2 This file is part of systemd.
4 Copyright 2015 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
20 #include <sys/mount.h>
21 #include <linux/magic.h>
23 #include "alloc-util.h"
30 #include "mount-util.h"
31 #include "nspawn-mount.h"
32 #include "parse-util.h"
33 #include "path-util.h"
36 #include "stat-util.h"
37 #include "string-util.h"
39 #include "user-util.h"
42 CustomMount
* custom_mount_add(CustomMount
**l
, unsigned *n
, CustomMountType t
) {
48 assert(t
< _CUSTOM_MOUNT_TYPE_MAX
);
50 c
= realloc(*l
, (*n
+ 1) * sizeof(CustomMount
));
58 *ret
= (CustomMount
) { .type
= t
};
63 void custom_mount_free_all(CustomMount
*l
, unsigned n
) {
66 for (i
= 0; i
< n
; i
++) {
67 CustomMount
*m
= l
+ i
;
74 (void) rm_rf(m
->work_dir
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
84 int custom_mount_compare(const void *a
, const void *b
) {
85 const CustomMount
*x
= a
, *y
= b
;
88 r
= path_compare(x
->destination
, y
->destination
);
92 if (x
->type
< y
->type
)
94 if (x
->type
> y
->type
)
100 int bind_mount_parse(CustomMount
**l
, unsigned *n
, const char *s
, bool read_only
) {
101 _cleanup_free_
char *source
= NULL
, *destination
= NULL
, *opts
= NULL
;
109 r
= extract_many_words(&p
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
, &source
, &destination
, NULL
);
116 destination
= strdup(source
);
121 if (r
== 2 && !isempty(p
)) {
127 if (!path_is_absolute(source
))
130 if (!path_is_absolute(destination
))
133 m
= custom_mount_add(l
, n
, CUSTOM_MOUNT_BIND
);
138 m
->destination
= destination
;
139 m
->read_only
= read_only
;
142 source
= destination
= opts
= NULL
;
146 int tmpfs_mount_parse(CustomMount
**l
, unsigned *n
, const char *s
) {
147 _cleanup_free_
char *path
= NULL
, *opts
= NULL
;
156 r
= extract_first_word(&p
, &path
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
163 opts
= strdup("mode=0755");
169 if (!path_is_absolute(path
))
172 m
= custom_mount_add(l
, n
, CUSTOM_MOUNT_TMPFS
);
176 m
->destination
= path
;
183 static int tmpfs_patch_options(
186 uid_t uid_shift
, uid_t uid_range
,
188 const char *selinux_apifs_context
,
193 if ((userns
&& uid_shift
!= 0) || patch_ids
) {
194 assert(uid_shift
!= UID_INVALID
);
197 (void) asprintf(&buf
, "%s,uid=" UID_FMT
",gid=" UID_FMT
, options
, uid_shift
, uid_shift
);
199 (void) asprintf(&buf
, "uid=" UID_FMT
",gid=" UID_FMT
, uid_shift
, uid_shift
);
207 if (selinux_apifs_context
) {
211 t
= strjoin(options
, ",context=\"", selinux_apifs_context
, "\"", NULL
);
213 t
= strjoin("context=\"", selinux_apifs_context
, "\"", NULL
);
224 if (!buf
&& options
) {
225 buf
= strdup(options
);
234 int mount_sysfs(const char *dest
) {
235 const char *full
, *top
, *x
;
238 top
= prefix_roota(dest
, "/sys");
239 r
= path_check_fstype(top
, SYSFS_MAGIC
);
241 return log_error_errno(r
, "Failed to determine filesystem type of %s: %m", top
);
242 /* /sys might already be mounted as sysfs by the outer child in the
243 * !netns case. In this case, it's all good. Don't touch it because we
244 * don't have the right to do so, see https://github.com/systemd/systemd/issues/1555.
249 full
= prefix_roota(top
, "/full");
251 (void) mkdir(full
, 0755);
253 r
= mount_verbose(LOG_ERR
, "sysfs", full
, "sysfs",
254 MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
258 FOREACH_STRING(x
, "block", "bus", "class", "dev", "devices", "kernel") {
259 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
261 from
= prefix_root(full
, x
);
265 to
= prefix_root(top
, x
);
269 (void) mkdir(to
, 0755);
271 r
= mount_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
275 r
= mount_verbose(LOG_ERR
, NULL
, to
, NULL
,
276 MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_REMOUNT
, NULL
);
281 r
= umount_verbose(full
);
286 return log_error_errno(errno
, "Failed to remove %s: %m", full
);
288 x
= prefix_roota(top
, "/fs/kdbus");
289 (void) mkdir_p(x
, 0755);
291 /* Create mountpoint for cgroups. Otherwise we are not allowed since we
292 * remount /sys read-only.
294 if (cg_ns_supported()) {
295 x
= prefix_roota(top
, "/fs/cgroup");
296 (void) mkdir_p(x
, 0755);
299 return mount_verbose(LOG_ERR
, NULL
, top
, NULL
,
300 MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_REMOUNT
, NULL
);
303 static int mkdir_userns(const char *path
, mode_t mode
, bool in_userns
, uid_t uid_shift
) {
308 r
= mkdir(path
, mode
);
309 if (r
< 0 && errno
!= EEXIST
)
313 r
= lchown(path
, uid_shift
, uid_shift
);
321 static int mkdir_userns_p(const char *prefix
, const char *path
, mode_t mode
, bool in_userns
, uid_t uid_shift
) {
327 if (prefix
&& !path_startswith(path
, prefix
))
330 /* create every parent directory in the path, except the last component */
331 p
= path
+ strspn(path
, "/");
333 char t
[strlen(path
) + 1];
335 e
= p
+ strcspn(p
, "/");
336 p
= e
+ strspn(e
, "/");
338 /* Is this the last component? If so, then we're done */
342 memcpy(t
, path
, e
- path
);
345 if (prefix
&& path_startswith(prefix
, t
))
348 r
= mkdir_userns(t
, mode
, in_userns
, uid_shift
);
353 return mkdir_userns(path
, mode
, in_userns
, uid_shift
);
356 int mount_all(const char *dest
,
357 bool use_userns
, bool in_userns
,
359 uid_t uid_shift
, uid_t uid_range
,
360 const char *selinux_apifs_context
) {
362 typedef struct MountPoint
{
373 static const MountPoint mount_table
[] = {
374 { "proc", "/proc", "proc", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, true, true, false },
375 { "/proc/sys", "/proc/sys", NULL
, NULL
, MS_BIND
, true, true, false }, /* Bind mount first ...*/
376 { "/proc/sys/net", "/proc/sys/net", NULL
, NULL
, MS_BIND
, true, true, true }, /* (except for this) */
377 { NULL
, "/proc/sys", NULL
, NULL
, MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_REMOUNT
, true, true, false }, /* ... then, make it r/o */
378 { "/proc/sysrq-trigger", "/proc/sysrq-trigger", NULL
, NULL
, MS_BIND
, false, true, false }, /* Bind mount first ...*/
379 { NULL
, "/proc/sysrq-trigger", NULL
, NULL
, MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_REMOUNT
, false, true, false }, /* ... then, make it r/o */
380 { "tmpfs", "/sys", "tmpfs", "mode=755", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, true, false, true },
381 { "sysfs", "/sys", "sysfs", NULL
, MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, true, false, false },
382 { "tmpfs", "/dev", "tmpfs", "mode=755", MS_NOSUID
|MS_STRICTATIME
, true, false, false },
383 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
, true, false, false },
384 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
, true, false, false },
385 { "tmpfs", "/tmp", "tmpfs", "mode=1777", MS_STRICTATIME
, true, true, false },
387 { "/sys/fs/selinux", "/sys/fs/selinux", NULL
, NULL
, MS_BIND
, false, false, false }, /* Bind mount first */
388 { NULL
, "/sys/fs/selinux", NULL
, NULL
, MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_REMOUNT
, false, false, false }, /* Then, make it r/o */
395 for (k
= 0; k
< ELEMENTSOF(mount_table
); k
++) {
396 _cleanup_free_
char *where
= NULL
, *options
= NULL
;
399 if (in_userns
!= mount_table
[k
].in_userns
)
402 if (!use_netns
&& mount_table
[k
].use_netns
)
405 where
= prefix_root(dest
, mount_table
[k
].where
);
409 r
= path_is_mount_point(where
, AT_SYMLINK_FOLLOW
);
410 if (r
< 0 && r
!= -ENOENT
)
411 return log_error_errno(r
, "Failed to detect whether %s is a mount point: %m", where
);
413 /* Skip this entry if it is not a remount. */
414 if (mount_table
[k
].what
&& r
> 0)
417 r
= mkdir_userns_p(dest
, where
, 0755, in_userns
, uid_shift
);
418 if (r
< 0 && r
!= -EEXIST
) {
419 if (mount_table
[k
].fatal
)
420 return log_error_errno(r
, "Failed to create directory %s: %m", where
);
422 log_debug_errno(r
, "Failed to create directory %s: %m", where
);
426 o
= mount_table
[k
].options
;
427 if (streq_ptr(mount_table
[k
].type
, "tmpfs")) {
429 r
= tmpfs_patch_options(o
, use_userns
, 0, uid_range
, true, selinux_apifs_context
, &options
);
431 r
= tmpfs_patch_options(o
, use_userns
, uid_shift
, uid_range
, false, selinux_apifs_context
, &options
);
438 r
= mount_verbose(mount_table
[k
].fatal
? LOG_ERR
: LOG_WARNING
,
442 mount_table
[k
].flags
,
444 if (r
< 0 && mount_table
[k
].fatal
)
451 static int parse_mount_bind_options(const char *options
, unsigned long *mount_flags
, char **mount_opts
) {
452 const char *p
= options
;
453 unsigned long flags
= *mount_flags
;
459 _cleanup_free_
char *word
= NULL
;
460 int r
= extract_first_word(&p
, &word
, ",", 0);
462 return log_error_errno(r
, "Failed to extract mount option: %m");
466 if (streq(word
, "rbind"))
468 else if (streq(word
, "norbind"))
471 log_error("Invalid bind mount option: %s", word
);
476 *mount_flags
= flags
;
477 /* in the future mount_opts will hold string options for mount(2) */
483 static int mount_bind(const char *dest
, CustomMount
*m
) {
484 struct stat source_st
, dest_st
;
486 unsigned long mount_flags
= MS_BIND
| MS_REC
;
487 _cleanup_free_
char *mount_opts
= NULL
;
493 r
= parse_mount_bind_options(m
->options
, &mount_flags
, &mount_opts
);
498 if (stat(m
->source
, &source_st
) < 0)
499 return log_error_errno(errno
, "Failed to stat %s: %m", m
->source
);
501 where
= prefix_roota(dest
, m
->destination
);
503 if (stat(where
, &dest_st
) >= 0) {
504 if (S_ISDIR(source_st
.st_mode
) && !S_ISDIR(dest_st
.st_mode
)) {
505 log_error("Cannot bind mount directory %s on file %s.", m
->source
, where
);
509 if (!S_ISDIR(source_st
.st_mode
) && S_ISDIR(dest_st
.st_mode
)) {
510 log_error("Cannot bind mount file %s on directory %s.", m
->source
, where
);
514 } else if (errno
== ENOENT
) {
515 r
= mkdir_parents_label(where
, 0755);
517 return log_error_errno(r
, "Failed to make parents of %s: %m", where
);
519 /* Create the mount point. Any non-directory file can be
520 * mounted on any non-directory file (regular, fifo, socket,
523 if (S_ISDIR(source_st
.st_mode
))
524 r
= mkdir_label(where
, 0755);
528 return log_error_errno(r
, "Failed to create mount point %s: %m", where
);
531 return log_error_errno(errno
, "Failed to stat %s: %m", where
);
533 r
= mount_verbose(LOG_ERR
, m
->source
, where
, NULL
, mount_flags
, mount_opts
);
538 r
= bind_remount_recursive(where
, true, NULL
);
540 return log_error_errno(r
, "Read-only bind mount failed: %m");
546 static int mount_tmpfs(
549 bool userns
, uid_t uid_shift
, uid_t uid_range
,
550 const char *selinux_apifs_context
) {
552 const char *where
, *options
;
553 _cleanup_free_
char *buf
= NULL
;
559 where
= prefix_roota(dest
, m
->destination
);
561 r
= mkdir_p_label(where
, 0755);
562 if (r
< 0 && r
!= -EEXIST
)
563 return log_error_errno(r
, "Creating mount point for tmpfs %s failed: %m", where
);
565 r
= tmpfs_patch_options(m
->options
, userns
, uid_shift
, uid_range
, false, selinux_apifs_context
, &buf
);
568 options
= r
> 0 ? buf
: m
->options
;
570 return mount_verbose(LOG_ERR
, "tmpfs", where
, "tmpfs", MS_NODEV
|MS_STRICTATIME
, options
);
573 static char *joined_and_escaped_lower_dirs(char * const *lower
) {
574 _cleanup_strv_free_
char **sv
= NULL
;
576 sv
= strv_copy(lower
);
582 if (!strv_shell_escape(sv
, ",:"))
585 return strv_join(sv
, ":");
588 static int mount_overlay(const char *dest
, CustomMount
*m
) {
589 _cleanup_free_
char *lower
= NULL
;
590 const char *where
, *options
;
596 where
= prefix_roota(dest
, m
->destination
);
598 r
= mkdir_label(where
, 0755);
599 if (r
< 0 && r
!= -EEXIST
)
600 return log_error_errno(r
, "Creating mount point for overlay %s failed: %m", where
);
602 (void) mkdir_p_label(m
->source
, 0755);
604 lower
= joined_and_escaped_lower_dirs(m
->lower
);
609 _cleanup_free_
char *escaped_source
= NULL
;
611 escaped_source
= shell_escape(m
->source
, ",:");
615 options
= strjoina("lowerdir=", escaped_source
, ":", lower
);
617 _cleanup_free_
char *escaped_source
= NULL
, *escaped_work_dir
= NULL
;
620 (void) mkdir_label(m
->work_dir
, 0700);
622 escaped_source
= shell_escape(m
->source
, ",:");
625 escaped_work_dir
= shell_escape(m
->work_dir
, ",:");
626 if (!escaped_work_dir
)
629 options
= strjoina("lowerdir=", lower
, ",upperdir=", escaped_source
, ",workdir=", escaped_work_dir
);
632 return mount_verbose(LOG_ERR
, "overlay", where
, "overlay", m
->read_only
? MS_RDONLY
: 0, options
);
637 CustomMount
*mounts
, unsigned n
,
638 bool userns
, uid_t uid_shift
, uid_t uid_range
,
639 const char *selinux_apifs_context
) {
646 for (i
= 0; i
< n
; i
++) {
647 CustomMount
*m
= mounts
+ i
;
651 case CUSTOM_MOUNT_BIND
:
652 r
= mount_bind(dest
, m
);
655 case CUSTOM_MOUNT_TMPFS
:
656 r
= mount_tmpfs(dest
, m
, userns
, uid_shift
, uid_range
, selinux_apifs_context
);
659 case CUSTOM_MOUNT_OVERLAY
:
660 r
= mount_overlay(dest
, m
);
664 assert_not_reached("Unknown custom mount type");
674 /* Retrieve existing subsystems. This function is called in a new cgroup
677 static int get_controllers(Set
*subsystems
) {
678 _cleanup_fclose_
FILE *f
= NULL
;
683 f
= fopen("/proc/self/cgroup", "re");
685 return errno
== ENOENT
? -ESRCH
: -errno
;
687 FOREACH_LINE(line
, f
, return -errno
) {
691 l
= strchr(line
, ':');
702 if (STR_IN_SET(l
, "", "name=systemd"))
709 r
= set_consume(subsystems
, p
);
717 static int mount_legacy_cgroup_hierarchy(const char *dest
, const char *controller
, const char *hierarchy
,
718 CGroupUnified unified_requested
, bool read_only
) {
719 const char *to
, *fstype
, *opts
;
722 to
= strjoina(strempty(dest
), "/sys/fs/cgroup/", hierarchy
);
724 r
= path_is_mount_point(to
, 0);
725 if (r
< 0 && r
!= -ENOENT
)
726 return log_error_errno(r
, "Failed to determine if %s is mounted already: %m", to
);
732 /* The superblock mount options of the mount point need to be
733 * identical to the hosts', and hence writable... */
734 if (streq(controller
, SYSTEMD_CGROUP_CONTROLLER
)) {
735 if (unified_requested
>= CGROUP_UNIFIED_SYSTEMD
) {
740 opts
= "none,name=systemd,xattr";
747 r
= mount_verbose(LOG_ERR
, "cgroup", to
, fstype
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, opts
);
751 /* ... hence let's only make the bind mount read-only, not the superblock. */
753 r
= mount_verbose(LOG_ERR
, NULL
, to
, NULL
,
754 MS_BIND
|MS_REMOUNT
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_RDONLY
, NULL
);
762 /* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */
763 static int mount_legacy_cgns_supported(
764 CGroupUnified unified_requested
, bool userns
, uid_t uid_shift
,
765 uid_t uid_range
, const char *selinux_apifs_context
) {
766 _cleanup_set_free_free_ Set
*controllers
= NULL
;
767 const char *cgroup_root
= "/sys/fs/cgroup", *c
;
770 (void) mkdir_p(cgroup_root
, 0755);
772 /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
773 r
= path_is_mount_point(cgroup_root
, AT_SYMLINK_FOLLOW
);
775 return log_error_errno(r
, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
777 _cleanup_free_
char *options
= NULL
;
779 /* When cgroup namespaces are enabled and user namespaces are
780 * used then the mount of the cgroupfs is done *inside* the new
781 * user namespace. We're root in the new user namespace and the
782 * kernel will happily translate our uid/gid to the correct
783 * uid/gid as seen from e.g. /proc/1/mountinfo. So we simply
784 * pass uid 0 and not uid_shift to tmpfs_patch_options().
786 r
= tmpfs_patch_options("mode=755", userns
, 0, uid_range
, true, selinux_apifs_context
, &options
);
790 r
= mount_verbose(LOG_ERR
, "tmpfs", cgroup_root
, "tmpfs",
791 MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
, options
);
796 if (cg_all_unified() > 0)
797 goto skip_controllers
;
799 controllers
= set_new(&string_hash_ops
);
803 r
= get_controllers(controllers
);
805 return log_error_errno(r
, "Failed to determine cgroup controllers: %m");
808 _cleanup_free_
const char *controller
= NULL
;
810 controller
= set_steal_first(controllers
);
814 r
= mount_legacy_cgroup_hierarchy("", controller
, controller
, unified_requested
, !userns
);
818 /* When multiple hierarchies are co-mounted, make their
819 * constituting individual hierarchies a symlink to the
824 _cleanup_free_
char *target
= NULL
, *tok
= NULL
;
826 r
= extract_first_word(&c
, &tok
, ",", 0);
828 return log_error_errno(r
, "Failed to extract co-mounted cgroup controller: %m");
832 target
= prefix_root("/sys/fs/cgroup", tok
);
836 if (streq(controller
, tok
))
839 r
= symlink_idempotent(controller
, target
);
841 return log_error_errno(r
, "Invalid existing symlink for combined hierarchy: %m");
843 return log_error_errno(r
, "Failed to create symlink for combined hierarchy: %m");
848 r
= mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER
, "systemd", unified_requested
, false);
853 return mount_verbose(LOG_ERR
, NULL
, cgroup_root
, NULL
,
854 MS_REMOUNT
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
|MS_RDONLY
, "mode=755");
859 /* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */
860 static int mount_legacy_cgns_unsupported(
862 CGroupUnified unified_requested
, bool userns
, uid_t uid_shift
, uid_t uid_range
,
863 const char *selinux_apifs_context
) {
864 _cleanup_set_free_free_ Set
*controllers
= NULL
;
865 const char *cgroup_root
;
868 cgroup_root
= prefix_roota(dest
, "/sys/fs/cgroup");
870 (void) mkdir_p(cgroup_root
, 0755);
872 /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
873 r
= path_is_mount_point(cgroup_root
, AT_SYMLINK_FOLLOW
);
875 return log_error_errno(r
, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
877 _cleanup_free_
char *options
= NULL
;
879 r
= tmpfs_patch_options("mode=755", userns
, uid_shift
, uid_range
, false, selinux_apifs_context
, &options
);
883 r
= mount_verbose(LOG_ERR
, "tmpfs", cgroup_root
, "tmpfs",
884 MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
, options
);
889 if (cg_all_unified() > 0)
890 goto skip_controllers
;
892 controllers
= set_new(&string_hash_ops
);
896 r
= cg_kernel_controllers(controllers
);
898 return log_error_errno(r
, "Failed to determine cgroup controllers: %m");
901 _cleanup_free_
char *controller
= NULL
, *origin
= NULL
, *combined
= NULL
;
903 controller
= set_steal_first(controllers
);
907 origin
= prefix_root("/sys/fs/cgroup/", controller
);
911 r
= readlink_malloc(origin
, &combined
);
913 /* Not a symbolic link, but directly a single cgroup hierarchy */
915 r
= mount_legacy_cgroup_hierarchy(dest
, controller
, controller
, unified_requested
, true);
920 return log_error_errno(r
, "Failed to read link %s: %m", origin
);
922 _cleanup_free_
char *target
= NULL
;
924 target
= prefix_root(dest
, origin
);
928 /* A symbolic link, a combination of controllers in one hierarchy */
930 if (!filename_is_valid(combined
)) {
931 log_warning("Ignoring invalid combined hierarchy %s.", combined
);
935 r
= mount_legacy_cgroup_hierarchy(dest
, combined
, combined
, unified_requested
, true);
939 r
= symlink_idempotent(combined
, target
);
941 return log_error_errno(r
, "Invalid existing symlink for combined hierarchy: %m");
943 return log_error_errno(r
, "Failed to create symlink for combined hierarchy: %m");
948 r
= mount_legacy_cgroup_hierarchy(dest
, SYSTEMD_CGROUP_CONTROLLER
, "systemd", unified_requested
, false);
952 return mount_verbose(LOG_ERR
, NULL
, cgroup_root
, NULL
,
953 MS_REMOUNT
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
|MS_RDONLY
, "mode=755");
956 static int mount_unified_cgroups(const char *dest
) {
962 p
= prefix_roota(dest
, "/sys/fs/cgroup");
964 (void) mkdir_p(p
, 0755);
966 r
= path_is_mount_point(p
, AT_SYMLINK_FOLLOW
);
968 return log_error_errno(r
, "Failed to determine if %s is mounted already: %m", p
);
970 p
= prefix_roota(dest
, "/sys/fs/cgroup/cgroup.procs");
971 if (access(p
, F_OK
) >= 0)
974 return log_error_errno(errno
, "Failed to determine if mount point %s contains the unified cgroup hierarchy: %m", p
);
976 log_error("%s is already mounted but not a unified cgroup hierarchy. Refusing.", p
);
980 return mount_verbose(LOG_ERR
, "cgroup", p
, "cgroup2", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
985 CGroupUnified unified_requested
,
986 bool userns
, uid_t uid_shift
, uid_t uid_range
,
987 const char *selinux_apifs_context
,
990 if (unified_requested
>= CGROUP_UNIFIED_ALL
)
991 return mount_unified_cgroups(dest
);
993 return mount_legacy_cgns_supported(unified_requested
, userns
, uid_shift
, uid_range
, selinux_apifs_context
);
995 return mount_legacy_cgns_unsupported(dest
, unified_requested
, userns
, uid_shift
, uid_range
, selinux_apifs_context
);
998 int mount_systemd_cgroup_writable(
1000 CGroupUnified unified_requested
) {
1002 _cleanup_free_
char *own_cgroup_path
= NULL
;
1003 const char *systemd_root
, *systemd_own
;
1008 r
= cg_pid_get_path(NULL
, 0, &own_cgroup_path
);
1010 return log_error_errno(r
, "Failed to determine our own cgroup path: %m");
1012 /* If we are living in the top-level, then there's nothing to do... */
1013 if (path_equal(own_cgroup_path
, "/"))
1016 if (unified_requested
>= CGROUP_UNIFIED_ALL
) {
1017 systemd_own
= strjoina(dest
, "/sys/fs/cgroup", own_cgroup_path
);
1018 systemd_root
= prefix_roota(dest
, "/sys/fs/cgroup");
1020 systemd_own
= strjoina(dest
, "/sys/fs/cgroup/systemd", own_cgroup_path
);
1021 systemd_root
= prefix_roota(dest
, "/sys/fs/cgroup/systemd");
1024 /* Make our own cgroup a (writable) bind mount */
1025 r
= mount_verbose(LOG_ERR
, systemd_own
, systemd_own
, NULL
, MS_BIND
, NULL
);
1029 /* And then remount the systemd cgroup root read-only */
1030 return mount_verbose(LOG_ERR
, NULL
, systemd_root
, NULL
,
1031 MS_BIND
|MS_REMOUNT
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_RDONLY
, NULL
);
1034 int setup_volatile_state(
1035 const char *directory
,
1037 bool userns
, uid_t uid_shift
, uid_t uid_range
,
1038 const char *selinux_apifs_context
) {
1040 _cleanup_free_
char *buf
= NULL
;
1041 const char *p
, *options
;
1046 if (mode
!= VOLATILE_STATE
)
1049 /* --volatile=state means we simply overmount /var
1050 with a tmpfs, and the rest read-only. */
1052 r
= bind_remount_recursive(directory
, true, NULL
);
1054 return log_error_errno(r
, "Failed to remount %s read-only: %m", directory
);
1056 p
= prefix_roota(directory
, "/var");
1058 if (r
< 0 && errno
!= EEXIST
)
1059 return log_error_errno(errno
, "Failed to create %s: %m", directory
);
1061 options
= "mode=755";
1062 r
= tmpfs_patch_options(options
, userns
, uid_shift
, uid_range
, false, selinux_apifs_context
, &buf
);
1068 return mount_verbose(LOG_ERR
, "tmpfs", p
, "tmpfs", MS_STRICTATIME
, options
);
1072 const char *directory
,
1074 bool userns
, uid_t uid_shift
, uid_t uid_range
,
1075 const char *selinux_apifs_context
) {
1077 bool tmpfs_mounted
= false, bind_mounted
= false;
1078 char template[] = "/tmp/nspawn-volatile-XXXXXX";
1079 _cleanup_free_
char *buf
= NULL
;
1080 const char *f
, *t
, *options
;
1085 if (mode
!= VOLATILE_YES
)
1088 /* --volatile=yes means we mount a tmpfs to the root dir, and
1089 the original /usr to use inside it, and that read-only. */
1091 if (!mkdtemp(template))
1092 return log_error_errno(errno
, "Failed to create temporary directory: %m");
1094 options
= "mode=755";
1095 r
= tmpfs_patch_options(options
, userns
, uid_shift
, uid_range
, false, selinux_apifs_context
, &buf
);
1101 r
= mount_verbose(LOG_ERR
, "tmpfs", template, "tmpfs", MS_STRICTATIME
, options
);
1105 tmpfs_mounted
= true;
1107 f
= prefix_roota(directory
, "/usr");
1108 t
= prefix_roota(template, "/usr");
1111 if (r
< 0 && errno
!= EEXIST
) {
1112 r
= log_error_errno(errno
, "Failed to create %s: %m", t
);
1116 r
= mount_verbose(LOG_ERR
, f
, t
, NULL
, MS_BIND
|MS_REC
, NULL
);
1120 bind_mounted
= true;
1122 r
= bind_remount_recursive(t
, true, NULL
);
1124 log_error_errno(r
, "Failed to remount %s read-only: %m", t
);
1128 r
= mount_verbose(LOG_ERR
, template, directory
, NULL
, MS_MOVE
, NULL
);
1132 (void) rmdir(template);
1138 (void) umount_verbose(t
);
1141 (void) umount_verbose(template);
1142 (void) rmdir(template);
1146 VolatileMode
volatile_mode_from_string(const char *s
) {
1150 return _VOLATILE_MODE_INVALID
;
1152 b
= parse_boolean(s
);
1154 return VOLATILE_YES
;
1158 if (streq(s
, "state"))
1159 return VOLATILE_STATE
;
1161 return _VOLATILE_MODE_INVALID
;