1 /* SPDX-License-Identifier: LGPL-2.1+ */
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 "crypt-util.h"
24 #include "device-nodes.h"
25 #include "device-util.h"
26 #include "dissect-image.h"
32 #include "fsck-util.h"
34 #include "hexdecoct.h"
35 #include "hostname-util.h"
36 #include "id128-util.h"
38 #include "mount-util.h"
39 #include "mountpoint-util.h"
40 #include "namespace-util.h"
41 #include "nulstr-util.h"
43 #include "path-util.h"
44 #include "process-util.h"
45 #include "raw-clone.h"
46 #include "signal-util.h"
47 #include "stat-util.h"
48 #include "stdio-util.h"
49 #include "string-table.h"
50 #include "string-util.h"
52 #include "tmpfile-util.h"
53 #include "udev-util.h"
54 #include "user-util.h"
55 #include "xattr-util.h"
57 int probe_filesystem(const char *node
, char **ret_fstype
) {
58 /* Try to find device content type and return it in *ret_fstype. If nothing is found,
59 * 0/NULL will be returned. -EUCLEAN will be returned for ambiguous results, and an
60 * different error otherwise. */
63 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
68 b
= blkid_new_probe_from_filename(node
);
70 return errno_or_else(ENOMEM
);
72 blkid_probe_enable_superblocks(b
, 1);
73 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
);
76 r
= blkid_do_safeprobe(b
);
78 log_debug("No type detected on partition %s", node
);
82 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
),
83 "Results ambiguous for partition %s", node
);
85 return errno_or_else(EIO
);
87 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
109 /* Detect RPMB and Boot partitions, which are not listed by blkid.
110 * See https://github.com/systemd/systemd/issues/5806. */
111 static bool device_is_mmc_special_partition(sd_device
*d
) {
116 if (sd_device_get_sysname(d
, &sysname
) < 0)
119 return startswith(sysname
, "mmcblk") &&
120 (endswith(sysname
, "rpmb") || endswith(sysname
, "boot0") || endswith(sysname
, "boot1"));
123 static bool device_is_block(sd_device
*d
) {
128 if (sd_device_get_subsystem(d
, &ss
) < 0)
131 return streq(ss
, "block");
134 static int enumerator_for_parent(sd_device
*d
, sd_device_enumerator
**ret
) {
135 _cleanup_(sd_device_enumerator_unrefp
) sd_device_enumerator
*e
= NULL
;
141 r
= sd_device_enumerator_new(&e
);
145 r
= sd_device_enumerator_allow_uninitialized(e
);
149 r
= sd_device_enumerator_add_match_parent(e
, d
);
157 /* how many times to wait for the device nodes to appear */
158 #define N_DEVICE_NODE_LIST_ATTEMPTS 10
160 static int wait_for_partitions_to_appear(
163 unsigned num_partitions
,
164 DissectImageFlags flags
,
165 sd_device_enumerator
**ret_enumerator
) {
167 _cleanup_(sd_device_enumerator_unrefp
) sd_device_enumerator
*e
= NULL
;
174 assert(ret_enumerator
);
176 r
= enumerator_for_parent(d
, &e
);
180 /* Count the partitions enumerated by the kernel */
182 FOREACH_DEVICE(e
, q
) {
183 if (sd_device_get_devnum(q
, NULL
) < 0)
185 if (!device_is_block(q
))
187 if (device_is_mmc_special_partition(q
))
190 if (!FLAGS_SET(flags
, DISSECT_IMAGE_NO_UDEV
)) {
191 r
= device_wait_for_initialization(q
, "block", USEC_INFINITY
, NULL
);
199 if (n
== num_partitions
+ 1) {
200 *ret_enumerator
= TAKE_PTR(e
);
201 return 0; /* success! */
203 if (n
> num_partitions
+ 1)
204 return log_debug_errno(SYNTHETIC_ERRNO(EIO
),
205 "blkid and kernel partition lists do not match.");
207 /* The kernel has probed fewer partitions than blkid? Maybe the kernel prober is still running or it
208 * got EBUSY because udev already opened the device. Let's reprobe the device, which is a synchronous
209 * call that waits until probing is complete. */
211 for (unsigned j
= 0; ; j
++) {
215 if (ioctl(fd
, BLKRRPART
, 0) >= 0)
219 /* If we are running on a block device that has partition scanning off, return an
220 * explicit recognizable error about this, so that callers can generate a proper
221 * message explaining the situation. */
223 r
= blockdev_partscan_enabled(fd
);
227 return log_debug_errno(EPROTONOSUPPORT
,
228 "Device is a loop device and partition scanning is off!");
230 return -EINVAL
; /* original error */
235 /* If something else has the device open, such as an udev rule, the ioctl will return
236 * EBUSY. Since there's no way to wait until it isn't busy anymore, let's just wait a bit,
239 * This is really something they should fix in the kernel! */
240 (void) usleep(50 * USEC_PER_MSEC
);
244 return -EAGAIN
; /* no success yet, try again */
247 static int loop_wait_for_partitions_to_appear(
250 unsigned num_partitions
,
251 DissectImageFlags flags
,
252 sd_device_enumerator
**ret_enumerator
) {
253 _cleanup_(sd_device_unrefp
) sd_device
*device
= NULL
;
258 assert(ret_enumerator
);
260 log_debug("Waiting for device (parent + %d partitions) to appear...", num_partitions
);
262 if (!FLAGS_SET(flags
, DISSECT_IMAGE_NO_UDEV
)) {
263 r
= device_wait_for_initialization(d
, "block", USEC_INFINITY
, &device
);
267 device
= sd_device_ref(d
);
269 for (unsigned i
= 0; i
< N_DEVICE_NODE_LIST_ATTEMPTS
; i
++) {
270 r
= wait_for_partitions_to_appear(fd
, device
, num_partitions
, flags
, ret_enumerator
);
275 return log_debug_errno(SYNTHETIC_ERRNO(ENXIO
),
276 "Kernel partitions dit not appear within %d attempts",
277 N_DEVICE_NODE_LIST_ATTEMPTS
);
280 static void check_partition_flags(
282 unsigned long long pflags
,
283 unsigned long long supported
) {
287 /* Mask away all flags supported by this partition's type and the three flags the UEFI spec defines generically */
288 pflags
&= ~(supported
| GPT_FLAG_REQUIRED_PARTITION
| GPT_FLAG_NO_BLOCK_IO_PROTOCOL
| GPT_FLAG_LEGACY_BIOS_BOOTABLE
);
293 /* If there are other bits set, then log about it, to make things discoverable */
294 for (unsigned i
= 0; i
< sizeof(pflags
) * 8; i
++) {
295 unsigned long long bit
= 1ULL << i
;
296 if (!FLAGS_SET(pflags
, bit
))
299 log_debug("Unexpected partition flag %llu set on %s!", bit
, node
);
307 const void *root_hash
,
308 size_t root_hash_size
,
309 const char *verity_data
,
310 const MountOptions
*mount_options
,
311 DissectImageFlags flags
,
312 DissectedImage
**ret
) {
315 sd_id128_t root_uuid
= SD_ID128_NULL
, verity_uuid
= SD_ID128_NULL
;
316 _cleanup_(sd_device_enumerator_unrefp
) sd_device_enumerator
*e
= NULL
;
317 bool is_gpt
, is_mbr
, generic_rw
, multiple_generic
= false;
318 _cleanup_(sd_device_unrefp
) sd_device
*d
= NULL
;
319 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
320 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
321 _cleanup_free_
char *generic_node
= NULL
;
322 sd_id128_t generic_uuid
= SD_ID128_NULL
;
323 const char *pttype
= NULL
;
332 assert(root_hash
|| root_hash_size
== 0);
333 assert(!((flags
& DISSECT_IMAGE_GPT_ONLY
) && (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)));
335 /* Probes a disk image, and returns information about what it found in *ret.
337 * Returns -ENOPKG if no suitable partition table or file system could be found.
338 * Returns -EADDRNOTAVAIL if a root hash was specified but no matching root/verity partitions found. */
341 /* If a root hash is supplied, then we use the root partition that has a UUID that match the first
342 * 128bit of the root hash. And we use the verity partition that has a UUID that match the final
345 if (root_hash_size
< sizeof(sd_id128_t
))
348 memcpy(&root_uuid
, root_hash
, sizeof(sd_id128_t
));
349 memcpy(&verity_uuid
, (const uint8_t*) root_hash
+ root_hash_size
- sizeof(sd_id128_t
), sizeof(sd_id128_t
));
351 if (sd_id128_is_null(root_uuid
))
353 if (sd_id128_is_null(verity_uuid
))
357 if (fstat(fd
, &st
) < 0)
360 if (!S_ISBLK(st
.st_mode
))
363 b
= blkid_new_probe();
368 r
= blkid_probe_set_device(b
, fd
, 0, 0);
370 return errno_or_else(ENOMEM
);
372 if ((flags
& DISSECT_IMAGE_GPT_ONLY
) == 0) {
373 /* Look for file system superblocks, unless we only shall look for GPT partition tables */
374 blkid_probe_enable_superblocks(b
, 1);
375 blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
|BLKID_SUBLKS_USAGE
);
378 blkid_probe_enable_partitions(b
, 1);
379 blkid_probe_set_partitions_flags(b
, BLKID_PARTS_ENTRY_DETAILS
);
382 r
= blkid_do_safeprobe(b
);
383 if (IN_SET(r
, -2, 1))
384 return log_debug_errno(SYNTHETIC_ERRNO(ENOPKG
), "Failed to identify any partition table.");
386 return errno_or_else(EIO
);
388 m
= new0(DissectedImage
, 1);
392 r
= sd_device_new_from_devnum(&d
, 'b', st
.st_rdev
);
396 if ((!(flags
& DISSECT_IMAGE_GPT_ONLY
) &&
397 (flags
& DISSECT_IMAGE_REQUIRE_ROOT
)) ||
398 (flags
& DISSECT_IMAGE_NO_PARTITION_TABLE
)) {
399 const char *usage
= NULL
;
401 (void) blkid_probe_lookup_value(b
, "USAGE", &usage
, NULL
);
402 if (STRPTR_IN_SET(usage
, "filesystem", "crypto")) {
403 _cleanup_free_
char *t
= NULL
, *n
= NULL
, *o
= NULL
;
404 const char *fstype
= NULL
, *options
= NULL
;
406 /* OK, we have found a file system, that's our root partition then. */
407 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
415 r
= device_path_make_major_minor(st
.st_mode
, st
.st_rdev
, &n
);
419 m
->single_file_system
= true;
420 m
->verity
= root_hash
&& verity_data
;
421 m
->can_verity
= !!verity_data
;
423 options
= mount_options_from_part(mount_options
, PARTITION_ROOT
);
430 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
434 .architecture
= _ARCHITECTURE_INVALID
,
435 .fstype
= TAKE_PTR(t
),
437 .mount_options
= TAKE_PTR(o
),
440 m
->encrypted
= streq_ptr(fstype
, "crypto_LUKS");
442 /* Even on a single partition we need to wait for udev to create the
443 * /dev/block/X:Y symlink to /dev/loopZ */
444 r
= loop_wait_for_partitions_to_appear(fd
, d
, 0, flags
, &e
);
453 (void) blkid_probe_lookup_value(b
, "PTTYPE", &pttype
, NULL
);
457 is_gpt
= streq_ptr(pttype
, "gpt");
458 is_mbr
= streq_ptr(pttype
, "dos");
460 if (!is_gpt
&& ((flags
& DISSECT_IMAGE_GPT_ONLY
) || !is_mbr
))
464 pl
= blkid_probe_get_partitions(b
);
466 return errno_or_else(ENOMEM
);
468 r
= loop_wait_for_partitions_to_appear(fd
, d
, blkid_partlist_numof_partitions(pl
), flags
, &e
);
472 FOREACH_DEVICE(e
, q
) {
473 unsigned long long pflags
;
479 r
= sd_device_get_devnum(q
, &qn
);
483 if (st
.st_rdev
== qn
)
486 if (!device_is_block(q
))
489 if (device_is_mmc_special_partition(q
))
492 r
= sd_device_get_devname(q
, &node
);
496 pp
= blkid_partlist_devno_to_partition(pl
, qn
);
500 pflags
= blkid_partition_get_flags(pp
);
502 nr
= blkid_partition_get_partno(pp
);
507 int designator
= _PARTITION_DESIGNATOR_INVALID
, architecture
= _ARCHITECTURE_INVALID
;
508 const char *stype
, *sid
, *fstype
= NULL
;
509 sd_id128_t type_id
, id
;
512 sid
= blkid_partition_get_uuid(pp
);
515 if (sd_id128_from_string(sid
, &id
) < 0)
518 stype
= blkid_partition_get_type_string(pp
);
521 if (sd_id128_from_string(stype
, &type_id
) < 0)
524 if (sd_id128_equal(type_id
, GPT_HOME
)) {
526 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
528 if (pflags
& GPT_FLAG_NO_AUTO
)
531 designator
= PARTITION_HOME
;
532 rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
533 } else if (sd_id128_equal(type_id
, GPT_SRV
)) {
535 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
537 if (pflags
& GPT_FLAG_NO_AUTO
)
540 designator
= PARTITION_SRV
;
541 rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
542 } else if (sd_id128_equal(type_id
, GPT_ESP
)) {
544 /* Note that we don't check the GPT_FLAG_NO_AUTO flag for the ESP, as it is not defined
545 * there. We instead check the GPT_FLAG_NO_BLOCK_IO_PROTOCOL, as recommended by the
546 * UEFI spec (See "12.3.3 Number and Location of System Partitions"). */
548 if (pflags
& GPT_FLAG_NO_BLOCK_IO_PROTOCOL
)
551 designator
= PARTITION_ESP
;
554 } else if (sd_id128_equal(type_id
, GPT_XBOOTLDR
)) {
556 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
558 if (pflags
& GPT_FLAG_NO_AUTO
)
561 designator
= PARTITION_XBOOTLDR
;
562 rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
564 #ifdef GPT_ROOT_NATIVE
565 else if (sd_id128_equal(type_id
, GPT_ROOT_NATIVE
)) {
567 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
569 if (pflags
& GPT_FLAG_NO_AUTO
)
572 /* If a root ID is specified, ignore everything but the root id */
573 if (!sd_id128_is_null(root_uuid
) && !sd_id128_equal(root_uuid
, id
))
576 designator
= PARTITION_ROOT
;
577 architecture
= native_architecture();
578 rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
579 } else if (sd_id128_equal(type_id
, GPT_ROOT_NATIVE_VERITY
)) {
581 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
583 if (pflags
& GPT_FLAG_NO_AUTO
)
586 m
->can_verity
= true;
588 /* Ignore verity unless a root hash is specified */
589 if (sd_id128_is_null(verity_uuid
) || !sd_id128_equal(verity_uuid
, id
))
592 designator
= PARTITION_ROOT_VERITY
;
593 fstype
= "DM_verity_hash";
594 architecture
= native_architecture();
598 #ifdef GPT_ROOT_SECONDARY
599 else if (sd_id128_equal(type_id
, GPT_ROOT_SECONDARY
)) {
601 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
603 if (pflags
& GPT_FLAG_NO_AUTO
)
606 /* If a root ID is specified, ignore everything but the root id */
607 if (!sd_id128_is_null(root_uuid
) && !sd_id128_equal(root_uuid
, id
))
610 designator
= PARTITION_ROOT_SECONDARY
;
611 architecture
= SECONDARY_ARCHITECTURE
;
612 rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
613 } else if (sd_id128_equal(type_id
, GPT_ROOT_SECONDARY_VERITY
)) {
615 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
617 if (pflags
& GPT_FLAG_NO_AUTO
)
620 m
->can_verity
= true;
622 /* Ignore verity unless root has is specified */
623 if (sd_id128_is_null(verity_uuid
) || !sd_id128_equal(verity_uuid
, id
))
626 designator
= PARTITION_ROOT_SECONDARY_VERITY
;
627 fstype
= "DM_verity_hash";
628 architecture
= SECONDARY_ARCHITECTURE
;
632 else if (sd_id128_equal(type_id
, GPT_SWAP
)) {
634 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
);
636 if (pflags
& GPT_FLAG_NO_AUTO
)
639 designator
= PARTITION_SWAP
;
641 } else if (sd_id128_equal(type_id
, GPT_LINUX_GENERIC
)) {
643 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
645 if (pflags
& GPT_FLAG_NO_AUTO
)
649 multiple_generic
= true;
652 generic_rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
654 generic_node
= strdup(node
);
659 } else if (sd_id128_equal(type_id
, GPT_TMP
)) {
661 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
663 if (pflags
& GPT_FLAG_NO_AUTO
)
666 designator
= PARTITION_TMP
;
667 rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
669 } else if (sd_id128_equal(type_id
, GPT_VAR
)) {
671 check_partition_flags(node
, pflags
, GPT_FLAG_NO_AUTO
|GPT_FLAG_READ_ONLY
);
673 if (pflags
& GPT_FLAG_NO_AUTO
)
676 if (!FLAGS_SET(flags
, DISSECT_IMAGE_RELAX_VAR_CHECK
)) {
679 /* For /var we insist that the uuid of the partition matches the
680 * HMAC-SHA256 of the /var GPT partition type uuid, keyed by machine
681 * ID. Why? Unlike the other partitions /var is inherently
682 * installation specific, hence we need to be careful not to mount it
683 * in the wrong installation. By hashing the partition UUID from
684 * /etc/machine-id we can securely bind the partition to the
687 r
= sd_id128_get_machine_app_specific(GPT_VAR
, &var_uuid
);
691 if (!sd_id128_equal(var_uuid
, id
)) {
692 log_debug("Found a /var/ partition, but its UUID didn't match our expectations, ignoring.");
697 designator
= PARTITION_VAR
;
698 rw
= !(pflags
& GPT_FLAG_READ_ONLY
);
701 if (designator
!= _PARTITION_DESIGNATOR_INVALID
) {
702 _cleanup_free_
char *t
= NULL
, *n
= NULL
, *o
= NULL
;
703 const char *options
= NULL
;
706 if (m
->partitions
[designator
].found
)
719 options
= mount_options_from_part(mount_options
, designator
);
726 m
->partitions
[designator
] = (DissectedPartition
) {
730 .architecture
= architecture
,
732 .fstype
= TAKE_PTR(t
),
734 .mount_options
= TAKE_PTR(o
),
740 switch (blkid_partition_get_type(pp
)) {
742 case 0x83: /* Linux partition */
744 if (pflags
!= 0x80) /* Bootable flag */
748 multiple_generic
= true;
752 generic_node
= strdup(node
);
759 case 0xEA: { /* Boot Loader Spec extended $BOOT partition */
760 _cleanup_free_
char *n
= NULL
, *o
= NULL
;
761 sd_id128_t id
= SD_ID128_NULL
;
762 const char *sid
, *options
= NULL
;
765 if (m
->partitions
[PARTITION_XBOOTLDR
].found
)
768 sid
= blkid_partition_get_uuid(pp
);
770 (void) sd_id128_from_string(sid
, &id
);
776 options
= mount_options_from_part(mount_options
, PARTITION_XBOOTLDR
);
783 m
->partitions
[PARTITION_XBOOTLDR
] = (DissectedPartition
) {
787 .architecture
= _ARCHITECTURE_INVALID
,
790 .mount_options
= TAKE_PTR(o
),
798 if (!m
->partitions
[PARTITION_ROOT
].found
) {
799 /* No root partition found? Then let's see if ther's one for the secondary architecture. And if not
800 * either, then check if there's a single generic one, and use that. */
802 if (m
->partitions
[PARTITION_ROOT_VERITY
].found
)
803 return -EADDRNOTAVAIL
;
805 if (m
->partitions
[PARTITION_ROOT_SECONDARY
].found
) {
806 m
->partitions
[PARTITION_ROOT
] = m
->partitions
[PARTITION_ROOT_SECONDARY
];
807 zero(m
->partitions
[PARTITION_ROOT_SECONDARY
]);
809 m
->partitions
[PARTITION_ROOT_VERITY
] = m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY
];
810 zero(m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY
]);
812 } else if (flags
& DISSECT_IMAGE_REQUIRE_ROOT
) {
813 _cleanup_free_
char *o
= NULL
;
814 const char *options
= NULL
;
816 /* If the root hash was set, then we won't fall back to a generic node, because the
817 * root hash decides. */
819 return -EADDRNOTAVAIL
;
821 /* If we didn't find a generic node, then we can't fix this up either */
825 /* If we didn't find a properly marked root partition, but we did find a single suitable
826 * generic Linux partition, then use this as root partition, if the caller asked for it. */
827 if (multiple_generic
)
830 options
= mount_options_from_part(mount_options
, PARTITION_ROOT
);
837 m
->partitions
[PARTITION_ROOT
] = (DissectedPartition
) {
840 .partno
= generic_nr
,
841 .architecture
= _ARCHITECTURE_INVALID
,
842 .node
= TAKE_PTR(generic_node
),
843 .uuid
= generic_uuid
,
844 .mount_options
= TAKE_PTR(o
),
850 if (!m
->partitions
[PARTITION_ROOT_VERITY
].found
|| !m
->partitions
[PARTITION_ROOT
].found
)
851 return -EADDRNOTAVAIL
;
853 /* If we found the primary root with the hash, then we definitely want to suppress any secondary root
854 * (which would be weird, after all the root hash should only be assigned to one pair of
856 m
->partitions
[PARTITION_ROOT_SECONDARY
].found
= false;
857 m
->partitions
[PARTITION_ROOT_SECONDARY_VERITY
].found
= false;
859 /* If we found a verity setup, then the root partition is necessarily read-only. */
860 m
->partitions
[PARTITION_ROOT
].rw
= false;
868 /* Fill in file system types if we don't know them yet. */
869 for (i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
870 DissectedPartition
*p
= m
->partitions
+ i
;
875 if (!p
->fstype
&& p
->node
) {
876 r
= probe_filesystem(p
->node
, &p
->fstype
);
877 if (r
< 0 && r
!= -EUCLEAN
)
881 if (streq_ptr(p
->fstype
, "crypto_LUKS"))
884 if (p
->fstype
&& fstype_is_ro(p
->fstype
))
896 DissectedImage
* dissected_image_unref(DissectedImage
*m
) {
902 for (i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
903 free(m
->partitions
[i
].fstype
);
904 free(m
->partitions
[i
].node
);
905 free(m
->partitions
[i
].decrypted_fstype
);
906 free(m
->partitions
[i
].decrypted_node
);
907 free(m
->partitions
[i
].mount_options
);
911 strv_free(m
->machine_info
);
912 strv_free(m
->os_release
);
917 static int is_loop_device(const char *path
) {
918 char s
[SYS_BLOCK_PATH_MAX("/../loop/")];
923 if (stat(path
, &st
) < 0)
926 if (!S_ISBLK(st
.st_mode
))
929 xsprintf_sys_block_path(s
, "/loop/", st
.st_dev
);
930 if (access(s
, F_OK
) < 0) {
934 /* The device itself isn't a loop device, but maybe it's a partition and its parent is? */
935 xsprintf_sys_block_path(s
, "/../loop/", st
.st_dev
);
936 if (access(s
, F_OK
) < 0)
937 return errno
== ENOENT
? false : -errno
;
943 static int run_fsck(const char *node
, const char *fstype
) {
950 r
= fsck_exists(fstype
);
952 log_debug_errno(r
, "Couldn't determine whether fsck for %s exists, proceeding anyway.", fstype
);
956 log_debug("Not checking partition %s, as fsck for %s does not exist.", node
, fstype
);
960 r
= safe_fork("(fsck)", FORK_RESET_SIGNALS
|FORK_CLOSE_ALL_FDS
|FORK_RLIMIT_NOFILE_SAFE
|FORK_DEATHSIG
|FORK_NULL_STDIO
, &pid
);
962 return log_debug_errno(r
, "Failed to fork off fsck: %m");
965 execl("/sbin/fsck", "/sbin/fsck", "-aT", node
, NULL
);
966 log_debug_errno(errno
, "Failed to execl() fsck: %m");
967 _exit(FSCK_OPERATIONAL_ERROR
);
970 exit_status
= wait_for_terminate_and_check("fsck", pid
, 0);
972 return log_debug_errno(exit_status
, "Failed to fork off /sbin/fsck: %m");
974 if ((exit_status
& ~FSCK_ERROR_CORRECTED
) != FSCK_SUCCESS
) {
975 log_debug("fsck failed with exit status %i.", exit_status
);
977 if ((exit_status
& (FSCK_SYSTEM_SHOULD_REBOOT
|FSCK_ERRORS_LEFT_UNCORRECTED
)) != 0)
978 return log_debug_errno(SYNTHETIC_ERRNO(EUCLEAN
), "File system is corrupted, refusing.");
980 log_debug("Ignoring fsck error.");
986 static int mount_partition(
987 DissectedPartition
*m
,
989 const char *directory
,
991 DissectImageFlags flags
) {
993 _cleanup_free_
char *chased
= NULL
, *options
= NULL
;
994 const char *p
, *node
, *fstype
;
1001 node
= m
->decrypted_node
?: m
->node
;
1002 fstype
= m
->decrypted_fstype
?: m
->fstype
;
1004 if (!m
->found
|| !node
|| !fstype
)
1007 /* 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. */
1008 if (streq_ptr(fstype
, "crypto_LUKS"))
1011 rw
= m
->rw
&& !(flags
& DISSECT_IMAGE_READ_ONLY
);
1013 if (FLAGS_SET(flags
, DISSECT_IMAGE_FSCK
) && rw
) {
1014 r
= run_fsck(node
, fstype
);
1020 if (!FLAGS_SET(flags
, DISSECT_IMAGE_READ_ONLY
)) {
1021 /* Automatically create missing mount points, if necessary. */
1022 r
= mkdir_p_root(where
, directory
, uid_shift
, (gid_t
) uid_shift
, 0755);
1027 r
= chase_symlinks(directory
, where
, CHASE_PREFIX_ROOT
, &chased
, NULL
);
1035 /* If requested, turn on discard support. */
1036 if (fstype_can_discard(fstype
) &&
1037 ((flags
& DISSECT_IMAGE_DISCARD
) ||
1038 ((flags
& DISSECT_IMAGE_DISCARD_ON_LOOP
) && is_loop_device(m
->node
)))) {
1039 options
= strdup("discard");
1044 if (uid_is_valid(uid_shift
) && uid_shift
!= 0 && fstype_can_uid_gid(fstype
)) {
1045 _cleanup_free_
char *uid_option
= NULL
;
1047 if (asprintf(&uid_option
, "uid=" UID_FMT
",gid=" GID_FMT
, uid_shift
, (gid_t
) uid_shift
) < 0)
1050 if (!strextend_with_separator(&options
, ",", uid_option
, NULL
))
1054 if (!isempty(m
->mount_options
))
1055 if (!strextend_with_separator(&options
, ",", m
->mount_options
, NULL
))
1058 if (FLAGS_SET(flags
, DISSECT_IMAGE_MKDIR
)) {
1059 r
= mkdir_p(p
, 0755);
1064 r
= mount_verbose(LOG_DEBUG
, node
, p
, fstype
, MS_NODEV
|(rw
? 0 : MS_RDONLY
), options
);
1071 int dissected_image_mount(DissectedImage
*m
, const char *where
, uid_t uid_shift
, DissectImageFlags flags
) {
1072 int r
, xbootldr_mounted
;
1079 * -ENXIO → No root partition found
1080 * -EMEDIUMTYPE → DISSECT_IMAGE_VALIDATE_OS set but no os-release file found
1081 * -EUNATCH → Encrypted partition found for which no dm-crypt was set up yet
1082 * -EUCLEAN → fsck for file system failed
1083 * -EBUSY → File system already mounted/used elsewhere (kernel)
1086 if (!m
->partitions
[PARTITION_ROOT
].found
)
1089 if ((flags
& DISSECT_IMAGE_MOUNT_NON_ROOT_ONLY
) == 0) {
1090 r
= mount_partition(m
->partitions
+ PARTITION_ROOT
, where
, NULL
, uid_shift
, flags
);
1094 if (flags
& DISSECT_IMAGE_VALIDATE_OS
) {
1095 r
= path_is_os_tree(where
);
1099 return -EMEDIUMTYPE
;
1103 if (flags
& DISSECT_IMAGE_MOUNT_ROOT_ONLY
)
1106 /* Mask DISSECT_IMAGE_MKDIR for all subdirs: the idea is that only the top-level mount point is
1107 * created if needed, but the image itself not modified. */
1108 flags
&= ~DISSECT_IMAGE_MKDIR
;
1110 r
= mount_partition(m
->partitions
+ PARTITION_HOME
, where
, "/home", uid_shift
, flags
);
1114 r
= mount_partition(m
->partitions
+ PARTITION_SRV
, where
, "/srv", uid_shift
, flags
);
1118 r
= mount_partition(m
->partitions
+ PARTITION_VAR
, where
, "/var", uid_shift
, flags
);
1122 r
= mount_partition(m
->partitions
+ PARTITION_TMP
, where
, "/var/tmp", uid_shift
, flags
);
1126 xbootldr_mounted
= mount_partition(m
->partitions
+ PARTITION_XBOOTLDR
, where
, "/boot", uid_shift
, flags
);
1127 if (xbootldr_mounted
< 0)
1128 return xbootldr_mounted
;
1130 if (m
->partitions
[PARTITION_ESP
].found
) {
1131 int esp_done
= false;
1133 /* Mount the ESP to /efi if it exists. If it doesn't exist, use /boot instead, but only if it
1134 * exists and is empty, and we didn't already mount the XBOOTLDR partition into it. */
1136 r
= chase_symlinks("/efi", where
, CHASE_PREFIX_ROOT
, NULL
, NULL
);
1141 /* /efi doesn't exist. Let's see if /boot is suitable then */
1143 if (!xbootldr_mounted
) {
1144 _cleanup_free_
char *p
= NULL
;
1146 r
= chase_symlinks("/boot", where
, CHASE_PREFIX_ROOT
, &p
, NULL
);
1150 } else if (dir_is_empty(p
) > 0) {
1151 /* It exists and is an empty directory. Let's mount the ESP there. */
1152 r
= mount_partition(m
->partitions
+ PARTITION_ESP
, where
, "/boot", uid_shift
, flags
);
1162 /* OK, let's mount the ESP now to /efi (possibly creating the dir if missing) */
1164 r
= mount_partition(m
->partitions
+ PARTITION_ESP
, where
, "/efi", uid_shift
, flags
);
1173 int dissected_image_mount_and_warn(DissectedImage
*m
, const char *where
, uid_t uid_shift
, DissectImageFlags flags
) {
1179 r
= dissected_image_mount(m
, where
, uid_shift
, flags
);
1181 return log_error_errno(r
, "Not root file system found in image.");
1182 if (r
== -EMEDIUMTYPE
)
1183 return log_error_errno(r
, "No suitable os-release file in image found.");
1185 return log_error_errno(r
, "Encrypted file system discovered, but decryption not requested.");
1187 return log_error_errno(r
, "File system check on image failed.");
1189 return log_error_errno(r
, "File system already mounted elsewhere.");
1191 return log_error_errno(r
, "Failed to mount image: %m");
1196 #if HAVE_LIBCRYPTSETUP
1197 typedef struct DecryptedPartition
{
1198 struct crypt_device
*device
;
1201 } DecryptedPartition
;
1203 struct DecryptedImage
{
1204 DecryptedPartition
*decrypted
;
1210 DecryptedImage
* decrypted_image_unref(DecryptedImage
* d
) {
1211 #if HAVE_LIBCRYPTSETUP
1218 for (i
= 0; i
< d
->n_decrypted
; i
++) {
1219 DecryptedPartition
*p
= d
->decrypted
+ i
;
1221 if (p
->device
&& p
->name
&& !p
->relinquished
) {
1222 r
= crypt_deactivate(p
->device
, p
->name
);
1224 log_debug_errno(r
, "Failed to deactivate encrypted partition %s", p
->name
);
1228 crypt_free(p
->device
);
1237 #if HAVE_LIBCRYPTSETUP
1239 static int make_dm_name_and_node(const void *original_node
, const char *suffix
, char **ret_name
, char **ret_node
) {
1240 _cleanup_free_
char *name
= NULL
, *node
= NULL
;
1243 assert(original_node
);
1248 base
= strrchr(original_node
, '/');
1250 base
= original_node
;
1256 name
= strjoin(base
, suffix
);
1259 if (!filename_is_valid(name
))
1262 node
= path_join(crypt_get_dir(), name
);
1266 *ret_name
= TAKE_PTR(name
);
1267 *ret_node
= TAKE_PTR(node
);
1272 static int decrypt_partition(
1273 DissectedPartition
*m
,
1274 const char *passphrase
,
1275 DissectImageFlags flags
,
1276 DecryptedImage
*d
) {
1278 _cleanup_free_
char *node
= NULL
, *name
= NULL
;
1279 _cleanup_(crypt_freep
) struct crypt_device
*cd
= NULL
;
1285 if (!m
->found
|| !m
->node
|| !m
->fstype
)
1288 if (!streq(m
->fstype
, "crypto_LUKS"))
1294 r
= make_dm_name_and_node(m
->node
, "-decrypted", &name
, &node
);
1298 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_allocated
, d
->n_decrypted
+ 1))
1301 r
= crypt_init(&cd
, m
->node
);
1303 return log_debug_errno(r
, "Failed to initialize dm-crypt: %m");
1305 cryptsetup_enable_logging(cd
);
1307 r
= crypt_load(cd
, CRYPT_LUKS
, NULL
);
1309 return log_debug_errno(r
, "Failed to load LUKS metadata: %m");
1311 r
= crypt_activate_by_passphrase(cd
, name
, CRYPT_ANY_SLOT
, passphrase
, strlen(passphrase
),
1312 ((flags
& DISSECT_IMAGE_READ_ONLY
) ? CRYPT_ACTIVATE_READONLY
: 0) |
1313 ((flags
& DISSECT_IMAGE_DISCARD_ON_CRYPTO
) ? CRYPT_ACTIVATE_ALLOW_DISCARDS
: 0));
1315 log_debug_errno(r
, "Failed to activate LUKS device: %m");
1316 return r
== -EPERM
? -EKEYREJECTED
: r
;
1319 d
->decrypted
[d
->n_decrypted
].name
= TAKE_PTR(name
);
1320 d
->decrypted
[d
->n_decrypted
].device
= TAKE_PTR(cd
);
1323 m
->decrypted_node
= TAKE_PTR(node
);
1328 static int verity_can_reuse(const void *root_hash
, size_t root_hash_size
, bool has_sig
, const char *name
, struct crypt_device
**ret_cd
) {
1329 /* If the same volume was already open, check that the root hashes match, and reuse it if they do */
1330 _cleanup_free_
char *root_hash_existing
= NULL
;
1331 _cleanup_(crypt_freep
) struct crypt_device
*cd
= NULL
;
1332 struct crypt_params_verity crypt_params
= {};
1333 size_t root_hash_existing_size
= root_hash_size
;
1338 r
= crypt_init_by_name(&cd
, name
);
1340 return log_debug_errno(r
, "Error opening verity device, crypt_init_by_name failed: %m");
1341 r
= crypt_get_verity_info(cd
, &crypt_params
);
1343 return log_debug_errno(r
, "Error opening verity device, crypt_get_verity_info failed: %m");
1344 root_hash_existing
= malloc0(root_hash_size
);
1345 if (!root_hash_existing
)
1347 r
= crypt_volume_key_get(cd
, CRYPT_ANY_SLOT
, root_hash_existing
, &root_hash_existing_size
, NULL
, 0);
1349 return log_debug_errno(r
, "Error opening verity device, crypt_volume_key_get failed: %m");
1350 if (root_hash_size
!= root_hash_existing_size
|| memcmp(root_hash_existing
, root_hash
, root_hash_size
) != 0)
1351 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but root hashes are different.");
1352 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1353 /* Ensure that, if signatures are supported, we only reuse the device if the previous mount
1354 * used the same settings, so that a previous unsigned mount will not be reused if the user
1355 * asks to use signing for the new one, and viceversa. */
1356 if (has_sig
!= !!(crypt_params
.flags
& CRYPT_VERITY_ROOT_HASH_SIGNATURE
))
1357 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Error opening verity device, it already exists but signature settings are not the same.");
1360 *ret_cd
= TAKE_PTR(cd
);
1364 static inline void dm_deferred_remove_clean(char *name
) {
1367 (void) crypt_deactivate_by_name(NULL
, name
, CRYPT_DEACTIVATE_DEFERRED
);
1370 DEFINE_TRIVIAL_CLEANUP_FUNC(char *, dm_deferred_remove_clean
);
1372 static int verity_partition(
1373 DissectedPartition
*m
,
1374 DissectedPartition
*v
,
1375 const void *root_hash
,
1376 size_t root_hash_size
,
1377 const char *verity_data
,
1378 const char *root_hash_sig_path
,
1379 const void *root_hash_sig
,
1380 size_t root_hash_sig_size
,
1381 DissectImageFlags flags
,
1382 DecryptedImage
*d
) {
1384 _cleanup_free_
char *node
= NULL
, *name
= NULL
, *hash_sig_from_file
= NULL
;
1385 _cleanup_(crypt_freep
) struct crypt_device
*cd
= NULL
;
1386 _cleanup_(dm_deferred_remove_cleanp
) char *restore_deferred_remove
= NULL
;
1390 assert(v
|| verity_data
);
1395 if (!m
->found
|| !m
->node
|| !m
->fstype
)
1398 if (!v
->found
|| !v
->node
|| !v
->fstype
)
1401 if (!streq(v
->fstype
, "DM_verity_hash"))
1405 if (FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
)) {
1406 /* Use the roothash, which is unique per volume, as the device node name, so that it can be reused */
1407 _cleanup_free_
char *root_hash_encoded
= NULL
;
1408 root_hash_encoded
= hexmem(root_hash
, root_hash_size
);
1409 if (!root_hash_encoded
)
1411 r
= make_dm_name_and_node(root_hash_encoded
, "-verity", &name
, &node
);
1413 r
= make_dm_name_and_node(m
->node
, "-verity", &name
, &node
);
1417 if (!root_hash_sig
&& root_hash_sig_path
) {
1418 r
= read_full_file_full(AT_FDCWD
, root_hash_sig_path
, 0, &hash_sig_from_file
, &root_hash_sig_size
);
1423 r
= crypt_init(&cd
, verity_data
?: v
->node
);
1427 cryptsetup_enable_logging(cd
);
1429 r
= crypt_load(cd
, CRYPT_VERITY
, NULL
);
1433 r
= crypt_set_data_device(cd
, m
->node
);
1437 if (!GREEDY_REALLOC0(d
->decrypted
, d
->n_allocated
, d
->n_decrypted
+ 1))
1440 /* If activating fails because the device already exists, check the metadata and reuse it if it matches.
1441 * In case of ENODEV/ENOENT, which can happen if another process is activating at the exact same time,
1442 * retry a few times before giving up. */
1443 for (unsigned i
= 0; i
< N_DEVICE_NODE_LIST_ATTEMPTS
; i
++) {
1444 if (root_hash_sig
|| hash_sig_from_file
) {
1445 #if HAVE_CRYPT_ACTIVATE_BY_SIGNED_KEY
1446 r
= crypt_activate_by_signed_key(cd
, name
, root_hash
, root_hash_size
, root_hash_sig
?: hash_sig_from_file
, root_hash_sig_size
, CRYPT_ACTIVATE_READONLY
);
1448 r
= log_debug_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "activation of verity device with signature requested, but not supported by cryptsetup due to missing crypt_activate_by_signed_key()");
1451 r
= crypt_activate_by_volume_key(cd
, name
, root_hash
, root_hash_size
, CRYPT_ACTIVATE_READONLY
);
1452 /* libdevmapper can return EINVAL when the device is already in the activation stage.
1453 * There's no way to distinguish this situation from a genuine error due to invalid
1454 * parameters, so immediately fall back to activating the device with a unique name.
1455 * Improvements in libcrypsetup can ensure this never happens: https://gitlab.com/cryptsetup/cryptsetup/-/merge_requests/96 */
1456 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
1457 return verity_partition(m
, v
, root_hash
, root_hash_size
, verity_data
, NULL
, root_hash_sig
?: hash_sig_from_file
, root_hash_sig_size
, flags
& ~DISSECT_IMAGE_VERITY_SHARE
, d
);
1460 -EEXIST
, /* Volume is already open and ready to be used */
1461 -EBUSY
, /* Volume is being opened but not ready, crypt_init_by_name can fetch details */
1462 -ENODEV
/* Volume is being opened but not ready, crypt_init_by_name would fail, try to open again */))
1464 if (IN_SET(r
, -EEXIST
, -EBUSY
)) {
1465 struct crypt_device
*existing_cd
= NULL
;
1467 if (!restore_deferred_remove
){
1468 /* To avoid races, disable automatic removal on umount while setting up the new device. Restore it on failure. */
1469 r
= dm_deferred_remove_cancel(name
);
1470 /* If activation returns EBUSY there might be no deferred removal to cancel, that's fine */
1471 if (r
< 0 && r
!= -ENXIO
)
1472 return log_debug_errno(r
, "Disabling automated deferred removal for verity device %s failed: %m", node
);
1474 restore_deferred_remove
= strdup(name
);
1475 if (!restore_deferred_remove
)
1480 r
= verity_can_reuse(root_hash
, root_hash_size
, !!root_hash_sig
|| !!hash_sig_from_file
, name
, &existing_cd
);
1481 /* Same as above, -EINVAL can randomly happen when it actually means -EEXIST */
1482 if (r
== -EINVAL
&& FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
1483 return verity_partition(m
, v
, root_hash
, root_hash_size
, verity_data
, NULL
, root_hash_sig
?: hash_sig_from_file
, root_hash_sig_size
, flags
& ~DISSECT_IMAGE_VERITY_SHARE
, d
);
1484 if (!IN_SET(r
, 0, -ENODEV
, -ENOENT
, -EBUSY
))
1485 return log_debug_errno(r
, "Checking whether existing verity device %s can be reused failed: %m", node
);
1487 /* devmapper might say that the device exists, but the devlink might not yet have been
1488 * created. Check and wait for the udev event in that case. */
1489 r
= device_wait_for_devlink(node
, "block", 100 * USEC_PER_MSEC
, NULL
);
1490 /* Fallback to activation with a unique device if it's taking too long */
1491 if (r
== -ETIMEDOUT
)
1504 /* Device is being opened by another process, but it has not finished yet, yield for 2ms */
1505 (void) usleep(2 * USEC_PER_MSEC
);
1508 /* An existing verity device was reported by libcryptsetup/libdevmapper, but we can't use it at this time.
1509 * Fall back to activating it with a unique device name. */
1510 if (r
!= 0 && FLAGS_SET(flags
, DISSECT_IMAGE_VERITY_SHARE
))
1511 return verity_partition(m
, v
, root_hash
, root_hash_size
, verity_data
, NULL
, root_hash_sig
?: hash_sig_from_file
, root_hash_sig_size
, flags
& ~DISSECT_IMAGE_VERITY_SHARE
, d
);
1513 /* Everything looks good and we'll be able to mount the device, so deferred remove will be re-enabled at that point. */
1514 restore_deferred_remove
= mfree(restore_deferred_remove
);
1516 d
->decrypted
[d
->n_decrypted
].name
= TAKE_PTR(name
);
1517 d
->decrypted
[d
->n_decrypted
].device
= TAKE_PTR(cd
);
1520 m
->decrypted_node
= TAKE_PTR(node
);
1526 int dissected_image_decrypt(
1528 const char *passphrase
,
1529 const void *root_hash
,
1530 size_t root_hash_size
,
1531 const char *verity_data
,
1532 const char *root_hash_sig_path
,
1533 const void *root_hash_sig
,
1534 size_t root_hash_sig_size
,
1535 DissectImageFlags flags
,
1536 DecryptedImage
**ret
) {
1538 #if HAVE_LIBCRYPTSETUP
1539 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*d
= NULL
;
1545 assert(root_hash
|| root_hash_size
== 0);
1549 * = 0 → There was nothing to decrypt
1550 * > 0 → Decrypted successfully
1551 * -ENOKEY → There's something to decrypt but no key was supplied
1552 * -EKEYREJECTED → Passed key was not correct
1555 if (root_hash
&& root_hash_size
< sizeof(sd_id128_t
))
1558 if (!m
->encrypted
&& !m
->verity
) {
1563 #if HAVE_LIBCRYPTSETUP
1564 d
= new0(DecryptedImage
, 1);
1568 for (i
= 0; i
< _PARTITION_DESIGNATOR_MAX
; i
++) {
1569 DissectedPartition
*p
= m
->partitions
+ i
;
1575 r
= decrypt_partition(p
, passphrase
, flags
, d
);
1579 k
= PARTITION_VERITY_OF(i
);
1581 r
= verity_partition(p
, m
->partitions
+ k
, root_hash
, root_hash_size
, verity_data
, root_hash_sig_path
, root_hash_sig
, root_hash_sig_size
, flags
| DISSECT_IMAGE_VERITY_SHARE
, d
);
1586 if (!p
->decrypted_fstype
&& p
->decrypted_node
) {
1587 r
= probe_filesystem(p
->decrypted_node
, &p
->decrypted_fstype
);
1588 if (r
< 0 && r
!= -EUCLEAN
)
1601 int dissected_image_decrypt_interactively(
1603 const char *passphrase
,
1604 const void *root_hash
,
1605 size_t root_hash_size
,
1606 const char *verity_data
,
1607 const char *root_hash_sig_path
,
1608 const void *root_hash_sig
,
1609 size_t root_hash_sig_size
,
1610 DissectImageFlags flags
,
1611 DecryptedImage
**ret
) {
1613 _cleanup_strv_free_erase_
char **z
= NULL
;
1620 r
= dissected_image_decrypt(m
, passphrase
, root_hash
, root_hash_size
, verity_data
, root_hash_sig_path
, root_hash_sig
, root_hash_sig_size
, flags
, ret
);
1623 if (r
== -EKEYREJECTED
)
1624 log_error_errno(r
, "Incorrect passphrase, try again!");
1625 else if (r
!= -ENOKEY
)
1626 return log_error_errno(r
, "Failed to decrypt image: %m");
1629 return log_error_errno(SYNTHETIC_ERRNO(EKEYREJECTED
),
1630 "Too many retries.");
1634 r
= ask_password_auto("Please enter image passphrase:", NULL
, "dissect", "dissect", USEC_INFINITY
, 0, &z
);
1636 return log_error_errno(r
, "Failed to query for passphrase: %m");
1642 int decrypted_image_relinquish(DecryptedImage
*d
) {
1644 #if HAVE_LIBCRYPTSETUP
1651 /* Turns on automatic removal after the last use ended for all DM devices of this image, and sets a boolean so
1652 * that we don't clean it up ourselves either anymore */
1654 #if HAVE_LIBCRYPTSETUP
1655 for (i
= 0; i
< d
->n_decrypted
; i
++) {
1656 DecryptedPartition
*p
= d
->decrypted
+ i
;
1658 if (p
->relinquished
)
1661 r
= crypt_deactivate_by_name(NULL
, p
->name
, CRYPT_DEACTIVATE_DEFERRED
);
1663 return log_debug_errno(r
, "Failed to mark %s for auto-removal: %m", p
->name
);
1665 p
->relinquished
= true;
1672 int verity_metadata_load(
1674 const char *root_hash_path
,
1675 void **ret_roothash
,
1676 size_t *ret_roothash_size
,
1677 char **ret_verity_data
,
1678 char **ret_roothashsig
) {
1680 _cleanup_free_
char *verity_filename
= NULL
, *roothashsig_filename
= NULL
;
1681 _cleanup_free_
void *roothash_decoded
= NULL
;
1682 size_t roothash_decoded_size
= 0;
1687 if (is_device_path(image
)) {
1688 /* If we are asked to load the root hash for a device node, exit early */
1690 *ret_roothash
= NULL
;
1691 if (ret_roothash_size
)
1692 *ret_roothash_size
= 0;
1693 if (ret_verity_data
)
1694 *ret_verity_data
= NULL
;
1695 if (ret_roothashsig
)
1696 *ret_roothashsig
= NULL
;
1700 if (ret_verity_data
) {
1703 verity_filename
= new(char, strlen(image
) + STRLEN(".verity") + 1);
1704 if (!verity_filename
)
1706 strcpy(verity_filename
, image
);
1707 e
= endswith(verity_filename
, ".raw");
1709 strcpy(e
, ".verity");
1711 strcat(verity_filename
, ".verity");
1713 r
= access(verity_filename
, F_OK
);
1715 if (errno
!= ENOENT
)
1717 verity_filename
= mfree(verity_filename
);
1721 if (ret_roothashsig
) {
1724 /* Follow naming convention recommended by the relevant RFC:
1725 * https://tools.ietf.org/html/rfc5751#section-3.2.1 */
1726 roothashsig_filename
= new(char, strlen(image
) + STRLEN(".roothash.p7s") + 1);
1727 if (!roothashsig_filename
)
1729 strcpy(roothashsig_filename
, image
);
1730 e
= endswith(roothashsig_filename
, ".raw");
1732 strcpy(e
, ".roothash.p7s");
1734 strcat(roothashsig_filename
, ".roothash.p7s");
1736 r
= access(roothashsig_filename
, R_OK
);
1738 if (errno
!= ENOENT
)
1740 roothashsig_filename
= mfree(roothashsig_filename
);
1745 _cleanup_free_
char *text
= NULL
;
1746 assert(ret_roothash_size
);
1748 if (root_hash_path
) {
1749 /* We have the path to a roothash to load and decode, eg: RootHash=/foo/bar.roothash */
1750 r
= read_one_line_file(root_hash_path
, &text
);
1754 r
= getxattr_malloc(image
, "user.verity.roothash", &text
, true);
1758 if (!IN_SET(r
, -ENODATA
, -EOPNOTSUPP
, -ENOENT
))
1761 fn
= newa(char, strlen(image
) + STRLEN(".roothash") + 1);
1762 n
= stpcpy(fn
, image
);
1763 e
= endswith(fn
, ".raw");
1767 strcpy(n
, ".roothash");
1769 r
= read_one_line_file(fn
, &text
);
1770 if (r
< 0 && r
!= -ENOENT
)
1776 r
= unhexmem(text
, strlen(text
), &roothash_decoded
, &roothash_decoded_size
);
1779 if (roothash_decoded_size
< sizeof(sd_id128_t
))
1785 *ret_roothash
= TAKE_PTR(roothash_decoded
);
1786 *ret_roothash_size
= roothash_decoded_size
;
1788 if (ret_verity_data
)
1789 *ret_verity_data
= TAKE_PTR(verity_filename
);
1790 if (roothashsig_filename
)
1791 *ret_roothashsig
= TAKE_PTR(roothashsig_filename
);
1796 int dissected_image_acquire_metadata(DissectedImage
*m
) {
1806 static const char *const paths
[_META_MAX
] = {
1807 [META_HOSTNAME
] = "/etc/hostname\0",
1808 [META_MACHINE_ID
] = "/etc/machine-id\0",
1809 [META_MACHINE_INFO
] = "/etc/machine-info\0",
1810 [META_OS_RELEASE
] = "/etc/os-release\0"
1811 "/usr/lib/os-release\0",
1814 _cleanup_strv_free_
char **machine_info
= NULL
, **os_release
= NULL
;
1815 _cleanup_close_pair_
int error_pipe
[2] = { -1, -1 };
1816 _cleanup_(rmdir_and_freep
) char *t
= NULL
;
1817 _cleanup_(sigkill_waitp
) pid_t child
= 0;
1818 sd_id128_t machine_id
= SD_ID128_NULL
;
1819 _cleanup_free_
char *hostname
= NULL
;
1820 unsigned n_meta_initialized
= 0, k
;
1821 int fds
[2 * _META_MAX
], r
, v
;
1824 BLOCK_SIGNALS(SIGCHLD
);
1828 for (; n_meta_initialized
< _META_MAX
; n_meta_initialized
++)
1829 if (pipe2(fds
+ 2*n_meta_initialized
, O_CLOEXEC
) < 0) {
1834 r
= mkdtemp_malloc("/tmp/dissect-XXXXXX", &t
);
1838 if (pipe2(error_pipe
, O_CLOEXEC
) < 0) {
1843 r
= safe_fork("(sd-dissect)", FORK_RESET_SIGNALS
|FORK_DEATHSIG
|FORK_NEW_MOUNTNS
|FORK_MOUNTNS_SLAVE
, &child
);
1847 error_pipe
[0] = safe_close(error_pipe
[0]);
1849 r
= dissected_image_mount(m
, t
, UID_INVALID
, DISSECT_IMAGE_READ_ONLY
|DISSECT_IMAGE_MOUNT_ROOT_ONLY
|DISSECT_IMAGE_VALIDATE_OS
);
1851 /* Let parent know the error */
1852 (void) write(error_pipe
[1], &r
, sizeof(r
));
1854 log_debug_errno(r
, "Failed to mount dissected image: %m");
1855 _exit(EXIT_FAILURE
);
1858 for (k
= 0; k
< _META_MAX
; k
++) {
1859 _cleanup_close_
int fd
= -ENOENT
;
1862 fds
[2*k
] = safe_close(fds
[2*k
]);
1864 NULSTR_FOREACH(p
, paths
[k
]) {
1865 fd
= chase_symlinks_and_open(p
, t
, CHASE_PREFIX_ROOT
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
, NULL
);
1870 log_debug_errno(fd
, "Failed to read %s file of image, ignoring: %m", paths
[k
]);
1871 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
1875 r
= copy_bytes(fd
, fds
[2*k
+1], (uint64_t) -1, 0);
1877 (void) write(error_pipe
[1], &r
, sizeof(r
));
1878 _exit(EXIT_FAILURE
);
1881 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
1884 _exit(EXIT_SUCCESS
);
1887 error_pipe
[1] = safe_close(error_pipe
[1]);
1889 for (k
= 0; k
< _META_MAX
; k
++) {
1890 _cleanup_fclose_
FILE *f
= NULL
;
1892 fds
[2*k
+1] = safe_close(fds
[2*k
+1]);
1894 f
= take_fdopen(&fds
[2*k
], "r");
1903 r
= read_etc_hostname_stream(f
, &hostname
);
1905 log_debug_errno(r
, "Failed to read /etc/hostname: %m");
1909 case META_MACHINE_ID
: {
1910 _cleanup_free_
char *line
= NULL
;
1912 r
= read_line(f
, LONG_LINE_MAX
, &line
);
1914 log_debug_errno(r
, "Failed to read /etc/machine-id: %m");
1916 r
= sd_id128_from_string(line
, &machine_id
);
1918 log_debug_errno(r
, "Image contains invalid /etc/machine-id: %s", line
);
1920 log_debug("/etc/machine-id file is empty.");
1922 log_debug("/etc/machine-id has unexpected length %i.", r
);
1927 case META_MACHINE_INFO
:
1928 r
= load_env_file_pairs(f
, "machine-info", &machine_info
);
1930 log_debug_errno(r
, "Failed to read /etc/machine-info: %m");
1934 case META_OS_RELEASE
:
1935 r
= load_env_file_pairs(f
, "os-release", &os_release
);
1937 log_debug_errno(r
, "Failed to read OS release file: %m");
1943 r
= wait_for_terminate_and_check("(sd-dissect)", child
, 0);
1948 n
= read(error_pipe
[0], &v
, sizeof(v
));
1952 return v
; /* propagate error sent to us from child */
1956 if (r
!= EXIT_SUCCESS
)
1959 free_and_replace(m
->hostname
, hostname
);
1960 m
->machine_id
= machine_id
;
1961 strv_free_and_replace(m
->machine_info
, machine_info
);
1962 strv_free_and_replace(m
->os_release
, os_release
);
1965 for (k
= 0; k
< n_meta_initialized
; k
++)
1966 safe_close_pair(fds
+ 2*k
);
1971 int dissect_image_and_warn(
1974 const void *root_hash
,
1975 size_t root_hash_size
,
1976 const char *verity_data
,
1977 const MountOptions
*mount_options
,
1978 DissectImageFlags flags
,
1979 DissectedImage
**ret
) {
1981 _cleanup_free_
char *buffer
= NULL
;
1985 r
= fd_get_path(fd
, &buffer
);
1992 r
= dissect_image(fd
, root_hash
, root_hash_size
, verity_data
, mount_options
, flags
, ret
);
1997 return log_error_errno(r
, "Dissecting images is not supported, compiled without blkid support.");
2000 return log_error_errno(r
, "Couldn't identify a suitable partition table or file system in '%s'.", name
);
2002 case -EADDRNOTAVAIL
:
2003 return log_error_errno(r
, "No root partition for specified root hash found in '%s'.", name
);
2006 return log_error_errno(r
, "Multiple suitable root partitions found in image '%s'.", name
);
2009 return log_error_errno(r
, "No suitable root partition found in image '%s'.", name
);
2011 case -EPROTONOSUPPORT
:
2012 return log_error_errno(r
, "Device '%s' is loopback block device with partition scanning turned off, please turn it on.", name
);
2016 return log_error_errno(r
, "Failed to dissect image '%s': %m", name
);
2022 bool dissected_image_can_do_verity(const DissectedImage
*image
, unsigned partition_designator
) {
2023 if (image
->single_file_system
)
2024 return partition_designator
== PARTITION_ROOT
&& image
->can_verity
;
2026 return PARTITION_VERITY_OF(partition_designator
) >= 0;
2029 bool dissected_image_has_verity(const DissectedImage
*image
, unsigned partition_designator
) {
2032 if (image
->single_file_system
)
2033 return partition_designator
== PARTITION_ROOT
&& image
->verity
;
2035 k
= PARTITION_VERITY_OF(partition_designator
);
2036 return k
>= 0 && image
->partitions
[k
].found
;
2039 MountOptions
* mount_options_free_all(MountOptions
*options
) {
2042 while ((m
= options
)) {
2043 LIST_REMOVE(mount_options
, options
, m
);
2051 const char* mount_options_from_part(const MountOptions
*options
, int designator
) {
2054 LIST_FOREACH(mount_options
, m
, (MountOptions
*)options
)
2055 if (designator
== m
->partition_designator
&& !isempty(m
->options
))
2061 int mount_image_privately_interactively(
2063 DissectImageFlags flags
,
2064 char **ret_directory
,
2065 LoopDevice
**ret_loop_device
,
2066 DecryptedImage
**ret_decrypted_image
) {
2068 _cleanup_(loop_device_unrefp
) LoopDevice
*d
= NULL
;
2069 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
2070 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
2071 _cleanup_(rmdir_and_freep
) char *created_dir
= NULL
;
2072 _cleanup_free_
char *temp
= NULL
;
2075 /* Mounts an OS image at a temporary place, inside a newly created mount namespace of our own. This
2076 * is used by tools such as systemd-tmpfiles or systemd-firstboot to operate on some disk image
2080 assert(ret_directory
);
2081 assert(ret_loop_device
);
2082 assert(ret_decrypted_image
);
2084 r
= tempfn_random_child(NULL
, program_invocation_short_name
, &temp
);
2086 return log_error_errno(r
, "Failed to generate temporary mount directory: %m");
2088 r
= loop_device_make_by_path(
2090 FLAGS_SET(flags
, DISSECT_IMAGE_READ_ONLY
) ? O_RDONLY
: O_RDWR
,
2091 FLAGS_SET(flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
2094 return log_error_errno(r
, "Failed to set up loopback device: %m");
2096 r
= dissect_image_and_warn(d
->fd
, image
, NULL
, 0, NULL
, NULL
, flags
, &dissected_image
);
2100 r
= dissected_image_decrypt_interactively(dissected_image
, NULL
, NULL
, 0, NULL
, NULL
, NULL
, 0, flags
, &decrypted_image
);
2104 r
= detach_mount_namespace();
2106 return log_error_errno(r
, "Failed to detach mount namespace: %m");
2108 r
= mkdir_p(temp
, 0700);
2110 return log_error_errno(r
, "Failed to create mount point: %m");
2112 created_dir
= TAKE_PTR(temp
);
2114 r
= dissected_image_mount_and_warn(dissected_image
, created_dir
, UID_INVALID
, flags
);
2118 if (decrypted_image
) {
2119 r
= decrypted_image_relinquish(decrypted_image
);
2121 return log_error_errno(r
, "Failed to relinquish DM devices: %m");
2124 loop_device_relinquish(d
);
2126 *ret_directory
= TAKE_PTR(created_dir
);
2127 *ret_loop_device
= TAKE_PTR(d
);
2128 *ret_decrypted_image
= TAKE_PTR(decrypted_image
);
2133 static const char *const partition_designator_table
[] = {
2134 [PARTITION_ROOT
] = "root",
2135 [PARTITION_ROOT_SECONDARY
] = "root-secondary",
2136 [PARTITION_HOME
] = "home",
2137 [PARTITION_SRV
] = "srv",
2138 [PARTITION_ESP
] = "esp",
2139 [PARTITION_XBOOTLDR
] = "xbootldr",
2140 [PARTITION_SWAP
] = "swap",
2141 [PARTITION_ROOT_VERITY
] = "root-verity",
2142 [PARTITION_ROOT_SECONDARY_VERITY
] = "root-secondary-verity",
2143 [PARTITION_TMP
] = "tmp",
2144 [PARTITION_VAR
] = "var",
2147 DEFINE_STRING_TABLE_LOOKUP(partition_designator
, int);