAS_INIT_SYSTEM
: {
3187 nr
= fread(&b
, 1, sizeof(b
), f
);
3188 if (nr
!= sizeof(b
))
3189 log_debug_errno(errno_or_else(EIO
), "Failed to read has-init-system boolean: %m");
3191 has_init_system
= b
;
3197 r
= wait_for_terminate_and_check("(sd-dissect)", child
, 0);
3202 n
= read(error_pipe
[0], &v
, sizeof(v
));
3206 return v
; /* propagate error sent to us from child */
3210 if (r
!= EXIT_SUCCESS
)
3213 free_and_replace(m
->hostname
, hostname
);
3214 m
->machine_id
= machine_id
;
3215 strv_free_and_replace(m
->machine_info
, machine_info
);
3216 strv_free_and_replace(m
->os_release
, os_release
);
3217 strv_free_and_replace(m
->initrd_release
, initrd_release
);
3218 strv_free_and_replace(m
->extension_release
, extension_release
);
3219 m
->has_init_system
= has_init_system
;
3222 for (unsigned k
= 0; k
< n_meta_initialized
; k
++)
3223 safe_close_pair(fds
+ 2*k
);
3228 Architecture
dissected_image_architecture(DissectedImage
*img
) {
3231 if (img
->partitions
[PARTITION_ROOT
].found
&&
3232 img
->partitions
[PARTITION_ROOT
].architecture
>= 0)
3233 return img
->partitions
[PARTITION_ROOT
].architecture
;
3235 if (img
->partitions
[PARTITION_USR
].found
&&
3236 img
->partitions
[PARTITION_USR
].architecture
>= 0)
3237 return img
->partitions
[PARTITION_USR
].architecture
;
3239 return _ARCHITECTURE_INVALID
;
3242 int dissect_loop_device(
3244 const VeritySettings
*verity
,
3245 const MountOptions
*mount_options
,
3246 DissectImageFlags flags
,
3247 DissectedImage
**ret
) {
3250 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
3256 r
= dissected_image_new(loop
->backing_file
?: loop
->node
, &m
);
3260 m
->loop
= loop_device_ref(loop
);
3261 m
->sector_size
= m
->loop
->sector_size
;
3263 r
= dissect_image(m
, loop
->fd
, loop
->node
, verity
, mount_options
, flags
);
3274 int dissect_loop_device_and_warn(
3276 const VeritySettings
*verity
,
3277 const MountOptions
*mount_options
,
3278 DissectImageFlags flags
,
3279 DissectedImage
**ret
) {
3285 assert(loop
->fd
>= 0);
3287 name
= ASSERT_PTR(loop
->backing_file
?: loop
->node
);
3289 r
= dissect_loop_device(loop
, verity
, mount_options
, flags
, ret
);
3293 return log_error_errno(r
, "Dissecting images is not supported, compiled without blkid support.");
3296 return log_error_errno(r
, "%s: Couldn't identify a suitable partition table or file system.", name
);
3299 return log_error_errno(r
, "%s: The image does not pass validation.", name
);
3301 case -EADDRNOTAVAIL
:
3302 return log_error_errno(r
, "%s: No root partition for specified root hash found.", name
);
3305 return log_error_errno(r
, "%s: Multiple suitable root partitions found in image.", name
);
3308 return log_error_errno(r
, "%s: No suitable root partition found in image.", name
);
3310 case -EPROTONOSUPPORT
:
3311 return log_error_errno(r
, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name
);
3314 return log_error_errno(r
, "%s: Image is not a block device.", name
);
3317 return log_error_errno(r
,
3318 "Combining partitioned images (such as '%s') with external Verity data (such as '%s') not supported. "
3319 "(Consider setting $SYSTEMD_DISSECT_VERITY_SIDECAR=0 to disable automatic discovery of external Verity data.)",
3320 name
, strna(verity
? verity
->data_path
: NULL
));
3324 return log_error_errno(r
, "Failed to dissect image '%s': %m", name
);
3330 bool dissected_image_verity_candidate(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3333 /* Checks if this partition could theoretically do Verity. For non-partitioned images this only works
3334 * if there's an external verity file supplied, for which we can consult .has_verity. For partitioned
3335 * images we only check the partition type.
3337 * This call is used to decide whether to suppress or show a verity column in tabular output of the
3340 if (image
->single_file_system
)
3341 return partition_designator
== PARTITION_ROOT
&& image
->has_verity
;
3343 return partition_verity_of(partition_designator
) >= 0;
3346 bool dissected_image_verity_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3347 PartitionDesignator k
;
3351 /* Checks if this partition has verity data available that we can activate. For non-partitioned this
3352 * works for the root partition, for others only if the associated verity partition was found. */
3354 if (!image
->verity_ready
)
3357 if (image
->single_file_system
)
3358 return partition_designator
== PARTITION_ROOT
;
3360 k
= partition_verity_of(partition_designator
);
3361 return k
>= 0 && image
->partitions
[k
].found
;
3364 bool dissected_image_verity_sig_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3365 PartitionDesignator k
;
3369 /* Checks if this partition has verity signature data available that we can use. */
3371 if (!image
->verity_sig_ready
)
3374 if (image
->single_file_system
)
3375 return partition_designator
== PARTITION_ROOT
;
3377 k
= partition_verity_sig_of(partition_designator
);
3378 return k
>= 0 && image
->partitions
[k
].found
;
3381 MountOptions
* mount_options_free_all(MountOptions
*options
) {
3384 while ((m
= options
)) {
3385 LIST_REMOVE(mount_options
, options
, m
);
3393 const char* mount_options_from_designator(const MountOptions
*options
, PartitionDesignator designator
) {
3394 LIST_FOREACH(mount_options
, m
, options
)
3395 if (designator
== m
->partition_designator
&& !isempty(m
->options
))
3401 int mount_image_privately_interactively(
3403 DissectImageFlags flags
,
3404 char **ret_directory
,
3406 LoopDevice
**ret_loop_device
) {
3408 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3409 _cleanup_(loop_device_unrefp
) LoopDevice
*d
= NULL
;
3410 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3411 _cleanup_(rmdir_and_freep
) char *created_dir
= NULL
;
3412 _cleanup_free_
char *temp
= NULL
;
3415 /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
3416 * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
3420 assert(ret_directory
);
3421 assert(ret_loop_device
);
3423 /* We intend to mount this right-away, hence add the partitions if needed and pin them. */
3424 flags
|= DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
3425 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
3427 r
= verity_settings_load(&verity
, image
, NULL
, NULL
);
3429 return log_error_errno(r
, "Failed to load root hash data: %m");
3431 r
= tempfn_random_child(NULL
, program_invocation_short_name
, &temp
);
3433 return log_error_errno(r
, "Failed to generate temporary mount directory: %m");
3435 r
= loop_device_make_by_path(
3437 FLAGS_SET(flags
, DISSECT_IMAGE_DEVICE_READ_ONLY
) ? O_RDONLY
: O_RDWR
,
3438 /* sector_size= */ UINT32_MAX
,
3439 FLAGS_SET(flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
3443 return log_error_errno(r
, "Failed to set up loopback device for %s: %m", image
);
3445 r
= dissect_loop_device_and_warn(d
, &verity
, NULL
, flags
, &dissected_image
);
3449 r
= dissected_image_load_verity_sig_partition(dissected_image
, d
->fd
, &verity
);
3453 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, &verity
, flags
);
3457 r
= detach_mount_namespace();
3459 return log_error_errno(r
, "Failed to detach mount namespace: %m");
3461 r
= mkdir_p(temp
, 0700);
3463 return log_error_errno(r
, "Failed to create mount point: %m");
3465 created_dir
= TAKE_PTR(temp
);
3467 r
= dissected_image_mount_and_warn(dissected_image
, created_dir
, UID_INVALID
, UID_INVALID
, flags
);
3471 r
= loop_device_flock(d
, LOCK_UN
);
3475 r
= dissected_image_relinquish(dissected_image
);
3477 return log_error_errno(r
, "Failed to relinquish DM and loopback block devices: %m");
3480 _cleanup_close_
int dir_fd
= -EBADF
;
3482 dir_fd
= open(created_dir
, O_CLOEXEC
|O_DIRECTORY
);
3484 return log_error_errno(errno
, "Failed to open mount point directory: %m");
3486 *ret_dir_fd
= TAKE_FD(dir_fd
);
3489 *ret_directory
= TAKE_PTR(created_dir
);
3490 *ret_loop_device
= TAKE_PTR(d
);
3495 static bool mount_options_relax_extension_release_checks(const MountOptions
*options
) {
3499 return string_contains_word(mount_options_from_designator(options
, PARTITION_ROOT
), ",", "x-systemd.relax-extension-release-check") ||
3500 string_contains_word(mount_options_from_designator(options
, PARTITION_USR
), ",", "x-systemd.relax-extension-release-check") ||
3501 string_contains_word(options
->options
, ",", "x-systemd.relax-extension-release-check");
3504 int verity_dissect_and_mount(
3508 const MountOptions
*options
,
3509 const char *required_host_os_release_id
,
3510 const char *required_host_os_release_version_id
,
3511 const char *required_host_os_release_sysext_level
,
3512 const char *required_sysext_scope
) {
3514 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
3515 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3516 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3517 DissectImageFlags dissect_image_flags
;
3518 bool relax_extension_release_check
;
3524 relax_extension_release_check
= mount_options_relax_extension_release_checks(options
);
3526 /* We might get an FD for the image, but we use the original path to look for the dm-verity files */
3527 r
= verity_settings_load(&verity
, src
, NULL
, NULL
);
3529 return log_debug_errno(r
, "Failed to load root hash: %m");
3531 dissect_image_flags
= (verity
.data_path
? DISSECT_IMAGE_NO_PARTITION_TABLE
: 0) |
3532 (relax_extension_release_check
? DISSECT_IMAGE_RELAX_SYSEXT_CHECK
: 0) |
3533 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
3534 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
3536 /* Note that we don't use loop_device_make here, as the FD is most likely O_PATH which would not be
3537 * accepted by LOOP_CONFIGURE, so just let loop_device_make_by_path reopen it as a regular FD. */
3538 r
= loop_device_make_by_path(
3539 src_fd
>= 0 ? FORMAT_PROC_FD_PATH(src_fd
) : src
,
3540 /* open_flags= */ -1,
3541 /* sector_size= */ UINT32_MAX
,
3542 verity
.data_path
? 0 : LO_FLAGS_PARTSCAN
,
3546 return log_debug_errno(r
, "Failed to create loop device for image: %m");
3548 r
= dissect_loop_device(
3552 dissect_image_flags
,
3554 /* No partition table? Might be a single-filesystem image, try again */
3555 if (!verity
.data_path
&& r
== -ENOPKG
)
3556 r
= dissect_loop_device(
3560 dissect_image_flags
| DISSECT_IMAGE_NO_PARTITION_TABLE
,
3563 return log_debug_errno(r
, "Failed to dissect image: %m");
3565 r
= dissected_image_load_verity_sig_partition(dissected_image
, loop_device
->fd
, &verity
);
3569 r
= dissected_image_decrypt(
3573 dissect_image_flags
);
3575 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
3577 r
= mkdir_p_label(dest
, 0755);
3579 return log_debug_errno(r
, "Failed to create destination directory %s: %m", dest
);
3580 r
= umount_recursive(dest
, 0);
3582 return log_debug_errno(r
, "Failed to umount under destination directory %s: %m", dest
);
3584 r
= dissected_image_mount(dissected_image
, dest
, UID_INVALID
, UID_INVALID
, dissect_image_flags
);
3586 return log_debug_errno(r
, "Failed to mount image: %m");
3588 r
= loop_device_flock(loop_device
, LOCK_UN
);
3590 return log_debug_errno(r
, "Failed to unlock loopback device: %m");
3592 /* If we got os-release values from the caller, then we need to match them with the image's
3593 * extension-release.d/ content. Return -EINVAL if there's any mismatch.
3594 * First, check the distro ID. If that matches, then check the new SYSEXT_LEVEL value if
3595 * available, or else fallback to VERSION_ID. If neither is present (eg: rolling release),
3596 * then a simple match on the ID will be performed. */
3597 if (required_host_os_release_id
) {
3598 _cleanup_strv_free_
char **extension_release
= NULL
;
3600 assert(!isempty(required_host_os_release_id
));
3602 r
= load_extension_release_pairs(dest
, dissected_image
->image_name
, relax_extension_release_check
, &extension_release
);
3604 return log_debug_errno(r
, "Failed to parse image %s extension-release metadata: %m", dissected_image
->image_name
);
3606 r
= extension_release_validate(
3607 dissected_image
->image_name
,
3608 required_host_os_release_id
,
3609 required_host_os_release_version_id
,
3610 required_host_os_release_sysext_level
,
3611 required_sysext_scope
,
3614 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE
), "Image %s extension-release metadata does not match the root's", dissected_image
->image_name
);
3616 return log_debug_errno(r
, "Failed to compare image %s extension-release metadata with the root's os-release: %m", dissected_image
->image_name
);
3619 r
= dissected_image_relinquish(dissected_image
);
3621 return log_debug_errno(r
, "Failed to relinquish dissected image: %m");