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 "mkdir-label.h"
54 #include "mount-util.h"
55 #include "mountpoint-util.h"
56 #include "namespace-util.h"
57 #include "nulstr-util.h"
58 #include "openssl-util.h"
60 #include "path-util.h"
61 #include "process-util.h"
62 #include "raw-clone.h"
63 #include "resize-fs.h"
64 #include "signal-util.h"
65 #include "sparse-endian.h"
66 #include "stat-util.h"
67 #include "stdio-util.h"
68 #include "string-table.h"
69 #include "string-util.h"
71 #include "tmpfile-util.h"
72 #include "udev-util.h"
73 #include "user-util.h"
74 #include "xattr-util.h"
76 /* how many times to wait for the device nodes to appear */
77 #define N_DEVICE_NODE_LIST_ATTEMPTS 10
79 int dissect_fstype_ok(const char *fstype
) {
83 /* When we automatically mount file systems, be a bit conservative by default what we are willing to
84 * mount, just as an extra safety net to not mount with badly maintained legacy file system
87 e
= secure_getenv("SYSTEMD_DISSECT_FILE_SYSTEMS");
89 _cleanup_strv_free_
char **l
= NULL
;
91 l
= strv_split(e
, ":");
95 b
= strv_contains(l
, fstype
);
97 b
= STR_IN_SET(fstype
,
107 log_debug("File system type '%s' is not allowed to be mounted as result of automatic dissection.", fstype
);
111 int probe_sector_size(int fd
, uint32_t *ret
) {
120 le64_t alternate_lba
;
121 le64_t first_usable_lba
;
122 le64_t last_usable_lba
;
123 sd_id128_t disk_guid
;
124 le64_t partition_entry_lba
;
125 le32_t number_of_partition_entries
;
126 le32_t size_of_partition_entry
;
127 le32_t partition_entry_array_crc32
;
130 /* Disk images might be for 512B or for 4096 sector sizes, let's try to auto-detect that by searching
131 * for the GPT headers at the relevant byte offsets */
133 assert_cc(sizeof(struct gpt_header
) == 92);
135 /* We expect a sector size in the range 512…4096. The GPT header is located in the second
136 * sector. Hence it could be at byte 512 at the earliest, and at byte 4096 at the latest. And we must
137 * read with granularity of the largest sector size we care about. Which means 8K. */
138 uint8_t sectors
[2 * 4096];
145 n
= pread(fd
, sectors
, sizeof(sectors
), 0);
148 if (n
!= sizeof(sectors
)) /* too short? */
151 /* Let's see if we find the GPT partition header with various expected sector sizes */
152 for (uint32_t sz
= 512; sz
<= 4096; sz
<<= 1) {
153 struct gpt_header
*p
;
155 assert(sizeof(sectors
) >= sz
* 2);
156 p
= (struct gpt_header
*) (sectors
+ sz
);
158 if (memcmp(p
->signature
, (const char[8]) { 'E', 'F', 'I', ' ', 'P', 'A', 'R', 'T' }, 8) != 0)
161 if (le32toh(p
->revision
) != UINT32_C(0x00010000)) /* the only known revision of the spec: 1.0 */
164 if (le32toh(p
->header_size
) < sizeof(struct gpt_header
))
167 if (le32toh(p
->header_size
) > 4096) /* larger than a sector? something is off… */
170 if (le64toh(p
->my_lba
) != 1) /* this sector must claim to be at sector offset 1 */
174 return log_debug_errno(SYNTHETIC_ERRNO(ENOTUNIQ
),
175 "Detected valid partition table at offsets matching multiple sector sizes, refusing.");
181 log_debug("Determined sector size %" PRIu32
" based on discovered partition table.", found
);
183 return 1; /* indicate we *did* find it */
187 log_debug("Couldn't find any partition table to derive sector size of.");
188 *ret
= 512; /* pick the traditional default */
189 return 0; /* indicate we didn't find it */
192 int probe_sector_size_prefer_ioctl(int fd
, uint32_t *ret
) {
198 /* Just like probe_sector_size(), but if we are looking at a block device, will use the already
199 * configured sector size rather than probing by contents */
201 if (fstat(fd
, &st
) < 0)
204 if (S_ISBLK(st
.st_mode
))
205 return blockdev_get_sector_size(fd
, ret
);
207 return probe_sector_size(fd
, ret
);
210 int probe_filesystem_full(
217 /* Try to find device content type and return it in *ret_fstype. If nothing is found,
218 * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and a
219 * different error otherwise. */
222 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
223 _cleanup_free_
char *path_by_fd
= NULL
;
224 _cleanup_close_
int fd_close
= -EBADF
;
228 assert(fd
>= 0 || path
);
232 fd_close
= open(path
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
240 r
= fd_get_path(fd
, &path_by_fd
);
247 if (size
== 0) /* empty size? nothing found! */
250 b
= blkid_new_probe();
255 r
= blkid_probe_set_device(
259 size
== UINT64_MAX
? 0 : size
); /* when blkid sees size=0 it understands "everything". We prefer using UINT64_MAX for that */
261 return errno_or_else(ENOMEM
);
263 blkid_probe_enable_superblocks(b
, 1);
264 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
);
267 r
= blkid_do_safeprobe(b
);
268 if (r
== _BLKID_SAFEPROBE_NOT_FOUND
)
270 if (r
== _BLKID_SAFEPROBE_AMBIGUOUS
)
271 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
),
272 "Results ambiguous for partition %s", path
);
273 if (r
== _BLKID_SAFEPROBE_ERROR
)
274 return log_debug_errno(errno_or_else(EIO
), "Failed to probe partition %s: %m", path
);
276 assert(r
== _BLKID_SAFEPROBE_FOUND
);
278 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
283 log_debug("Probed fstype '%s' on partition %s.", fstype
, path
);
294 log_debug("No type detected on partition %s", path
);
303 static int dissected_image_probe_filesystems(DissectedImage
*m
, int fd
) {
308 /* Fill in file system types if we don't know them yet. */
310 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
311 DissectedPartition
*p
= m
->partitions
+ i
;
317 /* If we have an fd referring to the partition block device, use that. Otherwise go
318 * via the whole block device or backing regular file, and read via offset. */
319 if (p
->mount_node_fd
>= 0)
320 r
= probe_filesystem_full(p
->mount_node_fd
, p
->node
, 0, UINT64_MAX
, &p
->fstype
);
322 r
= probe_filesystem_full(fd
, p
->node
, p
->offset
, p
->size
, &p
->fstype
);
327 if (streq_ptr(p
->fstype
, "crypto_LUKS"))
330 if (p
->fstype
&& fstype_is_ro(p
->fstype
))
340 static void check_partition_flags(
342 unsigned long long pflags
,
343 unsigned long long supported
) {
347 /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
348 pflags
&= ~(supported
|
349 SD_GPT_FLAG_REQUIRED_PARTITION
|
350 SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL
|
351 SD_GPT_FLAG_LEGACY_BIOS_BOOTABLE
);
356 /* If there are other bits set, then log about it, to make things discoverable */
357 for (unsigned i
= 0; i
< sizeof(pflags
) * 8; i
++) {
358 unsigned long long bit
= 1ULL << i
;
359 if (!FLAGS_SET(pflags
, bit
))
362 log_debug("Unexpected partition flag %llu set on %s!", bit
, node
);
368 static int dissected_image_new(const char *path
, DissectedImage
**ret
) {
369 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
370 _cleanup_free_
char *name
= NULL
;
376 _cleanup_free_
char *filename
= NULL
;
378 r
= path_extract_filename(path
, &filename
);
382 r
= raw_strip_suffixes(filename
, &name
);
386 if (!image_name_is_valid(name
)) {
387 log_debug("Image name %s is not valid, ignoring.", strna(name
));
392 m
= new(DissectedImage
, 1);
396 *m
= (DissectedImage
) {
397 .has_init_system
= -1,
398 .image_name
= TAKE_PTR(name
),
401 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++)
402 m
->partitions
[i
] = DISSECTED_PARTITION_NULL
;
409 static void dissected_partition_done(DissectedPartition
*p
) {
415 free(p
->decrypted_fstype
);
416 free(p
->decrypted_node
);
417 free(p
->mount_options
);
418 safe_close(p
->mount_node_fd
);
420 *p
= DISSECTED_PARTITION_NULL
;
424 static int make_partition_devname(
425 const char *whole_devname
,
428 DissectImageFlags flags
,
431 _cleanup_free_
char *s
= NULL
;
434 assert(whole_devname
);
435 assert(nr
!= 0); /* zero is not a valid partition nr */
438 if (!FLAGS_SET(flags
, DISSECT_IMAGE_DISKSEQ_DEVNODE
) || diskseq
== 0) {
440 /* Given a whole block device node name (e.g. /dev/sda or /dev/loop7) generate a partition
441 * device name (e.g. /dev/sda7 or /dev/loop7p5). The rule the kernel uses is simple: if whole
442 * block device node name ends in a digit, then suffix a 'p', followed by the partition
443 * number. Otherwise, just suffix the partition number without any 'p'. */
445 if (nr
< 0) { /* whole disk? */
446 s
= strdup(whole_devname
);
450 size_t l
= strlen(whole_devname
);
451 if (l
< 1) /* underflow check for the subtraction below */
454 bool need_p
= ascii_isdigit(whole_devname
[l
-1]); /* Last char a digit? */
456 if (asprintf(&s
, "%s%s%i", whole_devname
, need_p
? "p" : "", nr
) < 0)
460 if (nr
< 0) /* whole disk? */
461 r
= asprintf(&s
, "/dev/disk/by-diskseq/%" PRIu64
, diskseq
);
463 r
= asprintf(&s
, "/dev/disk/by-diskseq/%" PRIu64
"-part%i", diskseq
, nr
);
472 static int open_partition(
475 const LoopDevice
*loop
) {
477 _cleanup_(sd_device_unrefp
) sd_device
*dev
= NULL
;
478 _cleanup_close_
int fd
= -EBADF
;
485 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
489 /* Check if the block device is a child of (or equivalent to) the originally provided one. */
490 r
= block_device_new_from_fd(fd
, is_partition
? BLOCK_DEVICE_LOOKUP_WHOLE_DISK
: 0, &dev
);
494 r
= sd_device_get_devnum(dev
, &devnum
);
498 if (loop
->devno
!= devnum
)
501 /* Also check diskseq. */
502 if (loop
->diskseq
!= 0) {
505 r
= fd_get_diskseq(fd
, &diskseq
);
509 if (loop
->diskseq
!= diskseq
)
513 log_debug("Opened %s (fd=%i, whole_block_devnum=" DEVNUM_FORMAT_STR
", diskseq=%" PRIu64
").",
514 node
, fd
, DEVNUM_FORMAT_VAL(loop
->devno
), loop
->diskseq
);
518 static int compare_arch(Architecture a
, Architecture b
) {
522 if (a
== native_architecture())
525 if (b
== native_architecture())
528 #ifdef ARCHITECTURE_SECONDARY
529 if (a
== ARCHITECTURE_SECONDARY
)
532 if (b
== ARCHITECTURE_SECONDARY
)
539 static int dissect_image(
543 const VeritySettings
*verity
,
544 const MountOptions
*mount_options
,
545 DissectImageFlags flags
) {
547 sd_id128_t root_uuid
= SD_ID128_NULL
, root_verity_uuid
= SD_ID128_NULL
;
548 sd_id128_t usr_uuid
= SD_ID128_NULL
, usr_verity_uuid
= SD_ID128_NULL
;
549 bool is_gpt
, is_mbr
, multiple_generic
= false,
550 generic_rw
= false, /* initialize to appease gcc */
551 generic_growfs
= false;
552 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
553 _cleanup_free_
char *generic_node
= NULL
;
554 sd_id128_t generic_uuid
= SD_ID128_NULL
;
555 const char *pttype
= NULL
, *sptuuid
= NULL
;
557 int r
, generic_nr
= -1, n_partitions
;
562 assert(!verity
|| verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
563 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
564 assert(!verity
|| verity
->root_hash_sig
|| verity
->root_hash_sig_size
== 0);
565 assert(!verity
|| (verity
->root_hash
|| !verity
->root_hash_sig
));
566 assert(!((flags
& DISSECT_IMAGE_GPT_ONLY
) && (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)));
567 assert(m
->sector_size
> 0);
569 /* Probes a disk image, and returns information about what it found in *ret.
571 * Returns -ENOPKG if no suitable partition table or file system could be found.
572 * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found.
573 * Returns -ENXIO if we couldn't find any partition suitable as root or /usr partition
574 * Returns -ENOTUNIQ if we only found multiple generic partitions and thus don't know what to do with that */
576 uint64_t diskseq
= m
->loop
? m
->loop
->diskseq
: 0;
578 if (verity
&& verity
->root_hash
) {
579 sd_id128_t fsuuid
, vuuid
;
581 /* If a root hash is supplied, then we use the root partition that has a UUID that match the
582 * first 128bit of the root hash. And we use the verity partition that has a UUID that match
583 * the final 128bit. */
585 if (verity
->root_hash_size
< sizeof(sd_id128_t
))
588 memcpy(&fsuuid
, verity
->root_hash
, sizeof(sd_id128_t
));
589 memcpy(&vuuid
, (const uint8_t*) verity
->root_hash
+ verity
->root_hash_size
- sizeof(sd_id128_t
), sizeof(sd_id128_t
));
591 if (sd_id128_is_null(fsuuid
))
593 if (sd_id128_is_null(vuuid
))
596 /* If the verity data declares it's for the /usr partition, then search for that, in all
597 * other cases assume it's for the root partition. */
598 if (verity
->designator
== PARTITION_USR
) {
600 usr_verity_uuid
= vuuid
;
603 root_verity_uuid
= vuuid
;
607 b
= blkid_new_probe();
612 r
= blkid_probe_set_device(b
, fd
, 0, 0);
614 return errno_or_else(ENOMEM
);
617 r
= blkid_probe_set_sectorsize(b
, m
->sector_size
);
619 return errno_or_else(EIO
);
621 if ((flags
& DISSECT_IMAGE_GPT_ONLY
) == 0) {
622 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
623 blkid_probe_enable_superblocks(b
, 1);
624 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
|BLKID_SUBLKS_USAGE
|BLKID_SUBLKS_UUID
);
627 blkid_probe_enable_partitions(b
, 1);
628 blkid_probe_set_partitions_flags(b
, BLKID_PARTS_ENTRY_DETAILS
);
631 r
= blkid_do_safeprobe(b
);
632 if (r
== _BLKID_SAFEPROBE_ERROR
)
633 return errno_or_else(EIO
);
634 if (IN_SET(r
, _BLKID_SAFEPROBE_AMBIGUOUS
, _BLKID_SAFEPROBE_NOT_FOUND
))
635 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG
), "Failed to identify any partition table.");
637 assert(r
== _BLKID_SAFEPROBE_FOUND
);
639 if ((!(flags
& DISSECT_IMAGE_GPT_ONLY
) &&
640 (flags
& DISSECT_IMAGE_GENERIC_ROOT
)) ||
641 (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)) {
642 const char *usage
= NULL
;
644 /* If flags permit this, also allow using non-partitioned single-filesystem images */
646 (void) blkid_probe_lookup_value(b
, "USAGE", &usage
, NULL
);
647 if (STRPTR_IN_SET(usage
, "filesystem", "crypto")) {
648 _cleanup_free_
char *t
= NULL
, *n
= NULL
, *o
= NULL
;
649 const char *fstype
= NULL
, *options
= NULL
, *suuid
= NULL
;
650 _cleanup_close_
int mount_node_fd
= -EBADF
;
651 sd_id128_t uuid
= SD_ID128_NULL
;
653 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
654 mount_node_fd
= open_partition(devname
, /* is_partition = */ false, m
->loop
);
655 if (mount_node_fd
< 0)
656 return mount_node_fd
;
659 /* OK, we have found a file system, that's our root partition then. */
660 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
661 (void) blkid_probe_lookup_value(b
, "UUID", &suuid
, NULL
);
670 /* blkid will return FAT's serial number as UUID, hence it is quite possible
671 * that parsing this will fail. We'll ignore the ID, since it's just too
672 * short to be useful as tru identifier. */
673 r
= sd_id128_from_string(suuid
, &uuid
);
675 log_debug_errno(r
, "Failed to parse file system UUID '%s', ignoring: %m", suuid
);
678 r
= make_partition_devname(devname
, diskseq
, -1, flags
, &n
);
682 m
->single_file_system
= true;
683 m
->encrypted
= streq_ptr(fstype
, "crypto_LUKS");
685 m
->has_verity
= verity
&& verity
->data_path
;
686 m
->verity_ready
= verity_settings_data_covers(verity
, PARTITION_ROOT
);
688 m
->has_verity_sig
= false; /* signature not embedded, must be specified */
689 m
->verity_sig_ready
= m
->verity_ready
&& verity
->root_hash_sig
;
691 m
->image_uuid
= uuid
;
693 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
700 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
702 .rw
= !m
->verity_ready
&& !fstype_is_ro(fstype
),
704 .architecture
= _ARCHITECTURE_INVALID
,
705 .fstype
= TAKE_PTR(t
),
707 .mount_options
= TAKE_PTR(o
),
708 .mount_node_fd
= TAKE_FD(mount_node_fd
),
717 (void) blkid_probe_lookup_value(b
, "PTTYPE", &pttype
, NULL
);
721 is_gpt
= streq_ptr(pttype
, "gpt");
722 is_mbr
= streq_ptr(pttype
, "dos");
724 if (!is_gpt
&& ((flags
& DISSECT_IMAGE_GPT_ONLY
) || !is_mbr
))
727 /* We support external verity data partitions only if the image has no partition table */
728 if (verity
&& verity
->data_path
)
731 if (FLAGS_SET(flags
, DISSECT_IMAGE_ADD_PARTITION_DEVICES
)) {
732 /* Safety check: refuse block devices that carry a partition table but for which the kernel doesn't
733 * do partition scanning. */
734 r
= blockdev_partscan_enabled(fd
);
738 return -EPROTONOSUPPORT
;
741 (void) blkid_probe_lookup_value(b
, "PTUUID", &sptuuid
, NULL
);
743 r
= sd_id128_from_string(sptuuid
, &m
->image_uuid
);
745 log_debug_errno(r
, "Failed to parse partition table UUID '%s', ignoring: %m", sptuuid
);
749 pl
= blkid_probe_get_partitions(b
);
751 return errno_or_else(ENOMEM
);
754 n_partitions
= blkid_partlist_numof_partitions(pl
);
755 if (n_partitions
< 0)
756 return errno_or_else(EIO
);
758 for (int i
= 0; i
< n_partitions
; i
++) {
759 _cleanup_free_
char *node
= NULL
;
760 unsigned long long pflags
;
761 blkid_loff_t start
, size
;
766 pp
= blkid_partlist_get_partition(pl
, i
);
768 return errno_or_else(EIO
);
770 pflags
= blkid_partition_get_flags(pp
);
773 nr
= blkid_partition_get_partno(pp
);
775 return errno_or_else(EIO
);
778 start
= blkid_partition_get_start(pp
);
780 return errno_or_else(EIO
);
782 assert((uint64_t) start
< UINT64_MAX
/512);
785 size
= blkid_partition_get_size(pp
);
787 return errno_or_else(EIO
);
789 assert((uint64_t) size
< UINT64_MAX
/512);
791 /* While probing we need the non-diskseq device node name to access the thing, hence mask off
792 * DISSECT_IMAGE_DISKSEQ_DEVNODE. */
793 r
= make_partition_devname(devname
, diskseq
, nr
, flags
& ~DISSECT_IMAGE_DISKSEQ_DEVNODE
, &node
);
797 /* So here's the thing: after the main ("whole") block device popped up it might take a while
798 * before the kernel fully probed the partition table. Waiting for that to finish is icky in
799 * userspace. So here's what we do instead. We issue the BLKPG_ADD_PARTITION ioctl to add the
800 * partition ourselves, racing against the kernel. Good thing is: if this call fails with
801 * EBUSY then the kernel was quicker than us, and that's totally OK, the outcome is good for
802 * us: the device node will exist. If OTOH our call was successful we won the race. Which is
803 * also good as the outcome is the same: the partition block device exists, and we can use
806 * Kernel returns EBUSY if there's already a partition by that number or an overlapping
807 * partition already existent. */
809 if (FLAGS_SET(flags
, DISSECT_IMAGE_ADD_PARTITION_DEVICES
)) {
810 r
= block_device_add_partition(fd
, node
, nr
, (uint64_t) start
* 512, (uint64_t) size
* 512);
813 return log_debug_errno(r
, "BLKPG_ADD_PARTITION failed: %m");
815 log_debug_errno(r
, "Kernel was quicker than us in adding partition %i.", nr
);
817 log_debug("We were quicker than kernel in adding partition %i.", nr
);
821 const char *fstype
= NULL
, *label
;
822 sd_id128_t type_id
, id
;
823 GptPartitionType type
;
824 bool rw
= true, growfs
= false;
826 r
= blkid_partition_get_uuid_id128(pp
, &id
);
828 log_debug_errno(r
, "Failed to read partition UUID, ignoring: %m");
832 r
= blkid_partition_get_type_id128(pp
, &type_id
);
834 log_debug_errno(r
, "Failed to read partition type UUID, ignoring: %m");
838 type
= gpt_partition_type_from_uuid(type_id
);
840 label
= blkid_partition_get_name(pp
); /* libblkid returns NULL here if empty */
842 if (IN_SET(type
.designator
,
848 check_partition_flags(node
, pflags
,
849 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
851 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
854 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
855 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
857 } else if (type
.designator
== PARTITION_ESP
) {
859 /* Note that we don't check the SD_GPT_FLAG_NO_AUTO flag for the ESP, as it is
860 * not defined there. We instead check the SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as
861 * recommended by the UEFI spec (See "12.3.3 Number and Location of System
864 if (pflags
& SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL
)
869 } else if (type
.designator
== PARTITION_ROOT
) {
871 check_partition_flags(node
, pflags
,
872 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
874 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
877 /* If a root ID is specified, ignore everything but the root id */
878 if (!sd_id128_is_null(root_uuid
) && !sd_id128_equal(root_uuid
, id
))
881 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
882 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
884 } else if (type
.designator
== PARTITION_ROOT_VERITY
) {
886 check_partition_flags(node
, pflags
,
887 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
889 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
892 m
->has_verity
= true;
894 /* If no verity configuration is specified, then don't do verity */
897 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
900 /* If root hash is specified, then ignore everything but the root id */
901 if (!sd_id128_is_null(root_verity_uuid
) && !sd_id128_equal(root_verity_uuid
, id
))
904 fstype
= "DM_verity_hash";
907 } else if (type
.designator
== PARTITION_ROOT_VERITY_SIG
) {
909 check_partition_flags(node
, pflags
,
910 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
912 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
915 m
->has_verity_sig
= true;
919 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
922 fstype
= "verity_hash_signature";
925 } else if (type
.designator
== PARTITION_USR
) {
927 check_partition_flags(node
, pflags
,
928 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
930 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
933 /* If a usr ID is specified, ignore everything but the usr id */
934 if (!sd_id128_is_null(usr_uuid
) && !sd_id128_equal(usr_uuid
, id
))
937 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
938 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
940 } else if (type
.designator
== PARTITION_USR_VERITY
) {
942 check_partition_flags(node
, pflags
,
943 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
945 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
948 m
->has_verity
= true;
952 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
955 /* If usr hash is specified, then ignore everything but the usr id */
956 if (!sd_id128_is_null(usr_verity_uuid
) && !sd_id128_equal(usr_verity_uuid
, id
))
959 fstype
= "DM_verity_hash";
962 } else if (type
.designator
== PARTITION_USR_VERITY_SIG
) {
964 check_partition_flags(node
, pflags
,
965 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
967 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
970 m
->has_verity_sig
= true;
974 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
977 fstype
= "verity_hash_signature";
980 } else if (type
.designator
== PARTITION_SWAP
) {
982 check_partition_flags(node
, pflags
, SD_GPT_FLAG_NO_AUTO
);
984 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
989 /* We don't have a designator for SD_GPT_LINUX_GENERIC so check the UUID instead. */
990 } else if (sd_id128_equal(type
.uuid
, SD_GPT_LINUX_GENERIC
)) {
992 check_partition_flags(node
, pflags
,
993 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
995 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
999 multiple_generic
= true;
1002 generic_rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1003 generic_growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1005 generic_node
= TAKE_PTR(node
);
1008 } else if (type
.designator
== PARTITION_VAR
) {
1010 check_partition_flags(node
, pflags
,
1011 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
1013 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
1016 if (!FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_VAR_CHECK
)) {
1017 sd_id128_t var_uuid
;
1019 /* For /var we insist that the uuid of the partition matches the
1020 * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
1021 * ID. Why? Unlike the other partitions /var is inherently
1022 * installation specific, hence we need to be careful not to mount it
1023 * in the wrong installation. By hashing the partition UUID from
1024 * /etc/machine-id we can securely bind the partition to the
1027 r
= sd_id128_get_machine_app_specific(SD_GPT_VAR
, &var_uuid
);
1031 if (!sd_id128_equal(var_uuid
, id
)) {
1032 log_debug("Found a /var/ partition, but its UUID didn't match our expectations "
1033 "(found: " SD_ID128_UUID_FORMAT_STR
", expected: " SD_ID128_UUID_FORMAT_STR
"), ignoring.",
1034 SD_ID128_FORMAT_VAL(id
), SD_ID128_FORMAT_VAL(var_uuid
));
1039 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
1040 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
1043 if (type
.designator
!= _PARTITION_DESIGNATOR_INVALID
) {
1044 _cleanup_free_
char *t
= NULL
, *o
= NULL
, *l
= NULL
, *n
= NULL
;
1045 _cleanup_close_
int mount_node_fd
= -EBADF
;
1046 const char *options
= NULL
;
1048 if (m
->partitions
[type
.designator
].found
) {
1049 /* For most partition types the first one we see wins. Except for the
1050 * rootfs and /usr, where we do a version compare of the label, and
1051 * let the newest version win. This permits a simple A/B versioning
1052 * scheme in OS images. */
1054 if (compare_arch(type
.arch
, m
->partitions
[type
.designator
].architecture
) <= 0)
1057 if (!partition_designator_is_versioned(type
.designator
) ||
1058 strverscmp_improved(m
->partitions
[type
.designator
].label
, label
) >= 0)
1061 dissected_partition_done(m
->partitions
+ type
.designator
);
1064 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
) &&
1065 type
.designator
!= PARTITION_SWAP
) {
1066 mount_node_fd
= open_partition(node
, /* is_partition = */ true, m
->loop
);
1067 if (mount_node_fd
< 0)
1068 return mount_node_fd
;
1071 r
= make_partition_devname(devname
, diskseq
, nr
, flags
, &n
);
1087 options
= mount_options_from_designator(mount_options
, type
.designator
);
1089 o
= strdup(options
);
1094 m
->partitions
[type
.designator
] = (DissectedPartition
) {
1099 .architecture
= type
.arch
,
1100 .node
= TAKE_PTR(n
),
1101 .fstype
= TAKE_PTR(t
),
1102 .label
= TAKE_PTR(l
),
1104 .mount_options
= TAKE_PTR(o
),
1105 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1106 .offset
= (uint64_t) start
* 512,
1107 .size
= (uint64_t) size
* 512,
1108 .gpt_flags
= pflags
,
1112 } else if (is_mbr
) {
1114 switch (blkid_partition_get_type(pp
)) {
1116 case 0x83: /* Linux partition */
1118 if (pflags
!= 0x80) /* Bootable flag */
1122 multiple_generic
= true;
1126 generic_growfs
= false;
1127 generic_node
= TAKE_PTR(node
);
1132 case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
1133 _cleanup_close_
int mount_node_fd
= -EBADF
;
1134 _cleanup_free_
char *o
= NULL
, *n
= NULL
;
1135 sd_id128_t id
= SD_ID128_NULL
;
1136 const char *options
= NULL
;
1138 /* First one wins */
1139 if (m
->partitions
[PARTITION_XBOOTLDR
].found
)
1142 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
1143 mount_node_fd
= open_partition(node
, /* is_partition = */ true, m
->loop
);
1144 if (mount_node_fd
< 0)
1145 return mount_node_fd
;
1148 (void) blkid_partition_get_uuid_id128(pp
, &id
);
1150 r
= make_partition_devname(devname
, diskseq
, nr
, flags
, &n
);
1154 options
= mount_options_from_designator(mount_options
, PARTITION_XBOOTLDR
);
1156 o
= strdup(options
);
1161 m
->partitions
[PARTITION_XBOOTLDR
] = (DissectedPartition
) {
1166 .architecture
= _ARCHITECTURE_INVALID
,
1167 .node
= TAKE_PTR(n
),
1169 .mount_options
= TAKE_PTR(o
),
1170 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1171 .offset
= (uint64_t) start
* 512,
1172 .size
= (uint64_t) size
* 512,
1180 if (!m
->partitions
[PARTITION_ROOT
].found
&&
1181 (m
->partitions
[PARTITION_ROOT_VERITY
].found
||
1182 m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
))
1183 return -EADDRNOTAVAIL
; /* Verity found but no matching rootfs? Something is off, refuse. */
1185 /* Hmm, we found a signature partition but no Verity data? Something is off. */
1186 if (m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_ROOT_VERITY
].found
)
1187 return -EADDRNOTAVAIL
;
1189 if (!m
->partitions
[PARTITION_USR
].found
&&
1190 (m
->partitions
[PARTITION_USR_VERITY
].found
||
1191 m
->partitions
[PARTITION_USR_VERITY_SIG
].found
))
1192 return -EADDRNOTAVAIL
; /* as above */
1195 if (m
->partitions
[PARTITION_USR_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_USR_VERITY
].found
)
1196 return -EADDRNOTAVAIL
;
1198 /* If root and /usr are combined then insist that the architecture matches */
1199 if (m
->partitions
[PARTITION_ROOT
].found
&&
1200 m
->partitions
[PARTITION_USR
].found
&&
1201 (m
->partitions
[PARTITION_ROOT
].architecture
>= 0 &&
1202 m
->partitions
[PARTITION_USR
].architecture
>= 0 &&
1203 m
->partitions
[PARTITION_ROOT
].architecture
!= m
->partitions
[PARTITION_USR
].architecture
))
1204 return -EADDRNOTAVAIL
;
1206 if (!m
->partitions
[PARTITION_ROOT
].found
&&
1207 !m
->partitions
[PARTITION_USR
].found
&&
1208 (flags
& DISSECT_IMAGE_GENERIC_ROOT
) &&
1209 (!verity
|| !verity
->root_hash
|| verity
->designator
!= PARTITION_USR
)) {
1211 /* OK, we found nothing usable, then check if there's a single generic partition, and use
1212 * that. If the root hash was set however, then we won't fall back to a generic node, because
1213 * the root hash decides. */
1215 /* If we didn't find a properly marked root partition, but we did find a single suitable
1216 * generic Linux partition, then use this as root partition, if the caller asked for it. */
1217 if (multiple_generic
)
1220 /* If we didn't find a generic node, then we can't fix this up either */
1222 _cleanup_close_
int mount_node_fd
= -EBADF
;
1223 _cleanup_free_
char *o
= NULL
, *n
= NULL
;
1224 const char *options
;
1226 if (FLAGS_SET(flags
, DISSECT_IMAGE_PIN_PARTITION_DEVICES
)) {
1227 mount_node_fd
= open_partition(generic_node
, /* is_partition = */ true, m
->loop
);
1228 if (mount_node_fd
< 0)
1229 return mount_node_fd
;
1232 r
= make_partition_devname(devname
, diskseq
, generic_nr
, flags
, &n
);
1236 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
1238 o
= strdup(options
);
1243 assert(generic_nr
>= 0);
1244 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
1247 .growfs
= generic_growfs
,
1248 .partno
= generic_nr
,
1249 .architecture
= _ARCHITECTURE_INVALID
,
1250 .node
= TAKE_PTR(n
),
1251 .uuid
= generic_uuid
,
1252 .mount_options
= TAKE_PTR(o
),
1253 .mount_node_fd
= TAKE_FD(mount_node_fd
),
1254 .offset
= UINT64_MAX
,
1260 /* 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 */
1261 if (FLAGS_SET(flags
, DISSECT_IMAGE_REQUIRE_ROOT
) &&
1262 !(m
->partitions
[PARTITION_ROOT
].found
|| (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
1265 if (m
->partitions
[PARTITION_ROOT_VERITY
].found
) {
1266 /* We only support one verity partition per image, i.e. can't do for both /usr and root fs */
1267 if (m
->partitions
[PARTITION_USR_VERITY
].found
)
1270 /* We don't support verity enabled root with a split out /usr. Neither with nor without
1271 * verity there. (Note that we do support verity-less root with verity-full /usr, though.) */
1272 if (m
->partitions
[PARTITION_USR
].found
)
1273 return -EADDRNOTAVAIL
;
1277 /* If a verity designator is specified, then insist that the matching partition exists */
1278 if (verity
->designator
>= 0 && !m
->partitions
[verity
->designator
].found
)
1279 return -EADDRNOTAVAIL
;
1281 bool have_verity_sig_partition
=
1282 m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR_VERITY_SIG
: PARTITION_ROOT_VERITY_SIG
].found
;
1284 if (verity
->root_hash
) {
1285 /* If we have an explicit root hash and found the partitions for it, then we are ready to use
1286 * Verity, set things up for it */
1288 if (verity
->designator
< 0 || verity
->designator
== PARTITION_ROOT
) {
1289 if (!m
->partitions
[PARTITION_ROOT_VERITY
].found
|| !m
->partitions
[PARTITION_ROOT
].found
)
1290 return -EADDRNOTAVAIL
;
1292 /* If we found a verity setup, then the root partition is necessarily read-only. */
1293 m
->partitions
[PARTITION_ROOT
].rw
= false;
1294 m
->verity_ready
= true;
1297 assert(verity
->designator
== PARTITION_USR
);
1299 if (!m
->partitions
[PARTITION_USR_VERITY
].found
|| !m
->partitions
[PARTITION_USR
].found
)
1300 return -EADDRNOTAVAIL
;
1302 m
->partitions
[PARTITION_USR
].rw
= false;
1303 m
->verity_ready
= true;
1306 if (m
->verity_ready
)
1307 m
->verity_sig_ready
= verity
->root_hash_sig
|| have_verity_sig_partition
;
1309 } else if (have_verity_sig_partition
) {
1311 /* If we found an embedded signature partition, we are ready, too. */
1313 m
->verity_ready
= m
->verity_sig_ready
= true;
1314 m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR
: PARTITION_ROOT
].rw
= false;
1318 r
= dissected_image_probe_filesystems(m
, fd
);
1326 int dissect_image_file(
1328 const VeritySettings
*verity
,
1329 const MountOptions
*mount_options
,
1330 DissectImageFlags flags
,
1331 DissectedImage
**ret
) {
1334 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
1335 _cleanup_close_
int fd
= -EBADF
;
1341 fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
1345 r
= fd_verify_regular(fd
);
1349 r
= dissected_image_new(path
, &m
);
1353 r
= probe_sector_size(fd
, &m
->sector_size
);
1357 r
= dissect_image(m
, fd
, path
, verity
, mount_options
, flags
);
1368 DissectedImage
* dissected_image_unref(DissectedImage
*m
) {
1372 /* First, clear dissected partitions. */
1373 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++)
1374 dissected_partition_done(m
->partitions
+ i
);
1376 /* Second, free decrypted images. This must be after dissected_partition_done(), as freeing
1377 * DecryptedImage may try to deactivate partitions. */
1378 decrypted_image_unref(m
->decrypted_image
);
1380 /* Third, unref LoopDevice. This must be called after the above two, as freeing LoopDevice may try to
1381 * remove existing partitions on the loopback block device. */
1382 loop_device_unref(m
->loop
);
1384 free(m
->image_name
);
1386 strv_free(m
->machine_info
);
1387 strv_free(m
->os_release
);
1388 strv_free(m
->initrd_release
);
1389 strv_free(m
->extension_release
);
1394 static int is_loop_device(const char *path
) {
1395 char s
[SYS_BLOCK_PATH_MAX("/../loop/")];
1400 if (stat(path
, &st
) < 0)
1403 if (!S_ISBLK(st
.st_mode
))
1406 xsprintf_sys_block_path(s
, "/loop/", st
.st_dev
);
1407 if (access(s
, F_OK
) < 0) {
1408 if (errno
!= ENOENT
)
1411 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
1412 xsprintf_sys_block_path(s
, "/../loop/", st
.st_dev
);
1413 if (access(s
, F_OK
) < 0)
1414 return errno
== ENOENT
? false : -errno
;
1420 static int run_fsck(int node_fd
, const char *fstype
) {
1424 assert(node_fd
>= 0);
1427 r
= fsck_exists_for_fstype(fstype
);
1429 log_debug_errno(r
, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype
);
1433 log_debug("Not checking partition %s, as fsck for %s does not exist.", FORMAT_PROC_FD_PATH(node_fd
), fstype
);
1439 &node_fd
, 1, /* Leave the node fd open */
1440 FORK_RESET_SIGNALS
|FORK_CLOSE_ALL_FDS
|FORK_RLIMIT_NOFILE_SAFE
|FORK_DEATHSIG
|FORK_NULL_STDIO
|FORK_CLOEXEC_OFF
,
1443 return log_debug_errno(r
, "Failed to fork off fsck: %m");
1446 execl("/sbin/fsck", "/sbin/fsck", "-aT", FORMAT_PROC_FD_PATH(node_fd
), NULL
);
1448 log_debug_errno(errno
, "Failed to execl() fsck: %m");
1449 _exit(FSCK_OPERATIONAL_ERROR
);
1452 exit_status
= wait_for_terminate_and_check("fsck", pid
, 0);
1453 if (exit_status
< 0)
1454 return log_debug_errno(exit_status
, "Failed to fork off /sbin/fsck: %m");
1456 if ((exit_status
& ~FSCK_ERROR_CORRECTED
) != FSCK_SUCCESS
) {
1457 log_debug("fsck failed with exit status %i.", exit_status
);
1459 if ((exit_status
& (FSCK_SYSTEM_SHOULD_REBOOT
|FSCK_ERRORS_LEFT_UNCORRECTED
)) != 0)
1460 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
), "File system is corrupted, refusing.");
1462 log_debug("Ignoring fsck error.");
1468 static int fs_grow(const char *node_path
, const char *mount_path
) {
1469 _cleanup_close_
int mount_fd
= -EBADF
, node_fd
= -EBADF
;
1470 uint64_t size
, newsize
;
1473 node_fd
= open(node_path
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
1475 return log_debug_errno(errno
, "Failed to open node device %s: %m", node_path
);
1477 if (ioctl(node_fd
, BLKGETSIZE64
, &size
) != 0)
1478 return log_debug_errno(errno
, "Failed to get block device size of %s: %m", node_path
);
1480 mount_fd
= open(mount_path
, O_RDONLY
|O_DIRECTORY
|O_CLOEXEC
);
1482 return log_debug_errno(errno
, "Failed to open mountd file system %s: %m", mount_path
);
1484 log_debug("Resizing \"%s\" to %"PRIu64
" bytes...", mount_path
, size
);
1485 r
= resize_fs(mount_fd
, size
, &newsize
);
1487 return log_debug_errno(r
, "Failed to resize \"%s\" to %"PRIu64
" bytes: %m", mount_path
, size
);
1489 if (newsize
== size
)
1490 log_debug("Successfully resized \"%s\" to %s bytes.",
1491 mount_path
, FORMAT_BYTES(newsize
));
1493 assert(newsize
< size
);
1494 log_debug("Successfully resized \"%s\" to %s bytes (%"PRIu64
" bytes lost due to blocksize).",
1495 mount_path
, FORMAT_BYTES(newsize
), size
- newsize
);
1501 static int mount_partition(
1502 DissectedPartition
*m
,
1504 const char *directory
,
1507 DissectImageFlags flags
) {
1509 _cleanup_free_
char *chased
= NULL
, *options
= NULL
;
1510 const char *p
, *node
, *fstype
;
1511 bool rw
, remap_uid_gid
= false;
1517 if (m
->mount_node_fd
< 0)
1520 /* Use decrypted node and matching fstype if available, otherwise use the original device */
1521 node
= FORMAT_PROC_FD_PATH(m
->mount_node_fd
);
1522 fstype
= m
->decrypted_node
? m
->decrypted_fstype
: m
->fstype
;
1525 return -EAFNOSUPPORT
;
1526 r
= dissect_fstype_ok(fstype
);
1530 return -EIDRM
; /* Recognizable error */
1532 /* We are looking at an encrypted partition? This either means stacked encryption, or the caller
1533 * didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error for this
1535 if (streq(fstype
, "crypto_LUKS"))
1538 rw
= m
->rw
&& !(flags
& DISSECT_IMAGE_MOUNT_READ_ONLY
);
1540 if (FLAGS_SET(flags
, DISSECT_IMAGE_FSCK
) && rw
) {
1541 r
= run_fsck(m
->mount_node_fd
, fstype
);
1547 /* Automatically create missing mount points inside the image, if necessary. */
1548 r
= mkdir_p_root(where
, directory
, uid_shift
, (gid_t
) uid_shift
, 0755);
1549 if (r
< 0 && r
!= -EROFS
)
1552 r
= chase_symlinks(directory
, where
, CHASE_PREFIX_ROOT
, &chased
, NULL
);
1558 /* Create top-level mount if missing – but only if this is asked for. This won't modify the
1559 * image (as the branch above does) but the host hierarchy, and the created directory might
1560 * survive our mount in the host hierarchy hence. */
1561 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
1562 r
= mkdir_p(where
, 0755);
1570 /* If requested, turn on discard support. */
1571 if (fstype_can_discard(fstype
) &&
1572 ((flags
& DISSECT_IMAGE_DISCARD
) ||
1573 ((flags
& DISSECT_IMAGE_DISCARD_ON_LOOP
) && is_loop_device(m
->node
) > 0))) {
1574 options
= strdup("discard");
1579 if (uid_is_valid(uid_shift
) && uid_shift
!= 0) {
1581 if (fstype_can_uid_gid(fstype
)) {
1582 _cleanup_free_
char *uid_option
= NULL
;
1584 if (asprintf(&uid_option
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
1587 if (!strextend_with_separator(&options
, ",", uid_option
))
1589 } else if (FLAGS_SET(flags
, DISSECT_IMAGE_MOUNT_IDMAPPED
))
1590 remap_uid_gid
= true;
1593 if (!isempty(m
->mount_options
))
1594 if (!strextend_with_separator(&options
, ",", m
->mount_options
))
1597 /* So, when you request MS_RDONLY from ext4, then this means nothing. It happily still writes to the
1598 * backing storage. What's worse, the BLKRO[GS]ET flag and (in case of loopback devices)
1599 * LO_FLAGS_READ_ONLY don't mean anything, they affect userspace accesses only, and write accesses
1600 * from the upper file system still get propagated through to the underlying file system,
1601 * unrestricted. To actually get ext4/xfs/btrfs to stop writing to the device we need to specify
1602 * "norecovery" as mount option, in addition to MS_RDONLY. Yes, this sucks, since it means we need to
1603 * carry a per file system table here.
1605 * Note that this means that we might not be able to mount corrupted file systems as read-only
1606 * anymore (since in some cases the kernel implementations will refuse mounting when corrupted,
1607 * read-only and "norecovery" is specified). But I think for the case of automatically determined
1608 * mount options for loopback devices this is the right choice, since otherwise using the same
1609 * loopback file twice even in read-only mode, is going to fail badly sooner or later. The usecase of
1610 * making reuse of the immutable images "just work" is more relevant to us than having read-only
1611 * access that actually modifies stuff work on such image files. Or to say this differently: if
1612 * people want their file systems to be fixed up they should just open them in writable mode, where
1613 * all these problems don't exist. */
1614 if (!rw
&& STRPTR_IN_SET(fstype
, "ext3", "ext4", "xfs", "btrfs"))
1615 if (!strextend_with_separator(&options
, ",", "norecovery"))
1618 r
= mount_nofollow_verbose(LOG_DEBUG
, node
, p
, fstype
, MS_NODEV
|(rw
? 0 : MS_RDONLY
), options
);
1622 if (rw
&& m
->growfs
&& FLAGS_SET(flags
, DISSECT_IMAGE_GROWFS
))
1623 (void) fs_grow(node
, p
);
1625 if (remap_uid_gid
) {
1626 r
= remount_idmap(p
, uid_shift
, uid_range
, UID_INVALID
, REMOUNT_IDMAPPING_HOST_ROOT
);
1634 static int mount_root_tmpfs(const char *where
, uid_t uid_shift
, DissectImageFlags flags
) {
1635 _cleanup_free_
char *options
= NULL
;
1640 /* For images that contain /usr/ but no rootfs, let's mount rootfs as tmpfs */
1642 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
1643 r
= mkdir_p(where
, 0755);
1648 if (uid_is_valid(uid_shift
)) {
1649 if (asprintf(&options
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
1653 r
= mount_nofollow_verbose(LOG_DEBUG
, "rootfs", where
, "tmpfs", MS_NODEV
, options
);
1660 int dissected_image_mount(
1665 DissectImageFlags flags
) {
1667 int r
, xbootldr_mounted
;
1674 * -ENXIO → No root partition found
1675 * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release/extension-release file found
1676 * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet
1677 * -EUCLEAN → fsck for file system failed
1678 * -EBUSY → File system already mounted/used elsewhere (kernel)
1679 * -EAFNOSUPPORT → File system type not supported or not known
1680 * -EIDRM → File system is not among allowlisted "common" file systems
1683 if (!(m
->partitions
[PARTITION_ROOT
].found
||
1684 (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
1685 return -ENXIO
; /* Require a root fs or at least a /usr/ fs (the latter is subject to a flag of its own) */
1687 if ((flags
& DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
) == 0) {
1689 /* First mount the root fs. If there's none we use a tmpfs. */
1690 if (m
->partitions
[PARTITION_ROOT
].found
)
1691 r
= mount_partition(m
->partitions
+ PARTITION_ROOT
, where
, NULL
, uid_shift
, uid_range
, flags
);
1693 r
= mount_root_tmpfs(where
, uid_shift
, flags
);
1697 /* For us mounting root always means mounting /usr as well */
1698 r
= mount_partition(m
->partitions
+ PARTITION_USR
, where
, "/usr", uid_shift
, uid_range
, flags
);
1702 if ((flags
& (DISSECT_IMAGE_VALIDATE_OS
|DISSECT_IMAGE_VALIDATE_OS_EXT
)) != 0) {
1703 /* If either one of the validation flags are set, ensure that the image qualifies
1704 * as one or the other (or both). */
1707 if (FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS
)) {
1708 r
= path_is_os_tree(where
);
1714 if (!ok
&& FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS_EXT
)) {
1715 r
= path_is_extension_tree(where
, m
->image_name
, FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_SYSEXT_CHECK
));
1727 if (flags
& DISSECT_IMAGE_MOUNT_ROOT_ONLY
)
1730 r
= mount_partition(m
->partitions
+ PARTITION_HOME
, where
, "/home", uid_shift
, uid_range
, flags
);
1734 r
= mount_partition(m
->partitions
+ PARTITION_SRV
, where
, "/srv", uid_shift
, uid_range
, flags
);
1738 r
= mount_partition(m
->partitions
+ PARTITION_VAR
, where
, "/var", uid_shift
, uid_range
, flags
);
1742 r
= mount_partition(m
->partitions
+ PARTITION_TMP
, where
, "/var/tmp", uid_shift
, uid_range
, flags
);
1746 xbootldr_mounted
= mount_partition(m
->partitions
+ PARTITION_XBOOTLDR
, where
, "/boot", uid_shift
, uid_range
, flags
);
1747 if (xbootldr_mounted
< 0)
1748 return xbootldr_mounted
;
1750 if (m
->partitions
[PARTITION_ESP
].found
) {
1751 int esp_done
= false;
1753 /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it
1754 * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */
1756 r
= chase_symlinks("/efi", where
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
1761 /* /efi doesn't exist. Let's see if /boot is suitable then */
1763 if (!xbootldr_mounted
) {
1764 _cleanup_free_
char *p
= NULL
;
1766 r
= chase_symlinks("/boot", where
, CHASE_PREFIX_ROOT
, &p
, NULL
);
1770 } else if (dir_is_empty(p
, /* ignore_hidden_or_backup= */ false) > 0) {
1771 /* It exists and is an empty directory. Let's mount the ESP there. */
1772 r
= mount_partition(m
->partitions
+ PARTITION_ESP
, where
, "/boot", uid_shift
, uid_range
, flags
);
1782 /* OK, let's mount the ESP now to /efi (possibly creating the dir if missing) */
1784 r
= mount_partition(m
->partitions
+ PARTITION_ESP
, where
, "/efi", uid_shift
, uid_range
, flags
);
1793 int dissected_image_mount_and_warn(
1798 DissectImageFlags flags
) {
1805 r
= dissected_image_mount(m
, where
, uid_shift
, uid_range
, flags
);
1807 return log_error_errno(r
, "Not root file system found in image.");
1808 if (r
== -EMEDIUMTYPE
)
1809 return log_error_errno(r
, "No suitable os-release/extension-release file in image found.");
1811 return log_error_errno(r
, "Encrypted file system discovered, but decryption not requested.");
1813 return log_error_errno(r
, "File system check on image failed.");
1815 return log_error_errno(r
, "File system already mounted elsewhere.");
1816 if (r
== -EAFNOSUPPORT
)
1817 return log_error_errno(r
, "File system type not supported or not known.");
1819 return log_error_errno(r
, "File system is too uncommon, refused.");
1821 return log_error_errno(r
, "Failed to mount image: %m");
1826 #if HAVE_LIBCRYPTSETUP
1827 struct DecryptedPartition
{
1828 struct crypt_device
*device
;
1834 typedef struct DecryptedPartition DecryptedPartition
;
1836 struct DecryptedImage
{
1838 DecryptedPartition
*decrypted
;
1842 static DecryptedImage
* decrypted_image_free(DecryptedImage
*d
) {
1843 #if HAVE_LIBCRYPTSETUP
1849 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
1850 DecryptedPartition
*p
= d
->decrypted
+ i
;
1852 if (p
->device
&& p
->name
&& !p
->relinquished
) {
1853 _cleanup_free_
char *node
= NULL
;
1855 node
= path_join("/dev/mapper", p
->name
);
1857 r
= btrfs_forget_device(node
);
1858 if (r
< 0 && r
!= -ENOENT
)
1859 log_debug_errno(r
, "Failed to forget btrfs device %s, ignoring: %m", node
);
1863 /* Let's deactivate lazily, as the dm volume may be already/still used by other processes. */
1864 r
= sym_crypt_deactivate_by_name(p
->device
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
1866 log_debug_errno(r
, "Failed to deactivate encrypted partition %s", p
->name
);
1870 sym_crypt_free(p
->device
);
1880 DEFINE_TRIVIAL_REF_UNREF_FUNC(DecryptedImage
, decrypted_image
, decrypted_image_free
);
1882 #if HAVE_LIBCRYPTSETUP
1883 static int decrypted_image_new(DecryptedImage
**ret
) {
1884 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
1888 d
= new(DecryptedImage
, 1);
1892 *d
= (DecryptedImage
) {
1900 static int make_dm_name_and_node(const void *original_node
, const char *suffix
, char **ret_name
, char **ret_node
) {
1901 _cleanup_free_
char *name
= NULL
, *node
= NULL
;
1904 assert(original_node
);
1909 base
= strrchr(original_node
, '/');
1911 base
= original_node
;
1917 name
= strjoin(base
, suffix
);
1920 if (!filename_is_valid(name
))
1923 node
= path_join(sym_crypt_get_dir(), name
);
1927 *ret_name
= TAKE_PTR(name
);
1928 *ret_node
= TAKE_PTR(node
);
1933 static int decrypt_partition(
1934 DissectedPartition
*m
,
1935 const char *passphrase
,
1936 DissectImageFlags flags
,
1937 DecryptedImage
*d
) {
1939 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
1940 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
1941 _cleanup_close_
int fd
= -EBADF
;
1947 if (!m
->found
|| !m
->node
|| !m
->fstype
)
1950 if (!streq(m
->fstype
, "crypto_LUKS"))
1956 r
= dlopen_cryptsetup();
1960 r
= make_dm_name_and_node(m
->node
, "-decrypted", &name
, &node
);
1964 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
1967 r
= sym_crypt_init(&cd
, m
->node
);
1969 return log_debug_errno(r
, "Failed to initialize dm-crypt: %m");
1971 cryptsetup_enable_logging(cd
);
1973 r
= sym_crypt_load(cd
, CRYPT_LUKS
, NULL
);
1975 return log_debug_errno(r
, "Failed to load LUKS metadata: %m");
1977 r
= sym_crypt_activate_by_passphrase(cd
, name
, CRYPT_ANY_SLOT
, passphrase
, strlen(passphrase
),
1978 ((flags
& DISSECT_IMAGE_DEVICE_READ_ONLY
) ? CRYPT_ACTIVATE_READONLY
: 0) |
1979 ((flags
& DISSECT_IMAGE_DISCARD_ON_CRYPTO
) ? CRYPT_ACTIVATE_ALLOW_DISCARDS
: 0));
1981 log_debug_errno(r
, "Failed to activate LUKS device: %m");
1982 return r
== -EPERM
? -EKEYREJECTED
: r
;
1985 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
1987 return log_debug_errno(errno
, "Failed to open %s: %m", node
);
1989 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
1990 .name
= TAKE_PTR(name
),
1991 .device
= TAKE_PTR(cd
),
1994 m
->decrypted_node
= TAKE_PTR(node
);
1995 close_and_replace(m
->mount_node_fd
, fd
);
2000 static int verity_can_reuse(
2001 const VeritySettings
*verity
,
2003 struct crypt_device
**ret_cd
) {
2005 /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
2006 _cleanup_free_
char *root_hash_existing
= NULL
;
2007 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2008 struct crypt_params_verity crypt_params
= {};
2009 size_t root_hash_existing_size
;
2016 r
= sym_crypt_init_by_name(&cd
, name
);
2018 return log_debug_errno(r
, "Error opening verity device, crypt_init_by_name failed: %m");
2020 cryptsetup_enable_logging(cd
);
2022 r
= sym_crypt_get_verity_info(cd
, &crypt_params
);
2024 return log_debug_errno(r
, "Error opening verity device, crypt_get_verity_info failed: %m");
2026 root_hash_existing_size
= verity
->root_hash_size
;
2027 root_hash_existing
= malloc0(root_hash_existing_size
);
2028 if (!root_hash_existing
)
2031 r
= sym_crypt_volume_key_get(cd
, CRYPT_ANY_SLOT
, root_hash_existing
, &root_hash_existing_size
, NULL
, 0);
2033 return log_debug_errno(r
, "Error opening verity device, crypt_volume_key_get failed: %m");
2034 if (verity
->root_hash_size
!= root_hash_existing_size
||
2035 memcmp(root_hash_existing
, verity
->root_hash
, verity
->root_hash_size
) != 0)
2036 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but root hashes are different.");
2038 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
2039 /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the
2040 * same settings, so that a previous unsigned mount will not be reused if the user asks to use
2041 * signing for the new one, and vice versa. */
2042 if (!!verity
->root_hash_sig
!= !!(crypt_params
.flags
& CRYPT_VERITY_ROOT_HASH_SIGNATURE
))
2043 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but signature settings are not the same.");
2046 *ret_cd
= TAKE_PTR(cd
);
2050 static inline char* dm_deferred_remove_clean(char *name
) {
2054 (void) sym_crypt_deactivate_by_name(NULL
, name
, CRYPT_DEACTIVATE_DEFERRED
);
2057 DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean
);
2059 static int validate_signature_userspace(const VeritySettings
*verity
) {
2061 _cleanup_(sk_X509_free_allp
) STACK_OF(X509
) *sk
= NULL
;
2062 _cleanup_strv_free_
char **certs
= NULL
;
2063 _cleanup_(PKCS7_freep
) PKCS7
*p7
= NULL
;
2064 _cleanup_free_
char *s
= NULL
;
2065 _cleanup_(BIO_freep
) BIO
*bio
= NULL
; /* 'bio' must be freed first, 's' second, hence keep this order
2066 * of declaration in place, please */
2067 const unsigned char *d
;
2071 assert(verity
->root_hash
);
2072 assert(verity
->root_hash_sig
);
2074 /* Because installing a signature certificate into the kernel chain is so messy, let's optionally do
2075 * userspace validation. */
2077 r
= conf_files_list_nulstr(&certs
, ".crt", NULL
, CONF_FILES_REGULAR
|CONF_FILES_FILTER_MASKED
, CONF_PATHS_NULSTR("verity.d"));
2079 return log_debug_errno(r
, "Failed to enumerate certificates: %m");
2080 if (strv_isempty(certs
)) {
2081 log_debug("No userspace dm-verity certificates found.");
2085 d
= verity
->root_hash_sig
;
2086 p7
= d2i_PKCS7(NULL
, &d
, (long) verity
->root_hash_sig_size
);
2088 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to parse PKCS7 DER signature data.");
2090 s
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2092 return log_oom_debug();
2094 bio
= BIO_new_mem_buf(s
, strlen(s
));
2096 return log_oom_debug();
2098 sk
= sk_X509_new_null();
2100 return log_oom_debug();
2102 STRV_FOREACH(i
, certs
) {
2103 _cleanup_(X509_freep
) X509
*c
= NULL
;
2104 _cleanup_fclose_
FILE *f
= NULL
;
2106 f
= fopen(*i
, "re");
2108 log_debug_errno(errno
, "Failed to open '%s', ignoring: %m", *i
);
2112 c
= PEM_read_X509(f
, NULL
, NULL
, NULL
);
2114 log_debug("Failed to load X509 certificate '%s', ignoring.", *i
);
2118 if (sk_X509_push(sk
, c
) == 0)
2119 return log_oom_debug();
2124 r
= PKCS7_verify(p7
, sk
, NULL
, bio
, NULL
, PKCS7_NOINTERN
|PKCS7_NOVERIFY
);
2126 log_debug("Userspace PKCS#7 validation succeeded.");
2128 log_debug("Userspace PKCS#7 validation failed: %s", ERR_error_string(ERR_get_error(), NULL
));
2132 log_debug("Not doing client-side validation of dm-verity root hash signatures, OpenSSL support disabled.");
2137 static int do_crypt_activate_verity(
2138 struct crypt_device
*cd
,
2140 const VeritySettings
*verity
) {
2142 bool check_signature
;
2149 if (verity
->root_hash_sig
) {
2150 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIGNATURE");
2151 if (r
< 0 && r
!= -ENXIO
)
2152 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIGNATURE");
2154 check_signature
= r
!= 0;
2156 check_signature
= false;
2158 if (check_signature
) {
2160 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
2161 /* First, if we have support for signed keys in the kernel, then try that first. */
2162 r
= sym_crypt_activate_by_signed_key(
2166 verity
->root_hash_size
,
2167 verity
->root_hash_sig
,
2168 verity
->root_hash_sig_size
,
2169 CRYPT_ACTIVATE_READONLY
);
2173 log_debug("Validation of dm-verity signature failed via the kernel, trying userspace validation instead.");
2175 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.",
2176 program_invocation_short_name
);
2179 /* So this didn't work via the kernel, then let's try userspace validation instead. If that
2180 * works we'll try to activate without telling the kernel the signature. */
2182 r
= validate_signature_userspace(verity
);
2186 return log_debug_errno(SYNTHETIC_ERRNO(ENOKEY
),
2187 "Activation of signed Verity volume worked neither via the kernel nor in userspace, can't activate.");
2190 return sym_crypt_activate_by_volume_key(
2194 verity
->root_hash_size
,
2195 CRYPT_ACTIVATE_READONLY
);
2198 static usec_t
verity_timeout(void) {
2199 usec_t t
= 100 * USEC_PER_MSEC
;
2203 /* On slower machines, like non-KVM vm, setting up device may take a long time.
2204 * Let's make the timeout configurable. */
2206 e
= getenv("SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC");
2210 r
= parse_sec(e
, &t
);
2213 "Failed to parse timeout specified in $SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC, "
2214 "using the default timeout (%s).",
2215 FORMAT_TIMESPAN(t
, USEC_PER_MSEC
));
2220 static int verity_partition(
2221 PartitionDesignator designator
,
2222 DissectedPartition
*m
,
2223 DissectedPartition
*v
,
2224 const VeritySettings
*verity
,
2225 DissectImageFlags flags
,
2226 DecryptedImage
*d
) {
2228 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2229 _cleanup_(dm_deferred_remove_cleanp
) char *restore_deferred_remove
= NULL
;
2230 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
2231 _cleanup_close_
int mount_node_fd
= -EBADF
;
2235 assert(v
|| (verity
&& verity
->data_path
));
2237 if (!verity
|| !verity
->root_hash
)
2239 if (!((verity
->designator
< 0 && designator
== PARTITION_ROOT
) ||
2240 (verity
->designator
== designator
)))
2243 if (!m
->found
|| !m
->node
|| !m
->fstype
)
2245 if (!verity
->data_path
) {
2246 if (!v
->found
|| !v
->node
|| !v
->fstype
)
2249 if (!streq(v
->fstype
, "DM_verity_hash"))
2253 r
= dlopen_cryptsetup();
2257 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
2258 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
2259 _cleanup_free_
char *root_hash_encoded
= NULL
;
2261 root_hash_encoded
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2262 if (!root_hash_encoded
)
2265 r
= make_dm_name_and_node(root_hash_encoded
, "-verity", &name
, &node
);
2267 r
= make_dm_name_and_node(m
->node
, "-verity", &name
, &node
);
2271 r
= sym_crypt_init(&cd
, verity
->data_path
?: v
->node
);
2275 cryptsetup_enable_logging(cd
);
2277 r
= sym_crypt_load(cd
, CRYPT_VERITY
, NULL
);
2281 r
= sym_crypt_set_data_device(cd
, m
->node
);
2285 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
2288 /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
2289 * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
2290 * retry a few times before giving up. */
2291 for (unsigned i
= 0; i
< N_DEVICE_NODE_LIST_ATTEMPTS
; i
++) {
2292 _cleanup_(sym_crypt_freep
) struct crypt_device
*existing_cd
= NULL
;
2293 _cleanup_close_
int fd
= -EBADF
;
2295 /* First, check if the device already exists. */
2296 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2297 if (fd
< 0 && !ERRNO_IS_DEVICE_ABSENT(errno
))
2298 return log_debug_errno(errno
, "Failed to open verity device %s: %m", node
);
2300 goto check
; /* The device already exists. Let's check it. */
2302 /* The symlink to the device node does not exist yet. Assume not activated, and let's activate it. */
2303 r
= do_crypt_activate_verity(cd
, name
, verity
);
2305 goto try_open
; /* The device is activated. Let's open it. */
2306 /* libdevmapper can return EINVAL when the device is already in the activation stage.
2307 * There's no way to distinguish this situation from a genuine error due to invalid
2308 * parameters, so immediately fall back to activating the device with a unique name.
2309 * Improvements in libcrypsetup can ensure this never happens:
2310 * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
2311 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2313 if (r
== -ENODEV
) /* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */
2316 -EEXIST
, /* Volume has already been opened and ready to be used. */
2317 -EBUSY
/* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */))
2318 return log_debug_errno(r
, "Failed to activate verity device %s: %m", node
);
2321 if (!restore_deferred_remove
){
2322 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
2323 r
= dm_deferred_remove_cancel(name
);
2324 /* -EBUSY and -ENXIO: the device has already been removed or being removed. We cannot
2325 * use the device, try to open again. See target_message() in drivers/md/dm-ioctl.c
2326 * and dm_cancel_deferred_remove() in drivers/md/dm.c */
2327 if (IN_SET(r
, -EBUSY
, -ENXIO
))
2330 return log_debug_errno(r
, "Failed to disable automated deferred removal for verity device %s: %m", node
);
2332 restore_deferred_remove
= strdup(name
);
2333 if (!restore_deferred_remove
)
2334 return log_oom_debug();
2337 r
= verity_can_reuse(verity
, name
, &existing_cd
);
2338 /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
2339 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2342 -ENOENT
, /* Removed?? */
2343 -EBUSY
, /* Volume is being opened but not ready, crypt_init_by_name() can fetch details. */
2344 -ENODEV
/* Volume is being opened but not ready, crypt_init_by_name() would fail, try to open again. */ ))
2347 return log_debug_errno(r
, "Failed to check if existing verity device %s can be reused: %m", node
);
2350 /* devmapper might say that the device exists, but the devlink might not yet have been
2351 * created. Check and wait for the udev event in that case. */
2352 r
= device_wait_for_devlink(node
, "block", verity_timeout(), NULL
);
2353 /* Fallback to activation with a unique device if it's taking too long */
2354 if (r
== -ETIMEDOUT
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2357 return log_debug_errno(r
, "Failed to wait device node symlink %s: %m", node
);
2362 /* Now, the device is activated and devlink is created. Let's open it. */
2363 fd
= open(node
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
2365 if (!ERRNO_IS_DEVICE_ABSENT(errno
))
2366 return log_debug_errno(errno
, "Failed to open verity device %s: %m", node
);
2368 /* The device has already been removed?? */
2373 mount_node_fd
= TAKE_FD(fd
);
2375 crypt_free_and_replace(cd
, existing_cd
);
2380 /* Device is being removed by another process. Let's wait for a while. */
2381 (void) usleep(2 * USEC_PER_MSEC
);
2384 /* All trials failed or a conflicting verity device exists. Let's try to activate with a unique name. */
2385 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
2386 /* Before trying to activate with unique name, we need to free crypt_device object.
2387 * Otherwise, we get error from libcryptsetup like the following:
2389 * systemd[1234]: Cannot use device /dev/loop5 which is in use (already mapped or mounted).
2394 return verity_partition(designator
, m
, v
, verity
, flags
& ~DISSECT_IMAGE_VERITY_SHARE
, d
);
2397 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
), "All attempts to activate verity device %s failed.", name
);
2400 /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
2401 restore_deferred_remove
= mfree(restore_deferred_remove
);
2403 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
2404 .name
= TAKE_PTR(name
),
2405 .device
= TAKE_PTR(cd
),
2408 m
->decrypted_node
= TAKE_PTR(node
);
2409 close_and_replace(m
->mount_node_fd
, mount_node_fd
);
2415 int dissected_image_decrypt(
2417 const char *passphrase
,
2418 const VeritySettings
*verity
,
2419 DissectImageFlags flags
) {
2421 #if HAVE_LIBCRYPTSETUP
2422 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
2427 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
2431 * = 0 → There was nothing to decrypt
2432 * > 0 → Decrypted successfully
2433 * -ENOKEY → There's something to decrypt but no key was supplied
2434 * -EKEYREJECTED → Passed key was not correct
2437 if (verity
&& verity
->root_hash
&& verity
->root_hash_size
< sizeof(sd_id128_t
))
2440 if (!m
->encrypted
&& !m
->verity_ready
)
2443 #if HAVE_LIBCRYPTSETUP
2444 r
= decrypted_image_new(&d
);
2448 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
2449 DissectedPartition
*p
= m
->partitions
+ i
;
2450 PartitionDesignator k
;
2455 r
= decrypt_partition(p
, passphrase
, flags
, d
);
2459 k
= partition_verity_of(i
);
2461 r
= verity_partition(i
, p
, m
->partitions
+ k
, verity
, flags
| DISSECT_IMAGE_VERITY_SHARE
, d
);
2466 if (!p
->decrypted_fstype
&& p
->mount_node_fd
>= 0 && p
->decrypted_node
) {
2467 r
= probe_filesystem_full(p
->mount_node_fd
, p
->decrypted_node
, 0, UINT64_MAX
, &p
->decrypted_fstype
);
2468 if (r
< 0 && r
!= -EUCLEAN
)
2473 m
->decrypted_image
= TAKE_PTR(d
);
2481 int dissected_image_decrypt_interactively(
2483 const char *passphrase
,
2484 const VeritySettings
*verity
,
2485 DissectImageFlags flags
) {
2487 _cleanup_strv_free_erase_
char **z
= NULL
;
2494 r
= dissected_image_decrypt(m
, passphrase
, verity
, flags
);
2497 if (r
== -EKEYREJECTED
)
2498 log_error_errno(r
, "Incorrect passphrase, try again!");
2499 else if (r
!= -ENOKEY
)
2500 return log_error_errno(r
, "Failed to decrypt image: %m");
2503 return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED
),
2504 "Too many retries.");
2508 r
= ask_password_auto("Please enter image passphrase:", NULL
, "dissect", "dissect", "dissect.passphrase", USEC_INFINITY
, 0, &z
);
2510 return log_error_errno(r
, "Failed to query for passphrase: %m");
2516 static int decrypted_image_relinquish(DecryptedImage
*d
) {
2519 /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a
2520 * boolean so that we don't clean it up ourselves either anymore */
2522 #if HAVE_LIBCRYPTSETUP
2525 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
2526 DecryptedPartition
*p
= d
->decrypted
+ i
;
2528 if (p
->relinquished
)
2531 r
= sym_crypt_deactivate_by_name(NULL
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
2533 return log_debug_errno(r
, "Failed to mark %s for auto-removal: %m", p
->name
);
2535 p
->relinquished
= true;
2542 int dissected_image_relinquish(DissectedImage
*m
) {
2547 if (m
->decrypted_image
) {
2548 r
= decrypted_image_relinquish(m
->decrypted_image
);
2554 loop_device_relinquish(m
->loop
);
2559 static char *build_auxiliary_path(const char *image
, const char *suffix
) {
2566 e
= endswith(image
, ".raw");
2568 return strjoin(e
, suffix
);
2570 n
= new(char, e
- image
+ strlen(suffix
) + 1);
2574 strcpy(mempcpy(n
, image
, e
- image
), suffix
);
2578 void verity_settings_done(VeritySettings
*v
) {
2581 v
->root_hash
= mfree(v
->root_hash
);
2582 v
->root_hash_size
= 0;
2584 v
->root_hash_sig
= mfree(v
->root_hash_sig
);
2585 v
->root_hash_sig_size
= 0;
2587 v
->data_path
= mfree(v
->data_path
);
2590 int verity_settings_load(
2591 VeritySettings
*verity
,
2593 const char *root_hash_path
,
2594 const char *root_hash_sig_path
) {
2596 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
2597 size_t root_hash_size
= 0, root_hash_sig_size
= 0;
2598 _cleanup_free_
char *verity_data_path
= NULL
;
2599 PartitionDesignator designator
;
2604 assert(verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
2606 /* If we are asked to load the root hash for a device node, exit early */
2607 if (is_device_path(image
))
2610 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIDECAR");
2611 if (r
< 0 && r
!= -ENXIO
)
2612 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIDECAR, ignoring: %m");
2616 designator
= verity
->designator
;
2618 /* We only fill in what isn't already filled in */
2620 if (!verity
->root_hash
) {
2621 _cleanup_free_
char *text
= NULL
;
2623 if (root_hash_path
) {
2624 /* If explicitly specified it takes precedence */
2625 r
= read_one_line_file(root_hash_path
, &text
);
2630 designator
= PARTITION_ROOT
;
2632 /* Otherwise look for xattr and separate file, and first for the data for root and if
2633 * that doesn't exist for /usr */
2635 if (designator
< 0 || designator
== PARTITION_ROOT
) {
2636 r
= getxattr_malloc(image
, "user.verity.roothash", &text
);
2638 _cleanup_free_
char *p
= NULL
;
2640 if (r
!= -ENOENT
&& !ERRNO_IS_XATTR_ABSENT(r
))
2643 p
= build_auxiliary_path(image
, ".roothash");
2647 r
= read_one_line_file(p
, &text
);
2648 if (r
< 0 && r
!= -ENOENT
)
2653 designator
= PARTITION_ROOT
;
2656 if (!text
&& (designator
< 0 || designator
== PARTITION_USR
)) {
2657 /* So in the "roothash" xattr/file name above the "root" of course primarily
2658 * refers to the root of the Verity Merkle tree. But coincidentally it also
2659 * is the hash for the *root* file system, i.e. the "root" neatly refers to
2660 * two distinct concepts called "root". Taking benefit of this happy
2661 * coincidence we call the file with the root hash for the /usr/ file system
2662 * `usrhash`, because `usrroothash` or `rootusrhash` would just be too
2663 * confusing. We thus drop the reference to the root of the Merkle tree, and
2664 * just indicate which file system it's about. */
2665 r
= getxattr_malloc(image
, "user.verity.usrhash", &text
);
2667 _cleanup_free_
char *p
= NULL
;
2669 if (r
!= -ENOENT
&& !ERRNO_IS_XATTR_ABSENT(r
))
2672 p
= build_auxiliary_path(image
, ".usrhash");
2676 r
= read_one_line_file(p
, &text
);
2677 if (r
< 0 && r
!= -ENOENT
)
2682 designator
= PARTITION_USR
;
2687 r
= unhexmem(text
, strlen(text
), &root_hash
, &root_hash_size
);
2690 if (root_hash_size
< sizeof(sd_id128_t
))
2695 if ((root_hash
|| verity
->root_hash
) && !verity
->root_hash_sig
) {
2696 if (root_hash_sig_path
) {
2697 r
= read_full_file(root_hash_sig_path
, (char**) &root_hash_sig
, &root_hash_sig_size
);
2698 if (r
< 0 && r
!= -ENOENT
)
2702 designator
= PARTITION_ROOT
;
2704 if (designator
< 0 || designator
== PARTITION_ROOT
) {
2705 _cleanup_free_
char *p
= NULL
;
2707 /* Follow naming convention recommended by the relevant RFC:
2708 * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
2709 p
= build_auxiliary_path(image
, ".roothash.p7s");
2713 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
2714 if (r
< 0 && r
!= -ENOENT
)
2717 designator
= PARTITION_ROOT
;
2720 if (!root_hash_sig
&& (designator
< 0 || designator
== PARTITION_USR
)) {
2721 _cleanup_free_
char *p
= NULL
;
2723 p
= build_auxiliary_path(image
, ".usrhash.p7s");
2727 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
2728 if (r
< 0 && r
!= -ENOENT
)
2731 designator
= PARTITION_USR
;
2735 if (root_hash_sig
&& root_hash_sig_size
== 0) /* refuse empty size signatures */
2739 if (!verity
->data_path
) {
2740 _cleanup_free_
char *p
= NULL
;
2742 p
= build_auxiliary_path(image
, ".verity");
2746 if (access(p
, F_OK
) < 0) {
2747 if (errno
!= ENOENT
)
2750 verity_data_path
= TAKE_PTR(p
);
2754 verity
->root_hash
= TAKE_PTR(root_hash
);
2755 verity
->root_hash_size
= root_hash_size
;
2758 if (root_hash_sig
) {
2759 verity
->root_hash_sig
= TAKE_PTR(root_hash_sig
);
2760 verity
->root_hash_sig_size
= root_hash_sig_size
;
2763 if (verity_data_path
)
2764 verity
->data_path
= TAKE_PTR(verity_data_path
);
2766 if (verity
->designator
< 0)
2767 verity
->designator
= designator
;
2772 int dissected_image_load_verity_sig_partition(
2775 VeritySettings
*verity
) {
2777 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
2778 _cleanup_(json_variant_unrefp
) JsonVariant
*v
= NULL
;
2779 size_t root_hash_size
, root_hash_sig_size
;
2780 _cleanup_free_
char *buf
= NULL
;
2781 PartitionDesignator d
;
2782 DissectedPartition
*p
;
2783 JsonVariant
*rh
, *sig
;
2792 if (verity
->root_hash
&& verity
->root_hash_sig
) /* Already loaded? */
2795 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_EMBEDDED");
2796 if (r
< 0 && r
!= -ENXIO
)
2797 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_EMBEDDED, ignoring: %m");
2801 d
= partition_verity_sig_of(verity
->designator
< 0 ? PARTITION_ROOT
: verity
->designator
);
2804 p
= m
->partitions
+ d
;
2807 if (p
->offset
== UINT64_MAX
|| p
->size
== UINT64_MAX
)
2810 if (p
->size
> 4*1024*1024) /* Signature data cannot possible be larger than 4M, refuse that */
2813 buf
= new(char, p
->size
+1);
2817 n
= pread(fd
, buf
, p
->size
, p
->offset
);
2820 if ((uint64_t) n
!= p
->size
)
2823 e
= memchr(buf
, 0, p
->size
);
2825 /* If we found a NUL byte then the rest of the data must be NUL too */
2826 if (!memeqzero(e
, p
->size
- (e
- buf
)))
2827 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature data contains embedded NUL byte.");
2831 r
= json_parse(buf
, 0, &v
, NULL
, NULL
);
2833 return log_debug_errno(r
, "Failed to parse signature JSON data: %m");
2835 rh
= json_variant_by_key(v
, "rootHash");
2837 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'rootHash' field.");
2838 if (!json_variant_is_string(rh
))
2839 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'rootHash' field of signature JSON object is not a string.");
2841 r
= unhexmem(json_variant_string(rh
), SIZE_MAX
, &root_hash
, &root_hash_size
);
2843 return log_debug_errno(r
, "Failed to parse root hash field: %m");
2845 /* Check if specified root hash matches if it is specified */
2846 if (verity
->root_hash
&&
2847 memcmp_nn(verity
->root_hash
, verity
->root_hash_size
, root_hash
, root_hash_size
) != 0) {
2848 _cleanup_free_
char *a
= NULL
, *b
= NULL
;
2850 a
= hexmem(root_hash
, root_hash_size
);
2851 b
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2853 return log_debug_errno(r
, "Root hash in signature JSON data (%s) doesn't match configured hash (%s).", strna(a
), strna(b
));
2856 sig
= json_variant_by_key(v
, "signature");
2858 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'signature' field.");
2859 if (!json_variant_is_string(sig
))
2860 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'signature' field of signature JSON object is not a string.");
2862 r
= unbase64mem(json_variant_string(sig
), SIZE_MAX
, &root_hash_sig
, &root_hash_sig_size
);
2864 return log_debug_errno(r
, "Failed to parse signature field: %m");
2866 free_and_replace(verity
->root_hash
, root_hash
);
2867 verity
->root_hash_size
= root_hash_size
;
2869 free_and_replace(verity
->root_hash_sig
, root_hash_sig
);
2870 verity
->root_hash_sig_size
= root_hash_sig_size
;
2875 int dissected_image_acquire_metadata(DissectedImage
*m
, DissectImageFlags extra_flags
) {
2882 META_INITRD_RELEASE
,
2883 META_EXTENSION_RELEASE
,
2884 META_HAS_INIT_SYSTEM
,
2888 static const char *const paths
[_META_MAX
] = {
2889 [META_HOSTNAME
] = "/etc/hostname\0",
2890 [META_MACHINE_ID
] = "/etc/machine-id\0",
2891 [META_MACHINE_INFO
] = "/etc/machine-info\0",
2892 [META_OS_RELEASE
] = ("/etc/os-release\0"
2893 "/usr/lib/os-release\0"),
2894 [META_INITRD_RELEASE
] = ("/etc/initrd-release\0"
2895 "/usr/lib/initrd-release\0"),
2896 [META_EXTENSION_RELEASE
] = "extension-release\0", /* Used only for logging. */
2897 [META_HAS_INIT_SYSTEM
] = "has-init-system\0", /* ditto */
2900 _cleanup_strv_free_
char **machine_info
= NULL
, **os_release
= NULL
, **initrd_release
= NULL
, **extension_release
= NULL
;
2901 _cleanup_close_pair_
int error_pipe
[2] = PIPE_EBADF
;
2902 _cleanup_(rmdir_and_freep
) char *t
= NULL
;
2903 _cleanup_(sigkill_waitp
) pid_t child
= 0;
2904 sd_id128_t machine_id
= SD_ID128_NULL
;
2905 _cleanup_free_
char *hostname
= NULL
;
2906 unsigned n_meta_initialized
= 0;
2907 int fds
[2 * _META_MAX
], r
, v
;
2908 int has_init_system
= -1;
2911 BLOCK_SIGNALS(SIGCHLD
);
2915 for (; n_meta_initialized
< _META_MAX
; n_meta_initialized
++) {
2916 if (!paths
[n_meta_initialized
]) {
2917 fds
[2*n_meta_initialized
] = fds
[2*n_meta_initialized
+1] = -EBADF
;
2921 if (pipe2(fds
+ 2*n_meta_initialized
, O_CLOEXEC
) < 0) {
2927 r
= mkdtemp_malloc("/tmp/dissect-XXXXXX", &t
);
2931 if (pipe2(error_pipe
, O_CLOEXEC
) < 0) {
2936 r
= safe_fork("(sd-dissect)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_NEW_MOUNTNS
|FORK_MOUNTNS_SLAVE
, &child
);
2940 /* Child in a new mount namespace */
2941 error_pipe
[0] = safe_close(error_pipe
[0]);
2943 r
= dissected_image_mount(
2949 DISSECT_IMAGE_READ_ONLY
|
2950 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
2951 DISSECT_IMAGE_USR_NO_ROOT
);
2953 log_debug_errno(r
, "Failed to mount dissected image: %m");
2957 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
2958 _cleanup_close_
int fd
= -ENOENT
;
2963 fds
[2*k
] = safe_close(fds
[2*k
]);
2967 case META_EXTENSION_RELEASE
:
2968 /* As per the os-release spec, if the image is an extension it will have a file
2969 * named after the image name in extension-release.d/ - we use the image name
2970 * and try to resolve it with the extension-release helpers, as sometimes
2971 * the image names are mangled on deployment and do not match anymore.
2972 * Unlike other paths this is not fixed, and the image name
2973 * can be mangled on deployment, so by calling into the helper
2974 * we allow a fallback that matches on the first extension-release
2975 * file found in the directory, if one named after the image cannot
2976 * be found first. */
2977 r
= open_extension_release(t
, m
->image_name
, /* relax_extension_release_check= */ false, NULL
, &fd
);
2979 fd
= r
; /* Propagate the error. */
2982 case META_HAS_INIT_SYSTEM
: {
2985 FOREACH_STRING(init
,
2986 "/usr/lib/systemd/systemd", /* systemd on /usr merged system */
2987 "/lib/systemd/systemd", /* systemd on /usr non-merged systems */
2988 "/sbin/init") { /* traditional path the Linux kernel invokes */
2990 r
= chase_symlinks(init
, t
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
2993 log_debug_errno(r
, "Failed to resolve %s, ignoring: %m", init
);
3000 r
= loop_write(fds
[2*k
+1], &found
, sizeof(found
), false);
3008 NULSTR_FOREACH(p
, paths
[k
]) {
3009 fd
= chase_symlinks_and_open(p
, t
, CHASE_PREFIX_ROOT
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
, NULL
);
3016 log_debug_errno(fd
, "Failed to read %s file of image, ignoring: %m", paths
[k
]);
3017 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3021 r
= copy_bytes(fd
, fds
[2*k
+1], UINT64_MAX
, 0);
3025 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3028 _exit(EXIT_SUCCESS
);
3031 /* Let parent know the error */
3032 (void) write(error_pipe
[1], &r
, sizeof(r
));
3033 _exit(EXIT_FAILURE
);
3036 error_pipe
[1] = safe_close(error_pipe
[1]);
3038 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
3039 _cleanup_fclose_
FILE *f
= NULL
;
3044 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3046 f
= take_fdopen(&fds
[2*k
], "r");
3055 r
= read_etc_hostname_stream(f
, &hostname
);
3057 log_debug_errno(r
, "Failed to read /etc/hostname of image: %m");
3061 case META_MACHINE_ID
: {
3062 _cleanup_free_
char *line
= NULL
;
3064 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3066 log_debug_errno(r
, "Failed to read /etc/machine-id of image: %m");
3068 r
= sd_id128_from_string(line
, &machine_id
);
3070 log_debug_errno(r
, "Image contains invalid /etc/machine-id: %s", line
);
3072 log_debug("/etc/machine-id file of image is empty.");
3073 else if (streq(line
, "uninitialized"))
3074 log_debug("/etc/machine-id file of image is uninitialized (likely aborted first boot).");
3076 log_debug("/etc/machine-id file of image has unexpected length %i.", r
);
3081 case META_MACHINE_INFO
:
3082 r
= load_env_file_pairs(f
, "machine-info", &machine_info
);
3084 log_debug_errno(r
, "Failed to read /etc/machine-info of image: %m");
3088 case META_OS_RELEASE
:
3089 r
= load_env_file_pairs(f
, "os-release", &os_release
);
3091 log_debug_errno(r
, "Failed to read OS release file of image: %m");
3095 case META_INITRD_RELEASE
:
3096 r
= load_env_file_pairs(f
, "initrd-release", &initrd_release
);
3098 log_debug_errno(r
, "Failed to read initrd release file of image: %m");
3102 case META_EXTENSION_RELEASE
:
3103 r
= load_env_file_pairs(f
, "extension-release", &extension_release
);
3105 log_debug_errno(r
, "Failed to read extension release file of image: %m");
3109 case META_HAS_INIT_SYSTEM
: {
3114 nr
= fread(&b
, 1, sizeof(b
), f
);
3115 if (nr
!= sizeof(b
))
3116 log_debug_errno(errno_or_else(EIO
), "Failed to read has-init-system boolean: %m");
3118 has_init_system
= b
;
3124 r
= wait_for_terminate_and_check("(sd-dissect)", child
, 0);
3129 n
= read(error_pipe
[0], &v
, sizeof(v
));
3133 return v
; /* propagate error sent to us from child */
3137 if (r
!= EXIT_SUCCESS
)
3140 free_and_replace(m
->hostname
, hostname
);
3141 m
->machine_id
= machine_id
;
3142 strv_free_and_replace(m
->machine_info
, machine_info
);
3143 strv_free_and_replace(m
->os_release
, os_release
);
3144 strv_free_and_replace(m
->initrd_release
, initrd_release
);
3145 strv_free_and_replace(m
->extension_release
, extension_release
);
3146 m
->has_init_system
= has_init_system
;
3149 for (unsigned k
= 0; k
< n_meta_initialized
; k
++)
3150 safe_close_pair(fds
+ 2*k
);
3155 int dissect_loop_device(
3157 const VeritySettings
*verity
,
3158 const MountOptions
*mount_options
,
3159 DissectImageFlags flags
,
3160 DissectedImage
**ret
) {
3163 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
3169 r
= dissected_image_new(loop
->backing_file
?: loop
->node
, &m
);
3173 m
->loop
= loop_device_ref(loop
);
3174 m
->sector_size
= m
->loop
->sector_size
;
3176 r
= dissect_image(m
, loop
->fd
, loop
->node
, verity
, mount_options
, flags
);
3187 int dissect_loop_device_and_warn(
3189 const VeritySettings
*verity
,
3190 const MountOptions
*mount_options
,
3191 DissectImageFlags flags
,
3192 DissectedImage
**ret
) {
3198 assert(loop
->fd
>= 0);
3200 name
= ASSERT_PTR(loop
->backing_file
?: loop
->node
);
3202 r
= dissect_loop_device(loop
, verity
, mount_options
, flags
, ret
);
3206 return log_error_errno(r
, "Dissecting images is not supported, compiled without blkid support.");
3209 return log_error_errno(r
, "%s: Couldn't identify a suitable partition table or file system.", name
);
3212 return log_error_errno(r
, "%s: The image does not pass validation.", name
);
3214 case -EADDRNOTAVAIL
:
3215 return log_error_errno(r
, "%s: No root partition for specified root hash found.", name
);
3218 return log_error_errno(r
, "%s: Multiple suitable root partitions found in image.", name
);
3221 return log_error_errno(r
, "%s: No suitable root partition found in image.", name
);
3223 case -EPROTONOSUPPORT
:
3224 return log_error_errno(r
, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name
);
3227 return log_error_errno(r
, "%s: Image is not a block device.", name
);
3230 return log_error_errno(r
,
3231 "Combining partitioned images (such as '%s') with external Verity data (such as '%s') not supported. "
3232 "(Consider setting $SYSTEMD_DISSECT_VERITY_SIDECAR=0 to disable automatic discovery of external Verity data.)",
3233 name
, strna(verity
? verity
->data_path
: NULL
));
3237 return log_error_errno(r
, "Failed to dissect image '%s': %m", name
);
3243 bool dissected_image_verity_candidate(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3246 /* Checks if this partition could theoretically do Verity. For non-partitioned images this only works
3247 * if there's an external verity file supplied, for which we can consult .has_verity. For partitioned
3248 * images we only check the partition type.
3250 * This call is used to decide whether to suppress or show a verity column in tabular output of the
3253 if (image
->single_file_system
)
3254 return partition_designator
== PARTITION_ROOT
&& image
->has_verity
;
3256 return partition_verity_of(partition_designator
) >= 0;
3259 bool dissected_image_verity_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3260 PartitionDesignator k
;
3264 /* Checks if this partition has verity data available that we can activate. For non-partitioned this
3265 * works for the root partition, for others only if the associated verity partition was found. */
3267 if (!image
->verity_ready
)
3270 if (image
->single_file_system
)
3271 return partition_designator
== PARTITION_ROOT
;
3273 k
= partition_verity_of(partition_designator
);
3274 return k
>= 0 && image
->partitions
[k
].found
;
3277 bool dissected_image_verity_sig_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3278 PartitionDesignator k
;
3282 /* Checks if this partition has verity signature data available that we can use. */
3284 if (!image
->verity_sig_ready
)
3287 if (image
->single_file_system
)
3288 return partition_designator
== PARTITION_ROOT
;
3290 k
= partition_verity_sig_of(partition_designator
);
3291 return k
>= 0 && image
->partitions
[k
].found
;
3294 MountOptions
* mount_options_free_all(MountOptions
*options
) {
3297 while ((m
= options
)) {
3298 LIST_REMOVE(mount_options
, options
, m
);
3306 const char* mount_options_from_designator(const MountOptions
*options
, PartitionDesignator designator
) {
3307 LIST_FOREACH(mount_options
, m
, options
)
3308 if (designator
== m
->partition_designator
&& !isempty(m
->options
))
3314 int mount_image_privately_interactively(
3316 DissectImageFlags flags
,
3317 char **ret_directory
,
3318 LoopDevice
**ret_loop_device
) {
3320 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3321 _cleanup_(loop_device_unrefp
) LoopDevice
*d
= NULL
;
3322 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3323 _cleanup_(rmdir_and_freep
) char *created_dir
= NULL
;
3324 _cleanup_free_
char *temp
= NULL
;
3327 /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
3328 * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
3332 assert(ret_directory
);
3333 assert(ret_loop_device
);
3335 /* We intend to mount this right-away, hence add the partitions if needed and pin them. */
3336 flags
|= DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
3337 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
3339 r
= verity_settings_load(&verity
, image
, NULL
, NULL
);
3341 return log_error_errno(r
, "Failed to load root hash data: %m");
3343 r
= tempfn_random_child(NULL
, program_invocation_short_name
, &temp
);
3345 return log_error_errno(r
, "Failed to generate temporary mount directory: %m");
3347 r
= loop_device_make_by_path(
3349 FLAGS_SET(flags
, DISSECT_IMAGE_DEVICE_READ_ONLY
) ? O_RDONLY
: O_RDWR
,
3350 /* sector_size= */ UINT32_MAX
,
3351 FLAGS_SET(flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
3355 return log_error_errno(r
, "Failed to set up loopback device for %s: %m", image
);
3357 r
= dissect_loop_device_and_warn(d
, &verity
, NULL
, flags
, &dissected_image
);
3361 r
= dissected_image_load_verity_sig_partition(dissected_image
, d
->fd
, &verity
);
3365 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, &verity
, flags
);
3369 r
= detach_mount_namespace();
3371 return log_error_errno(r
, "Failed to detach mount namespace: %m");
3373 r
= mkdir_p(temp
, 0700);
3375 return log_error_errno(r
, "Failed to create mount point: %m");
3377 created_dir
= TAKE_PTR(temp
);
3379 r
= dissected_image_mount_and_warn(dissected_image
, created_dir
, UID_INVALID
, UID_INVALID
, flags
);
3383 r
= loop_device_flock(d
, LOCK_UN
);
3387 r
= dissected_image_relinquish(dissected_image
);
3389 return log_error_errno(r
, "Failed to relinquish DM and loopback block devices: %m");
3391 *ret_directory
= TAKE_PTR(created_dir
);
3392 *ret_loop_device
= TAKE_PTR(d
);
3397 static bool mount_options_relax_extension_release_checks(const MountOptions
*options
) {
3401 return string_contains_word(mount_options_from_designator(options
, PARTITION_ROOT
), ",", "x-systemd.relax-extension-release-check") ||
3402 string_contains_word(mount_options_from_designator(options
, PARTITION_USR
), ",", "x-systemd.relax-extension-release-check") ||
3403 string_contains_word(options
->options
, ",", "x-systemd.relax-extension-release-check");
3406 int verity_dissect_and_mount(
3410 const MountOptions
*options
,
3411 const char *required_host_os_release_id
,
3412 const char *required_host_os_release_version_id
,
3413 const char *required_host_os_release_sysext_level
,
3414 const char *required_sysext_scope
) {
3416 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
3417 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3418 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3419 DissectImageFlags dissect_image_flags
;
3420 bool relax_extension_release_check
;
3426 relax_extension_release_check
= mount_options_relax_extension_release_checks(options
);
3428 /* We might get an FD for the image, but we use the original path to look for the dm-verity files */
3429 r
= verity_settings_load(&verity
, src
, NULL
, NULL
);
3431 return log_debug_errno(r
, "Failed to load root hash: %m");
3433 dissect_image_flags
= (verity
.data_path
? DISSECT_IMAGE_NO_PARTITION_TABLE
: 0) |
3434 (relax_extension_release_check
? DISSECT_IMAGE_RELAX_SYSEXT_CHECK
: 0) |
3435 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
3436 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
3438 /* Note that we don't use loop_device_make here, as the FD is most likely O_PATH which would not be
3439 * accepted by LOOP_CONFIGURE, so just let loop_device_make_by_path reopen it as a regular FD. */
3440 r
= loop_device_make_by_path(
3441 src_fd
>= 0 ? FORMAT_PROC_FD_PATH(src_fd
) : src
,
3442 /* open_flags= */ -1,
3443 /* sector_size= */ UINT32_MAX
,
3444 verity
.data_path
? 0 : LO_FLAGS_PARTSCAN
,
3448 return log_debug_errno(r
, "Failed to create loop device for image: %m");
3450 r
= dissect_loop_device(
3454 dissect_image_flags
,
3456 /* No partition table? Might be a single-filesystem image, try again */
3457 if (!verity
.data_path
&& r
== -ENOPKG
)
3458 r
= dissect_loop_device(
3462 dissect_image_flags
| DISSECT_IMAGE_NO_PARTITION_TABLE
,
3465 return log_debug_errno(r
, "Failed to dissect image: %m");
3467 r
= dissected_image_load_verity_sig_partition(dissected_image
, loop_device
->fd
, &verity
);
3471 r
= dissected_image_decrypt(
3475 dissect_image_flags
);
3477 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
3479 r
= mkdir_p_label(dest
, 0755);
3481 return log_debug_errno(r
, "Failed to create destination directory %s: %m", dest
);
3482 r
= umount_recursive(dest
, 0);
3484 return log_debug_errno(r
, "Failed to umount under destination directory %s: %m", dest
);
3486 r
= dissected_image_mount(dissected_image
, dest
, UID_INVALID
, UID_INVALID
, dissect_image_flags
);
3488 return log_debug_errno(r
, "Failed to mount image: %m");
3490 r
= loop_device_flock(loop_device
, LOCK_UN
);
3492 return log_debug_errno(r
, "Failed to unlock loopback device: %m");
3494 /* If we got os-release values from the caller, then we need to match them with the image's
3495 * extension-release.d/ content. Return -EINVAL if there's any mismatch.
3496 * First, check the distro ID. If that matches, then check the new SYSEXT_LEVEL value if
3497 * available, or else fallback to VERSION_ID. If neither is present (eg: rolling release),
3498 * then a simple match on the ID will be performed. */
3499 if (required_host_os_release_id
) {
3500 _cleanup_strv_free_
char **extension_release
= NULL
;
3502 assert(!isempty(required_host_os_release_id
));
3504 r
= load_extension_release_pairs(dest
, dissected_image
->image_name
, relax_extension_release_check
, &extension_release
);
3506 return log_debug_errno(r
, "Failed to parse image %s extension-release metadata: %m", dissected_image
->image_name
);
3508 r
= extension_release_validate(
3509 dissected_image
->image_name
,
3510 required_host_os_release_id
,
3511 required_host_os_release_version_id
,
3512 required_host_os_release_sysext_level
,
3513 required_sysext_scope
,
3516 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE
), "Image %s extension-release metadata does not match the root's", dissected_image
->image_name
);
3518 return log_debug_errno(r
, "Failed to compare image %s extension-release metadata with the root's os-release: %m", dissected_image
->image_name
);
3521 r
= dissected_image_relinquish(dissected_image
);
3523 return log_debug_errno(r
, "Failed to relinquish dissected image: %m");