1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2010 Lennart Poettering
12 #include <sys/mount.h>
17 #include "alloc-util.h"
18 #include "base-filesystem.h"
19 #include "dev-setup.h"
23 #include "loop-util.h"
24 #include "loopback-setup.h"
27 #include "mount-util.h"
28 #include "namespace.h"
29 #include "path-util.h"
30 #include "selinux-util.h"
31 #include "socket-util.h"
32 #include "stat-util.h"
33 #include "string-table.h"
34 #include "string-util.h"
36 #include "umask-util.h"
37 #include "user-util.h"
40 #define DEV_MOUNT_OPTIONS (MS_NOSUID|MS_STRICTATIME|MS_NOEXEC)
42 typedef enum MountMode
{
43 /* This is ordered by priority! */
58 typedef struct MountEntry
{
59 const char *path_const
; /* Memory allocated on stack or static */
61 bool ignore
:1; /* Ignore if path does not exist? */
62 bool has_prefix
:1; /* Already is prefixed by the root dir? */
63 bool read_only
:1; /* Shall this mount point be read-only? */
64 bool applied
:1; /* Already applied */
65 char *path_malloc
; /* Use this instead of 'path_const' if we had to allocate memory */
66 const char *source_const
; /* The source path, for bind mounts */
68 const char *options_const
;/* Mount options for tmpfs */
70 unsigned long flags
; /* Mount flags used by EMPTY_DIR and TMPFS. Do not include MS_RDONLY here, but please use read_only. */
74 /* If MountAPIVFS= is used, let's mount /sys and /proc into the it, but only as a fallback if the user hasn't mounted
75 * something there already. These mounts are hence overriden by any other explicitly configured mounts. */
76 static const MountEntry apivfs_table
[] = {
77 { "/proc", PROCFS
, false },
78 { "/dev", BIND_DEV
, false },
79 { "/sys", SYSFS
, false },
82 /* ProtectKernelTunables= option and the related filesystem APIs */
83 static const MountEntry protect_kernel_tunables_table
[] = {
84 { "/proc/acpi", READONLY
, true },
85 { "/proc/apm", READONLY
, true }, /* Obsolete API, there's no point in permitting access to this, ever */
86 { "/proc/asound", READONLY
, true },
87 { "/proc/bus", READONLY
, true },
88 { "/proc/fs", READONLY
, true },
89 { "/proc/irq", READONLY
, true },
90 { "/proc/kallsyms", INACCESSIBLE
, true },
91 { "/proc/kcore", INACCESSIBLE
, true },
92 { "/proc/latency_stats", READONLY
, true },
93 { "/proc/mtrr", READONLY
, true },
94 { "/proc/scsi", READONLY
, true },
95 { "/proc/sys", READONLY
, false },
96 { "/proc/sysrq-trigger", READONLY
, true },
97 { "/proc/timer_stats", READONLY
, true },
98 { "/sys", READONLY
, false },
99 { "/sys/fs/bpf", READONLY
, true },
100 { "/sys/fs/cgroup", READWRITE
, false }, /* READONLY is set by ProtectControlGroups= option */
101 { "/sys/fs/selinux", READWRITE
, true },
102 { "/sys/kernel/debug", READONLY
, true },
103 { "/sys/kernel/tracing", READONLY
, true },
106 /* ProtectKernelModules= option */
107 static const MountEntry protect_kernel_modules_table
[] = {
109 { "/lib/modules", INACCESSIBLE
, true },
111 { "/usr/lib/modules", INACCESSIBLE
, true },
115 * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of
116 * system should be protected by ProtectSystem=
118 static const MountEntry protect_home_read_only_table
[] = {
119 { "/home", READONLY
, true },
120 { "/run/user", READONLY
, true },
121 { "/root", READONLY
, true },
124 /* ProtectHome=tmpfs table */
125 static const MountEntry protect_home_tmpfs_table
[] = {
126 { "/home", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755", .flags
= MS_NODEV
|MS_STRICTATIME
},
127 { "/run/user", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755", .flags
= MS_NODEV
|MS_STRICTATIME
},
128 { "/root", TMPFS
, true, .read_only
= true, .options_const
= "mode=0700", .flags
= MS_NODEV
|MS_STRICTATIME
},
131 /* ProtectHome=yes table */
132 static const MountEntry protect_home_yes_table
[] = {
133 { "/home", INACCESSIBLE
, true },
134 { "/run/user", INACCESSIBLE
, true },
135 { "/root", INACCESSIBLE
, true },
138 /* ProtectSystem=yes table */
139 static const MountEntry protect_system_yes_table
[] = {
140 { "/usr", READONLY
, false },
141 { "/boot", READONLY
, true },
142 { "/efi", READONLY
, true },
144 { "/lib", READONLY
, true },
145 { "/lib64", READONLY
, true },
146 { "/bin", READONLY
, true },
148 { "/sbin", READONLY
, true },
153 /* ProtectSystem=full includes ProtectSystem=yes */
154 static const MountEntry protect_system_full_table
[] = {
155 { "/usr", READONLY
, false },
156 { "/boot", READONLY
, true },
157 { "/efi", READONLY
, true },
158 { "/etc", READONLY
, false },
160 { "/lib", READONLY
, true },
161 { "/lib64", READONLY
, true },
162 { "/bin", READONLY
, true },
164 { "/sbin", READONLY
, true },
170 * ProtectSystem=strict table. In this strict mode, we mount everything
171 * read-only, except for /proc, /dev, /sys which are the kernel API VFS,
172 * which are left writable, but PrivateDevices= + ProtectKernelTunables=
173 * protect those, and these options should be fully orthogonal.
174 * (And of course /home and friends are also left writable, as ProtectHome=
175 * shall manage those, orthogonally).
177 static const MountEntry protect_system_strict_table
[] = {
178 { "/", READONLY
, false },
179 { "/proc", READWRITE
, false }, /* ProtectKernelTunables= */
180 { "/sys", READWRITE
, false }, /* ProtectKernelTunables= */
181 { "/dev", READWRITE
, false }, /* PrivateDevices= */
182 { "/home", READWRITE
, true }, /* ProtectHome= */
183 { "/run/user", READWRITE
, true }, /* ProtectHome= */
184 { "/root", READWRITE
, true }, /* ProtectHome= */
187 static const char *mount_entry_path(const MountEntry
*p
) {
190 /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that,
191 * otherwise the stack/static ->path field is returned. */
193 return p
->path_malloc
?: p
->path_const
;
196 static bool mount_entry_read_only(const MountEntry
*p
) {
199 return p
->read_only
|| IN_SET(p
->mode
, READONLY
, INACCESSIBLE
);
202 static const char *mount_entry_source(const MountEntry
*p
) {
205 return p
->source_malloc
?: p
->source_const
;
208 static const char *mount_entry_options(const MountEntry
*p
) {
211 return p
->options_malloc
?: p
->options_const
;
214 static void mount_entry_done(MountEntry
*p
) {
217 p
->path_malloc
= mfree(p
->path_malloc
);
218 p
->source_malloc
= mfree(p
->source_malloc
);
219 p
->options_malloc
= mfree(p
->options_malloc
);
222 static int append_access_mounts(MountEntry
**p
, char **strv
, MountMode mode
, bool forcibly_require_prefix
) {
227 /* Adds a list of user-supplied READWRITE/READONLY/INACCESSIBLE entries */
229 STRV_FOREACH(i
, strv
) {
230 bool ignore
= false, needs_prefix
= false;
233 /* Look for any prefixes */
234 if (startswith(e
, "-")) {
238 if (startswith(e
, "+")) {
243 if (!path_is_absolute(e
))
246 *((*p
)++) = (MountEntry
) {
250 .has_prefix
= !needs_prefix
&& !forcibly_require_prefix
,
257 static int append_empty_dir_mounts(MountEntry
**p
, char **strv
) {
262 /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the
263 * "/private/" boundary directories for DynamicUser=1. */
265 STRV_FOREACH(i
, strv
) {
267 *((*p
)++) = (MountEntry
) {
273 .options_const
= "mode=755",
274 .flags
= MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
,
281 static int append_bind_mounts(MountEntry
**p
, const BindMount
*binds
, size_t n
) {
286 for (i
= 0; i
< n
; i
++) {
287 const BindMount
*b
= binds
+ i
;
289 *((*p
)++) = (MountEntry
) {
290 .path_const
= b
->destination
,
291 .mode
= b
->recursive
? BIND_MOUNT_RECURSIVE
: BIND_MOUNT
,
292 .read_only
= b
->read_only
,
293 .source_const
= b
->source
,
294 .ignore
= b
->ignore_enoent
,
301 static int append_tmpfs_mounts(MountEntry
**p
, const TemporaryFileSystem
*tmpfs
, size_t n
) {
307 for (i
= 0; i
< n
; i
++) {
308 const TemporaryFileSystem
*t
= tmpfs
+ i
;
309 _cleanup_free_
char *o
= NULL
, *str
= NULL
;
310 unsigned long flags
= MS_NODEV
|MS_STRICTATIME
;
313 if (!path_is_absolute(t
->path
))
316 if (!isempty(t
->options
)) {
317 str
= strjoin("mode=0755,", t
->options
);
321 r
= mount_option_mangle(str
, MS_NODEV
|MS_STRICTATIME
, &flags
, &o
);
325 ro
= !!(flags
& MS_RDONLY
);
330 *((*p
)++) = (MountEntry
) {
331 .path_const
= t
->path
,
344 static int append_static_mounts(MountEntry
**p
, const MountEntry
*mounts
, size_t n
, bool ignore_protect
) {
350 /* Adds a list of static pre-defined entries */
352 for (i
= 0; i
< n
; i
++)
353 *((*p
)++) = (MountEntry
) {
354 .path_const
= mount_entry_path(mounts
+i
),
355 .mode
= mounts
[i
].mode
,
356 .ignore
= mounts
[i
].ignore
|| ignore_protect
,
362 static int append_protect_home(MountEntry
**p
, ProtectHome protect_home
, bool ignore_protect
) {
365 switch (protect_home
) {
367 case PROTECT_HOME_NO
:
370 case PROTECT_HOME_READ_ONLY
:
371 return append_static_mounts(p
, protect_home_read_only_table
, ELEMENTSOF(protect_home_read_only_table
), ignore_protect
);
373 case PROTECT_HOME_TMPFS
:
374 return append_static_mounts(p
, protect_home_tmpfs_table
, ELEMENTSOF(protect_home_tmpfs_table
), ignore_protect
);
376 case PROTECT_HOME_YES
:
377 return append_static_mounts(p
, protect_home_yes_table
, ELEMENTSOF(protect_home_yes_table
), ignore_protect
);
380 assert_not_reached("Unexpected ProtectHome= value");
384 static int append_protect_system(MountEntry
**p
, ProtectSystem protect_system
, bool ignore_protect
) {
387 switch (protect_system
) {
389 case PROTECT_SYSTEM_NO
:
392 case PROTECT_SYSTEM_STRICT
:
393 return append_static_mounts(p
, protect_system_strict_table
, ELEMENTSOF(protect_system_strict_table
), ignore_protect
);
395 case PROTECT_SYSTEM_YES
:
396 return append_static_mounts(p
, protect_system_yes_table
, ELEMENTSOF(protect_system_yes_table
), ignore_protect
);
398 case PROTECT_SYSTEM_FULL
:
399 return append_static_mounts(p
, protect_system_full_table
, ELEMENTSOF(protect_system_full_table
), ignore_protect
);
402 assert_not_reached("Unexpected ProtectSystem= value");
406 static int mount_path_compare(const void *a
, const void *b
) {
407 const MountEntry
*p
= a
, *q
= b
;
410 /* If the paths are not equal, then order prefixes first */
411 d
= path_compare(mount_entry_path(p
), mount_entry_path(q
));
415 /* If the paths are equal, check the mode */
416 if (p
->mode
< q
->mode
)
418 if (p
->mode
> q
->mode
)
424 static int prefix_where_needed(MountEntry
*m
, size_t n
, const char *root_directory
) {
427 /* Prefixes all paths in the bind mount table with the root directory if it is specified and the entry needs
433 for (i
= 0; i
< n
; i
++) {
439 s
= prefix_root(root_directory
, mount_entry_path(m
+i
));
443 free_and_replace(m
[i
].path_malloc
, s
);
444 m
[i
].has_prefix
= true;
450 static void drop_duplicates(MountEntry
*m
, size_t *n
) {
451 MountEntry
*f
, *t
, *previous
;
456 /* Drops duplicate entries. Expects that the array is properly ordered already. */
458 for (f
= m
, t
= m
, previous
= NULL
; f
< m
+ *n
; f
++) {
460 /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare()
461 * above. Note that we only drop duplicates that haven't been mounted yet. */
463 path_equal(mount_entry_path(f
), mount_entry_path(previous
)) &&
464 !f
->applied
&& !previous
->applied
) {
465 log_debug("%s is duplicate.", mount_entry_path(f
));
466 previous
->read_only
= previous
->read_only
|| mount_entry_read_only(f
); /* Propagate the read-only flag to the remaining entry */
479 static void drop_inaccessible(MountEntry
*m
, size_t *n
) {
481 const char *clear
= NULL
;
486 /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly
487 * ordered already. */
489 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
491 /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop
492 * it, as inaccessible paths really should drop the entire subtree. */
493 if (clear
&& path_startswith(mount_entry_path(f
), clear
)) {
494 log_debug("%s is masked by %s.", mount_entry_path(f
), clear
);
499 clear
= f
->mode
== INACCESSIBLE
? mount_entry_path(f
) : NULL
;
508 static void drop_nop(MountEntry
*m
, size_t *n
) {
514 /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the
515 * list is ordered by prefixes. */
517 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
519 /* Only suppress such subtrees for READONLY and READWRITE entries */
520 if (IN_SET(f
->mode
, READONLY
, READWRITE
)) {
524 /* Now let's find the first parent of the entry we are looking at. */
525 for (p
= t
-1; p
>= m
; p
--) {
526 if (path_startswith(mount_entry_path(f
), mount_entry_path(p
))) {
532 /* We found it, let's see if it's the same mode, if so, we can drop this entry */
533 if (found
&& p
->mode
== f
->mode
) {
534 log_debug("%s is redundant by %s", mount_entry_path(f
), mount_entry_path(p
));
547 static void drop_outside_root(const char *root_directory
, MountEntry
*m
, size_t *n
) {
557 /* Drops all mounts that are outside of the root directory. */
559 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
561 if (!path_startswith(mount_entry_path(f
), root_directory
)) {
562 log_debug("%s is outside of root directory.", mount_entry_path(f
));
574 static int clone_device_node(const char *d
, const char *temporary_mount
, bool *make_devnode
) {
579 if (stat(d
, &st
) < 0) {
585 if (!S_ISBLK(st
.st_mode
) &&
586 !S_ISCHR(st
.st_mode
))
592 dn
= strjoina(temporary_mount
, d
);
595 mac_selinux_create_file_prepare(d
, st
.st_mode
);
596 r
= mknod(dn
, st
.st_mode
, st
.st_rdev
);
597 mac_selinux_create_file_clear();
602 return log_debug_errno(errno
, "mknod failed for %s: %m", d
);
604 *make_devnode
= false;
607 /* We're about to fallback to bind-mounting the device
608 * node. So create a dummy bind-mount target. */
609 mac_selinux_create_file_prepare(d
, 0);
610 r
= mknod(dn
, S_IFREG
, 0);
611 mac_selinux_create_file_clear();
613 if (r
< 0 && errno
!= EEXIST
)
614 return log_debug_errno(errno
, "mknod fallback failed for %s: %m", d
);
616 /* Fallback to bind-mounting:
617 * The assumption here is that all used device nodes carry standard
618 * properties. Specifically, the devices nodes we bind-mount should
619 * either be owned by root:root or root:tty (e.g. /dev/tty, /dev/ptmx)
620 * and should not carry ACLs. */
621 if (mount(d
, dn
, NULL
, MS_BIND
, NULL
) < 0)
622 return log_debug_errno(errno
, "mount failed for %s: %m", d
);
627 static int mount_private_dev(MountEntry
*m
) {
628 static const char devnodes
[] =
636 char temporary_mount
[] = "/tmp/namespace-dev-XXXXXX";
637 const char *d
, *dev
= NULL
, *devpts
= NULL
, *devshm
= NULL
, *devhugepages
= NULL
, *devmqueue
= NULL
, *devlog
= NULL
, *devptmx
= NULL
;
638 bool can_mknod
= true;
639 _cleanup_umask_ mode_t u
;
646 if (!mkdtemp(temporary_mount
))
649 dev
= strjoina(temporary_mount
, "/dev");
650 (void) mkdir(dev
, 0755);
651 if (mount("tmpfs", dev
, "tmpfs", DEV_MOUNT_OPTIONS
, "mode=755") < 0) {
656 devpts
= strjoina(temporary_mount
, "/dev/pts");
657 (void) mkdir(devpts
, 0755);
658 if (mount("/dev/pts", devpts
, NULL
, MS_BIND
, NULL
) < 0) {
663 /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx
664 * when /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible
665 * thus, in that case make a clone
667 * in nspawn and other containers it will be a symlink, in that case make it a symlink
669 r
= is_symlink("/dev/ptmx");
673 devptmx
= strjoina(temporary_mount
, "/dev/ptmx");
674 if (symlink("pts/ptmx", devptmx
) < 0) {
679 r
= clone_device_node("/dev/ptmx", temporary_mount
, &can_mknod
);
684 devshm
= strjoina(temporary_mount
, "/dev/shm");
685 (void) mkdir(devshm
, 0755);
686 r
= mount("/dev/shm", devshm
, NULL
, MS_BIND
, NULL
);
692 devmqueue
= strjoina(temporary_mount
, "/dev/mqueue");
693 (void) mkdir(devmqueue
, 0755);
694 (void) mount("/dev/mqueue", devmqueue
, NULL
, MS_BIND
, NULL
);
696 devhugepages
= strjoina(temporary_mount
, "/dev/hugepages");
697 (void) mkdir(devhugepages
, 0755);
698 (void) mount("/dev/hugepages", devhugepages
, NULL
, MS_BIND
, NULL
);
700 devlog
= strjoina(temporary_mount
, "/dev/log");
701 (void) symlink("/run/systemd/journal/dev-log", devlog
);
703 NULSTR_FOREACH(d
, devnodes
) {
704 r
= clone_device_node(d
, temporary_mount
, &can_mknod
);
705 /* ENXIO means the the *source* is not a device file, skip creation in that case */
706 if (r
< 0 && r
!= -ENXIO
)
710 dev_setup(temporary_mount
, UID_INVALID
, GID_INVALID
);
712 /* Create the /dev directory if missing. It is more likely to be
713 * missing when the service is started with RootDirectory. This is
714 * consistent with mount units creating the mount points when missing.
716 (void) mkdir_p_label(mount_entry_path(m
), 0755);
718 /* Unmount everything in old /dev */
719 umount_recursive(mount_entry_path(m
), 0);
720 if (mount(dev
, mount_entry_path(m
), NULL
, MS_MOVE
, NULL
) < 0) {
726 rmdir(temporary_mount
);
738 umount(devhugepages
);
745 rmdir(temporary_mount
);
750 static int mount_bind_dev(const MountEntry
*m
) {
755 /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the service's
756 * /dev. This is only used when RootDirectory= is set. */
758 (void) mkdir_p_label(mount_entry_path(m
), 0755);
760 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
762 return log_debug_errno(r
, "Unable to determine whether /dev is already mounted: %m");
763 if (r
> 0) /* make this a NOP if /dev is already a mount point */
766 if (mount("/dev", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
767 return log_debug_errno(errno
, "Failed to bind mount %s: %m", mount_entry_path(m
));
772 static int mount_sysfs(const MountEntry
*m
) {
777 (void) mkdir_p_label(mount_entry_path(m
), 0755);
779 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
781 return log_debug_errno(r
, "Unable to determine whether /sys is already mounted: %m");
782 if (r
> 0) /* make this a NOP if /sys is already a mount point */
785 /* Bind mount the host's version so that we get all child mounts of it, too. */
786 if (mount("/sys", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
787 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
792 static int mount_procfs(const MountEntry
*m
) {
797 (void) mkdir_p_label(mount_entry_path(m
), 0755);
799 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
801 return log_debug_errno(r
, "Unable to determine whether /proc is already mounted: %m");
802 if (r
> 0) /* make this a NOP if /proc is already a mount point */
805 /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in one */
806 if (mount("proc", mount_entry_path(m
), "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
) < 0)
807 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
812 static int mount_tmpfs(const MountEntry
*m
) {
815 /* First, get rid of everything that is below if there is anything. Then, overmount with our new tmpfs */
817 (void) mkdir_p_label(mount_entry_path(m
), 0755);
818 (void) umount_recursive(mount_entry_path(m
), 0);
820 if (mount("tmpfs", mount_entry_path(m
), "tmpfs", m
->flags
, mount_entry_options(m
)) < 0)
821 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
826 static int follow_symlink(
827 const char *root_directory
,
830 _cleanup_free_
char *target
= NULL
;
833 /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we
834 * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at
835 * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the
836 * end and already have a fully normalized name. */
838 r
= chase_symlinks(mount_entry_path(m
), root_directory
, CHASE_STEP
|CHASE_NONEXISTENT
, &target
);
840 return log_debug_errno(r
, "Failed to chase symlinks '%s': %m", mount_entry_path(m
));
841 if (r
> 0) /* Reached the end, nothing more to resolve */
844 if (m
->n_followed
>= CHASE_SYMLINKS_MAX
) { /* put a boundary on things */
845 log_debug("Symlink loop on '%s'.", mount_entry_path(m
));
849 log_debug("Followed mount entry path symlink %s → %s.", mount_entry_path(m
), target
);
851 free_and_replace(m
->path_malloc
, target
);
852 m
->has_prefix
= true;
859 static int apply_mount(
860 const char *root_directory
,
863 bool rbind
= true, make
= false;
869 log_debug("Applying namespace mount on %s", mount_entry_path(m
));
876 /* First, get rid of everything that is below if there
877 * is anything... Then, overmount it with an
878 * inaccessible path. */
879 (void) umount_recursive(mount_entry_path(m
), 0);
881 if (lstat(mount_entry_path(m
), &target
) < 0) {
882 if (errno
== ENOENT
&& m
->ignore
)
885 return log_debug_errno(errno
, "Failed to lstat() %s to determine what to mount over it: %m", mount_entry_path(m
));
888 what
= mode_to_inaccessible_node(target
.st_mode
);
890 log_debug("File type not supported for inaccessible mounts. Note that symlinks are not allowed");
898 r
= path_is_mount_point(mount_entry_path(m
), root_directory
, 0);
899 if (r
== -ENOENT
&& m
->ignore
)
902 return log_debug_errno(r
, "Failed to determine whether %s is already a mount point: %m", mount_entry_path(m
));
903 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. */
905 /* This isn't a mount point yet, let's make it one. */
906 what
= mount_entry_path(m
);
913 case BIND_MOUNT_RECURSIVE
: {
914 _cleanup_free_
char *chased
= NULL
;
916 /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note that bind
917 * mount source paths are always relative to the host root, hence we pass NULL as root directory to
918 * chase_symlinks() here. */
920 r
= chase_symlinks(mount_entry_source(m
), NULL
, CHASE_TRAIL_SLASH
, &chased
);
921 if (r
== -ENOENT
&& m
->ignore
) {
922 log_debug_errno(r
, "Path %s does not exist, ignoring.", mount_entry_source(m
));
926 return log_debug_errno(r
, "Failed to follow symlinks on %s: %m", mount_entry_source(m
));
928 log_debug("Followed source symlinks %s → %s.", mount_entry_source(m
), chased
);
930 free_and_replace(m
->source_malloc
, chased
);
932 what
= mount_entry_source(m
);
939 return mount_tmpfs(m
);
942 what
= mount_entry_source(m
);
947 return mount_private_dev(m
);
950 return mount_bind_dev(m
);
953 return mount_sysfs(m
);
956 return mount_procfs(m
);
959 assert_not_reached("Unknown mode");
964 if (mount(what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
) < 0) {
965 bool try_again
= false;
968 if (r
== -ENOENT
&& make
) {
971 /* Hmm, either the source or the destination are missing. Let's see if we can create the destination, then try again */
973 if (stat(what
, &st
) < 0)
974 log_debug_errno(errno
, "Mount point source '%s' is not accessible: %m", what
);
978 (void) mkdir_parents(mount_entry_path(m
), 0755);
980 if (S_ISDIR(st
.st_mode
))
981 q
= mkdir(mount_entry_path(m
), 0755) < 0 ? -errno
: 0;
983 q
= touch(mount_entry_path(m
));
986 log_debug_errno(q
, "Failed to create destination mount point node '%s': %m", mount_entry_path(m
));
993 if (mount(what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
) < 0)
1000 return log_debug_errno(r
, "Failed to mount %s to %s: %m", what
, mount_entry_path(m
));
1003 log_debug("Successfully mounted %s to %s", what
, mount_entry_path(m
));
1007 static int make_read_only(const MountEntry
*m
, char **blacklist
, FILE *proc_self_mountinfo
) {
1011 assert(proc_self_mountinfo
);
1013 if (mount_entry_read_only(m
)) {
1014 if (IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
)) {
1015 /* Make superblock readonly */
1016 if (mount(NULL
, mount_entry_path(m
), NULL
, MS_REMOUNT
| MS_RDONLY
| m
->flags
, mount_entry_options(m
)) < 0)
1019 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), true, blacklist
, proc_self_mountinfo
);
1020 } else if (m
->mode
== PRIVATE_DEV
) {
1021 /* Superblock can be readonly but the submounts can't */
1022 if (mount(NULL
, mount_entry_path(m
), NULL
, MS_REMOUNT
|DEV_MOUNT_OPTIONS
|MS_RDONLY
, NULL
) < 0)
1027 /* Not that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked read-only
1028 * already stays this way. This improves compatibility with container managers, where we won't attempt to undo
1029 * read-only mounts already applied. */
1031 if (r
== -ENOENT
&& m
->ignore
)
1037 static bool namespace_info_mount_apivfs(const char *root_directory
, const NamespaceInfo
*ns_info
) {
1041 * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
1042 * since to protect the API VFS mounts, they need to be around in the
1043 * first place... and RootDirectory= or RootImage= need to be set.
1046 /* root_directory should point to a mount point */
1047 return root_directory
&&
1048 (ns_info
->mount_apivfs
||
1049 ns_info
->protect_control_groups
||
1050 ns_info
->protect_kernel_tunables
);
1053 static size_t namespace_calculate_mounts(
1054 const char* root_directory
,
1055 const NamespaceInfo
*ns_info
,
1056 char** read_write_paths
,
1057 char** read_only_paths
,
1058 char** inaccessible_paths
,
1059 char** empty_directories
,
1060 size_t n_bind_mounts
,
1061 size_t n_temporary_filesystems
,
1062 const char* tmp_dir
,
1063 const char* var_tmp_dir
,
1064 ProtectHome protect_home
,
1065 ProtectSystem protect_system
) {
1067 size_t protect_home_cnt
;
1068 size_t protect_system_cnt
=
1069 (protect_system
== PROTECT_SYSTEM_STRICT
?
1070 ELEMENTSOF(protect_system_strict_table
) :
1071 ((protect_system
== PROTECT_SYSTEM_FULL
) ?
1072 ELEMENTSOF(protect_system_full_table
) :
1073 ((protect_system
== PROTECT_SYSTEM_YES
) ?
1074 ELEMENTSOF(protect_system_yes_table
) : 0)));
1077 (protect_home
== PROTECT_HOME_YES
?
1078 ELEMENTSOF(protect_home_yes_table
) :
1079 ((protect_home
== PROTECT_HOME_READ_ONLY
) ?
1080 ELEMENTSOF(protect_home_read_only_table
) :
1081 ((protect_home
== PROTECT_HOME_TMPFS
) ?
1082 ELEMENTSOF(protect_home_tmpfs_table
) : 0)));
1084 return !!tmp_dir
+ !!var_tmp_dir
+
1085 strv_length(read_write_paths
) +
1086 strv_length(read_only_paths
) +
1087 strv_length(inaccessible_paths
) +
1088 strv_length(empty_directories
) +
1090 n_temporary_filesystems
+
1091 ns_info
->private_dev
+
1092 (ns_info
->protect_kernel_tunables
? ELEMENTSOF(protect_kernel_tunables_table
) : 0) +
1093 (ns_info
->protect_control_groups
? 1 : 0) +
1094 (ns_info
->protect_kernel_modules
? ELEMENTSOF(protect_kernel_modules_table
) : 0) +
1095 protect_home_cnt
+ protect_system_cnt
+
1096 (namespace_info_mount_apivfs(root_directory
, ns_info
) ? ELEMENTSOF(apivfs_table
) : 0);
1099 static void normalize_mounts(const char *root_directory
, MountEntry
*mounts
, size_t *n_mounts
) {
1101 assert(mounts
|| *n_mounts
== 0);
1103 qsort_safe(mounts
, *n_mounts
, sizeof(MountEntry
), mount_path_compare
);
1105 drop_duplicates(mounts
, n_mounts
);
1106 drop_outside_root(root_directory
, mounts
, n_mounts
);
1107 drop_inaccessible(mounts
, n_mounts
);
1108 drop_nop(mounts
, n_mounts
);
1111 int setup_namespace(
1112 const char* root_directory
,
1113 const char* root_image
,
1114 const NamespaceInfo
*ns_info
,
1115 char** read_write_paths
,
1116 char** read_only_paths
,
1117 char** inaccessible_paths
,
1118 char** empty_directories
,
1119 const BindMount
*bind_mounts
,
1120 size_t n_bind_mounts
,
1121 const TemporaryFileSystem
*temporary_filesystems
,
1122 size_t n_temporary_filesystems
,
1123 const char* tmp_dir
,
1124 const char* var_tmp_dir
,
1125 ProtectHome protect_home
,
1126 ProtectSystem protect_system
,
1127 unsigned long mount_flags
,
1128 DissectImageFlags dissect_image_flags
) {
1130 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
1131 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
1132 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
1133 _cleanup_free_
void *root_hash
= NULL
;
1134 MountEntry
*m
, *mounts
= NULL
;
1135 size_t root_hash_size
= 0;
1136 bool make_slave
= false;
1139 bool require_prefix
= false;
1144 if (mount_flags
== 0)
1145 mount_flags
= MS_SHARED
;
1148 dissect_image_flags
|= DISSECT_IMAGE_REQUIRE_ROOT
;
1150 if (protect_system
== PROTECT_SYSTEM_STRICT
&&
1151 protect_home
!= PROTECT_HOME_NO
&&
1152 strv_isempty(read_write_paths
))
1153 dissect_image_flags
|= DISSECT_IMAGE_READ_ONLY
;
1155 r
= loop_device_make_by_path(root_image
,
1156 dissect_image_flags
& DISSECT_IMAGE_READ_ONLY
? O_RDONLY
: O_RDWR
,
1161 r
= root_hash_load(root_image
, &root_hash
, &root_hash_size
);
1165 r
= dissect_image(loop_device
->fd
, root_hash
, root_hash_size
, dissect_image_flags
, &dissected_image
);
1169 r
= dissected_image_decrypt(dissected_image
, NULL
, root_hash
, root_hash_size
, dissect_image_flags
, &decrypted_image
);
1175 root
= root_directory
;
1176 else if (root_image
|| n_bind_mounts
> 0 || n_temporary_filesystems
> 0) {
1178 /* If we are booting from an image, create a mount point for the image, if it's still missing. We use
1179 * the same mount point for all images, which is safe, since they all live in their own namespaces
1180 * after all, and hence won't see each other. We also use such a root directory whenever there are bind
1181 * mounts configured, so that their source mounts are never obstructed by mounts we already applied
1182 * while we are applying them. */
1184 root
= "/run/systemd/unit-root";
1185 (void) mkdir_label(root
, 0700);
1186 require_prefix
= true;
1190 n_mounts
= namespace_calculate_mounts(
1198 n_temporary_filesystems
,
1199 tmp_dir
, var_tmp_dir
,
1200 protect_home
, protect_system
);
1202 /* Set mount slave mode */
1203 if (root
|| n_mounts
> 0)
1207 m
= mounts
= (MountEntry
*) alloca0(n_mounts
* sizeof(MountEntry
));
1208 r
= append_access_mounts(&m
, read_write_paths
, READWRITE
, require_prefix
);
1212 r
= append_access_mounts(&m
, read_only_paths
, READONLY
, require_prefix
);
1216 r
= append_access_mounts(&m
, inaccessible_paths
, INACCESSIBLE
, require_prefix
);
1220 r
= append_empty_dir_mounts(&m
, empty_directories
);
1224 r
= append_bind_mounts(&m
, bind_mounts
, n_bind_mounts
);
1228 r
= append_tmpfs_mounts(&m
, temporary_filesystems
, n_temporary_filesystems
);
1233 *(m
++) = (MountEntry
) {
1234 .path_const
= "/tmp",
1235 .mode
= PRIVATE_TMP
,
1236 .source_const
= tmp_dir
,
1241 *(m
++) = (MountEntry
) {
1242 .path_const
= "/var/tmp",
1243 .mode
= PRIVATE_TMP
,
1244 .source_const
= var_tmp_dir
,
1248 if (ns_info
->private_dev
) {
1249 *(m
++) = (MountEntry
) {
1250 .path_const
= "/dev",
1251 .mode
= PRIVATE_DEV
,
1255 if (ns_info
->protect_kernel_tunables
) {
1256 r
= append_static_mounts(&m
, protect_kernel_tunables_table
, ELEMENTSOF(protect_kernel_tunables_table
), ns_info
->ignore_protect_paths
);
1261 if (ns_info
->protect_kernel_modules
) {
1262 r
= append_static_mounts(&m
, protect_kernel_modules_table
, ELEMENTSOF(protect_kernel_modules_table
), ns_info
->ignore_protect_paths
);
1267 if (ns_info
->protect_control_groups
) {
1268 *(m
++) = (MountEntry
) {
1269 .path_const
= "/sys/fs/cgroup",
1274 r
= append_protect_home(&m
, protect_home
, ns_info
->ignore_protect_paths
);
1278 r
= append_protect_system(&m
, protect_system
, false);
1282 if (namespace_info_mount_apivfs(root
, ns_info
)) {
1283 r
= append_static_mounts(&m
, apivfs_table
, ELEMENTSOF(apivfs_table
), ns_info
->ignore_protect_paths
);
1288 assert(mounts
+ n_mounts
== m
);
1290 /* Prepend the root directory where that's necessary */
1291 r
= prefix_where_needed(mounts
, n_mounts
, root
);
1295 normalize_mounts(root_directory
, mounts
, &n_mounts
);
1298 if (unshare(CLONE_NEWNS
) < 0) {
1304 /* Remount / as SLAVE so that nothing now mounted in the namespace
1305 shows up in the parent */
1306 if (mount(NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
) < 0) {
1313 /* A root image is specified, mount it to the right place */
1314 r
= dissected_image_mount(dissected_image
, root
, UID_INVALID
, dissect_image_flags
);
1318 if (decrypted_image
) {
1319 r
= decrypted_image_relinquish(decrypted_image
);
1324 loop_device_relinquish(loop_device
);
1326 } else if (root_directory
) {
1328 /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
1329 r
= path_is_mount_point(root
, NULL
, AT_SYMLINK_FOLLOW
);
1333 if (mount(root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0) {
1341 /* Let's mount the main root directory to the root directory to use */
1342 if (mount("/", root
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0) {
1348 /* Try to set up the new root directory before mounting anything else there. */
1349 if (root_image
|| root_directory
)
1350 (void) base_filesystem_create(root
, UID_INVALID
, GID_INVALID
);
1353 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
1357 /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of /proc.
1358 * For example, this is the case with the option: 'InaccessiblePaths=/proc' */
1359 proc_self_mountinfo
= fopen("/proc/self/mountinfo", "re");
1360 if (!proc_self_mountinfo
) {
1365 /* First round, establish all mounts we need */
1369 for (m
= mounts
; m
< mounts
+ n_mounts
; ++m
) {
1374 r
= follow_symlink(root
, m
);
1378 /* We hit a symlinked mount point. The entry got rewritten and might point to a
1379 * very different place now. Let's normalize the changed list, and start from
1380 * the beginning. After all to mount the entry at the new location we might
1381 * need some other mounts first */
1386 r
= apply_mount(root
, m
);
1396 normalize_mounts(root_directory
, mounts
, &n_mounts
);
1399 /* Create a blacklist we can pass to bind_mount_recursive() */
1400 blacklist
= newa(char*, n_mounts
+1);
1401 for (j
= 0; j
< n_mounts
; j
++)
1402 blacklist
[j
] = (char*) mount_entry_path(mounts
+j
);
1403 blacklist
[j
] = NULL
;
1405 /* Second round, flip the ro bits if necessary. */
1406 for (m
= mounts
; m
< mounts
+ n_mounts
; ++m
) {
1407 r
= make_read_only(m
, blacklist
, proc_self_mountinfo
);
1414 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
1415 r
= mount_move_root(root
);
1420 /* Remount / as the desired mode. Note that this will not
1421 * reestablish propagation from our side to the host, since
1422 * what's disconnected is disconnected. */
1423 if (mount(NULL
, "/", NULL
, mount_flags
| MS_REC
, NULL
) < 0) {
1431 for (m
= mounts
; m
< mounts
+ n_mounts
; m
++)
1432 mount_entry_done(m
);
1437 void bind_mount_free_many(BindMount
*b
, size_t n
) {
1440 assert(b
|| n
== 0);
1442 for (i
= 0; i
< n
; i
++) {
1444 free(b
[i
].destination
);
1450 int bind_mount_add(BindMount
**b
, size_t *n
, const BindMount
*item
) {
1451 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
1458 s
= strdup(item
->source
);
1462 d
= strdup(item
->destination
);
1466 c
= reallocarray(*b
, *n
+ 1, sizeof(BindMount
));
1472 c
[(*n
) ++] = (BindMount
) {
1473 .source
= TAKE_PTR(s
),
1474 .destination
= TAKE_PTR(d
),
1475 .read_only
= item
->read_only
,
1476 .recursive
= item
->recursive
,
1477 .ignore_enoent
= item
->ignore_enoent
,
1483 void temporary_filesystem_free_many(TemporaryFileSystem
*t
, size_t n
) {
1486 assert(t
|| n
== 0);
1488 for (i
= 0; i
< n
; i
++) {
1496 int temporary_filesystem_add(
1497 TemporaryFileSystem
**t
,
1500 const char *options
) {
1502 _cleanup_free_
char *p
= NULL
, *o
= NULL
;
1503 TemporaryFileSystem
*c
;
1513 if (!isempty(options
)) {
1514 o
= strdup(options
);
1519 c
= reallocarray(*t
, *n
+ 1, sizeof(TemporaryFileSystem
));
1525 c
[(*n
) ++] = (TemporaryFileSystem
) {
1526 .path
= TAKE_PTR(p
),
1527 .options
= TAKE_PTR(o
),
1533 static int setup_one_tmp_dir(const char *id
, const char *prefix
, char **path
) {
1534 _cleanup_free_
char *x
= NULL
;
1535 char bid
[SD_ID128_STRING_MAX
];
1543 /* We include the boot id in the directory so that after a
1544 * reboot we can easily identify obsolete directories. */
1546 r
= sd_id128_get_boot(&boot_id
);
1550 x
= strjoin(prefix
, "/systemd-private-", sd_id128_to_string(boot_id
, bid
), "-", id
, "-XXXXXX");
1554 RUN_WITH_UMASK(0077)
1558 RUN_WITH_UMASK(0000) {
1561 y
= strjoina(x
, "/tmp");
1563 if (mkdir(y
, 0777 | S_ISVTX
) < 0)
1567 *path
= TAKE_PTR(x
);
1572 int setup_tmp_dirs(const char *id
, char **tmp_dir
, char **var_tmp_dir
) {
1578 assert(var_tmp_dir
);
1580 r
= setup_one_tmp_dir(id
, "/tmp", &a
);
1584 r
= setup_one_tmp_dir(id
, "/var/tmp", &b
);
1588 t
= strjoina(a
, "/tmp");
1602 int setup_netns(int netns_storage_socket
[2]) {
1603 _cleanup_close_
int netns
= -1;
1606 assert(netns_storage_socket
);
1607 assert(netns_storage_socket
[0] >= 0);
1608 assert(netns_storage_socket
[1] >= 0);
1610 /* We use the passed socketpair as a storage buffer for our
1611 * namespace reference fd. Whatever process runs this first
1612 * shall create a new namespace, all others should just join
1613 * it. To serialize that we use a file lock on the socket
1616 * It's a bit crazy, but hey, works great! */
1618 if (lockf(netns_storage_socket
[0], F_LOCK
, 0) < 0)
1621 netns
= receive_one_fd(netns_storage_socket
[0], MSG_DONTWAIT
);
1622 if (netns
== -EAGAIN
) {
1623 /* Nothing stored yet, so let's create a new namespace */
1625 if (unshare(CLONE_NEWNET
) < 0) {
1632 netns
= open("/proc/self/ns/net", O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
1640 } else if (netns
< 0) {
1645 /* Yay, found something, so let's join the namespace */
1646 if (setns(netns
, CLONE_NEWNET
) < 0) {
1654 q
= send_one_fd(netns_storage_socket
[1], netns
, MSG_DONTWAIT
);
1661 (void) lockf(netns_storage_socket
[0], F_ULOCK
, 0);
1665 bool ns_type_supported(NamespaceType type
) {
1666 const char *t
, *ns_proc
;
1668 t
= namespace_type_to_string(type
);
1669 if (!t
) /* Don't know how to translate this? Then it's not supported */
1672 ns_proc
= strjoina("/proc/self/ns/", t
);
1673 return access(ns_proc
, F_OK
) == 0;
1676 static const char *const protect_home_table
[_PROTECT_HOME_MAX
] = {
1677 [PROTECT_HOME_NO
] = "no",
1678 [PROTECT_HOME_YES
] = "yes",
1679 [PROTECT_HOME_READ_ONLY
] = "read-only",
1680 [PROTECT_HOME_TMPFS
] = "tmpfs",
1683 DEFINE_STRING_TABLE_LOOKUP(protect_home
, ProtectHome
);
1685 ProtectHome
protect_home_or_bool_from_string(const char *s
) {
1688 r
= parse_boolean(s
);
1690 return PROTECT_HOME_YES
;
1692 return PROTECT_HOME_NO
;
1694 return protect_home_from_string(s
);
1697 static const char *const protect_system_table
[_PROTECT_SYSTEM_MAX
] = {
1698 [PROTECT_SYSTEM_NO
] = "no",
1699 [PROTECT_SYSTEM_YES
] = "yes",
1700 [PROTECT_SYSTEM_FULL
] = "full",
1701 [PROTECT_SYSTEM_STRICT
] = "strict",
1704 DEFINE_STRING_TABLE_LOOKUP(protect_system
, ProtectSystem
);
1706 ProtectSystem
protect_system_or_bool_from_string(const char *s
) {
1709 r
= parse_boolean(s
);
1711 return PROTECT_SYSTEM_YES
;
1713 return PROTECT_SYSTEM_NO
;
1715 return protect_system_from_string(s
);
1718 static const char* const namespace_type_table
[] = {
1719 [NAMESPACE_MOUNT
] = "mnt",
1720 [NAMESPACE_CGROUP
] = "cgroup",
1721 [NAMESPACE_UTS
] = "uts",
1722 [NAMESPACE_IPC
] = "ipc",
1723 [NAMESPACE_USER
] = "user",
1724 [NAMESPACE_PID
] = "pid",
1725 [NAMESPACE_NET
] = "net",
1728 DEFINE_STRING_TABLE_LOOKUP(namespace_type
, NamespaceType
);