1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 Copyright 2010 Lennart Poettering
10 #include <sys/mount.h>
15 #include "alloc-util.h"
16 #include "base-filesystem.h"
17 #include "dev-setup.h"
21 #include "loop-util.h"
22 #include "loopback-setup.h"
25 #include "mount-util.h"
26 #include "namespace.h"
27 #include "path-util.h"
28 #include "selinux-util.h"
29 #include "socket-util.h"
30 #include "stat-util.h"
31 #include "string-table.h"
32 #include "string-util.h"
34 #include "umask-util.h"
35 #include "user-util.h"
38 #define DEV_MOUNT_OPTIONS (MS_NOSUID|MS_STRICTATIME|MS_NOEXEC)
40 typedef enum MountMode
{
41 /* This is ordered by priority! */
56 typedef struct MountEntry
{
57 const char *path_const
; /* Memory allocated on stack or static */
59 bool ignore
:1; /* Ignore if path does not exist? */
60 bool has_prefix
:1; /* Already is prefixed by the root dir? */
61 bool read_only
:1; /* Shall this mount point be read-only? */
62 bool applied
:1; /* Already applied */
63 char *path_malloc
; /* Use this instead of 'path_const' if we had to allocate memory */
64 const char *source_const
; /* The source path, for bind mounts */
66 const char *options_const
;/* Mount options for tmpfs */
68 unsigned long flags
; /* Mount flags used by EMPTY_DIR and TMPFS. Do not include MS_RDONLY here, but please use read_only. */
72 /* If MountAPIVFS= is used, let's mount /sys and /proc into the it, but only as a fallback if the user hasn't mounted
73 * something there already. These mounts are hence overriden by any other explicitly configured mounts. */
74 static const MountEntry apivfs_table
[] = {
75 { "/proc", PROCFS
, false },
76 { "/dev", BIND_DEV
, false },
77 { "/sys", SYSFS
, false },
80 /* ProtectKernelTunables= option and the related filesystem APIs */
81 static const MountEntry protect_kernel_tunables_table
[] = {
82 { "/proc/acpi", READONLY
, true },
83 { "/proc/apm", READONLY
, true }, /* Obsolete API, there's no point in permitting access to this, ever */
84 { "/proc/asound", READONLY
, true },
85 { "/proc/bus", READONLY
, true },
86 { "/proc/fs", READONLY
, true },
87 { "/proc/irq", READONLY
, true },
88 { "/proc/kallsyms", INACCESSIBLE
, true },
89 { "/proc/kcore", INACCESSIBLE
, true },
90 { "/proc/latency_stats", READONLY
, true },
91 { "/proc/mtrr", READONLY
, true },
92 { "/proc/scsi", READONLY
, true },
93 { "/proc/sys", READONLY
, false },
94 { "/proc/sysrq-trigger", READONLY
, true },
95 { "/proc/timer_stats", READONLY
, true },
96 { "/sys", READONLY
, false },
97 { "/sys/fs/bpf", READONLY
, true },
98 { "/sys/fs/cgroup", READWRITE
, false }, /* READONLY is set by ProtectControlGroups= option */
99 { "/sys/fs/selinux", READWRITE
, true },
100 { "/sys/kernel/debug", READONLY
, true },
101 { "/sys/kernel/tracing", READONLY
, true },
104 /* ProtectKernelModules= option */
105 static const MountEntry protect_kernel_modules_table
[] = {
107 { "/lib/modules", INACCESSIBLE
, true },
109 { "/usr/lib/modules", INACCESSIBLE
, true },
113 * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of
114 * system should be protected by ProtectSystem=
116 static const MountEntry protect_home_read_only_table
[] = {
117 { "/home", READONLY
, true },
118 { "/run/user", READONLY
, true },
119 { "/root", READONLY
, true },
122 /* ProtectHome=tmpfs table */
123 static const MountEntry protect_home_tmpfs_table
[] = {
124 { "/home", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755", .flags
= MS_NODEV
|MS_STRICTATIME
},
125 { "/run/user", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755", .flags
= MS_NODEV
|MS_STRICTATIME
},
126 { "/root", TMPFS
, true, .read_only
= true, .options_const
= "mode=0700", .flags
= MS_NODEV
|MS_STRICTATIME
},
129 /* ProtectHome=yes table */
130 static const MountEntry protect_home_yes_table
[] = {
131 { "/home", INACCESSIBLE
, true },
132 { "/run/user", INACCESSIBLE
, true },
133 { "/root", INACCESSIBLE
, true },
136 /* ProtectSystem=yes table */
137 static const MountEntry protect_system_yes_table
[] = {
138 { "/usr", READONLY
, false },
139 { "/boot", READONLY
, true },
140 { "/efi", READONLY
, true },
142 { "/lib", READONLY
, true },
143 { "/lib64", READONLY
, true },
144 { "/bin", READONLY
, true },
146 { "/sbin", READONLY
, true },
151 /* ProtectSystem=full includes ProtectSystem=yes */
152 static const MountEntry protect_system_full_table
[] = {
153 { "/usr", READONLY
, false },
154 { "/boot", READONLY
, true },
155 { "/efi", READONLY
, true },
156 { "/etc", READONLY
, false },
158 { "/lib", READONLY
, true },
159 { "/lib64", READONLY
, true },
160 { "/bin", READONLY
, true },
162 { "/sbin", READONLY
, true },
168 * ProtectSystem=strict table. In this strict mode, we mount everything
169 * read-only, except for /proc, /dev, /sys which are the kernel API VFS,
170 * which are left writable, but PrivateDevices= + ProtectKernelTunables=
171 * protect those, and these options should be fully orthogonal.
172 * (And of course /home and friends are also left writable, as ProtectHome=
173 * shall manage those, orthogonally).
175 static const MountEntry protect_system_strict_table
[] = {
176 { "/", READONLY
, false },
177 { "/proc", READWRITE
, false }, /* ProtectKernelTunables= */
178 { "/sys", READWRITE
, false }, /* ProtectKernelTunables= */
179 { "/dev", READWRITE
, false }, /* PrivateDevices= */
180 { "/home", READWRITE
, true }, /* ProtectHome= */
181 { "/run/user", READWRITE
, true }, /* ProtectHome= */
182 { "/root", READWRITE
, true }, /* ProtectHome= */
185 static const char *mount_entry_path(const MountEntry
*p
) {
188 /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that,
189 * otherwise the stack/static ->path field is returned. */
191 return p
->path_malloc
?: p
->path_const
;
194 static bool mount_entry_read_only(const MountEntry
*p
) {
197 return p
->read_only
|| IN_SET(p
->mode
, READONLY
, INACCESSIBLE
);
200 static const char *mount_entry_source(const MountEntry
*p
) {
203 return p
->source_malloc
?: p
->source_const
;
206 static const char *mount_entry_options(const MountEntry
*p
) {
209 return p
->options_malloc
?: p
->options_const
;
212 static void mount_entry_done(MountEntry
*p
) {
215 p
->path_malloc
= mfree(p
->path_malloc
);
216 p
->source_malloc
= mfree(p
->source_malloc
);
217 p
->options_malloc
= mfree(p
->options_malloc
);
220 static int append_access_mounts(MountEntry
**p
, char **strv
, MountMode mode
, bool forcibly_require_prefix
) {
225 /* Adds a list of user-supplied READWRITE/READONLY/INACCESSIBLE entries */
227 STRV_FOREACH(i
, strv
) {
228 bool ignore
= false, needs_prefix
= false;
231 /* Look for any prefixes */
232 if (startswith(e
, "-")) {
236 if (startswith(e
, "+")) {
241 if (!path_is_absolute(e
))
244 *((*p
)++) = (MountEntry
) {
248 .has_prefix
= !needs_prefix
&& !forcibly_require_prefix
,
255 static int append_empty_dir_mounts(MountEntry
**p
, char **strv
) {
260 /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the
261 * "/private/" boundary directories for DynamicUser=1. */
263 STRV_FOREACH(i
, strv
) {
265 *((*p
)++) = (MountEntry
) {
271 .options_const
= "mode=755",
272 .flags
= MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
,
279 static int append_bind_mounts(MountEntry
**p
, const BindMount
*binds
, size_t n
) {
284 for (i
= 0; i
< n
; i
++) {
285 const BindMount
*b
= binds
+ i
;
287 *((*p
)++) = (MountEntry
) {
288 .path_const
= b
->destination
,
289 .mode
= b
->recursive
? BIND_MOUNT_RECURSIVE
: BIND_MOUNT
,
290 .read_only
= b
->read_only
,
291 .source_const
= b
->source
,
292 .ignore
= b
->ignore_enoent
,
299 static int append_tmpfs_mounts(MountEntry
**p
, const TemporaryFileSystem
*tmpfs
, size_t n
) {
305 for (i
= 0; i
< n
; i
++) {
306 const TemporaryFileSystem
*t
= tmpfs
+ i
;
307 _cleanup_free_
char *o
= NULL
, *str
= NULL
;
308 unsigned long flags
= MS_NODEV
|MS_STRICTATIME
;
311 if (!path_is_absolute(t
->path
))
314 if (!isempty(t
->options
)) {
315 str
= strjoin("mode=0755,", t
->options
);
319 r
= mount_option_mangle(str
, MS_NODEV
|MS_STRICTATIME
, &flags
, &o
);
323 ro
= flags
& MS_RDONLY
;
328 *((*p
)++) = (MountEntry
) {
329 .path_const
= t
->path
,
342 static int append_static_mounts(MountEntry
**p
, const MountEntry
*mounts
, size_t n
, bool ignore_protect
) {
348 /* Adds a list of static pre-defined entries */
350 for (i
= 0; i
< n
; i
++)
351 *((*p
)++) = (MountEntry
) {
352 .path_const
= mount_entry_path(mounts
+i
),
353 .mode
= mounts
[i
].mode
,
354 .ignore
= mounts
[i
].ignore
|| ignore_protect
,
360 static int append_protect_home(MountEntry
**p
, ProtectHome protect_home
, bool ignore_protect
) {
363 switch (protect_home
) {
365 case PROTECT_HOME_NO
:
368 case PROTECT_HOME_READ_ONLY
:
369 return append_static_mounts(p
, protect_home_read_only_table
, ELEMENTSOF(protect_home_read_only_table
), ignore_protect
);
371 case PROTECT_HOME_TMPFS
:
372 return append_static_mounts(p
, protect_home_tmpfs_table
, ELEMENTSOF(protect_home_tmpfs_table
), ignore_protect
);
374 case PROTECT_HOME_YES
:
375 return append_static_mounts(p
, protect_home_yes_table
, ELEMENTSOF(protect_home_yes_table
), ignore_protect
);
378 assert_not_reached("Unexpected ProtectHome= value");
382 static int append_protect_system(MountEntry
**p
, ProtectSystem protect_system
, bool ignore_protect
) {
385 switch (protect_system
) {
387 case PROTECT_SYSTEM_NO
:
390 case PROTECT_SYSTEM_STRICT
:
391 return append_static_mounts(p
, protect_system_strict_table
, ELEMENTSOF(protect_system_strict_table
), ignore_protect
);
393 case PROTECT_SYSTEM_YES
:
394 return append_static_mounts(p
, protect_system_yes_table
, ELEMENTSOF(protect_system_yes_table
), ignore_protect
);
396 case PROTECT_SYSTEM_FULL
:
397 return append_static_mounts(p
, protect_system_full_table
, ELEMENTSOF(protect_system_full_table
), ignore_protect
);
400 assert_not_reached("Unexpected ProtectSystem= value");
404 static int mount_path_compare(const void *a
, const void *b
) {
405 const MountEntry
*p
= a
, *q
= b
;
408 /* If the paths are not equal, then order prefixes first */
409 d
= path_compare(mount_entry_path(p
), mount_entry_path(q
));
413 /* If the paths are equal, check the mode */
414 if (p
->mode
< q
->mode
)
416 if (p
->mode
> q
->mode
)
422 static int prefix_where_needed(MountEntry
*m
, size_t n
, const char *root_directory
) {
425 /* Prefixes all paths in the bind mount table with the root directory if it is specified and the entry needs
431 for (i
= 0; i
< n
; i
++) {
437 s
= prefix_root(root_directory
, mount_entry_path(m
+i
));
441 free_and_replace(m
[i
].path_malloc
, s
);
442 m
[i
].has_prefix
= true;
448 static void drop_duplicates(MountEntry
*m
, size_t *n
) {
449 MountEntry
*f
, *t
, *previous
;
454 /* Drops duplicate entries. Expects that the array is properly ordered already. */
456 for (f
= m
, t
= m
, previous
= NULL
; f
< m
+ *n
; f
++) {
458 /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare()
459 * above. Note that we only drop duplicates that haven't been mounted yet. */
461 path_equal(mount_entry_path(f
), mount_entry_path(previous
)) &&
462 !f
->applied
&& !previous
->applied
) {
463 log_debug("%s is duplicate.", mount_entry_path(f
));
464 previous
->read_only
= previous
->read_only
|| mount_entry_read_only(f
); /* Propagate the read-only flag to the remaining entry */
477 static void drop_inaccessible(MountEntry
*m
, size_t *n
) {
479 const char *clear
= NULL
;
484 /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly
485 * ordered already. */
487 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
489 /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop
490 * it, as inaccessible paths really should drop the entire subtree. */
491 if (clear
&& path_startswith(mount_entry_path(f
), clear
)) {
492 log_debug("%s is masked by %s.", mount_entry_path(f
), clear
);
497 clear
= f
->mode
== INACCESSIBLE
? mount_entry_path(f
) : NULL
;
506 static void drop_nop(MountEntry
*m
, size_t *n
) {
512 /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the
513 * list is ordered by prefixes. */
515 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
517 /* Only suppress such subtrees for READONLY and READWRITE entries */
518 if (IN_SET(f
->mode
, READONLY
, READWRITE
)) {
522 /* Now let's find the first parent of the entry we are looking at. */
523 for (p
= t
-1; p
>= m
; p
--) {
524 if (path_startswith(mount_entry_path(f
), mount_entry_path(p
))) {
530 /* We found it, let's see if it's the same mode, if so, we can drop this entry */
531 if (found
&& p
->mode
== f
->mode
) {
532 log_debug("%s is redundant by %s", mount_entry_path(f
), mount_entry_path(p
));
545 static void drop_outside_root(const char *root_directory
, MountEntry
*m
, size_t *n
) {
555 /* Drops all mounts that are outside of the root directory. */
557 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
559 if (!path_startswith(mount_entry_path(f
), root_directory
)) {
560 log_debug("%s is outside of root directory.", mount_entry_path(f
));
572 static int clone_device_node(const char *d
, const char *temporary_mount
, bool *make_devnode
) {
577 if (stat(d
, &st
) < 0) {
583 if (!S_ISBLK(st
.st_mode
) &&
584 !S_ISCHR(st
.st_mode
))
590 dn
= strjoina(temporary_mount
, d
);
593 mac_selinux_create_file_prepare(d
, st
.st_mode
);
594 r
= mknod(dn
, st
.st_mode
, st
.st_rdev
);
595 mac_selinux_create_file_clear();
600 return log_debug_errno(errno
, "mknod failed for %s: %m", d
);
602 *make_devnode
= false;
605 /* We're about to fallback to bind-mounting the device
606 * node. So create a dummy bind-mount target. */
607 mac_selinux_create_file_prepare(d
, 0);
608 r
= mknod(dn
, S_IFREG
, 0);
609 mac_selinux_create_file_clear();
611 if (r
< 0 && errno
!= EEXIST
)
612 return log_debug_errno(errno
, "mknod fallback failed for %s: %m", d
);
614 /* Fallback to bind-mounting:
615 * The assumption here is that all used device nodes carry standard
616 * properties. Specifically, the devices nodes we bind-mount should
617 * either be owned by root:root or root:tty (e.g. /dev/tty, /dev/ptmx)
618 * and should not carry ACLs. */
619 if (mount(d
, dn
, NULL
, MS_BIND
, NULL
) < 0)
620 return log_debug_errno(errno
, "mount failed for %s: %m", d
);
625 static int mount_private_dev(MountEntry
*m
) {
626 static const char devnodes
[] =
634 char temporary_mount
[] = "/tmp/namespace-dev-XXXXXX";
635 const char *d
, *dev
= NULL
, *devpts
= NULL
, *devshm
= NULL
, *devhugepages
= NULL
, *devmqueue
= NULL
, *devlog
= NULL
, *devptmx
= NULL
;
636 bool can_mknod
= true;
637 _cleanup_umask_ mode_t u
;
644 if (!mkdtemp(temporary_mount
))
647 dev
= strjoina(temporary_mount
, "/dev");
648 (void) mkdir(dev
, 0755);
649 if (mount("tmpfs", dev
, "tmpfs", DEV_MOUNT_OPTIONS
, "mode=755") < 0) {
654 devpts
= strjoina(temporary_mount
, "/dev/pts");
655 (void) mkdir(devpts
, 0755);
656 if (mount("/dev/pts", devpts
, NULL
, MS_BIND
, NULL
) < 0) {
661 /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx
662 * when /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible
663 * thus, in that case make a clone
665 * in nspawn and other containers it will be a symlink, in that case make it a symlink
667 r
= is_symlink("/dev/ptmx");
671 devptmx
= strjoina(temporary_mount
, "/dev/ptmx");
672 if (symlink("pts/ptmx", devptmx
) < 0) {
677 r
= clone_device_node("/dev/ptmx", temporary_mount
, &can_mknod
);
682 devshm
= strjoina(temporary_mount
, "/dev/shm");
683 (void) mkdir(devshm
, 0755);
684 r
= mount("/dev/shm", devshm
, NULL
, MS_BIND
, NULL
);
690 devmqueue
= strjoina(temporary_mount
, "/dev/mqueue");
691 (void) mkdir(devmqueue
, 0755);
692 (void) mount("/dev/mqueue", devmqueue
, NULL
, MS_BIND
, NULL
);
694 devhugepages
= strjoina(temporary_mount
, "/dev/hugepages");
695 (void) mkdir(devhugepages
, 0755);
696 (void) mount("/dev/hugepages", devhugepages
, NULL
, MS_BIND
, NULL
);
698 devlog
= strjoina(temporary_mount
, "/dev/log");
699 (void) symlink("/run/systemd/journal/dev-log", devlog
);
701 NULSTR_FOREACH(d
, devnodes
) {
702 r
= clone_device_node(d
, temporary_mount
, &can_mknod
);
703 /* ENXIO means the the *source* is not a device file, skip creation in that case */
704 if (r
< 0 && r
!= -ENXIO
)
708 dev_setup(temporary_mount
, UID_INVALID
, GID_INVALID
);
710 /* Create the /dev directory if missing. It is more likely to be
711 * missing when the service is started with RootDirectory. This is
712 * consistent with mount units creating the mount points when missing.
714 (void) mkdir_p_label(mount_entry_path(m
), 0755);
716 /* Unmount everything in old /dev */
717 umount_recursive(mount_entry_path(m
), 0);
718 if (mount(dev
, mount_entry_path(m
), NULL
, MS_MOVE
, NULL
) < 0) {
724 rmdir(temporary_mount
);
736 umount(devhugepages
);
743 rmdir(temporary_mount
);
748 static int mount_bind_dev(const MountEntry
*m
) {
753 /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the service's
754 * /dev. This is only used when RootDirectory= is set. */
756 (void) mkdir_p_label(mount_entry_path(m
), 0755);
758 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
760 return log_debug_errno(r
, "Unable to determine whether /dev is already mounted: %m");
761 if (r
> 0) /* make this a NOP if /dev is already a mount point */
764 if (mount("/dev", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
765 return log_debug_errno(errno
, "Failed to bind mount %s: %m", mount_entry_path(m
));
770 static int mount_sysfs(const MountEntry
*m
) {
775 (void) mkdir_p_label(mount_entry_path(m
), 0755);
777 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
779 return log_debug_errno(r
, "Unable to determine whether /sys is already mounted: %m");
780 if (r
> 0) /* make this a NOP if /sys is already a mount point */
783 /* Bind mount the host's version so that we get all child mounts of it, too. */
784 if (mount("/sys", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
785 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
790 static int mount_procfs(const MountEntry
*m
) {
795 (void) mkdir_p_label(mount_entry_path(m
), 0755);
797 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
799 return log_debug_errno(r
, "Unable to determine whether /proc is already mounted: %m");
800 if (r
> 0) /* make this a NOP if /proc is already a mount point */
803 /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in one */
804 if (mount("proc", mount_entry_path(m
), "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
) < 0)
805 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
810 static int mount_tmpfs(const MountEntry
*m
) {
813 /* First, get rid of everything that is below if there is anything. Then, overmount with our new tmpfs */
815 (void) mkdir_p_label(mount_entry_path(m
), 0755);
816 (void) umount_recursive(mount_entry_path(m
), 0);
818 if (mount("tmpfs", mount_entry_path(m
), "tmpfs", m
->flags
, mount_entry_options(m
)) < 0)
819 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
824 static int follow_symlink(
825 const char *root_directory
,
828 _cleanup_free_
char *target
= NULL
;
831 /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we
832 * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at
833 * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the
834 * end and already have a fully normalized name. */
836 r
= chase_symlinks(mount_entry_path(m
), root_directory
, CHASE_STEP
|CHASE_NONEXISTENT
, &target
);
838 return log_debug_errno(r
, "Failed to chase symlinks '%s': %m", mount_entry_path(m
));
839 if (r
> 0) /* Reached the end, nothing more to resolve */
842 if (m
->n_followed
>= CHASE_SYMLINKS_MAX
) { /* put a boundary on things */
843 log_debug("Symlink loop on '%s'.", mount_entry_path(m
));
847 log_debug("Followed mount entry path symlink %s → %s.", mount_entry_path(m
), target
);
849 free_and_replace(m
->path_malloc
, target
);
850 m
->has_prefix
= true;
857 static int apply_mount(
858 const char *root_directory
,
861 bool rbind
= true, make
= false;
867 log_debug("Applying namespace mount on %s", mount_entry_path(m
));
874 /* First, get rid of everything that is below if there
875 * is anything... Then, overmount it with an
876 * inaccessible path. */
877 (void) umount_recursive(mount_entry_path(m
), 0);
879 if (lstat(mount_entry_path(m
), &target
) < 0) {
880 if (errno
== ENOENT
&& m
->ignore
)
883 return log_debug_errno(errno
, "Failed to lstat() %s to determine what to mount over it: %m", mount_entry_path(m
));
886 what
= mode_to_inaccessible_node(target
.st_mode
);
888 log_debug("File type not supported for inaccessible mounts. Note that symlinks are not allowed");
896 r
= path_is_mount_point(mount_entry_path(m
), root_directory
, 0);
897 if (r
== -ENOENT
&& m
->ignore
)
900 return log_debug_errno(r
, "Failed to determine whether %s is already a mount point: %m", mount_entry_path(m
));
901 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. */
903 /* This isn't a mount point yet, let's make it one. */
904 what
= mount_entry_path(m
);
911 case BIND_MOUNT_RECURSIVE
: {
912 _cleanup_free_
char *chased
= NULL
;
914 /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note that bind
915 * mount source paths are always relative to the host root, hence we pass NULL as root directory to
916 * chase_symlinks() here. */
918 r
= chase_symlinks(mount_entry_source(m
), NULL
, CHASE_TRAIL_SLASH
, &chased
);
919 if (r
== -ENOENT
&& m
->ignore
) {
920 log_debug_errno(r
, "Path %s does not exist, ignoring.", mount_entry_source(m
));
924 return log_debug_errno(r
, "Failed to follow symlinks on %s: %m", mount_entry_source(m
));
926 log_debug("Followed source symlinks %s → %s.", mount_entry_source(m
), chased
);
928 free_and_replace(m
->source_malloc
, chased
);
930 what
= mount_entry_source(m
);
937 return mount_tmpfs(m
);
940 what
= mount_entry_source(m
);
945 return mount_private_dev(m
);
948 return mount_bind_dev(m
);
951 return mount_sysfs(m
);
954 return mount_procfs(m
);
957 assert_not_reached("Unknown mode");
962 if (mount(what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
) < 0) {
963 bool try_again
= false;
966 if (r
== -ENOENT
&& make
) {
969 /* Hmm, either the source or the destination are missing. Let's see if we can create the destination, then try again */
971 if (stat(what
, &st
) < 0)
972 log_debug_errno(errno
, "Mount point source '%s' is not accessible: %m", what
);
976 (void) mkdir_parents(mount_entry_path(m
), 0755);
978 if (S_ISDIR(st
.st_mode
))
979 q
= mkdir(mount_entry_path(m
), 0755) < 0 ? -errno
: 0;
981 q
= touch(mount_entry_path(m
));
984 log_debug_errno(q
, "Failed to create destination mount point node '%s': %m", mount_entry_path(m
));
991 if (mount(what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
) < 0)
998 return log_debug_errno(r
, "Failed to mount %s to %s: %m", what
, mount_entry_path(m
));
1001 log_debug("Successfully mounted %s to %s", what
, mount_entry_path(m
));
1005 static int make_read_only(const MountEntry
*m
, char **blacklist
, FILE *proc_self_mountinfo
) {
1009 assert(proc_self_mountinfo
);
1011 if (mount_entry_read_only(m
)) {
1012 if (IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
)) {
1013 /* Make superblock readonly */
1014 if (mount(NULL
, mount_entry_path(m
), NULL
, MS_REMOUNT
| MS_RDONLY
| m
->flags
, mount_entry_options(m
)) < 0)
1017 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), true, blacklist
, proc_self_mountinfo
);
1018 } else if (m
->mode
== PRIVATE_DEV
) {
1019 /* Superblock can be readonly but the submounts can't */
1020 if (mount(NULL
, mount_entry_path(m
), NULL
, MS_REMOUNT
|DEV_MOUNT_OPTIONS
|MS_RDONLY
, NULL
) < 0)
1025 /* Not that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked read-only
1026 * already stays this way. This improves compatibility with container managers, where we won't attempt to undo
1027 * read-only mounts already applied. */
1029 if (r
== -ENOENT
&& m
->ignore
)
1035 static bool namespace_info_mount_apivfs(const char *root_directory
, const NamespaceInfo
*ns_info
) {
1039 * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
1040 * since to protect the API VFS mounts, they need to be around in the
1041 * first place... and RootDirectory= or RootImage= need to be set.
1044 /* root_directory should point to a mount point */
1045 return root_directory
&&
1046 (ns_info
->mount_apivfs
||
1047 ns_info
->protect_control_groups
||
1048 ns_info
->protect_kernel_tunables
);
1051 static size_t namespace_calculate_mounts(
1052 const char* root_directory
,
1053 const NamespaceInfo
*ns_info
,
1054 char** read_write_paths
,
1055 char** read_only_paths
,
1056 char** inaccessible_paths
,
1057 char** empty_directories
,
1058 size_t n_bind_mounts
,
1059 size_t n_temporary_filesystems
,
1060 const char* tmp_dir
,
1061 const char* var_tmp_dir
,
1062 ProtectHome protect_home
,
1063 ProtectSystem protect_system
) {
1065 size_t protect_home_cnt
;
1066 size_t protect_system_cnt
=
1067 (protect_system
== PROTECT_SYSTEM_STRICT
?
1068 ELEMENTSOF(protect_system_strict_table
) :
1069 ((protect_system
== PROTECT_SYSTEM_FULL
) ?
1070 ELEMENTSOF(protect_system_full_table
) :
1071 ((protect_system
== PROTECT_SYSTEM_YES
) ?
1072 ELEMENTSOF(protect_system_yes_table
) : 0)));
1075 (protect_home
== PROTECT_HOME_YES
?
1076 ELEMENTSOF(protect_home_yes_table
) :
1077 ((protect_home
== PROTECT_HOME_READ_ONLY
) ?
1078 ELEMENTSOF(protect_home_read_only_table
) :
1079 ((protect_home
== PROTECT_HOME_TMPFS
) ?
1080 ELEMENTSOF(protect_home_tmpfs_table
) : 0)));
1082 return !!tmp_dir
+ !!var_tmp_dir
+
1083 strv_length(read_write_paths
) +
1084 strv_length(read_only_paths
) +
1085 strv_length(inaccessible_paths
) +
1086 strv_length(empty_directories
) +
1088 n_temporary_filesystems
+
1089 ns_info
->private_dev
+
1090 (ns_info
->protect_kernel_tunables
? ELEMENTSOF(protect_kernel_tunables_table
) : 0) +
1091 (ns_info
->protect_control_groups
? 1 : 0) +
1092 (ns_info
->protect_kernel_modules
? ELEMENTSOF(protect_kernel_modules_table
) : 0) +
1093 protect_home_cnt
+ protect_system_cnt
+
1094 (namespace_info_mount_apivfs(root_directory
, ns_info
) ? ELEMENTSOF(apivfs_table
) : 0);
1097 static void normalize_mounts(const char *root_directory
, MountEntry
*mounts
, size_t *n_mounts
) {
1099 assert(mounts
|| *n_mounts
== 0);
1101 qsort_safe(mounts
, *n_mounts
, sizeof(MountEntry
), mount_path_compare
);
1103 drop_duplicates(mounts
, n_mounts
);
1104 drop_outside_root(root_directory
, mounts
, n_mounts
);
1105 drop_inaccessible(mounts
, n_mounts
);
1106 drop_nop(mounts
, n_mounts
);
1109 int setup_namespace(
1110 const char* root_directory
,
1111 const char* root_image
,
1112 const NamespaceInfo
*ns_info
,
1113 char** read_write_paths
,
1114 char** read_only_paths
,
1115 char** inaccessible_paths
,
1116 char** empty_directories
,
1117 const BindMount
*bind_mounts
,
1118 size_t n_bind_mounts
,
1119 const TemporaryFileSystem
*temporary_filesystems
,
1120 size_t n_temporary_filesystems
,
1121 const char* tmp_dir
,
1122 const char* var_tmp_dir
,
1123 ProtectHome protect_home
,
1124 ProtectSystem protect_system
,
1125 unsigned long mount_flags
,
1126 DissectImageFlags dissect_image_flags
) {
1128 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
1129 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
1130 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
1131 _cleanup_free_
void *root_hash
= NULL
;
1132 MountEntry
*m
, *mounts
= NULL
;
1133 size_t root_hash_size
= 0;
1137 bool require_prefix
= false;
1142 if (mount_flags
== 0)
1143 mount_flags
= MS_SHARED
;
1146 dissect_image_flags
|= DISSECT_IMAGE_REQUIRE_ROOT
;
1148 if (protect_system
== PROTECT_SYSTEM_STRICT
&&
1149 protect_home
!= PROTECT_HOME_NO
&&
1150 strv_isempty(read_write_paths
))
1151 dissect_image_flags
|= DISSECT_IMAGE_READ_ONLY
;
1153 r
= loop_device_make_by_path(root_image
,
1154 dissect_image_flags
& DISSECT_IMAGE_READ_ONLY
? O_RDONLY
: O_RDWR
,
1159 r
= root_hash_load(root_image
, &root_hash
, &root_hash_size
);
1163 r
= dissect_image(loop_device
->fd
, root_hash
, root_hash_size
, dissect_image_flags
, &dissected_image
);
1167 r
= dissected_image_decrypt(dissected_image
, NULL
, root_hash
, root_hash_size
, dissect_image_flags
, &decrypted_image
);
1173 root
= root_directory
;
1174 else if (root_image
|| n_bind_mounts
> 0 || n_temporary_filesystems
> 0) {
1176 /* If we are booting from an image, create a mount point for the image, if it's still missing. We use
1177 * the same mount point for all images, which is safe, since they all live in their own namespaces
1178 * after all, and hence won't see each other. We also use such a root directory whenever there are bind
1179 * mounts configured, so that their source mounts are never obstructed by mounts we already applied
1180 * while we are applying them. */
1182 root
= "/run/systemd/unit-root";
1183 (void) mkdir_label(root
, 0700);
1184 require_prefix
= true;
1188 n_mounts
= namespace_calculate_mounts(
1196 n_temporary_filesystems
,
1197 tmp_dir
, var_tmp_dir
,
1198 protect_home
, protect_system
);
1200 /* Set mount slave mode */
1201 make_slave
= root
|| n_mounts
> 0 || ns_info
->private_mounts
;
1204 m
= mounts
= (MountEntry
*) alloca0(n_mounts
* sizeof(MountEntry
));
1205 r
= append_access_mounts(&m
, read_write_paths
, READWRITE
, require_prefix
);
1209 r
= append_access_mounts(&m
, read_only_paths
, READONLY
, require_prefix
);
1213 r
= append_access_mounts(&m
, inaccessible_paths
, INACCESSIBLE
, require_prefix
);
1217 r
= append_empty_dir_mounts(&m
, empty_directories
);
1221 r
= append_bind_mounts(&m
, bind_mounts
, n_bind_mounts
);
1225 r
= append_tmpfs_mounts(&m
, temporary_filesystems
, n_temporary_filesystems
);
1230 *(m
++) = (MountEntry
) {
1231 .path_const
= "/tmp",
1232 .mode
= PRIVATE_TMP
,
1233 .source_const
= tmp_dir
,
1238 *(m
++) = (MountEntry
) {
1239 .path_const
= "/var/tmp",
1240 .mode
= PRIVATE_TMP
,
1241 .source_const
= var_tmp_dir
,
1245 if (ns_info
->private_dev
) {
1246 *(m
++) = (MountEntry
) {
1247 .path_const
= "/dev",
1248 .mode
= PRIVATE_DEV
,
1252 if (ns_info
->protect_kernel_tunables
) {
1253 r
= append_static_mounts(&m
, protect_kernel_tunables_table
, ELEMENTSOF(protect_kernel_tunables_table
), ns_info
->ignore_protect_paths
);
1258 if (ns_info
->protect_kernel_modules
) {
1259 r
= append_static_mounts(&m
, protect_kernel_modules_table
, ELEMENTSOF(protect_kernel_modules_table
), ns_info
->ignore_protect_paths
);
1264 if (ns_info
->protect_control_groups
) {
1265 *(m
++) = (MountEntry
) {
1266 .path_const
= "/sys/fs/cgroup",
1271 r
= append_protect_home(&m
, protect_home
, ns_info
->ignore_protect_paths
);
1275 r
= append_protect_system(&m
, protect_system
, false);
1279 if (namespace_info_mount_apivfs(root
, ns_info
)) {
1280 r
= append_static_mounts(&m
, apivfs_table
, ELEMENTSOF(apivfs_table
), ns_info
->ignore_protect_paths
);
1285 assert(mounts
+ n_mounts
== m
);
1287 /* Prepend the root directory where that's necessary */
1288 r
= prefix_where_needed(mounts
, n_mounts
, root
);
1292 normalize_mounts(root_directory
, mounts
, &n_mounts
);
1295 if (unshare(CLONE_NEWNS
) < 0) {
1301 /* Remount / as SLAVE so that nothing now mounted in the namespace
1302 shows up in the parent */
1303 if (mount(NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
) < 0) {
1310 /* A root image is specified, mount it to the right place */
1311 r
= dissected_image_mount(dissected_image
, root
, UID_INVALID
, dissect_image_flags
);
1315 if (decrypted_image
) {
1316 r
= decrypted_image_relinquish(decrypted_image
);
1321 loop_device_relinquish(loop_device
);
1323 } else if (root_directory
) {
1325 /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
1326 r
= path_is_mount_point(root
, NULL
, AT_SYMLINK_FOLLOW
);
1330 if (mount(root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0) {
1338 /* Let's mount the main root directory to the root directory to use */
1339 if (mount("/", root
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0) {
1345 /* Try to set up the new root directory before mounting anything else there. */
1346 if (root_image
|| root_directory
)
1347 (void) base_filesystem_create(root
, UID_INVALID
, GID_INVALID
);
1350 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
1354 /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of /proc.
1355 * For example, this is the case with the option: 'InaccessiblePaths=/proc' */
1356 proc_self_mountinfo
= fopen("/proc/self/mountinfo", "re");
1357 if (!proc_self_mountinfo
) {
1362 /* First round, establish all mounts we need */
1366 for (m
= mounts
; m
< mounts
+ n_mounts
; ++m
) {
1371 r
= follow_symlink(root
, m
);
1375 /* We hit a symlinked mount point. The entry got rewritten and might point to a
1376 * very different place now. Let's normalize the changed list, and start from
1377 * the beginning. After all to mount the entry at the new location we might
1378 * need some other mounts first */
1383 r
= apply_mount(root
, m
);
1393 normalize_mounts(root_directory
, mounts
, &n_mounts
);
1396 /* Create a blacklist we can pass to bind_mount_recursive() */
1397 blacklist
= newa(char*, n_mounts
+1);
1398 for (j
= 0; j
< n_mounts
; j
++)
1399 blacklist
[j
] = (char*) mount_entry_path(mounts
+j
);
1400 blacklist
[j
] = NULL
;
1402 /* Second round, flip the ro bits if necessary. */
1403 for (m
= mounts
; m
< mounts
+ n_mounts
; ++m
) {
1404 r
= make_read_only(m
, blacklist
, proc_self_mountinfo
);
1411 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
1412 r
= mount_move_root(root
);
1417 /* Remount / as the desired mode. Note that this will not
1418 * reestablish propagation from our side to the host, since
1419 * what's disconnected is disconnected. */
1420 if (mount(NULL
, "/", NULL
, mount_flags
| MS_REC
, NULL
) < 0) {
1428 for (m
= mounts
; m
< mounts
+ n_mounts
; m
++)
1429 mount_entry_done(m
);
1434 void bind_mount_free_many(BindMount
*b
, size_t n
) {
1437 assert(b
|| n
== 0);
1439 for (i
= 0; i
< n
; i
++) {
1441 free(b
[i
].destination
);
1447 int bind_mount_add(BindMount
**b
, size_t *n
, const BindMount
*item
) {
1448 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
1455 s
= strdup(item
->source
);
1459 d
= strdup(item
->destination
);
1463 c
= reallocarray(*b
, *n
+ 1, sizeof(BindMount
));
1469 c
[(*n
) ++] = (BindMount
) {
1470 .source
= TAKE_PTR(s
),
1471 .destination
= TAKE_PTR(d
),
1472 .read_only
= item
->read_only
,
1473 .recursive
= item
->recursive
,
1474 .ignore_enoent
= item
->ignore_enoent
,
1480 void temporary_filesystem_free_many(TemporaryFileSystem
*t
, size_t n
) {
1483 assert(t
|| n
== 0);
1485 for (i
= 0; i
< n
; i
++) {
1493 int temporary_filesystem_add(
1494 TemporaryFileSystem
**t
,
1497 const char *options
) {
1499 _cleanup_free_
char *p
= NULL
, *o
= NULL
;
1500 TemporaryFileSystem
*c
;
1510 if (!isempty(options
)) {
1511 o
= strdup(options
);
1516 c
= reallocarray(*t
, *n
+ 1, sizeof(TemporaryFileSystem
));
1522 c
[(*n
) ++] = (TemporaryFileSystem
) {
1523 .path
= TAKE_PTR(p
),
1524 .options
= TAKE_PTR(o
),
1530 static int setup_one_tmp_dir(const char *id
, const char *prefix
, char **path
) {
1531 _cleanup_free_
char *x
= NULL
;
1532 char bid
[SD_ID128_STRING_MAX
];
1540 /* We include the boot id in the directory so that after a
1541 * reboot we can easily identify obsolete directories. */
1543 r
= sd_id128_get_boot(&boot_id
);
1547 x
= strjoin(prefix
, "/systemd-private-", sd_id128_to_string(boot_id
, bid
), "-", id
, "-XXXXXX");
1551 RUN_WITH_UMASK(0077)
1555 RUN_WITH_UMASK(0000) {
1558 y
= strjoina(x
, "/tmp");
1560 if (mkdir(y
, 0777 | S_ISVTX
) < 0)
1564 *path
= TAKE_PTR(x
);
1569 int setup_tmp_dirs(const char *id
, char **tmp_dir
, char **var_tmp_dir
) {
1575 assert(var_tmp_dir
);
1577 r
= setup_one_tmp_dir(id
, "/tmp", &a
);
1581 r
= setup_one_tmp_dir(id
, "/var/tmp", &b
);
1585 t
= strjoina(a
, "/tmp");
1599 int setup_netns(int netns_storage_socket
[2]) {
1600 _cleanup_close_
int netns
= -1;
1603 assert(netns_storage_socket
);
1604 assert(netns_storage_socket
[0] >= 0);
1605 assert(netns_storage_socket
[1] >= 0);
1607 /* We use the passed socketpair as a storage buffer for our
1608 * namespace reference fd. Whatever process runs this first
1609 * shall create a new namespace, all others should just join
1610 * it. To serialize that we use a file lock on the socket
1613 * It's a bit crazy, but hey, works great! */
1615 if (lockf(netns_storage_socket
[0], F_LOCK
, 0) < 0)
1618 netns
= receive_one_fd(netns_storage_socket
[0], MSG_DONTWAIT
);
1619 if (netns
== -EAGAIN
) {
1620 /* Nothing stored yet, so let's create a new namespace */
1622 if (unshare(CLONE_NEWNET
) < 0) {
1629 netns
= open("/proc/self/ns/net", O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
1637 } else if (netns
< 0) {
1642 /* Yay, found something, so let's join the namespace */
1643 if (setns(netns
, CLONE_NEWNET
) < 0) {
1651 q
= send_one_fd(netns_storage_socket
[1], netns
, MSG_DONTWAIT
);
1658 (void) lockf(netns_storage_socket
[0], F_ULOCK
, 0);
1662 bool ns_type_supported(NamespaceType type
) {
1663 const char *t
, *ns_proc
;
1665 t
= namespace_type_to_string(type
);
1666 if (!t
) /* Don't know how to translate this? Then it's not supported */
1669 ns_proc
= strjoina("/proc/self/ns/", t
);
1670 return access(ns_proc
, F_OK
) == 0;
1673 static const char *const protect_home_table
[_PROTECT_HOME_MAX
] = {
1674 [PROTECT_HOME_NO
] = "no",
1675 [PROTECT_HOME_YES
] = "yes",
1676 [PROTECT_HOME_READ_ONLY
] = "read-only",
1677 [PROTECT_HOME_TMPFS
] = "tmpfs",
1680 DEFINE_STRING_TABLE_LOOKUP(protect_home
, ProtectHome
);
1682 ProtectHome
protect_home_or_bool_from_string(const char *s
) {
1685 r
= parse_boolean(s
);
1687 return PROTECT_HOME_YES
;
1689 return PROTECT_HOME_NO
;
1691 return protect_home_from_string(s
);
1694 static const char *const protect_system_table
[_PROTECT_SYSTEM_MAX
] = {
1695 [PROTECT_SYSTEM_NO
] = "no",
1696 [PROTECT_SYSTEM_YES
] = "yes",
1697 [PROTECT_SYSTEM_FULL
] = "full",
1698 [PROTECT_SYSTEM_STRICT
] = "strict",
1701 DEFINE_STRING_TABLE_LOOKUP(protect_system
, ProtectSystem
);
1703 ProtectSystem
protect_system_or_bool_from_string(const char *s
) {
1706 r
= parse_boolean(s
);
1708 return PROTECT_SYSTEM_YES
;
1710 return PROTECT_SYSTEM_NO
;
1712 return protect_system_from_string(s
);
1715 static const char* const namespace_type_table
[] = {
1716 [NAMESPACE_MOUNT
] = "mnt",
1717 [NAMESPACE_CGROUP
] = "cgroup",
1718 [NAMESPACE_UTS
] = "uts",
1719 [NAMESPACE_IPC
] = "ipc",
1720 [NAMESPACE_USER
] = "user",
1721 [NAMESPACE_PID
] = "pid",
1722 [NAMESPACE_NET
] = "net",
1725 DEFINE_STRING_TABLE_LOOKUP(namespace_type
, NamespaceType
);