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/prctl.h>
14 #include "sd-device.h"
17 #include "architecture.h"
18 #include "ask-password-api.h"
19 #include "blkid-util.h"
20 #include "blockdev-util.h"
22 #include "cryptsetup-util.h"
24 #include "device-nodes.h"
25 #include "device-util.h"
26 #include "discover-image.h"
27 #include "dissect-image.h"
31 #include "extension-release.h"
35 #include "fsck-util.h"
37 #include "hexdecoct.h"
38 #include "hostname-setup.h"
39 #include "id128-util.h"
40 #include "import-util.h"
42 #include "mount-util.h"
43 #include "mountpoint-util.h"
44 #include "namespace-util.h"
45 #include "nulstr-util.h"
47 #include "path-util.h"
48 #include "process-util.h"
49 #include "raw-clone.h"
50 #include "resize-fs.h"
51 #include "signal-util.h"
52 #include "stat-util.h"
53 #include "stdio-util.h"
54 #include "string-table.h"
55 #include "string-util.h"
57 #include "tmpfile-util.h"
58 #include "udev-util.h"
59 #include "user-util.h"
60 #include "xattr-util.h"
62 /* how many times to wait for the device nodes to appear */
63 #define N_DEVICE_NODE_LIST_ATTEMPTS 10
65 int probe_filesystem(const char *node
, char **ret_fstype
) {
66 /* Try to find device content type and return it in *ret_fstype. If nothing is found,
67 * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and an
68 * different error otherwise. */
71 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
76 b
= blkid_new_probe_from_filename(node
);
78 return errno_or_else(ENOMEM
);
80 blkid_probe_enable_superblocks(b
, 1);
81 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
);
84 r
= blkid_do_safeprobe(b
);
86 log_debug("No type detected on partition %s", node
);
90 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
),
91 "Results ambiguous for partition %s", node
);
93 return errno_or_else(EIO
);
95 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
117 static int enumerator_for_parent(sd_device
*d
, sd_device_enumerator
**ret
) {
118 _cleanup_(sd_device_enumerator_unrefp
) sd_device_enumerator
*e
= NULL
;
124 r
= sd_device_enumerator_new(&e
);
128 r
= sd_device_enumerator_add_match_subsystem(e
, "block", true);
132 r
= sd_device_enumerator_add_match_parent(e
, d
);
136 r
= sd_device_enumerator_add_match_sysattr(e
, "partition", NULL
, true);
144 static int device_is_partition(
146 sd_device
*expected_parent
,
147 blkid_partition pp
) {
149 const char *v
, *parent_syspath
, *expected_parent_syspath
;
150 blkid_loff_t bsize
, bstart
;
151 uint64_t size
, start
;
152 int partno
, bpartno
, r
;
156 assert(expected_parent
);
159 r
= sd_device_get_subsystem(d
, &v
);
162 if (!streq(v
, "block"))
165 if (sd_device_get_devtype(d
, &v
) < 0 || !streq(v
, "partition"))
168 r
= sd_device_get_parent(d
, &parent
);
170 return false; /* Doesn't have a parent? No relevant to us */
172 r
= sd_device_get_syspath(parent
, &parent_syspath
); /* Check parent of device of this action */
176 r
= sd_device_get_syspath(expected_parent
, &expected_parent_syspath
); /* Check parent of device we are looking for */
180 if (!path_equal(parent_syspath
, expected_parent_syspath
))
181 return false; /* Has a different parent than what we need, not interesting to us */
183 /* On kernel uevents we may find the partition number in the PARTN= field. Let's use that preferably,
184 * since it's cheaper and more importantly: the sysfs attribute "partition" appears to become
185 * available late, hence let's use the property instead, which is available at the moment we see the
187 r
= sd_device_get_property_value(d
, "PARTN", &v
);
189 r
= sd_device_get_sysattr_value(d
, "partition", &v
);
193 r
= safe_atoi(v
, &partno
);
198 bpartno
= blkid_partition_get_partno(pp
);
200 return errno_or_else(EIO
);
202 if (partno
!= bpartno
)
205 r
= sd_device_get_sysattr_value(d
, "start", &v
);
208 r
= safe_atou64(v
, &start
);
213 bstart
= blkid_partition_get_start(pp
);
215 return errno_or_else(EIO
);
217 if (start
!= (uint64_t) bstart
)
220 r
= sd_device_get_sysattr_value(d
, "size", &v
);
223 r
= safe_atou64(v
, &size
);
228 bsize
= blkid_partition_get_size(pp
);
230 return errno_or_else(EIO
);
232 if (size
!= (uint64_t) bsize
)
238 static int find_partition(
241 usec_t timestamp_not_before
,
242 DissectImageFlags flags
,
245 _cleanup_(sd_device_enumerator_unrefp
) sd_device_enumerator
*e
= NULL
;
253 r
= enumerator_for_parent(parent
, &e
);
257 FOREACH_DEVICE(e
, q
) {
260 if (!FLAGS_SET(flags
, DISSECT_IMAGE_NO_UDEV
)) {
261 r
= sd_device_get_usec_initialized(q
, &usec
);
262 if (r
== -EBUSY
) /* Not initialized yet */
267 if (timestamp_not_before
!= USEC_INFINITY
&&
268 usec
< timestamp_not_before
) /* udev database entry older than our attachment? Then it's not ours */
272 r
= device_is_partition(q
, parent
, pp
);
276 *ret
= sd_device_ref(q
);
285 sd_device
*parent_device
;
286 blkid_partition blkidp
;
289 uint64_t uevent_seqnum_not_before
;
290 usec_t timestamp_not_before
;
291 DissectImageFlags flags
;
294 static inline void wait_data_done(struct wait_data
*d
) {
295 sd_device_unref(d
->found
);
298 static int device_monitor_handler(sd_device_monitor
*monitor
, sd_device
*device
, void *userdata
) {
299 struct wait_data
*w
= userdata
;
304 if (device_for_action(device
, SD_DEVICE_REMOVE
))
307 if (w
->diskseq
!= 0) {
310 /* If w->diskseq is non-zero, then we must have a disk seqnum */
311 r
= sd_device_get_diskseq(device
, &diskseq
);
313 log_debug_errno(r
, "Dropping event because it has no diskseq, but waiting for %" PRIu64
, w
->diskseq
);
316 if (diskseq
< w
->diskseq
) {
317 log_debug("Dropping event because diskseq too old (%" PRIu64
" < %" PRIu64
")",
318 diskseq
, w
->diskseq
);
321 if (diskseq
> w
->diskseq
) {
323 goto finish
; /* Newer than what we were expecting, so we missed it, stop waiting */
325 } else if (w
->uevent_seqnum_not_before
!= UINT64_MAX
) {
328 r
= sd_device_get_seqnum(device
, &seqnum
);
332 if (seqnum
<= w
->uevent_seqnum_not_before
) { /* From an older use of this loop device */
333 log_debug("Dropping event because seqnum too old (%" PRIu64
" <= %" PRIu64
")",
334 seqnum
, w
->uevent_seqnum_not_before
);
339 r
= device_is_partition(device
, w
->parent_device
, w
->blkidp
);
342 if (r
== 0) /* Not the one we need */
345 /* It's the one we need! Yay! */
347 w
->found
= sd_device_ref(device
);
351 return sd_event_exit(sd_device_monitor_get_event(monitor
), r
);
354 static int timeout_handler(sd_event_source
*s
, uint64_t usec
, void *userdata
) {
355 struct wait_data
*w
= userdata
;
360 /* Why partition not appeared within the timeout? We may lost some uevent, as some properties
361 * were not ready when we received uevent... Not sure, but anyway, let's try to find the
362 * partition again before give up. */
364 r
= find_partition(w
->parent_device
, w
->blkidp
, w
->timestamp_not_before
, w
->flags
, &w
->found
);
366 return log_debug_errno(SYNTHETIC_ERRNO(ETIMEDOUT
),
367 "Partition still not appeared after timeout reached.");
369 return log_debug_errno(r
, "Failed to find partition: %m");
371 log_debug("Partition appeared after timeout reached.");
372 return sd_event_exit(sd_event_source_get_event(s
), 0);
375 static int retry_handler(sd_event_source
*s
, uint64_t usec
, void *userdata
) {
376 struct wait_data
*w
= userdata
;
381 r
= find_partition(w
->parent_device
, w
->blkidp
, w
->timestamp_not_before
, w
->flags
, &w
->found
);
384 return log_debug_errno(r
, "Failed to find partition: %m");
386 log_debug("Partition found by a periodic search.");
387 return sd_event_exit(sd_event_source_get_event(s
), 0);
390 r
= sd_event_source_set_time_relative(s
, 500 * USEC_PER_MSEC
);
394 return sd_event_source_set_enabled(s
, SD_EVENT_ONESHOT
);
397 static int wait_for_partition_device(
402 uint64_t uevent_seqnum_not_before
,
403 usec_t timestamp_not_before
,
404 DissectImageFlags flags
,
407 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timeout_source
= NULL
, *retry_source
= NULL
;
408 _cleanup_(sd_device_monitor_unrefp
) sd_device_monitor
*monitor
= NULL
;
409 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
416 r
= find_partition(parent
, pp
, timestamp_not_before
, flags
, ret
);
420 r
= sd_event_new(&event
);
424 r
= sd_device_monitor_new(&monitor
);
428 r
= sd_device_monitor_filter_add_match_subsystem_devtype(monitor
, "block", "partition");
432 r
= sd_device_monitor_filter_add_match_parent(monitor
, parent
, true);
436 r
= sd_device_monitor_filter_add_match_sysattr(monitor
, "partition", NULL
, true);
440 r
= sd_device_monitor_attach_event(monitor
, event
);
444 _cleanup_(wait_data_done
) struct wait_data w
= {
445 .parent_device
= parent
,
448 .uevent_seqnum_not_before
= uevent_seqnum_not_before
,
449 .timestamp_not_before
= timestamp_not_before
,
453 r
= sd_device_monitor_start(monitor
, device_monitor_handler
, &w
);
457 /* Check again, the partition might have appeared in the meantime */
458 r
= find_partition(parent
, pp
, timestamp_not_before
, flags
, ret
);
462 if (deadline
!= USEC_INFINITY
) {
463 r
= sd_event_add_time(
464 event
, &timeout_source
,
465 CLOCK_MONOTONIC
, deadline
, 0,
466 timeout_handler
, &w
);
470 r
= sd_event_source_set_exit_on_failure(timeout_source
, true);
475 /* If we don't have a disk sequence number then we cannot do exact matching,
476 * and we cannot know if we missed it or if it has not been sent yet, so set
477 * up additional retries to increase the chances of receiving the event. */
479 r
= sd_event_add_time_relative(
480 event
, &retry_source
,
481 CLOCK_MONOTONIC
, 500 * USEC_PER_MSEC
, 0,
486 r
= sd_event_source_set_exit_on_failure(retry_source
, true);
491 r
= sd_event_loop(event
);
496 *ret
= TAKE_PTR(w
.found
);
500 static void check_partition_flags(
502 unsigned long long pflags
,
503 unsigned long long supported
) {
507 /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
508 pflags
&= ~(supported
| GPT_FLAG_REQUIRED_PARTITION
| GPT_FLAG_NO_BLOCK_IO_PROTOCOL
| GPT_FLAG_LEGACY_BIOS_BOOTABLE
);
513 /* If there are other bits set, then log about it, to make things discoverable */
514 for (unsigned i
= 0; i
< sizeof(pflags
) * 8; i
++) {
515 unsigned long long bit
= 1ULL << i
;
516 if (!FLAGS_SET(pflags
, bit
))
519 log_debug("Unexpected partition flag %llu set on %s!", bit
, node
);
523 static int device_wait_for_initialization_harder(
525 const char *subsystem
,
529 usec_t start
, left
, retrigger_timeout
;
532 start
= now(CLOCK_MONOTONIC
);
533 left
= usec_sub_unsigned(deadline
, start
);
536 const char *sn
= NULL
;
538 (void) sd_device_get_sysname(device
, &sn
);
539 log_device_debug(device
,
540 "Waiting for device '%s' to initialize for %s.", strna(sn
), FORMAT_TIMESPAN(left
, 0));
543 if (left
!= USEC_INFINITY
)
544 retrigger_timeout
= CLAMP(left
/ 4, 1 * USEC_PER_SEC
, 5 * USEC_PER_SEC
); /* A fourth of the total timeout, but let's clamp to 1s…5s range */
546 retrigger_timeout
= 2 * USEC_PER_SEC
;
549 usec_t local_deadline
, n
;
552 n
= now(CLOCK_MONOTONIC
);
555 /* Find next deadline, when we'll retrigger */
556 local_deadline
= start
+
557 DIV_ROUND_UP(n
- start
, retrigger_timeout
) * retrigger_timeout
;
559 if (deadline
!= USEC_INFINITY
&& deadline
<= local_deadline
) {
560 local_deadline
= deadline
;
565 r
= device_wait_for_initialization(device
, subsystem
, local_deadline
, ret
);
566 if (r
>= 0 && DEBUG_LOGGING
) {
567 const char *sn
= NULL
;
569 (void) sd_device_get_sysname(device
, &sn
);
570 log_device_debug(device
,
571 "Successfully waited for device '%s' to initialize for %s.",
573 FORMAT_TIMESPAN(usec_sub_unsigned(now(CLOCK_MONOTONIC
), start
), 0));
576 if (r
!= -ETIMEDOUT
|| last_try
)
580 log_device_debug(device
,
581 "Device didn't initialize within %s, assuming lost event. Retriggering device.",
582 FORMAT_TIMESPAN(usec_sub_unsigned(now(CLOCK_MONOTONIC
), start
), 0));
584 r
= sd_device_trigger(device
, SD_DEVICE_CHANGE
);
591 #define DEVICE_TIMEOUT_USEC (45 * USEC_PER_SEC)
593 static void dissected_partition_done(DissectedPartition
*p
) {
599 free(p
->decrypted_fstype
);
600 free(p
->decrypted_node
);
601 free(p
->mount_options
);
603 *p
= (DissectedPartition
) {
611 const VeritySettings
*verity
,
612 const MountOptions
*mount_options
,
614 uint64_t uevent_seqnum_not_before
,
615 usec_t timestamp_not_before
,
616 DissectImageFlags flags
,
617 DissectedImage
**ret
) {
620 #ifdef GPT_ROOT_NATIVE
621 sd_id128_t root_uuid
= SD_ID128_NULL
, root_verity_uuid
= SD_ID128_NULL
;
623 #ifdef GPT_USR_NATIVE
624 sd_id128_t usr_uuid
= SD_ID128_NULL
, usr_verity_uuid
= SD_ID128_NULL
;
626 bool is_gpt
, is_mbr
, multiple_generic
= false,
627 generic_rw
= false, /* initialize to appease gcc */
628 generic_growfs
= false;
629 _cleanup_(sd_device_unrefp
) sd_device
*d
= NULL
;
630 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
631 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
632 _cleanup_free_
char *generic_node
= NULL
;
633 sd_id128_t generic_uuid
= SD_ID128_NULL
;
634 const char *pttype
= NULL
, *sysname
= NULL
;
636 int r
, generic_nr
= -1, n_partitions
;
642 assert(!verity
|| verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
643 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
644 assert(!verity
|| verity
->root_hash_sig
|| verity
->root_hash_sig_size
== 0);
645 assert(!verity
|| (verity
->root_hash
|| !verity
->root_hash_sig
));
646 assert(!((flags
& DISSECT_IMAGE_GPT_ONLY
) && (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)));
648 /* Probes a disk image, and returns information about what it found in *ret.
650 * Returns -ENOPKG if no suitable partition table or file system could be found.
651 * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found.
652 * Returns -ENXIO if we couldn't find any partition suitable as root or /usr partition
653 * Returns -ENOTUNIQ if we only found multiple generic partitions and thus don't know what to do with that */
655 if (verity
&& verity
->root_hash
) {
656 sd_id128_t fsuuid
, vuuid
;
658 /* If a root hash is supplied, then we use the root partition that has a UUID that match the
659 * first 128bit of the root hash. And we use the verity partition that has a UUID that match
660 * the final 128bit. */
662 if (verity
->root_hash_size
< sizeof(sd_id128_t
))
665 memcpy(&fsuuid
, verity
->root_hash
, sizeof(sd_id128_t
));
666 memcpy(&vuuid
, (const uint8_t*) verity
->root_hash
+ verity
->root_hash_size
- sizeof(sd_id128_t
), sizeof(sd_id128_t
));
668 if (sd_id128_is_null(fsuuid
))
670 if (sd_id128_is_null(vuuid
))
673 /* If the verity data declares it's for the /usr partition, then search for that, in all
674 * other cases assume it's for the root partition. */
675 #ifdef GPT_USR_NATIVE
676 if (verity
->designator
== PARTITION_USR
) {
678 usr_verity_uuid
= vuuid
;
681 #ifdef GPT_ROOT_NATIVE
683 root_verity_uuid
= vuuid
;
685 #ifdef GPT_USR_NATIVE
690 if (fstat(fd
, &st
) < 0)
693 if (!S_ISBLK(st
.st_mode
))
696 r
= sd_device_new_from_stat_rdev(&d
, &st
);
700 if (!FLAGS_SET(flags
, DISSECT_IMAGE_NO_UDEV
)) {
701 _cleanup_(sd_device_unrefp
) sd_device
*initialized
= NULL
;
703 /* If udev support is enabled, then let's wait for the device to be initialized before we doing anything. */
705 r
= device_wait_for_initialization_harder(
708 usec_add(now(CLOCK_MONOTONIC
), DEVICE_TIMEOUT_USEC
),
714 d
= TAKE_PTR(initialized
);
717 b
= blkid_new_probe();
722 r
= blkid_probe_set_device(b
, fd
, 0, 0);
724 return errno_or_else(ENOMEM
);
726 if ((flags
& DISSECT_IMAGE_GPT_ONLY
) == 0) {
727 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
728 blkid_probe_enable_superblocks(b
, 1);
729 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
|BLKID_SUBLKS_USAGE
);
732 blkid_probe_enable_partitions(b
, 1);
733 blkid_probe_set_partitions_flags(b
, BLKID_PARTS_ENTRY_DETAILS
);
736 r
= blkid_do_safeprobe(b
);
737 if (IN_SET(r
, -2, 1))
738 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG
), "Failed to identify any partition table.");
740 return errno_or_else(EIO
);
742 m
= new0(DissectedImage
, 1);
746 r
= sd_device_get_sysname(d
, &sysname
);
748 return log_debug_errno(r
, "Failed to get device sysname: %m");
749 if (startswith(sysname
, "loop")) {
750 _cleanup_free_
char *name_stripped
= NULL
;
751 const char *full_path
;
753 r
= sd_device_get_sysattr_value(d
, "loop/backing_file", &full_path
);
755 log_debug_errno(r
, "Failed to lookup image name via loop device backing file sysattr, ignoring: %m");
757 r
= raw_strip_suffixes(basename(full_path
), &name_stripped
);
762 free_and_replace(m
->image_name
, name_stripped
);
764 r
= free_and_strdup(&m
->image_name
, sysname
);
769 if (!image_name_is_valid(m
->image_name
)) {
770 log_debug("Image name %s is not valid, ignoring", strempty(m
->image_name
));
771 m
->image_name
= mfree(m
->image_name
);
774 if ((!(flags
& DISSECT_IMAGE_GPT_ONLY
) &&
775 (flags
& DISSECT_IMAGE_GENERIC_ROOT
)) ||
776 (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)) {
777 const char *usage
= NULL
;
779 /* If flags permit this, also allow using non-partitioned single-filesystem images */
781 (void) blkid_probe_lookup_value(b
, "USAGE", &usage
, NULL
);
782 if (STRPTR_IN_SET(usage
, "filesystem", "crypto")) {
783 const char *fstype
= NULL
, *options
= NULL
, *devname
= NULL
;
784 _cleanup_free_
char *t
= NULL
, *n
= NULL
, *o
= NULL
;
786 /* OK, we have found a file system, that's our root partition then. */
787 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
795 r
= sd_device_get_devname(d
, &devname
);
803 m
->single_file_system
= true;
804 m
->encrypted
= streq_ptr(fstype
, "crypto_LUKS");
806 m
->has_verity
= verity
&& verity
->data_path
;
807 m
->verity_ready
= m
->has_verity
&&
809 (verity
->designator
< 0 || verity
->designator
== PARTITION_ROOT
);
811 m
->has_verity_sig
= false; /* signature not embedded, must be specified */
812 m
->verity_sig_ready
= m
->verity_ready
&&
813 verity
->root_hash_sig
;
815 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
822 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
824 .rw
= !m
->verity_ready
&& !fstype_is_ro(fstype
),
826 .architecture
= _ARCHITECTURE_INVALID
,
827 .fstype
= TAKE_PTR(t
),
829 .mount_options
= TAKE_PTR(o
),
839 (void) blkid_probe_lookup_value(b
, "PTTYPE", &pttype
, NULL
);
843 is_gpt
= streq_ptr(pttype
, "gpt");
844 is_mbr
= streq_ptr(pttype
, "dos");
846 if (!is_gpt
&& ((flags
& DISSECT_IMAGE_GPT_ONLY
) || !is_mbr
))
849 /* We support external verity data partitions only if the image has no partition table */
850 if (verity
&& verity
->data_path
)
853 /* Safety check: refuse block devices that carry a partition table but for which the kernel doesn't
854 * do partition scanning. */
855 r
= blockdev_partscan_enabled(fd
);
859 return -EPROTONOSUPPORT
;
862 pl
= blkid_probe_get_partitions(b
);
864 return errno_or_else(ENOMEM
);
867 n_partitions
= blkid_partlist_numof_partitions(pl
);
868 if (n_partitions
< 0)
869 return errno_or_else(EIO
);
871 deadline
= usec_add(now(CLOCK_MONOTONIC
), DEVICE_TIMEOUT_USEC
);
872 for (int i
= 0; i
< n_partitions
; i
++) {
873 _cleanup_(sd_device_unrefp
) sd_device
*q
= NULL
;
874 unsigned long long pflags
;
875 blkid_loff_t start
, size
;
881 pp
= blkid_partlist_get_partition(pl
, i
);
883 return errno_or_else(EIO
);
885 r
= wait_for_partition_device(d
, pp
, deadline
, diskseq
, uevent_seqnum_not_before
, timestamp_not_before
, flags
, &q
);
889 r
= sd_device_get_devname(q
, &node
);
893 pflags
= blkid_partition_get_flags(pp
);
896 nr
= blkid_partition_get_partno(pp
);
898 return errno_or_else(EIO
);
901 start
= blkid_partition_get_start(pp
);
903 return errno_or_else(EIO
);
905 assert((uint64_t) start
< UINT64_MAX
/512);
908 size
= blkid_partition_get_size(pp
);
910 return errno_or_else(EIO
);
912 assert((uint64_t) size
< UINT64_MAX
/512);
915 PartitionDesignator designator
= _PARTITION_DESIGNATOR_INVALID
;
916 int architecture
= _ARCHITECTURE_INVALID
;
917 const char *stype
, *sid
, *fstype
= NULL
, *label
;
918 sd_id128_t type_id
, id
;
919 bool rw
= true, growfs
= false;
921 sid
= blkid_partition_get_uuid(pp
);
924 if (sd_id128_from_string(sid
, &id
) < 0)
927 stype
= blkid_partition_get_type_string(pp
);
930 if (sd_id128_from_string(stype
, &type_id
) < 0)
933 label
= blkid_partition_get_name(pp
); /* libblkid returns NULL here if empty */
935 if (sd_id128_equal(type_id
, GPT_HOME
)) {
937 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
|GPT_FLAG_GROWFS
);
939 if (pflags
& GPT_FLAG_NO_AUTO
)
942 designator
= PARTITION_HOME
;
943 rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
944 growfs
= FLAGS_SET(pflags
, GPT_FLAG_GROWFS
);
946 } else if (sd_id128_equal(type_id
, GPT_SRV
)) {
948 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
|GPT_FLAG_GROWFS
);
950 if (pflags
& GPT_FLAG_NO_AUTO
)
953 designator
= PARTITION_SRV
;
954 rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
955 growfs
= FLAGS_SET(pflags
, GPT_FLAG_GROWFS
);
957 } else if (sd_id128_equal(type_id
, GPT_ESP
)) {
959 /* Note that we don't check the GPT_FLAG_NO_AUTO flag for the ESP, as it is
960 * not defined there. We instead check the GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as
961 * recommended by the UEFI spec (See "12.3.3 Number and Location of System
964 if (pflags
& GPT_FLAG_NO_BLOCK_IO_PROTOCOL
)
967 designator
= PARTITION_ESP
;
970 } else if (sd_id128_equal(type_id
, GPT_XBOOTLDR
)) {
972 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
|GPT_FLAG_GROWFS
);
974 if (pflags
& GPT_FLAG_NO_AUTO
)
977 designator
= PARTITION_XBOOTLDR
;
978 rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
979 growfs
= FLAGS_SET(pflags
, GPT_FLAG_GROWFS
);
981 #ifdef GPT_ROOT_NATIVE
982 else if (sd_id128_equal(type_id
, GPT_ROOT_NATIVE
)) {
984 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
|GPT_FLAG_GROWFS
);
986 if (pflags
& GPT_FLAG_NO_AUTO
)
989 /* If a root ID is specified, ignore everything but the root id */
990 if (!sd_id128_is_null(root_uuid
) && !sd_id128_equal(root_uuid
, id
))
993 designator
= PARTITION_ROOT
;
994 architecture
= native_architecture();
995 rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
996 growfs
= FLAGS_SET(pflags
, GPT_FLAG_GROWFS
);
998 } else if (sd_id128_equal(type_id
, GPT_ROOT_NATIVE_VERITY
)) {
1000 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
1002 if (pflags
& GPT_FLAG_NO_AUTO
)
1005 m
->has_verity
= true;
1007 /* If no verity configuration is specified, then don't do verity */
1010 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
1013 /* If root hash is specified, then ignore everything but the root id */
1014 if (!sd_id128_is_null(root_verity_uuid
) && !sd_id128_equal(root_verity_uuid
, id
))
1017 designator
= PARTITION_ROOT_VERITY
;
1018 fstype
= "DM_verity_hash";
1019 architecture
= native_architecture();
1022 } else if (sd_id128_equal(type_id
, GPT_ROOT_NATIVE_VERITY_SIG
)) {
1024 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
1026 if (pflags
& GPT_FLAG_NO_AUTO
)
1029 m
->has_verity_sig
= true;
1031 /* If root hash is specified explicitly, then ignore any embedded signature */
1034 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
1036 if (verity
->root_hash
)
1039 designator
= PARTITION_ROOT_VERITY_SIG
;
1040 fstype
= "verity_hash_signature";
1041 architecture
= native_architecture();
1045 #ifdef GPT_ROOT_SECONDARY
1046 else if (sd_id128_equal(type_id
, GPT_ROOT_SECONDARY
)) {
1048 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
|GPT_FLAG_GROWFS
);
1050 if (pflags
& GPT_FLAG_NO_AUTO
)
1053 /* If a root ID is specified, ignore everything but the root id */
1054 if (!sd_id128_is_null(root_uuid
) && !sd_id128_equal(root_uuid
, id
))
1057 designator
= PARTITION_ROOT_SECONDARY
;
1058 architecture
= SECONDARY_ARCHITECTURE
;
1059 rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
1060 growfs
= FLAGS_SET(pflags
, GPT_FLAG_GROWFS
);
1062 } else if (sd_id128_equal(type_id
, GPT_ROOT_SECONDARY_VERITY
)) {
1064 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
1066 if (pflags
& GPT_FLAG_NO_AUTO
)
1069 m
->has_verity
= true;
1071 /* Don't do verity if no verity config is passed in */
1074 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
1077 /* If root hash is specified, then ignore everything but the root id */
1078 if (!sd_id128_is_null(root_verity_uuid
) && !sd_id128_equal(root_verity_uuid
, id
))
1081 designator
= PARTITION_ROOT_SECONDARY_VERITY
;
1082 fstype
= "DM_verity_hash";
1083 architecture
= SECONDARY_ARCHITECTURE
;
1086 } else if (sd_id128_equal(type_id
, GPT_ROOT_SECONDARY_VERITY_SIG
)) {
1088 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
1090 if (pflags
& GPT_FLAG_NO_AUTO
)
1093 m
->has_verity_sig
= true;
1095 /* If root hash is specified explicitly, then ignore any embedded signature */
1098 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_ROOT
)
1100 if (verity
->root_hash
)
1103 designator
= PARTITION_ROOT_SECONDARY_VERITY_SIG
;
1104 fstype
= "verity_hash_signature";
1105 architecture
= native_architecture();
1109 #ifdef GPT_USR_NATIVE
1110 else if (sd_id128_equal(type_id
, GPT_USR_NATIVE
)) {
1112 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
|GPT_FLAG_GROWFS
);
1114 if (pflags
& GPT_FLAG_NO_AUTO
)
1117 /* If a usr ID is specified, ignore everything but the usr id */
1118 if (!sd_id128_is_null(usr_uuid
) && !sd_id128_equal(usr_uuid
, id
))
1121 designator
= PARTITION_USR
;
1122 architecture
= native_architecture();
1123 rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
1124 growfs
= FLAGS_SET(pflags
, GPT_FLAG_GROWFS
);
1126 } else if (sd_id128_equal(type_id
, GPT_USR_NATIVE_VERITY
)) {
1128 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
1130 if (pflags
& GPT_FLAG_NO_AUTO
)
1133 m
->has_verity
= true;
1137 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
1140 /* If usr hash is specified, then ignore everything but the usr id */
1141 if (!sd_id128_is_null(usr_verity_uuid
) && !sd_id128_equal(usr_verity_uuid
, id
))
1144 designator
= PARTITION_USR_VERITY
;
1145 fstype
= "DM_verity_hash";
1146 architecture
= native_architecture();
1149 } else if (sd_id128_equal(type_id
, GPT_USR_NATIVE_VERITY_SIG
)) {
1151 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
1153 if (pflags
& GPT_FLAG_NO_AUTO
)
1156 m
->has_verity_sig
= true;
1158 /* If usr hash is specified explicitly, then ignore any embedded signature */
1161 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
1163 if (verity
->root_hash
)
1166 designator
= PARTITION_USR_VERITY_SIG
;
1167 fstype
= "verity_hash_signature";
1168 architecture
= native_architecture();
1172 #ifdef GPT_USR_SECONDARY
1173 else if (sd_id128_equal(type_id
, GPT_USR_SECONDARY
)) {
1175 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
|GPT_FLAG_GROWFS
);
1177 if (pflags
& GPT_FLAG_NO_AUTO
)
1180 /* If a usr ID is specified, ignore everything but the usr id */
1181 if (!sd_id128_is_null(usr_uuid
) && !sd_id128_equal(usr_uuid
, id
))
1184 designator
= PARTITION_USR_SECONDARY
;
1185 architecture
= SECONDARY_ARCHITECTURE
;
1186 rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
1187 growfs
= FLAGS_SET(pflags
, GPT_FLAG_GROWFS
);
1189 } else if (sd_id128_equal(type_id
, GPT_USR_SECONDARY_VERITY
)) {
1191 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
1193 if (pflags
& GPT_FLAG_NO_AUTO
)
1196 m
->has_verity
= true;
1200 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
1203 /* If usr hash is specified, then ignore everything but the root id */
1204 if (!sd_id128_is_null(usr_verity_uuid
) && !sd_id128_equal(usr_verity_uuid
, id
))
1207 designator
= PARTITION_USR_SECONDARY_VERITY
;
1208 fstype
= "DM_verity_hash";
1209 architecture
= SECONDARY_ARCHITECTURE
;
1212 } else if (sd_id128_equal(type_id
, GPT_USR_SECONDARY_VERITY_SIG
)) {
1214 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
1216 if (pflags
& GPT_FLAG_NO_AUTO
)
1219 m
->has_verity_sig
= true;
1221 /* If usr hash is specified explicitly, then ignore any embedded signature */
1224 if (verity
->designator
>= 0 && verity
->designator
!= PARTITION_USR
)
1226 if (verity
->root_hash
)
1229 designator
= PARTITION_USR_SECONDARY_VERITY_SIG
;
1230 fstype
= "verity_hash_signature";
1231 architecture
= native_architecture();
1235 else if (sd_id128_equal(type_id
, GPT_SWAP
)) {
1237 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
);
1239 if (pflags
& GPT_FLAG_NO_AUTO
)
1242 designator
= PARTITION_SWAP
;
1244 } else if (sd_id128_equal(type_id
, GPT_LINUX_GENERIC
)) {
1246 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
|GPT_FLAG_GROWFS
);
1248 if (pflags
& GPT_FLAG_NO_AUTO
)
1252 multiple_generic
= true;
1255 generic_rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
1256 generic_growfs
= FLAGS_SET(pflags
, GPT_FLAG_GROWFS
);
1258 generic_node
= strdup(node
);
1263 } else if (sd_id128_equal(type_id
, GPT_TMP
)) {
1265 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
|GPT_FLAG_GROWFS
);
1267 if (pflags
& GPT_FLAG_NO_AUTO
)
1270 designator
= PARTITION_TMP
;
1271 rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
1272 growfs
= FLAGS_SET(pflags
, GPT_FLAG_GROWFS
);
1274 } else if (sd_id128_equal(type_id
, GPT_VAR
)) {
1276 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
|GPT_FLAG_GROWFS
);
1278 if (pflags
& GPT_FLAG_NO_AUTO
)
1281 if (!FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_VAR_CHECK
)) {
1282 sd_id128_t var_uuid
;
1284 /* For /var we insist that the uuid of the partition matches the
1285 * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
1286 * ID. Why? Unlike the other partitions /var is inherently
1287 * installation specific, hence we need to be careful not to mount it
1288 * in the wrong installation. By hashing the partition UUID from
1289 * /etc/machine-id we can securely bind the partition to the
1292 r
= sd_id128_get_machine_app_specific(GPT_VAR
, &var_uuid
);
1296 if (!sd_id128_equal(var_uuid
, id
)) {
1297 log_debug("Found a /var/ partition, but its UUID didn't match our expectations, ignoring.");
1302 designator
= PARTITION_VAR
;
1303 rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
1304 growfs
= FLAGS_SET(pflags
, GPT_FLAG_GROWFS
);
1307 if (designator
!= _PARTITION_DESIGNATOR_INVALID
) {
1308 _cleanup_free_
char *t
= NULL
, *n
= NULL
, *o
= NULL
, *l
= NULL
;
1309 const char *options
= NULL
;
1311 if (m
->partitions
[designator
].found
) {
1312 /* For most partition types the first one we see wins. Except for the
1313 * rootfs and /usr, where we do a version compare of the label, and
1314 * let the newest version win. This permits a simple A/B versioning
1315 * scheme in OS images. */
1317 if (!PARTITION_DESIGNATOR_VERSIONED(designator
) ||
1318 strverscmp_improved(m
->partitions
[designator
].label
, label
) >= 0)
1321 dissected_partition_done(m
->partitions
+ designator
);
1340 options
= mount_options_from_designator(mount_options
, designator
);
1342 o
= strdup(options
);
1347 m
->partitions
[designator
] = (DissectedPartition
) {
1352 .architecture
= architecture
,
1353 .node
= TAKE_PTR(n
),
1354 .fstype
= TAKE_PTR(t
),
1355 .label
= TAKE_PTR(l
),
1357 .mount_options
= TAKE_PTR(o
),
1358 .offset
= (uint64_t) start
* 512,
1359 .size
= (uint64_t) size
* 512,
1363 } else if (is_mbr
) {
1365 switch (blkid_partition_get_type(pp
)) {
1367 case 0x83: /* Linux partition */
1369 if (pflags
!= 0x80) /* Bootable flag */
1373 multiple_generic
= true;
1377 generic_growfs
= false;
1378 generic_node
= strdup(node
);
1385 case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
1386 _cleanup_free_
char *n
= NULL
, *o
= NULL
;
1387 sd_id128_t id
= SD_ID128_NULL
;
1388 const char *sid
, *options
= NULL
;
1390 /* First one wins */
1391 if (m
->partitions
[PARTITION_XBOOTLDR
].found
)
1394 sid
= blkid_partition_get_uuid(pp
);
1396 (void) sd_id128_from_string(sid
, &id
);
1402 options
= mount_options_from_designator(mount_options
, PARTITION_XBOOTLDR
);
1404 o
= strdup(options
);
1409 m
->partitions
[PARTITION_XBOOTLDR
] = (DissectedPartition
) {
1414 .architecture
= _ARCHITECTURE_INVALID
,
1415 .node
= TAKE_PTR(n
),
1417 .mount_options
= TAKE_PTR(o
),
1418 .offset
= (uint64_t) start
* 512,
1419 .size
= (uint64_t) size
* 512,
1427 if (m
->partitions
[PARTITION_ROOT
].found
) {
1428 /* If we found the primary arch, then invalidate the secondary arch to avoid any ambiguities,
1429 * since we never want to mount the secondary arch in this case. */
1430 m
->partitions
[PARTITION_ROOT_SECONDARY
].found
= false;
1431 m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY
].found
= false;
1432 m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY_SIG
].found
= false;
1433 m
->partitions
[PARTITION_USR_SECONDARY
].found
= false;
1434 m
->partitions
[PARTITION_USR_SECONDARY_VERITY
].found
= false;
1435 m
->partitions
[PARTITION_USR_SECONDARY_VERITY_SIG
].found
= false;
1437 } else if (m
->partitions
[PARTITION_ROOT_VERITY
].found
||
1438 m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
)
1439 return -EADDRNOTAVAIL
; /* Verity found but no matching rootfs? Something is off, refuse. */
1441 else if (m
->partitions
[PARTITION_ROOT_SECONDARY
].found
) {
1443 /* No root partition found but there's one for the secondary architecture? Then upgrade
1444 * secondary arch to first */
1446 m
->partitions
[PARTITION_ROOT
] = m
->partitions
[PARTITION_ROOT_SECONDARY
];
1447 zero(m
->partitions
[PARTITION_ROOT_SECONDARY
]);
1448 m
->partitions
[PARTITION_ROOT_VERITY
] = m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY
];
1449 zero(m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY
]);
1450 m
->partitions
[PARTITION_ROOT_VERITY_SIG
] = m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY_SIG
];
1451 zero(m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY_SIG
]);
1453 m
->partitions
[PARTITION_USR
] = m
->partitions
[PARTITION_USR_SECONDARY
];
1454 zero(m
->partitions
[PARTITION_USR_SECONDARY
]);
1455 m
->partitions
[PARTITION_USR_VERITY
] = m
->partitions
[PARTITION_USR_SECONDARY_VERITY
];
1456 zero(m
->partitions
[PARTITION_USR_SECONDARY_VERITY
]);
1457 m
->partitions
[PARTITION_USR_VERITY_SIG
] = m
->partitions
[PARTITION_USR_SECONDARY_VERITY_SIG
];
1458 zero(m
->partitions
[PARTITION_USR_SECONDARY_VERITY_SIG
]);
1460 } else if (m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY
].found
||
1461 m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY_SIG
].found
)
1462 return -EADDRNOTAVAIL
; /* as above */
1464 /* Hmm, we found a signature partition but no Verity data? Something is off. */
1465 if (m
->partitions
[PARTITION_ROOT_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_ROOT_VERITY
].found
)
1466 return -EADDRNOTAVAIL
;
1468 if (m
->partitions
[PARTITION_USR
].found
) {
1469 /* Invalidate secondary arch /usr/ if we found the primary arch */
1470 m
->partitions
[PARTITION_USR_SECONDARY
].found
= false;
1471 m
->partitions
[PARTITION_USR_SECONDARY_VERITY
].found
= false;
1472 m
->partitions
[PARTITION_USR_SECONDARY_VERITY_SIG
].found
= false;
1474 } else if (m
->partitions
[PARTITION_USR_VERITY
].found
||
1475 m
->partitions
[PARTITION_USR_VERITY_SIG
].found
)
1476 return -EADDRNOTAVAIL
; /* as above */
1478 else if (m
->partitions
[PARTITION_USR_SECONDARY
].found
) {
1480 /* Upgrade secondary arch to primary */
1481 m
->partitions
[PARTITION_USR
] = m
->partitions
[PARTITION_USR_SECONDARY
];
1482 zero(m
->partitions
[PARTITION_USR_SECONDARY
]);
1483 m
->partitions
[PARTITION_USR_VERITY
] = m
->partitions
[PARTITION_USR_SECONDARY_VERITY
];
1484 zero(m
->partitions
[PARTITION_USR_SECONDARY_VERITY
]);
1485 m
->partitions
[PARTITION_USR_VERITY_SIG
] = m
->partitions
[PARTITION_USR_SECONDARY_VERITY_SIG
];
1486 zero(m
->partitions
[PARTITION_USR_SECONDARY_VERITY_SIG
]);
1488 } else if (m
->partitions
[PARTITION_USR_SECONDARY_VERITY
].found
||
1489 m
->partitions
[PARTITION_USR_SECONDARY_VERITY_SIG
].found
)
1490 return -EADDRNOTAVAIL
; /* as above */
1492 /* Hmm, we found a signature partition but no Verity data? Something is off. */
1493 if (m
->partitions
[PARTITION_USR_VERITY_SIG
].found
&& !m
->partitions
[PARTITION_USR_VERITY
].found
)
1494 return -EADDRNOTAVAIL
;
1496 /* If root and /usr are combined then insist that the architecture matches */
1497 if (m
->partitions
[PARTITION_ROOT
].found
&&
1498 m
->partitions
[PARTITION_USR
].found
&&
1499 (m
->partitions
[PARTITION_ROOT
].architecture
>= 0 &&
1500 m
->partitions
[PARTITION_USR
].architecture
>= 0 &&
1501 m
->partitions
[PARTITION_ROOT
].architecture
!= m
->partitions
[PARTITION_USR
].architecture
))
1502 return -EADDRNOTAVAIL
;
1504 if (!m
->partitions
[PARTITION_ROOT
].found
&&
1505 !m
->partitions
[PARTITION_USR
].found
&&
1506 (flags
& DISSECT_IMAGE_GENERIC_ROOT
) &&
1507 (!verity
|| !verity
->root_hash
|| verity
->designator
!= PARTITION_USR
)) {
1509 /* OK, we found nothing usable, then check if there's a single generic one distro, and use
1510 * that. If the root hash was set however, then we won't fall back to a generic node, because
1511 * the root hash decides. */
1513 /* If we didn't find a properly marked root partition, but we did find a single suitable
1514 * generic Linux partition, then use this as root partition, if the caller asked for it. */
1515 if (multiple_generic
)
1518 /* If we didn't find a generic node, then we can't fix this up either */
1520 _cleanup_free_
char *o
= NULL
;
1521 const char *options
;
1523 options
= mount_options_from_designator(mount_options
, PARTITION_ROOT
);
1525 o
= strdup(options
);
1530 assert(generic_nr
>= 0);
1531 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
1534 .growfs
= generic_growfs
,
1535 .partno
= generic_nr
,
1536 .architecture
= _ARCHITECTURE_INVALID
,
1537 .node
= TAKE_PTR(generic_node
),
1538 .uuid
= generic_uuid
,
1539 .mount_options
= TAKE_PTR(o
),
1540 .offset
= UINT64_MAX
,
1546 /* 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 */
1547 if (FLAGS_SET(flags
, DISSECT_IMAGE_REQUIRE_ROOT
) &&
1548 !(m
->partitions
[PARTITION_ROOT
].found
|| (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
1551 if (m
->partitions
[PARTITION_ROOT_VERITY
].found
) {
1552 /* We only support one verity partition per image, i.e. can't do for both /usr and root fs */
1553 if (m
->partitions
[PARTITION_USR_VERITY
].found
)
1556 /* We don't support verity enabled root with a split out /usr. Neither with nor without
1557 * verity there. (Note that we do support verity-less root with verity-full /usr, though.) */
1558 if (m
->partitions
[PARTITION_USR
].found
)
1559 return -EADDRNOTAVAIL
;
1563 /* If a verity designator is specified, then insist that the matching partition exists */
1564 if (verity
->designator
>= 0 && !m
->partitions
[verity
->designator
].found
)
1565 return -EADDRNOTAVAIL
;
1567 if (verity
->root_hash
) {
1568 /* If we have an explicit root hash and found the partitions for it, then we are ready to use
1569 * Verity, set things up for it */
1571 if (verity
->designator
< 0 || verity
->designator
== PARTITION_ROOT
) {
1572 if (!m
->partitions
[PARTITION_ROOT_VERITY
].found
|| !m
->partitions
[PARTITION_ROOT
].found
)
1573 return -EADDRNOTAVAIL
;
1575 /* If we found a verity setup, then the root partition is necessarily read-only. */
1576 m
->partitions
[PARTITION_ROOT
].rw
= false;
1577 m
->verity_ready
= true;
1580 assert(verity
->designator
== PARTITION_USR
);
1582 if (!m
->partitions
[PARTITION_USR_VERITY
].found
|| !m
->partitions
[PARTITION_USR
].found
)
1583 return -EADDRNOTAVAIL
;
1585 m
->partitions
[PARTITION_USR
].rw
= false;
1586 m
->verity_ready
= true;
1589 if (m
->verity_ready
)
1590 m
->verity_sig_ready
= !!verity
->root_hash_sig
;
1592 } else if (m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR_VERITY_SIG
: PARTITION_ROOT_VERITY_SIG
].found
) {
1594 /* If we found an embedded signature partition, we are ready, too. */
1596 m
->verity_ready
= m
->verity_sig_ready
= true;
1597 m
->partitions
[verity
->designator
== PARTITION_USR
? PARTITION_USR
: PARTITION_ROOT
].rw
= false;
1601 blkid_free_probe(b
);
1604 /* Fill in file system types if we don't know them yet. */
1605 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
1606 DissectedPartition
*p
= m
->partitions
+ i
;
1611 if (!p
->fstype
&& p
->node
) {
1612 r
= probe_filesystem(p
->node
, &p
->fstype
);
1613 if (r
< 0 && r
!= -EUCLEAN
)
1617 if (streq_ptr(p
->fstype
, "crypto_LUKS"))
1618 m
->encrypted
= true;
1620 if (p
->fstype
&& fstype_is_ro(p
->fstype
))
1634 DissectedImage
* dissected_image_unref(DissectedImage
*m
) {
1638 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++)
1639 dissected_partition_done(m
->partitions
+ i
);
1641 free(m
->image_name
);
1643 strv_free(m
->machine_info
);
1644 strv_free(m
->os_release
);
1645 strv_free(m
->extension_release
);
1650 static int is_loop_device(const char *path
) {
1651 char s
[SYS_BLOCK_PATH_MAX("/../loop/")];
1656 if (stat(path
, &st
) < 0)
1659 if (!S_ISBLK(st
.st_mode
))
1662 xsprintf_sys_block_path(s
, "/loop/", st
.st_dev
);
1663 if (access(s
, F_OK
) < 0) {
1664 if (errno
!= ENOENT
)
1667 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
1668 xsprintf_sys_block_path(s
, "/../loop/", st
.st_dev
);
1669 if (access(s
, F_OK
) < 0)
1670 return errno
== ENOENT
? false : -errno
;
1676 static int run_fsck(const char *node
, const char *fstype
) {
1683 r
= fsck_exists(fstype
);
1685 log_debug_errno(r
, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype
);
1689 log_debug("Not checking partition %s, as fsck for %s does not exist.", node
, fstype
);
1693 r
= safe_fork("(fsck)", FORK_RESET_SIGNALS
|FORK_CLOSE_ALL_FDS
|FORK_RLIMIT_NOFILE_SAFE
|FORK_DEATHSIG
|FORK_NULL_STDIO
, &pid
);
1695 return log_debug_errno(r
, "Failed to fork off fsck: %m");
1698 execl("/sbin/fsck", "/sbin/fsck", "-aT", node
, NULL
);
1700 log_debug_errno(errno
, "Failed to execl() fsck: %m");
1701 _exit(FSCK_OPERATIONAL_ERROR
);
1704 exit_status
= wait_for_terminate_and_check("fsck", pid
, 0);
1705 if (exit_status
< 0)
1706 return log_debug_errno(exit_status
, "Failed to fork off /sbin/fsck: %m");
1708 if ((exit_status
& ~FSCK_ERROR_CORRECTED
) != FSCK_SUCCESS
) {
1709 log_debug("fsck failed with exit status %i.", exit_status
);
1711 if ((exit_status
& (FSCK_SYSTEM_SHOULD_REBOOT
|FSCK_ERRORS_LEFT_UNCORRECTED
)) != 0)
1712 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
), "File system is corrupted, refusing.");
1714 log_debug("Ignoring fsck error.");
1720 static int fs_grow(const char *node_path
, const char *mount_path
) {
1721 _cleanup_close_
int mount_fd
= -1, node_fd
= -1;
1722 uint64_t size
, newsize
;
1725 node_fd
= open(node_path
, O_RDONLY
|O_CLOEXEC
|O_NONBLOCK
|O_NOCTTY
);
1727 return log_debug_errno(errno
, "Failed to open node device %s: %m", node_path
);
1729 if (ioctl(node_fd
, BLKGETSIZE64
, &size
) != 0)
1730 return log_debug_errno(errno
, "Failed to get block device size of %s: %m", node_path
);
1732 mount_fd
= open(mount_path
, O_RDONLY
|O_DIRECTORY
|O_CLOEXEC
);
1734 return log_debug_errno(errno
, "Failed to open mountd file system %s: %m", mount_path
);
1736 log_debug("Resizing \"%s\" to %"PRIu64
" bytes...", mount_path
, size
);
1737 r
= resize_fs(mount_fd
, size
, &newsize
);
1739 return log_debug_errno(r
, "Failed to resize \"%s\" to %"PRIu64
" bytes: %m", mount_path
, size
);
1741 if (newsize
== size
)
1742 log_debug("Successfully resized \"%s\" to %s bytes.",
1743 mount_path
, FORMAT_BYTES(newsize
));
1745 assert(newsize
< size
);
1746 log_debug("Successfully resized \"%s\" to %s bytes (%"PRIu64
" bytes lost due to blocksize).",
1747 mount_path
, FORMAT_BYTES(newsize
), size
- newsize
);
1753 static int mount_partition(
1754 DissectedPartition
*m
,
1756 const char *directory
,
1759 DissectImageFlags flags
) {
1761 _cleanup_free_
char *chased
= NULL
, *options
= NULL
;
1762 const char *p
, *node
, *fstype
;
1763 bool rw
, remap_uid_gid
= false;
1769 /* Use decrypted node and matching fstype if available, otherwise use the original device */
1770 node
= m
->decrypted_node
?: m
->node
;
1771 fstype
= m
->decrypted_node
? m
->decrypted_fstype
: m
->fstype
;
1773 if (!m
->found
|| !node
)
1776 return -EAFNOSUPPORT
;
1778 /* We are looking at an encrypted partition? This either means stacked encryption, or the caller didn't call dissected_image_decrypt() beforehand. Let's return a recognizable error for this case. */
1779 if (streq(fstype
, "crypto_LUKS"))
1782 rw
= m
->rw
&& !(flags
& DISSECT_IMAGE_MOUNT_READ_ONLY
);
1784 if (FLAGS_SET(flags
, DISSECT_IMAGE_FSCK
) && rw
) {
1785 r
= run_fsck(node
, fstype
);
1791 /* Automatically create missing mount points inside the image, if necessary. */
1792 r
= mkdir_p_root(where
, directory
, uid_shift
, (gid_t
) uid_shift
, 0755);
1793 if (r
< 0 && r
!= -EROFS
)
1796 r
= chase_symlinks(directory
, where
, CHASE_PREFIX_ROOT
, &chased
, NULL
);
1802 /* Create top-level mount if missing – but only if this is asked for. This won't modify the
1803 * image (as the branch above does) but the host hierarchy, and the created directory might
1804 * survive our mount in the host hierarchy hence. */
1805 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
1806 r
= mkdir_p(where
, 0755);
1814 /* If requested, turn on discard support. */
1815 if (fstype_can_discard(fstype
) &&
1816 ((flags
& DISSECT_IMAGE_DISCARD
) ||
1817 ((flags
& DISSECT_IMAGE_DISCARD_ON_LOOP
) && is_loop_device(m
->node
) > 0))) {
1818 options
= strdup("discard");
1823 if (uid_is_valid(uid_shift
) && uid_shift
!= 0) {
1825 if (fstype_can_uid_gid(fstype
)) {
1826 _cleanup_free_
char *uid_option
= NULL
;
1828 if (asprintf(&uid_option
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
1831 if (!strextend_with_separator(&options
, ",", uid_option
))
1833 } else if (FLAGS_SET(flags
, DISSECT_IMAGE_MOUNT_IDMAPPED
))
1834 remap_uid_gid
= true;
1837 if (!isempty(m
->mount_options
))
1838 if (!strextend_with_separator(&options
, ",", m
->mount_options
))
1841 /* So, when you request MS_RDONLY from ext4, then this means nothing. It happily still writes to the
1842 * backing storage. What's worse, the BLKRO[GS]ET flag and (in case of loopback devices)
1843 * LO_FLAGS_READ_ONLY don't mean anything, they affect userspace accesses only, and write accesses
1844 * from the upper file system still get propagated through to the underlying file system,
1845 * unrestricted. To actually get ext4/xfs/btrfs to stop writing to the device we need to specify
1846 * "norecovery" as mount option, in addition to MS_RDONLY. Yes, this sucks, since it means we need to
1847 * carry a per file system table here.
1849 * Note that this means that we might not be able to mount corrupted file systems as read-only
1850 * anymore (since in some cases the kernel implementations will refuse mounting when corrupted,
1851 * read-only and "norecovery" is specified). But I think for the case of automatically determined
1852 * mount options for loopback devices this is the right choice, since otherwise using the same
1853 * loopback file twice even in read-only mode, is going to fail badly sooner or later. The usecase of
1854 * making reuse of the immutable images "just work" is more relevant to us than having read-only
1855 * access that actually modifies stuff work on such image files. Or to say this differently: if
1856 * people want their file systems to be fixed up they should just open them in writable mode, where
1857 * all these problems don't exist. */
1858 if (!rw
&& STRPTR_IN_SET(fstype
, "ext3", "ext4", "xfs", "btrfs"))
1859 if (!strextend_with_separator(&options
, ",", "norecovery"))
1862 r
= mount_nofollow_verbose(LOG_DEBUG
, node
, p
, fstype
, MS_NODEV
|(rw
? 0 : MS_RDONLY
), options
);
1866 if (rw
&& m
->growfs
&& FLAGS_SET(flags
, DISSECT_IMAGE_GROWFS
))
1867 (void) fs_grow(node
, p
);
1869 if (remap_uid_gid
) {
1870 r
= remount_idmap(p
, uid_shift
, uid_range
);
1878 static int mount_root_tmpfs(const char *where
, uid_t uid_shift
, DissectImageFlags flags
) {
1879 _cleanup_free_
char *options
= NULL
;
1884 /* For images that contain /usr/ but no rootfs, let's mount rootfs as tmpfs */
1886 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
1887 r
= mkdir_p(where
, 0755);
1892 if (uid_is_valid(uid_shift
)) {
1893 if (asprintf(&options
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
1897 r
= mount_nofollow_verbose(LOG_DEBUG
, "rootfs", where
, "tmpfs", MS_NODEV
, options
);
1904 int dissected_image_mount(
1909 DissectImageFlags flags
) {
1911 int r
, xbootldr_mounted
;
1918 * -ENXIO → No root partition found
1919 * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release/extension-release file found
1920 * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet
1921 * -EUCLEAN → fsck for file system failed
1922 * -EBUSY → File system already mounted/used elsewhere (kernel)
1923 * -EAFNOSUPPORT → File system type not supported or not known
1926 if (!(m
->partitions
[PARTITION_ROOT
].found
||
1927 (m
->partitions
[PARTITION_USR
].found
&& FLAGS_SET(flags
, DISSECT_IMAGE_USR_NO_ROOT
))))
1928 return -ENXIO
; /* Require a root fs or at least a /usr/ fs (the latter is subject to a flag of its own) */
1930 if ((flags
& DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
) == 0) {
1932 /* First mount the root fs. If there's none we use a tmpfs. */
1933 if (m
->partitions
[PARTITION_ROOT
].found
)
1934 r
= mount_partition(m
->partitions
+ PARTITION_ROOT
, where
, NULL
, uid_shift
, uid_range
, flags
);
1936 r
= mount_root_tmpfs(where
, uid_shift
, flags
);
1940 /* For us mounting root always means mounting /usr as well */
1941 r
= mount_partition(m
->partitions
+ PARTITION_USR
, where
, "/usr", uid_shift
, uid_range
, flags
);
1945 if ((flags
& (DISSECT_IMAGE_VALIDATE_OS
|DISSECT_IMAGE_VALIDATE_OS_EXT
)) != 0) {
1946 /* If either one of the validation flags are set, ensure that the image qualifies
1947 * as one or the other (or both). */
1950 if (FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS
)) {
1951 r
= path_is_os_tree(where
);
1957 if (!ok
&& FLAGS_SET(flags
, DISSECT_IMAGE_VALIDATE_OS_EXT
)) {
1958 r
= path_is_extension_tree(where
, m
->image_name
);
1970 if (flags
& DISSECT_IMAGE_MOUNT_ROOT_ONLY
)
1973 r
= mount_partition(m
->partitions
+ PARTITION_HOME
, where
, "/home", uid_shift
, uid_range
, flags
);
1977 r
= mount_partition(m
->partitions
+ PARTITION_SRV
, where
, "/srv", uid_shift
, uid_range
, flags
);
1981 r
= mount_partition(m
->partitions
+ PARTITION_VAR
, where
, "/var", uid_shift
, uid_range
, flags
);
1985 r
= mount_partition(m
->partitions
+ PARTITION_TMP
, where
, "/var/tmp", uid_shift
, uid_range
, flags
);
1989 xbootldr_mounted
= mount_partition(m
->partitions
+ PARTITION_XBOOTLDR
, where
, "/boot", uid_shift
, uid_range
, flags
);
1990 if (xbootldr_mounted
< 0)
1991 return xbootldr_mounted
;
1993 if (m
->partitions
[PARTITION_ESP
].found
) {
1994 int esp_done
= false;
1996 /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it
1997 * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */
1999 r
= chase_symlinks("/efi", where
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
2004 /* /efi doesn't exist. Let's see if /boot is suitable then */
2006 if (!xbootldr_mounted
) {
2007 _cleanup_free_
char *p
= NULL
;
2009 r
= chase_symlinks("/boot", where
, CHASE_PREFIX_ROOT
, &p
, NULL
);
2013 } else if (dir_is_empty(p
) > 0) {
2014 /* It exists and is an empty directory. Let's mount the ESP there. */
2015 r
= mount_partition(m
->partitions
+ PARTITION_ESP
, where
, "/boot", uid_shift
, uid_range
, flags
);
2025 /* OK, let's mount the ESP now to /efi (possibly creating the dir if missing) */
2027 r
= mount_partition(m
->partitions
+ PARTITION_ESP
, where
, "/efi", uid_shift
, uid_range
, flags
);
2036 int dissected_image_mount_and_warn(
2041 DissectImageFlags flags
) {
2048 r
= dissected_image_mount(m
, where
, uid_shift
, uid_range
, flags
);
2050 return log_error_errno(r
, "Not root file system found in image.");
2051 if (r
== -EMEDIUMTYPE
)
2052 return log_error_errno(r
, "No suitable os-release/extension-release file in image found.");
2054 return log_error_errno(r
, "Encrypted file system discovered, but decryption not requested.");
2056 return log_error_errno(r
, "File system check on image failed.");
2058 return log_error_errno(r
, "File system already mounted elsewhere.");
2059 if (r
== -EAFNOSUPPORT
)
2060 return log_error_errno(r
, "File system type not supported or not known.");
2062 return log_error_errno(r
, "Failed to mount image: %m");
2067 #if HAVE_LIBCRYPTSETUP
2068 typedef struct DecryptedPartition
{
2069 struct crypt_device
*device
;
2072 } DecryptedPartition
;
2074 struct DecryptedImage
{
2075 DecryptedPartition
*decrypted
;
2080 DecryptedImage
* decrypted_image_unref(DecryptedImage
* d
) {
2081 #if HAVE_LIBCRYPTSETUP
2087 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
2088 DecryptedPartition
*p
= d
->decrypted
+ i
;
2090 if (p
->device
&& p
->name
&& !p
->relinquished
) {
2091 r
= sym_crypt_deactivate_by_name(p
->device
, p
->name
, 0);
2093 log_debug_errno(r
, "Failed to deactivate encrypted partition %s", p
->name
);
2097 sym_crypt_free(p
->device
);
2107 #if HAVE_LIBCRYPTSETUP
2109 static int make_dm_name_and_node(const void *original_node
, const char *suffix
, char **ret_name
, char **ret_node
) {
2110 _cleanup_free_
char *name
= NULL
, *node
= NULL
;
2113 assert(original_node
);
2118 base
= strrchr(original_node
, '/');
2120 base
= original_node
;
2126 name
= strjoin(base
, suffix
);
2129 if (!filename_is_valid(name
))
2132 node
= path_join(sym_crypt_get_dir(), name
);
2136 *ret_name
= TAKE_PTR(name
);
2137 *ret_node
= TAKE_PTR(node
);
2142 static int decrypt_partition(
2143 DissectedPartition
*m
,
2144 const char *passphrase
,
2145 DissectImageFlags flags
,
2146 DecryptedImage
*d
) {
2148 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
2149 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2155 if (!m
->found
|| !m
->node
|| !m
->fstype
)
2158 if (!streq(m
->fstype
, "crypto_LUKS"))
2164 r
= dlopen_cryptsetup();
2168 r
= make_dm_name_and_node(m
->node
, "-decrypted", &name
, &node
);
2172 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
2175 r
= sym_crypt_init(&cd
, m
->node
);
2177 return log_debug_errno(r
, "Failed to initialize dm-crypt: %m");
2179 cryptsetup_enable_logging(cd
);
2181 r
= sym_crypt_load(cd
, CRYPT_LUKS
, NULL
);
2183 return log_debug_errno(r
, "Failed to load LUKS metadata: %m");
2185 r
= sym_crypt_activate_by_passphrase(cd
, name
, CRYPT_ANY_SLOT
, passphrase
, strlen(passphrase
),
2186 ((flags
& DISSECT_IMAGE_DEVICE_READ_ONLY
) ? CRYPT_ACTIVATE_READONLY
: 0) |
2187 ((flags
& DISSECT_IMAGE_DISCARD_ON_CRYPTO
) ? CRYPT_ACTIVATE_ALLOW_DISCARDS
: 0));
2189 log_debug_errno(r
, "Failed to activate LUKS device: %m");
2190 return r
== -EPERM
? -EKEYREJECTED
: r
;
2193 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
2194 .name
= TAKE_PTR(name
),
2195 .device
= TAKE_PTR(cd
),
2198 m
->decrypted_node
= TAKE_PTR(node
);
2203 static int verity_can_reuse(
2204 const VeritySettings
*verity
,
2206 struct crypt_device
**ret_cd
) {
2208 /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
2209 _cleanup_free_
char *root_hash_existing
= NULL
;
2210 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2211 struct crypt_params_verity crypt_params
= {};
2212 size_t root_hash_existing_size
;
2219 r
= sym_crypt_init_by_name(&cd
, name
);
2221 return log_debug_errno(r
, "Error opening verity device, crypt_init_by_name failed: %m");
2223 cryptsetup_enable_logging(cd
);
2225 r
= sym_crypt_get_verity_info(cd
, &crypt_params
);
2227 return log_debug_errno(r
, "Error opening verity device, crypt_get_verity_info failed: %m");
2229 root_hash_existing_size
= verity
->root_hash_size
;
2230 root_hash_existing
= malloc0(root_hash_existing_size
);
2231 if (!root_hash_existing
)
2234 r
= sym_crypt_volume_key_get(cd
, CRYPT_ANY_SLOT
, root_hash_existing
, &root_hash_existing_size
, NULL
, 0);
2236 return log_debug_errno(r
, "Error opening verity device, crypt_volume_key_get failed: %m");
2237 if (verity
->root_hash_size
!= root_hash_existing_size
||
2238 memcmp(root_hash_existing
, verity
->root_hash
, verity
->root_hash_size
) != 0)
2239 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but root hashes are different.");
2241 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
2242 /* Ensure that, if signatures are supported, we only reuse the device if the previous mount used the
2243 * same settings, so that a previous unsigned mount will not be reused if the user asks to use
2244 * signing for the new one, and vice versa. */
2245 if (!!verity
->root_hash_sig
!= !!(crypt_params
.flags
& CRYPT_VERITY_ROOT_HASH_SIGNATURE
))
2246 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but signature settings are not the same.");
2249 *ret_cd
= TAKE_PTR(cd
);
2253 static inline char* dm_deferred_remove_clean(char *name
) {
2257 (void) sym_crypt_deactivate_by_name(NULL
, name
, CRYPT_DEACTIVATE_DEFERRED
);
2260 DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean
);
2262 static int verity_partition(
2263 PartitionDesignator designator
,
2264 DissectedPartition
*m
,
2265 DissectedPartition
*v
,
2266 const VeritySettings
*verity
,
2267 DissectImageFlags flags
,
2268 DecryptedImage
*d
) {
2270 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
2271 _cleanup_(dm_deferred_remove_cleanp
) char *restore_deferred_remove
= NULL
;
2272 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
2276 assert(v
|| (verity
&& verity
->data_path
));
2278 if (!verity
|| !verity
->root_hash
)
2280 if (!((verity
->designator
< 0 && designator
== PARTITION_ROOT
) ||
2281 (verity
->designator
== designator
)))
2284 if (!m
->found
|| !m
->node
|| !m
->fstype
)
2286 if (!verity
->data_path
) {
2287 if (!v
->found
|| !v
->node
|| !v
->fstype
)
2290 if (!streq(v
->fstype
, "DM_verity_hash"))
2294 r
= dlopen_cryptsetup();
2298 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
2299 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
2300 _cleanup_free_
char *root_hash_encoded
= NULL
;
2302 root_hash_encoded
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2303 if (!root_hash_encoded
)
2306 r
= make_dm_name_and_node(root_hash_encoded
, "-verity", &name
, &node
);
2308 r
= make_dm_name_and_node(m
->node
, "-verity", &name
, &node
);
2312 r
= sym_crypt_init(&cd
, verity
->data_path
?: v
->node
);
2316 cryptsetup_enable_logging(cd
);
2318 r
= sym_crypt_load(cd
, CRYPT_VERITY
, NULL
);
2322 r
= sym_crypt_set_data_device(cd
, m
->node
);
2326 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_decrypted
+ 1))
2329 /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
2330 * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
2331 * retry a few times before giving up. */
2332 for (unsigned i
= 0; i
< N_DEVICE_NODE_LIST_ATTEMPTS
; i
++) {
2333 if (verity
->root_hash_sig
) {
2334 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
2335 r
= sym_crypt_activate_by_signed_key(
2339 verity
->root_hash_size
,
2340 verity
->root_hash_sig
,
2341 verity
->root_hash_sig_size
,
2342 CRYPT_ACTIVATE_READONLY
);
2344 r
= log_debug_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
),
2345 "Activation of verity device with signature requested, but not supported by %s due to missing crypt_activate_by_signed_key().", program_invocation_short_name
);
2348 r
= sym_crypt_activate_by_volume_key(
2352 verity
->root_hash_size
,
2353 CRYPT_ACTIVATE_READONLY
);
2354 /* libdevmapper can return EINVAL when the device is already in the activation stage.
2355 * There's no way to distinguish this situation from a genuine error due to invalid
2356 * parameters, so immediately fall back to activating the device with a unique name.
2357 * Improvements in libcrypsetup can ensure this never happens:
2358 * https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
2359 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2360 return verity_partition(designator
, m
, v
, verity
, flags
& ~DISSECT_IMAGE_VERITY_SHARE
, d
);
2363 -EEXIST
, /* Volume is already open and ready to be used */
2364 -EBUSY
, /* Volume is being opened but not ready, crypt_init_by_name can fetch details */
2365 -ENODEV
/* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */))
2367 if (IN_SET(r
, -EEXIST
, -EBUSY
)) {
2368 struct crypt_device
*existing_cd
= NULL
;
2370 if (!restore_deferred_remove
){
2371 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
2372 r
= dm_deferred_remove_cancel(name
);
2373 /* If activation returns EBUSY there might be no deferred removal to cancel, that's fine */
2374 if (r
< 0 && r
!= -ENXIO
)
2375 return log_debug_errno(r
, "Disabling automated deferred removal for verity device %s failed: %m", node
);
2377 restore_deferred_remove
= strdup(name
);
2378 if (!restore_deferred_remove
)
2383 r
= verity_can_reuse(verity
, name
, &existing_cd
);
2384 /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
2385 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2386 return verity_partition(designator
, m
, v
, verity
, flags
& ~DISSECT_IMAGE_VERITY_SHARE
, d
);
2387 if (!IN_SET(r
, 0, -ENODEV
, -ENOENT
, -EBUSY
))
2388 return log_debug_errno(r
, "Checking whether existing verity device %s can be reused failed: %m", node
);
2390 /* devmapper might say that the device exists, but the devlink might not yet have been
2391 * created. Check and wait for the udev event in that case. */
2392 r
= device_wait_for_devlink(node
, "block", usec_add(now(CLOCK_MONOTONIC
), 100 * USEC_PER_MSEC
), NULL
);
2393 /* Fallback to activation with a unique device if it's taking too long */
2394 if (r
== -ETIMEDOUT
)
2407 /* Device is being opened by another process, but it has not finished yet, yield for 2ms */
2408 (void) usleep(2 * USEC_PER_MSEC
);
2411 /* An existing verity device was reported by libcryptsetup/libdevmapper, but we can't use it at this time.
2412 * Fall back to activating it with a unique device name. */
2413 if (r
!= 0 && FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
2414 return verity_partition(designator
, m
, v
, verity
, flags
& ~DISSECT_IMAGE_VERITY_SHARE
, d
);
2416 /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
2417 restore_deferred_remove
= mfree(restore_deferred_remove
);
2419 d
->decrypted
[d
->n_decrypted
++] = (DecryptedPartition
) {
2420 .name
= TAKE_PTR(name
),
2421 .device
= TAKE_PTR(cd
),
2424 m
->decrypted_node
= TAKE_PTR(node
);
2430 int dissected_image_decrypt(
2432 const char *passphrase
,
2433 const VeritySettings
*verity
,
2434 DissectImageFlags flags
,
2435 DecryptedImage
**ret
) {
2437 #if HAVE_LIBCRYPTSETUP
2438 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
2443 assert(!verity
|| verity
->root_hash
|| verity
->root_hash_size
== 0);
2447 * = 0 → There was nothing to decrypt
2448 * > 0 → Decrypted successfully
2449 * -ENOKEY → There's something to decrypt but no key was supplied
2450 * -EKEYREJECTED → Passed key was not correct
2453 if (verity
&& verity
->root_hash
&& verity
->root_hash_size
< sizeof(sd_id128_t
))
2456 if (!m
->encrypted
&& !m
->verity_ready
) {
2461 #if HAVE_LIBCRYPTSETUP
2462 d
= new0(DecryptedImage
, 1);
2466 for (PartitionDesignator i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
2467 DissectedPartition
*p
= m
->partitions
+ i
;
2468 PartitionDesignator k
;
2473 r
= decrypt_partition(p
, passphrase
, flags
, d
);
2477 k
= PARTITION_VERITY_OF(i
);
2479 r
= verity_partition(i
, p
, m
->partitions
+ k
, verity
, flags
| DISSECT_IMAGE_VERITY_SHARE
, d
);
2484 if (!p
->decrypted_fstype
&& p
->decrypted_node
) {
2485 r
= probe_filesystem(p
->decrypted_node
, &p
->decrypted_fstype
);
2486 if (r
< 0 && r
!= -EUCLEAN
)
2499 int dissected_image_decrypt_interactively(
2501 const char *passphrase
,
2502 const VeritySettings
*verity
,
2503 DissectImageFlags flags
,
2504 DecryptedImage
**ret
) {
2506 _cleanup_strv_free_erase_
char **z
= NULL
;
2513 r
= dissected_image_decrypt(m
, passphrase
, verity
, flags
, ret
);
2516 if (r
== -EKEYREJECTED
)
2517 log_error_errno(r
, "Incorrect passphrase, try again!");
2518 else if (r
!= -ENOKEY
)
2519 return log_error_errno(r
, "Failed to decrypt image: %m");
2522 return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED
),
2523 "Too many retries.");
2527 r
= ask_password_auto("Please enter image passphrase:", NULL
, "dissect", "dissect", "dissect.passphrase", USEC_INFINITY
, 0, &z
);
2529 return log_error_errno(r
, "Failed to query for passphrase: %m");
2535 int decrypted_image_relinquish(DecryptedImage
*d
) {
2538 /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a
2539 * boolean so that we don't clean it up ourselves either anymore */
2541 #if HAVE_LIBCRYPTSETUP
2544 for (size_t i
= 0; i
< d
->n_decrypted
; i
++) {
2545 DecryptedPartition
*p
= d
->decrypted
+ i
;
2547 if (p
->relinquished
)
2550 r
= sym_crypt_deactivate_by_name(NULL
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
2552 return log_debug_errno(r
, "Failed to mark %s for auto-removal: %m", p
->name
);
2554 p
->relinquished
= true;
2561 static char *build_auxiliary_path(const char *image
, const char *suffix
) {
2568 e
= endswith(image
, ".raw");
2570 return strjoin(e
, suffix
);
2572 n
= new(char, e
- image
+ strlen(suffix
) + 1);
2576 strcpy(mempcpy(n
, image
, e
- image
), suffix
);
2580 void verity_settings_done(VeritySettings
*v
) {
2583 v
->root_hash
= mfree(v
->root_hash
);
2584 v
->root_hash_size
= 0;
2586 v
->root_hash_sig
= mfree(v
->root_hash_sig
);
2587 v
->root_hash_sig_size
= 0;
2589 v
->data_path
= mfree(v
->data_path
);
2592 int verity_settings_load(
2593 VeritySettings
*verity
,
2595 const char *root_hash_path
,
2596 const char *root_hash_sig_path
) {
2598 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
2599 size_t root_hash_size
= 0, root_hash_sig_size
= 0;
2600 _cleanup_free_
char *verity_data_path
= NULL
;
2601 PartitionDesignator designator
;
2606 assert(verity
->designator
< 0 || IN_SET(verity
->designator
, PARTITION_ROOT
, PARTITION_USR
));
2608 /* If we are asked to load the root hash for a device node, exit early */
2609 if (is_device_path(image
))
2612 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_SIDECAR");
2613 if (r
< 0 && r
!= -ENXIO
)
2614 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_SIDECAR, ignoring: %m");
2618 designator
= verity
->designator
;
2620 /* We only fill in what isn't already filled in */
2622 if (!verity
->root_hash
) {
2623 _cleanup_free_
char *text
= NULL
;
2625 if (root_hash_path
) {
2626 /* If explicitly specified it takes precedence */
2627 r
= read_one_line_file(root_hash_path
, &text
);
2632 designator
= PARTITION_ROOT
;
2634 /* Otherwise look for xattr and separate file, and first for the data for root and if
2635 * that doesn't exist for /usr */
2637 if (designator
< 0 || designator
== PARTITION_ROOT
) {
2638 r
= getxattr_malloc(image
, "user.verity.roothash", &text
, true);
2640 _cleanup_free_
char *p
= NULL
;
2642 if (!IN_SET(r
, -ENODATA
, -ENOENT
) && !ERRNO_IS_NOT_SUPPORTED(r
))
2645 p
= build_auxiliary_path(image
, ".roothash");
2649 r
= read_one_line_file(p
, &text
);
2650 if (r
< 0 && r
!= -ENOENT
)
2655 designator
= PARTITION_ROOT
;
2658 if (!text
&& (designator
< 0 || designator
== PARTITION_USR
)) {
2659 /* So in the "roothash" xattr/file name above the "root" of course primarily
2660 * refers to the root of the Verity Merkle tree. But coincidentally it also
2661 * is the hash for the *root* file system, i.e. the "root" neatly refers to
2662 * two distinct concepts called "root". Taking benefit of this happy
2663 * coincidence we call the file with the root hash for the /usr/ file system
2664 * `usrhash`, because `usrroothash` or `rootusrhash` would just be too
2665 * confusing. We thus drop the reference to the root of the Merkle tree, and
2666 * just indicate which file system it's about. */
2667 r
= getxattr_malloc(image
, "user.verity.usrhash", &text
, true);
2669 _cleanup_free_
char *p
= NULL
;
2671 if (!IN_SET(r
, -ENODATA
, -ENOENT
) && !ERRNO_IS_NOT_SUPPORTED(r
))
2674 p
= build_auxiliary_path(image
, ".usrhash");
2678 r
= read_one_line_file(p
, &text
);
2679 if (r
< 0 && r
!= -ENOENT
)
2684 designator
= PARTITION_USR
;
2689 r
= unhexmem(text
, strlen(text
), &root_hash
, &root_hash_size
);
2692 if (root_hash_size
< sizeof(sd_id128_t
))
2697 if ((root_hash
|| verity
->root_hash
) && !verity
->root_hash_sig
) {
2698 if (root_hash_sig_path
) {
2699 r
= read_full_file(root_hash_sig_path
, (char**) &root_hash_sig
, &root_hash_sig_size
);
2700 if (r
< 0 && r
!= -ENOENT
)
2704 designator
= PARTITION_ROOT
;
2706 if (designator
< 0 || designator
== PARTITION_ROOT
) {
2707 _cleanup_free_
char *p
= NULL
;
2709 /* Follow naming convention recommended by the relevant RFC:
2710 * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
2711 p
= build_auxiliary_path(image
, ".roothash.p7s");
2715 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
2716 if (r
< 0 && r
!= -ENOENT
)
2719 designator
= PARTITION_ROOT
;
2722 if (!root_hash_sig
&& (designator
< 0 || designator
== PARTITION_USR
)) {
2723 _cleanup_free_
char *p
= NULL
;
2725 p
= build_auxiliary_path(image
, ".usrhash.p7s");
2729 r
= read_full_file(p
, (char**) &root_hash_sig
, &root_hash_sig_size
);
2730 if (r
< 0 && r
!= -ENOENT
)
2733 designator
= PARTITION_USR
;
2737 if (root_hash_sig
&& root_hash_sig_size
== 0) /* refuse empty size signatures */
2741 if (!verity
->data_path
) {
2742 _cleanup_free_
char *p
= NULL
;
2744 p
= build_auxiliary_path(image
, ".verity");
2748 if (access(p
, F_OK
) < 0) {
2749 if (errno
!= ENOENT
)
2752 verity_data_path
= TAKE_PTR(p
);
2756 verity
->root_hash
= TAKE_PTR(root_hash
);
2757 verity
->root_hash_size
= root_hash_size
;
2760 if (root_hash_sig
) {
2761 verity
->root_hash_sig
= TAKE_PTR(root_hash_sig
);
2762 verity
->root_hash_sig_size
= root_hash_sig_size
;
2765 if (verity_data_path
)
2766 verity
->data_path
= TAKE_PTR(verity_data_path
);
2768 if (verity
->designator
< 0)
2769 verity
->designator
= designator
;
2774 int dissected_image_load_verity_sig_partition(
2777 VeritySettings
*verity
) {
2779 _cleanup_free_
void *root_hash
= NULL
, *root_hash_sig
= NULL
;
2780 _cleanup_(json_variant_unrefp
) JsonVariant
*v
= NULL
;
2781 size_t root_hash_size
, root_hash_sig_size
;
2782 _cleanup_free_
char *buf
= NULL
;
2783 PartitionDesignator d
;
2784 DissectedPartition
*p
;
2785 JsonVariant
*rh
, *sig
;
2794 if (verity
->root_hash
&& verity
->root_hash_sig
) /* Already loaded? */
2797 r
= getenv_bool_secure("SYSTEMD_DISSECT_VERITY_EMBEDDED");
2798 if (r
< 0 && r
!= -ENXIO
)
2799 log_debug_errno(r
, "Failed to parse $SYSTEMD_DISSECT_VERITY_EMBEDDED, ignoring: %m");
2803 d
= PARTITION_VERITY_SIG_OF(verity
->designator
< 0 ? PARTITION_ROOT
: verity
->designator
);
2806 p
= m
->partitions
+ d
;
2809 if (p
->offset
== UINT64_MAX
|| p
->size
== UINT64_MAX
)
2812 if (p
->size
> 4*1024*1024) /* Signature data cannot possible be larger than 4M, refuse that */
2815 buf
= new(char, p
->size
+1);
2819 n
= pread(fd
, buf
, p
->size
, p
->offset
);
2822 if ((uint64_t) n
!= p
->size
)
2825 e
= memchr(buf
, 0, p
->size
);
2827 /* If we found a NUL byte then the rest of the data must be NUL too */
2828 if (!memeqzero(e
, p
->size
- (e
- buf
)))
2829 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature data contains embedded NUL byte.");
2833 r
= json_parse(buf
, 0, &v
, NULL
, NULL
);
2835 return log_debug_errno(r
, "Failed to parse signature JSON data: %m");
2837 rh
= json_variant_by_key(v
, "rootHash");
2839 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'rootHash' field.");
2840 if (!json_variant_is_string(rh
))
2841 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'rootHash' field of signature JSON object is not a string.");
2843 r
= unhexmem(json_variant_string(rh
), SIZE_MAX
, &root_hash
, &root_hash_size
);
2845 return log_debug_errno(r
, "Failed to parse root hash field: %m");
2847 /* Check if specified root hash matches if it is specified */
2848 if (verity
->root_hash
&&
2849 memcmp_nn(verity
->root_hash
, verity
->root_hash_size
, root_hash
, root_hash_size
) != 0) {
2850 _cleanup_free_
char *a
= NULL
, *b
= NULL
;
2852 a
= hexmem(root_hash
, root_hash_size
);
2853 b
= hexmem(verity
->root_hash
, verity
->root_hash_size
);
2855 return log_debug_errno(r
, "Root hash in signature JSON data (%s) doesn't match configured hash (%s).", strna(a
), strna(b
));
2858 sig
= json_variant_by_key(v
, "signature");
2860 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Signature JSON object lacks 'signature' field.");
2861 if (!json_variant_is_string(sig
))
2862 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'signature' field of signature JSON object is not a string.");
2864 r
= unbase64mem(json_variant_string(sig
), SIZE_MAX
, &root_hash_sig
, &root_hash_sig_size
);
2866 return log_debug_errno(r
, "Failed to parse signature field: %m");
2868 free_and_replace(verity
->root_hash
, root_hash
);
2869 verity
->root_hash_size
= root_hash_size
;
2871 free_and_replace(verity
->root_hash_sig
, root_hash_sig
);
2872 verity
->root_hash_sig_size
= root_hash_sig_size
;
2877 int dissected_image_acquire_metadata(DissectedImage
*m
) {
2884 META_EXTENSION_RELEASE
,
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_EXTENSION_RELEASE
] = "extension-release\0", /* Used only for logging. */
2897 _cleanup_strv_free_
char **machine_info
= NULL
, **os_release
= NULL
, **extension_release
= NULL
;
2898 _cleanup_close_pair_
int error_pipe
[2] = { -1, -1 };
2899 _cleanup_(rmdir_and_freep
) char *t
= NULL
;
2900 _cleanup_(sigkill_waitp
) pid_t child
= 0;
2901 sd_id128_t machine_id
= SD_ID128_NULL
;
2902 _cleanup_free_
char *hostname
= NULL
;
2903 unsigned n_meta_initialized
= 0;
2904 int fds
[2 * _META_MAX
], r
, v
;
2907 BLOCK_SIGNALS(SIGCHLD
);
2911 for (; n_meta_initialized
< _META_MAX
; n_meta_initialized
++) {
2912 if (!paths
[n_meta_initialized
]) {
2913 fds
[2*n_meta_initialized
] = fds
[2*n_meta_initialized
+1] = -1;
2917 if (pipe2(fds
+ 2*n_meta_initialized
, O_CLOEXEC
) < 0) {
2923 r
= mkdtemp_malloc("/tmp/dissect-XXXXXX", &t
);
2927 if (pipe2(error_pipe
, O_CLOEXEC
) < 0) {
2932 r
= safe_fork("(sd-dissect)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_NEW_MOUNTNS
|FORK_MOUNTNS_SLAVE
, &child
);
2936 error_pipe
[0] = safe_close(error_pipe
[0]);
2938 r
= dissected_image_mount(
2943 DISSECT_IMAGE_READ_ONLY
|
2944 DISSECT_IMAGE_MOUNT_ROOT_ONLY
|
2945 DISSECT_IMAGE_VALIDATE_OS
|
2946 DISSECT_IMAGE_VALIDATE_OS_EXT
|
2947 DISSECT_IMAGE_USR_NO_ROOT
);
2949 /* Let parent know the error */
2950 (void) write(error_pipe
[1], &r
, sizeof(r
));
2952 log_debug_errno(r
, "Failed to mount dissected image: %m");
2953 _exit(EXIT_FAILURE
);
2956 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
2957 _cleanup_close_
int fd
= -ENOENT
;
2963 fds
[2*k
] = safe_close(fds
[2*k
]);
2965 if (k
== META_EXTENSION_RELEASE
) {
2966 /* As per the os-release spec, if the image is an extension it will have a file
2967 * named after the image name in extension-release.d/ - we use the image name
2968 * and try to resolve it with the extension-release helpers, as sometimes
2969 * the image names are mangled on deployment and do not match anymore.
2970 * Unlike other paths this is not fixed, and the image name
2971 * can be mangled on deployment, so by calling into the helper
2972 * we allow a fallback that matches on the first extension-release
2973 * file found in the directory, if one named after the image cannot
2974 * be found first. */
2975 r
= open_extension_release(t
, m
->image_name
, NULL
, &fd
);
2977 fd
= r
; /* Propagate the error. */
2979 NULSTR_FOREACH(p
, paths
[k
]) {
2980 fd
= chase_symlinks_and_open(p
, t
, CHASE_PREFIX_ROOT
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
, NULL
);
2985 log_debug_errno(fd
, "Failed to read %s file of image, ignoring: %m", paths
[k
]);
2986 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
2990 r
= copy_bytes(fd
, fds
[2*k
+1], UINT64_MAX
, 0);
2992 (void) write(error_pipe
[1], &r
, sizeof(r
));
2993 _exit(EXIT_FAILURE
);
2996 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
2999 _exit(EXIT_SUCCESS
);
3002 error_pipe
[1] = safe_close(error_pipe
[1]);
3004 for (unsigned k
= 0; k
< _META_MAX
; k
++) {
3005 _cleanup_fclose_
FILE *f
= NULL
;
3010 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
3012 f
= take_fdopen(&fds
[2*k
], "r");
3021 r
= read_etc_hostname_stream(f
, &hostname
);
3023 log_debug_errno(r
, "Failed to read /etc/hostname: %m");
3027 case META_MACHINE_ID
: {
3028 _cleanup_free_
char *line
= NULL
;
3030 r
= read_line(f
, LONG_LINE_MAX
, &line
);
3032 log_debug_errno(r
, "Failed to read /etc/machine-id: %m");
3034 r
= sd_id128_from_string(line
, &machine_id
);
3036 log_debug_errno(r
, "Image contains invalid /etc/machine-id: %s", line
);
3038 log_debug("/etc/machine-id file is empty.");
3039 else if (streq(line
, "uninitialized"))
3040 log_debug("/etc/machine-id file is uninitialized (likely aborted first boot).");
3042 log_debug("/etc/machine-id has unexpected length %i.", r
);
3047 case META_MACHINE_INFO
:
3048 r
= load_env_file_pairs(f
, "machine-info", &machine_info
);
3050 log_debug_errno(r
, "Failed to read /etc/machine-info: %m");
3054 case META_OS_RELEASE
:
3055 r
= load_env_file_pairs(f
, "os-release", &os_release
);
3057 log_debug_errno(r
, "Failed to read OS release file: %m");
3061 case META_EXTENSION_RELEASE
:
3062 r
= load_env_file_pairs(f
, "extension-release", &extension_release
);
3064 log_debug_errno(r
, "Failed to read extension release file: %m");
3070 r
= wait_for_terminate_and_check("(sd-dissect)", child
, 0);
3075 n
= read(error_pipe
[0], &v
, sizeof(v
));
3079 return v
; /* propagate error sent to us from child */
3083 if (r
!= EXIT_SUCCESS
)
3086 free_and_replace(m
->hostname
, hostname
);
3087 m
->machine_id
= machine_id
;
3088 strv_free_and_replace(m
->machine_info
, machine_info
);
3089 strv_free_and_replace(m
->os_release
, os_release
);
3090 strv_free_and_replace(m
->extension_release
, extension_release
);
3093 for (unsigned k
= 0; k
< n_meta_initialized
; k
++)
3094 safe_close_pair(fds
+ 2*k
);
3099 int dissect_image_and_warn(
3102 const VeritySettings
*verity
,
3103 const MountOptions
*mount_options
,
3105 uint64_t uevent_seqnum_not_before
,
3106 usec_t timestamp_not_before
,
3107 DissectImageFlags flags
,
3108 DissectedImage
**ret
) {
3110 _cleanup_free_
char *buffer
= NULL
;
3114 r
= fd_get_path(fd
, &buffer
);
3121 r
= dissect_image(fd
, verity
, mount_options
, diskseq
, uevent_seqnum_not_before
, timestamp_not_before
, flags
, ret
);
3125 return log_error_errno(r
, "Dissecting images is not supported, compiled without blkid support.");
3128 return log_error_errno(r
, "Couldn't identify a suitable partition table or file system in '%s'.", name
);
3130 case -EADDRNOTAVAIL
:
3131 return log_error_errno(r
, "No root partition for specified root hash found in '%s'.", name
);
3134 return log_error_errno(r
, "Multiple suitable root partitions found in image '%s'.", name
);
3137 return log_error_errno(r
, "No suitable root partition found in image '%s'.", name
);
3139 case -EPROTONOSUPPORT
:
3140 return log_error_errno(r
, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name
);
3144 return log_error_errno(r
, "Failed to dissect image '%s': %m", name
);
3150 bool dissected_image_verity_candidate(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3153 /* Checks if this partition could theoretically do Verity. For non-partitioned images this only works
3154 * if there's an external verity file supplied, for which we can consult .has_verity. For partitioned
3155 * images we only check the partition type.
3157 * This call is used to decide whether to suppress or show a verity column in tabular output of the
3160 if (image
->single_file_system
)
3161 return partition_designator
== PARTITION_ROOT
&& image
->has_verity
;
3163 return PARTITION_VERITY_OF(partition_designator
) >= 0;
3166 bool dissected_image_verity_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3167 PartitionDesignator k
;
3171 /* Checks if this partition has verity data available that we can activate. For non-partitioned this
3172 * works for the root partition, for others only if the associated verity partition was found. */
3174 if (!image
->verity_ready
)
3177 if (image
->single_file_system
)
3178 return partition_designator
== PARTITION_ROOT
;
3180 k
= PARTITION_VERITY_OF(partition_designator
);
3181 return k
>= 0 && image
->partitions
[k
].found
;
3184 bool dissected_image_verity_sig_ready(const DissectedImage
*image
, PartitionDesignator partition_designator
) {
3185 PartitionDesignator k
;
3189 /* Checks if this partition has verity signature data available that we can use. */
3191 if (!image
->verity_sig_ready
)
3194 if (image
->single_file_system
)
3195 return partition_designator
== PARTITION_ROOT
;
3197 k
= PARTITION_VERITY_SIG_OF(partition_designator
);
3198 return k
>= 0 && image
->partitions
[k
].found
;
3201 MountOptions
* mount_options_free_all(MountOptions
*options
) {
3204 while ((m
= options
)) {
3205 LIST_REMOVE(mount_options
, options
, m
);
3213 const char* mount_options_from_designator(const MountOptions
*options
, PartitionDesignator designator
) {
3214 const MountOptions
*m
;
3216 LIST_FOREACH(mount_options
, m
, options
)
3217 if (designator
== m
->partition_designator
&& !isempty(m
->options
))
3223 int mount_image_privately_interactively(
3225 DissectImageFlags flags
,
3226 char **ret_directory
,
3227 LoopDevice
**ret_loop_device
,
3228 DecryptedImage
**ret_decrypted_image
) {
3230 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3231 _cleanup_(loop_device_unrefp
) LoopDevice
*d
= NULL
;
3232 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
3233 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3234 _cleanup_(rmdir_and_freep
) char *created_dir
= NULL
;
3235 _cleanup_free_
char *temp
= NULL
;
3238 /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
3239 * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
3243 assert(ret_directory
);
3244 assert(ret_loop_device
);
3245 assert(ret_decrypted_image
);
3247 r
= verity_settings_load(&verity
, image
, NULL
, NULL
);
3249 return log_error_errno(r
, "Failed to load root hash data: %m");
3251 r
= tempfn_random_child(NULL
, program_invocation_short_name
, &temp
);
3253 return log_error_errno(r
, "Failed to generate temporary mount directory: %m");
3255 r
= loop_device_make_by_path(
3257 FLAGS_SET(flags
, DISSECT_IMAGE_DEVICE_READ_ONLY
) ? O_RDONLY
: O_RDWR
,
3258 FLAGS_SET(flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
3261 return log_error_errno(r
, "Failed to set up loopback device for %s: %m", image
);
3263 r
= dissect_image_and_warn(d
->fd
, image
, &verity
, NULL
, d
->diskseq
, d
->uevent_seqnum_not_before
, d
->timestamp_not_before
, flags
, &dissected_image
);
3267 r
= dissected_image_load_verity_sig_partition(dissected_image
, d
->fd
, &verity
);
3271 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, &verity
, flags
, &decrypted_image
);
3275 r
= detach_mount_namespace();
3277 return log_error_errno(r
, "Failed to detach mount namespace: %m");
3279 r
= mkdir_p(temp
, 0700);
3281 return log_error_errno(r
, "Failed to create mount point: %m");
3283 created_dir
= TAKE_PTR(temp
);
3285 r
= dissected_image_mount_and_warn(dissected_image
, created_dir
, UID_INVALID
, UID_INVALID
, flags
);
3289 if (decrypted_image
) {
3290 r
= decrypted_image_relinquish(decrypted_image
);
3292 return log_error_errno(r
, "Failed to relinquish DM devices: %m");
3295 loop_device_relinquish(d
);
3297 *ret_directory
= TAKE_PTR(created_dir
);
3298 *ret_loop_device
= TAKE_PTR(d
);
3299 *ret_decrypted_image
= TAKE_PTR(decrypted_image
);
3304 static const char *const partition_designator_table
[] = {
3305 [PARTITION_ROOT
] = "root",
3306 [PARTITION_ROOT_SECONDARY
] = "root-secondary",
3307 [PARTITION_USR
] = "usr",
3308 [PARTITION_USR_SECONDARY
] = "usr-secondary",
3309 [PARTITION_HOME
] = "home",
3310 [PARTITION_SRV
] = "srv",
3311 [PARTITION_ESP
] = "esp",
3312 [PARTITION_XBOOTLDR
] = "xbootldr",
3313 [PARTITION_SWAP
] = "swap",
3314 [PARTITION_ROOT_VERITY
] = "root-verity",
3315 [PARTITION_ROOT_SECONDARY_VERITY
] = "root-secondary-verity",
3316 [PARTITION_USR_VERITY
] = "usr-verity",
3317 [PARTITION_USR_SECONDARY_VERITY
] = "usr-secondary-verity",
3318 [PARTITION_ROOT_VERITY_SIG
] = "root-verity-sig",
3319 [PARTITION_ROOT_SECONDARY_VERITY_SIG
] = "root-secondary-verity-sig",
3320 [PARTITION_USR_VERITY_SIG
] = "usr-verity-sig",
3321 [PARTITION_USR_SECONDARY_VERITY_SIG
] = "usr-secondary-verity-sig",
3322 [PARTITION_TMP
] = "tmp",
3323 [PARTITION_VAR
] = "var",
3326 int verity_dissect_and_mount(
3329 const MountOptions
*options
,
3330 const char *required_host_os_release_id
,
3331 const char *required_host_os_release_version_id
,
3332 const char *required_host_os_release_sysext_level
) {
3334 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
3335 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
3336 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
3337 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
3338 DissectImageFlags dissect_image_flags
;
3344 r
= verity_settings_load(&verity
, src
, NULL
, NULL
);
3346 return log_debug_errno(r
, "Failed to load root hash: %m");
3348 dissect_image_flags
= verity
.data_path
? DISSECT_IMAGE_NO_PARTITION_TABLE
: 0;
3350 r
= loop_device_make_by_path(
3353 verity
.data_path
? 0 : LO_FLAGS_PARTSCAN
,
3356 return log_debug_errno(r
, "Failed to create loop device for image: %m");
3362 loop_device
->diskseq
,
3363 loop_device
->uevent_seqnum_not_before
,
3364 loop_device
->timestamp_not_before
,
3365 dissect_image_flags
,
3367 /* No partition table? Might be a single-filesystem image, try again */
3368 if (!verity
.data_path
&& r
== -ENOPKG
)
3373 loop_device
->diskseq
,
3374 loop_device
->uevent_seqnum_not_before
,
3375 loop_device
->timestamp_not_before
,
3376 dissect_image_flags
| DISSECT_IMAGE_NO_PARTITION_TABLE
,
3379 return log_debug_errno(r
, "Failed to dissect image: %m");
3381 r
= dissected_image_load_verity_sig_partition(dissected_image
, loop_device
->fd
, &verity
);
3385 r
= dissected_image_decrypt(
3389 dissect_image_flags
,
3392 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
3394 r
= mkdir_p_label(dest
, 0755);
3396 return log_debug_errno(r
, "Failed to create destination directory %s: %m", dest
);
3397 r
= umount_recursive(dest
, 0);
3399 return log_debug_errno(r
, "Failed to umount under destination directory %s: %m", dest
);
3401 r
= dissected_image_mount(dissected_image
, dest
, UID_INVALID
, UID_INVALID
, dissect_image_flags
);
3403 return log_debug_errno(r
, "Failed to mount image: %m");
3405 /* If we got os-release values from the caller, then we need to match them with the image's
3406 * extension-release.d/ content. Return -EINVAL if there's any mismatch.
3407 * First, check the distro ID. If that matches, then check the new SYSEXT_LEVEL value if
3408 * available, or else fallback to VERSION_ID. */
3409 if (required_host_os_release_id
&&
3410 (required_host_os_release_version_id
|| required_host_os_release_sysext_level
)) {
3411 _cleanup_strv_free_
char **extension_release
= NULL
;
3413 r
= load_extension_release_pairs(dest
, dissected_image
->image_name
, &extension_release
);
3415 return log_debug_errno(r
, "Failed to parse image %s extension-release metadata: %m", dissected_image
->image_name
);
3417 r
= extension_release_validate(
3418 dissected_image
->image_name
,
3419 required_host_os_release_id
,
3420 required_host_os_release_version_id
,
3421 required_host_os_release_sysext_level
,
3424 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE
), "Image %s extension-release metadata does not match the root's", dissected_image
->image_name
);
3426 return log_debug_errno(r
, "Failed to compare image %s extension-release metadata with the root's os-release: %m", dissected_image
->image_name
);
3429 if (decrypted_image
) {
3430 r
= decrypted_image_relinquish(decrypted_image
);
3432 return log_debug_errno(r
, "Failed to relinquish decrypted image: %m");
3435 loop_device_relinquish(loop_device
);
3440 DEFINE_STRING_TABLE_LOOKUP(partition_designator
, PartitionDesignator
);