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 int mount_all(const char *dest
,
304 bool use_userns
, bool in_userns
,
306 uid_t uid_shift
, uid_t uid_range
,
307 const char *selinux_apifs_context
) {
309 typedef struct MountPoint
{
320 static const MountPoint mount_table
[] = {
321 { "proc", "/proc", "proc", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, true, true, false },
322 { "/proc/sys", "/proc/sys", NULL
, NULL
, MS_BIND
, true, true, false }, /* Bind mount first ...*/
323 { "/proc/sys/net", "/proc/sys/net", NULL
, NULL
, MS_BIND
, true, true, true }, /* (except for this) */
324 { 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 */
325 { "/proc/sysrq-trigger", "/proc/sysrq-trigger", NULL
, NULL
, MS_BIND
, false, true, false }, /* Bind mount first ...*/
326 { 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 */
327 { "tmpfs", "/sys", "tmpfs", "mode=755", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, true, false, true },
328 { "sysfs", "/sys", "sysfs", NULL
, MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, true, false, false },
329 { "tmpfs", "/dev", "tmpfs", "mode=755", MS_NOSUID
|MS_STRICTATIME
, true, false, false },
330 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
, true, false, false },
331 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
, true, false, false },
332 { "tmpfs", "/tmp", "tmpfs", "mode=1777", MS_STRICTATIME
, true, true, false },
334 { "/sys/fs/selinux", "/sys/fs/selinux", NULL
, NULL
, MS_BIND
, false, false, false }, /* Bind mount first */
335 { 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 */
342 for (k
= 0; k
< ELEMENTSOF(mount_table
); k
++) {
343 _cleanup_free_
char *where
= NULL
, *options
= NULL
;
346 if (in_userns
!= mount_table
[k
].in_userns
)
349 if (!use_netns
&& mount_table
[k
].use_netns
)
352 where
= prefix_root(dest
, mount_table
[k
].where
);
356 r
= path_is_mount_point(where
, AT_SYMLINK_FOLLOW
);
357 if (r
< 0 && r
!= -ENOENT
)
358 return log_error_errno(r
, "Failed to detect whether %s is a mount point: %m", where
);
360 /* Skip this entry if it is not a remount. */
361 if (mount_table
[k
].what
&& r
> 0)
364 r
= mkdir_p(where
, 0755);
365 if (r
< 0 && r
!= -EEXIST
) {
366 if (mount_table
[k
].fatal
)
367 return log_error_errno(r
, "Failed to create directory %s: %m", where
);
369 log_debug_errno(r
, "Failed to create directory %s: %m", where
);
373 o
= mount_table
[k
].options
;
374 if (streq_ptr(mount_table
[k
].type
, "tmpfs")) {
376 r
= tmpfs_patch_options(o
, use_userns
, 0, uid_range
, true, selinux_apifs_context
, &options
);
378 r
= tmpfs_patch_options(o
, use_userns
, uid_shift
, uid_range
, false, selinux_apifs_context
, &options
);
385 r
= mount_verbose(mount_table
[k
].fatal
? LOG_ERR
: LOG_WARNING
,
389 mount_table
[k
].flags
,
391 if (r
< 0 && mount_table
[k
].fatal
)
398 static int parse_mount_bind_options(const char *options
, unsigned long *mount_flags
, char **mount_opts
) {
399 const char *p
= options
;
400 unsigned long flags
= *mount_flags
;
406 _cleanup_free_
char *word
= NULL
;
407 int r
= extract_first_word(&p
, &word
, ",", 0);
409 return log_error_errno(r
, "Failed to extract mount option: %m");
413 if (streq(word
, "rbind"))
415 else if (streq(word
, "norbind"))
418 log_error("Invalid bind mount option: %s", word
);
423 *mount_flags
= flags
;
424 /* in the future mount_opts will hold string options for mount(2) */
430 static int mount_bind(const char *dest
, CustomMount
*m
) {
431 struct stat source_st
, dest_st
;
433 unsigned long mount_flags
= MS_BIND
| MS_REC
;
434 _cleanup_free_
char *mount_opts
= NULL
;
440 r
= parse_mount_bind_options(m
->options
, &mount_flags
, &mount_opts
);
445 if (stat(m
->source
, &source_st
) < 0)
446 return log_error_errno(errno
, "Failed to stat %s: %m", m
->source
);
448 where
= prefix_roota(dest
, m
->destination
);
450 if (stat(where
, &dest_st
) >= 0) {
451 if (S_ISDIR(source_st
.st_mode
) && !S_ISDIR(dest_st
.st_mode
)) {
452 log_error("Cannot bind mount directory %s on file %s.", m
->source
, where
);
456 if (!S_ISDIR(source_st
.st_mode
) && S_ISDIR(dest_st
.st_mode
)) {
457 log_error("Cannot bind mount file %s on directory %s.", m
->source
, where
);
461 } else if (errno
== ENOENT
) {
462 r
= mkdir_parents_label(where
, 0755);
464 return log_error_errno(r
, "Failed to make parents of %s: %m", where
);
466 /* Create the mount point. Any non-directory file can be
467 * mounted on any non-directory file (regular, fifo, socket,
470 if (S_ISDIR(source_st
.st_mode
))
471 r
= mkdir_label(where
, 0755);
475 return log_error_errno(r
, "Failed to create mount point %s: %m", where
);
478 return log_error_errno(errno
, "Failed to stat %s: %m", where
);
480 r
= mount_verbose(LOG_ERR
, m
->source
, where
, NULL
, mount_flags
, mount_opts
);
485 r
= bind_remount_recursive(where
, true, NULL
);
487 return log_error_errno(r
, "Read-only bind mount failed: %m");
493 static int mount_tmpfs(
496 bool userns
, uid_t uid_shift
, uid_t uid_range
,
497 const char *selinux_apifs_context
) {
499 const char *where
, *options
;
500 _cleanup_free_
char *buf
= NULL
;
506 where
= prefix_roota(dest
, m
->destination
);
508 r
= mkdir_p_label(where
, 0755);
509 if (r
< 0 && r
!= -EEXIST
)
510 return log_error_errno(r
, "Creating mount point for tmpfs %s failed: %m", where
);
512 r
= tmpfs_patch_options(m
->options
, userns
, uid_shift
, uid_range
, false, selinux_apifs_context
, &buf
);
515 options
= r
> 0 ? buf
: m
->options
;
517 return mount_verbose(LOG_ERR
, "tmpfs", where
, "tmpfs", MS_NODEV
|MS_STRICTATIME
, options
);
520 static char *joined_and_escaped_lower_dirs(char * const *lower
) {
521 _cleanup_strv_free_
char **sv
= NULL
;
523 sv
= strv_copy(lower
);
529 if (!strv_shell_escape(sv
, ",:"))
532 return strv_join(sv
, ":");
535 static int mount_overlay(const char *dest
, CustomMount
*m
) {
536 _cleanup_free_
char *lower
= NULL
;
537 const char *where
, *options
;
543 where
= prefix_roota(dest
, m
->destination
);
545 r
= mkdir_label(where
, 0755);
546 if (r
< 0 && r
!= -EEXIST
)
547 return log_error_errno(r
, "Creating mount point for overlay %s failed: %m", where
);
549 (void) mkdir_p_label(m
->source
, 0755);
551 lower
= joined_and_escaped_lower_dirs(m
->lower
);
556 _cleanup_free_
char *escaped_source
= NULL
;
558 escaped_source
= shell_escape(m
->source
, ",:");
562 options
= strjoina("lowerdir=", escaped_source
, ":", lower
);
564 _cleanup_free_
char *escaped_source
= NULL
, *escaped_work_dir
= NULL
;
567 (void) mkdir_label(m
->work_dir
, 0700);
569 escaped_source
= shell_escape(m
->source
, ",:");
572 escaped_work_dir
= shell_escape(m
->work_dir
, ",:");
573 if (!escaped_work_dir
)
576 options
= strjoina("lowerdir=", lower
, ",upperdir=", escaped_source
, ",workdir=", escaped_work_dir
);
579 return mount_verbose(LOG_ERR
, "overlay", where
, "overlay", m
->read_only
? MS_RDONLY
: 0, options
);
584 CustomMount
*mounts
, unsigned n
,
585 bool userns
, uid_t uid_shift
, uid_t uid_range
,
586 const char *selinux_apifs_context
) {
593 for (i
= 0; i
< n
; i
++) {
594 CustomMount
*m
= mounts
+ i
;
598 case CUSTOM_MOUNT_BIND
:
599 r
= mount_bind(dest
, m
);
602 case CUSTOM_MOUNT_TMPFS
:
603 r
= mount_tmpfs(dest
, m
, userns
, uid_shift
, uid_range
, selinux_apifs_context
);
606 case CUSTOM_MOUNT_OVERLAY
:
607 r
= mount_overlay(dest
, m
);
611 assert_not_reached("Unknown custom mount type");
621 /* Retrieve existing subsystems. This function is called in a new cgroup
624 static int get_controllers(Set
*subsystems
) {
625 _cleanup_fclose_
FILE *f
= NULL
;
630 f
= fopen("/proc/self/cgroup", "re");
632 return errno
== ENOENT
? -ESRCH
: -errno
;
634 FOREACH_LINE(line
, f
, return -errno
) {
638 l
= strchr(line
, ':');
649 if (STR_IN_SET(l
, "", "name=systemd"))
656 r
= set_consume(subsystems
, p
);
664 static int mount_legacy_cgroup_hierarchy(const char *dest
, const char *controller
, const char *hierarchy
,
665 CGroupUnified unified_requested
, bool read_only
) {
666 const char *to
, *fstype
, *opts
;
669 to
= strjoina(strempty(dest
), "/sys/fs/cgroup/", hierarchy
);
671 r
= path_is_mount_point(to
, 0);
672 if (r
< 0 && r
!= -ENOENT
)
673 return log_error_errno(r
, "Failed to determine if %s is mounted already: %m", to
);
679 /* The superblock mount options of the mount point need to be
680 * identical to the hosts', and hence writable... */
681 if (streq(controller
, SYSTEMD_CGROUP_CONTROLLER
)) {
682 if (unified_requested
>= CGROUP_UNIFIED_SYSTEMD
) {
687 opts
= "none,name=systemd,xattr";
694 r
= mount_verbose(LOG_ERR
, "cgroup", to
, fstype
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, opts
);
698 /* ... hence let's only make the bind mount read-only, not the superblock. */
700 r
= mount_verbose(LOG_ERR
, NULL
, to
, NULL
,
701 MS_BIND
|MS_REMOUNT
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_RDONLY
, NULL
);
709 /* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */
710 static int mount_legacy_cgns_supported(
711 CGroupUnified unified_requested
, bool userns
, uid_t uid_shift
,
712 uid_t uid_range
, const char *selinux_apifs_context
) {
713 _cleanup_set_free_free_ Set
*controllers
= NULL
;
714 const char *cgroup_root
= "/sys/fs/cgroup", *c
;
717 (void) mkdir_p(cgroup_root
, 0755);
719 /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
720 r
= path_is_mount_point(cgroup_root
, AT_SYMLINK_FOLLOW
);
722 return log_error_errno(r
, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
724 _cleanup_free_
char *options
= NULL
;
726 /* When cgroup namespaces are enabled and user namespaces are
727 * used then the mount of the cgroupfs is done *inside* the new
728 * user namespace. We're root in the new user namespace and the
729 * kernel will happily translate our uid/gid to the correct
730 * uid/gid as seen from e.g. /proc/1/mountinfo. So we simply
731 * pass uid 0 and not uid_shift to tmpfs_patch_options().
733 r
= tmpfs_patch_options("mode=755", userns
, 0, uid_range
, true, selinux_apifs_context
, &options
);
737 r
= mount_verbose(LOG_ERR
, "tmpfs", cgroup_root
, "tmpfs",
738 MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
, options
);
743 if (cg_all_unified() > 0)
744 goto skip_controllers
;
746 controllers
= set_new(&string_hash_ops
);
750 r
= get_controllers(controllers
);
752 return log_error_errno(r
, "Failed to determine cgroup controllers: %m");
755 _cleanup_free_
const char *controller
= NULL
;
757 controller
= set_steal_first(controllers
);
761 r
= mount_legacy_cgroup_hierarchy("", controller
, controller
, unified_requested
, !userns
);
765 /* When multiple hierarchies are co-mounted, make their
766 * constituting individual hierarchies a symlink to the
771 _cleanup_free_
char *target
= NULL
, *tok
= NULL
;
773 r
= extract_first_word(&c
, &tok
, ",", 0);
775 return log_error_errno(r
, "Failed to extract co-mounted cgroup controller: %m");
779 target
= prefix_root("/sys/fs/cgroup", tok
);
783 if (streq(controller
, tok
))
786 r
= symlink_idempotent(controller
, target
);
788 return log_error_errno(r
, "Invalid existing symlink for combined hierarchy: %m");
790 return log_error_errno(r
, "Failed to create symlink for combined hierarchy: %m");
795 r
= mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER
, "systemd", unified_requested
, false);
800 return mount_verbose(LOG_ERR
, NULL
, cgroup_root
, NULL
,
801 MS_REMOUNT
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
|MS_RDONLY
, "mode=755");
806 /* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */
807 static int mount_legacy_cgns_unsupported(
809 CGroupUnified unified_requested
, bool userns
, uid_t uid_shift
, uid_t uid_range
,
810 const char *selinux_apifs_context
) {
811 _cleanup_set_free_free_ Set
*controllers
= NULL
;
812 const char *cgroup_root
;
815 cgroup_root
= prefix_roota(dest
, "/sys/fs/cgroup");
817 (void) mkdir_p(cgroup_root
, 0755);
819 /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
820 r
= path_is_mount_point(cgroup_root
, AT_SYMLINK_FOLLOW
);
822 return log_error_errno(r
, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
824 _cleanup_free_
char *options
= NULL
;
826 r
= tmpfs_patch_options("mode=755", userns
, uid_shift
, uid_range
, false, selinux_apifs_context
, &options
);
830 r
= mount_verbose(LOG_ERR
, "tmpfs", cgroup_root
, "tmpfs",
831 MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
, options
);
836 if (cg_all_unified() > 0)
837 goto skip_controllers
;
839 controllers
= set_new(&string_hash_ops
);
843 r
= cg_kernel_controllers(controllers
);
845 return log_error_errno(r
, "Failed to determine cgroup controllers: %m");
848 _cleanup_free_
char *controller
= NULL
, *origin
= NULL
, *combined
= NULL
;
850 controller
= set_steal_first(controllers
);
854 origin
= prefix_root("/sys/fs/cgroup/", controller
);
858 r
= readlink_malloc(origin
, &combined
);
860 /* Not a symbolic link, but directly a single cgroup hierarchy */
862 r
= mount_legacy_cgroup_hierarchy(dest
, controller
, controller
, unified_requested
, true);
867 return log_error_errno(r
, "Failed to read link %s: %m", origin
);
869 _cleanup_free_
char *target
= NULL
;
871 target
= prefix_root(dest
, origin
);
875 /* A symbolic link, a combination of controllers in one hierarchy */
877 if (!filename_is_valid(combined
)) {
878 log_warning("Ignoring invalid combined hierarchy %s.", combined
);
882 r
= mount_legacy_cgroup_hierarchy(dest
, combined
, combined
, unified_requested
, true);
886 r
= symlink_idempotent(combined
, target
);
888 return log_error_errno(r
, "Invalid existing symlink for combined hierarchy: %m");
890 return log_error_errno(r
, "Failed to create symlink for combined hierarchy: %m");
895 r
= mount_legacy_cgroup_hierarchy(dest
, SYSTEMD_CGROUP_CONTROLLER
, "systemd", unified_requested
, false);
899 return mount_verbose(LOG_ERR
, NULL
, cgroup_root
, NULL
,
900 MS_REMOUNT
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
|MS_RDONLY
, "mode=755");
903 static int mount_unified_cgroups(const char *dest
) {
909 p
= prefix_roota(dest
, "/sys/fs/cgroup");
911 (void) mkdir_p(p
, 0755);
913 r
= path_is_mount_point(p
, AT_SYMLINK_FOLLOW
);
915 return log_error_errno(r
, "Failed to determine if %s is mounted already: %m", p
);
917 p
= prefix_roota(dest
, "/sys/fs/cgroup/cgroup.procs");
918 if (access(p
, F_OK
) >= 0)
921 return log_error_errno(errno
, "Failed to determine if mount point %s contains the unified cgroup hierarchy: %m", p
);
923 log_error("%s is already mounted but not a unified cgroup hierarchy. Refusing.", p
);
927 return mount_verbose(LOG_ERR
, "cgroup", p
, "cgroup2", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
932 CGroupUnified unified_requested
,
933 bool userns
, uid_t uid_shift
, uid_t uid_range
,
934 const char *selinux_apifs_context
,
937 if (unified_requested
>= CGROUP_UNIFIED_ALL
)
938 return mount_unified_cgroups(dest
);
940 return mount_legacy_cgns_supported(unified_requested
, userns
, uid_shift
, uid_range
, selinux_apifs_context
);
942 return mount_legacy_cgns_unsupported(dest
, unified_requested
, userns
, uid_shift
, uid_range
, selinux_apifs_context
);
945 int mount_systemd_cgroup_writable(
947 CGroupUnified unified_requested
) {
949 _cleanup_free_
char *own_cgroup_path
= NULL
;
950 const char *systemd_root
, *systemd_own
;
955 r
= cg_pid_get_path(NULL
, 0, &own_cgroup_path
);
957 return log_error_errno(r
, "Failed to determine our own cgroup path: %m");
959 /* If we are living in the top-level, then there's nothing to do... */
960 if (path_equal(own_cgroup_path
, "/"))
963 if (unified_requested
>= CGROUP_UNIFIED_ALL
) {
964 systemd_own
= strjoina(dest
, "/sys/fs/cgroup", own_cgroup_path
);
965 systemd_root
= prefix_roota(dest
, "/sys/fs/cgroup");
967 systemd_own
= strjoina(dest
, "/sys/fs/cgroup/systemd", own_cgroup_path
);
968 systemd_root
= prefix_roota(dest
, "/sys/fs/cgroup/systemd");
971 /* Make our own cgroup a (writable) bind mount */
972 r
= mount_verbose(LOG_ERR
, systemd_own
, systemd_own
, NULL
, MS_BIND
, NULL
);
976 /* And then remount the systemd cgroup root read-only */
977 return mount_verbose(LOG_ERR
, NULL
, systemd_root
, NULL
,
978 MS_BIND
|MS_REMOUNT
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_RDONLY
, NULL
);
981 int setup_volatile_state(
982 const char *directory
,
984 bool userns
, uid_t uid_shift
, uid_t uid_range
,
985 const char *selinux_apifs_context
) {
987 _cleanup_free_
char *buf
= NULL
;
988 const char *p
, *options
;
993 if (mode
!= VOLATILE_STATE
)
996 /* --volatile=state means we simply overmount /var
997 with a tmpfs, and the rest read-only. */
999 r
= bind_remount_recursive(directory
, true, NULL
);
1001 return log_error_errno(r
, "Failed to remount %s read-only: %m", directory
);
1003 p
= prefix_roota(directory
, "/var");
1005 if (r
< 0 && errno
!= EEXIST
)
1006 return log_error_errno(errno
, "Failed to create %s: %m", directory
);
1008 options
= "mode=755";
1009 r
= tmpfs_patch_options(options
, userns
, uid_shift
, uid_range
, false, selinux_apifs_context
, &buf
);
1015 return mount_verbose(LOG_ERR
, "tmpfs", p
, "tmpfs", MS_STRICTATIME
, options
);
1019 const char *directory
,
1021 bool userns
, uid_t uid_shift
, uid_t uid_range
,
1022 const char *selinux_apifs_context
) {
1024 bool tmpfs_mounted
= false, bind_mounted
= false;
1025 char template[] = "/tmp/nspawn-volatile-XXXXXX";
1026 _cleanup_free_
char *buf
= NULL
;
1027 const char *f
, *t
, *options
;
1032 if (mode
!= VOLATILE_YES
)
1035 /* --volatile=yes means we mount a tmpfs to the root dir, and
1036 the original /usr to use inside it, and that read-only. */
1038 if (!mkdtemp(template))
1039 return log_error_errno(errno
, "Failed to create temporary directory: %m");
1041 options
= "mode=755";
1042 r
= tmpfs_patch_options(options
, userns
, uid_shift
, uid_range
, false, selinux_apifs_context
, &buf
);
1048 r
= mount_verbose(LOG_ERR
, "tmpfs", template, "tmpfs", MS_STRICTATIME
, options
);
1052 tmpfs_mounted
= true;
1054 f
= prefix_roota(directory
, "/usr");
1055 t
= prefix_roota(template, "/usr");
1058 if (r
< 0 && errno
!= EEXIST
) {
1059 r
= log_error_errno(errno
, "Failed to create %s: %m", t
);
1063 r
= mount_verbose(LOG_ERR
, f
, t
, NULL
, MS_BIND
|MS_REC
, NULL
);
1067 bind_mounted
= true;
1069 r
= bind_remount_recursive(t
, true, NULL
);
1071 log_error_errno(r
, "Failed to remount %s read-only: %m", t
);
1075 r
= mount_verbose(LOG_ERR
, template, directory
, NULL
, MS_MOVE
, NULL
);
1079 (void) rmdir(template);
1085 (void) umount_verbose(t
);
1088 (void) umount_verbose(template);
1089 (void) rmdir(template);
1093 VolatileMode
volatile_mode_from_string(const char *s
) {
1097 return _VOLATILE_MODE_INVALID
;
1099 b
= parse_boolean(s
);
1101 return VOLATILE_YES
;
1105 if (streq(s
, "state"))
1106 return VOLATILE_STATE
;
1108 return _VOLATILE_MODE_INVALID
;