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
);
196 if (asprintf(&buf
, "%s%suid=" UID_FMT
",gid=" UID_FMT
,
197 options
?: "", options
? "," : "",
198 uid_shift
, uid_shift
) < 0)
205 if (selinux_apifs_context
) {
208 t
= strjoin(options
?: "", options
? "," : "",
209 "context=\"", selinux_apifs_context
, "\"");
218 if (!buf
&& options
) {
219 buf
= strdup(options
);
228 int mount_sysfs(const char *dest
, MountSettingsMask mount_settings
) {
229 const char *full
, *top
, *x
;
231 unsigned long extra_flags
= 0;
233 top
= prefix_roota(dest
, "/sys");
234 r
= path_check_fstype(top
, SYSFS_MAGIC
);
236 return log_error_errno(r
, "Failed to determine filesystem type of %s: %m", top
);
237 /* /sys might already be mounted as sysfs by the outer child in the
238 * !netns case. In this case, it's all good. Don't touch it because we
239 * don't have the right to do so, see https://github.com/systemd/systemd/issues/1555.
244 full
= prefix_roota(top
, "/full");
246 (void) mkdir(full
, 0755);
248 if (mount_settings
& MOUNT_APPLY_APIVFS_RO
)
249 extra_flags
|= MS_RDONLY
;
251 r
= mount_verbose(LOG_ERR
, "sysfs", full
, "sysfs",
252 MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|extra_flags
, NULL
);
256 FOREACH_STRING(x
, "block", "bus", "class", "dev", "devices", "kernel") {
257 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
259 from
= prefix_root(full
, x
);
263 to
= prefix_root(top
, x
);
267 (void) mkdir(to
, 0755);
269 r
= mount_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
273 r
= mount_verbose(LOG_ERR
, NULL
, to
, NULL
,
274 MS_BIND
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_REMOUNT
|extra_flags
, NULL
);
279 r
= umount_verbose(full
);
284 return log_error_errno(errno
, "Failed to remove %s: %m", full
);
286 x
= prefix_roota(top
, "/fs/kdbus");
287 (void) mkdir_p(x
, 0755);
289 /* Create mountpoint for cgroups. Otherwise we are not allowed since we
290 * remount /sys read-only.
292 if (cg_ns_supported()) {
293 x
= prefix_roota(top
, "/fs/cgroup");
294 (void) mkdir_p(x
, 0755);
297 return mount_verbose(LOG_ERR
, NULL
, top
, NULL
,
298 MS_BIND
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_REMOUNT
|extra_flags
, NULL
);
301 static int mkdir_userns(const char *path
, mode_t mode
, bool in_userns
, uid_t uid_shift
) {
306 r
= mkdir(path
, mode
);
307 if (r
< 0 && errno
!= EEXIST
)
311 r
= lchown(path
, uid_shift
, uid_shift
);
319 static int mkdir_userns_p(const char *prefix
, const char *path
, mode_t mode
, bool in_userns
, uid_t uid_shift
) {
325 if (prefix
&& !path_startswith(path
, prefix
))
328 /* create every parent directory in the path, except the last component */
329 p
= path
+ strspn(path
, "/");
331 char t
[strlen(path
) + 1];
333 e
= p
+ strcspn(p
, "/");
334 p
= e
+ strspn(e
, "/");
336 /* Is this the last component? If so, then we're done */
340 memcpy(t
, path
, e
- path
);
343 if (prefix
&& path_startswith(prefix
, t
))
346 r
= mkdir_userns(t
, mode
, in_userns
, uid_shift
);
351 return mkdir_userns(path
, mode
, in_userns
, uid_shift
);
354 int mount_all(const char *dest
,
355 MountSettingsMask mount_settings
,
356 uid_t uid_shift
, uid_t uid_range
,
357 const char *selinux_apifs_context
) {
359 typedef struct MountPoint
{
365 MountSettingsMask mount_settings
;
368 static const MountPoint mount_table
[] = {
369 /* inner child mounts */
370 { "proc", "/proc", "proc", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, MOUNT_FATAL
|MOUNT_IN_USERNS
},
371 { "/proc/sys", "/proc/sys", NULL
, NULL
, MS_BIND
, MOUNT_FATAL
|MOUNT_IN_USERNS
|MOUNT_APPLY_APIVFS_RO
}, /* Bind mount first ...*/
372 { "/proc/sys/net", "/proc/sys/net", NULL
, NULL
, MS_BIND
, MOUNT_FATAL
|MOUNT_IN_USERNS
|MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
}, /* (except for this) */
373 { NULL
, "/proc/sys", NULL
, NULL
, MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_REMOUNT
, MOUNT_FATAL
|MOUNT_IN_USERNS
|MOUNT_APPLY_APIVFS_RO
}, /* ... then, make it r/o */
374 { "/proc/sysrq-trigger", "/proc/sysrq-trigger", NULL
, NULL
, MS_BIND
, MOUNT_IN_USERNS
|MOUNT_APPLY_APIVFS_RO
}, /* Bind mount first ...*/
375 { NULL
, "/proc/sysrq-trigger", NULL
, NULL
, MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_REMOUNT
, MOUNT_IN_USERNS
|MOUNT_APPLY_APIVFS_RO
}, /* ... then, make it r/o */
376 { "tmpfs", "/tmp", "tmpfs", "mode=1777", MS_STRICTATIME
, MOUNT_FATAL
|MOUNT_IN_USERNS
},
378 /* outer child mounts */
379 { "tmpfs", "/sys", "tmpfs", "mode=755", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, MOUNT_FATAL
|MOUNT_APPLY_APIVFS_NETNS
},
380 { "sysfs", "/sys", "sysfs", NULL
, MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, MOUNT_FATAL
|MOUNT_APPLY_APIVFS_RO
}, /* skipped if above was mounted */
381 { "sysfs", "/sys", "sysfs", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, MOUNT_FATAL
}, /* skipped if above was mounted */
383 { "tmpfs", "/dev", "tmpfs", "mode=755", MS_NOSUID
|MS_STRICTATIME
, MOUNT_FATAL
},
384 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
, MOUNT_FATAL
},
385 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
, MOUNT_FATAL
},
387 { "/sys/fs/selinux", "/sys/fs/selinux", NULL
, NULL
, MS_BIND
, 0 }, /* Bind mount first */
388 { NULL
, "/sys/fs/selinux", NULL
, NULL
, MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_REMOUNT
, 0 }, /* Then, make it r/o */
394 bool use_userns
= (mount_settings
& MOUNT_USE_USERNS
);
395 bool netns
= (mount_settings
& MOUNT_APPLY_APIVFS_NETNS
);
396 bool ro
= (mount_settings
& MOUNT_APPLY_APIVFS_RO
);
397 bool in_userns
= (mount_settings
& MOUNT_IN_USERNS
);
399 for (k
= 0; k
< ELEMENTSOF(mount_table
); k
++) {
400 _cleanup_free_
char *where
= NULL
, *options
= NULL
;
402 bool fatal
= (mount_table
[k
].mount_settings
& MOUNT_FATAL
);
404 if (in_userns
!= (bool)(mount_table
[k
].mount_settings
& MOUNT_IN_USERNS
))
407 if (!netns
&& (bool)(mount_table
[k
].mount_settings
& MOUNT_APPLY_APIVFS_NETNS
))
410 if (!ro
&& (bool)(mount_table
[k
].mount_settings
& MOUNT_APPLY_APIVFS_RO
))
413 where
= prefix_root(dest
, mount_table
[k
].where
);
417 r
= path_is_mount_point(where
, AT_SYMLINK_FOLLOW
);
418 if (r
< 0 && r
!= -ENOENT
)
419 return log_error_errno(r
, "Failed to detect whether %s is a mount point: %m", where
);
421 /* Skip this entry if it is not a remount. */
422 if (mount_table
[k
].what
&& r
> 0)
425 r
= mkdir_userns_p(dest
, where
, 0755, in_userns
, uid_shift
);
426 if (r
< 0 && r
!= -EEXIST
) {
428 return log_error_errno(r
, "Failed to create directory %s: %m", where
);
430 log_debug_errno(r
, "Failed to create directory %s: %m", where
);
434 o
= mount_table
[k
].options
;
435 if (streq_ptr(mount_table
[k
].type
, "tmpfs")) {
437 r
= tmpfs_patch_options(o
, use_userns
, 0, uid_range
, true, selinux_apifs_context
, &options
);
439 r
= tmpfs_patch_options(o
, use_userns
, uid_shift
, uid_range
, false, selinux_apifs_context
, &options
);
446 r
= mount_verbose(fatal
? LOG_ERR
: LOG_DEBUG
,
450 mount_table
[k
].flags
,
459 static int parse_mount_bind_options(const char *options
, unsigned long *mount_flags
, char **mount_opts
) {
460 const char *p
= options
;
461 unsigned long flags
= *mount_flags
;
467 _cleanup_free_
char *word
= NULL
;
468 int r
= extract_first_word(&p
, &word
, ",", 0);
470 return log_error_errno(r
, "Failed to extract mount option: %m");
474 if (streq(word
, "rbind"))
476 else if (streq(word
, "norbind"))
479 log_error("Invalid bind mount option: %s", word
);
484 *mount_flags
= flags
;
485 /* in the future mount_opts will hold string options for mount(2) */
491 static int mount_bind(const char *dest
, CustomMount
*m
) {
492 struct stat source_st
, dest_st
;
494 unsigned long mount_flags
= MS_BIND
| MS_REC
;
495 _cleanup_free_
char *mount_opts
= NULL
;
501 r
= parse_mount_bind_options(m
->options
, &mount_flags
, &mount_opts
);
506 if (stat(m
->source
, &source_st
) < 0)
507 return log_error_errno(errno
, "Failed to stat %s: %m", m
->source
);
509 where
= prefix_roota(dest
, m
->destination
);
511 if (stat(where
, &dest_st
) >= 0) {
512 if (S_ISDIR(source_st
.st_mode
) && !S_ISDIR(dest_st
.st_mode
)) {
513 log_error("Cannot bind mount directory %s on file %s.", m
->source
, where
);
517 if (!S_ISDIR(source_st
.st_mode
) && S_ISDIR(dest_st
.st_mode
)) {
518 log_error("Cannot bind mount file %s on directory %s.", m
->source
, where
);
522 } else if (errno
== ENOENT
) {
523 r
= mkdir_parents_label(where
, 0755);
525 return log_error_errno(r
, "Failed to make parents of %s: %m", where
);
527 /* Create the mount point. Any non-directory file can be
528 * mounted on any non-directory file (regular, fifo, socket,
531 if (S_ISDIR(source_st
.st_mode
))
532 r
= mkdir_label(where
, 0755);
536 return log_error_errno(r
, "Failed to create mount point %s: %m", where
);
539 return log_error_errno(errno
, "Failed to stat %s: %m", where
);
541 r
= mount_verbose(LOG_ERR
, m
->source
, where
, NULL
, mount_flags
, mount_opts
);
546 r
= bind_remount_recursive(where
, true, NULL
);
548 return log_error_errno(r
, "Read-only bind mount failed: %m");
554 static int mount_tmpfs(
557 bool userns
, uid_t uid_shift
, uid_t uid_range
,
558 const char *selinux_apifs_context
) {
560 const char *where
, *options
;
561 _cleanup_free_
char *buf
= NULL
;
567 where
= prefix_roota(dest
, m
->destination
);
569 r
= mkdir_p_label(where
, 0755);
570 if (r
< 0 && r
!= -EEXIST
)
571 return log_error_errno(r
, "Creating mount point for tmpfs %s failed: %m", where
);
573 r
= tmpfs_patch_options(m
->options
, userns
, uid_shift
, uid_range
, false, selinux_apifs_context
, &buf
);
576 options
= r
> 0 ? buf
: m
->options
;
578 return mount_verbose(LOG_ERR
, "tmpfs", where
, "tmpfs", MS_NODEV
|MS_STRICTATIME
, options
);
581 static char *joined_and_escaped_lower_dirs(char * const *lower
) {
582 _cleanup_strv_free_
char **sv
= NULL
;
584 sv
= strv_copy(lower
);
590 if (!strv_shell_escape(sv
, ",:"))
593 return strv_join(sv
, ":");
596 static int mount_overlay(const char *dest
, CustomMount
*m
) {
597 _cleanup_free_
char *lower
= NULL
;
598 const char *where
, *options
;
604 where
= prefix_roota(dest
, m
->destination
);
606 r
= mkdir_label(where
, 0755);
607 if (r
< 0 && r
!= -EEXIST
)
608 return log_error_errno(r
, "Creating mount point for overlay %s failed: %m", where
);
610 (void) mkdir_p_label(m
->source
, 0755);
612 lower
= joined_and_escaped_lower_dirs(m
->lower
);
617 _cleanup_free_
char *escaped_source
= NULL
;
619 escaped_source
= shell_escape(m
->source
, ",:");
623 options
= strjoina("lowerdir=", escaped_source
, ":", lower
);
625 _cleanup_free_
char *escaped_source
= NULL
, *escaped_work_dir
= NULL
;
628 (void) mkdir_label(m
->work_dir
, 0700);
630 escaped_source
= shell_escape(m
->source
, ",:");
633 escaped_work_dir
= shell_escape(m
->work_dir
, ",:");
634 if (!escaped_work_dir
)
637 options
= strjoina("lowerdir=", lower
, ",upperdir=", escaped_source
, ",workdir=", escaped_work_dir
);
640 return mount_verbose(LOG_ERR
, "overlay", where
, "overlay", m
->read_only
? MS_RDONLY
: 0, options
);
645 CustomMount
*mounts
, unsigned n
,
646 bool userns
, uid_t uid_shift
, uid_t uid_range
,
647 const char *selinux_apifs_context
) {
654 for (i
= 0; i
< n
; i
++) {
655 CustomMount
*m
= mounts
+ i
;
659 case CUSTOM_MOUNT_BIND
:
660 r
= mount_bind(dest
, m
);
663 case CUSTOM_MOUNT_TMPFS
:
664 r
= mount_tmpfs(dest
, m
, userns
, uid_shift
, uid_range
, selinux_apifs_context
);
667 case CUSTOM_MOUNT_OVERLAY
:
668 r
= mount_overlay(dest
, m
);
672 assert_not_reached("Unknown custom mount type");
682 /* Retrieve existing subsystems. This function is called in a new cgroup
685 static int get_controllers(Set
*subsystems
) {
686 _cleanup_fclose_
FILE *f
= NULL
;
691 f
= fopen("/proc/self/cgroup", "re");
693 return errno
== ENOENT
? -ESRCH
: -errno
;
695 FOREACH_LINE(line
, f
, return -errno
) {
699 l
= strchr(line
, ':');
710 if (STR_IN_SET(l
, "", "name=systemd"))
717 r
= set_consume(subsystems
, p
);
725 static int mount_legacy_cgroup_hierarchy(const char *dest
, const char *controller
, const char *hierarchy
,
726 CGroupUnified unified_requested
, bool read_only
) {
727 const char *to
, *fstype
, *opts
;
730 to
= strjoina(strempty(dest
), "/sys/fs/cgroup/", hierarchy
);
732 r
= path_is_mount_point(to
, 0);
733 if (r
< 0 && r
!= -ENOENT
)
734 return log_error_errno(r
, "Failed to determine if %s is mounted already: %m", to
);
740 /* The superblock mount options of the mount point need to be
741 * identical to the hosts', and hence writable... */
742 if (streq(controller
, SYSTEMD_CGROUP_CONTROLLER
)) {
743 if (unified_requested
>= CGROUP_UNIFIED_SYSTEMD
) {
748 opts
= "none,name=systemd,xattr";
755 r
= mount_verbose(LOG_ERR
, "cgroup", to
, fstype
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, opts
);
759 /* ... hence let's only make the bind mount read-only, not the superblock. */
761 r
= mount_verbose(LOG_ERR
, NULL
, to
, NULL
,
762 MS_BIND
|MS_REMOUNT
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_RDONLY
, NULL
);
770 /* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */
771 static int mount_legacy_cgns_supported(
772 CGroupUnified unified_requested
, bool userns
, uid_t uid_shift
,
773 uid_t uid_range
, const char *selinux_apifs_context
) {
774 _cleanup_set_free_free_ Set
*controllers
= NULL
;
775 const char *cgroup_root
= "/sys/fs/cgroup", *c
;
778 (void) mkdir_p(cgroup_root
, 0755);
780 /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
781 r
= path_is_mount_point(cgroup_root
, AT_SYMLINK_FOLLOW
);
783 return log_error_errno(r
, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
785 _cleanup_free_
char *options
= NULL
;
787 /* When cgroup namespaces are enabled and user namespaces are
788 * used then the mount of the cgroupfs is done *inside* the new
789 * user namespace. We're root in the new user namespace and the
790 * kernel will happily translate our uid/gid to the correct
791 * uid/gid as seen from e.g. /proc/1/mountinfo. So we simply
792 * pass uid 0 and not uid_shift to tmpfs_patch_options().
794 r
= tmpfs_patch_options("mode=755", userns
, 0, uid_range
, true, selinux_apifs_context
, &options
);
798 r
= mount_verbose(LOG_ERR
, "tmpfs", cgroup_root
, "tmpfs",
799 MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
, options
);
804 if (cg_all_unified() > 0)
805 goto skip_controllers
;
807 controllers
= set_new(&string_hash_ops
);
811 r
= get_controllers(controllers
);
813 return log_error_errno(r
, "Failed to determine cgroup controllers: %m");
816 _cleanup_free_
const char *controller
= NULL
;
818 controller
= set_steal_first(controllers
);
822 r
= mount_legacy_cgroup_hierarchy("", controller
, controller
, unified_requested
, !userns
);
826 /* When multiple hierarchies are co-mounted, make their
827 * constituting individual hierarchies a symlink to the
832 _cleanup_free_
char *target
= NULL
, *tok
= NULL
;
834 r
= extract_first_word(&c
, &tok
, ",", 0);
836 return log_error_errno(r
, "Failed to extract co-mounted cgroup controller: %m");
840 target
= prefix_root("/sys/fs/cgroup", tok
);
844 if (streq(controller
, tok
))
847 r
= symlink_idempotent(controller
, target
);
849 return log_error_errno(r
, "Invalid existing symlink for combined hierarchy: %m");
851 return log_error_errno(r
, "Failed to create symlink for combined hierarchy: %m");
856 r
= mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER
, "systemd", unified_requested
, false);
861 return mount_verbose(LOG_ERR
, NULL
, cgroup_root
, NULL
,
862 MS_REMOUNT
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
|MS_RDONLY
, "mode=755");
867 /* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */
868 static int mount_legacy_cgns_unsupported(
870 CGroupUnified unified_requested
, bool userns
, uid_t uid_shift
, uid_t uid_range
,
871 const char *selinux_apifs_context
) {
872 _cleanup_set_free_free_ Set
*controllers
= NULL
;
873 const char *cgroup_root
;
876 cgroup_root
= prefix_roota(dest
, "/sys/fs/cgroup");
878 (void) mkdir_p(cgroup_root
, 0755);
880 /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
881 r
= path_is_mount_point(cgroup_root
, AT_SYMLINK_FOLLOW
);
883 return log_error_errno(r
, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
885 _cleanup_free_
char *options
= NULL
;
887 r
= tmpfs_patch_options("mode=755", userns
, uid_shift
, uid_range
, false, selinux_apifs_context
, &options
);
891 r
= mount_verbose(LOG_ERR
, "tmpfs", cgroup_root
, "tmpfs",
892 MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
, options
);
897 if (cg_all_unified() > 0)
898 goto skip_controllers
;
900 controllers
= set_new(&string_hash_ops
);
904 r
= cg_kernel_controllers(controllers
);
906 return log_error_errno(r
, "Failed to determine cgroup controllers: %m");
909 _cleanup_free_
char *controller
= NULL
, *origin
= NULL
, *combined
= NULL
;
911 controller
= set_steal_first(controllers
);
915 origin
= prefix_root("/sys/fs/cgroup/", controller
);
919 r
= readlink_malloc(origin
, &combined
);
921 /* Not a symbolic link, but directly a single cgroup hierarchy */
923 r
= mount_legacy_cgroup_hierarchy(dest
, controller
, controller
, unified_requested
, true);
928 return log_error_errno(r
, "Failed to read link %s: %m", origin
);
930 _cleanup_free_
char *target
= NULL
;
932 target
= prefix_root(dest
, origin
);
936 /* A symbolic link, a combination of controllers in one hierarchy */
938 if (!filename_is_valid(combined
)) {
939 log_warning("Ignoring invalid combined hierarchy %s.", combined
);
943 r
= mount_legacy_cgroup_hierarchy(dest
, combined
, combined
, unified_requested
, true);
947 r
= symlink_idempotent(combined
, target
);
949 return log_error_errno(r
, "Invalid existing symlink for combined hierarchy: %m");
951 return log_error_errno(r
, "Failed to create symlink for combined hierarchy: %m");
956 r
= mount_legacy_cgroup_hierarchy(dest
, SYSTEMD_CGROUP_CONTROLLER
, "systemd", unified_requested
, false);
960 return mount_verbose(LOG_ERR
, NULL
, cgroup_root
, NULL
,
961 MS_REMOUNT
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
|MS_RDONLY
, "mode=755");
964 static int mount_unified_cgroups(const char *dest
) {
970 p
= prefix_roota(dest
, "/sys/fs/cgroup");
972 (void) mkdir_p(p
, 0755);
974 r
= path_is_mount_point(p
, AT_SYMLINK_FOLLOW
);
976 return log_error_errno(r
, "Failed to determine if %s is mounted already: %m", p
);
978 p
= prefix_roota(dest
, "/sys/fs/cgroup/cgroup.procs");
979 if (access(p
, F_OK
) >= 0)
982 return log_error_errno(errno
, "Failed to determine if mount point %s contains the unified cgroup hierarchy: %m", p
);
984 log_error("%s is already mounted but not a unified cgroup hierarchy. Refusing.", p
);
988 return mount_verbose(LOG_ERR
, "cgroup", p
, "cgroup2", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
993 CGroupUnified unified_requested
,
994 bool userns
, uid_t uid_shift
, uid_t uid_range
,
995 const char *selinux_apifs_context
,
998 if (unified_requested
>= CGROUP_UNIFIED_ALL
)
999 return mount_unified_cgroups(dest
);
1001 return mount_legacy_cgns_supported(unified_requested
, userns
, uid_shift
, uid_range
, selinux_apifs_context
);
1003 return mount_legacy_cgns_unsupported(dest
, unified_requested
, userns
, uid_shift
, uid_range
, selinux_apifs_context
);
1006 int mount_systemd_cgroup_writable(
1008 CGroupUnified unified_requested
) {
1010 _cleanup_free_
char *own_cgroup_path
= NULL
;
1011 const char *systemd_root
, *systemd_own
;
1016 r
= cg_pid_get_path(NULL
, 0, &own_cgroup_path
);
1018 return log_error_errno(r
, "Failed to determine our own cgroup path: %m");
1020 /* If we are living in the top-level, then there's nothing to do... */
1021 if (path_equal(own_cgroup_path
, "/"))
1024 if (unified_requested
>= CGROUP_UNIFIED_ALL
) {
1025 systemd_own
= strjoina(dest
, "/sys/fs/cgroup", own_cgroup_path
);
1026 systemd_root
= prefix_roota(dest
, "/sys/fs/cgroup");
1028 systemd_own
= strjoina(dest
, "/sys/fs/cgroup/systemd", own_cgroup_path
);
1029 systemd_root
= prefix_roota(dest
, "/sys/fs/cgroup/systemd");
1032 /* Make our own cgroup a (writable) bind mount */
1033 r
= mount_verbose(LOG_ERR
, systemd_own
, systemd_own
, NULL
, MS_BIND
, NULL
);
1037 /* And then remount the systemd cgroup root read-only */
1038 return mount_verbose(LOG_ERR
, NULL
, systemd_root
, NULL
,
1039 MS_BIND
|MS_REMOUNT
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_RDONLY
, NULL
);
1042 int setup_volatile_state(
1043 const char *directory
,
1045 bool userns
, uid_t uid_shift
, uid_t uid_range
,
1046 const char *selinux_apifs_context
) {
1048 _cleanup_free_
char *buf
= NULL
;
1049 const char *p
, *options
;
1054 if (mode
!= VOLATILE_STATE
)
1057 /* --volatile=state means we simply overmount /var
1058 with a tmpfs, and the rest read-only. */
1060 r
= bind_remount_recursive(directory
, true, NULL
);
1062 return log_error_errno(r
, "Failed to remount %s read-only: %m", directory
);
1064 p
= prefix_roota(directory
, "/var");
1066 if (r
< 0 && errno
!= EEXIST
)
1067 return log_error_errno(errno
, "Failed to create %s: %m", directory
);
1069 options
= "mode=755";
1070 r
= tmpfs_patch_options(options
, userns
, uid_shift
, uid_range
, false, selinux_apifs_context
, &buf
);
1076 return mount_verbose(LOG_ERR
, "tmpfs", p
, "tmpfs", MS_STRICTATIME
, options
);
1080 const char *directory
,
1082 bool userns
, uid_t uid_shift
, uid_t uid_range
,
1083 const char *selinux_apifs_context
) {
1085 bool tmpfs_mounted
= false, bind_mounted
= false;
1086 char template[] = "/tmp/nspawn-volatile-XXXXXX";
1087 _cleanup_free_
char *buf
= NULL
;
1088 const char *f
, *t
, *options
;
1093 if (mode
!= VOLATILE_YES
)
1096 /* --volatile=yes means we mount a tmpfs to the root dir, and
1097 the original /usr to use inside it, and that read-only. */
1099 if (!mkdtemp(template))
1100 return log_error_errno(errno
, "Failed to create temporary directory: %m");
1102 options
= "mode=755";
1103 r
= tmpfs_patch_options(options
, userns
, uid_shift
, uid_range
, false, selinux_apifs_context
, &buf
);
1109 r
= mount_verbose(LOG_ERR
, "tmpfs", template, "tmpfs", MS_STRICTATIME
, options
);
1113 tmpfs_mounted
= true;
1115 f
= prefix_roota(directory
, "/usr");
1116 t
= prefix_roota(template, "/usr");
1119 if (r
< 0 && errno
!= EEXIST
) {
1120 r
= log_error_errno(errno
, "Failed to create %s: %m", t
);
1124 r
= mount_verbose(LOG_ERR
, f
, t
, NULL
, MS_BIND
|MS_REC
, NULL
);
1128 bind_mounted
= true;
1130 r
= bind_remount_recursive(t
, true, NULL
);
1132 log_error_errno(r
, "Failed to remount %s read-only: %m", t
);
1136 r
= mount_verbose(LOG_ERR
, template, directory
, NULL
, MS_MOVE
, NULL
);
1140 (void) rmdir(template);
1146 (void) umount_verbose(t
);
1149 (void) umount_verbose(template);
1150 (void) rmdir(template);
1154 VolatileMode
volatile_mode_from_string(const char *s
) {
1158 return _VOLATILE_MODE_INVALID
;
1160 b
= parse_boolean(s
);
1162 return VOLATILE_YES
;
1166 if (streq(s
, "state"))
1167 return VOLATILE_STATE
;
1169 return _VOLATILE_MODE_INVALID
;