1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
6 #include <linux/loop.h>
10 #include <sys/ioctl.h>
14 #include "alloc-util.h"
15 #include "btrfs-util.h"
16 #include "chattr-util.h"
18 #include "dirent-util.h"
19 #include "discover-image.h"
20 #include "dissect-image.h"
26 #include "hostname-setup.h"
27 #include "id128-util.h"
28 #include "lockfile-util.h"
30 #include "loop-util.h"
33 #include "nulstr-util.h"
35 #include "path-util.h"
37 #include "string-table.h"
38 #include "string-util.h"
40 #include "time-util.h"
42 #include "xattr-util.h"
44 static const char* const image_search_path
[_IMAGE_CLASS_MAX
] = {
45 [IMAGE_MACHINE
] = "/etc/machines\0" /* only place symlinks here */
46 "/run/machines\0" /* and here too */
47 "/var/lib/machines\0" /* the main place for images */
48 "/var/lib/container\0" /* legacy */
49 "/usr/local/lib/machines\0"
50 "/usr/lib/machines\0",
52 [IMAGE_PORTABLE
] = "/etc/portables\0" /* only place symlinks here */
53 "/run/portables\0" /* and here too */
54 "/var/lib/portables\0" /* the main place for images */
55 "/usr/local/lib/portables\0"
56 "/usr/lib/portables\0",
58 [IMAGE_EXTENSION
] = "/etc/extensions\0" /* only place symlinks here */
59 "/run/extensions\0" /* and here too */
60 "/var/lib/extensions\0" /* the main place for images */
61 "/usr/local/lib/extensions\0"
62 "/usr/lib/extensions\0",
65 static Image
*image_free(Image
*i
) {
72 strv_free(i
->machine_info
);
73 strv_free(i
->os_release
);
74 strv_free(i
->extension_release
);
79 DEFINE_TRIVIAL_REF_UNREF_FUNC(Image
, image
, image_free
);
80 DEFINE_HASH_OPS_WITH_VALUE_DESTRUCTOR(image_hash_ops
, char, string_hash_func
, string_compare_func
,
83 static char **image_settings_path(Image
*image
) {
84 _cleanup_strv_free_
char **l
= NULL
;
94 fn
= strjoina(image
->name
, ".nspawn");
96 FOREACH_STRING(s
, "/etc/systemd/nspawn", "/run/systemd/nspawn") {
97 l
[i
] = path_join(s
, fn
);
104 l
[i
] = file_in_same_dir(image
->path
, fn
);
111 static char *image_roothash_path(Image
*image
) {
116 fn
= strjoina(image
->name
, ".roothash");
118 return file_in_same_dir(image
->path
, fn
);
121 static int image_new(
125 const char *filename
,
131 _cleanup_(image_unrefp
) Image
*i
= NULL
;
134 assert(t
< _IMAGE_TYPE_MAX
);
146 .read_only
= read_only
,
150 .usage_exclusive
= UINT64_MAX
,
152 .limit_exclusive
= UINT64_MAX
,
155 i
->name
= strdup(pretty
);
159 i
->path
= path_join(path
, filename
);
163 path_simplify(i
->path
);
170 static int extract_pretty(const char *path
, const char *suffix
, char **ret
) {
171 _cleanup_free_
char *name
= NULL
;
178 p
= last_path_component(path
);
181 name
= strndup(p
, n
);
188 e
= endswith(name
, suffix
);
195 if (!image_name_is_valid(name
))
198 *ret
= TAKE_PTR(name
);
202 static int image_make(
206 const char *filename
,
207 const struct stat
*st
,
210 _cleanup_free_
char *pretty_buffer
= NULL
, *parent
= NULL
;
215 assert(dfd
>= 0 || dfd
== AT_FDCWD
);
216 assert(path
|| dfd
== AT_FDCWD
);
219 /* We explicitly *do* follow symlinks here, since we want to allow symlinking trees, raw files and block
220 * devices into /var/lib/machines/, and treat them normally.
222 * This function returns -ENOENT if we can't find the image after all, and -EMEDIUMTYPE if it's not a file we
226 if (fstatat(dfd
, filename
, &stbuf
, 0) < 0)
234 (void) safe_getcwd(&parent
);
236 (void) fd_get_path(dfd
, &parent
);
240 (path
&& path_startswith(path
, "/usr")) ||
241 (faccessat(dfd
, filename
, W_OK
, AT_EACCESS
) < 0 && errno
== EROFS
);
243 if (S_ISDIR(st
->st_mode
)) {
244 _cleanup_close_
int fd
= -1;
245 unsigned file_attr
= 0;
252 r
= extract_pretty(filename
, NULL
, &pretty_buffer
);
256 pretty
= pretty_buffer
;
259 fd
= openat(dfd
, filename
, O_CLOEXEC
|O_NOCTTY
|O_DIRECTORY
);
263 if (btrfs_might_be_subvol(st
)) {
265 r
= btrfs_is_filesystem(fd
);
269 BtrfsSubvolInfo info
;
271 /* It's a btrfs subvolume */
273 r
= btrfs_subvol_get_info_fd(fd
, 0, &info
);
277 r
= image_new(IMAGE_SUBVOLUME
,
281 info
.read_only
|| read_only
,
288 if (btrfs_quota_scan_ongoing(fd
) == 0) {
289 BtrfsQuotaInfo quota
;
291 r
= btrfs_subvol_get_subtree_quota_fd(fd
, 0, "a
);
293 (*ret
)->usage
= quota
.referenced
;
294 (*ret
)->usage_exclusive
= quota
.exclusive
;
296 (*ret
)->limit
= quota
.referenced_max
;
297 (*ret
)->limit_exclusive
= quota
.exclusive_max
;
305 /* Get directory creation time (not available everywhere, but that's OK */
306 (void) fd_getcrtime(dfd
, &crtime
);
308 /* If the IMMUTABLE bit is set, we consider the directory read-only. Since the ioctl is not
309 * supported everywhere we ignore failures. */
310 (void) read_attr_fd(fd
, &file_attr
);
312 /* It's just a normal directory. */
313 r
= image_new(IMAGE_DIRECTORY
,
317 read_only
|| (file_attr
& FS_IMMUTABLE_FL
),
319 0, /* we don't use mtime of stat() here, since it's not the time of last change of the tree, but only of the top-level dir */
326 } else if (S_ISREG(st
->st_mode
) && endswith(filename
, ".raw")) {
329 /* It's a RAW disk image */
334 (void) fd_getcrtime_at(dfd
, filename
, &crtime
, 0);
337 r
= extract_pretty(filename
, ".raw", &pretty_buffer
);
341 pretty
= pretty_buffer
;
344 r
= image_new(IMAGE_RAW
,
348 !(st
->st_mode
& 0222) || read_only
,
350 timespec_load(&st
->st_mtim
),
355 (*ret
)->usage
= (*ret
)->usage_exclusive
= st
->st_blocks
* 512;
356 (*ret
)->limit
= (*ret
)->limit_exclusive
= st
->st_size
;
360 } else if (S_ISBLK(st
->st_mode
)) {
361 _cleanup_close_
int block_fd
= -1;
362 uint64_t size
= UINT64_MAX
;
370 r
= extract_pretty(filename
, NULL
, &pretty_buffer
);
374 pretty
= pretty_buffer
;
377 block_fd
= openat(dfd
, filename
, O_RDONLY
|O_NONBLOCK
|O_CLOEXEC
|O_NOCTTY
);
379 log_debug_errno(errno
, "Failed to open block device %s/%s, ignoring: %m", path
?: strnull(parent
), filename
);
381 /* Refresh stat data after opening the node */
382 if (fstat(block_fd
, &stbuf
) < 0)
386 if (!S_ISBLK(st
->st_mode
)) /* Verify that what we opened is actually what we think it is */
392 if (ioctl(block_fd
, BLKROGET
, &state
) < 0)
393 log_debug_errno(errno
, "Failed to issue BLKROGET on device %s/%s, ignoring: %m", path
?: strnull(parent
), filename
);
398 if (ioctl(block_fd
, BLKGETSIZE64
, &size
) < 0)
399 log_debug_errno(errno
, "Failed to issue BLKGETSIZE64 on device %s/%s, ignoring: %m", path
?: strnull(parent
), filename
);
401 block_fd
= safe_close(block_fd
);
404 r
= image_new(IMAGE_BLOCK
,
408 !(st
->st_mode
& 0222) || read_only
,
415 if (!IN_SET(size
, 0, UINT64_MAX
))
416 (*ret
)->usage
= (*ret
)->usage_exclusive
= (*ret
)->limit
= (*ret
)->limit_exclusive
= size
;
424 int image_find(ImageClass
class,
433 assert(class < _IMAGE_CLASS_MAX
);
436 /* There are no images with invalid names */
437 if (!image_name_is_valid(name
))
440 NULSTR_FOREACH(path
, image_search_path
[class]) {
441 _cleanup_free_
char *resolved
= NULL
;
442 _cleanup_closedir_
DIR *d
= NULL
;
446 r
= chase_symlinks_and_opendir(path
, root
, CHASE_PREFIX_ROOT
, &resolved
, &d
);
452 /* As mentioned above, we follow symlinks on this fstatat(), because we want to permit people
453 * to symlink block devices into the search path. (For now, we disable that when operating
454 * relative to some root directory.) */
455 flags
= root
? AT_SYMLINK_NOFOLLOW
: 0;
456 if (fstatat(dirfd(d
), name
, &st
, flags
) < 0) {
457 _cleanup_free_
char *raw
= NULL
;
462 raw
= strjoin(name
, ".raw");
466 if (fstatat(dirfd(d
), raw
, &st
, flags
) < 0) {
473 if (!S_ISREG(st
.st_mode
))
476 r
= image_make(name
, dirfd(d
), resolved
, raw
, &st
, ret
);
479 if (!S_ISDIR(st
.st_mode
) && !S_ISBLK(st
.st_mode
))
482 r
= image_make(name
, dirfd(d
), resolved
, name
, &st
, ret
);
484 if (IN_SET(r
, -ENOENT
, -EMEDIUMTYPE
))
490 (*ret
)->discoverable
= true;
495 if (class == IMAGE_MACHINE
&& streq(name
, ".host")) {
496 r
= image_make(".host", AT_FDCWD
, NULL
, empty_to_root(root
), NULL
, ret
);
501 (*ret
)->discoverable
= true;
509 int image_from_path(const char *path
, Image
**ret
) {
511 /* Note that we don't set the 'discoverable' field of the returned object, because we don't check here whether
512 * the image is in the image search path. And if it is we don't know if the path we used is actually not
513 * overridden by another, different image earlier in the search path */
515 if (path_equal(path
, "/"))
516 return image_make(".host", AT_FDCWD
, NULL
, "/", NULL
, ret
);
518 return image_make(NULL
, AT_FDCWD
, NULL
, path
, NULL
, ret
);
521 int image_find_harder(ImageClass
class, const char *name_or_path
, const char *root
, Image
**ret
) {
522 if (image_name_is_valid(name_or_path
))
523 return image_find(class, name_or_path
, root
, ret
);
525 return image_from_path(name_or_path
, ret
);
537 assert(class < _IMAGE_CLASS_MAX
);
540 NULSTR_FOREACH(path
, image_search_path
[class]) {
541 _cleanup_free_
char *resolved
= NULL
;
542 _cleanup_closedir_
DIR *d
= NULL
;
545 r
= chase_symlinks_and_opendir(path
, root
, CHASE_PREFIX_ROOT
, &resolved
, &d
);
551 FOREACH_DIRENT_ALL(de
, d
, return -errno
) {
552 _cleanup_(image_unrefp
) Image
*image
= NULL
;
553 _cleanup_free_
char *truncated
= NULL
;
558 if (dot_or_dot_dot(de
->d_name
))
561 /* As mentioned above, we follow symlinks on this fstatat(), because we want to
562 * permit people to symlink block devices into the search path. */
563 flags
= root
? AT_SYMLINK_NOFOLLOW
: 0;
564 if (fstatat(dirfd(d
), de
->d_name
, &st
, flags
) < 0) {
571 if (S_ISREG(st
.st_mode
)) {
574 e
= endswith(de
->d_name
, ".raw");
578 truncated
= strndup(de
->d_name
, e
- de
->d_name
);
583 } else if (S_ISDIR(st
.st_mode
) || S_ISBLK(st
.st_mode
))
588 if (!image_name_is_valid(pretty
))
591 if (hashmap_contains(h
, pretty
))
594 r
= image_make(pretty
, dirfd(d
), resolved
, de
->d_name
, &st
, &image
);
595 if (IN_SET(r
, -ENOENT
, -EMEDIUMTYPE
))
600 image
->discoverable
= true;
602 r
= hashmap_put(h
, image
->name
, image
);
610 if (class == IMAGE_MACHINE
&& !hashmap_contains(h
, ".host")) {
611 _cleanup_(image_unrefp
) Image
*image
= NULL
;
613 r
= image_make(".host", AT_FDCWD
, NULL
, empty_to_root("/"), NULL
, &image
);
617 image
->discoverable
= true;
619 r
= hashmap_put(h
, image
->name
, image
);
629 int image_remove(Image
*i
) {
630 _cleanup_(release_lock_file
) LockFile global_lock
= LOCK_FILE_INIT
, local_lock
= LOCK_FILE_INIT
;
631 _cleanup_strv_free_
char **settings
= NULL
;
632 _cleanup_free_
char *roothash
= NULL
;
638 if (IMAGE_IS_VENDOR(i
) || IMAGE_IS_HOST(i
))
641 settings
= image_settings_path(i
);
645 roothash
= image_roothash_path(i
);
649 /* Make sure we don't interfere with a running nspawn */
650 r
= image_path_lock(i
->path
, LOCK_EX
|LOCK_NB
, &global_lock
, &local_lock
);
656 case IMAGE_SUBVOLUME
:
658 /* Let's unlink first, maybe it is a symlink? If that works we are happy. Otherwise, let's get out the
660 if (unlink(i
->path
) < 0) {
661 r
= btrfs_subvol_remove(i
->path
, BTRFS_REMOVE_RECURSIVE
|BTRFS_REMOVE_QUOTA
);
668 case IMAGE_DIRECTORY
:
669 /* Allow deletion of read-only directories */
670 (void) chattr_path(i
->path
, 0, FS_IMMUTABLE_FL
, NULL
);
671 r
= rm_rf(i
->path
, REMOVE_ROOT
|REMOVE_PHYSICAL
|REMOVE_SUBVOLUME
);
679 /* If this is inside of /dev, then it's a real block device, hence let's not touch the device node
680 * itself (but let's remove the stuff stored alongside it). If it's anywhere else, let's try to unlink
681 * the thing (it's most likely a symlink after all). */
683 if (path_startswith(i
->path
, "/dev"))
688 if (unlink(i
->path
) < 0)
696 STRV_FOREACH(j
, settings
) {
697 if (unlink(*j
) < 0 && errno
!= ENOENT
)
698 log_debug_errno(errno
, "Failed to unlink %s, ignoring: %m", *j
);
701 if (unlink(roothash
) < 0 && errno
!= ENOENT
)
702 log_debug_errno(errno
, "Failed to unlink %s, ignoring: %m", roothash
);
707 static int rename_auxiliary_file(const char *path
, const char *new_name
, const char *suffix
) {
708 _cleanup_free_
char *rs
= NULL
;
711 fn
= strjoina(new_name
, suffix
);
713 rs
= file_in_same_dir(path
, fn
);
717 return rename_noreplace(AT_FDCWD
, path
, AT_FDCWD
, rs
);
720 int image_rename(Image
*i
, const char *new_name
) {
721 _cleanup_(release_lock_file
) LockFile global_lock
= LOCK_FILE_INIT
, local_lock
= LOCK_FILE_INIT
, name_lock
= LOCK_FILE_INIT
;
722 _cleanup_free_
char *new_path
= NULL
, *nn
= NULL
, *roothash
= NULL
;
723 _cleanup_strv_free_
char **settings
= NULL
;
724 unsigned file_attr
= 0;
730 if (!image_name_is_valid(new_name
))
733 if (IMAGE_IS_VENDOR(i
) || IMAGE_IS_HOST(i
))
736 settings
= image_settings_path(i
);
740 roothash
= image_roothash_path(i
);
744 /* Make sure we don't interfere with a running nspawn */
745 r
= image_path_lock(i
->path
, LOCK_EX
|LOCK_NB
, &global_lock
, &local_lock
);
749 /* Make sure nobody takes the new name, between the time we
750 * checked it is currently unused in all search paths, and the
751 * time we take possession of it */
752 r
= image_name_lock(new_name
, LOCK_EX
|LOCK_NB
, &name_lock
);
756 r
= image_find(IMAGE_MACHINE
, new_name
, NULL
, NULL
);
764 case IMAGE_DIRECTORY
:
765 /* Turn of the immutable bit while we rename the image, so that we can rename it */
766 (void) read_attr_path(i
->path
, &file_attr
);
768 if (file_attr
& FS_IMMUTABLE_FL
)
769 (void) chattr_path(i
->path
, 0, FS_IMMUTABLE_FL
, NULL
);
772 case IMAGE_SUBVOLUME
:
773 new_path
= file_in_same_dir(i
->path
, new_name
);
778 /* Refuse renaming raw block devices in /dev, the names are picked by udev after all. */
779 if (path_startswith(i
->path
, "/dev"))
782 new_path
= file_in_same_dir(i
->path
, new_name
);
788 fn
= strjoina(new_name
, ".raw");
789 new_path
= file_in_same_dir(i
->path
, fn
);
800 nn
= strdup(new_name
);
804 r
= rename_noreplace(AT_FDCWD
, i
->path
, AT_FDCWD
, new_path
);
808 /* Restore the immutable bit, if it was set before */
809 if (file_attr
& FS_IMMUTABLE_FL
)
810 (void) chattr_path(new_path
, FS_IMMUTABLE_FL
, FS_IMMUTABLE_FL
, NULL
);
812 free_and_replace(i
->path
, new_path
);
813 free_and_replace(i
->name
, nn
);
815 STRV_FOREACH(j
, settings
) {
816 r
= rename_auxiliary_file(*j
, new_name
, ".nspawn");
817 if (r
< 0 && r
!= -ENOENT
)
818 log_debug_errno(r
, "Failed to rename settings file %s, ignoring: %m", *j
);
821 r
= rename_auxiliary_file(roothash
, new_name
, ".roothash");
822 if (r
< 0 && r
!= -ENOENT
)
823 log_debug_errno(r
, "Failed to rename roothash file %s, ignoring: %m", roothash
);
828 static int clone_auxiliary_file(const char *path
, const char *new_name
, const char *suffix
) {
829 _cleanup_free_
char *rs
= NULL
;
832 fn
= strjoina(new_name
, suffix
);
834 rs
= file_in_same_dir(path
, fn
);
838 return copy_file_atomic(path
, rs
, 0664, 0, 0, COPY_REFLINK
);
841 int image_clone(Image
*i
, const char *new_name
, bool read_only
) {
842 _cleanup_(release_lock_file
) LockFile name_lock
= LOCK_FILE_INIT
;
843 _cleanup_strv_free_
char **settings
= NULL
;
844 _cleanup_free_
char *roothash
= NULL
;
845 const char *new_path
;
851 if (!image_name_is_valid(new_name
))
854 settings
= image_settings_path(i
);
858 roothash
= image_roothash_path(i
);
862 /* Make sure nobody takes the new name, between the time we
863 * checked it is currently unused in all search paths, and the
864 * time we take possession of it */
865 r
= image_name_lock(new_name
, LOCK_EX
|LOCK_NB
, &name_lock
);
869 r
= image_find(IMAGE_MACHINE
, new_name
, NULL
, NULL
);
877 case IMAGE_SUBVOLUME
:
878 case IMAGE_DIRECTORY
:
879 /* If we can we'll always try to create a new btrfs subvolume here, even if the source is a plain
882 new_path
= strjoina("/var/lib/machines/", new_name
);
884 r
= btrfs_subvol_snapshot(i
->path
, new_path
,
885 (read_only
? BTRFS_SNAPSHOT_READ_ONLY
: 0) |
886 BTRFS_SNAPSHOT_FALLBACK_COPY
|
887 BTRFS_SNAPSHOT_FALLBACK_DIRECTORY
|
888 BTRFS_SNAPSHOT_FALLBACK_IMMUTABLE
|
889 BTRFS_SNAPSHOT_RECURSIVE
|
890 BTRFS_SNAPSHOT_QUOTA
);
892 /* Enable "subtree" quotas for the copy, if we didn't copy any quota from the source. */
893 (void) btrfs_subvol_auto_qgroup(new_path
, 0, true);
898 new_path
= strjoina("/var/lib/machines/", new_name
, ".raw");
900 r
= copy_file_atomic(i
->path
, new_path
, read_only
? 0444 : 0644, FS_NOCOW_FL
, FS_NOCOW_FL
, COPY_REFLINK
|COPY_CRTIME
);
911 STRV_FOREACH(j
, settings
) {
912 r
= clone_auxiliary_file(*j
, new_name
, ".nspawn");
913 if (r
< 0 && r
!= -ENOENT
)
914 log_debug_errno(r
, "Failed to clone settings %s, ignoring: %m", *j
);
917 r
= clone_auxiliary_file(roothash
, new_name
, ".roothash");
918 if (r
< 0 && r
!= -ENOENT
)
919 log_debug_errno(r
, "Failed to clone root hash file %s, ignoring: %m", roothash
);
924 int image_read_only(Image
*i
, bool b
) {
925 _cleanup_(release_lock_file
) LockFile global_lock
= LOCK_FILE_INIT
, local_lock
= LOCK_FILE_INIT
;
930 if (IMAGE_IS_VENDOR(i
) || IMAGE_IS_HOST(i
))
933 /* Make sure we don't interfere with a running nspawn */
934 r
= image_path_lock(i
->path
, LOCK_EX
|LOCK_NB
, &global_lock
, &local_lock
);
940 case IMAGE_SUBVOLUME
:
942 /* Note that we set the flag only on the top-level
943 * subvolume of the image. */
945 r
= btrfs_subvol_set_read_only(i
->path
, b
);
951 case IMAGE_DIRECTORY
:
952 /* For simple directory trees we cannot use the access
953 mode of the top-level directory, since it has an
954 effect on the container itself. However, we can
955 use the "immutable" flag, to at least make the
956 top-level directory read-only. It's not as good as
957 a read-only subvolume, but at least something, and
958 we can read the value back. */
960 r
= chattr_path(i
->path
, b
? FS_IMMUTABLE_FL
: 0, FS_IMMUTABLE_FL
, NULL
);
969 if (stat(i
->path
, &st
) < 0)
972 if (chmod(i
->path
, (st
.st_mode
& 0444) | (b
? 0000 : 0200)) < 0)
975 /* If the images is now read-only, it's a good time to
976 * defrag it, given that no write patterns will
977 * fragment it again. */
979 (void) btrfs_defrag(i
->path
);
984 _cleanup_close_
int fd
= -1;
988 fd
= open(i
->path
, O_CLOEXEC
|O_RDONLY
|O_NONBLOCK
|O_NOCTTY
);
992 if (fstat(fd
, &st
) < 0)
994 if (!S_ISBLK(st
.st_mode
))
997 if (ioctl(fd
, BLKROSET
, &state
) < 0)
1010 int image_path_lock(const char *path
, int operation
, LockFile
*global
, LockFile
*local
) {
1011 _cleanup_free_
char *p
= NULL
;
1012 LockFile t
= LOCK_FILE_INIT
;
1021 /* Locks an image path. This actually creates two locks: one "local" one, next to the image path
1022 * itself, which might be shared via NFS. And another "global" one, in /run, that uses the
1023 * device/inode number. This has the benefit that we can even lock a tree that is a mount point,
1026 if (!path_is_absolute(path
))
1029 switch (operation
& (LOCK_SH
|LOCK_EX
)) {
1040 if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1041 *local
= *global
= (LockFile
) LOCK_FILE_INIT
;
1045 /* Prohibit taking exclusive locks on the host image. We can't allow this, since we ourselves are
1046 * running off it after all, and we don't want any images to manipulate the host image. We make an
1047 * exception for shared locks however: we allow those (and make them NOPs since there's no point in
1048 * taking them if there can't be exclusive locks). Strictly speaking these are questionable as well,
1049 * since it means changes made to the host might propagate to the container as they happen (and a
1050 * shared lock kinda suggests that no changes happen at all while it is in place), but it's too
1051 * useful not to allow read-only containers off the host root, hence let's support this, and trust
1052 * the user to do the right thing with this. */
1053 if (path_equal(path
, "/")) {
1057 *local
= *global
= (LockFile
) LOCK_FILE_INIT
;
1061 if (stat(path
, &st
) >= 0) {
1062 if (S_ISBLK(st
.st_mode
))
1063 r
= asprintf(&p
, "/run/systemd/nspawn/locks/block-%u:%u", major(st
.st_rdev
), minor(st
.st_rdev
));
1064 else if (S_ISDIR(st
.st_mode
) || S_ISREG(st
.st_mode
))
1065 r
= asprintf(&p
, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st
.st_dev
, (unsigned long) st
.st_ino
);
1072 /* For block devices we don't need the "local" lock, as the major/minor lock above should be
1073 * sufficient, since block devices are host local anyway. */
1074 if (!path_startswith(path
, "/dev/")) {
1075 r
= make_lock_file_for(path
, operation
, &t
);
1077 if (!exclusive
&& r
== -EROFS
)
1078 log_debug_errno(r
, "Failed to create shared lock for '%s', ignoring: %m", path
);
1085 (void) mkdir_p("/run/systemd/nspawn/locks", 0700);
1087 r
= make_lock_file(p
, operation
, global
);
1089 release_lock_file(&t
);
1093 *global
= (LockFile
) LOCK_FILE_INIT
;
1099 int image_set_limit(Image
*i
, uint64_t referenced_max
) {
1102 if (IMAGE_IS_VENDOR(i
) || IMAGE_IS_HOST(i
))
1105 if (i
->type
!= IMAGE_SUBVOLUME
)
1108 /* We set the quota both for the subvolume as well as for the
1109 * subtree. The latter is mostly for historical reasons, since
1110 * we didn't use to have a concept of subtree quota, and hence
1111 * only modified the subvolume quota. */
1113 (void) btrfs_qgroup_set_limit(i
->path
, 0, referenced_max
);
1114 (void) btrfs_subvol_auto_qgroup(i
->path
, 0, true);
1115 return btrfs_subvol_set_subtree_quota_limit(i
->path
, 0, referenced_max
);
1118 int image_read_metadata(Image
*i
) {
1119 _cleanup_(release_lock_file
) LockFile global_lock
= LOCK_FILE_INIT
, local_lock
= LOCK_FILE_INIT
;
1124 r
= image_path_lock(i
->path
, LOCK_SH
|LOCK_NB
, &global_lock
, &local_lock
);
1130 case IMAGE_SUBVOLUME
:
1131 case IMAGE_DIRECTORY
: {
1132 _cleanup_strv_free_
char **machine_info
= NULL
, **os_release
= NULL
, **extension_release
= NULL
;
1133 sd_id128_t machine_id
= SD_ID128_NULL
;
1134 _cleanup_free_
char *hostname
= NULL
;
1135 _cleanup_free_
char *path
= NULL
;
1137 r
= chase_symlinks("/etc/hostname", i
->path
, CHASE_PREFIX_ROOT
|CHASE_TRAIL_SLASH
, &path
, NULL
);
1138 if (r
< 0 && r
!= -ENOENT
)
1139 log_debug_errno(r
, "Failed to chase /etc/hostname in image %s: %m", i
->name
);
1141 r
= read_etc_hostname(path
, &hostname
);
1143 log_debug_errno(errno
, "Failed to read /etc/hostname of image %s: %m", i
->name
);
1148 r
= chase_symlinks("/etc/machine-id", i
->path
, CHASE_PREFIX_ROOT
|CHASE_TRAIL_SLASH
, &path
, NULL
);
1149 if (r
< 0 && r
!= -ENOENT
)
1150 log_debug_errno(r
, "Failed to chase /etc/machine-id in image %s: %m", i
->name
);
1152 _cleanup_close_
int fd
= -1;
1154 fd
= open(path
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
1156 log_debug_errno(errno
, "Failed to open %s: %m", path
);
1158 r
= id128_read_fd(fd
, ID128_PLAIN
, &machine_id
);
1160 log_debug_errno(r
, "Image %s contains invalid machine ID.", i
->name
);
1166 r
= chase_symlinks("/etc/machine-info", i
->path
, CHASE_PREFIX_ROOT
|CHASE_TRAIL_SLASH
, &path
, NULL
);
1167 if (r
< 0 && r
!= -ENOENT
)
1168 log_debug_errno(r
, "Failed to chase /etc/machine-info in image %s: %m", i
->name
);
1170 r
= load_env_file_pairs(NULL
, path
, &machine_info
);
1172 log_debug_errno(r
, "Failed to parse machine-info data of %s: %m", i
->name
);
1175 r
= load_os_release_pairs(i
->path
, &os_release
);
1177 log_debug_errno(r
, "Failed to read os-release in image, ignoring: %m");
1179 r
= load_extension_release_pairs(i
->path
, i
->name
, &extension_release
);
1181 log_debug_errno(r
, "Failed to read extension-release in image, ignoring: %m");
1183 free_and_replace(i
->hostname
, hostname
);
1184 i
->machine_id
= machine_id
;
1185 strv_free_and_replace(i
->machine_info
, machine_info
);
1186 strv_free_and_replace(i
->os_release
, os_release
);
1187 strv_free_and_replace(i
->extension_release
, extension_release
);
1194 _cleanup_(loop_device_unrefp
) LoopDevice
*d
= NULL
;
1195 _cleanup_(dissected_image_unrefp
) DissectedImage
*m
= NULL
;
1197 r
= loop_device_make_by_path(i
->path
, O_RDONLY
, LO_FLAGS_PARTSCAN
, &d
);
1204 d
->uevent_seqnum_not_before
,
1205 d
->timestamp_not_before
,
1206 DISSECT_IMAGE_GENERIC_ROOT
|
1207 DISSECT_IMAGE_REQUIRE_ROOT
|
1208 DISSECT_IMAGE_RELAX_VAR_CHECK
|
1209 DISSECT_IMAGE_USR_NO_ROOT
,
1214 r
= dissected_image_acquire_metadata(m
);
1218 free_and_replace(i
->hostname
, m
->hostname
);
1219 i
->machine_id
= m
->machine_id
;
1220 strv_free_and_replace(i
->machine_info
, m
->machine_info
);
1221 strv_free_and_replace(i
->os_release
, m
->os_release
);
1222 strv_free_and_replace(i
->extension_release
, m
->extension_release
);
1231 i
->metadata_valid
= true;
1236 int image_name_lock(const char *name
, int operation
, LockFile
*ret
) {
1242 /* Locks an image name, regardless of the precise path used. */
1244 if (streq(name
, ".host"))
1247 if (!image_name_is_valid(name
))
1250 if (getenv_bool("SYSTEMD_NSPAWN_LOCK") == 0) {
1251 *ret
= (LockFile
) LOCK_FILE_INIT
;
1255 (void) mkdir_p("/run/systemd/nspawn/locks", 0700);
1257 p
= strjoina("/run/systemd/nspawn/locks/name-", name
);
1258 return make_lock_file(p
, operation
, ret
);
1261 bool image_in_search_path(
1264 const char *image
) {
1270 NULSTR_FOREACH(path
, image_search_path
[class]) {
1274 if (!empty_or_root(root
)) {
1275 q
= path_startswith(path
, root
);
1281 p
= path_startswith(q
, path
);
1285 /* Make sure there's a filename following */
1286 k
= strcspn(p
, "/");
1292 /* Accept trailing slashes */
1293 if (p
[strspn(p
, "/")] == 0)
1301 static const char* const image_type_table
[_IMAGE_TYPE_MAX
] = {
1302 [IMAGE_DIRECTORY
] = "directory",
1303 [IMAGE_SUBVOLUME
] = "subvolume",
1304 [IMAGE_RAW
] = "raw",
1305 [IMAGE_BLOCK
] = "block",
1308 DEFINE_STRING_TABLE_LOOKUP(image_type
, ImageType
);