1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2010 Lennart Poettering
7 systemd is free software; you can redistribute it and/or modify it
8 under the terms of the GNU Lesser General Public License as published by
9 the Free Software Foundation; either version 2.1 of the License, or
10 (at your option) any later version.
12 systemd is distributed in the hope that it will be useful, but
13 WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 Lesser General Public License for more details.
17 You should have received a copy of the GNU Lesser General Public License
18 along with systemd; If not, see <http://www.gnu.org/licenses/>.
25 #include <sys/mount.h>
30 #include "alloc-util.h"
31 #include "base-filesystem.h"
32 #include "dev-setup.h"
36 #include "loop-util.h"
37 #include "loopback-setup.h"
40 #include "mount-util.h"
41 #include "namespace.h"
42 #include "path-util.h"
43 #include "selinux-util.h"
44 #include "socket-util.h"
45 #include "stat-util.h"
46 #include "string-table.h"
47 #include "string-util.h"
49 #include "umask-util.h"
50 #include "user-util.h"
53 #define DEV_MOUNT_OPTIONS (MS_NOSUID|MS_STRICTATIME|MS_NOEXEC)
55 typedef enum MountMode
{
56 /* This is ordered by priority! */
70 typedef struct MountEntry
{
71 const char *path_const
; /* Memory allocated on stack or static */
73 bool ignore
:1; /* Ignore if path does not exist? */
74 bool has_prefix
:1; /* Already is prefixed by the root dir? */
75 bool read_only
:1; /* Shall this mount point be read-only? */
76 char *path_malloc
; /* Use this instead of 'path_const' if we had to allocate memory */
77 const char *source_const
; /* The source path, for bind mounts */
81 /* If MountAPIVFS= is used, let's mount /sys and /proc into the it, but only as a fallback if the user hasn't mounted
82 * something there already. These mounts are hence overriden by any other explicitly configured mounts. */
83 static const MountEntry apivfs_table
[] = {
84 { "/proc", PROCFS
, false },
85 { "/dev", BIND_DEV
, false },
86 { "/sys", SYSFS
, false },
89 /* ProtectKernelTunables= option and the related filesystem APIs */
90 static const MountEntry protect_kernel_tunables_table
[] = {
91 { "/proc/sys", READONLY
, false },
92 { "/proc/sysrq-trigger", READONLY
, true },
93 { "/proc/latency_stats", READONLY
, true },
94 { "/proc/mtrr", READONLY
, true },
95 { "/proc/apm", READONLY
, true }, /* Obsolete API, there's no point in permitting access to this, ever */
96 { "/proc/acpi", READONLY
, true },
97 { "/proc/timer_stats", READONLY
, true },
98 { "/proc/asound", READONLY
, true },
99 { "/proc/bus", READONLY
, true },
100 { "/proc/fs", READONLY
, true },
101 { "/proc/irq", READONLY
, true },
102 { "/sys", READONLY
, false },
103 { "/sys/kernel/debug", READONLY
, true },
104 { "/sys/kernel/tracing", READONLY
, true },
105 { "/sys/fs/cgroup", READWRITE
, false }, /* READONLY is set by ProtectControlGroups= option */
106 { "/sys/fs/selinux", READWRITE
, true },
109 /* ProtectKernelModules= option */
110 static const MountEntry protect_kernel_modules_table
[] = {
112 { "/lib/modules", INACCESSIBLE
, true },
114 { "/usr/lib/modules", INACCESSIBLE
, true },
118 * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of
119 * system should be protected by ProtectSystem=
121 static const MountEntry protect_home_read_only_table
[] = {
122 { "/home", READONLY
, true },
123 { "/run/user", READONLY
, true },
124 { "/root", READONLY
, true },
127 /* ProtectHome=yes table */
128 static const MountEntry protect_home_yes_table
[] = {
129 { "/home", INACCESSIBLE
, true },
130 { "/run/user", INACCESSIBLE
, true },
131 { "/root", INACCESSIBLE
, true },
134 /* ProtectSystem=yes table */
135 static const MountEntry protect_system_yes_table
[] = {
136 { "/usr", READONLY
, false },
137 { "/boot", READONLY
, true },
138 { "/efi", READONLY
, true },
141 /* ProtectSystem=full includes ProtectSystem=yes */
142 static const MountEntry protect_system_full_table
[] = {
143 { "/usr", READONLY
, false },
144 { "/boot", READONLY
, true },
145 { "/efi", READONLY
, true },
146 { "/etc", READONLY
, false },
150 * ProtectSystem=strict table. In this strict mode, we mount everything
151 * read-only, except for /proc, /dev, /sys which are the kernel API VFS,
152 * which are left writable, but PrivateDevices= + ProtectKernelTunables=
153 * protect those, and these options should be fully orthogonal.
154 * (And of course /home and friends are also left writable, as ProtectHome=
155 * shall manage those, orthogonally).
157 static const MountEntry protect_system_strict_table
[] = {
158 { "/", READONLY
, false },
159 { "/proc", READWRITE
, false }, /* ProtectKernelTunables= */
160 { "/sys", READWRITE
, false }, /* ProtectKernelTunables= */
161 { "/dev", READWRITE
, false }, /* PrivateDevices= */
162 { "/home", READWRITE
, true }, /* ProtectHome= */
163 { "/run/user", READWRITE
, true }, /* ProtectHome= */
164 { "/root", READWRITE
, true }, /* ProtectHome= */
167 static const char *mount_entry_path(const MountEntry
*p
) {
170 /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that,
171 * otherwise the stack/static ->path field is returned. */
173 return p
->path_malloc
?: p
->path_const
;
176 static bool mount_entry_read_only(const MountEntry
*p
) {
179 return p
->read_only
|| IN_SET(p
->mode
, READONLY
, INACCESSIBLE
);
182 static const char *mount_entry_source(const MountEntry
*p
) {
185 return p
->source_malloc
?: p
->source_const
;
188 static void mount_entry_done(MountEntry
*p
) {
191 p
->path_malloc
= mfree(p
->path_malloc
);
192 p
->source_malloc
= mfree(p
->source_malloc
);
195 static int append_access_mounts(MountEntry
**p
, char **strv
, MountMode mode
, bool forcibly_require_prefix
) {
200 /* Adds a list of user-supplied READWRITE/READONLY/INACCESSIBLE entries */
202 STRV_FOREACH(i
, strv
) {
203 bool ignore
= false, needs_prefix
= false;
206 /* Look for any prefixes */
207 if (startswith(e
, "-")) {
211 if (startswith(e
, "+")) {
216 if (!path_is_absolute(e
))
219 *((*p
)++) = (MountEntry
) {
223 .has_prefix
= !needs_prefix
&& !forcibly_require_prefix
,
230 static int append_empty_dir_mounts(MountEntry
**p
, char **strv
) {
235 /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the
236 * "/private/" boundary directories for DynamicUser=1. */
238 STRV_FOREACH(i
, strv
) {
240 *((*p
)++) = (MountEntry
) {
252 static int append_bind_mounts(MountEntry
**p
, const BindMount
*binds
, unsigned n
) {
257 for (i
= 0; i
< n
; i
++) {
258 const BindMount
*b
= binds
+ i
;
260 *((*p
)++) = (MountEntry
) {
261 .path_const
= b
->destination
,
262 .mode
= b
->recursive
? BIND_MOUNT_RECURSIVE
: BIND_MOUNT
,
263 .read_only
= b
->read_only
,
264 .source_const
= b
->source
,
271 static int append_static_mounts(MountEntry
**p
, const MountEntry
*mounts
, unsigned n
, bool ignore_protect
) {
277 /* Adds a list of static pre-defined entries */
279 for (i
= 0; i
< n
; i
++)
280 *((*p
)++) = (MountEntry
) {
281 .path_const
= mount_entry_path(mounts
+i
),
282 .mode
= mounts
[i
].mode
,
283 .ignore
= mounts
[i
].ignore
|| ignore_protect
,
289 static int append_protect_home(MountEntry
**p
, ProtectHome protect_home
, bool ignore_protect
) {
292 switch (protect_home
) {
294 case PROTECT_HOME_NO
:
297 case PROTECT_HOME_READ_ONLY
:
298 return append_static_mounts(p
, protect_home_read_only_table
, ELEMENTSOF(protect_home_read_only_table
), ignore_protect
);
300 case PROTECT_HOME_YES
:
301 return append_static_mounts(p
, protect_home_yes_table
, ELEMENTSOF(protect_home_yes_table
), ignore_protect
);
304 assert_not_reached("Unexpected ProtectHome= value");
308 static int append_protect_system(MountEntry
**p
, ProtectSystem protect_system
, bool ignore_protect
) {
311 switch (protect_system
) {
313 case PROTECT_SYSTEM_NO
:
316 case PROTECT_SYSTEM_STRICT
:
317 return append_static_mounts(p
, protect_system_strict_table
, ELEMENTSOF(protect_system_strict_table
), ignore_protect
);
319 case PROTECT_SYSTEM_YES
:
320 return append_static_mounts(p
, protect_system_yes_table
, ELEMENTSOF(protect_system_yes_table
), ignore_protect
);
322 case PROTECT_SYSTEM_FULL
:
323 return append_static_mounts(p
, protect_system_full_table
, ELEMENTSOF(protect_system_full_table
), ignore_protect
);
326 assert_not_reached("Unexpected ProtectSystem= value");
330 static int mount_path_compare(const void *a
, const void *b
) {
331 const MountEntry
*p
= a
, *q
= b
;
334 /* If the paths are not equal, then order prefixes first */
335 d
= path_compare(mount_entry_path(p
), mount_entry_path(q
));
339 /* If the paths are equal, check the mode */
340 if (p
->mode
< q
->mode
)
343 if (p
->mode
> q
->mode
)
349 static int prefix_where_needed(MountEntry
*m
, unsigned n
, const char *root_directory
) {
352 /* Prefixes all paths in the bind mount table with the root directory if it is specified and the entry needs
358 for (i
= 0; i
< n
; i
++) {
364 s
= prefix_root(root_directory
, mount_entry_path(m
+i
));
368 free_and_replace(m
[i
].path_malloc
, s
);
369 m
[i
].has_prefix
= true;
375 static void drop_duplicates(MountEntry
*m
, unsigned *n
) {
376 MountEntry
*f
, *t
, *previous
;
381 /* Drops duplicate entries. Expects that the array is properly ordered already. */
383 for (f
= m
, t
= m
, previous
= NULL
; f
< m
+ *n
; f
++) {
385 /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare()
387 if (previous
&& path_equal(mount_entry_path(f
), mount_entry_path(previous
))) {
388 log_debug("%s is duplicate.", mount_entry_path(f
));
389 previous
->read_only
= previous
->read_only
|| mount_entry_read_only(f
); /* Propagate the read-only flag to the remaining entry */
402 static void drop_inaccessible(MountEntry
*m
, unsigned *n
) {
404 const char *clear
= NULL
;
409 /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly
410 * ordered already. */
412 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
414 /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop
415 * it, as inaccessible paths really should drop the entire subtree. */
416 if (clear
&& path_startswith(mount_entry_path(f
), clear
)) {
417 log_debug("%s is masked by %s.", mount_entry_path(f
), clear
);
422 clear
= f
->mode
== INACCESSIBLE
? mount_entry_path(f
) : NULL
;
431 static void drop_nop(MountEntry
*m
, unsigned *n
) {
437 /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the
438 * list is ordered by prefixes. */
440 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
442 /* Only suppress such subtrees for READONLY and READWRITE entries */
443 if (IN_SET(f
->mode
, READONLY
, READWRITE
)) {
447 /* Now let's find the first parent of the entry we are looking at. */
448 for (p
= t
-1; p
>= m
; p
--) {
449 if (path_startswith(mount_entry_path(f
), mount_entry_path(p
))) {
455 /* We found it, let's see if it's the same mode, if so, we can drop this entry */
456 if (found
&& p
->mode
== f
->mode
) {
457 log_debug("%s is redundant by %s", mount_entry_path(f
), mount_entry_path(p
));
470 static void drop_outside_root(const char *root_directory
, MountEntry
*m
, unsigned *n
) {
480 /* Drops all mounts that are outside of the root directory. */
482 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
484 if (!path_startswith(mount_entry_path(f
), root_directory
)) {
485 log_debug("%s is outside of root directory.", mount_entry_path(f
));
497 static int clone_device_node(const char *d
, const char *temporary_mount
) {
502 if (stat(d
, &st
) < 0) {
508 if (!S_ISBLK(st
.st_mode
) &&
509 !S_ISCHR(st
.st_mode
))
515 dn
= strjoina(temporary_mount
, d
);
517 mac_selinux_create_file_prepare(d
, st
.st_mode
);
518 r
= mknod(dn
, st
.st_mode
, st
.st_rdev
);
519 mac_selinux_create_file_clear();
521 return log_debug_errno(errno
, "mknod failed for %s: %m", d
);
526 static int mount_private_dev(MountEntry
*m
) {
527 static const char devnodes
[] =
535 char temporary_mount
[] = "/tmp/namespace-dev-XXXXXX";
536 const char *d
, *dev
= NULL
, *devpts
= NULL
, *devshm
= NULL
, *devhugepages
= NULL
, *devmqueue
= NULL
, *devlog
= NULL
, *devptmx
= NULL
;
537 _cleanup_umask_ mode_t u
;
544 if (!mkdtemp(temporary_mount
))
547 dev
= strjoina(temporary_mount
, "/dev");
548 (void) mkdir(dev
, 0755);
549 if (mount("tmpfs", dev
, "tmpfs", DEV_MOUNT_OPTIONS
, "mode=755") < 0) {
554 devpts
= strjoina(temporary_mount
, "/dev/pts");
555 (void) mkdir(devpts
, 0755);
556 if (mount("/dev/pts", devpts
, NULL
, MS_BIND
, NULL
) < 0) {
561 /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx
562 * when /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible
563 * thus, in that case make a clone
565 * in nspawn and other containers it will be a symlink, in that case make it a symlink
567 r
= is_symlink("/dev/ptmx");
571 devptmx
= strjoina(temporary_mount
, "/dev/ptmx");
572 if (symlink("pts/ptmx", devptmx
) < 0) {
577 r
= clone_device_node("/dev/ptmx", temporary_mount
);
586 devshm
= strjoina(temporary_mount
, "/dev/shm");
587 (void) mkdir(devshm
, 0755);
588 r
= mount("/dev/shm", devshm
, NULL
, MS_BIND
, NULL
);
594 devmqueue
= strjoina(temporary_mount
, "/dev/mqueue");
595 (void) mkdir(devmqueue
, 0755);
596 (void) mount("/dev/mqueue", devmqueue
, NULL
, MS_BIND
, NULL
);
598 devhugepages
= strjoina(temporary_mount
, "/dev/hugepages");
599 (void) mkdir(devhugepages
, 0755);
600 (void) mount("/dev/hugepages", devhugepages
, NULL
, MS_BIND
, NULL
);
602 devlog
= strjoina(temporary_mount
, "/dev/log");
603 (void) symlink("/run/systemd/journal/dev-log", devlog
);
605 NULSTR_FOREACH(d
, devnodes
) {
606 r
= clone_device_node(d
, temporary_mount
);
611 dev_setup(temporary_mount
, UID_INVALID
, GID_INVALID
);
613 /* Create the /dev directory if missing. It is more likely to be
614 * missing when the service is started with RootDirectory. This is
615 * consistent with mount units creating the mount points when missing.
617 (void) mkdir_p_label(mount_entry_path(m
), 0755);
619 /* Unmount everything in old /dev */
620 umount_recursive(mount_entry_path(m
), 0);
621 if (mount(dev
, mount_entry_path(m
), NULL
, MS_MOVE
, NULL
) < 0) {
627 rmdir(temporary_mount
);
639 umount(devhugepages
);
646 rmdir(temporary_mount
);
651 static int mount_bind_dev(const MountEntry
*m
) {
656 /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the service's
657 * /dev. This is only used when RootDirectory= is set. */
659 (void) mkdir_p_label(mount_entry_path(m
), 0755);
661 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
663 return log_debug_errno(r
, "Unable to determine whether /dev is already mounted: %m");
664 if (r
> 0) /* make this a NOP if /dev is already a mount point */
667 if (mount("/dev", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
668 return log_debug_errno(errno
, "Failed to bind mount %s: %m", mount_entry_path(m
));
673 static int mount_sysfs(const MountEntry
*m
) {
678 (void) mkdir_p_label(mount_entry_path(m
), 0755);
680 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
682 return log_debug_errno(r
, "Unable to determine whether /sys is already mounted: %m");
683 if (r
> 0) /* make this a NOP if /sys is already a mount point */
686 /* Bind mount the host's version so that we get all child mounts of it, too. */
687 if (mount("/sys", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
688 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
693 static int mount_procfs(const MountEntry
*m
) {
698 (void) mkdir_p_label(mount_entry_path(m
), 0755);
700 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
702 return log_debug_errno(r
, "Unable to determine whether /proc is already mounted: %m");
703 if (r
> 0) /* make this a NOP if /proc is already a mount point */
706 /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in one */
707 if (mount("proc", mount_entry_path(m
), "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
) < 0)
708 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
713 static int mount_empty_dir(const MountEntry
*m
) {
716 /* First, get rid of everything that is below if there is anything. Then, overmount with our new empty dir */
718 (void) mkdir_p_label(mount_entry_path(m
), 0755);
719 (void) umount_recursive(mount_entry_path(m
), 0);
721 if (mount("tmpfs", mount_entry_path(m
), "tmpfs", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
, "mode=755") < 0)
722 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
727 static int mount_entry_chase(
728 const char *root_directory
,
739 /* Since mount() will always follow symlinks and we need to take the different root directory into account we
740 * chase the symlinks on our own first. This is called for the destination path, as well as the source path (if
741 * that applies). The result is stored in "location". */
745 BIND_MOUNT_RECURSIVE
,
752 flags
|= CHASE_NONEXISTENT
;
754 r
= chase_symlinks(path
, root_directory
, flags
, &chased
);
755 if (r
== -ENOENT
&& m
->ignore
) {
756 log_debug_errno(r
, "Path %s does not exist, ignoring.", path
);
760 return log_debug_errno(r
, "Failed to follow symlinks on %s: %m", path
);
762 log_debug("Followed symlinks %s → %s.", path
, chased
);
770 static int apply_mount(
771 const char *root_directory
,
774 bool rbind
= true, make
= false;
780 r
= mount_entry_chase(root_directory
, m
, mount_entry_path(m
), &m
->path_malloc
);
784 log_debug("Applying namespace mount on %s", mount_entry_path(m
));
791 /* First, get rid of everything that is below if there
792 * is anything... Then, overmount it with an
793 * inaccessible path. */
794 (void) umount_recursive(mount_entry_path(m
), 0);
796 if (lstat(mount_entry_path(m
), &target
) < 0)
797 return log_debug_errno(errno
, "Failed to lstat() %s to determine what to mount over it: %m", mount_entry_path(m
));
799 what
= mode_to_inaccessible_node(target
.st_mode
);
801 log_debug("File type not supported for inaccessible mounts. Note that symlinks are not allowed");
809 r
= path_is_mount_point(mount_entry_path(m
), root_directory
, 0);
811 return log_debug_errno(r
, "Failed to determine whether %s is already a mount point: %m", mount_entry_path(m
));
812 if (r
> 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY bit for the mount point if needed. */
814 /* This isn't a mount point yet, let's make it one. */
815 what
= mount_entry_path(m
);
822 case BIND_MOUNT_RECURSIVE
:
823 /* Also chase the source mount */
825 r
= mount_entry_chase(root_directory
, m
, mount_entry_source(m
), &m
->source_malloc
);
829 what
= mount_entry_source(m
);
834 return mount_empty_dir(m
);
837 what
= mount_entry_source(m
);
842 return mount_private_dev(m
);
845 return mount_bind_dev(m
);
848 return mount_sysfs(m
);
851 return mount_procfs(m
);
854 assert_not_reached("Unknown mode");
859 if (mount(what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
) < 0) {
860 bool try_again
= false;
863 if (r
== -ENOENT
&& make
) {
866 /* Hmm, either the source or the destination are missing. Let's see if we can create the destination, then try again */
868 if (stat(what
, &st
) >= 0) {
870 (void) mkdir_parents(mount_entry_path(m
), 0755);
872 if (S_ISDIR(st
.st_mode
))
873 try_again
= mkdir(mount_entry_path(m
), 0755) >= 0;
875 try_again
= touch(mount_entry_path(m
)) >= 0;
880 if (mount(what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
) < 0)
887 return log_debug_errno(r
, "Failed to mount %s to %s: %m", what
, mount_entry_path(m
));
890 log_debug("Successfully mounted %s to %s", what
, mount_entry_path(m
));
894 static int make_read_only(const MountEntry
*m
, char **blacklist
, FILE *proc_self_mountinfo
) {
898 assert(proc_self_mountinfo
);
900 if (mount_entry_read_only(m
))
901 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), true, blacklist
, proc_self_mountinfo
);
902 else if (m
->mode
== PRIVATE_DEV
) { /* Superblock can be readonly but the submounts can't */
903 if (mount(NULL
, mount_entry_path(m
), NULL
, MS_REMOUNT
|DEV_MOUNT_OPTIONS
|MS_RDONLY
, NULL
) < 0)
908 /* Not that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked read-only
909 * already stays this way. This improves compatibility with container managers, where we won't attempt to undo
910 * read-only mounts already applied. */
912 if (r
== -ENOENT
&& m
->ignore
)
918 static bool namespace_info_mount_apivfs(const char *root_directory
, const NamespaceInfo
*ns_info
) {
922 * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
923 * since to protect the API VFS mounts, they need to be around in the
924 * first place... and RootDirectory= or RootImage= need to be set.
927 /* root_directory should point to a mount point */
928 return root_directory
&&
929 (ns_info
->mount_apivfs
||
930 ns_info
->protect_control_groups
||
931 ns_info
->protect_kernel_tunables
);
934 static unsigned namespace_calculate_mounts(
935 const char* root_directory
,
936 const NamespaceInfo
*ns_info
,
937 char** read_write_paths
,
938 char** read_only_paths
,
939 char** inaccessible_paths
,
940 char** empty_directories
,
941 const BindMount
*bind_mounts
,
942 unsigned n_bind_mounts
,
944 const char* var_tmp_dir
,
945 ProtectHome protect_home
,
946 ProtectSystem protect_system
) {
948 unsigned protect_home_cnt
;
949 unsigned protect_system_cnt
=
950 (protect_system
== PROTECT_SYSTEM_STRICT
?
951 ELEMENTSOF(protect_system_strict_table
) :
952 ((protect_system
== PROTECT_SYSTEM_FULL
) ?
953 ELEMENTSOF(protect_system_full_table
) :
954 ((protect_system
== PROTECT_SYSTEM_YES
) ?
955 ELEMENTSOF(protect_system_yes_table
) : 0)));
958 (protect_home
== PROTECT_HOME_YES
?
959 ELEMENTSOF(protect_home_yes_table
) :
960 ((protect_home
== PROTECT_HOME_READ_ONLY
) ?
961 ELEMENTSOF(protect_home_read_only_table
) : 0));
963 return !!tmp_dir
+ !!var_tmp_dir
+
964 strv_length(read_write_paths
) +
965 strv_length(read_only_paths
) +
966 strv_length(inaccessible_paths
) +
967 strv_length(empty_directories
) +
969 ns_info
->private_dev
+
970 (ns_info
->protect_kernel_tunables
? ELEMENTSOF(protect_kernel_tunables_table
) : 0) +
971 (ns_info
->protect_control_groups
? 1 : 0) +
972 (ns_info
->protect_kernel_modules
? ELEMENTSOF(protect_kernel_modules_table
) : 0) +
973 protect_home_cnt
+ protect_system_cnt
+
974 (namespace_info_mount_apivfs(root_directory
, ns_info
) ? ELEMENTSOF(apivfs_table
) : 0);
978 const char* root_directory
,
979 const char* root_image
,
980 const NamespaceInfo
*ns_info
,
981 char** read_write_paths
,
982 char** read_only_paths
,
983 char** inaccessible_paths
,
984 char** empty_directories
,
985 const BindMount
*bind_mounts
,
986 unsigned n_bind_mounts
,
988 const char* var_tmp_dir
,
989 ProtectHome protect_home
,
990 ProtectSystem protect_system
,
991 unsigned long mount_flags
,
992 DissectImageFlags dissect_image_flags
) {
994 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
995 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
996 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
997 _cleanup_free_
void *root_hash
= NULL
;
998 MountEntry
*m
, *mounts
= NULL
;
999 size_t root_hash_size
= 0;
1000 bool make_slave
= false;
1003 bool require_prefix
= false;
1008 if (mount_flags
== 0)
1009 mount_flags
= MS_SHARED
;
1012 dissect_image_flags
|= DISSECT_IMAGE_REQUIRE_ROOT
;
1014 if (protect_system
== PROTECT_SYSTEM_STRICT
&& strv_isempty(read_write_paths
))
1015 dissect_image_flags
|= DISSECT_IMAGE_READ_ONLY
;
1017 r
= loop_device_make_by_path(root_image
,
1018 dissect_image_flags
& DISSECT_IMAGE_READ_ONLY
? O_RDONLY
: O_RDWR
,
1023 r
= root_hash_load(root_image
, &root_hash
, &root_hash_size
);
1027 r
= dissect_image(loop_device
->fd
, root_hash
, root_hash_size
, dissect_image_flags
, &dissected_image
);
1031 r
= dissected_image_decrypt(dissected_image
, NULL
, root_hash
, root_hash_size
, dissect_image_flags
, &decrypted_image
);
1037 root
= root_directory
;
1038 else if (root_image
|| n_bind_mounts
> 0) {
1040 /* If we are booting from an image, create a mount point for the image, if it's still missing. We use
1041 * the same mount point for all images, which is safe, since they all live in their own namespaces
1042 * after all, and hence won't see each other. We also use such a root directory whenever there are bind
1043 * mounts configured, so that their source mounts are never obstructed by mounts we already applied
1044 * while we are applying them. */
1046 root
= "/run/systemd/unit-root";
1047 (void) mkdir_label(root
, 0700);
1048 require_prefix
= true;
1052 n_mounts
= namespace_calculate_mounts(
1059 bind_mounts
, n_bind_mounts
,
1060 tmp_dir
, var_tmp_dir
,
1061 protect_home
, protect_system
);
1063 /* Set mount slave mode */
1064 if (root
|| n_mounts
> 0)
1068 m
= mounts
= (MountEntry
*) alloca0(n_mounts
* sizeof(MountEntry
));
1069 r
= append_access_mounts(&m
, read_write_paths
, READWRITE
, require_prefix
);
1073 r
= append_access_mounts(&m
, read_only_paths
, READONLY
, require_prefix
);
1077 r
= append_access_mounts(&m
, inaccessible_paths
, INACCESSIBLE
, require_prefix
);
1081 r
= append_empty_dir_mounts(&m
, empty_directories
);
1085 r
= append_bind_mounts(&m
, bind_mounts
, n_bind_mounts
);
1090 *(m
++) = (MountEntry
) {
1091 .path_const
= "/tmp",
1092 .mode
= PRIVATE_TMP
,
1093 .source_const
= tmp_dir
,
1098 *(m
++) = (MountEntry
) {
1099 .path_const
= "/var/tmp",
1100 .mode
= PRIVATE_TMP
,
1101 .source_const
= var_tmp_dir
,
1105 if (ns_info
->private_dev
) {
1106 *(m
++) = (MountEntry
) {
1107 .path_const
= "/dev",
1108 .mode
= PRIVATE_DEV
,
1112 if (ns_info
->protect_kernel_tunables
) {
1113 r
= append_static_mounts(&m
, protect_kernel_tunables_table
, ELEMENTSOF(protect_kernel_tunables_table
), ns_info
->ignore_protect_paths
);
1118 if (ns_info
->protect_kernel_modules
) {
1119 r
= append_static_mounts(&m
, protect_kernel_modules_table
, ELEMENTSOF(protect_kernel_modules_table
), ns_info
->ignore_protect_paths
);
1124 if (ns_info
->protect_control_groups
) {
1125 *(m
++) = (MountEntry
) {
1126 .path_const
= "/sys/fs/cgroup",
1131 r
= append_protect_home(&m
, protect_home
, ns_info
->ignore_protect_paths
);
1135 r
= append_protect_system(&m
, protect_system
, false);
1139 if (namespace_info_mount_apivfs(root
, ns_info
)) {
1140 r
= append_static_mounts(&m
, apivfs_table
, ELEMENTSOF(apivfs_table
), ns_info
->ignore_protect_paths
);
1145 assert(mounts
+ n_mounts
== m
);
1147 /* Prepend the root directory where that's necessary */
1148 r
= prefix_where_needed(mounts
, n_mounts
, root
);
1152 qsort(mounts
, n_mounts
, sizeof(MountEntry
), mount_path_compare
);
1154 drop_duplicates(mounts
, &n_mounts
);
1155 drop_outside_root(root
, mounts
, &n_mounts
);
1156 drop_inaccessible(mounts
, &n_mounts
);
1157 drop_nop(mounts
, &n_mounts
);
1160 if (unshare(CLONE_NEWNS
) < 0) {
1166 /* Remount / as SLAVE so that nothing now mounted in the namespace
1167 shows up in the parent */
1168 if (mount(NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
) < 0) {
1175 /* A root image is specified, mount it to the right place */
1176 r
= dissected_image_mount(dissected_image
, root
, UID_INVALID
, dissect_image_flags
);
1180 if (decrypted_image
) {
1181 r
= decrypted_image_relinquish(decrypted_image
);
1186 loop_device_relinquish(loop_device
);
1188 } else if (root_directory
) {
1190 /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
1191 r
= path_is_mount_point(root
, NULL
, AT_SYMLINK_FOLLOW
);
1195 if (mount(root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0) {
1203 /* Let's mount the main root directory to the root directory to use */
1204 if (mount("/", root
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0) {
1210 /* Try to set up the new root directory before mounting anything else there. */
1211 if (root_image
|| root_directory
)
1212 (void) base_filesystem_create(root
, UID_INVALID
, GID_INVALID
);
1215 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
1219 /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of /proc.
1220 * For example, this is the case with the option: 'InaccessiblePaths=/proc' */
1221 proc_self_mountinfo
= fopen("/proc/self/mountinfo", "re");
1222 if (!proc_self_mountinfo
) {
1227 /* First round, add in all special mounts we need */
1228 for (m
= mounts
; m
< mounts
+ n_mounts
; ++m
) {
1229 r
= apply_mount(root
, m
);
1234 /* Create a blacklist we can pass to bind_mount_recursive() */
1235 blacklist
= newa(char*, n_mounts
+1);
1236 for (j
= 0; j
< n_mounts
; j
++)
1237 blacklist
[j
] = (char*) mount_entry_path(mounts
+j
);
1238 blacklist
[j
] = NULL
;
1240 /* Second round, flip the ro bits if necessary. */
1241 for (m
= mounts
; m
< mounts
+ n_mounts
; ++m
) {
1242 r
= make_read_only(m
, blacklist
, proc_self_mountinfo
);
1249 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
1250 r
= mount_move_root(root
);
1255 /* Remount / as the desired mode. Note that this will not
1256 * reestablish propagation from our side to the host, since
1257 * what's disconnected is disconnected. */
1258 if (mount(NULL
, "/", NULL
, mount_flags
| MS_REC
, NULL
) < 0) {
1266 for (m
= mounts
; m
< mounts
+ n_mounts
; m
++)
1267 mount_entry_done(m
);
1272 void bind_mount_free_many(BindMount
*b
, unsigned n
) {
1275 assert(b
|| n
== 0);
1277 for (i
= 0; i
< n
; i
++) {
1279 free(b
[i
].destination
);
1285 int bind_mount_add(BindMount
**b
, unsigned *n
, const BindMount
*item
) {
1286 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
1293 s
= strdup(item
->source
);
1297 d
= strdup(item
->destination
);
1301 c
= realloc_multiply(*b
, sizeof(BindMount
), *n
+ 1);
1307 c
[(*n
) ++] = (BindMount
) {
1310 .read_only
= item
->read_only
,
1311 .recursive
= item
->recursive
,
1312 .ignore_enoent
= item
->ignore_enoent
,
1319 static int setup_one_tmp_dir(const char *id
, const char *prefix
, char **path
) {
1320 _cleanup_free_
char *x
= NULL
;
1321 char bid
[SD_ID128_STRING_MAX
];
1329 /* We include the boot id in the directory so that after a
1330 * reboot we can easily identify obsolete directories. */
1332 r
= sd_id128_get_boot(&boot_id
);
1336 x
= strjoin(prefix
, "/systemd-private-", sd_id128_to_string(boot_id
, bid
), "-", id
, "-XXXXXX");
1340 RUN_WITH_UMASK(0077)
1344 RUN_WITH_UMASK(0000) {
1347 y
= strjoina(x
, "/tmp");
1349 if (mkdir(y
, 0777 | S_ISVTX
) < 0)
1359 int setup_tmp_dirs(const char *id
, char **tmp_dir
, char **var_tmp_dir
) {
1365 assert(var_tmp_dir
);
1367 r
= setup_one_tmp_dir(id
, "/tmp", &a
);
1371 r
= setup_one_tmp_dir(id
, "/var/tmp", &b
);
1375 t
= strjoina(a
, "/tmp");
1389 int setup_netns(int netns_storage_socket
[2]) {
1390 _cleanup_close_
int netns
= -1;
1393 assert(netns_storage_socket
);
1394 assert(netns_storage_socket
[0] >= 0);
1395 assert(netns_storage_socket
[1] >= 0);
1397 /* We use the passed socketpair as a storage buffer for our
1398 * namespace reference fd. Whatever process runs this first
1399 * shall create a new namespace, all others should just join
1400 * it. To serialize that we use a file lock on the socket
1403 * It's a bit crazy, but hey, works great! */
1405 if (lockf(netns_storage_socket
[0], F_LOCK
, 0) < 0)
1408 netns
= receive_one_fd(netns_storage_socket
[0], MSG_DONTWAIT
);
1409 if (netns
== -EAGAIN
) {
1410 /* Nothing stored yet, so let's create a new namespace */
1412 if (unshare(CLONE_NEWNET
) < 0) {
1419 netns
= open("/proc/self/ns/net", O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
1427 } else if (netns
< 0) {
1432 /* Yay, found something, so let's join the namespace */
1433 if (setns(netns
, CLONE_NEWNET
) < 0) {
1441 q
= send_one_fd(netns_storage_socket
[1], netns
, MSG_DONTWAIT
);
1448 (void) lockf(netns_storage_socket
[0], F_ULOCK
, 0);
1452 bool ns_type_supported(NamespaceType type
) {
1453 const char *t
, *ns_proc
;
1455 t
= namespace_type_to_string(type
);
1456 if (!t
) /* Don't know how to translate this? Then it's not supported */
1459 ns_proc
= strjoina("/proc/self/ns/", t
);
1460 return access(ns_proc
, F_OK
) == 0;
1463 static const char *const protect_home_table
[_PROTECT_HOME_MAX
] = {
1464 [PROTECT_HOME_NO
] = "no",
1465 [PROTECT_HOME_YES
] = "yes",
1466 [PROTECT_HOME_READ_ONLY
] = "read-only",
1469 DEFINE_STRING_TABLE_LOOKUP(protect_home
, ProtectHome
);
1471 ProtectHome
parse_protect_home_or_bool(const char *s
) {
1474 r
= parse_boolean(s
);
1476 return PROTECT_HOME_YES
;
1478 return PROTECT_HOME_NO
;
1480 return protect_home_from_string(s
);
1483 static const char *const protect_system_table
[_PROTECT_SYSTEM_MAX
] = {
1484 [PROTECT_SYSTEM_NO
] = "no",
1485 [PROTECT_SYSTEM_YES
] = "yes",
1486 [PROTECT_SYSTEM_FULL
] = "full",
1487 [PROTECT_SYSTEM_STRICT
] = "strict",
1490 DEFINE_STRING_TABLE_LOOKUP(protect_system
, ProtectSystem
);
1492 ProtectSystem
parse_protect_system_or_bool(const char *s
) {
1495 r
= parse_boolean(s
);
1497 return PROTECT_SYSTEM_YES
;
1499 return PROTECT_SYSTEM_NO
;
1501 return protect_system_from_string(s
);
1504 static const char* const namespace_type_table
[] = {
1505 [NAMESPACE_MOUNT
] = "mnt",
1506 [NAMESPACE_CGROUP
] = "cgroup",
1507 [NAMESPACE_UTS
] = "uts",
1508 [NAMESPACE_IPC
] = "ipc",
1509 [NAMESPACE_USER
] = "user",
1510 [NAMESPACE_PID
] = "pid",
1511 [NAMESPACE_NET
] = "net",
1514 DEFINE_STRING_TABLE_LOOKUP(namespace_type
, NamespaceType
);