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 "chase-symlinks.h"
29 #include "conf-files.h"
31 #include "cryptsetup-util.h"
33 #include "device-nodes.h"
34 #include "device-util.h"
35 #include "discover-image.h"
36 #include "dissect-image.h"
40 #include "extension-release.h"
44 #include "fsck-util.h"
46 #include "hexdecoct.h"
47 #include "hostname-setup.h"
48 #include "id128-util.h"
49 #include "import-util.h"
51 #include "mkdir-label.h"
52 #include "mount-util.h"
53 #include "mountpoint-util.h"
54 #include "namespace-util.h"
55 #include "nulstr-util.h"
56 #include "openssl-util.h"
58 #include "path-util.h"
59 #include "process-util.h"
60 #include "raw-clone.h"
61 #include "resize-fs.h"
62 #include "signal-util.h"
63 #include "stat-util.h"
64 #include "stdio-util.h"
65 #include "string-table.h"
66 #include "string-util.h"
68 #include "tmpfile-util.h"
69 #include "udev-util.h"
70 #include "user-util.h"
71 #include "xattr-util.h"
73 /* how many times to wait for the device nodes to appear */
74 #define N_DEVICE_NODE_LIST_ATTEMPTS 10
76 int probe_filesystem(const char *node
, char **ret_fstype
) {
77 /* Try to find device content type and return it in *ret_fstype. If nothing is found,
78 * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and an
79 * different error otherwise. */
82 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
87 b
= blkid_new_probe_from_filename(node
);
89 return errno_or_else(ENOMEM
);
91 blkid_probe_enable_superblocks(b
, 1);
92 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
);
95 r
= blkid_do_safeprobe(b
);
99 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
),
100 "Results ambiguous for partition %s", node
);
102 return log_debug_errno(errno_or_else(EIO
), "Failed to probe partition %s: %m", node
);
104 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
109 log_debug("Probed fstype '%s' on partition %s.", fstype
, node
);
120 log_debug("No type detected on partition %s", node
);
129 static void check_partition_flags(
131 unsigned long long pflags
,
132 unsigned long long supported
) {
136 /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
137 pflags
&= ~(supported
|
138 SD_GPT_FLAG_REQUIRED_PARTITION
|
139 SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL
|
140 SD_GPT_FLAG_LEGACY_BIOS_BOOTABLE
);
145 /* If there are other bits set, then log about it, to make things discoverable */
146 for (unsigned i
= 0; i
< sizeof(pflags
) * 8; i
++) {
147 unsigned long long bit
= 1ULL << i
;
148 if (!FLAGS_SET(pflags
, bit
))
151 log_debug("Unexpected partition flag %llu set on %s!", bit
, node
);
157 static int dissected_image_new(const char *path
, DissectedImage
**ret
) {
158 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
159 _cleanup_free_
char *name
= NULL
;
165 _cleanup_free_
char *filename
= NULL
;
167 r
= path_extract_filename(path
, &filename
);
171 r
= raw_strip_suffixes(filename
, &name
);
175 if (!image_name_is_valid(name
)) {
176 log_debug("Image name %s is not valid, ignoring.", strna(name
));
181 m
= new(DissectedImage
, 1);
185 *m
= (DissectedImage
) {
186 .has_init_system
= -1,
187 .image_name
= TAKE_PTR(name
),
190 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++)
191 m
->partitions
[i
] = DISSECTED_PARTITION_NULL
;
198 static void dissected_partition_done(DissectedPartition
*p
) {
204 free(p
->decrypted_fstype
);
205 free(p
->decrypted_node
);
206 free(p
->mount_options
);
208 *p
= DISSECTED_PARTITION_NULL
;
212 static int make_partition_devname(
213 const char *whole_devname
,
219 assert(whole_devname
);
222 /* Given a whole block device node name (e.g. /dev/sda or /dev/loop7) generate a partition device
223 * name (e.g. /dev/sda7 or /dev/loop7p5). The rule the kernel uses is simple: if whole block device
224 * node name ends in a digit, then suffix a 'p', followed by the partition number. Otherwise, just
225 * suffix the partition number without any 'p'. */
227 if (isempty(whole_devname
)) /* Make sure there *is* a last char */
230 need_p
= ascii_isdigit(whole_devname
[strlen(whole_devname
)-1]); /* Last char a digit? */
232 return asprintf(ret
, "%s%s%i", whole_devname
, need_p
? "p" : "", nr
);
239 const char *image_path
,
240 const VeritySettings
*verity
,
241 const MountOptions
*mount_options
,
242 DissectImageFlags flags
,
243 DissectedImage
**ret
) {
246 sd_id128_t root_uuid
= SD_ID128_NULL
, root_verity_uuid
= SD_ID128_NULL
;
247 sd_id128_t usr_uuid
= SD_ID128_NULL
, usr_verity_uuid
= SD_ID128_NULL
;
248 bool is_gpt
, is_mbr
, multiple_generic
= false,
249 generic_rw
= false, /* initialize to appease gcc */
250 generic_growfs
= false;
251 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
252 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
253 _cleanup_free_
char *generic_node
= NULL
;
254 sd_id128_t generic_uuid
= SD_ID128_NULL
;
255 const char *pttype
= NULL
;
257 int r
, generic_nr
= -1, n_partitions
;
262 assert(!verity
|| verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
263 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
264 assert(!verity
|| verity
->root_hash_sig
|| verity
->root_hash_sig_size
== 0);
265 assert(!verity
|| (verity
->root_hash
|| !verity
->root_hash_sig
));
266 assert(!((flags
& DISSECT_IMAGE_GPT_ONLY
) && (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)));
268 /* Probes a disk image, and returns information about what it found in *ret.
270 * Returns -ENOPKG if no suitable partition table or file system could be found.
271 * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found.
272 * Returns -ENXIO if we couldn't find any partition suitable as root or /usr partition
273 * Returns -ENOTUNIQ if we only found multiple generic partitions and thus don't know what to do with that */
275 if (verity
&& verity
->root_hash
) {
276 sd_id128_t fsuuid
, vuuid
;
278 /* If a root hash is supplied, then we use the root partition that has a UUID that match the
279 * first 128bit of the root hash. And we use the verity partition that has a UUID that match
280 * the final 128bit. */
282 if (verity
->root_hash_size
< sizeof(sd_id128_t
))
285 memcpy(&fsuuid
, verity
->root_hash
, sizeof(sd_id128_t
));
286 memcpy(&vuuid
, (const uint8_t*) verity
->root_hash
+ verity
->root_hash_size
- sizeof(sd_id128_t
), sizeof(sd_id128_t
));
288 if (sd_id128_is_null(fsuuid
))
290 if (sd_id128_is_null(vuuid
))
293 /* If the verity data declares it's for the /usr partition, then search for that, in all
294 * other cases assume it's for the root partition. */
295 if (verity
->designator
== PARTITION_USR
) {
297 usr_verity_uuid
= vuuid
;
300 root_verity_uuid
= vuuid
;
304 b
= blkid_new_probe();
309 r
= blkid_probe_set_device(b
, fd
, 0, 0);
311 return errno_or_else(ENOMEM
);
313 if ((flags
& DISSECT_IMAGE_GPT_ONLY
) == 0) {
314 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
315 blkid_probe_enable_superblocks(b
, 1);
316 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
|BLKID_SUBLKS_USAGE
);
319 blkid_probe_enable_partitions(b
, 1);
320 blkid_probe_set_partitions_flags(b
, BLKID_PARTS_ENTRY_DETAILS
);
323 r
= blkid_do_safeprobe(b
);
324 if (IN_SET(r
, -2, 1))
325 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG
), "Failed to identify any partition table.");
327 return errno_or_else(EIO
);
329 r
= dissected_image_new(image_path
, &m
);
333 if ((!(flags
& DISSECT_IMAGE_GPT_ONLY
) &&
334 (flags
& DISSECT_IMAGE_GENERIC_ROOT
)) ||
335 (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)) {
336 const char *usage
= NULL
;
338 /* If flags permit this, also allow using non-partitioned single-filesystem images */
340 (void) blkid_probe_lookup_value(b
, "USAGE", &usage
, NULL
);
341 if (STRPTR_IN_SET(usage
, "filesystem", "crypto")) {
342 _cleanup_free_
char *t
= NULL
, *n
= NULL
, *o
= NULL
;
343 const char *fstype
= NULL
, *options
= NULL
;
345 /* OK, we have found a file system, that's our root partition then. */
346 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
358 m
->single_file_system
= true;
359 m
->encrypted
= streq_ptr(fstype
, "crypto_LUKS");
361 m
->has_verity
= verity
&& verity
->data_path
;
362 m
->verity_ready
= m
->has_verity
&&
364 (verity
->designator
< 0 || verity
->designator
== PARTITION_ROOT
);
366 m
->has_verity_sig
= false; /* signature not embedded, must be specified */
367 m
->verity_sig_ready
= m
->verity_ready
&&
368 verity
->root_hash_sig
;
370 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
377 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
379 .rw
= !m
->verity_ready
&& !fstype_is_ro(fstype
),
381 .architecture
= _ARCHITECTURE_INVALID
,
382 .fstype
= TAKE_PTR(t
),
384 .mount_options
= TAKE_PTR(o
),
394 (void) blkid_probe_lookup_value(b
, "PTTYPE", &pttype
, NULL
);
398 is_gpt
= streq_ptr(pttype
, "gpt");
399 is_mbr
= streq_ptr(pttype
, "dos");
401 if (!is_gpt
&& ((flags
& DISSECT_IMAGE_GPT_ONLY
) || !is_mbr
))
404 /* We support external verity data partitions only if the image has no partition table */
405 if (verity
&& verity
->data_path
)
408 /* Safety check: refuse block devices that carry a partition table but for which the kernel doesn't
409 * do partition scanning. */
410 r
= blockdev_partscan_enabled(fd
);
414 return -EPROTONOSUPPORT
;
417 pl
= blkid_probe_get_partitions(b
);
419 return errno_or_else(ENOMEM
);
422 n_partitions
= blkid_partlist_numof_partitions(pl
);
423 if (n_partitions
< 0)
424 return errno_or_else(EIO
);
426 for (int i
= 0; i
< n_partitions
; i
++) {
427 _cleanup_free_
char *node
= NULL
;
428 unsigned long long pflags
;
429 blkid_loff_t start
, size
;
434 pp
= blkid_partlist_get_partition(pl
, i
);
436 return errno_or_else(EIO
);
438 pflags
= blkid_partition_get_flags(pp
);
441 nr
= blkid_partition_get_partno(pp
);
443 return errno_or_else(EIO
);
446 start
= blkid_partition_get_start(pp
);
448 return errno_or_else(EIO
);
450 assert((uint64_t) start
< UINT64_MAX
/512);
453 size
= blkid_partition_get_size(pp
);
455 return errno_or_else(EIO
);
457 assert((uint64_t) size
< UINT64_MAX
/512);
459 r
= make_partition_devname(devname
, nr
, &node
);
463 /* So here's the thing: after the main ("whole") block device popped up it might take a while
464 * before the kernel fully probed the partition table. Waiting for that to finish is icky in
465 * userspace. So here's what we do instead. We issue the BLKPG_ADD_PARTITION ioctl to add the
466 * partition ourselves, racing against the kernel. Good thing is: if this call fails with
467 * EBUSY then the kernel was quicker than us, and that's totally OK, the outcome is good for
468 * us: the device node will exist. If OTOH our call was successful we won the race. Which is
469 * also good as the outcome is the same: the partition block device exists, and we can use
472 * Kernel returns EBUSY if there's already a partition by that number or an overlapping
473 * partition already existent. */
475 r
= block_device_add_partition(fd
, node
, nr
, (uint64_t) start
* 512, (uint64_t) size
* 512);
478 return log_debug_errno(r
, "BLKPG_ADD_PARTITION failed: %m");
480 log_debug_errno(r
, "Kernel was quicker than us in adding partition %i.", nr
);
482 log_debug("We were quicker than kernel in adding partition %i.", nr
);
485 PartitionDesignator designator
= _PARTITION_DESIGNATOR_INVALID
;
486 Architecture architecture
= _ARCHITECTURE_INVALID
;
487 const char *stype
, *sid
, *fstype
= NULL
, *label
;
488 sd_id128_t type_id
, id
;
489 bool rw
= true, growfs
= false;
491 sid
= blkid_partition_get_uuid(pp
);
494 if (sd_id128_from_string(sid
, &id
) < 0)
497 stype
= blkid_partition_get_type_string(pp
);
500 if (sd_id128_from_string(stype
, &type_id
) < 0)
503 label
= blkid_partition_get_name(pp
); /* libblkid returns NULL here if empty */
505 if (sd_id128_equal(type_id
, SD_GPT_HOME
)) {
507 check_partition_flags(node
, pflags
,
508 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
510 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
513 designator
= PARTITION_HOME
;
514 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
515 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
517 } else if (sd_id128_equal(type_id
, SD_GPT_SRV
)) {
519 check_partition_flags(node
, pflags
,
520 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
522 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
525 designator
= PARTITION_SRV
;
526 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
527 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
529 } else if (sd_id128_equal(type_id
, SD_GPT_ESP
)) {
531 /* Note that we don't check the SD_GPT_FLAG_NO_AUTO flag for the ESP, as it is
532 * not defined there. We instead check the SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as
533 * recommended by the UEFI spec (See "12.3.3 Number and Location of System
536 if (pflags
& SD_GPT_FLAG_NO_BLOCK_IO_PROTOCOL
)
539 designator
= PARTITION_ESP
;
542 } else if (sd_id128_equal(type_id
, SD_GPT_XBOOTLDR
)) {
544 check_partition_flags(node
, pflags
,
545 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
547 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
550 designator
= PARTITION_XBOOTLDR
;
551 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
552 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
554 } else if (gpt_partition_type_is_root(type_id
)) {
556 check_partition_flags(node
, pflags
,
557 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
559 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
562 /* If a root ID is specified, ignore everything but the root id */
563 if (!sd_id128_is_null(root_uuid
) && !sd_id128_equal(root_uuid
, id
))
566 assert_se((architecture
= gpt_partition_type_uuid_to_arch(type_id
)) >= 0);
567 designator
= PARTITION_ROOT_OF_ARCH(architecture
);
568 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
569 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
571 } else if (gpt_partition_type_is_root_verity(type_id
)) {
573 check_partition_flags(node
, pflags
,
574 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
576 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
579 m
->has_verity
= true;
581 /* If no verity configuration is specified, then don't do verity */
584 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
587 /* If root hash is specified, then ignore everything but the root id */
588 if (!sd_id128_is_null(root_verity_uuid
) && !sd_id128_equal(root_verity_uuid
, id
))
591 assert_se((architecture
= gpt_partition_type_uuid_to_arch(type_id
)) >= 0);
592 designator
= PARTITION_VERITY_OF(PARTITION_ROOT_OF_ARCH(architecture
));
593 fstype
= "DM_verity_hash";
596 } else if (gpt_partition_type_is_root_verity_sig(type_id
)) {
598 check_partition_flags(node
, pflags
,
599 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
601 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
604 m
->has_verity_sig
= true;
606 /* If root hash is specified explicitly, then ignore any embedded signature */
609 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
611 if (verity
->root_hash
)
614 assert_se((architecture
= gpt_partition_type_uuid_to_arch(type_id
)) >= 0);
615 designator
= PARTITION_VERITY_SIG_OF(PARTITION_ROOT_OF_ARCH(architecture
));
616 fstype
= "verity_hash_signature";
619 } else if (gpt_partition_type_is_usr(type_id
)) {
621 check_partition_flags(node
, pflags
,
622 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
624 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
627 /* If a usr ID is specified, ignore everything but the usr id */
628 if (!sd_id128_is_null(usr_uuid
) && !sd_id128_equal(usr_uuid
, id
))
631 assert_se((architecture
= gpt_partition_type_uuid_to_arch(type_id
)) >= 0);
632 designator
= PARTITION_USR_OF_ARCH(architecture
);
633 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
634 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
636 } else if (gpt_partition_type_is_usr_verity(type_id
)) {
638 check_partition_flags(node
, pflags
,
639 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
641 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
644 m
->has_verity
= true;
648 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
651 /* If usr hash is specified, then ignore everything but the usr id */
652 if (!sd_id128_is_null(usr_verity_uuid
) && !sd_id128_equal(usr_verity_uuid
, id
))
655 assert_se((architecture
= gpt_partition_type_uuid_to_arch(type_id
)) >= 0);
656 designator
= PARTITION_VERITY_OF(PARTITION_USR_OF_ARCH(architecture
));
657 fstype
= "DM_verity_hash";
660 } else if (gpt_partition_type_is_usr_verity_sig(type_id
)) {
662 check_partition_flags(node
, pflags
,
663 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
);
665 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
668 m
->has_verity_sig
= true;
670 /* If usr hash is specified explicitly, then ignore any embedded signature */
673 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
675 if (verity
->root_hash
)
678 assert_se((architecture
= gpt_partition_type_uuid_to_arch(type_id
)) >= 0);
679 designator
= PARTITION_VERITY_SIG_OF(PARTITION_USR_OF_ARCH(architecture
));
680 fstype
= "verity_hash_signature";
683 } else if (sd_id128_equal(type_id
, SD_GPT_SWAP
)) {
685 check_partition_flags(node
, pflags
, SD_GPT_FLAG_NO_AUTO
);
687 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
690 designator
= PARTITION_SWAP
;
692 } else if (sd_id128_equal(type_id
, SD_GPT_LINUX_GENERIC
)) {
694 check_partition_flags(node
, pflags
,
695 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
697 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
701 multiple_generic
= true;
704 generic_rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
705 generic_growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
707 generic_node
= strdup(node
);
712 } else if (sd_id128_equal(type_id
, SD_GPT_TMP
)) {
714 check_partition_flags(node
, pflags
,
715 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
717 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
720 designator
= PARTITION_TMP
;
721 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
722 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
724 } else if (sd_id128_equal(type_id
, SD_GPT_VAR
)) {
726 check_partition_flags(node
, pflags
,
727 SD_GPT_FLAG_NO_AUTO
| SD_GPT_FLAG_READ_ONLY
| SD_GPT_FLAG_GROWFS
);
729 if (pflags
& SD_GPT_FLAG_NO_AUTO
)
732 if (!FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_VAR_CHECK
)) {
735 /* For /var we insist that the uuid of the partition matches the
736 * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
737 * ID. Why? Unlike the other partitions /var is inherently
738 * installation specific, hence we need to be careful not to mount it
739 * in the wrong installation. By hashing the partition UUID from
740 * /etc/machine-id we can securely bind the partition to the
743 r
= sd_id128_get_machine_app_specific(SD_GPT_VAR
, &var_uuid
);
747 if (!sd_id128_equal(var_uuid
, id
)) {
748 log_debug("Found a /var/ partition, but its UUID didn't match our expectations, ignoring.");
753 designator
= PARTITION_VAR
;
754 rw
= !(pflags
& SD_GPT_FLAG_READ_ONLY
);
755 growfs
= FLAGS_SET(pflags
, SD_GPT_FLAG_GROWFS
);
758 if (designator
!= _PARTITION_DESIGNATOR_INVALID
) {
759 _cleanup_free_
char *t
= NULL
, *o
= NULL
, *l
= NULL
;
760 const char *options
= NULL
;
762 if (m
->partitions
[designator
].found
) {
763 /* For most partition types the first one we see wins. Except for the
764 * rootfs and /usr, where we do a version compare of the label, and
765 * let the newest version win. This permits a simple A/B versioning
766 * scheme in OS images. */
768 if (!PARTITION_DESIGNATOR_VERSIONED(designator
) ||
769 strverscmp_improved(m
->partitions
[designator
].label
, label
) >= 0)
772 dissected_partition_done(m
->partitions
+ designator
);
787 options
= mount_options_from_designator(mount_options
, designator
);
794 m
->partitions
[designator
] = (DissectedPartition
) {
799 .architecture
= architecture
,
800 .node
= TAKE_PTR(node
),
801 .fstype
= TAKE_PTR(t
),
802 .label
= TAKE_PTR(l
),
804 .mount_options
= TAKE_PTR(o
),
805 .offset
= (uint64_t) start
* 512,
806 .size
= (uint64_t) size
* 512,
812 switch (blkid_partition_get_type(pp
)) {
814 case 0x83: /* Linux partition */
816 if (pflags
!= 0x80) /* Bootable flag */
820 multiple_generic
= true;
824 generic_growfs
= false;
825 generic_node
= strdup(node
);
832 case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
833 _cleanup_free_
char *o
= NULL
;
834 sd_id128_t id
= SD_ID128_NULL
;
835 const char *sid
, *options
= NULL
;
838 if (m
->partitions
[PARTITION_XBOOTLDR
].found
)
841 sid
= blkid_partition_get_uuid(pp
);
843 (void) sd_id128_from_string(sid
, &id
);
845 options
= mount_options_from_designator(mount_options
, PARTITION_XBOOTLDR
);
852 m
->partitions
[PARTITION_XBOOTLDR
] = (DissectedPartition
) {
857 .architecture
= _ARCHITECTURE_INVALID
,
858 .node
= TAKE_PTR(node
),
860 .mount_options
= TAKE_PTR(o
),
861 .offset
= (uint64_t) start
* 512,
862 .size
= (uint64_t) size
* 512,
870 if (m
->partitions
[PARTITION_ROOT
].found
) {
871 /* If we found the primary arch, then invalidate the secondary and other arch to avoid any
872 * ambiguities, since we never want to mount the secondary or other arch in this case. */
873 m
->partitions
[PARTITION_ROOT_SECONDARY
].found
= false;
874 m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY
].found
= false;
875 m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY_SIG
].found
= false;
876 m
->partitions
[PARTITION_USR_SECONDARY
].found
= false;
877 m
->partitions
[PARTITION_USR_SECONDARY_VERITY
].found
= false;
878 m
->partitions
[PARTITION_USR_SECONDARY_VERITY_SIG
].found
= false;
880 m
->partitions
[PARTITION_ROOT_OTHER
].found
= false;
881 m
->partitions
[PARTITION_ROOT_OTHER_VERITY
].found
= false;
882 m
->partitions
[PARTITION_ROOT_OTHER_VERITY_SIG
].found
= false;
883 m
->partitions
[PARTITION_USR_OTHER
].found
= false;
884 m
->partitions
[PARTITION_USR_OTHER_VERITY
].found
= false;
885 m
->partitions
[PARTITION_USR_OTHER_VERITY_SIG
].found
= false;
887 } else if (m
->partitions
[PARTITION_ROOT_VERITY
].found
||
888 m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
)
889 return -EADDRNOTAVAIL
; /* Verity found but no matching rootfs? Something is off, refuse. */
891 else if (m
->partitions
[PARTITION_ROOT_SECONDARY
].found
) {
893 /* No root partition found but there's one for the secondary architecture? Then upgrade
894 * secondary arch to first and invalidate the other arch. */
896 log_debug("No root partition found of the native architecture, falling back to a root "
897 "partition of the secondary architecture.");
899 m
->partitions
[PARTITION_ROOT
] = m
->partitions
[PARTITION_ROOT_SECONDARY
];
900 zero(m
->partitions
[PARTITION_ROOT_SECONDARY
]);
901 m
->partitions
[PARTITION_ROOT_VERITY
] = m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY
];
902 zero(m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY
]);
903 m
->partitions
[PARTITION_ROOT_VERITY_SIG
] = m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY_SIG
];
904 zero(m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY_SIG
]);
906 m
->partitions
[PARTITION_USR
] = m
->partitions
[PARTITION_USR_SECONDARY
];
907 zero(m
->partitions
[PARTITION_USR_SECONDARY
]);
908 m
->partitions
[PARTITION_USR_VERITY
] = m
->partitions
[PARTITION_USR_SECONDARY_VERITY
];
909 zero(m
->partitions
[PARTITION_USR_SECONDARY_VERITY
]);
910 m
->partitions
[PARTITION_USR_VERITY_SIG
] = m
->partitions
[PARTITION_USR_SECONDARY_VERITY_SIG
];
911 zero(m
->partitions
[PARTITION_USR_SECONDARY_VERITY_SIG
]);
913 m
->partitions
[PARTITION_ROOT_OTHER
].found
= false;
914 m
->partitions
[PARTITION_ROOT_OTHER_VERITY
].found
= false;
915 m
->partitions
[PARTITION_ROOT_OTHER_VERITY_SIG
].found
= false;
916 m
->partitions
[PARTITION_USR_OTHER
].found
= false;
917 m
->partitions
[PARTITION_USR_OTHER_VERITY
].found
= false;
918 m
->partitions
[PARTITION_USR_OTHER_VERITY_SIG
].found
= false;
920 } else if (m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY
].found
||
921 m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY_SIG
].found
)
922 return -EADDRNOTAVAIL
; /* as above */
924 else if (m
->partitions
[PARTITION_ROOT_OTHER
].found
) {
926 /* No root or secondary partition found but there's one for another architecture? Then
927 * upgrade the other architecture to first. */
929 log_debug("No root partition found of the native architecture or the secondary architecture, "
930 "falling back to a root partition of a non-native architecture (%s).",
931 architecture_to_string(m
->partitions
[PARTITION_ROOT_OTHER
].architecture
));
933 m
->partitions
[PARTITION_ROOT
] = m
->partitions
[PARTITION_ROOT_OTHER
];
934 zero(m
->partitions
[PARTITION_ROOT_OTHER
]);
935 m
->partitions
[PARTITION_ROOT_VERITY
] = m
->partitions
[PARTITION_ROOT_OTHER_VERITY
];
936 zero(m
->partitions
[PARTITION_ROOT_OTHER_VERITY
]);
937 m
->partitions
[PARTITION_ROOT_VERITY_SIG
] = m
->partitions
[PARTITION_ROOT_OTHER_VERITY_SIG
];
938 zero(m
->partitions
[PARTITION_ROOT_OTHER_VERITY_SIG
]);
940 m
->partitions
[PARTITION_USR
] = m
->partitions
[PARTITION_USR_OTHER
];
941 zero(m
->partitions
[PARTITION_USR_OTHER
]);
942 m
->partitions
[PARTITION_USR_VERITY
] = m
->partitions
[PARTITION_USR_OTHER_VERITY
];
943 zero(m
->partitions
[PARTITION_USR_OTHER_VERITY
]);
944 m
->partitions
[PARTITION_USR_VERITY_SIG
] = m
->partitions
[PARTITION_USR_OTHER_VERITY_SIG
];
945 zero(m
->partitions
[PARTITION_USR_OTHER_VERITY_SIG
]);
948 /* Hmm, we found a signature partition but no Verity data? Something is off. */
949 if (m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_ROOT_VERITY
].found
)
950 return -EADDRNOTAVAIL
;
952 if (m
->partitions
[PARTITION_USR
].found
) {
953 /* Invalidate secondary and other arch /usr/ if we found the primary arch */
954 m
->partitions
[PARTITION_USR_SECONDARY
].found
= false;
955 m
->partitions
[PARTITION_USR_SECONDARY_VERITY
].found
= false;
956 m
->partitions
[PARTITION_USR_SECONDARY_VERITY_SIG
].found
= false;
958 m
->partitions
[PARTITION_USR_OTHER
].found
= false;
959 m
->partitions
[PARTITION_USR_OTHER_VERITY
].found
= false;
960 m
->partitions
[PARTITION_USR_OTHER_VERITY_SIG
].found
= false;
962 } else if (m
->partitions
[PARTITION_USR_VERITY
].found
||
963 m
->partitions
[PARTITION_USR_VERITY_SIG
].found
)
964 return -EADDRNOTAVAIL
; /* as above */
966 else if (m
->partitions
[PARTITION_USR_SECONDARY
].found
) {
968 log_debug("No usr partition found of the native architecture, falling back to a usr "
969 "partition of the secondary architecture.");
971 /* Upgrade secondary arch to primary */
972 m
->partitions
[PARTITION_USR
] = m
->partitions
[PARTITION_USR_SECONDARY
];
973 zero(m
->partitions
[PARTITION_USR_SECONDARY
]);
974 m
->partitions
[PARTITION_USR_VERITY
] = m
->partitions
[PARTITION_USR_SECONDARY_VERITY
];
975 zero(m
->partitions
[PARTITION_USR_SECONDARY_VERITY
]);
976 m
->partitions
[PARTITION_USR_VERITY_SIG
] = m
->partitions
[PARTITION_USR_SECONDARY_VERITY_SIG
];
977 zero(m
->partitions
[PARTITION_USR_SECONDARY_VERITY_SIG
]);
979 m
->partitions
[PARTITION_USR_OTHER
].found
= false;
980 m
->partitions
[PARTITION_USR_OTHER_VERITY
].found
= false;
981 m
->partitions
[PARTITION_USR_OTHER_VERITY_SIG
].found
= false;
983 } else if (m
->partitions
[PARTITION_USR_SECONDARY_VERITY
].found
||
984 m
->partitions
[PARTITION_USR_SECONDARY_VERITY_SIG
].found
)
985 return -EADDRNOTAVAIL
; /* as above */
987 else if (m
->partitions
[PARTITION_USR_OTHER
].found
) {
989 log_debug("No usr partition found of the native architecture or the secondary architecture, "
990 "falling back to a usr partition of a non-native architecture (%s).",
991 architecture_to_string(m
->partitions
[PARTITION_ROOT_OTHER
].architecture
));
993 /* Upgrade other arch to primary */
994 m
->partitions
[PARTITION_USR
] = m
->partitions
[PARTITION_USR_OTHER
];
995 zero(m
->partitions
[PARTITION_USR_OTHER
]);
996 m
->partitions
[PARTITION_USR_VERITY
] = m
->partitions
[PARTITION_USR_OTHER_VERITY
];
997 zero(m
->partitions
[PARTITION_USR_OTHER_VERITY
]);
998 m
->partitions
[PARTITION_USR_VERITY_SIG
] = m
->partitions
[PARTITION_USR_OTHER_VERITY_SIG
];
999 zero(m
->partitions
[PARTITION_USR_OTHER_VERITY_SIG
]);
1002 /* Hmm, we found a signature partition but no Verity data? Something is off. */
1003 if (m
->partitions
[PARTITION_USR_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_USR_VERITY
].found
)
1004 return -EADDRNOTAVAIL
;
1006 /* If root and /usr are combined then insist that the architecture matches */
1007 if (m
->partitions
[PARTITION_ROOT
].found
&&
1008 m
->partitions
[PARTITION_USR
].found
&&
1009 (m
->partitions
[PARTITION_ROOT
].architecture
>= 0 &&
1010 m
->partitions
[PARTITION_USR
].architecture
>= 0 &&
1011 m
->partitions
[PARTITION_ROOT
].architecture
!= m
->partitions
[PARTITION_USR
].architecture
))
1012 return -EADDRNOTAVAIL
;
1014 if (!m
->partitions
[PARTITION_ROOT
].found
&&
1015 !m
->partitions
[PARTITION_USR
].found
&&
1016 (flags
& DISSECT_IMAGE_GENERIC_ROOT
) &&
1017 (!verity
|| !verity
->root_hash
|| verity
->designator
!= PARTITION_USR
)) {
1019 /* OK, we found nothing usable, then check if there's a single generic partition, and use
1020 * that. If the root hash was set however, then we won't fall back to a generic node, because
1021 * the root hash decides. */
1023 /* If we didn't find a properly marked root partition, but we did find a single suitable
1024 * generic Linux partition, then use this as root partition, if the caller asked for it. */
1025 if (multiple_generic
)
1028 /* If we didn't find a generic node, then we can't fix this up either */
1030 _cleanup_free_
char *o
= NULL
;
1031 const char *options
;
1033 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
1035 o
= strdup(options
);
1040 assert(generic_nr
>= 0);
1041 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
1044 .growfs
= generic_growfs
,
1045 .partno
= generic_nr
,
1046 .architecture
= _ARCHITECTURE_INVALID
,
1047 .node
= TAKE_PTR(generic_node
),
1048 .uuid
= generic_uuid
,
1049 .mount_options
= TAKE_PTR(o
),
1050 .offset
= UINT64_MAX
,
1056 /* 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 */
1057 if (FLAGS_SET(flags
, DISSECT_IMAGE_REQUIRE_ROOT
) &&
1058 !(m
->partitions
[PARTITION_ROOT
].found
|| (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
1061 if (m
->partitions
[PARTITION_ROOT_VERITY
].found
) {
1062 /* We only support one verity partition per image, i.e. can't do for both /usr and root fs */
1063 if (m
->partitions
[PARTITION_USR_VERITY
].found
)
1066 /* We don't support verity enabled root with a split out /usr. Neither with nor without
1067 * verity there. (Note that we do support verity-less root with verity-full /usr, though.) */
1068 if (m
->partitions
[PARTITION_USR
].found
)
1069 return -EADDRNOTAVAIL
;
1073 /* If a verity designator is specified, then insist that the matching partition exists */
1074 if (verity
->designator
>= 0 && !m
->partitions
[verity
->designator
].found
)
1075 return -EADDRNOTAVAIL
;
1077 if (verity
->root_hash
) {
1078 /* If we have an explicit root hash and found the partitions for it, then we are ready to use
1079 * Verity, set things up for it */
1081 if (verity
->designator
< 0 || verity
->designator
== PARTITION_ROOT
) {
1082 if (!m
->partitions
[PARTITION_ROOT_VERITY
].found
|| !m
->partitions
[PARTITION_ROOT
].found
)
1083 return -EADDRNOTAVAIL
;
1085 /* If we found a verity setup, then the root partition is necessarily read-only. */
1086 m
->partitions
[PARTITION_ROOT
].rw
= false;
1087 m
->verity_ready
= true;
1090 assert(verity
->designator
== PARTITION_USR
);
1092 if (!m
->partitions
[PARTITION_USR_VERITY
].found
|| !m
->partitions
[PARTITION_USR
].found
)
1093 return -EADDRNOTAVAIL
;
1095 m
->partitions
[PARTITION_USR
].rw
= false;
1096 m
->verity_ready
= true;
1099 if (m
->verity_ready
)
1100 m
->verity_sig_ready
= verity
->root_hash_sig
;
1102 } else if (m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR_VERITY_SIG
: PARTITION_ROOT_VERITY_SIG
].found
) {
1104 /* If we found an embedded signature partition, we are ready, too. */
1106 m
->verity_ready
= m
->verity_sig_ready
= true;
1107 m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR
: PARTITION_ROOT
].rw
= false;
1111 blkid_free_probe(b
);
1114 /* Fill in file system types if we don't know them yet. */
1115 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
1116 DissectedPartition
*p
= m
->partitions
+ i
;
1121 if (!p
->fstype
&& p
->node
) {
1122 r
= probe_filesystem(p
->node
, &p
->fstype
);
1123 if (r
< 0 && r
!= -EUCLEAN
)
1127 if (streq_ptr(p
->fstype
, "crypto_LUKS"))
1128 m
->encrypted
= true;
1130 if (p
->fstype
&& fstype_is_ro(p
->fstype
))
1144 DissectedImage
* dissected_image_unref(DissectedImage
*m
) {
1148 /* First, clear dissected partitions. */
1149 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++)
1150 dissected_partition_done(m
->partitions
+ i
);
1152 /* Second, free decrypted images. This must be after dissected_partition_done(), as freeing
1153 * DecryptedImage may try to deactivate partitions. */
1154 decrypted_image_unref(m
->decrypted_image
);
1156 /* Third, unref LoopDevice. This must be called after the above two, as freeing LoopDevice may try to
1157 * remove existing partitions on the loopback block device. */
1158 loop_device_unref(m
->loop
);
1160 free(m
->image_name
);
1162 strv_free(m
->machine_info
);
1163 strv_free(m
->os_release
);
1164 strv_free(m
->extension_release
);
1169 static int is_loop_device(const char *path
) {
1170 char s
[SYS_BLOCK_PATH_MAX("/../loop/")];
1175 if (stat(path
, &st
) < 0)
1178 if (!S_ISBLK(st
.st_mode
))
1181 xsprintf_sys_block_path(s
, "/loop/", st
.st_dev
);
1182 if (access(s
, F_OK
) < 0) {
1183 if (errno
!= ENOENT
)
1186 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
1187 xsprintf_sys_block_path(s
, "/../loop/", st
.st_dev
);
1188 if (access(s
, F_OK
) < 0)
1189 return errno
== ENOENT
? false : -errno
;
1195 static int run_fsck(const char *node
, const char *fstype
) {
1202 r
= fsck_exists(fstype
);
1204 log_debug_errno(r
, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype
);
1208 log_debug("Not checking partition %s, as fsck for %s does not exist.", node
, fstype
);
1212 r
= safe_fork("(fsck)", FORK_RESET_SIGNALS
|FORK_CLOSE_ALL_FDS
|FORK_RLIMIT_NOFILE_SAFE
|FORK_DEATHSIG
|FORK_NULL_STDIO
, &pid
);
1214 return log_debug_errno(r
, "Failed to fork off fsck: %m");
1217 execl("/sbin/fsck", "/sbin/fsck", "-aT", node
, NULL
);
1219 log_debug_errno(errno
, "Failed to execl() fsck: %m");
1220 _exit(FSCK_OPERATIONAL_ERROR
);
1223 exit_status
= wait_for_terminate_and_check("fsck", pid
, 0);
1224 if (exit_status
< 0)
1225 return log_debug_errno(exit_status
, "Failed to fork off /sbin/fsck: %m");
1227 if ((exit_status
& ~FSCK_ERROR_CORRECTED
) != FSCK_SUCCESS
) {
1228 log_debug("fsck failed with exit status %i.", exit_status
);
1230 if ((exit_status
& (FSCK_SYSTEM_SHOULD_REBOOT
|FSCK_ERRORS_LEFT_UNCORRECTED
)) != 0)
1231 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
), "File system is corrupted, refusing.");
1233 log_debug("Ignoring fsck error.");
1239 static int fs_grow(const char *node_path
, const char *mount_path
) {
1240 _cleanup_close_
int mount_fd
= -1, node_fd
= -1;
1241 uint64_t size
, newsize
;
1244 node_fd
= open(node_path
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
1246 return log_debug_errno(errno
, "Failed to open node device %s: %m", node_path
);
1248 if (ioctl(node_fd
, BLKGETSIZE64
, &size
) != 0)
1249 return log_debug_errno(errno
, "Failed to get block device size of %s: %m", node_path
);
1251 mount_fd
= open(mount_path
, O_RDONLY
|O_DIRECTORY
|O_CLOEXEC
);
1253 return log_debug_errno(errno
, "Failed to open mountd file system %s: %m", mount_path
);
1255 log_debug("Resizing \"%s\" to %"PRIu64
" bytes...", mount_path
, size
);
1256 r
= resize_fs(mount_fd
, size
, &newsize
);
1258 return log_debug_errno(r
, "Failed to resize \"%s\" to %"PRIu64
" bytes: %m", mount_path
, size
);
1260 if (newsize
== size
)
1261 log_debug("Successfully resized \"%s\" to %s bytes.",
1262 mount_path
, FORMAT_BYTES(newsize
));
1264 assert(newsize
< size
);
1265 log_debug("Successfully resized \"%s\" to %s bytes (%"PRIu64
" bytes lost due to blocksize).",
1266 mount_path
, FORMAT_BYTES(newsize
), size
- newsize
);
1272 static int mount_partition(
1273 DissectedPartition
*m
,
1275 const char *directory
,
1278 DissectImageFlags flags
) {
1280 _cleanup_free_
char *chased
= NULL
, *options
= NULL
;
1281 const char *p
, *node
, *fstype
;
1282 bool rw
, remap_uid_gid
= false;
1288 /* Use decrypted node and matching fstype if available, otherwise use the original device */
1289 node
= m
->decrypted_node
?: m
->node
;
1290 fstype
= m
->decrypted_node
? m
->decrypted_fstype
: m
->fstype
;
1292 if (!m
->found
|| !node
)
1295 return -EAFNOSUPPORT
;
1297 /* We are looking at an encrypted partition? This either means stacked encryption, or the caller
1298 * didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error for this
1300 if (streq(fstype
, "crypto_LUKS"))
1303 rw
= m
->rw
&& !(flags
& DISSECT_IMAGE_MOUNT_READ_ONLY
);
1305 if (FLAGS_SET(flags
, DISSECT_IMAGE_FSCK
) && rw
) {
1306 r
= run_fsck(node
, fstype
);
1312 /* Automatically create missing mount points inside the image, if necessary. */
1313 r
= mkdir_p_root(where
, directory
, uid_shift
, (gid_t
) uid_shift
, 0755);
1314 if (r
< 0 && r
!= -EROFS
)
1317 r
= chase_symlinks(directory
, where
, CHASE_PREFIX_ROOT
, &chased
, NULL
);
1323 /* Create top-level mount if missing – but only if this is asked for. This won't modify the
1324 * image (as the branch above does) but the host hierarchy, and the created directory might
1325 * survive our mount in the host hierarchy hence. */
1326 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
1327 r
= mkdir_p(where
, 0755);
1335 /* If requested, turn on discard support. */
1336 if (fstype_can_discard(fstype
) &&
1337 ((flags
& DISSECT_IMAGE_DISCARD
) ||
1338 ((flags
& DISSECT_IMAGE_DISCARD_ON_LOOP
) && is_loop_device(m
->node
) > 0))) {
1339 options
= strdup("discard");
1344 if (uid_is_valid(uid_shift
) && uid_shift
!= 0) {
1346 if (fstype_can_uid_gid(fstype
)) {
1347 _cleanup_free_
char *uid_option
= NULL
;
1349 if (asprintf(&uid_option
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
1352 if (!strextend_with_separator(&options
, ",", uid_option
))
1354 } else if (FLAGS_SET(flags
, DISSECT_IMAGE_MOUNT_IDMAPPED
))
1355 remap_uid_gid
= true;
1358 if (!isempty(m
->mount_options
))
1359 if (!strextend_with_separator(&options
, ",", m
->mount_options
))
1362 /* So, when you request MS_RDONLY from ext4, then this means nothing. It happily still writes to the
1363 * backing storage. What's worse, the BLKRO[GS]ET flag and (in case of loopback devices)
1364 * LO_FLAGS_READ_ONLY don't mean anything, they affect userspace accesses only, and write accesses
1365 * from the upper file system still get propagated through to the underlying file system,
1366 * unrestricted. To actually get ext4/xfs/btrfs to stop writing to the device we need to specify
1367 * "norecovery" as mount option, in addition to MS_RDONLY. Yes, this sucks, since it means we need to
1368 * carry a per file system table here.
1370 * Note that this means that we might not be able to mount corrupted file systems as read-only
1371 * anymore (since in some cases the kernel implementations will refuse mounting when corrupted,
1372 * read-only and "norecovery" is specified). But I think for the case of automatically determined
1373 * mount options for loopback devices this is the right choice, since otherwise using the same
1374 * loopback file twice even in read-only mode, is going to fail badly sooner or later. The usecase of
1375 * making reuse of the immutable images "just work" is more relevant to us than having read-only
1376 * access that actually modifies stuff work on such image files. Or to say this differently: if
1377 * people want their file systems to be fixed up they should just open them in writable mode, where
1378 * all these problems don't exist. */
1379 if (!rw
&& STRPTR_IN_SET(fstype
, "ext3", "ext4", "xfs", "btrfs"))
1380 if (!strextend_with_separator(&options
, ",", "norecovery"))
1383 r
= mount_nofollow_verbose(LOG_DEBUG
, node
, p
, fstype
, MS_NODEV
|(rw
? 0 : MS_RDONLY
), options
);
1387 if (rw
&& m
->growfs
&& FLAGS_SET(flags
, DISSECT_IMAGE_GROWFS
))
1388 (void) fs_grow(node
, p
);
1390 if (remap_uid_gid
) {
1391 r
= remount_idmap(p
, uid_shift
, uid_range
, UID_INVALID
, REMOUNT_IDMAPPING_HOST_ROOT
);
1399 static int mount_root_tmpfs(const char *where
, uid_t uid_shift
, DissectImageFlags flags
) {
1400 _cleanup_free_
char *options
= NULL
;
1405 /* For images that contain /usr/ but no rootfs, let's mount rootfs as tmpfs */
1407 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
1408 r
= mkdir_p(where
, 0755);
1413 if (uid_is_valid(uid_shift
)) {
1414 if (asprintf(&options
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
1418 r
= mount_nofollow_verbose(LOG_DEBUG
, "rootfs", where
, "tmpfs", MS_NODEV
, options
);
1425 int dissected_image_mount(
1430 DissectImageFlags flags
) {
1432 int r
, xbootldr_mounted
;
1439 * -ENXIO → No root partition found
1440 * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release/extension-release file found
1441 * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet
1442 * -EUCLEAN → fsck for file system failed
1443 * -EBUSY → File system already mounted/used elsewhere (kernel)
1444 * -EAFNOSUPPORT → File system type not supported or not known
1447 if (!(m
->partitions
[PARTITION_ROOT
].found
||
1448 (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
1449 return -ENXIO
; /* Require a root fs or at least a /usr/ fs (the latter is subject to a flag of its own) */
1451 if ((flags
& DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
) == 0) {
1453 /* First mount the root fs. If there's none we use a tmpfs. */
1454 if (m
->partitions
[PARTITION_ROOT
].found
)
1455 r
= mount_partition(m
->partitions
+ PARTITION_ROOT
, where
, NULL
, uid_shift
, uid_range
, flags
);
1457 r
= mount_root_tmpfs(where
, uid_shift
, flags
);
1461 /* For us mounting root always means mounting /usr as well */
1462 r
= mount_partition(m
->partitions
+ PARTITION_USR
, where
, "/usr", uid_shift
, uid_range
, flags
);
1466 if ((flags
& (DISSECT_IMAGE_VALIDATE_OS
|DISSECT_IMAGE_VALIDATE_OS_EXT
)) != 0) {
1467 /* If either one of the validation flags are set, ensure that the image qualifies
1468 * as one or the other (or both). */
1471 if (FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS
)) {
1472 r
= path_is_os_tree(where
);
1478 if (!ok
&& FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS_EXT
)) {
1479 r
= path_is_extension_tree(where
, m
->image_name
);
1491 if (flags
& DISSECT_IMAGE_MOUNT_ROOT_ONLY
)
1494 r
= mount_partition(m
->partitions
+ PARTITION_HOME
, where
, "/home", uid_shift
, uid_range
, flags
);
1498 r
= mount_partition(m
->partitions
+ PARTITION_SRV
, where
, "/srv", uid_shift
, uid_range
, flags
);
1502 r
= mount_partition(m
->partitions
+ PARTITION_VAR
, where
, "/var", uid_shift
, uid_range
, flags
);
1506 r
= mount_partition(m
->partitions
+ PARTITION_TMP
, where
, "/var/tmp", uid_shift
, uid_range
, flags
);
1510 xbootldr_mounted
= mount_partition(m
->partitions
+ PARTITION_XBOOTLDR
, where
, "/boot", uid_shift
, uid_range
, flags
);
1511 if (xbootldr_mounted
< 0)
1512 return xbootldr_mounted
;
1514 if (m
->partitions
[PARTITION_ESP
].found
) {
1515 int esp_done
= false;
1517 /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it
1518 * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */
1520 r
= chase_symlinks("/efi", where
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
1525 /* /efi doesn't exist. Let's see if /boot is suitable then */
1527 if (!xbootldr_mounted
) {
1528 _cleanup_free_
char *p
= NULL
;
1530 r
= chase_symlinks("/boot", where
, CHASE_PREFIX_ROOT
, &p
, NULL
);
1534 } else if (dir_is_empty(p
, /* ignore_hidden_or_backup= */ false) > 0) {
1535 /* It exists and is an empty directory. Let's mount the ESP there. */
1536 r
= mount_partition(m
->partitions
+ PARTITION_ESP
, where
, "/boot", uid_shift
, uid_range
, flags
);
1546 /* OK, let's mount the ESP now to /efi (possibly creating the dir if missing) */
1548 r
= mount_partition(m
->partitions
+ PARTITION_ESP
, where
, "/efi", uid_shift
, uid_range
, flags
);
1557 int dissected_image_mount_and_warn(
1562 DissectImageFlags flags
) {
1569 r
= dissected_image_mount(m
, where
, uid_shift
, uid_range
, flags
);
1571 return log_error_errno(r
, "Not root file system found in image.");
1572 if (r
== -EMEDIUMTYPE
)
1573 return log_error_errno(r
, "No suitable os-release/extension-release file in image found.");
1575 return log_error_errno(r
, "Encrypted file system discovered, but decryption not requested.");
1577 return log_error_errno(r
, "File system check on image failed.");
1579 return log_error_errno(r
, "File system already mounted elsewhere.");
1580 if (r
== -EAFNOSUPPORT
)
1581 return log_error_errno(r
, "File system type not supported or not known.");
1583 return log_error_errno(r
, "Failed to mount image: %m");
1588 #if HAVE_LIBCRYPTSETUP
1589 struct DecryptedPartition
{
1590 struct crypt_device
*device
;
1596 typedef struct DecryptedPartition DecryptedPartition
;
1598 struct DecryptedImage
{
1600 DecryptedPartition
*decrypted
;
1604 static DecryptedImage
* decrypted_image_free(DecryptedImage
*d
) {
1605 #if HAVE_LIBCRYPTSETUP
1611 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
1612 DecryptedPartition
*p
= d
->decrypted
+ i
;
1614 if (p
->device
&& p
->name
&& !p
->relinquished
) {
1615 /* Let's deactivate lazily, as the dm volume may be already/still used by other processes. */
1616 r
= sym_crypt_deactivate_by_name(p
->device
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
1618 log_debug_errno(r
, "Failed to deactivate encrypted partition %s", p
->name
);
1622 sym_crypt_free(p
->device
);
1632 DEFINE_TRIVIAL_REF_UNREF_FUNC(DecryptedImage
, decrypted_image
, decrypted_image_free
);
1634 #if HAVE_LIBCRYPTSETUP
1635 static int decrypted_image_new(DecryptedImage
**ret
) {
1636 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
1640 d
= new(DecryptedImage
, 1);
1644 *d
= (DecryptedImage
) {
1652 static int make_dm_name_and_node(const void *original_node
, const char *suffix
, char **ret_name
, char **ret_node
) {
1653 _cleanup_free_
char *name
= NULL
, *node
= NULL
;
1656 assert(original_node
);
1661 base
= strrchr(original_node
, '/');
1663 base
= original_node
;
1669 name
= strjoin(base
, suffix
);
1672 if (!filename_is_valid(name
))
1675 node
= path_join(sym_crypt_get_dir(), name
);
1679 *ret_name
= TAKE_PTR(name
);
1680 *ret_node
= TAKE_PTR(node
);
1685 static int decrypt_partition(
1686 DissectedPartition
*m
,
1687 const char *passphrase
,
1688 DissectImageFlags flags
,
1689 DecryptedImage
*d
) {
1691 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
1692 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
1698 if (!m
->found
|| !m
->node
|| !m
->fstype
)
1701 if (!streq(m
->fstype
, "crypto_LUKS"))
1707 r
= dlopen_cryptsetup();
1711 r
= make_dm_name_and_node(m
->node
, "-decrypted", &name
, &node
);
1715 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
1718 r
= sym_crypt_init(&cd
, m
->node
);
1720 return log_debug_errno(r
, "Failed to initialize dm-crypt: %m");
1722 cryptsetup_enable_logging(cd
);
1724 r
= sym_crypt_load(cd
, CRYPT_LUKS
, NULL
);
1726 return log_debug_errno(r
, "Failed to load LUKS metadata: %m");
1728 r
= sym_crypt_activate_by_passphrase(cd
, name
, CRYPT_ANY_SLOT
, passphrase
, strlen(passphrase
),
1729 ((flags
& DISSECT_IMAGE_DEVICE_READ_ONLY
) ? CRYPT_ACTIVATE_READONLY
: 0) |
1730 ((flags
& DISSECT_IMAGE_DISCARD_ON_CRYPTO
) ? CRYPT_ACTIVATE_ALLOW_DISCARDS
: 0));
1732 log_debug_errno(r
, "Failed to activate LUKS device: %m");
1733 return r
== -EPERM
? -EKEYREJECTED
: r
;
1736 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
1737 .name
= TAKE_PTR(name
),
1738 .device
= TAKE_PTR(cd
),
1741 m
->decrypted_node
= TAKE_PTR(node
);
1746 static int verity_can_reuse(
1747 const VeritySettings
*verity
,
1749 struct crypt_device
**ret_cd
) {
1751 /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
1752 _cleanup_free_
char *root_hash_existing
= NULL
;
1753 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
1754 struct crypt_params_verity crypt_params
= {};
1755 size_t root_hash_existing_size
;
1762 r
= sym_crypt_init_by_name(&cd
, name
);
1764 return log_debug_errno(r
, "Error opening verity device, crypt_init_by_name failed: %m");
1766 cryptsetup_enable_logging(cd
);
1768 r
= sym_crypt_get_verity_info(cd
, &crypt_params
);
1770 return log_debug_errno(r
, "Error opening verity device, crypt_get_verity_info failed: %m");
1772 root_hash_existing_size
= verity
->root_hash_size
;
1773 root_hash_existing
= malloc0(root_hash_existing_size
);
1774 if (!root_hash_existing
)
1777 r
= sym_crypt_volume_key_get(cd
, CRYPT_ANY_SLOT
, root_hash_existing
, &root_hash_existing_size
, NULL
, 0);
1779 return log_debug_errno(r
, "Error opening verity device, crypt_volume_key_get failed: %m");
1780 if (verity
->root_hash_size
!= root_hash_existing_size
||
1781 memcmp(root_hash_existing
, verity
->root_hash
, verity
->root_hash_size
) != 0)
1782 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but root hashes are different.");
1784 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1785 /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the
1786 * same settings, so that a previous unsigned mount will not be reused if the user asks to use
1787 * signing for the new one, and vice versa. */
1788 if (!!verity
->root_hash_sig
!= !!(crypt_params
.flags
& CRYPT_VERITY_ROOT_HASH_SIGNATURE
))
1789 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but signature settings are not the same.");
1792 *ret_cd
= TAKE_PTR(cd
);
1796 static inline char* dm_deferred_remove_clean(char *name
) {
1800 (void) sym_crypt_deactivate_by_name(NULL
, name
, CRYPT_DEACTIVATE_DEFERRED
);
1803 DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean
);
1805 static int validate_signature_userspace(const VeritySettings
*verity
) {
1807 _cleanup_(sk_X509_free_allp
) STACK_OF(X509
) *sk
= NULL
;
1808 _cleanup_strv_free_
char **certs
= NULL
;
1809 _cleanup_(PKCS7_freep
) PKCS7
*p7
= NULL
;
1810 _cleanup_free_
char *s
= NULL
;
1811 _cleanup_(BIO_freep
) BIO
*bio
= NULL
; /* 'bio' must be freed first, 's' second, hence keep this order
1812 * of declaration in place, please */
1813 const unsigned char *d
;
1817 assert(verity
->root_hash
);
1818 assert(verity
->root_hash_sig
);
1820 /* Because installing a signature certificate into the kernel chain is so messy, let's optionally do
1821 * userspace validation. */
1823 r
= conf_files_list_nulstr(&certs
, ".crt", NULL
, CONF_FILES_REGULAR
|CONF_FILES_FILTER_MASKED
, CONF_PATHS_NULSTR("verity.d"));
1825 return log_debug_errno(r
, "Failed to enumerate certificates: %m");
1826 if (strv_isempty(certs
)) {
1827 log_debug("No userspace dm-verity certificates found.");
1831 d
= verity
->root_hash_sig
;
1832 p7
= d2i_PKCS7(NULL
, &d
, (long) verity
->root_hash_sig_size
);
1834 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to parse PKCS7 DER signature data.");
1836 s
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
1838 return log_oom_debug();
1840 bio
= BIO_new_mem_buf(s
, strlen(s
));
1842 return log_oom_debug();
1844 sk
= sk_X509_new_null();
1846 return log_oom_debug();
1848 STRV_FOREACH(i
, certs
) {
1849 _cleanup_(X509_freep
) X509
*c
= NULL
;
1850 _cleanup_fclose_
FILE *f
= NULL
;
1852 f
= fopen(*i
, "re");
1854 log_debug_errno(errno
, "Failed to open '%s', ignoring: %m", *i
);
1858 c
= PEM_read_X509(f
, NULL
, NULL
, NULL
);
1860 log_debug("Failed to load X509 certificate '%s', ignoring.", *i
);
1864 if (sk_X509_push(sk
, c
) == 0)
1865 return log_oom_debug();
1870 r
= PKCS7_verify(p7
, sk
, NULL
, bio
, NULL
, PKCS7_NOINTERN
|PKCS7_NOVERIFY
);
1872 log_debug("Userspace PKCS#7 validation succeeded.");
1874 log_debug("Userspace PKCS#7 validation failed: %s", ERR_error_string(ERR_get_error(), NULL
));
1878 log_debug("Not doing client-side validation of dm-verity root hash signatures, OpenSSL support disabled.");
1883 static int do_crypt_activate_verity(
1884 struct crypt_device
*cd
,
1886 const VeritySettings
*verity
) {
1888 bool check_signature
;
1895 if (verity
->root_hash_sig
) {
1896 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIGNATURE");
1897 if (r
< 0 && r
!= -ENXIO
)
1898 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIGNATURE");
1900 check_signature
= r
!= 0;
1902 check_signature
= false;
1904 if (check_signature
) {
1906 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1907 /* First, if we have support for signed keys in the kernel, then try that first. */
1908 r
= sym_crypt_activate_by_signed_key(
1912 verity
->root_hash_size
,
1913 verity
->root_hash_sig
,
1914 verity
->root_hash_sig_size
,
1915 CRYPT_ACTIVATE_READONLY
);
1919 log_debug("Validation of dm-verity signature failed via the kernel, trying userspace validation instead.");
1921 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.",
1922 program_invocation_short_name
);
1925 /* So this didn't work via the kernel, then let's try userspace validation instead. If that
1926 * works we'll try to activate without telling the kernel the signature. */
1928 r
= validate_signature_userspace(verity
);
1932 return log_debug_errno(SYNTHETIC_ERRNO(ENOKEY
),
1933 "Activation of signed Verity volume worked neither via the kernel nor in userspace, can't activate.");
1936 return sym_crypt_activate_by_volume_key(
1940 verity
->root_hash_size
,
1941 CRYPT_ACTIVATE_READONLY
);
1944 static usec_t
verity_timeout(void) {
1945 usec_t t
= 100 * USEC_PER_MSEC
;
1949 /* On slower machines, like non-KVM vm, setting up device may take a long time.
1950 * Let's make the timeout configurable. */
1952 e
= getenv("SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC");
1956 r
= parse_sec(e
, &t
);
1959 "Failed to parse timeout specified in $SYSTEMD_DISSECT_VERITY_TIMEOUT_SEC, "
1960 "using the default timeout (%s).",
1961 FORMAT_TIMESPAN(t
, USEC_PER_MSEC
));
1966 static int verity_partition(
1967 PartitionDesignator designator
,
1968 DissectedPartition
*m
,
1969 DissectedPartition
*v
,
1970 const VeritySettings
*verity
,
1971 DissectImageFlags flags
,
1972 DecryptedImage
*d
) {
1974 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
1975 _cleanup_(dm_deferred_remove_cleanp
) char *restore_deferred_remove
= NULL
;
1976 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
1980 assert(v
|| (verity
&& verity
->data_path
));
1982 if (!verity
|| !verity
->root_hash
)
1984 if (!((verity
->designator
< 0 && designator
== PARTITION_ROOT
) ||
1985 (verity
->designator
== designator
)))
1988 if (!m
->found
|| !m
->node
|| !m
->fstype
)
1990 if (!verity
->data_path
) {
1991 if (!v
->found
|| !v
->node
|| !v
->fstype
)
1994 if (!streq(v
->fstype
, "DM_verity_hash"))
1998 r
= dlopen_cryptsetup();
2002 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
2003 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
2004 _cleanup_free_
char *root_hash_encoded
= NULL
;
2006 root_hash_encoded
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2007 if (!root_hash_encoded
)
2010 r
= make_dm_name_and_node(root_hash_encoded
, "-verity", &name
, &node
);
2012 r
= make_dm_name_and_node(m
->node
, "-verity", &name
, &node
);
2016 r
= sym_crypt_init(&cd
, verity
->data_path
?: v
->node
);
2020 cryptsetup_enable_logging(cd
);
2022 r
= sym_crypt_load(cd
, CRYPT_VERITY
, NULL
);
2026 r
= sym_crypt_set_data_device(cd
, m
->node
);
2030 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
2033 /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
2034 * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
2035 * retry a few times before giving up. */
2036 for (unsigned i
= 0; i
< N_DEVICE_NODE_LIST_ATTEMPTS
; i
++) {
2038 r
= do_crypt_activate_verity(cd
, name
, verity
);
2039 /* libdevmapper can return EINVAL when the device is already in the activation stage.
2040 * There's no way to distinguish this situation from a genuine error due to invalid
2041 * parameters, so immediately fall back to activating the device with a unique name.
2042 * Improvements in libcrypsetup can ensure this never happens:
2043 * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
2044 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2046 if (r
< 0 && !IN_SET(r
,
2047 -EEXIST
, /* Volume is already open and ready to be used */
2048 -EBUSY
, /* Volume is being opened but not ready, crypt_init_by_name can fetch details */
2049 -ENODEV
/* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */))
2051 if (IN_SET(r
, -EEXIST
, -EBUSY
)) {
2052 _cleanup_(sym_crypt_freep
) struct crypt_device
*existing_cd
= NULL
;
2054 if (!restore_deferred_remove
){
2055 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
2056 r
= dm_deferred_remove_cancel(name
);
2057 /* If activation returns EBUSY there might be no deferred removal to cancel, that's fine */
2058 if (r
< 0 && r
!= -ENXIO
)
2059 return log_debug_errno(r
, "Disabling automated deferred removal for verity device %s failed: %m", node
);
2061 restore_deferred_remove
= strdup(name
);
2062 if (!restore_deferred_remove
)
2067 r
= verity_can_reuse(verity
, name
, &existing_cd
);
2068 /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
2069 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2071 if (r
< 0 && !IN_SET(r
, -ENODEV
, -ENOENT
, -EBUSY
))
2072 return log_debug_errno(r
, "Checking whether existing verity device %s can be reused failed: %m", node
);
2074 /* devmapper might say that the device exists, but the devlink might not yet have been
2075 * created. Check and wait for the udev event in that case. */
2076 r
= device_wait_for_devlink(node
, "block", verity_timeout(), NULL
);
2077 /* Fallback to activation with a unique device if it's taking too long */
2078 if (r
== -ETIMEDOUT
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2083 crypt_free_and_replace(cd
, existing_cd
);
2089 /* Device is being opened by another process, but it has not finished yet, yield for 2ms */
2090 (void) usleep(2 * USEC_PER_MSEC
);
2093 /* All trials failed or a conflicting verity device exists. Let's try to activate with a unique name. */
2094 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
2095 /* Before trying to activate with unique name, we need to free crypt_device object.
2096 * Otherwise, we get error from libcryptsetup like the following:
2098 * systemd[1234]: Cannot use device /dev/loop5 which is in use (already mapped or mounted).
2103 return verity_partition(designator
, m
, v
, verity
, flags
& ~DISSECT_IMAGE_VERITY_SHARE
, d
);
2106 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
), "All attempts to activate verity device %s failed.", name
);
2109 /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
2110 restore_deferred_remove
= mfree(restore_deferred_remove
);
2112 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
2113 .name
= TAKE_PTR(name
),
2114 .device
= TAKE_PTR(cd
),
2117 m
->decrypted_node
= TAKE_PTR(node
);
2123 int dissected_image_decrypt(
2125 const char *passphrase
,
2126 const VeritySettings
*verity
,
2127 DissectImageFlags flags
) {
2129 #if HAVE_LIBCRYPTSETUP
2130 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
2135 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
2139 * = 0 → There was nothing to decrypt
2140 * > 0 → Decrypted successfully
2141 * -ENOKEY → There's something to decrypt but no key was supplied
2142 * -EKEYREJECTED → Passed key was not correct
2145 if (verity
&& verity
->root_hash
&& verity
->root_hash_size
< sizeof(sd_id128_t
))
2148 if (!m
->encrypted
&& !m
->verity_ready
)
2151 #if HAVE_LIBCRYPTSETUP
2152 r
= decrypted_image_new(&d
);
2156 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
2157 DissectedPartition
*p
= m
->partitions
+ i
;
2158 PartitionDesignator k
;
2163 r
= decrypt_partition(p
, passphrase
, flags
, d
);
2167 k
= PARTITION_VERITY_OF(i
);
2169 r
= verity_partition(i
, p
, m
->partitions
+ k
, verity
, flags
| DISSECT_IMAGE_VERITY_SHARE
, d
);
2174 if (!p
->decrypted_fstype
&& p
->decrypted_node
) {
2175 r
= probe_filesystem(p
->decrypted_node
, &p
->decrypted_fstype
);
2176 if (r
< 0 && r
!= -EUCLEAN
)
2181 m
->decrypted_image
= TAKE_PTR(d
);
2189 int dissected_image_decrypt_interactively(
2191 const char *passphrase
,
2192 const VeritySettings
*verity
,
2193 DissectImageFlags flags
) {
2195 _cleanup_strv_free_erase_
char **z
= NULL
;
2202 r
= dissected_image_decrypt(m
, passphrase
, verity
, flags
);
2205 if (r
== -EKEYREJECTED
)
2206 log_error_errno(r
, "Incorrect passphrase, try again!");
2207 else if (r
!= -ENOKEY
)
2208 return log_error_errno(r
, "Failed to decrypt image: %m");
2211 return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED
),
2212 "Too many retries.");
2216 r
= ask_password_auto("Please enter image passphrase:", NULL
, "dissect", "dissect", "dissect.passphrase", USEC_INFINITY
, 0, &z
);
2218 return log_error_errno(r
, "Failed to query for passphrase: %m");
2224 static int decrypted_image_relinquish(DecryptedImage
*d
) {
2227 /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a
2228 * boolean so that we don't clean it up ourselves either anymore */
2230 #if HAVE_LIBCRYPTSETUP
2233 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
2234 DecryptedPartition
*p
= d
->decrypted
+ i
;
2236 if (p
->relinquished
)
2239 r
= sym_crypt_deactivate_by_name(NULL
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
2241 return log_debug_errno(r
, "Failed to mark %s for auto-removal: %m", p
->name
);
2243 p
->relinquished
= true;
2250 int dissected_image_relinquish(DissectedImage
*m
) {
2255 if (m
->decrypted_image
) {
2256 r
= decrypted_image_relinquish(m
->decrypted_image
);
2262 loop_device_relinquish(m
->loop
);
2267 static char *build_auxiliary_path(const char *image
, const char *suffix
) {
2274 e
= endswith(image
, ".raw");
2276 return strjoin(e
, suffix
);
2278 n
= new(char, e
- image
+ strlen(suffix
) + 1);
2282 strcpy(mempcpy(n
, image
, e
- image
), suffix
);
2286 void verity_settings_done(VeritySettings
*v
) {
2289 v
->root_hash
= mfree(v
->root_hash
);
2290 v
->root_hash_size
= 0;
2292 v
->root_hash_sig
= mfree(v
->root_hash_sig
);
2293 v
->root_hash_sig_size
= 0;
2295 v
->data_path
= mfree(v
->data_path
);
2298 int verity_settings_load(
2299 VeritySettings
*verity
,
2301 const char *root_hash_path
,
2302 const char *root_hash_sig_path
) {
2304 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
2305 size_t root_hash_size
= 0, root_hash_sig_size
= 0;
2306 _cleanup_free_
char *verity_data_path
= NULL
;
2307 PartitionDesignator designator
;
2312 assert(verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
2314 /* If we are asked to load the root hash for a device node, exit early */
2315 if (is_device_path(image
))
2318 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIDECAR");
2319 if (r
< 0 && r
!= -ENXIO
)
2320 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIDECAR, ignoring: %m");
2324 designator
= verity
->designator
;
2326 /* We only fill in what isn't already filled in */
2328 if (!verity
->root_hash
) {
2329 _cleanup_free_
char *text
= NULL
;
2331 if (root_hash_path
) {
2332 /* If explicitly specified it takes precedence */
2333 r
= read_one_line_file(root_hash_path
, &text
);
2338 designator
= PARTITION_ROOT
;
2340 /* Otherwise look for xattr and separate file, and first for the data for root and if
2341 * that doesn't exist for /usr */
2343 if (designator
< 0 || designator
== PARTITION_ROOT
) {
2344 r
= getxattr_malloc(image
, "user.verity.roothash", &text
);
2346 _cleanup_free_
char *p
= NULL
;
2348 if (!IN_SET(r
, -ENODATA
, -ENOENT
) && !ERRNO_IS_NOT_SUPPORTED(r
))
2351 p
= build_auxiliary_path(image
, ".roothash");
2355 r
= read_one_line_file(p
, &text
);
2356 if (r
< 0 && r
!= -ENOENT
)
2361 designator
= PARTITION_ROOT
;
2364 if (!text
&& (designator
< 0 || designator
== PARTITION_USR
)) {
2365 /* So in the "roothash" xattr/file name above the "root" of course primarily
2366 * refers to the root of the Verity Merkle tree. But coincidentally it also
2367 * is the hash for the *root* file system, i.e. the "root" neatly refers to
2368 * two distinct concepts called "root". Taking benefit of this happy
2369 * coincidence we call the file with the root hash for the /usr/ file system
2370 * `usrhash`, because `usrroothash` or `rootusrhash` would just be too
2371 * confusing. We thus drop the reference to the root of the Merkle tree, and
2372 * just indicate which file system it's about. */
2373 r
= getxattr_malloc(image
, "user.verity.usrhash", &text
);
2375 _cleanup_free_
char *p
= NULL
;
2377 if (!IN_SET(r
, -ENODATA
, -ENOENT
) && !ERRNO_IS_NOT_SUPPORTED(r
))
2380 p
= build_auxiliary_path(image
, ".usrhash");
2384 r
= read_one_line_file(p
, &text
);
2385 if (r
< 0 && r
!= -ENOENT
)
2390 designator
= PARTITION_USR
;
2395 r
= unhexmem(text
, strlen(text
), &root_hash
, &root_hash_size
);
2398 if (root_hash_size
< sizeof(sd_id128_t
))
2403 if ((root_hash
|| verity
->root_hash
) && !verity
->root_hash_sig
) {
2404 if (root_hash_sig_path
) {
2405 r
= read_full_file(root_hash_sig_path
, (char**) &root_hash_sig
, &root_hash_sig_size
);
2406 if (r
< 0 && r
!= -ENOENT
)
2410 designator
= PARTITION_ROOT
;
2412 if (designator
< 0 || designator
== PARTITION_ROOT
) {
2413 _cleanup_free_
char *p
= NULL
;
2415 /* Follow naming convention recommended by the relevant RFC:
2416 * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
2417 p
= build_auxiliary_path(image
, ".roothash.p7s");
2421 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
2422 if (r
< 0 && r
!= -ENOENT
)
2425 designator
= PARTITION_ROOT
;
2428 if (!root_hash_sig
&& (designator
< 0 || designator
== PARTITION_USR
)) {
2429 _cleanup_free_
char *p
= NULL
;
2431 p
= build_auxiliary_path(image
, ".usrhash.p7s");
2435 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
2436 if (r
< 0 && r
!= -ENOENT
)
2439 designator
= PARTITION_USR
;
2443 if (root_hash_sig
&& root_hash_sig_size
== 0) /* refuse empty size signatures */
2447 if (!verity
->data_path
) {
2448 _cleanup_free_
char *p
= NULL
;
2450 p
= build_auxiliary_path(image
, ".verity");
2454 if (access(p
, F_OK
) < 0) {
2455 if (errno
!= ENOENT
)
2458 verity_data_path
= TAKE_PTR(p
);
2462 verity
->root_hash
= TAKE_PTR(root_hash
);
2463 verity
->root_hash_size
= root_hash_size
;
2466 if (root_hash_sig
) {
2467 verity
->root_hash_sig
= TAKE_PTR(root_hash_sig
);
2468 verity
->root_hash_sig_size
= root_hash_sig_size
;
2471 if (verity_data_path
)
2472 verity
->data_path
= TAKE_PTR(verity_data_path
);
2474 if (verity
->designator
< 0)
2475 verity
->designator
= designator
;
2480 int dissected_image_load_verity_sig_partition(
2483 VeritySettings
*verity
) {
2485 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
2486 _cleanup_(json_variant_unrefp
) JsonVariant
*v
= NULL
;
2487 size_t root_hash_size
, root_hash_sig_size
;
2488 _cleanup_free_
char *buf
= NULL
;
2489 PartitionDesignator d
;
2490 DissectedPartition
*p
;
2491 JsonVariant
*rh
, *sig
;
2500 if (verity
->root_hash
&& verity
->root_hash_sig
) /* Already loaded? */
2503 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_EMBEDDED");
2504 if (r
< 0 && r
!= -ENXIO
)
2505 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_EMBEDDED, ignoring: %m");
2509 d
= PARTITION_VERITY_SIG_OF(verity
->designator
< 0 ? PARTITION_ROOT
: verity
->designator
);
2512 p
= m
->partitions
+ d
;
2515 if (p
->offset
== UINT64_MAX
|| p
->size
== UINT64_MAX
)
2518 if (p
->size
> 4*1024*1024) /* Signature data cannot possible be larger than 4M, refuse that */
2521 buf
= new(char, p
->size
+1);
2525 n
= pread(fd
, buf
, p
->size
, p
->offset
);
2528 if ((uint64_t) n
!= p
->size
)
2531 e
= memchr(buf
, 0, p
->size
);
2533 /* If we found a NUL byte then the rest of the data must be NUL too */
2534 if (!memeqzero(e
, p
->size
- (e
- buf
)))
2535 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature data contains embedded NUL byte.");
2539 r
= json_parse(buf
, 0, &v
, NULL
, NULL
);
2541 return log_debug_errno(r
, "Failed to parse signature JSON data: %m");
2543 rh
= json_variant_by_key(v
, "rootHash");
2545 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'rootHash' field.");
2546 if (!json_variant_is_string(rh
))
2547 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'rootHash' field of signature JSON object is not a string.");
2549 r
= unhexmem(json_variant_string(rh
), SIZE_MAX
, &root_hash
, &root_hash_size
);
2551 return log_debug_errno(r
, "Failed to parse root hash field: %m");
2553 /* Check if specified root hash matches if it is specified */
2554 if (verity
->root_hash
&&
2555 memcmp_nn(verity
->root_hash
, verity
->root_hash_size
, root_hash
, root_hash_size
) != 0) {
2556 _cleanup_free_
char *a
= NULL
, *b
= NULL
;
2558 a
= hexmem(root_hash
, root_hash_size
);
2559 b
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2561 return log_debug_errno(r
, "Root hash in signature JSON data (%s) doesn't match configured hash (%s).", strna(a
), strna(b
));
2564 sig
= json_variant_by_key(v
, "signature");
2566 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'signature' field.");
2567 if (!json_variant_is_string(sig
))
2568 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'signature' field of signature JSON object is not a string.");
2570 r
= unbase64mem(json_variant_string(sig
), SIZE_MAX
, &root_hash_sig
, &root_hash_sig_size
);
2572 return log_debug_errno(r
, "Failed to parse signature field: %m");
2574 free_and_replace(verity
->root_hash
, root_hash
);
2575 verity
->root_hash_size
= root_hash_size
;
2577 free_and_replace(verity
->root_hash_sig
, root_hash_sig
);
2578 verity
->root_hash_sig_size
= root_hash_sig_size
;
2583 int dissected_image_acquire_metadata(DissectedImage
*m
, DissectImageFlags extra_flags
) {
2590 META_EXTENSION_RELEASE
,
2591 META_HAS_INIT_SYSTEM
,
2595 static const char *const paths
[_META_MAX
] = {
2596 [META_HOSTNAME
] = "/etc/hostname\0",
2597 [META_MACHINE_ID
] = "/etc/machine-id\0",
2598 [META_MACHINE_INFO
] = "/etc/machine-info\0",
2599 [META_OS_RELEASE
] = ("/etc/os-release\0"
2600 "/usr/lib/os-release\0"),
2601 [META_EXTENSION_RELEASE
] = "extension-release\0", /* Used only for logging. */
2602 [META_HAS_INIT_SYSTEM
] = "has-init-system\0", /* ditto */
2605 _cleanup_strv_free_
char **machine_info
= NULL
, **os_release
= NULL
, **extension_release
= NULL
;
2606 _cleanup_close_pair_
int error_pipe
[2] = { -1, -1 };
2607 _cleanup_(rmdir_and_freep
) char *t
= NULL
;
2608 _cleanup_(sigkill_waitp
) pid_t child
= 0;
2609 sd_id128_t machine_id
= SD_ID128_NULL
;
2610 _cleanup_free_
char *hostname
= NULL
;
2611 unsigned n_meta_initialized
= 0;
2612 int fds
[2 * _META_MAX
], r
, v
;
2613 int has_init_system
= -1;
2616 BLOCK_SIGNALS(SIGCHLD
);
2620 for (; n_meta_initialized
< _META_MAX
; n_meta_initialized
++) {
2621 if (!paths
[n_meta_initialized
]) {
2622 fds
[2*n_meta_initialized
] = fds
[2*n_meta_initialized
+1] = -1;
2626 if (pipe2(fds
+ 2*n_meta_initialized
, O_CLOEXEC
) < 0) {
2632 r
= mkdtemp_malloc("/tmp/dissect-XXXXXX", &t
);
2636 if (pipe2(error_pipe
, O_CLOEXEC
) < 0) {
2641 r
= safe_fork("(sd-dissect)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_NEW_MOUNTNS
|FORK_MOUNTNS_SLAVE
, &child
);
2645 /* Child in a new mount namespace */
2646 error_pipe
[0] = safe_close(error_pipe
[0]);
2648 r
= dissected_image_mount(
2654 DISSECT_IMAGE_READ_ONLY
|
2655 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
2656 DISSECT_IMAGE_USR_NO_ROOT
);
2658 log_debug_errno(r
, "Failed to mount dissected image: %m");
2662 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
2663 _cleanup_close_
int fd
= -ENOENT
;
2669 fds
[2*k
] = safe_close(fds
[2*k
]);
2673 case META_EXTENSION_RELEASE
:
2674 /* As per the os-release spec, if the image is an extension it will have a file
2675 * named after the image name in extension-release.d/ - we use the image name
2676 * and try to resolve it with the extension-release helpers, as sometimes
2677 * the image names are mangled on deployment and do not match anymore.
2678 * Unlike other paths this is not fixed, and the image name
2679 * can be mangled on deployment, so by calling into the helper
2680 * we allow a fallback that matches on the first extension-release
2681 * file found in the directory, if one named after the image cannot
2682 * be found first. */
2683 r
= open_extension_release(t
, m
->image_name
, NULL
, &fd
);
2685 fd
= r
; /* Propagate the error. */
2688 case META_HAS_INIT_SYSTEM
: {
2691 FOREACH_STRING(init
,
2692 "/usr/lib/systemd/systemd", /* systemd on /usr merged system */
2693 "/lib/systemd/systemd", /* systemd on /usr non-merged systems */
2694 "/sbin/init") { /* traditional path the Linux kernel invokes */
2696 r
= chase_symlinks(init
, t
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
2699 log_debug_errno(r
, "Failed to resolve %s, ignoring: %m", init
);
2706 r
= loop_write(fds
[2*k
+1], &found
, sizeof(found
), false);
2714 NULSTR_FOREACH(p
, paths
[k
]) {
2715 fd
= chase_symlinks_and_open(p
, t
, CHASE_PREFIX_ROOT
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
, NULL
);
2722 log_debug_errno(fd
, "Failed to read %s file of image, ignoring: %m", paths
[k
]);
2723 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
2727 r
= copy_bytes(fd
, fds
[2*k
+1], UINT64_MAX
, 0);
2731 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
2734 _exit(EXIT_SUCCESS
);
2737 /* Let parent know the error */
2738 (void) write(error_pipe
[1], &r
, sizeof(r
));
2739 _exit(EXIT_FAILURE
);
2742 error_pipe
[1] = safe_close(error_pipe
[1]);
2744 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
2745 _cleanup_fclose_
FILE *f
= NULL
;
2750 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
2752 f
= take_fdopen(&fds
[2*k
], "r");
2761 r
= read_etc_hostname_stream(f
, &hostname
);
2763 log_debug_errno(r
, "Failed to read /etc/hostname of image: %m");
2767 case META_MACHINE_ID
: {
2768 _cleanup_free_
char *line
= NULL
;
2770 r
= read_line(f
, LONG_LINE_MAX
, &line
);
2772 log_debug_errno(r
, "Failed to read /etc/machine-id of image: %m");
2774 r
= sd_id128_from_string(line
, &machine_id
);
2776 log_debug_errno(r
, "Image contains invalid /etc/machine-id: %s", line
);
2778 log_debug("/etc/machine-id file of image is empty.");
2779 else if (streq(line
, "uninitialized"))
2780 log_debug("/etc/machine-id file of image is uninitialized (likely aborted first boot).");
2782 log_debug("/etc/machine-id file of image has unexpected length %i.", r
);
2787 case META_MACHINE_INFO
:
2788 r
= load_env_file_pairs(f
, "machine-info", &machine_info
);
2790 log_debug_errno(r
, "Failed to read /etc/machine-info of image: %m");
2794 case META_OS_RELEASE
:
2795 r
= load_env_file_pairs(f
, "os-release", &os_release
);
2797 log_debug_errno(r
, "Failed to read OS release file of image: %m");
2801 case META_EXTENSION_RELEASE
:
2802 r
= load_env_file_pairs(f
, "extension-release", &extension_release
);
2804 log_debug_errno(r
, "Failed to read extension release file of image: %m");
2808 case META_HAS_INIT_SYSTEM
: {
2813 nr
= fread(&b
, 1, sizeof(b
), f
);
2814 if (nr
!= sizeof(b
))
2815 log_debug_errno(errno_or_else(EIO
), "Failed to read has-init-system boolean: %m");
2817 has_init_system
= b
;
2823 r
= wait_for_terminate_and_check("(sd-dissect)", child
, 0);
2828 n
= read(error_pipe
[0], &v
, sizeof(v
));
2832 return v
; /* propagate error sent to us from child */
2836 if (r
!= EXIT_SUCCESS
)
2839 free_and_replace(m
->hostname
, hostname
);
2840 m
->machine_id
= machine_id
;
2841 strv_free_and_replace(m
->machine_info
, machine_info
);
2842 strv_free_and_replace(m
->os_release
, os_release
);
2843 strv_free_and_replace(m
->extension_release
, extension_release
);
2844 m
->has_init_system
= has_init_system
;
2847 for (unsigned k
= 0; k
< n_meta_initialized
; k
++)
2848 safe_close_pair(fds
+ 2*k
);
2853 int dissect_loop_device(
2855 const VeritySettings
*verity
,
2856 const MountOptions
*mount_options
,
2857 DissectImageFlags flags
,
2858 DissectedImage
**ret
) {
2860 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
2866 r
= dissect_image(loop
->fd
, loop
->node
, loop
->backing_file
?: loop
->node
, verity
, mount_options
, flags
, &m
);
2870 m
->loop
= loop_device_ref(loop
);
2876 int dissect_loop_device_and_warn(
2878 const VeritySettings
*verity
,
2879 const MountOptions
*mount_options
,
2880 DissectImageFlags flags
,
2881 DissectedImage
**ret
) {
2887 assert(loop
->fd
>= 0);
2889 name
= ASSERT_PTR(loop
->backing_file
?: loop
->node
);
2891 r
= dissect_loop_device(loop
, verity
, mount_options
, flags
, ret
);
2895 return log_error_errno(r
, "Dissecting images is not supported, compiled without blkid support.");
2898 return log_error_errno(r
, "%s: Couldn't identify a suitable partition table or file system.", name
);
2901 return log_error_errno(r
, "%s: The image does not pass validation.", name
);
2903 case -EADDRNOTAVAIL
:
2904 return log_error_errno(r
, "%s: No root partition for specified root hash found.", name
);
2907 return log_error_errno(r
, "%s: Multiple suitable root partitions found in image.", name
);
2910 return log_error_errno(r
, "%s: No suitable root partition found in image.", name
);
2912 case -EPROTONOSUPPORT
:
2913 return log_error_errno(r
, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name
);
2916 return log_error_errno(r
, "%s: Image is not a block device.", name
);
2919 return log_error_errno(r
,
2920 "Combining partitioned images (such as '%s') with external Verity data (such as '%s') not supported. "
2921 "(Consider setting $SYSTEMD_DISSECT_VERITY_SIDECAR=0 to disable automatic discovery of external Verity data.)",
2922 name
, strna(verity
? verity
->data_path
: NULL
));
2926 return log_error_errno(r
, "Failed to dissect image '%s': %m", name
);
2932 bool dissected_image_verity_candidate(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
2935 /* Checks if this partition could theoretically do Verity. For non-partitioned images this only works
2936 * if there's an external verity file supplied, for which we can consult .has_verity. For partitioned
2937 * images we only check the partition type.
2939 * This call is used to decide whether to suppress or show a verity column in tabular output of the
2942 if (image
->single_file_system
)
2943 return partition_designator
== PARTITION_ROOT
&& image
->has_verity
;
2945 return PARTITION_VERITY_OF(partition_designator
) >= 0;
2948 bool dissected_image_verity_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
2949 PartitionDesignator k
;
2953 /* Checks if this partition has verity data available that we can activate. For non-partitioned this
2954 * works for the root partition, for others only if the associated verity partition was found. */
2956 if (!image
->verity_ready
)
2959 if (image
->single_file_system
)
2960 return partition_designator
== PARTITION_ROOT
;
2962 k
= PARTITION_VERITY_OF(partition_designator
);
2963 return k
>= 0 && image
->partitions
[k
].found
;
2966 bool dissected_image_verity_sig_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
2967 PartitionDesignator k
;
2971 /* Checks if this partition has verity signature data available that we can use. */
2973 if (!image
->verity_sig_ready
)
2976 if (image
->single_file_system
)
2977 return partition_designator
== PARTITION_ROOT
;
2979 k
= PARTITION_VERITY_SIG_OF(partition_designator
);
2980 return k
>= 0 && image
->partitions
[k
].found
;
2983 MountOptions
* mount_options_free_all(MountOptions
*options
) {
2986 while ((m
= options
)) {
2987 LIST_REMOVE(mount_options
, options
, m
);
2995 const char* mount_options_from_designator(const MountOptions
*options
, PartitionDesignator designator
) {
2996 LIST_FOREACH(mount_options
, m
, options
)
2997 if (designator
== m
->partition_designator
&& !isempty(m
->options
))
3003 int mount_image_privately_interactively(
3005 DissectImageFlags flags
,
3006 char **ret_directory
,
3007 LoopDevice
**ret_loop_device
) {
3009 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3010 _cleanup_(loop_device_unrefp
) LoopDevice
*d
= NULL
;
3011 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3012 _cleanup_(rmdir_and_freep
) char *created_dir
= NULL
;
3013 _cleanup_free_
char *temp
= NULL
;
3016 /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
3017 * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
3021 assert(ret_directory
);
3022 assert(ret_loop_device
);
3024 r
= verity_settings_load(&verity
, image
, NULL
, NULL
);
3026 return log_error_errno(r
, "Failed to load root hash data: %m");
3028 r
= tempfn_random_child(NULL
, program_invocation_short_name
, &temp
);
3030 return log_error_errno(r
, "Failed to generate temporary mount directory: %m");
3032 r
= loop_device_make_by_path(
3034 FLAGS_SET(flags
, DISSECT_IMAGE_DEVICE_READ_ONLY
) ? O_RDONLY
: O_RDWR
,
3035 FLAGS_SET(flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
3039 return log_error_errno(r
, "Failed to set up loopback device for %s: %m", image
);
3041 r
= dissect_loop_device_and_warn(d
, &verity
, NULL
, flags
, &dissected_image
);
3045 r
= dissected_image_load_verity_sig_partition(dissected_image
, d
->fd
, &verity
);
3049 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, &verity
, flags
);
3053 r
= detach_mount_namespace();
3055 return log_error_errno(r
, "Failed to detach mount namespace: %m");
3057 r
= mkdir_p(temp
, 0700);
3059 return log_error_errno(r
, "Failed to create mount point: %m");
3061 created_dir
= TAKE_PTR(temp
);
3063 r
= dissected_image_mount_and_warn(dissected_image
, created_dir
, UID_INVALID
, UID_INVALID
, flags
);
3067 r
= loop_device_flock(d
, LOCK_UN
);
3071 r
= dissected_image_relinquish(dissected_image
);
3073 return log_error_errno(r
, "Failed to relinquish DM and loopback block devices: %m");
3075 *ret_directory
= TAKE_PTR(created_dir
);
3076 *ret_loop_device
= TAKE_PTR(d
);
3081 static const char *const partition_designator_table
[] = {
3082 [PARTITION_ROOT
] = "root",
3083 [PARTITION_ROOT_SECONDARY
] = "root-secondary",
3084 [PARTITION_ROOT_OTHER
] = "root-other",
3085 [PARTITION_USR
] = "usr",
3086 [PARTITION_USR_SECONDARY
] = "usr-secondary",
3087 [PARTITION_USR_OTHER
] = "usr-other",
3088 [PARTITION_HOME
] = "home",
3089 [PARTITION_SRV
] = "srv",
3090 [PARTITION_ESP
] = "esp",
3091 [PARTITION_XBOOTLDR
] = "xbootldr",
3092 [PARTITION_SWAP
] = "swap",
3093 [PARTITION_ROOT_VERITY
] = "root-verity",
3094 [PARTITION_ROOT_SECONDARY_VERITY
] = "root-secondary-verity",
3095 [PARTITION_ROOT_OTHER_VERITY
] = "root-other-verity",
3096 [PARTITION_USR_VERITY
] = "usr-verity",
3097 [PARTITION_USR_SECONDARY_VERITY
] = "usr-secondary-verity",
3098 [PARTITION_USR_OTHER_VERITY
] = "usr-other-verity",
3099 [PARTITION_ROOT_VERITY_SIG
] = "root-verity-sig",
3100 [PARTITION_ROOT_SECONDARY_VERITY_SIG
] = "root-secondary-verity-sig",
3101 [PARTITION_ROOT_OTHER_VERITY_SIG
] = "root-other-verity-sig",
3102 [PARTITION_USR_VERITY_SIG
] = "usr-verity-sig",
3103 [PARTITION_USR_SECONDARY_VERITY_SIG
] = "usr-secondary-verity-sig",
3104 [PARTITION_USR_OTHER_VERITY_SIG
] = "usr-other-verity-sig",
3105 [PARTITION_TMP
] = "tmp",
3106 [PARTITION_VAR
] = "var",
3109 int verity_dissect_and_mount(
3113 const MountOptions
*options
,
3114 const char *required_host_os_release_id
,
3115 const char *required_host_os_release_version_id
,
3116 const char *required_host_os_release_sysext_level
,
3117 const char *required_sysext_scope
) {
3119 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
3120 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3121 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3122 DissectImageFlags dissect_image_flags
;
3128 /* We might get an FD for the image, but we use the original path to look for the dm-verity files */
3129 r
= verity_settings_load(&verity
, src
, NULL
, NULL
);
3131 return log_debug_errno(r
, "Failed to load root hash: %m");
3133 dissect_image_flags
= verity
.data_path
? DISSECT_IMAGE_NO_PARTITION_TABLE
: 0;
3135 /* Note that we don't use loop_device_make here, as the FD is most likely O_PATH which would not be
3136 * accepted by LOOP_CONFIGURE, so just let loop_device_make_by_path reopen it as a regular FD. */
3137 r
= loop_device_make_by_path(
3138 src_fd
>= 0 ? FORMAT_PROC_FD_PATH(src_fd
) : src
,
3140 verity
.data_path
? 0 : LO_FLAGS_PARTSCAN
,
3144 return log_debug_errno(r
, "Failed to create loop device for image: %m");
3146 r
= dissect_loop_device(
3150 dissect_image_flags
,
3152 /* No partition table? Might be a single-filesystem image, try again */
3153 if (!verity
.data_path
&& r
== -ENOPKG
)
3154 r
= dissect_loop_device(
3158 dissect_image_flags
| DISSECT_IMAGE_NO_PARTITION_TABLE
,
3161 return log_debug_errno(r
, "Failed to dissect image: %m");
3163 r
= dissected_image_load_verity_sig_partition(dissected_image
, loop_device
->fd
, &verity
);
3167 r
= dissected_image_decrypt(
3171 dissect_image_flags
);
3173 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
3175 r
= mkdir_p_label(dest
, 0755);
3177 return log_debug_errno(r
, "Failed to create destination directory %s: %m", dest
);
3178 r
= umount_recursive(dest
, 0);
3180 return log_debug_errno(r
, "Failed to umount under destination directory %s: %m", dest
);
3182 r
= dissected_image_mount(dissected_image
, dest
, UID_INVALID
, UID_INVALID
, dissect_image_flags
);
3184 return log_debug_errno(r
, "Failed to mount image: %m");
3186 r
= loop_device_flock(loop_device
, LOCK_UN
);
3188 return log_debug_errno(r
, "Failed to unlock loopback device: %m");
3190 /* If we got os-release values from the caller, then we need to match them with the image's
3191 * extension-release.d/ content. Return -EINVAL if there's any mismatch.
3192 * First, check the distro ID. If that matches, then check the new SYSEXT_LEVEL value if
3193 * available, or else fallback to VERSION_ID. If neither is present (eg: rolling release),
3194 * then a simple match on the ID will be performed. */
3195 if (required_host_os_release_id
) {
3196 _cleanup_strv_free_
char **extension_release
= NULL
;
3198 assert(!isempty(required_host_os_release_id
));
3200 r
= load_extension_release_pairs(dest
, dissected_image
->image_name
, &extension_release
);
3202 return log_debug_errno(r
, "Failed to parse image %s extension-release metadata: %m", dissected_image
->image_name
);
3204 r
= extension_release_validate(
3205 dissected_image
->image_name
,
3206 required_host_os_release_id
,
3207 required_host_os_release_version_id
,
3208 required_host_os_release_sysext_level
,
3209 required_sysext_scope
,
3212 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE
), "Image %s extension-release metadata does not match the root's", dissected_image
->image_name
);
3214 return log_debug_errno(r
, "Failed to compare image %s extension-release metadata with the root's os-release: %m", dissected_image
->image_name
);
3217 r
= dissected_image_relinquish(dissected_image
);
3219 return log_debug_errno(r
, "Failed to relinquish dissected image: %m");
3224 DEFINE_STRING_TABLE_LOOKUP(partition_designator
, PartitionDesignator
);