1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
3 #include <linux/loop.h>
9 #if HAVE_VALGRIND_MEMCHECK_H
10 #include <valgrind/memcheck.h>
13 #include "sd-daemon.h"
14 #include "sd-device.h"
18 #include "blkid-util.h"
19 #include "blockdev-util.h"
20 #include "btrfs-util.h"
21 #include "chattr-util.h"
22 #include "device-util.h"
23 #include "devnum-util.h"
26 #include "errno-util.h"
28 #include "fdisk-util.h"
30 #include "filesystems.h"
32 #include "fsck-util.h"
33 #include "glyph-util.h"
35 #include "home-util.h"
36 #include "homework-luks.h"
37 #include "homework-mount.h"
39 #include "keyring-util.h"
40 #include "memory-util.h"
41 #include "missing_magic.h"
43 #include "mkfs-util.h"
44 #include "mount-util.h"
45 #include "openssl-util.h"
46 #include "parse-util.h"
47 #include "path-util.h"
48 #include "process-util.h"
49 #include "random-util.h"
50 #include "resize-fs.h"
52 #include "sync-util.h"
53 #include "tmpfile-util.h"
54 #include "udev-util.h"
55 #include "user-util.h"
57 /* Round down to the nearest 4K size. Given that newer hardware generally prefers 4K sectors, let's align our
58 * partitions to that too. In the worst case we'll waste 3.5K per partition that way, but I think I can live
60 #define DISK_SIZE_ROUND_DOWN(x) ((x) & ~UINT64_C(4095))
62 /* Rounds up to the nearest 4K boundary. Returns UINT64_MAX on overflow */
63 #define DISK_SIZE_ROUND_UP(x) \
66 _x > UINT64_MAX - 4095U ? UINT64_MAX : (_x + 4095U) & ~UINT64_C(4095); \
69 /* How much larger will the image on disk be than the fs inside it, i.e. the space we pay for the GPT and
70 * LUKS2 envelope. (As measured on cryptsetup 2.4.1) */
71 #define GPT_LUKS2_OVERHEAD UINT64_C(18874368)
73 static int resize_image_loop(UserRecord
*h
, HomeSetup
*setup
, uint64_t old_image_size
, uint64_t new_image_size
, uint64_t *ret_image_size
);
75 int run_mark_dirty(int fd
, bool b
) {
79 /* Sets or removes the 'user.home-dirty' xattr on the specified file. We use this to detect when a
80 * home directory was not properly unmounted. */
84 r
= fd_verify_regular(fd
);
89 ret
= fsetxattr(fd
, "user.home-dirty", &x
, 1, XATTR_CREATE
);
90 if (ret
< 0 && errno
!= EEXIST
)
91 return log_debug_errno(errno
, "Could not mark home directory as dirty: %m");
96 return log_debug_errno(r
, "Failed to synchronize image before marking it clean: %m");
98 ret
= fremovexattr(fd
, "user.home-dirty");
99 if (ret
< 0 && !ERRNO_IS_XATTR_ABSENT(errno
))
100 return log_debug_errno(errno
, "Could not mark home directory as clean: %m");
105 return log_debug_errno(r
, "Failed to synchronize dirty flag to disk: %m");
110 int run_mark_dirty_by_path(const char *path
, bool b
) {
111 _cleanup_close_
int fd
= -1;
115 fd
= open(path
, O_RDWR
|O_CLOEXEC
|O_NOCTTY
);
117 return log_debug_errno(errno
, "Failed to open %s to mark dirty or clean: %m", path
);
119 return run_mark_dirty(fd
, b
);
122 static int probe_file_system_by_fd(
125 sd_id128_t
*ret_uuid
) {
127 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
128 _cleanup_free_
char *s
= NULL
;
129 const char *fstype
= NULL
, *uuid
= NULL
;
137 b
= blkid_new_probe();
142 r
= blkid_probe_set_device(b
, fd
, 0, 0);
144 return errno
> 0 ? -errno
: -ENOMEM
;
146 (void) blkid_probe_enable_superblocks(b
, 1);
147 (void) blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
|BLKID_SUBLKS_UUID
);
150 r
= blkid_do_safeprobe(b
);
151 if (IN_SET(r
, -2, 1)) /* nothing found or ambiguous result */
154 return errno
> 0 ? -errno
: -EIO
;
156 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
160 (void) blkid_probe_lookup_value(b
, "UUID", &uuid
, NULL
);
164 r
= sd_id128_from_string(uuid
, &id
);
172 *ret_fstype
= TAKE_PTR(s
);
178 static int probe_file_system_by_path(const char *path
, char **ret_fstype
, sd_id128_t
*ret_uuid
) {
179 _cleanup_close_
int fd
= -1;
181 fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
|O_NONBLOCK
);
183 return negative_errno();
185 return probe_file_system_by_fd(fd
, ret_fstype
, ret_uuid
);
188 static int block_get_size_by_fd(int fd
, uint64_t *ret
) {
194 if (fstat(fd
, &st
) < 0)
197 if (!S_ISBLK(st
.st_mode
))
200 return RET_NERRNO(ioctl(fd
, BLKGETSIZE64
, ret
));
203 static int block_get_size_by_path(const char *path
, uint64_t *ret
) {
204 _cleanup_close_
int fd
= -1;
206 fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
|O_NONBLOCK
);
210 return block_get_size_by_fd(fd
, ret
);
213 static int run_fsck(const char *node
, const char *fstype
) {
220 r
= fsck_exists_for_fstype(fstype
);
222 return log_error_errno(r
, "Failed to check if fsck for file system %s exists: %m", fstype
);
224 log_warning("No fsck for file system %s installed, ignoring.", fstype
);
228 r
= safe_fork("(fsck)",
229 FORK_RESET_SIGNALS
|FORK_RLIMIT_NOFILE_SAFE
|FORK_DEATHSIG
|FORK_LOG
|FORK_STDOUT_TO_STDERR
|FORK_CLOSE_ALL_FDS
,
235 execl("/sbin/fsck", "/sbin/fsck", "-aTl", node
, NULL
);
237 log_error_errno(errno
, "Failed to execute fsck: %m");
238 _exit(FSCK_OPERATIONAL_ERROR
);
241 exit_status
= wait_for_terminate_and_check("fsck", fsck_pid
, WAIT_LOG_ABNORMAL
);
244 if ((exit_status
& ~FSCK_ERROR_CORRECTED
) != 0) {
245 log_warning("fsck failed with exit status %i.", exit_status
);
247 if ((exit_status
& (FSCK_SYSTEM_SHOULD_REBOOT
|FSCK_ERRORS_LEFT_UNCORRECTED
)) != 0)
248 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "File system is corrupted, refusing.");
250 log_warning("Ignoring fsck error.");
253 log_info("File system check completed.");
258 DEFINE_TRIVIAL_CLEANUP_FUNC_FULL(key_serial_t
, keyring_unlink
, -1);
260 static int upload_to_keyring(
262 const char *password
,
263 key_serial_t
*ret_key_serial
) {
265 _cleanup_free_
char *name
= NULL
;
271 /* If auto-shrink-on-logout is turned on, we need to keep the key we used to unlock the LUKS volume
272 * around, since we'll need it when automatically resizing (since we can't ask the user there
273 * again). We do this by uploading it into the kernel keyring, specifically the "session" one. This
274 * is done under the assumption systemd-homed gets its private per-session keyring (i.e. default
275 * service behaviour, given that KeyringMode=private is the default). It will survive between our
276 * systemd-homework invocations that way.
278 * If auto-shrink-on-logout is disabled we'll skip this step, to be frugal with sensitive data. */
280 if (user_record_auto_resize_mode(h
) != AUTO_RESIZE_SHRINK_AND_GROW
) { /* Won't need it */
282 *ret_key_serial
= -1;
286 name
= strjoin("homework-user-", h
->user_name
);
290 serial
= add_key("user", name
, password
, strlen(password
), KEY_SPEC_SESSION_KEYRING
);
295 *ret_key_serial
= serial
;
300 static int luks_try_passwords(
302 struct crypt_device
*cd
,
305 size_t *volume_key_size
,
306 key_serial_t
*ret_key_serial
) {
313 STRV_FOREACH(pp
, passwords
) {
314 size_t vks
= *volume_key_size
;
316 r
= sym_crypt_volume_key_get(
324 if (ret_key_serial
) {
325 /* If ret_key_serial is non-NULL, let's try to upload the password that
326 * worked, and return its serial. */
327 r
= upload_to_keyring(h
, *pp
, ret_key_serial
);
329 log_debug_errno(r
, "Failed to upload LUKS password to kernel keyring, ignoring: %m");
330 *ret_key_serial
= -1;
334 *volume_key_size
= vks
;
338 log_debug_errno(r
, "Password %zu didn't work for unlocking LUKS superblock: %m", (size_t) (pp
- passwords
));
344 static int luks_setup(
350 const char *cipher_mode
,
351 uint64_t volume_key_size
,
353 const PasswordCache
*cache
,
355 struct crypt_device
**ret
,
356 sd_id128_t
*ret_found_uuid
,
357 void **ret_volume_key
,
358 size_t *ret_volume_key_size
,
359 key_serial_t
*ret_key_serial
) {
361 _cleanup_(keyring_unlinkp
) key_serial_t key_serial
= -1;
362 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
363 _cleanup_(erase_and_freep
) void *vk
= NULL
;
374 r
= sym_crypt_init(&cd
, node
);
376 return log_error_errno(r
, "Failed to allocate libcryptsetup context: %m");
378 cryptsetup_enable_logging(cd
);
380 r
= sym_crypt_load(cd
, CRYPT_LUKS2
, NULL
);
382 return log_error_errno(r
, "Failed to load LUKS superblock: %m");
384 r
= sym_crypt_get_volume_key_size(cd
);
386 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine LUKS volume key size");
389 if (!sd_id128_is_null(uuid
) || ret_found_uuid
) {
392 s
= sym_crypt_get_uuid(cd
);
394 return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE
), "LUKS superblock has no UUID.");
396 r
= sd_id128_from_string(s
, &p
);
398 return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE
), "LUKS superblock has invalid UUID.");
400 /* Check that the UUID matches, if specified */
401 if (!sd_id128_is_null(uuid
) &&
402 !sd_id128_equal(uuid
, p
))
403 return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE
), "LUKS superblock has wrong UUID.");
406 if (cipher
&& !streq_ptr(cipher
, sym_crypt_get_cipher(cd
)))
407 return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE
), "LUKS superblock declares wrong cipher.");
409 if (cipher_mode
&& !streq_ptr(cipher_mode
, sym_crypt_get_cipher_mode(cd
)))
410 return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE
), "LUKS superblock declares wrong cipher mode.");
412 if (volume_key_size
!= UINT64_MAX
&& vks
!= volume_key_size
)
413 return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE
), "LUKS superblock declares wrong volume key size.");
420 FOREACH_POINTER(list
,
421 cache
? cache
->keyring_passswords
: NULL
,
422 cache
? cache
->pkcs11_passwords
: NULL
,
423 cache
? cache
->fido2_passwords
: NULL
,
425 r
= luks_try_passwords(h
, cd
, list
, vk
, &vks
, ret_key_serial
? &key_serial
: NULL
);
430 return log_error_errno(r
, "No valid password for LUKS superblock.");
432 return log_error_errno(r
, "Failed to unlock LUKS superblock: %m");
434 r
= sym_crypt_activate_by_volume_key(
438 discard
? CRYPT_ACTIVATE_ALLOW_DISCARDS
: 0);
440 return log_error_errno(r
, "Failed to unlock LUKS superblock: %m");
442 log_info("Setting up LUKS device /dev/mapper/%s completed.", dm_name
);
446 if (ret_found_uuid
) /* Return the UUID actually found if the caller wants to know */
449 *ret_volume_key
= TAKE_PTR(vk
);
450 if (ret_volume_key_size
)
451 *ret_volume_key_size
= vks
;
453 *ret_key_serial
= TAKE_KEY_SERIAL(key_serial
);
458 static int make_dm_names(UserRecord
*h
, HomeSetup
*setup
) {
460 assert(h
->user_name
);
463 if (!setup
->dm_name
) {
464 setup
->dm_name
= strjoin("home-", h
->user_name
);
469 if (!setup
->dm_node
) {
470 setup
->dm_node
= path_join("/dev/mapper/", setup
->dm_name
);
478 static int acquire_open_luks_device(
483 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
488 assert(!setup
->crypt_device
);
490 r
= dlopen_cryptsetup();
494 r
= make_dm_names(h
, setup
);
498 r
= sym_crypt_init_by_name(&cd
, setup
->dm_name
);
499 if ((ERRNO_IS_DEVICE_ABSENT(r
) || r
== -EINVAL
) && graceful
)
502 return log_error_errno(r
, "Failed to initialize cryptsetup context for %s: %m", setup
->dm_name
);
504 cryptsetup_enable_logging(cd
);
506 setup
->crypt_device
= TAKE_PTR(cd
);
510 static int luks_open(
513 const PasswordCache
*cache
,
514 sd_id128_t
*ret_found_uuid
,
515 void **ret_volume_key
,
516 size_t *ret_volume_key_size
) {
518 _cleanup_(erase_and_freep
) void *vk
= NULL
;
526 assert(!setup
->crypt_device
);
528 /* Opens a LUKS device that is already set up. Re-validates the password while doing so (which also
529 * provides us with the volume key, which we want). */
531 r
= acquire_open_luks_device(h
, setup
, /* graceful= */ false);
535 r
= sym_crypt_load(setup
->crypt_device
, CRYPT_LUKS2
, NULL
);
537 return log_error_errno(r
, "Failed to load LUKS superblock: %m");
539 r
= sym_crypt_get_volume_key_size(setup
->crypt_device
);
541 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine LUKS volume key size");
544 if (ret_found_uuid
) {
547 s
= sym_crypt_get_uuid(setup
->crypt_device
);
549 return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE
), "LUKS superblock has no UUID.");
551 r
= sd_id128_from_string(s
, &p
);
553 return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE
), "LUKS superblock has invalid UUID.");
561 FOREACH_POINTER(list
,
562 cache
? cache
->keyring_passswords
: NULL
,
563 cache
? cache
->pkcs11_passwords
: NULL
,
564 cache
? cache
->fido2_passwords
: NULL
,
566 r
= luks_try_passwords(h
, setup
->crypt_device
, list
, vk
, &vks
, NULL
);
571 return log_error_errno(r
, "No valid password for LUKS superblock.");
573 return log_error_errno(r
, "Failed to unlocks LUKS superblock: %m");
575 log_info("Discovered used LUKS device /dev/mapper/%s, and validated password.", setup
->dm_name
);
577 /* This is needed so that crypt_resize() can operate correctly for pre-existing LUKS devices. We need
578 * to tell libcryptsetup the volume key explicitly, so that it is in the kernel keyring. */
579 r
= sym_crypt_activate_by_volume_key(setup
->crypt_device
, NULL
, vk
, vks
, CRYPT_ACTIVATE_KEYRING_KEY
);
581 return log_error_errno(r
, "Failed to upload volume key again: %m");
583 log_info("Successfully re-activated LUKS device.");
588 *ret_volume_key
= TAKE_PTR(vk
);
589 if (ret_volume_key_size
)
590 *ret_volume_key_size
= vks
;
595 static int fs_validate(
599 sd_id128_t
*ret_found_uuid
) {
601 _cleanup_free_
char *fstype
= NULL
;
608 r
= probe_file_system_by_path(dm_node
, &fstype
, &u
);
610 return log_error_errno(r
, "Failed to probe file system: %m");
612 /* Limit the set of supported file systems a bit, as protection against little tested kernel file
613 * systems. Also, we only support the resize ioctls for these file systems. */
614 if (!supported_fstype(fstype
))
615 return log_error_errno(SYNTHETIC_ERRNO(EPROTONOSUPPORT
), "Image contains unsupported file system: %s", strna(fstype
));
617 if (!sd_id128_is_null(uuid
) &&
618 !sd_id128_equal(uuid
, u
))
619 return log_error_errno(SYNTHETIC_ERRNO(EMEDIUMTYPE
), "File system has wrong UUID.");
621 log_info("Probing file system completed (found %s).", fstype
);
623 *ret_fstype
= TAKE_PTR(fstype
);
625 if (ret_found_uuid
) /* Return the UUID actually found if the caller wants to know */
631 static int luks_validate(
634 sd_id128_t partition_uuid
,
635 sd_id128_t
*ret_partition_uuid
,
636 uint64_t *ret_offset
,
637 uint64_t *ret_size
) {
639 _cleanup_(blkid_free_probep
) blkid_probe b
= NULL
;
640 sd_id128_t found_partition_uuid
= SD_ID128_NULL
;
641 const char *fstype
= NULL
, *pttype
= NULL
;
642 blkid_loff_t offset
= 0, size
= 0;
652 b
= blkid_new_probe();
657 r
= blkid_probe_set_device(b
, fd
, 0, 0);
659 return errno
> 0 ? -errno
: -ENOMEM
;
661 (void) blkid_probe_enable_superblocks(b
, 1);
662 (void) blkid_probe_set_superblocks_flags(b
, BLKID_SUBLKS_TYPE
);
663 (void) blkid_probe_enable_partitions(b
, 1);
664 (void) blkid_probe_set_partitions_flags(b
, BLKID_PARTS_ENTRY_DETAILS
);
667 r
= blkid_do_safeprobe(b
);
668 if (IN_SET(r
, -2, 1)) /* nothing found or ambiguous result */
671 return errno
> 0 ? -errno
: -EIO
;
673 (void) blkid_probe_lookup_value(b
, "TYPE", &fstype
, NULL
);
674 if (streq_ptr(fstype
, "crypto_LUKS")) {
675 /* Directly a LUKS image */
677 *ret_size
= UINT64_MAX
; /* full disk */
678 *ret_partition_uuid
= SD_ID128_NULL
;
683 (void) blkid_probe_lookup_value(b
, "PTTYPE", &pttype
, NULL
);
684 if (!streq_ptr(pttype
, "gpt"))
688 pl
= blkid_probe_get_partitions(b
);
690 return errno
> 0 ? -errno
: -ENOMEM
;
693 n
= blkid_partlist_numof_partitions(pl
);
695 return errno
> 0 ? -errno
: -EIO
;
697 for (int i
= 0; i
< n
; i
++) {
699 sd_id128_t id
= SD_ID128_NULL
;
703 pp
= blkid_partlist_get_partition(pl
, i
);
705 return errno
> 0 ? -errno
: -EIO
;
707 if (sd_id128_string_equal(blkid_partition_get_type_string(pp
), SD_GPT_USER_HOME
) <= 0)
710 if (!streq_ptr(blkid_partition_get_name(pp
), label
))
713 sid
= blkid_partition_get_uuid(pp
);
715 r
= sd_id128_from_string(sid
, &id
);
717 log_debug_errno(r
, "Couldn't parse partition UUID %s, weird: %m", sid
);
719 if (!sd_id128_is_null(partition_uuid
) && !sd_id128_equal(id
, partition_uuid
))
726 offset
= blkid_partition_get_start(pp
);
727 size
= blkid_partition_get_size(pp
);
728 found_partition_uuid
= id
;
738 if ((uint64_t) offset
> UINT64_MAX
/ 512U)
742 if ((uint64_t) size
> UINT64_MAX
/ 512U)
745 *ret_offset
= offset
* 512U;
746 *ret_size
= size
* 512U;
747 *ret_partition_uuid
= found_partition_uuid
;
752 static int crypt_device_to_evp_cipher(struct crypt_device
*cd
, const EVP_CIPHER
**ret
) {
753 _cleanup_free_
char *cipher_name
= NULL
;
754 const char *cipher
, *cipher_mode
, *e
;
755 size_t key_size
, key_bits
;
756 const EVP_CIPHER
*cc
;
761 /* Let's find the right OpenSSL EVP_CIPHER object that matches the encryption settings of the LUKS
764 cipher
= sym_crypt_get_cipher(cd
);
766 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot get cipher from LUKS device.");
768 cipher_mode
= sym_crypt_get_cipher_mode(cd
);
770 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Cannot get cipher mode from LUKS device.");
772 e
= strchr(cipher_mode
, '-');
774 cipher_mode
= strndupa_safe(cipher_mode
, e
- cipher_mode
);
776 r
= sym_crypt_get_volume_key_size(cd
);
778 return log_error_errno(r
< 0 ? r
: SYNTHETIC_ERRNO(EINVAL
), "Cannot get volume key size from LUKS device.");
781 key_bits
= key_size
* 8;
782 if (streq(cipher_mode
, "xts"))
785 if (asprintf(&cipher_name
, "%s-%zu-%s", cipher
, key_bits
, cipher_mode
) < 0)
788 cc
= EVP_get_cipherbyname(cipher_name
);
790 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "Selected cipher mode '%s' not supported, can't encrypt JSON record.", cipher_name
);
792 /* Verify that our key length calculations match what OpenSSL thinks */
793 r
= EVP_CIPHER_key_length(cc
);
794 if (r
< 0 || (uint64_t) r
!= key_size
)
795 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Key size of selected cipher doesn't meet our expectations.");
801 static int luks_validate_home_record(
802 struct crypt_device
*cd
,
804 const void *volume_key
,
805 PasswordCache
*cache
,
806 UserRecord
**ret_luks_home_record
) {
813 for (int token
= 0; token
< sym_crypt_token_max(CRYPT_LUKS2
); token
++) {
814 _cleanup_(json_variant_unrefp
) JsonVariant
*v
= NULL
, *rr
= NULL
;
815 _cleanup_(EVP_CIPHER_CTX_freep
) EVP_CIPHER_CTX
*context
= NULL
;
816 _cleanup_(user_record_unrefp
) UserRecord
*lhr
= NULL
;
817 _cleanup_free_
void *encrypted
= NULL
, *iv
= NULL
;
818 size_t decrypted_size
, encrypted_size
, iv_size
;
819 int decrypted_size_out1
, decrypted_size_out2
;
820 _cleanup_free_
char *decrypted
= NULL
;
821 const char *text
, *type
;
822 crypt_token_info state
;
823 JsonVariant
*jr
, *jiv
;
824 unsigned line
, column
;
825 const EVP_CIPHER
*cc
;
827 state
= sym_crypt_token_status(cd
, token
, &type
);
828 if (state
== CRYPT_TOKEN_INACTIVE
) /* First unconfigured token, give up */
830 if (IN_SET(state
, CRYPT_TOKEN_INTERNAL
, CRYPT_TOKEN_INTERNAL_UNKNOWN
, CRYPT_TOKEN_EXTERNAL
))
832 if (state
!= CRYPT_TOKEN_EXTERNAL_UNKNOWN
)
833 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Unexpected token state of token %i: %i", token
, (int) state
);
835 if (!streq(type
, "systemd-homed"))
838 r
= sym_crypt_token_json_get(cd
, token
, &text
);
840 return log_error_errno(r
, "Failed to read LUKS token %i: %m", token
);
842 r
= json_parse(text
, JSON_PARSE_SENSITIVE
, &v
, &line
, &column
);
844 return log_error_errno(r
, "Failed to parse LUKS token JSON data %u:%u: %m", line
, column
);
846 jr
= json_variant_by_key(v
, "record");
848 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "LUKS token lacks 'record' field.");
849 jiv
= json_variant_by_key(v
, "iv");
851 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "LUKS token lacks 'iv' field.");
853 r
= json_variant_unbase64(jr
, &encrypted
, &encrypted_size
);
855 return log_error_errno(r
, "Failed to base64 decode record: %m");
857 r
= json_variant_unbase64(jiv
, &iv
, &iv_size
);
859 return log_error_errno(r
, "Failed to base64 decode IV: %m");
861 r
= crypt_device_to_evp_cipher(cd
, &cc
);
864 if (iv_size
> INT_MAX
|| EVP_CIPHER_iv_length(cc
) != (int) iv_size
)
865 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "IV size doesn't match.");
867 context
= EVP_CIPHER_CTX_new();
871 if (EVP_DecryptInit_ex(context
, cc
, NULL
, volume_key
, iv
) != 1)
872 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to initialize decryption context.");
874 decrypted_size
= encrypted_size
+ EVP_CIPHER_key_length(cc
) * 2;
875 decrypted
= new(char, decrypted_size
);
879 if (EVP_DecryptUpdate(context
, (uint8_t*) decrypted
, &decrypted_size_out1
, encrypted
, encrypted_size
) != 1)
880 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to decrypt JSON record.");
882 assert((size_t) decrypted_size_out1
<= decrypted_size
);
884 if (EVP_DecryptFinal_ex(context
, (uint8_t*) decrypted
+ decrypted_size_out1
, &decrypted_size_out2
) != 1)
885 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to finish decryption of JSON record.");
887 assert((size_t) decrypted_size_out1
+ (size_t) decrypted_size_out2
< decrypted_size
);
888 decrypted_size
= (size_t) decrypted_size_out1
+ (size_t) decrypted_size_out2
;
890 if (memchr(decrypted
, 0, decrypted_size
))
891 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Inner NUL byte in JSON record, refusing.");
893 decrypted
[decrypted_size
] = 0;
895 r
= json_parse(decrypted
, JSON_PARSE_SENSITIVE
, &rr
, NULL
, NULL
);
897 return log_error_errno(r
, "Failed to parse decrypted JSON record, refusing.");
899 lhr
= user_record_new();
903 r
= user_record_load(lhr
, rr
, USER_RECORD_LOAD_EMBEDDED
|USER_RECORD_PERMISSIVE
);
905 return log_error_errno(r
, "Failed to parse user record: %m");
907 if (!user_record_compatible(h
, lhr
))
908 return log_error_errno(SYNTHETIC_ERRNO(EREMCHG
), "LUKS home record not compatible with host record, refusing.");
910 r
= user_record_authenticate(lhr
, h
, cache
, /* strict_verify= */ true);
913 assert(r
> 0); /* Insist that a password was verified */
915 *ret_luks_home_record
= TAKE_PTR(lhr
);
919 return log_error_errno(SYNTHETIC_ERRNO(EBADMSG
), "Couldn't find home record in LUKS2 header, refusing.");
922 static int format_luks_token_text(
923 struct crypt_device
*cd
,
925 const void *volume_key
,
928 int r
, encrypted_size_out1
= 0, encrypted_size_out2
= 0, iv_size
, key_size
;
929 _cleanup_(EVP_CIPHER_CTX_freep
) EVP_CIPHER_CTX
*context
= NULL
;
930 _cleanup_(json_variant_unrefp
) JsonVariant
*v
= NULL
;
931 _cleanup_free_
void *iv
= NULL
, *encrypted
= NULL
;
932 size_t text_length
, encrypted_size
;
933 _cleanup_free_
char *text
= NULL
;
934 const EVP_CIPHER
*cc
;
941 r
= crypt_device_to_evp_cipher(cd
, &cc
);
945 key_size
= EVP_CIPHER_key_length(cc
);
946 iv_size
= EVP_CIPHER_iv_length(cc
);
949 iv
= malloc(iv_size
);
953 r
= crypto_random_bytes(iv
, iv_size
);
955 return log_error_errno(r
, "Failed to generate IV: %m");
958 context
= EVP_CIPHER_CTX_new();
962 if (EVP_EncryptInit_ex(context
, cc
, NULL
, volume_key
, iv
) != 1)
963 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to initialize encryption context.");
965 r
= json_variant_format(hr
->json
, 0, &text
);
967 return log_error_errno(r
, "Failed to format user record for LUKS: %m");
969 text_length
= strlen(text
);
970 encrypted_size
= text_length
+ 2*key_size
- 1;
972 encrypted
= malloc(encrypted_size
);
976 if (EVP_EncryptUpdate(context
, encrypted
, &encrypted_size_out1
, (uint8_t*) text
, text_length
) != 1)
977 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to encrypt JSON record.");
979 assert((size_t) encrypted_size_out1
<= encrypted_size
);
981 if (EVP_EncryptFinal_ex(context
, (uint8_t*) encrypted
+ encrypted_size_out1
, &encrypted_size_out2
) != 1)
982 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to finish encryption of JSON record. ");
984 assert((size_t) encrypted_size_out1
+ (size_t) encrypted_size_out2
<= encrypted_size
);
988 JSON_BUILD_PAIR("type", JSON_BUILD_CONST_STRING("systemd-homed")),
989 JSON_BUILD_PAIR("keyslots", JSON_BUILD_EMPTY_ARRAY
),
990 JSON_BUILD_PAIR("record", JSON_BUILD_BASE64(encrypted
, encrypted_size_out1
+ encrypted_size_out2
)),
991 JSON_BUILD_PAIR("iv", JSON_BUILD_BASE64(iv
, iv_size
))));
993 return log_error_errno(r
, "Failed to prepare LUKS JSON token object: %m");
995 r
= json_variant_format(v
, 0, ret
);
997 return log_error_errno(r
, "Failed to format encrypted user record for LUKS: %m");
1002 int home_store_header_identity_luks(
1005 UserRecord
*old_home
) {
1007 _cleanup_(user_record_unrefp
) UserRecord
*header_home
= NULL
;
1008 _cleanup_free_
char *text
= NULL
;
1013 if (!setup
->crypt_device
)
1016 assert(setup
->volume_key
);
1018 /* Let's store the user's identity record in the LUKS2 "token" header data fields, in an encrypted
1019 * fashion. Why that? If we'd rely on the record being embedded in the payload file system itself we
1020 * would have to mount the file system before we can validate the JSON record, its signatures and
1021 * whether it matches what we are looking for. However, kernel file system implementations are
1022 * generally not ready to be used on untrusted media. Hence let's store the record independently of
1023 * the file system, so that we can validate it first, and only then mount the file system. To keep
1024 * things simple we use the same encryption settings for this record as for the file system itself. */
1026 r
= user_record_clone(h
, USER_RECORD_EXTRACT_EMBEDDED
|USER_RECORD_PERMISSIVE
, &header_home
);
1028 return log_error_errno(r
, "Failed to determine new header record: %m");
1030 if (old_home
&& user_record_equal(old_home
, header_home
)) {
1031 log_debug("Not updating header home record.");
1035 r
= format_luks_token_text(setup
->crypt_device
, header_home
, setup
->volume_key
, &text
);
1039 for (int token
= 0; token
< sym_crypt_token_max(CRYPT_LUKS2
); token
++) {
1040 crypt_token_info state
;
1043 state
= sym_crypt_token_status(setup
->crypt_device
, token
, &type
);
1044 if (state
== CRYPT_TOKEN_INACTIVE
) /* First unconfigured token, we are done */
1046 if (IN_SET(state
, CRYPT_TOKEN_INTERNAL
, CRYPT_TOKEN_INTERNAL_UNKNOWN
, CRYPT_TOKEN_EXTERNAL
))
1047 continue; /* Not ours */
1048 if (state
!= CRYPT_TOKEN_EXTERNAL_UNKNOWN
)
1049 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Unexpected token state of token %i: %i", token
, (int) state
);
1051 if (!streq(type
, "systemd-homed"))
1054 r
= sym_crypt_token_json_set(setup
->crypt_device
, token
, text
);
1056 return log_error_errno(r
, "Failed to set JSON token for slot %i: %m", token
);
1058 /* Now, let's free the text so that for all further matching tokens we all crypt_json_token_set()
1059 * with a NULL text in order to invalidate the tokens. */
1064 return log_error_errno(SYNTHETIC_ERRNO(EBADMSG
), "Didn't find any record token to update.");
1066 log_info("Wrote LUKS header user record.");
1071 int run_fitrim(int root_fd
) {
1072 struct fstrim_range range
= {
1076 /* If discarding is on, discard everything right after mounting, so that the discard setting takes
1077 * effect on activation. (Also, optionally, trim on logout) */
1079 assert(root_fd
>= 0);
1081 if (ioctl(root_fd
, FITRIM
, &range
) < 0) {
1082 if (ERRNO_IS_NOT_SUPPORTED(errno
) || errno
== EBADF
) {
1083 log_debug_errno(errno
, "File system does not support FITRIM, not trimming.");
1087 return log_warning_errno(errno
, "Failed to invoke FITRIM, ignoring: %m");
1090 log_info("Discarded unused %s.", FORMAT_BYTES(range
.len
));
1094 int run_fallocate(int backing_fd
, const struct stat
*st
) {
1097 assert(backing_fd
>= 0);
1099 /* If discarding is off, let's allocate the whole image before mounting, so that the setting takes
1100 * effect on activation */
1103 if (fstat(backing_fd
, &stbuf
) < 0)
1104 return log_error_errno(errno
, "Failed to fstat(): %m");
1109 if (!S_ISREG(st
->st_mode
))
1112 if (st
->st_blocks
>= DIV_ROUND_UP(st
->st_size
, 512)) {
1113 log_info("Backing file is fully allocated already.");
1117 if (fallocate(backing_fd
, FALLOC_FL_KEEP_SIZE
, 0, st
->st_size
) < 0) {
1119 if (ERRNO_IS_NOT_SUPPORTED(errno
)) {
1120 log_debug_errno(errno
, "fallocate() not supported on file system, ignoring.");
1124 if (ERRNO_IS_DISK_SPACE(errno
)) {
1125 log_debug_errno(errno
, "Not enough disk space to fully allocate home.");
1126 return -ENOSPC
; /* make recognizable */
1129 return log_error_errno(errno
, "Failed to allocate backing file blocks: %m");
1132 log_info("Allocated additional %s.",
1133 FORMAT_BYTES((DIV_ROUND_UP(st
->st_size
, 512) - st
->st_blocks
) * 512));
1137 int run_fallocate_by_path(const char *backing_path
) {
1138 _cleanup_close_
int backing_fd
= -1;
1140 backing_fd
= open(backing_path
, O_RDWR
|O_CLOEXEC
|O_NOCTTY
|O_NONBLOCK
);
1142 return log_error_errno(errno
, "Failed to open '%s' for fallocate(): %m", backing_path
);
1144 return run_fallocate(backing_fd
, NULL
);
1147 static int lock_image_fd(int image_fd
, const char *ip
) {
1150 /* If the $SYSTEMD_LUKS_LOCK environment variable is set we'll take an exclusive BSD lock on the
1151 * image file, and send it to our parent. homed will keep it open to ensure no other instance of
1152 * homed (across the network or such) will also mount the file. */
1154 assert(image_fd
>= 0);
1157 r
= getenv_bool("SYSTEMD_LUKS_LOCK");
1161 return log_error_errno(r
, "Failed to parse $SYSTEMD_LUKS_LOCK environment variable: %m");
1165 if (flock(image_fd
, LOCK_EX
|LOCK_NB
) < 0) {
1167 if (errno
== EAGAIN
)
1168 log_error_errno(errno
, "Image file '%s' already locked, can't use.", ip
);
1170 log_error_errno(errno
, "Failed to lock image file '%s': %m", ip
);
1172 return errno
!= EAGAIN
? -errno
: -EADDRINUSE
; /* Make error recognizable */
1175 log_info("Successfully locked image file '%s'.", ip
);
1177 /* Now send it to our parent to keep safe while the home dir is active */
1178 r
= sd_pid_notify_with_fds(0, false, "SYSTEMD_LUKS_LOCK_FD=1", &image_fd
, 1);
1180 log_warning_errno(r
, "Failed to send LUKS lock fd to parent, ignoring: %m");
1185 static int open_image_file(
1187 const char *force_image_path
,
1188 struct stat
*ret_stat
) {
1190 _cleanup_close_
int image_fd
= -1;
1195 assert(h
|| force_image_path
);
1197 ip
= force_image_path
?: user_record_image_path(h
);
1199 image_fd
= open(ip
, O_RDWR
|O_CLOEXEC
|O_NOCTTY
|O_NONBLOCK
);
1201 return log_error_errno(errno
, "Failed to open image file %s: %m", ip
);
1203 if (fstat(image_fd
, &st
) < 0)
1204 return log_error_errno(errno
, "Failed to fstat() image file: %m");
1205 if (!S_ISREG(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
1206 return log_error_errno(
1207 S_ISDIR(st
.st_mode
) ? SYNTHETIC_ERRNO(EISDIR
) : SYNTHETIC_ERRNO(EBADFD
),
1208 "Image file %s is not a regular file or block device: %m", ip
);
1210 /* Locking block devices doesn't really make sense, as this might interfere with
1211 * udev's workings, and these locks aren't network propagated anyway, hence not what
1212 * we are after here. */
1213 if (S_ISREG(st
.st_mode
)) {
1214 r
= lock_image_fd(image_fd
, ip
);
1222 return TAKE_FD(image_fd
);
1225 int home_setup_luks(
1227 HomeSetupFlags flags
,
1228 const char *force_image_path
,
1230 PasswordCache
*cache
,
1231 UserRecord
**ret_luks_home
) {
1233 sd_id128_t found_partition_uuid
, found_fs_uuid
, found_luks_uuid
= SD_ID128_NULL
;
1234 _cleanup_(user_record_unrefp
) UserRecord
*luks_home
= NULL
;
1235 _cleanup_(erase_and_freep
) void *volume_key
= NULL
;
1236 size_t volume_key_size
= 0;
1237 uint64_t offset
, size
;
1243 assert(user_record_storage(h
) == USER_LUKS
);
1245 r
= dlopen_cryptsetup();
1249 r
= make_dm_names(h
, setup
);
1253 /* Reuse the image fd if it has already been opened by an earlier step */
1254 if (setup
->image_fd
< 0) {
1255 setup
->image_fd
= open_image_file(h
, force_image_path
, &st
);
1256 if (setup
->image_fd
< 0)
1257 return setup
->image_fd
;
1258 } else if (fstat(setup
->image_fd
, &st
) < 0)
1259 return log_error_errno(errno
, "Failed to stat image: %m");
1261 if (FLAGS_SET(flags
, HOME_SETUP_ALREADY_ACTIVATED
)) {
1262 struct loop_info64 info
;
1265 if (!setup
->crypt_device
) {
1276 if (ret_luks_home
) {
1277 r
= luks_validate_home_record(setup
->crypt_device
, h
, volume_key
, cache
, &luks_home
);
1282 n
= sym_crypt_get_device_name(setup
->crypt_device
);
1284 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine backing device for DM %s.", setup
->dm_name
);
1287 r
= loop_device_open_from_path(n
, O_RDWR
, LOCK_UN
, &setup
->loop
);
1289 return log_error_errno(r
, "Failed to open loopback device %s: %m", n
);
1292 if (ioctl(setup
->loop
->fd
, LOOP_GET_STATUS64
, &info
) < 0) {
1293 _cleanup_free_
char *sysfs
= NULL
;
1295 if (!IN_SET(errno
, ENOTTY
, EINVAL
))
1296 return log_error_errno(errno
, "Failed to get block device metrics of %s: %m", n
);
1298 if (ioctl(setup
->loop
->fd
, BLKGETSIZE64
, &size
) < 0)
1299 return log_error_errno(r
, "Failed to read block device size of %s: %m", n
);
1301 if (fstat(setup
->loop
->fd
, &st
) < 0)
1302 return log_error_errno(r
, "Failed to stat block device %s: %m", n
);
1303 assert(S_ISBLK(st
.st_mode
));
1305 if (asprintf(&sysfs
, "/sys/dev/block/" DEVNUM_FORMAT_STR
"/partition", DEVNUM_FORMAT_VAL(st
.st_rdev
)) < 0)
1308 if (access(sysfs
, F_OK
) < 0) {
1309 if (errno
!= ENOENT
)
1310 return log_error_errno(errno
, "Failed to determine whether %s exists: %m", sysfs
);
1314 _cleanup_free_
char *buffer
= NULL
;
1316 if (asprintf(&sysfs
, "/sys/dev/block/" DEVNUM_FORMAT_STR
"/start", DEVNUM_FORMAT_VAL(st
.st_rdev
)) < 0)
1319 r
= read_one_line_file(sysfs
, &buffer
);
1321 return log_error_errno(r
, "Failed to read partition start offset: %m");
1323 r
= safe_atou64(buffer
, &offset
);
1325 return log_error_errno(r
, "Failed to parse partition start offset: %m");
1327 if (offset
> UINT64_MAX
/ 512U)
1328 return log_error_errno(SYNTHETIC_ERRNO(E2BIG
), "Offset too large for 64 byte range, refusing.");
1333 #if HAVE_VALGRIND_MEMCHECK_H
1334 VALGRIND_MAKE_MEM_DEFINED(&info
, sizeof(info
));
1337 offset
= info
.lo_offset
;
1338 size
= info
.lo_sizelimit
;
1341 found_partition_uuid
= found_fs_uuid
= SD_ID128_NULL
;
1343 log_info("Discovered used loopback device %s.", setup
->loop
->node
);
1345 if (setup
->root_fd
< 0) {
1346 setup
->root_fd
= open(user_record_home_directory(h
), O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
|O_NOFOLLOW
);
1347 if (setup
->root_fd
< 0)
1348 return log_error_errno(errno
, "Failed to open home directory: %m");
1351 _cleanup_free_
char *fstype
= NULL
, *subdir
= NULL
;
1354 /* When we aren't reopening the home directory we are allocating it fresh, hence the relevant
1355 * objects can't be allocated yet. */
1356 assert(setup
->root_fd
< 0);
1357 assert(!setup
->crypt_device
);
1358 assert(!setup
->loop
);
1360 ip
= force_image_path
?: user_record_image_path(h
);
1362 subdir
= path_join(HOME_RUNTIME_WORK_DIR
, user_record_user_name_and_realm(h
));
1366 r
= luks_validate(setup
->image_fd
, user_record_user_name_and_realm(h
), h
->partition_uuid
, &found_partition_uuid
, &offset
, &size
);
1368 return log_error_errno(r
, "Failed to validate disk label: %m");
1370 /* Everything before this point left the image untouched. We are now starting to make
1371 * changes, hence mark the image dirty */
1372 if (run_mark_dirty(setup
->image_fd
, true) > 0)
1373 setup
->do_mark_clean
= true;
1375 if (!user_record_luks_discard(h
)) {
1376 r
= run_fallocate(setup
->image_fd
, &st
);
1381 r
= loop_device_make(setup
->image_fd
, O_RDWR
, offset
, size
, user_record_luks_sector_size(h
), 0, LOCK_UN
, &setup
->loop
);
1383 log_error_errno(r
, "Loopback block device support is not available on this system.");
1384 return -ENOLINK
; /* make recognizable */
1387 return log_error_errno(r
, "Failed to allocate loopback context: %m");
1389 log_info("Setting up loopback device %s completed.", setup
->loop
->node
?: ip
);
1392 setup
->loop
->node
?: ip
,
1396 h
->luks_cipher_mode
,
1397 h
->luks_volume_key_size
,
1400 user_record_luks_discard(h
) || user_record_luks_offline_discard(h
),
1401 &setup
->crypt_device
,
1405 &setup
->key_serial
);
1409 setup
->undo_dm
= true;
1411 if (ret_luks_home
) {
1412 r
= luks_validate_home_record(setup
->crypt_device
, h
, volume_key
, cache
, &luks_home
);
1417 r
= fs_validate(setup
->dm_node
, h
->file_system_uuid
, &fstype
, &found_fs_uuid
);
1421 r
= run_fsck(setup
->dm_node
, fstype
);
1425 r
= home_unshare_and_mount(setup
->dm_node
, fstype
, user_record_luks_discard(h
), user_record_mount_flags(h
), h
->luks_extra_mount_options
);
1429 setup
->undo_mount
= true;
1431 setup
->root_fd
= open(subdir
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
|O_NOFOLLOW
);
1432 if (setup
->root_fd
< 0)
1433 return log_error_errno(errno
, "Failed to open home directory: %m");
1435 if (user_record_luks_discard(h
))
1436 (void) run_fitrim(setup
->root_fd
);
1438 setup
->do_offline_fallocate
= !(setup
->do_offline_fitrim
= user_record_luks_offline_discard(h
));
1441 if (!sd_id128_is_null(found_partition_uuid
))
1442 setup
->found_partition_uuid
= found_partition_uuid
;
1443 if (!sd_id128_is_null(found_luks_uuid
))
1444 setup
->found_luks_uuid
= found_luks_uuid
;
1445 if (!sd_id128_is_null(found_fs_uuid
))
1446 setup
->found_fs_uuid
= found_fs_uuid
;
1448 setup
->partition_offset
= offset
;
1449 setup
->partition_size
= size
;
1452 erase_and_free(setup
->volume_key
);
1453 setup
->volume_key
= TAKE_PTR(volume_key
);
1454 setup
->volume_key_size
= volume_key_size
;
1458 *ret_luks_home
= TAKE_PTR(luks_home
);
1463 static void print_size_summary(uint64_t host_size
, uint64_t encrypted_size
, const struct statfs
*sfs
) {
1466 log_info("Image size is %s, file system size is %s, file system payload size is %s, file system free is %s.",
1467 FORMAT_BYTES(host_size
),
1468 FORMAT_BYTES(encrypted_size
),
1469 FORMAT_BYTES((uint64_t) sfs
->f_blocks
* (uint64_t) sfs
->f_frsize
),
1470 FORMAT_BYTES((uint64_t) sfs
->f_bfree
* (uint64_t) sfs
->f_frsize
));
1473 static int home_auto_grow_luks(
1476 PasswordCache
*cache
) {
1483 if (!IN_SET(user_record_auto_resize_mode(h
), AUTO_RESIZE_GROW
, AUTO_RESIZE_SHRINK_AND_GROW
))
1486 assert(setup
->root_fd
>= 0);
1488 if (fstatfs(setup
->root_fd
, &sfs
) < 0)
1489 return log_error_errno(errno
, "Failed to statfs home directory: %m");
1491 if (!fs_can_online_shrink_and_grow(sfs
.f_type
)) {
1492 log_debug("Not auto-grow file system, since selected file system cannot do both online shrink and grow.");
1496 log_debug("Initiating auto-grow...");
1498 return home_resize_luks(
1500 HOME_SETUP_ALREADY_ACTIVATED
|
1501 HOME_SETUP_RESIZE_DONT_SYNC_IDENTITIES
|
1502 HOME_SETUP_RESIZE_DONT_SHRINK
|
1503 HOME_SETUP_RESIZE_DONT_UNDO
,
1509 int home_activate_luks(
1511 HomeSetupFlags flags
,
1513 PasswordCache
*cache
,
1514 UserRecord
**ret_home
) {
1516 _cleanup_(user_record_unrefp
) UserRecord
*new_home
= NULL
, *luks_home_record
= NULL
;
1517 uint64_t host_size
, encrypted_size
;
1518 const char *hdo
, *hd
;
1523 assert(user_record_storage(h
) == USER_LUKS
);
1527 r
= dlopen_cryptsetup();
1531 assert_se(hdo
= user_record_home_directory(h
));
1532 hd
= strdupa_safe(hdo
); /* copy the string out, since it might change later in the home record object */
1534 r
= home_get_state_luks(h
, setup
);
1538 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
), "Device mapper device %s already exists, refusing.", setup
->dm_node
);
1540 r
= home_setup_luks(
1550 r
= home_auto_grow_luks(h
, setup
, cache
);
1554 r
= block_get_size_by_fd(setup
->loop
->fd
, &host_size
);
1556 return log_error_errno(r
, "Failed to get loopback block device size: %m");
1558 r
= block_get_size_by_path(setup
->dm_node
, &encrypted_size
);
1560 return log_error_errno(r
, "Failed to get LUKS block device size: %m");
1573 r
= home_extend_embedded_identity(new_home
, h
, setup
);
1577 setup
->root_fd
= safe_close(setup
->root_fd
);
1579 r
= home_move_mount(user_record_user_name_and_realm(h
), hd
);
1583 setup
->undo_mount
= false;
1584 setup
->do_offline_fitrim
= false;
1586 loop_device_relinquish(setup
->loop
);
1588 r
= sym_crypt_deactivate_by_name(NULL
, setup
->dm_name
, CRYPT_DEACTIVATE_DEFERRED
);
1590 log_warning_errno(r
, "Failed to relinquish DM device, ignoring: %m");
1592 setup
->undo_dm
= false;
1593 setup
->do_offline_fallocate
= false;
1594 setup
->do_mark_clean
= false;
1595 setup
->do_drop_caches
= false;
1596 TAKE_KEY_SERIAL(setup
->key_serial
); /* Leave key in kernel keyring */
1598 log_info("Activation completed.");
1600 print_size_summary(host_size
, encrypted_size
, &sfs
);
1602 *ret_home
= TAKE_PTR(new_home
);
1606 int home_deactivate_luks(UserRecord
*h
, HomeSetup
*setup
) {
1607 bool we_detached
= false;
1613 /* Note that the DM device and loopback device are set to auto-detach, hence strictly speaking we
1614 * don't have to explicitly have to detach them. However, we do that nonetheless (in case of the DM
1615 * device), to avoid races: by explicitly detaching them we know when the detaching is complete. We
1616 * don't bother about the loopback device because unlike the DM device it doesn't have a fixed
1619 if (!setup
->crypt_device
) {
1620 r
= acquire_open_luks_device(h
, setup
, /* graceful= */ true);
1622 return log_error_errno(r
, "Failed to initialize cryptsetup context for %s: %m", setup
->dm_name
);
1624 log_debug("LUKS device %s has already been detached.", setup
->dm_name
);
1627 if (setup
->crypt_device
) {
1628 log_info("Discovered used LUKS device %s.", setup
->dm_node
);
1630 cryptsetup_enable_logging(setup
->crypt_device
);
1632 r
= sym_crypt_deactivate_by_name(setup
->crypt_device
, setup
->dm_name
, 0);
1633 if (ERRNO_IS_DEVICE_ABSENT(r
) || r
== -EINVAL
)
1634 log_debug_errno(r
, "LUKS device %s is already detached.", setup
->dm_node
);
1636 return log_info_errno(r
, "LUKS device %s couldn't be deactivated: %m", setup
->dm_node
);
1638 log_info("LUKS device detaching completed.");
1643 (void) wait_for_block_device_gone(setup
, USEC_PER_SEC
* 30);
1644 setup
->undo_dm
= false;
1646 if (user_record_luks_offline_discard(h
))
1647 log_debug("Not allocating on logout.");
1649 (void) run_fallocate_by_path(user_record_image_path(h
));
1651 run_mark_dirty_by_path(user_record_image_path(h
), false);
1655 int home_trim_luks(UserRecord
*h
, HomeSetup
*setup
) {
1658 assert(setup
->root_fd
>= 0);
1660 if (!user_record_luks_offline_discard(h
)) {
1661 log_debug("Not trimming on logout.");
1665 (void) run_fitrim(setup
->root_fd
);
1669 static struct crypt_pbkdf_type
* build_good_pbkdf(struct crypt_pbkdf_type
*buffer
, UserRecord
*hr
) {
1673 *buffer
= (struct crypt_pbkdf_type
) {
1674 .hash
= user_record_luks_pbkdf_hash_algorithm(hr
),
1675 .type
= user_record_luks_pbkdf_type(hr
),
1676 .time_ms
= user_record_luks_pbkdf_time_cost_usec(hr
) / USEC_PER_MSEC
,
1677 .max_memory_kb
= user_record_luks_pbkdf_memory_cost(hr
) / 1024,
1678 .parallel_threads
= user_record_luks_pbkdf_parallel_threads(hr
),
1684 static struct crypt_pbkdf_type
* build_minimal_pbkdf(struct crypt_pbkdf_type
*buffer
, UserRecord
*hr
) {
1688 /* For PKCS#11 derived keys (which are generated randomly and are of high quality already) we use a
1690 *buffer
= (struct crypt_pbkdf_type
) {
1691 .hash
= user_record_luks_pbkdf_hash_algorithm(hr
),
1692 .type
= CRYPT_KDF_PBKDF2
,
1700 static int luks_format(
1702 const char *dm_name
,
1705 const PasswordCache
*cache
,
1706 char **effective_passwords
,
1709 struct crypt_device
**ret
) {
1711 _cleanup_(user_record_unrefp
) UserRecord
*reduced
= NULL
;
1712 _cleanup_(sym_crypt_freep
) struct crypt_device
*cd
= NULL
;
1713 _cleanup_(erase_and_freep
) void *volume_key
= NULL
;
1714 struct crypt_pbkdf_type good_pbkdf
, minimal_pbkdf
;
1715 _cleanup_free_
char *text
= NULL
;
1716 size_t volume_key_size
;
1724 r
= sym_crypt_init(&cd
, node
);
1726 return log_error_errno(r
, "Failed to allocate libcryptsetup context: %m");
1728 cryptsetup_enable_logging(cd
);
1730 /* Normally we'd, just leave volume key generation to libcryptsetup. However, we can't, since we
1731 * can't extract the volume key from the library again, but we need it in order to encrypt the JSON
1732 * record. Hence, let's generate it on our own, so that we can keep track of it. */
1734 volume_key_size
= user_record_luks_volume_key_size(hr
);
1735 volume_key
= malloc(volume_key_size
);
1739 r
= crypto_random_bytes(volume_key
, volume_key_size
);
1741 return log_error_errno(r
, "Failed to generate volume key: %m");
1743 #if HAVE_CRYPT_SET_METADATA_SIZE
1744 /* Increase the metadata space to 4M, the largest LUKS2 supports */
1745 r
= sym_crypt_set_metadata_size(cd
, 4096U*1024U, 0);
1747 return log_error_errno(r
, "Failed to change LUKS2 metadata size: %m");
1750 build_good_pbkdf(&good_pbkdf
, hr
);
1751 build_minimal_pbkdf(&minimal_pbkdf
, hr
);
1753 r
= sym_crypt_format(
1756 user_record_luks_cipher(hr
),
1757 user_record_luks_cipher_mode(hr
),
1758 SD_ID128_TO_UUID_STRING(uuid
),
1761 &(struct crypt_params_luks2
) {
1763 .subsystem
= "systemd-home",
1764 .sector_size
= user_record_luks_sector_size(hr
),
1765 .pbkdf
= &good_pbkdf
,
1768 return log_error_errno(r
, "Failed to format LUKS image: %m");
1770 log_info("LUKS formatting completed.");
1772 STRV_FOREACH(pp
, effective_passwords
) {
1774 if (password_cache_contains(cache
, *pp
)) { /* is this a fido2 or pkcs11 password? */
1775 log_debug("Using minimal PBKDF for slot %i", slot
);
1776 r
= sym_crypt_set_pbkdf_type(cd
, &minimal_pbkdf
);
1778 log_debug("Using good PBKDF for slot %i", slot
);
1779 r
= sym_crypt_set_pbkdf_type(cd
, &good_pbkdf
);
1782 return log_error_errno(r
, "Failed to tweak PBKDF for slot %i: %m", slot
);
1784 r
= sym_crypt_keyslot_add_by_volume_key(
1792 return log_error_errno(r
, "Failed to set up LUKS password for slot %i: %m", slot
);
1794 log_info("Writing password to LUKS keyslot %i completed.", slot
);
1798 r
= sym_crypt_activate_by_volume_key(
1803 discard
? CRYPT_ACTIVATE_ALLOW_DISCARDS
: 0);
1805 return log_error_errno(r
, "Failed to activate LUKS superblock: %m");
1807 log_info("LUKS activation by volume key succeeded.");
1809 r
= user_record_clone(hr
, USER_RECORD_EXTRACT_EMBEDDED
|USER_RECORD_PERMISSIVE
, &reduced
);
1811 return log_error_errno(r
, "Failed to prepare home record for LUKS: %m");
1813 r
= format_luks_token_text(cd
, reduced
, volume_key
, &text
);
1817 r
= sym_crypt_token_json_set(cd
, CRYPT_ANY_TOKEN
, text
);
1819 return log_error_errno(r
, "Failed to set LUKS JSON token: %m");
1821 log_info("Writing user record as LUKS token completed.");
1824 *ret
= TAKE_PTR(cd
);
1829 static int make_partition_table(
1833 uint64_t *ret_offset
,
1835 sd_id128_t
*ret_disk_uuid
) {
1837 _cleanup_(fdisk_unref_partitionp
) struct fdisk_partition
*p
= NULL
, *q
= NULL
;
1838 _cleanup_(fdisk_unref_parttypep
) struct fdisk_parttype
*t
= NULL
;
1839 _cleanup_(fdisk_unref_contextp
) struct fdisk_context
*c
= NULL
;
1840 _cleanup_free_
char *path
= NULL
, *disk_uuid_as_string
= NULL
;
1841 uint64_t offset
, size
, first_lba
, start
, last_lba
, end
;
1842 sd_id128_t disk_uuid
;
1850 t
= fdisk_new_parttype();
1854 r
= fdisk_parttype_set_typestr(t
, SD_GPT_USER_HOME_STR
);
1856 return log_error_errno(r
, "Failed to initialize partition type: %m");
1858 c
= fdisk_new_context();
1862 if (asprintf(&path
, "/proc/self/fd/%i", fd
) < 0)
1865 r
= fdisk_assign_device(c
, path
, 0);
1867 return log_error_errno(r
, "Failed to open device: %m");
1869 r
= fdisk_create_disklabel(c
, "gpt");
1871 return log_error_errno(r
, "Failed to create GPT disk label: %m");
1873 p
= fdisk_new_partition();
1877 r
= fdisk_partition_set_type(p
, t
);
1879 return log_error_errno(r
, "Failed to set partition type: %m");
1881 r
= fdisk_partition_partno_follow_default(p
, 1);
1883 return log_error_errno(r
, "Failed to place partition at first free partition index: %m");
1885 first_lba
= fdisk_get_first_lba(c
); /* Boundary where usable space starts */
1886 assert(first_lba
<= UINT64_MAX
/512);
1887 start
= DISK_SIZE_ROUND_UP(first_lba
* 512); /* Round up to multiple of 4K */
1889 log_debug("Starting partition at offset %" PRIu64
, start
);
1891 if (start
== UINT64_MAX
)
1892 return log_error_errno(SYNTHETIC_ERRNO(ERANGE
), "Overflow while rounding up start LBA.");
1894 last_lba
= fdisk_get_last_lba(c
); /* One sector before boundary where usable space ends */
1895 assert(last_lba
< UINT64_MAX
/512);
1896 end
= DISK_SIZE_ROUND_DOWN((last_lba
+ 1) * 512); /* Round down to multiple of 4K */
1899 return log_error_errno(SYNTHETIC_ERRNO(ERANGE
), "Resulting partition size zero or negative.");
1901 r
= fdisk_partition_set_start(p
, start
/ 512);
1903 return log_error_errno(r
, "Failed to place partition at offset %" PRIu64
": %m", start
);
1905 r
= fdisk_partition_set_size(p
, (end
- start
) / 512);
1907 return log_error_errno(r
, "Failed to end partition at offset %" PRIu64
": %m", end
);
1909 r
= fdisk_partition_set_name(p
, label
);
1911 return log_error_errno(r
, "Failed to set partition name: %m");
1913 r
= fdisk_partition_set_uuid(p
, SD_ID128_TO_UUID_STRING(uuid
));
1915 return log_error_errno(r
, "Failed to set partition UUID: %m");
1917 r
= fdisk_add_partition(c
, p
, NULL
);
1919 return log_error_errno(r
, "Failed to add partition: %m");
1921 r
= fdisk_write_disklabel(c
);
1923 return log_error_errno(r
, "Failed to write disk label: %m");
1925 r
= fdisk_get_disklabel_id(c
, &disk_uuid_as_string
);
1927 return log_error_errno(r
, "Failed to determine disk label UUID: %m");
1929 r
= sd_id128_from_string(disk_uuid_as_string
, &disk_uuid
);
1931 return log_error_errno(r
, "Failed to parse disk label UUID: %m");
1933 r
= fdisk_get_partition(c
, 0, &q
);
1935 return log_error_errno(r
, "Failed to read created partition metadata: %m");
1937 assert(fdisk_partition_has_start(q
));
1938 offset
= fdisk_partition_get_start(q
);
1939 if (offset
> UINT64_MAX
/ 512U)
1940 return log_error_errno(SYNTHETIC_ERRNO(ERANGE
), "Partition offset too large.");
1942 assert(fdisk_partition_has_size(q
));
1943 size
= fdisk_partition_get_size(q
);
1944 if (size
> UINT64_MAX
/ 512U)
1945 return log_error_errno(SYNTHETIC_ERRNO(ERANGE
), "Partition size too large.");
1947 *ret_offset
= offset
* 512U;
1948 *ret_size
= size
* 512U;
1949 *ret_disk_uuid
= disk_uuid
;
1954 static bool supported_fs_size(const char *fstype
, uint64_t host_size
) {
1957 m
= minimal_size_by_fs_name(fstype
);
1958 if (m
== UINT64_MAX
)
1961 return host_size
>= m
;
1964 static int wait_for_devlink(const char *path
) {
1965 _cleanup_close_
int inotify_fd
= -1;
1969 /* let's wait for a device link to show up in /dev, with a timeout. This is good to do since we
1970 * return a /dev/disk/by-uuid/… link to our callers and they likely want to access it right-away,
1971 * hence let's wait until udev has caught up with our changes, and wait for the symlink to be
1974 until
= usec_add(now(CLOCK_MONOTONIC
), 45 * USEC_PER_SEC
);
1977 _cleanup_free_
char *dn
= NULL
;
1980 if (laccess(path
, F_OK
) < 0) {
1981 if (errno
!= ENOENT
)
1982 return log_error_errno(errno
, "Failed to determine whether %s exists: %m", path
);
1984 return 0; /* Found it */
1986 if (inotify_fd
< 0) {
1987 /* We need to wait for the device symlink to show up, let's create an inotify watch for it */
1988 inotify_fd
= inotify_init1(IN_NONBLOCK
|IN_CLOEXEC
);
1990 return log_error_errno(errno
, "Failed to allocate inotify fd: %m");
1993 r
= path_extract_directory(path
, &dn
);
1995 return log_error_errno(r
, "Failed to extract directory from device node path '%s': %m", path
);
1997 _cleanup_free_
char *ndn
= NULL
;
1999 log_info("Watching %s", dn
);
2001 if (inotify_add_watch(inotify_fd
, dn
, IN_CREATE
|IN_MOVED_TO
|IN_ONLYDIR
|IN_DELETE_SELF
|IN_MOVE_SELF
) < 0) {
2002 if (errno
!= ENOENT
)
2003 return log_error_errno(errno
, "Failed to add watch on %s: %m", dn
);
2007 r
= path_extract_directory(dn
, &ndn
);
2008 if (r
== -EADDRNOTAVAIL
) /* Arrived at the top? */
2011 return log_error_errno(r
, "Failed to extract directory from device node path '%s': %m", dn
);
2013 free_and_replace(dn
, ndn
);
2016 w
= now(CLOCK_MONOTONIC
);
2018 return log_error_errno(SYNTHETIC_ERRNO(ETIMEDOUT
), "Device link %s still hasn't shown up, giving up.", path
);
2020 r
= fd_wait_for_event(inotify_fd
, POLLIN
, until
- w
);
2022 if (ERRNO_IS_TRANSIENT(r
))
2024 return log_error_errno(r
, "Failed to watch inotify: %m");
2027 (void) flush_fd(inotify_fd
);
2031 static int calculate_initial_image_size(UserRecord
*h
, int image_fd
, const char *fstype
, uint64_t *ret
) {
2032 uint64_t upper_boundary
, lower_boundary
;
2036 assert(image_fd
>= 0);
2039 if (fstatfs(image_fd
, &sfs
) < 0)
2040 return log_error_errno(errno
, "statfs() on image failed: %m");
2042 upper_boundary
= DISK_SIZE_ROUND_DOWN((uint64_t) sfs
.f_bsize
* sfs
.f_bavail
);
2044 if (h
->disk_size
!= UINT64_MAX
)
2045 *ret
= MIN(DISK_SIZE_ROUND_DOWN(h
->disk_size
), upper_boundary
);
2046 else if (h
->disk_size_relative
== UINT64_MAX
) {
2048 if (upper_boundary
> UINT64_MAX
/ USER_DISK_SIZE_DEFAULT_PERCENT
)
2049 return log_error_errno(SYNTHETIC_ERRNO(EOVERFLOW
), "Disk size too large.");
2051 *ret
= DISK_SIZE_ROUND_DOWN(upper_boundary
* USER_DISK_SIZE_DEFAULT_PERCENT
/ 100);
2053 log_info("Sizing home to %u%% of available disk space, which is %s.",
2054 USER_DISK_SIZE_DEFAULT_PERCENT
,
2055 FORMAT_BYTES(*ret
));
2057 *ret
= DISK_SIZE_ROUND_DOWN((uint64_t) ((double) upper_boundary
* (double) CLAMP(h
->disk_size_relative
, 0U, UINT32_MAX
) / (double) UINT32_MAX
));
2059 log_info("Sizing home to %" PRIu64
".%01" PRIu64
"%% of available disk space, which is %s.",
2060 (h
->disk_size_relative
* 100) / UINT32_MAX
,
2061 ((h
->disk_size_relative
* 1000) / UINT32_MAX
) % 10,
2062 FORMAT_BYTES(*ret
));
2065 lower_boundary
= minimal_size_by_fs_name(fstype
);
2066 if (lower_boundary
!= UINT64_MAX
) {
2067 assert(GPT_LUKS2_OVERHEAD
< UINT64_MAX
- lower_boundary
);
2068 lower_boundary
+= GPT_LUKS2_OVERHEAD
;
2070 if (lower_boundary
== UINT64_MAX
|| lower_boundary
< USER_DISK_SIZE_MIN
)
2071 lower_boundary
= USER_DISK_SIZE_MIN
;
2073 if (*ret
< lower_boundary
)
2074 *ret
= lower_boundary
;
2079 static int home_truncate(
2090 trunc
= user_record_luks_discard(h
);
2092 r
= fallocate(fd
, 0, 0, size
);
2093 if (r
< 0 && ERRNO_IS_NOT_SUPPORTED(errno
)) {
2094 /* Some file systems do not support fallocate(), let's gracefully degrade
2095 * (ZFS, reiserfs, …) and fall back to truncation */
2096 log_notice_errno(errno
, "Backing file system does not support fallocate(), falling back to ftruncate(), i.e. implicitly using non-discard mode.");
2102 r
= ftruncate(fd
, size
);
2105 if (ERRNO_IS_DISK_SPACE(errno
)) {
2106 log_debug_errno(errno
, "Not enough disk space to allocate home of size %s.", FORMAT_BYTES(size
));
2107 return -ENOSPC
; /* make recognizable */
2110 return log_error_errno(errno
, "Failed to truncate home image: %m");
2113 return !trunc
; /* Return == 0 if we managed to truncate, > 0 if we managed to allocate */
2116 int home_create_luks(
2119 const PasswordCache
*cache
,
2120 char **effective_passwords
,
2121 UserRecord
**ret_home
) {
2123 _cleanup_free_
char *subdir
= NULL
, *disk_uuid_path
= NULL
;
2124 uint64_t encrypted_size
,
2125 host_size
= 0, partition_offset
= 0, partition_size
= 0; /* Unnecessary initialization to appease gcc */
2126 _cleanup_(user_record_unrefp
) UserRecord
*new_home
= NULL
;
2127 sd_id128_t partition_uuid
, fs_uuid
, luks_uuid
, disk_uuid
;
2128 _cleanup_close_
int mount_fd
= -1;
2129 const char *fstype
, *ip
;
2134 assert(h
->storage
< 0 || h
->storage
== USER_LUKS
);
2136 assert(!setup
->temporary_image_path
);
2137 assert(setup
->image_fd
< 0);
2140 r
= dlopen_cryptsetup();
2144 assert_se(ip
= user_record_image_path(h
));
2146 fstype
= user_record_file_system_type(h
);
2147 if (!supported_fstype(fstype
))
2148 return log_error_errno(SYNTHETIC_ERRNO(EPROTONOSUPPORT
), "Unsupported file system type: %s", fstype
);
2150 r
= mkfs_exists(fstype
);
2152 return log_error_errno(r
, "Failed to check if mkfs binary for %s exists: %m", fstype
);
2154 if (h
->file_system_type
|| streq(fstype
, "ext4") || !supported_fstype("ext4"))
2155 return log_error_errno(SYNTHETIC_ERRNO(EPROTONOSUPPORT
), "mkfs binary for file system type %s does not exist.", fstype
);
2157 /* If the record does not explicitly declare a file system to use, and the compiled-in
2158 * default does not actually exist, than do an automatic fallback onto ext4, as the baseline
2159 * fs of Linux. We won't search for a working fs type here beyond ext4, i.e. nothing fancier
2160 * than a single, conservative fallback to baseline. This should be useful in minimal
2161 * environments where mkfs.btrfs or so are not made available, but mkfs.ext4 as Linux' most
2162 * boring, most basic fs is. */
2163 log_info("Formatting tool for compiled-in default file system %s not available, falling back to ext4 instead.", fstype
);
2167 if (sd_id128_is_null(h
->partition_uuid
)) {
2168 r
= sd_id128_randomize(&partition_uuid
);
2170 return log_error_errno(r
, "Failed to acquire partition UUID: %m");
2172 partition_uuid
= h
->partition_uuid
;
2174 if (sd_id128_is_null(h
->luks_uuid
)) {
2175 r
= sd_id128_randomize(&luks_uuid
);
2177 return log_error_errno(r
, "Failed to acquire LUKS UUID: %m");
2179 luks_uuid
= h
->luks_uuid
;
2181 if (sd_id128_is_null(h
->file_system_uuid
)) {
2182 r
= sd_id128_randomize(&fs_uuid
);
2184 return log_error_errno(r
, "Failed to acquire file system UUID: %m");
2186 fs_uuid
= h
->file_system_uuid
;
2188 r
= make_dm_names(h
, setup
);
2192 r
= access(setup
->dm_node
, F_OK
);
2194 if (errno
!= ENOENT
)
2195 return log_error_errno(errno
, "Failed to determine whether %s exists: %m", setup
->dm_node
);
2197 return log_error_errno(SYNTHETIC_ERRNO(EEXIST
), "Device mapper device %s already exists, refusing.", setup
->dm_node
);
2199 if (path_startswith(ip
, "/dev/")) {
2200 _cleanup_free_
char *sysfs
= NULL
;
2201 uint64_t block_device_size
;
2204 /* Let's place the home directory on a real device, i.e. an USB stick or such */
2206 setup
->image_fd
= open_image_file(h
, ip
, &st
);
2207 if (setup
->image_fd
< 0)
2208 return setup
->image_fd
;
2210 if (!S_ISBLK(st
.st_mode
))
2211 return log_error_errno(SYNTHETIC_ERRNO(ENOTBLK
), "Device is not a block device, refusing.");
2213 if (asprintf(&sysfs
, "/sys/dev/block/" DEVNUM_FORMAT_STR
"/partition", DEVNUM_FORMAT_VAL(st
.st_rdev
)) < 0)
2215 if (access(sysfs
, F_OK
) < 0) {
2216 if (errno
!= ENOENT
)
2217 return log_error_errno(errno
, "Failed to check whether %s exists: %m", sysfs
);
2219 return log_error_errno(SYNTHETIC_ERRNO(ENOTBLK
), "Operating on partitions is currently not supported, sorry. Please specify a top-level block device.");
2221 if (flock(setup
->image_fd
, LOCK_EX
) < 0) /* make sure udev doesn't read from it while we operate on the device */
2222 return log_error_errno(errno
, "Failed to lock block device %s: %m", ip
);
2224 if (ioctl(setup
->image_fd
, BLKGETSIZE64
, &block_device_size
) < 0)
2225 return log_error_errno(errno
, "Failed to read block device size: %m");
2227 if (h
->disk_size
== UINT64_MAX
) {
2229 /* If a relative disk size is requested, apply it relative to the block device size */
2230 if (h
->disk_size_relative
< UINT32_MAX
)
2231 host_size
= CLAMP(DISK_SIZE_ROUND_DOWN(block_device_size
* h
->disk_size_relative
/ UINT32_MAX
),
2232 USER_DISK_SIZE_MIN
, USER_DISK_SIZE_MAX
);
2234 host_size
= block_device_size
; /* Otherwise, take the full device */
2236 } else if (h
->disk_size
> block_device_size
)
2237 return log_error_errno(SYNTHETIC_ERRNO(EMSGSIZE
), "Selected disk size larger than backing block device, refusing.");
2239 host_size
= DISK_SIZE_ROUND_DOWN(h
->disk_size
);
2241 if (!supported_fs_size(fstype
, LESS_BY(host_size
, GPT_LUKS2_OVERHEAD
)))
2242 return log_error_errno(SYNTHETIC_ERRNO(ERANGE
),
2243 "Selected file system size too small for %s.", fstype
);
2245 /* After creation we should reference this partition by its UUID instead of the block
2246 * device. That's preferable since the user might have specified a device node such as
2247 * /dev/sdb to us, which might look very different when replugged. */
2248 if (asprintf(&disk_uuid_path
, "/dev/disk/by-uuid/" SD_ID128_UUID_FORMAT_STR
, SD_ID128_FORMAT_VAL(luks_uuid
)) < 0)
2251 if (user_record_luks_discard(h
) || user_record_luks_offline_discard(h
)) {
2252 /* If we want online or offline discard, discard once before we start using things. */
2254 if (ioctl(setup
->image_fd
, BLKDISCARD
, (uint64_t[]) { 0, block_device_size
}) < 0)
2255 log_full_errno(errno
== EOPNOTSUPP
? LOG_DEBUG
: LOG_WARNING
, errno
,
2256 "Failed to issue full-device BLKDISCARD on device, ignoring: %m");
2258 log_info("Full device discard completed.");
2261 _cleanup_free_
char *t
= NULL
;
2263 r
= mkdir_parents(ip
, 0755);
2265 return log_error_errno(r
, "Failed to create parent directory of %s: %m", ip
);
2267 r
= tempfn_random(ip
, "homework", &t
);
2269 return log_error_errno(r
, "Failed to derive temporary file name for %s: %m", ip
);
2271 setup
->image_fd
= open(t
, O_RDWR
|O_CREAT
|O_EXCL
|O_CLOEXEC
|O_NOCTTY
|O_NOFOLLOW
, 0600);
2272 if (setup
->image_fd
< 0)
2273 return log_error_errno(errno
, "Failed to create home image %s: %m", t
);
2275 setup
->temporary_image_path
= TAKE_PTR(t
);
2277 r
= chattr_full(t
, setup
->image_fd
, FS_NOCOW_FL
|FS_NOCOMP_FL
, FS_NOCOW_FL
|FS_NOCOMP_FL
, NULL
, NULL
, CHATTR_FALLBACK_BITWISE
);
2278 if (r
< 0 && r
!= -ENOANO
) /* ENOANO → some bits didn't work; which we skip logging about because chattr_full() already debug logs about those flags */
2279 log_full_errno(ERRNO_IS_NOT_SUPPORTED(r
) ? LOG_DEBUG
: LOG_WARNING
, r
,
2280 "Failed to set file attributes on %s, ignoring: %m", setup
->temporary_image_path
);
2282 r
= calculate_initial_image_size(h
, setup
->image_fd
, fstype
, &host_size
);
2286 r
= resize_image_loop(h
, setup
, 0, host_size
, &host_size
);
2290 log_info("Allocating image file completed.");
2293 r
= make_partition_table(
2295 user_record_user_name_and_realm(h
),
2303 log_info("Writing of partition table completed.");
2305 r
= loop_device_make(setup
->image_fd
, O_RDWR
, partition_offset
, partition_size
, user_record_luks_sector_size(h
), 0, LOCK_EX
, &setup
->loop
);
2307 if (r
== -ENOENT
) { /* this means /dev/loop-control doesn't exist, i.e. we are in a container
2308 * or similar and loopback bock devices are not available, return a
2309 * recognizable error in this case. */
2310 log_error_errno(r
, "Loopback block device support is not available on this system.");
2311 return -ENOLINK
; /* Make recognizable */
2314 return log_error_errno(r
, "Failed to set up loopback device for %s: %m", setup
->temporary_image_path
);
2317 log_info("Setting up loopback device %s completed.", setup
->loop
->node
?: ip
);
2319 r
= luks_format(setup
->loop
->node
,
2322 user_record_user_name_and_realm(h
),
2324 effective_passwords
,
2325 user_record_luks_discard(h
) || user_record_luks_offline_discard(h
),
2327 &setup
->crypt_device
);
2331 setup
->undo_dm
= true;
2333 r
= block_get_size_by_path(setup
->dm_node
, &encrypted_size
);
2335 return log_error_errno(r
, "Failed to get encrypted block device size: %m");
2337 log_info("Setting up LUKS device %s completed.", setup
->dm_node
);
2339 r
= make_filesystem(setup
->dm_node
, fstype
, user_record_user_name_and_realm(h
), NULL
, fs_uuid
, user_record_luks_discard(h
));
2343 log_info("Formatting file system completed.");
2345 r
= home_unshare_and_mount(setup
->dm_node
, fstype
, user_record_luks_discard(h
), user_record_mount_flags(h
), h
->luks_extra_mount_options
);
2349 setup
->undo_mount
= true;
2351 subdir
= path_join(HOME_RUNTIME_WORK_DIR
, user_record_user_name_and_realm(h
));
2355 /* Prefer using a btrfs subvolume if we can, fall back to directory otherwise */
2356 r
= btrfs_subvol_make_fallback(subdir
, 0700);
2358 return log_error_errno(r
, "Failed to create user directory in mounted image file: %m");
2360 setup
->root_fd
= open(subdir
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
|O_NOFOLLOW
);
2361 if (setup
->root_fd
< 0)
2362 return log_error_errno(errno
, "Failed to open user directory in mounted image file: %m");
2364 (void) home_shift_uid(setup
->root_fd
, NULL
, UID_NOBODY
, h
->uid
, &mount_fd
);
2366 if (mount_fd
>= 0) {
2367 /* If we have established a new mount, then we can use that as new root fd to our home directory. */
2368 safe_close(setup
->root_fd
);
2370 setup
->root_fd
= fd_reopen(mount_fd
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
);
2371 if (setup
->root_fd
< 0)
2372 return log_error_errno(setup
->root_fd
, "Unable to convert mount fd into proper directory fd: %m");
2374 mount_fd
= safe_close(mount_fd
);
2377 r
= home_populate(h
, setup
->root_fd
);
2381 r
= home_sync_and_statfs(setup
->root_fd
, &sfs
);
2385 r
= user_record_clone(h
, USER_RECORD_LOAD_MASK_SECRET
|USER_RECORD_LOG
|USER_RECORD_PERMISSIVE
, &new_home
);
2387 return log_error_errno(r
, "Failed to clone record: %m");
2389 r
= user_record_add_binding(
2392 disk_uuid_path
?: ip
,
2396 sym_crypt_get_cipher(setup
->crypt_device
),
2397 sym_crypt_get_cipher_mode(setup
->crypt_device
),
2398 luks_volume_key_size_convert(setup
->crypt_device
),
2404 return log_error_errno(r
, "Failed to add binding to record: %m");
2406 if (user_record_luks_offline_discard(h
)) {
2407 r
= run_fitrim(setup
->root_fd
);
2412 setup
->root_fd
= safe_close(setup
->root_fd
);
2414 r
= home_setup_undo_mount(setup
, LOG_ERR
);
2418 r
= home_setup_undo_dm(setup
, LOG_ERR
);
2422 setup
->loop
= loop_device_unref(setup
->loop
);
2424 if (!user_record_luks_offline_discard(h
)) {
2425 r
= run_fallocate(setup
->image_fd
, NULL
/* refresh stat() data */);
2430 /* Sync everything to disk before we move things into place under the final name. */
2431 if (fsync(setup
->image_fd
) < 0)
2432 return log_error_errno(r
, "Failed to synchronize image to disk: %m");
2435 /* Reread partition table if this is a block device */
2436 (void) ioctl(setup
->image_fd
, BLKRRPART
, 0);
2438 assert(setup
->temporary_image_path
);
2440 if (rename(setup
->temporary_image_path
, ip
) < 0)
2441 return log_error_errno(errno
, "Failed to rename image file: %m");
2443 setup
->temporary_image_path
= mfree(setup
->temporary_image_path
);
2445 /* If we operate on a file, sync the containing directory too. */
2446 r
= fsync_directory_of_file(setup
->image_fd
);
2448 return log_error_errno(r
, "Failed to synchronize directory of image file to disk: %m");
2450 log_info("Moved image file into place.");
2453 /* Let's close the image fd now. If we are operating on a real block device this will release the BSD
2454 * lock that ensures udev doesn't interfere with what we are doing */
2455 setup
->image_fd
= safe_close(setup
->image_fd
);
2458 (void) wait_for_devlink(disk_uuid_path
);
2460 log_info("Creation completed.");
2462 print_size_summary(host_size
, encrypted_size
, &sfs
);
2464 log_debug("GPT + LUKS2 overhead is %" PRIu64
" (expected %" PRIu64
")", host_size
- encrypted_size
, GPT_LUKS2_OVERHEAD
);
2466 *ret_home
= TAKE_PTR(new_home
);
2470 int home_get_state_luks(UserRecord
*h
, HomeSetup
*setup
) {
2476 r
= make_dm_names(h
, setup
);
2480 r
= access(setup
->dm_node
, F_OK
);
2481 if (r
< 0 && errno
!= ENOENT
)
2482 return log_error_errno(errno
, "Failed to determine whether %s exists: %m", setup
->dm_node
);
2492 static int can_resize_fs(int fd
, uint64_t old_size
, uint64_t new_size
) {
2497 /* Filter out bogus requests early */
2498 if (old_size
== 0 || old_size
== UINT64_MAX
||
2499 new_size
== 0 || new_size
== UINT64_MAX
)
2500 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Invalid resize parameters.");
2502 if ((old_size
& 511) != 0 || (new_size
& 511) != 0)
2503 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Resize parameters not multiple of 512.");
2505 if (fstatfs(fd
, &sfs
) < 0)
2506 return log_error_errno(errno
, "Failed to fstatfs() file system: %m");
2508 if (is_fs_type(&sfs
, BTRFS_SUPER_MAGIC
)) {
2510 if (new_size
< BTRFS_MINIMAL_SIZE
)
2511 return log_error_errno(SYNTHETIC_ERRNO(ERANGE
), "New file system size too small for btrfs (needs to be 256M at least.");
2513 /* btrfs can grow and shrink online */
2515 } else if (is_fs_type(&sfs
, XFS_SB_MAGIC
)) {
2517 if (new_size
< XFS_MINIMAL_SIZE
)
2518 return log_error_errno(SYNTHETIC_ERRNO(ERANGE
), "New file system size too small for xfs (needs to be 14M at least).");
2520 /* XFS can grow, but not shrink */
2521 if (new_size
< old_size
)
2522 return log_error_errno(SYNTHETIC_ERRNO(EMSGSIZE
), "Shrinking this type of file system is not supported.");
2524 } else if (is_fs_type(&sfs
, EXT4_SUPER_MAGIC
)) {
2526 if (new_size
< EXT4_MINIMAL_SIZE
)
2527 return log_error_errno(SYNTHETIC_ERRNO(ERANGE
), "New file system size too small for ext4 (needs to be 1M at least).");
2529 /* ext4 can grow online, and shrink offline */
2530 if (new_size
< old_size
)
2531 return CAN_RESIZE_OFFLINE
;
2534 return log_error_errno(SYNTHETIC_ERRNO(ESOCKTNOSUPPORT
), "Resizing this type of file system is not supported.");
2536 return CAN_RESIZE_ONLINE
;
2539 static int ext4_offline_resize_fs(
2543 unsigned long flags
,
2544 const char *extra_mount_options
) {
2546 _cleanup_free_
char *size_str
= NULL
;
2547 bool re_open
= false, re_mount
= false;
2548 pid_t resize_pid
, fsck_pid
;
2552 assert(setup
->dm_node
);
2554 /* First, unmount the file system */
2555 if (setup
->root_fd
>= 0) {
2556 setup
->root_fd
= safe_close(setup
->root_fd
);
2560 if (setup
->undo_mount
) {
2561 r
= home_setup_undo_mount(setup
, LOG_ERR
);
2568 log_info("Temporary unmounting of file system completed.");
2570 /* resize2fs requires that the file system is force checked first, do so. */
2571 r
= safe_fork("(e2fsck)",
2572 FORK_RESET_SIGNALS
|FORK_RLIMIT_NOFILE_SAFE
|FORK_DEATHSIG
|FORK_LOG
|FORK_STDOUT_TO_STDERR
|FORK_CLOSE_ALL_FDS
,
2578 execlp("e2fsck" ,"e2fsck", "-fp", setup
->dm_node
, NULL
);
2580 log_error_errno(errno
, "Failed to execute e2fsck: %m");
2581 _exit(EXIT_FAILURE
);
2584 exit_status
= wait_for_terminate_and_check("e2fsck", fsck_pid
, WAIT_LOG_ABNORMAL
);
2585 if (exit_status
< 0)
2587 if ((exit_status
& ~FSCK_ERROR_CORRECTED
) != 0) {
2588 log_warning("e2fsck failed with exit status %i.", exit_status
);
2590 if ((exit_status
& (FSCK_SYSTEM_SHOULD_REBOOT
|FSCK_ERRORS_LEFT_UNCORRECTED
)) != 0)
2591 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "File system is corrupted, refusing.");
2593 log_warning("Ignoring fsck error.");
2596 log_info("Forced file system check completed.");
2598 /* We use 512 sectors here, because resize2fs doesn't do byte sizes */
2599 if (asprintf(&size_str
, "%" PRIu64
"s", new_size
/ 512) < 0)
2602 /* Resize the thing */
2603 r
= safe_fork("(e2resize)",
2604 FORK_RESET_SIGNALS
|FORK_RLIMIT_NOFILE_SAFE
|FORK_DEATHSIG
|FORK_LOG
|FORK_WAIT
|FORK_STDOUT_TO_STDERR
|FORK_CLOSE_ALL_FDS
,
2610 execlp("resize2fs" ,"resize2fs", setup
->dm_node
, size_str
, NULL
);
2612 log_error_errno(errno
, "Failed to execute resize2fs: %m");
2613 _exit(EXIT_FAILURE
);
2616 log_info("Offline file system resize completed.");
2618 /* Re-establish mounts and reopen the directory */
2620 r
= home_mount_node(setup
->dm_node
, "ext4", discard
, flags
, extra_mount_options
);
2624 setup
->undo_mount
= true;
2628 setup
->root_fd
= open(HOME_RUNTIME_WORK_DIR
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
|O_NOFOLLOW
);
2629 if (setup
->root_fd
< 0)
2630 return log_error_errno(errno
, "Failed to reopen file system: %m");
2633 log_info("File system mounted again.");
2638 static int prepare_resize_partition(
2640 uint64_t partition_offset
,
2641 uint64_t old_partition_size
,
2642 sd_id128_t
*ret_disk_uuid
,
2643 struct fdisk_table
**ret_table
,
2644 struct fdisk_partition
**ret_partition
) {
2646 _cleanup_(fdisk_unref_contextp
) struct fdisk_context
*c
= NULL
;
2647 _cleanup_(fdisk_unref_tablep
) struct fdisk_table
*t
= NULL
;
2648 _cleanup_free_
char *path
= NULL
, *disk_uuid_as_string
= NULL
;
2649 struct fdisk_partition
*found
= NULL
;
2650 sd_id128_t disk_uuid
;
2651 size_t n_partitions
;
2655 assert(ret_disk_uuid
);
2658 assert((partition_offset
& 511) == 0);
2659 assert((old_partition_size
& 511) == 0);
2660 assert(UINT64_MAX
- old_partition_size
>= partition_offset
);
2662 if (partition_offset
== 0) {
2663 /* If the offset is at the beginning we assume no partition table, let's exit early. */
2664 log_debug("Not rewriting partition table, operating on naked device.");
2665 *ret_disk_uuid
= SD_ID128_NULL
;
2667 *ret_partition
= NULL
;
2671 c
= fdisk_new_context();
2675 if (asprintf(&path
, "/proc/self/fd/%i", fd
) < 0)
2678 r
= fdisk_assign_device(c
, path
, 0);
2680 return log_error_errno(r
, "Failed to open device: %m");
2682 if (!fdisk_is_labeltype(c
, FDISK_DISKLABEL_GPT
))
2683 return log_error_errno(SYNTHETIC_ERRNO(ENOMEDIUM
), "Disk has no GPT partition table.");
2685 r
= fdisk_get_disklabel_id(c
, &disk_uuid_as_string
);
2687 return log_error_errno(r
, "Failed to acquire disk UUID: %m");
2689 r
= sd_id128_from_string(disk_uuid_as_string
, &disk_uuid
);
2691 return log_error_errno(r
, "Failed parse disk UUID: %m");
2693 r
= fdisk_get_partitions(c
, &t
);
2695 return log_error_errno(r
, "Failed to acquire partition table: %m");
2697 n_partitions
= fdisk_table_get_nents(t
);
2698 for (size_t i
= 0; i
< n_partitions
; i
++) {
2699 struct fdisk_partition
*p
;
2701 p
= fdisk_table_get_partition(t
, i
);
2703 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Failed to read partition metadata: %m");
2705 if (fdisk_partition_is_used(p
) <= 0)
2707 if (fdisk_partition_has_start(p
) <= 0 || fdisk_partition_has_size(p
) <= 0 || fdisk_partition_has_end(p
) <= 0)
2708 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Found partition without a size.");
2710 if (fdisk_partition_get_start(p
) == partition_offset
/ 512U &&
2711 fdisk_partition_get_size(p
) == old_partition_size
/ 512U) {
2714 return log_error_errno(SYNTHETIC_ERRNO(ENOTUNIQ
), "Partition found twice, refusing.");
2717 } else if (fdisk_partition_get_end(p
) > partition_offset
/ 512U)
2718 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Can't extend, not last partition in image.");
2722 return log_error_errno(SYNTHETIC_ERRNO(ENOPKG
), "Failed to find matching partition to resize.");
2724 *ret_disk_uuid
= disk_uuid
;
2725 *ret_table
= TAKE_PTR(t
);
2726 *ret_partition
= found
;
2731 static int ask_cb(struct fdisk_context
*c
, struct fdisk_ask
*ask
, void *userdata
) {
2736 switch (fdisk_ask_get_type(ask
)) {
2738 case FDISK_ASKTYPE_STRING
:
2739 result
= new(char, 37);
2743 fdisk_ask_string_set_result(ask
, sd_id128_to_uuid_string(*(sd_id128_t
*) userdata
, result
));
2747 log_debug("Unexpected question from libfdisk, ignoring.");
2753 static int apply_resize_partition(
2755 sd_id128_t disk_uuids
,
2756 struct fdisk_table
*t
,
2757 struct fdisk_partition
*p
,
2758 size_t new_partition_size
) {
2760 _cleanup_(fdisk_unref_contextp
) struct fdisk_context
*c
= NULL
;
2761 _cleanup_free_
void *two_zero_lbas
= NULL
;
2762 _cleanup_free_
char *path
= NULL
;
2769 if (!t
) /* no partition table to apply, exit early */
2774 /* Before writing our partition patch the final size in */
2775 r
= fdisk_partition_size_explicit(p
, 1);
2777 return log_error_errno(r
, "Failed to enable explicit partition size: %m");
2779 r
= fdisk_partition_set_size(p
, new_partition_size
/ 512U);
2781 return log_error_errno(r
, "Failed to change partition size: %m");
2783 two_zero_lbas
= malloc0(1024U);
2787 /* libfdisk appears to get confused by the existing PMBR. Let's explicitly flush it out. */
2788 n
= pwrite(fd
, two_zero_lbas
, 1024U, 0);
2790 return log_error_errno(errno
, "Failed to wipe partition table: %m");
2792 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Short write while wiping partition table.");
2794 c
= fdisk_new_context();
2798 if (asprintf(&path
, "/proc/self/fd/%i", fd
) < 0)
2801 r
= fdisk_assign_device(c
, path
, 0);
2803 return log_error_errno(r
, "Failed to open device: %m");
2805 r
= fdisk_create_disklabel(c
, "gpt");
2807 return log_error_errno(r
, "Failed to create GPT disk label: %m");
2809 r
= fdisk_apply_table(c
, t
);
2811 return log_error_errno(r
, "Failed to apply partition table: %m");
2813 r
= fdisk_set_ask(c
, ask_cb
, &disk_uuids
);
2815 return log_error_errno(r
, "Failed to set libfdisk query function: %m");
2817 r
= fdisk_set_disklabel_id(c
);
2819 return log_error_errno(r
, "Failed to change disklabel ID: %m");
2821 r
= fdisk_write_disklabel(c
);
2823 return log_error_errno(r
, "Failed to write disk label: %m");
2828 /* Always keep at least 16M free, so that we can safely log in and update the user record while doing so */
2829 #define HOME_MIN_FREE (16U*1024U*1024U)
2831 static int get_smallest_fs_size(int fd
, uint64_t *ret
) {
2832 uint64_t minsz
, needed
;
2838 /* Determines the minimal disk size we might be able to shrink the file system referenced by the fd to. */
2840 if (syncfs(fd
) < 0) /* let's sync before we query the size, so that the values returned are accurate */
2841 return log_error_errno(errno
, "Failed to synchronize home file system: %m");
2843 if (fstatfs(fd
, &sfs
) < 0)
2844 return log_error_errno(errno
, "Failed to statfs() home file system: %m");
2846 /* Let's determine the minimal file system size of the used fstype */
2847 minsz
= minimal_size_by_fs_magic(sfs
.f_type
);
2848 if (minsz
== UINT64_MAX
)
2849 return log_error_errno(SYNTHETIC_ERRNO(EOPNOTSUPP
), "Don't know minimum file system size of file system type '%s' of home directory.", fs_type_to_string(sfs
.f_type
));
2851 if (minsz
< USER_DISK_SIZE_MIN
)
2852 minsz
= USER_DISK_SIZE_MIN
;
2854 if (sfs
.f_bfree
> sfs
.f_blocks
)
2855 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Detected amount of free blocks is greater than the total amount of file system blocks. Refusing.");
2857 /* Calculate how much disk space is currently in use. */
2858 needed
= sfs
.f_blocks
- sfs
.f_bfree
;
2859 if (needed
> UINT64_MAX
/ sfs
.f_bsize
)
2860 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "File system size out of range.");
2862 needed
*= sfs
.f_bsize
;
2864 /* Add some safety margin of free space we'll always keep */
2865 if (needed
> UINT64_MAX
- HOME_MIN_FREE
) /* Check for overflow */
2866 needed
= UINT64_MAX
;
2868 needed
+= HOME_MIN_FREE
;
2870 *ret
= DISK_SIZE_ROUND_UP(MAX(needed
, minsz
));
2874 static int get_largest_image_size(int fd
, const struct stat
*st
, uint64_t *ret
) {
2875 uint64_t used
, avail
, sum
;
2883 /* Determines the maximum file size we might be able to grow the image file referenced by the fd to. */
2885 r
= stat_verify_regular(st
);
2887 return log_error_errno(r
, "Image file is not a regular file, refusing: %m");
2890 return log_error_errno(errno
, "Failed to synchronize file system backing image file: %m");
2892 if (fstatfs(fd
, &sfs
) < 0)
2893 return log_error_errno(errno
, "Failed to statfs() image file: %m");
2895 used
= (uint64_t) st
->st_blocks
* 512;
2896 avail
= (uint64_t) sfs
.f_bsize
* sfs
.f_bavail
;
2898 if (avail
> UINT64_MAX
- used
)
2903 *ret
= DISK_SIZE_ROUND_DOWN(MIN(sum
, USER_DISK_SIZE_MAX
));
2907 static int resize_fs_loop(
2911 uint64_t old_fs_size
,
2912 uint64_t new_fs_size
,
2913 uint64_t *ret_fs_size
) {
2915 uint64_t current_fs_size
;
2916 unsigned n_iterations
= 0;
2921 assert(setup
->root_fd
>= 0);
2923 /* A bisection loop trying to find the closest size to what the user asked for. (Well, we bisect like
2924 * this only when we *shrink* the fs — if we grow the fs there's no need to bisect.) */
2926 current_fs_size
= old_fs_size
;
2927 for (uint64_t lower_boundary
= new_fs_size
, upper_boundary
= old_fs_size
, try_fs_size
= new_fs_size
;;) {
2932 /* Now resize the file system */
2933 if (resize_type
== CAN_RESIZE_ONLINE
) {
2934 r
= resize_fs(setup
->root_fd
, try_fs_size
, NULL
);
2936 if (!ERRNO_IS_DISK_SPACE(r
) || new_fs_size
> old_fs_size
) /* Not a disk space issue? Not trying to shrink? */
2937 return log_error_errno(r
, "Failed to resize file system: %m");
2939 log_debug_errno(r
, "Shrinking from %s to %s didn't work, not enough space for contained data.", FORMAT_BYTES(current_fs_size
), FORMAT_BYTES(try_fs_size
));
2942 log_debug("Successfully resized from %s to %s.", FORMAT_BYTES(current_fs_size
), FORMAT_BYTES(try_fs_size
));
2943 current_fs_size
= try_fs_size
;
2947 /* If we hit a disk space issue and are shrinking the fs, then maybe it helps to
2948 * increase the image size. */
2950 r
= ext4_offline_resize_fs(setup
, try_fs_size
, user_record_luks_discard(h
), user_record_mount_flags(h
), h
->luks_extra_mount_options
);
2954 /* For now, when we fail to shrink an ext4 image we'll not try again via the
2955 * bisection logic. We might add that later, but give this involves shelling out
2956 * multiple programs it's a bit too cumbersome to my taste. */
2959 current_fs_size
= try_fs_size
;
2962 if (new_fs_size
> old_fs_size
) /* If we are growing we are done after one iteration */
2965 /* If we are shrinking then let's adjust our bisection boundaries and try again. */
2967 upper_boundary
= MIN(upper_boundary
, try_fs_size
);
2969 lower_boundary
= MAX(lower_boundary
, try_fs_size
);
2971 /* OK, this attempt to shrink didn't work. Let's try between the old size and what worked. */
2972 if (lower_boundary
>= upper_boundary
) {
2973 log_debug("Image can't be shrunk further (range to try is empty).");
2977 /* Let's find a new value to try half-way between the lower boundary and the upper boundary
2979 try_fs_size
= DISK_SIZE_ROUND_DOWN(lower_boundary
+ (upper_boundary
- lower_boundary
) / 2);
2980 if (try_fs_size
<= lower_boundary
|| try_fs_size
>= upper_boundary
) {
2981 log_debug("Image can't be shrunk further (remaining range to try too small).");
2986 log_debug("Bisection loop completed after %u iterations.", n_iterations
);
2989 *ret_fs_size
= current_fs_size
;
2994 static int resize_image_loop(
2997 uint64_t old_image_size
,
2998 uint64_t new_image_size
,
2999 uint64_t *ret_image_size
) {
3001 uint64_t current_image_size
;
3002 unsigned n_iterations
= 0;
3007 assert(setup
->image_fd
>= 0);
3009 /* A bisection loop trying to find the closest size to what the user asked for. (Well, we bisect like
3010 * this only when we *grow* the image — if we shrink the image then there's no need to bisect.) */
3012 current_image_size
= old_image_size
;
3013 for (uint64_t lower_boundary
= old_image_size
, upper_boundary
= new_image_size
, try_image_size
= new_image_size
;;) {
3018 r
= home_truncate(h
, setup
->image_fd
, try_image_size
);
3020 if (!ERRNO_IS_DISK_SPACE(r
) || new_image_size
< old_image_size
) /* Not a disk space issue? Not trying to grow? */
3023 log_debug_errno(r
, "Growing from %s to %s didn't work, not enough space on backing disk.", FORMAT_BYTES(current_image_size
), FORMAT_BYTES(try_image_size
));
3025 } else if (r
> 0) { /* Success: allocation worked */
3026 log_debug("Resizing from %s to %s via allocation worked successfully.", FORMAT_BYTES(current_image_size
), FORMAT_BYTES(try_image_size
));
3027 current_image_size
= try_image_size
;
3029 } else { /* Success, but through truncation, not allocation. */
3030 log_debug("Resizing from %s to %s via truncation worked successfully.", FORMAT_BYTES(old_image_size
), FORMAT_BYTES(try_image_size
));
3031 current_image_size
= try_image_size
;
3032 break; /* there's no point in the bisection logic if this was plain truncation and
3033 * not allocation, let's exit immediately. */
3036 if (new_image_size
< old_image_size
) /* If we are shrinking we are done after one iteration */
3039 /* If we are growing then let's adjust our bisection boundaries and try again */
3041 lower_boundary
= MAX(lower_boundary
, try_image_size
);
3043 upper_boundary
= MIN(upper_boundary
, try_image_size
);
3045 if (lower_boundary
>= upper_boundary
) {
3046 log_debug("Image can't be grown further (range to try is empty).");
3050 try_image_size
= DISK_SIZE_ROUND_DOWN(lower_boundary
+ (upper_boundary
- lower_boundary
) / 2);
3051 if (try_image_size
<= lower_boundary
|| try_image_size
>= upper_boundary
) {
3052 log_debug("Image can't be grown further (remaining range to try too small).");
3057 log_debug("Bisection loop completed after %u iterations.", n_iterations
);
3060 *ret_image_size
= current_image_size
;
3065 int home_resize_luks(
3067 HomeSetupFlags flags
,
3069 PasswordCache
*cache
,
3070 UserRecord
**ret_home
) {
3072 uint64_t old_image_size
, new_image_size
, old_fs_size
, new_fs_size
, crypto_offset
, crypto_offset_bytes
,
3073 new_partition_size
, smallest_fs_size
, resized_fs_size
;
3074 _cleanup_(user_record_unrefp
) UserRecord
*header_home
= NULL
, *embedded_home
= NULL
, *new_home
= NULL
;
3075 _cleanup_(fdisk_unref_tablep
) struct fdisk_table
*table
= NULL
;
3076 struct fdisk_partition
*partition
= NULL
;
3077 _cleanup_close_
int opened_image_fd
= -1;
3078 _cleanup_free_
char *whole_disk
= NULL
;
3079 int r
, resize_type
, image_fd
= -1;
3080 sd_id128_t disk_uuid
;
3081 const char *ip
, *ipo
;
3085 INTENTION_DONT_KNOW
= 0, /* These happen to match the return codes of CMP() */
3086 INTENTION_SHRINK
= -1,
3088 } intention
= INTENTION_DONT_KNOW
;
3091 assert(user_record_storage(h
) == USER_LUKS
);
3094 r
= dlopen_cryptsetup();
3098 assert_se(ipo
= user_record_image_path(h
));
3099 ip
= strdupa_safe(ipo
); /* copy out since original might change later in home record object */
3101 if (setup
->image_fd
< 0) {
3102 setup
->image_fd
= open_image_file(h
, NULL
, &st
);
3103 if (setup
->image_fd
< 0)
3104 return setup
->image_fd
;
3106 if (fstat(setup
->image_fd
, &st
) < 0)
3107 return log_error_errno(errno
, "Failed to stat image file %s: %m", ip
);
3110 image_fd
= setup
->image_fd
;
3112 if (S_ISBLK(st
.st_mode
)) {
3115 r
= block_get_whole_disk(st
.st_rdev
, &parent
);
3117 return log_error_errno(r
, "Failed to acquire whole block device for %s: %m", ip
);
3119 /* If we shall resize a file system on a partition device, then let's figure out the
3120 * whole disk device and operate on that instead, since we need to rewrite the
3121 * partition table to resize the partition. */
3123 log_info("Operating on partition device %s, using parent device.", ip
);
3125 opened_image_fd
= r
= device_open_from_devnum(S_IFBLK
, parent
, O_RDWR
|O_CLOEXEC
|O_NOCTTY
|O_NONBLOCK
, &whole_disk
);
3127 return log_error_errno(r
, "Failed to open whole block device for %s: %m", ip
);
3129 image_fd
= opened_image_fd
;
3131 if (fstat(image_fd
, &st
) < 0)
3132 return log_error_errno(errno
, "Failed to stat whole block device %s: %m", whole_disk
);
3134 log_info("Operating on whole block device %s.", ip
);
3136 if (ioctl(image_fd
, BLKGETSIZE64
, &old_image_size
) < 0)
3137 return log_error_errno(errno
, "Failed to determine size of original block device: %m");
3139 if (flock(image_fd
, LOCK_EX
) < 0) /* make sure udev doesn't read from it while we operate on the device */
3140 return log_error_errno(errno
, "Failed to lock block device %s: %m", ip
);
3142 new_image_size
= old_image_size
; /* we can't resize physical block devices */
3144 r
= stat_verify_regular(&st
);
3146 return log_error_errno(r
, "Image %s is not a block device nor regular file: %m", ip
);
3148 old_image_size
= st
.st_size
;
3150 /* Note an asymetry here: when we operate on loopback files the specified disk size we get we
3151 * apply onto the loopback file as a whole. When we operate on block devices we instead apply
3152 * to the partition itself only. */
3154 if (FLAGS_SET(flags
, HOME_SETUP_RESIZE_MINIMIZE
)) {
3156 intention
= INTENTION_SHRINK
;
3158 uint64_t new_image_size_rounded
;
3160 new_image_size_rounded
= DISK_SIZE_ROUND_DOWN(h
->disk_size
);
3162 if (old_image_size
>= new_image_size_rounded
&& old_image_size
<= h
->disk_size
) {
3163 /* If exact match, or a match after we rounded down, don't do a thing */
3164 log_info("Image size already matching, skipping operation.");
3168 new_image_size
= new_image_size_rounded
;
3169 intention
= CMP(new_image_size
, old_image_size
); /* Is this a shrink */
3173 r
= home_setup_luks(
3179 FLAGS_SET(flags
, HOME_SETUP_RESIZE_DONT_SYNC_IDENTITIES
) ? NULL
: &header_home
);
3183 if (!FLAGS_SET(flags
, HOME_SETUP_RESIZE_DONT_SYNC_IDENTITIES
)) {
3184 r
= home_load_embedded_identity(h
, setup
->root_fd
, header_home
, USER_RECONCILE_REQUIRE_NEWER_OR_EQUAL
, cache
, &embedded_home
, &new_home
);
3189 r
= home_maybe_shift_uid(h
, flags
, setup
);
3193 log_info("offset = %" PRIu64
", size = %" PRIu64
", image = %" PRIu64
, setup
->partition_offset
, setup
->partition_size
, old_image_size
);
3195 if ((UINT64_MAX
- setup
->partition_offset
) < setup
->partition_size
||
3196 setup
->partition_offset
+ setup
->partition_size
> old_image_size
)
3197 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Old partition doesn't fit in backing storage, refusing.");
3199 if (S_ISREG(st
.st_mode
)) {
3200 uint64_t partition_table_extra
, largest_size
;
3202 partition_table_extra
= old_image_size
- setup
->partition_size
;
3204 r
= get_largest_image_size(setup
->image_fd
, &st
, &largest_size
);
3207 if (new_image_size
> largest_size
)
3208 new_image_size
= largest_size
;
3210 if (new_image_size
< partition_table_extra
)
3211 new_image_size
= partition_table_extra
;
3213 new_partition_size
= DISK_SIZE_ROUND_DOWN(new_image_size
- partition_table_extra
);
3215 assert(S_ISBLK(st
.st_mode
));
3217 if (FLAGS_SET(flags
, HOME_SETUP_RESIZE_MINIMIZE
)) {
3218 new_partition_size
= 0;
3219 intention
= INTENTION_SHRINK
;
3221 uint64_t new_partition_size_rounded
;
3223 new_partition_size_rounded
= DISK_SIZE_ROUND_DOWN(h
->disk_size
);
3225 if (setup
->partition_size
>= new_partition_size_rounded
&&
3226 setup
->partition_size
<= h
->disk_size
) {
3227 log_info("Partition size already matching, skipping operation.");
3231 new_partition_size
= new_partition_size_rounded
;
3232 intention
= CMP(new_partition_size
, setup
->partition_size
);
3236 if ((UINT64_MAX
- setup
->partition_offset
) < new_partition_size
||
3237 setup
->partition_offset
+ new_partition_size
> new_image_size
)
3238 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "New partition doesn't fit into backing storage, refusing.");
3240 crypto_offset
= sym_crypt_get_data_offset(setup
->crypt_device
);
3241 if (crypto_offset
> UINT64_MAX
/512U)
3242 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "LUKS2 data offset out of range, refusing.");
3243 crypto_offset_bytes
= (uint64_t) crypto_offset
* 512U;
3244 if (setup
->partition_size
<= crypto_offset_bytes
)
3245 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Weird, old crypto payload offset doesn't actually fit in partition size?");
3247 /* Make sure at least the LUKS header fit in */
3248 if (new_partition_size
<= crypto_offset_bytes
) {
3251 add
= DISK_SIZE_ROUND_UP(crypto_offset_bytes
) - new_partition_size
;
3252 new_partition_size
+= add
;
3253 if (S_ISREG(st
.st_mode
))
3254 new_image_size
+= add
;
3257 old_fs_size
= setup
->partition_size
- crypto_offset_bytes
;
3258 new_fs_size
= DISK_SIZE_ROUND_DOWN(new_partition_size
- crypto_offset_bytes
);
3260 r
= get_smallest_fs_size(setup
->root_fd
, &smallest_fs_size
);
3264 if (new_fs_size
< smallest_fs_size
) {
3267 add
= DISK_SIZE_ROUND_UP(smallest_fs_size
) - new_fs_size
;
3269 new_partition_size
+= add
;
3270 if (S_ISREG(st
.st_mode
))
3271 new_image_size
+= add
;
3274 if (new_fs_size
== old_fs_size
) {
3275 log_info("New file system size identical to old file system size, skipping operation.");
3279 if (FLAGS_SET(flags
, HOME_SETUP_RESIZE_DONT_GROW
) && new_fs_size
> old_fs_size
) {
3280 log_info("New file system size would be larger than old, but shrinking requested, skipping operation.");
3284 if (FLAGS_SET(flags
, HOME_SETUP_RESIZE_DONT_SHRINK
) && new_fs_size
< old_fs_size
) {
3285 log_info("New file system size would be smaller than old, but growing requested, skipping operation.");
3289 if (CMP(new_fs_size
, old_fs_size
) != intention
) {
3291 log_info("Shrink operation would enlarge file system, skipping operation.");
3293 assert(intention
> 0);
3294 log_info("Grow operation would shrink file system, skipping operation.");
3299 /* Before we start doing anything, let's figure out if we actually can */
3300 resize_type
= can_resize_fs(setup
->root_fd
, old_fs_size
, new_fs_size
);
3301 if (resize_type
< 0)
3303 if (resize_type
== CAN_RESIZE_OFFLINE
&& FLAGS_SET(flags
, HOME_SETUP_ALREADY_ACTIVATED
))
3304 return log_error_errno(SYNTHETIC_ERRNO(ETXTBSY
), "File systems of this type can only be resized offline, but is currently online.");
3306 log_info("Ready to resize image size %s %s %s, partition size %s %s %s, file system size %s %s %s.",
3307 FORMAT_BYTES(old_image_size
),
3308 special_glyph(SPECIAL_GLYPH_ARROW_RIGHT
),
3309 FORMAT_BYTES(new_image_size
),
3310 FORMAT_BYTES(setup
->partition_size
),
3311 special_glyph(SPECIAL_GLYPH_ARROW_RIGHT
),
3312 FORMAT_BYTES(new_partition_size
),
3313 FORMAT_BYTES(old_fs_size
),
3314 special_glyph(SPECIAL_GLYPH_ARROW_RIGHT
),
3315 FORMAT_BYTES(new_fs_size
));
3317 r
= prepare_resize_partition(
3319 setup
->partition_offset
,
3320 setup
->partition_size
,
3327 if (new_fs_size
> old_fs_size
) { /* → Grow */
3329 if (S_ISREG(st
.st_mode
)) {
3330 uint64_t resized_image_size
;
3332 /* Grow file size */
3333 r
= resize_image_loop(h
, setup
, old_image_size
, new_image_size
, &resized_image_size
);
3337 if (resized_image_size
== old_image_size
) {
3338 log_info("Couldn't change image size.");
3342 assert(resized_image_size
> old_image_size
);
3344 log_info("Growing of image file from %s to %s completed.", FORMAT_BYTES(old_image_size
), FORMAT_BYTES(resized_image_size
));
3346 if (resized_image_size
< new_image_size
) {
3349 /* If the growing we managed to do is smaller than what we wanted we need to
3350 * adjust the partition/file system sizes we are going for, too */
3351 sub
= new_image_size
- resized_image_size
;
3352 assert(new_partition_size
>= sub
);
3353 new_partition_size
-= sub
;
3354 assert(new_fs_size
>= sub
);
3358 new_image_size
= resized_image_size
;
3360 assert(S_ISBLK(st
.st_mode
));
3361 assert(new_image_size
== old_image_size
);
3364 /* Make sure loopback device sees the new bigger size */
3365 r
= loop_device_refresh_size(setup
->loop
, UINT64_MAX
, new_partition_size
);
3367 log_debug_errno(r
, "Device is not a loopback device, not refreshing size.");
3369 return log_error_errno(r
, "Failed to refresh loopback device size: %m");
3371 log_info("Refreshing loop device size completed.");
3373 r
= apply_resize_partition(image_fd
, disk_uuid
, table
, partition
, new_partition_size
);
3377 log_info("Growing of partition completed.");
3379 if (S_ISBLK(st
.st_mode
) && ioctl(image_fd
, BLKRRPART
, 0) < 0)
3380 log_debug_errno(errno
, "BLKRRPART failed on block device, ignoring: %m");
3382 /* Tell LUKS about the new bigger size too */
3383 r
= sym_crypt_resize(setup
->crypt_device
, setup
->dm_name
, new_fs_size
/ 512U);
3385 return log_error_errno(r
, "Failed to grow LUKS device: %m");
3387 log_info("LUKS device growing completed.");
3391 if (!FLAGS_SET(flags
, HOME_SETUP_RESIZE_DONT_SYNC_IDENTITIES
)) {
3392 r
= home_store_embedded_identity(new_home
, setup
->root_fd
, h
->uid
, embedded_home
);
3397 if (S_ISREG(st
.st_mode
)) {
3398 if (user_record_luks_discard(h
))
3399 /* Before we shrink, let's trim the file system, so that we need less space on disk during the shrinking */
3400 (void) run_fitrim(setup
->root_fd
);
3402 /* If discard is off, let's ensure all backing blocks are allocated, so that our resize operation doesn't fail half-way */
3403 r
= run_fallocate(image_fd
, &st
);
3410 /* Now try to resize the file system. The requested size might not always be possible, in which case
3411 * we'll try to get as close as we can get. The result is returned in 'resized_fs_size' */
3412 r
= resize_fs_loop(h
, setup
, resize_type
, old_fs_size
, new_fs_size
, &resized_fs_size
);
3416 if (resized_fs_size
== old_fs_size
) {
3417 log_info("Couldn't change file system size.");
3421 log_info("File system resizing from %s to %s completed.", FORMAT_BYTES(old_fs_size
), FORMAT_BYTES(resized_fs_size
));
3423 if (resized_fs_size
> new_fs_size
) {
3426 /* If the shrinking we managed to do is larger than what we wanted we need to adjust the partition/image sizes. */
3427 add
= resized_fs_size
- new_fs_size
;
3428 new_partition_size
+= add
;
3429 if (S_ISREG(st
.st_mode
))
3430 new_image_size
+= add
;
3433 new_fs_size
= resized_fs_size
;
3435 /* Immediately sync afterwards */
3436 r
= home_sync_and_statfs(setup
->root_fd
, NULL
);
3440 if (new_fs_size
< old_fs_size
) { /* → Shrink */
3442 /* Shrink the LUKS device now, matching the new file system size */
3443 r
= sym_crypt_resize(setup
->crypt_device
, setup
->dm_name
, new_fs_size
/ 512);
3445 return log_error_errno(r
, "Failed to shrink LUKS device: %m");
3447 log_info("LUKS device shrinking completed.");
3449 /* Refresh the loop devices size */
3450 r
= loop_device_refresh_size(setup
->loop
, UINT64_MAX
, new_partition_size
);
3452 log_debug_errno(r
, "Device is not a loopback device, not refreshing size.");
3454 return log_error_errno(r
, "Failed to refresh loopback device size: %m");
3456 log_info("Refreshing loop device size completed.");
3458 if (S_ISREG(st
.st_mode
)) {
3459 /* Shrink the image file */
3460 if (ftruncate(image_fd
, new_image_size
) < 0)
3461 return log_error_errno(errno
, "Failed to shrink image file %s: %m", ip
);
3463 log_info("Shrinking of image file completed.");
3465 assert(S_ISBLK(st
.st_mode
));
3466 assert(new_image_size
== old_image_size
);
3469 r
= apply_resize_partition(image_fd
, disk_uuid
, table
, partition
, new_partition_size
);
3473 log_info("Shrinking of partition completed.");
3475 if (S_ISBLK(st
.st_mode
) && ioctl(image_fd
, BLKRRPART
, 0) < 0)
3476 log_debug_errno(errno
, "BLKRRPART failed on block device, ignoring: %m");
3478 } else { /* → Grow */
3479 if (!FLAGS_SET(flags
, HOME_SETUP_RESIZE_DONT_SYNC_IDENTITIES
)) {
3480 r
= home_store_embedded_identity(new_home
, setup
->root_fd
, h
->uid
, embedded_home
);
3486 if (!FLAGS_SET(flags
, HOME_SETUP_RESIZE_DONT_SYNC_IDENTITIES
)) {
3487 r
= home_store_header_identity_luks(new_home
, setup
, header_home
);
3491 r
= home_extend_embedded_identity(new_home
, h
, setup
);
3496 if (user_record_luks_discard(h
))
3497 (void) run_fitrim(setup
->root_fd
);
3499 r
= home_sync_and_statfs(setup
->root_fd
, &sfs
);
3503 if (!FLAGS_SET(flags
, HOME_SETUP_RESIZE_DONT_UNDO
)) {
3504 r
= home_setup_done(setup
);
3509 log_info("Resizing completed.");
3511 print_size_summary(new_image_size
, new_fs_size
, &sfs
);
3514 *ret_home
= TAKE_PTR(new_home
);
3519 int home_passwd_luks(
3521 HomeSetupFlags flags
,
3523 const PasswordCache
*cache
, /* the passwords acquired via PKCS#11/FIDO2 security tokens */
3524 char **effective_passwords
/* new passwords */) {
3526 size_t volume_key_size
, max_key_slots
, n_effective
;
3527 _cleanup_(erase_and_freep
) void *volume_key
= NULL
;
3528 struct crypt_pbkdf_type good_pbkdf
, minimal_pbkdf
;
3534 assert(user_record_storage(h
) == USER_LUKS
);
3537 r
= dlopen_cryptsetup();
3541 type
= sym_crypt_get_type(setup
->crypt_device
);
3543 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine crypto device type.");
3545 r
= sym_crypt_keyslot_max(type
);
3547 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine number of key slots.");
3550 r
= sym_crypt_get_volume_key_size(setup
->crypt_device
);
3552 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to determine volume key size.");
3553 volume_key_size
= (size_t) r
;
3555 volume_key
= malloc(volume_key_size
);
3560 FOREACH_POINTER(list
,
3561 cache
? cache
->keyring_passswords
: NULL
,
3562 cache
? cache
->pkcs11_passwords
: NULL
,
3563 cache
? cache
->fido2_passwords
: NULL
,
3566 r
= luks_try_passwords(h
, setup
->crypt_device
, list
, volume_key
, &volume_key_size
, NULL
);
3571 return log_error_errno(SYNTHETIC_ERRNO(ENOKEY
), "Failed to unlock LUKS superblock with supplied passwords.");
3573 return log_error_errno(r
, "Failed to unlocks LUKS superblock: %m");
3575 n_effective
= strv_length(effective_passwords
);
3577 build_good_pbkdf(&good_pbkdf
, h
);
3578 build_minimal_pbkdf(&minimal_pbkdf
, h
);
3580 for (size_t i
= 0; i
< max_key_slots
; i
++) {
3581 r
= sym_crypt_keyslot_destroy(setup
->crypt_device
, i
);
3582 if (r
< 0 && !IN_SET(r
, -ENOENT
, -EINVAL
)) /* Returns EINVAL or ENOENT if there's no key in this slot already */
3583 return log_error_errno(r
, "Failed to destroy LUKS password: %m");
3585 if (i
>= n_effective
) {
3587 log_info("Destroyed LUKS key slot %zu.", i
);
3591 if (password_cache_contains(cache
, effective_passwords
[i
])) { /* Is this a FIDO2 or PKCS#11 password? */
3592 log_debug("Using minimal PBKDF for slot %zu", i
);
3593 r
= sym_crypt_set_pbkdf_type(setup
->crypt_device
, &minimal_pbkdf
);
3595 log_debug("Using good PBKDF for slot %zu", i
);
3596 r
= sym_crypt_set_pbkdf_type(setup
->crypt_device
, &good_pbkdf
);
3599 return log_error_errno(r
, "Failed to tweak PBKDF for slot %zu: %m", i
);
3601 r
= sym_crypt_keyslot_add_by_volume_key(
3602 setup
->crypt_device
,
3606 effective_passwords
[i
],
3607 strlen(effective_passwords
[i
]));
3609 return log_error_errno(r
, "Failed to set up LUKS password: %m");
3611 log_info("Updated LUKS key slot %zu.", i
);
3613 /* If we changed the password, then make sure to update the copy in the keyring, so that
3614 * auto-rebalance continues to work. We only do this if we operate on an active home dir. */
3615 if (i
== 0 && FLAGS_SET(flags
, HOME_SETUP_ALREADY_ACTIVATED
))
3616 upload_to_keyring(h
, effective_passwords
[i
], NULL
);
3622 int home_lock_luks(UserRecord
*h
, HomeSetup
*setup
) {
3628 assert(setup
->root_fd
< 0);
3629 assert(!setup
->crypt_device
);
3631 r
= acquire_open_luks_device(h
, setup
, /* graceful= */ false);
3635 log_info("Discovered used LUKS device %s.", setup
->dm_node
);
3637 assert_se(p
= user_record_home_directory(h
));
3638 r
= syncfs_path(AT_FDCWD
, p
);
3639 if (r
< 0) /* Snake oil, but let's better be safe than sorry */
3640 return log_error_errno(r
, "Failed to synchronize file system %s: %m", p
);
3642 log_info("File system synchronized.");
3644 /* Note that we don't invoke FIFREEZE here, it appears libcryptsetup/device-mapper already does that on its own for us */
3646 r
= sym_crypt_suspend(setup
->crypt_device
, setup
->dm_name
);
3648 return log_error_errno(r
, "Failed to suspend cryptsetup device: %s: %m", setup
->dm_node
);
3650 log_info("LUKS device suspended.");
3654 static int luks_try_resume(
3655 struct crypt_device
*cd
,
3656 const char *dm_name
,
3664 STRV_FOREACH(pp
, password
) {
3665 r
= sym_crypt_resume_by_passphrase(
3672 log_info("Resumed LUKS device %s.", dm_name
);
3676 log_debug_errno(r
, "Password %zu didn't work for resuming device: %m", (size_t) (pp
- password
));
3682 int home_unlock_luks(UserRecord
*h
, HomeSetup
*setup
, const PasswordCache
*cache
) {
3688 assert(!setup
->crypt_device
);
3690 r
= acquire_open_luks_device(h
, setup
, /* graceful= */ false);
3694 log_info("Discovered used LUKS device %s.", setup
->dm_node
);
3697 FOREACH_POINTER(list
,
3698 cache
? cache
->pkcs11_passwords
: NULL
,
3699 cache
? cache
->fido2_passwords
: NULL
,
3701 r
= luks_try_resume(setup
->crypt_device
, setup
->dm_name
, list
);
3706 return log_error_errno(r
, "No valid password for LUKS superblock.");
3708 return log_error_errno(r
, "Failed to resume LUKS superblock: %m");
3710 log_info("LUKS device resumed.");
3714 static int device_is_gone(HomeSetup
*setup
) {
3715 _cleanup_(sd_device_unrefp
) sd_device
*d
= NULL
;
3721 if (!setup
->dm_node
)
3724 if (stat(setup
->dm_node
, &st
) < 0) {
3725 if (errno
!= ENOENT
)
3726 return log_error_errno(errno
, "Failed to stat block device node %s: %m", setup
->dm_node
);
3731 r
= sd_device_new_from_stat_rdev(&d
, &st
);
3734 return log_error_errno(errno
, "Failed to allocate device object from block device node %s: %m", setup
->dm_node
);
3742 static int device_monitor_handler(sd_device_monitor
*monitor
, sd_device
*device
, void *userdata
) {
3743 HomeSetup
*setup
= ASSERT_PTR(userdata
);
3746 if (!device_for_action(device
, SD_DEVICE_REMOVE
))
3749 /* We don't really care for the device object passed to us, we just check if the device node still
3752 r
= device_is_gone(setup
);
3755 if (r
> 0) /* Yay! we are done! */
3756 (void) sd_event_exit(sd_device_monitor_get_event(monitor
), 0);
3761 int wait_for_block_device_gone(HomeSetup
*setup
, usec_t timeout_usec
) {
3762 _cleanup_(sd_device_monitor_unrefp
) sd_device_monitor
*m
= NULL
;
3763 _cleanup_(sd_event_unrefp
) sd_event
*event
= NULL
;
3768 /* So here's the thing: we enable "deferred deactivation" on our dm-crypt volumes. This means they
3769 * are automatically torn down once not used anymore (i.e. once unmounted). Which is great. It also
3770 * means that when we deactivate a home directory and try to tear down the volume that backs it, it
3771 * possibly is already torn down or in the process of being torn down, since we race against the
3772 * automatic tearing down. Which is fine, we handle errors from that. However, we lose the ability to
3773 * naturally wait for the tear down operation to complete: if we are not the ones who tear down the
3774 * device we are also not the ones who naturally block on that operation. Hence let's add some code
3775 * to actively wait for the device to go away, via sd-device. We'll call this whenever tearing down a
3776 * LUKS device, to ensure the device is really really gone before we proceed. Net effect: "homectl
3777 * deactivate foo && homectl activate foo" will work reliably, i.e. deactivation immediately followed
3778 * by activation will work. Also, by the time deactivation completes we can guarantee that all data
3779 * is sync'ed down to the lowest block layer as all higher levels are fully and entirely
3782 if (!setup
->dm_name
)
3785 assert(setup
->dm_node
);
3786 log_debug("Waiting until %s disappears.", setup
->dm_node
);
3788 r
= sd_event_new(&event
);
3790 return log_error_errno(r
, "Failed to allocate event loop: %m");
3792 r
= sd_device_monitor_new(&m
);
3794 return log_error_errno(r
, "Failed to allocate device monitor: %m");
3796 r
= sd_device_monitor_filter_add_match_subsystem_devtype(m
, "block", "disk");
3798 return log_error_errno(r
, "Failed to configure device monitor match: %m");
3800 r
= sd_device_monitor_attach_event(m
, event
);
3802 return log_error_errno(r
, "Failed to attach device monitor to event loop: %m");
3804 r
= sd_device_monitor_start(m
, device_monitor_handler
, setup
);
3806 return log_error_errno(r
, "Failed to start device monitor: %m");
3808 r
= device_is_gone(setup
);
3812 log_debug("%s has already disappeared before entering wait loop.", setup
->dm_node
);
3813 return 0; /* gone already */
3816 if (timeout_usec
!= USEC_INFINITY
) {
3817 r
= sd_event_add_time_relative(event
, NULL
, CLOCK_MONOTONIC
, timeout_usec
, 0, NULL
, NULL
);
3819 return log_error_errno(r
, "Failed to add timer event: %m");
3822 r
= sd_event_loop(event
);
3824 return log_error_errno(r
, "Failed to run event loop: %m");
3826 r
= device_is_gone(setup
);
3830 return log_error_errno(r
, "Device %s still around.", setup
->dm_node
);
3832 log_debug("Successfully waited until device %s disappeared.", setup
->dm_node
);
3836 int home_auto_shrink_luks(UserRecord
*h
, HomeSetup
*setup
, PasswordCache
*cache
) {
3841 assert(user_record_storage(h
) == USER_LUKS
);
3843 assert(setup
->root_fd
>= 0);
3845 if (user_record_auto_resize_mode(h
) != AUTO_RESIZE_SHRINK_AND_GROW
)
3848 if (fstatfs(setup
->root_fd
, &sfs
) < 0)
3849 return log_error_errno(errno
, "Failed to statfs home directory: %m");
3851 if (!fs_can_online_shrink_and_grow(sfs
.f_type
)) {
3852 log_debug("Not auto-shrinking file system, since selected file system cannot do both online shrink and grow.");
3856 r
= home_resize_luks(
3858 HOME_SETUP_ALREADY_ACTIVATED
|
3859 HOME_SETUP_RESIZE_DONT_SYNC_IDENTITIES
|
3860 HOME_SETUP_RESIZE_MINIMIZE
|
3861 HOME_SETUP_RESIZE_DONT_GROW
|
3862 HOME_SETUP_RESIZE_DONT_UNDO
,