1 /* SPDX-License-Identifier: LGPL-2.1+ */
4 #include <linux/loop.h>
11 #include "alloc-util.h"
12 #include "base-filesystem.h"
13 #include "dev-setup.h"
15 #include "format-util.h"
19 #include "loop-util.h"
20 #include "loopback-setup.h"
22 #include "mount-util.h"
23 #include "mountpoint-util.h"
24 #include "namespace-util.h"
25 #include "namespace.h"
26 #include "nulstr-util.h"
27 #include "path-util.h"
28 #include "selinux-util.h"
29 #include "socket-util.h"
30 #include "sort-util.h"
31 #include "stat-util.h"
32 #include "string-table.h"
33 #include "string-util.h"
35 #include "tmpfile-util.h"
36 #include "umask-util.h"
37 #include "user-util.h"
39 #define DEV_MOUNT_OPTIONS (MS_NOSUID|MS_STRICTATIME|MS_NOEXEC)
41 typedef enum MountMode
{
42 /* This is ordered by priority! */
57 READWRITE_IMPLICIT
, /* Should have the lowest priority. */
61 typedef struct MountEntry
{
62 const char *path_const
; /* Memory allocated on stack or static */
64 bool ignore
:1; /* Ignore if path does not exist? */
65 bool has_prefix
:1; /* Already is prefixed by the root dir? */
66 bool read_only
:1; /* Shall this mount point be read-only? */
67 bool nosuid
:1; /* Shall set MS_NOSUID on the mount itself */
68 bool applied
:1; /* Already applied */
69 char *path_malloc
; /* Use this instead of 'path_const' if we had to allocate memory */
70 const char *source_const
; /* The source path, for bind mounts or images */
72 const char *options_const
;/* Mount options for tmpfs */
74 unsigned long flags
; /* Mount flags used by EMPTY_DIR and TMPFS. Do not include MS_RDONLY here, but please use read_only. */
76 LIST_HEAD(MountOptions
, image_options
);
79 /* If MountAPIVFS= is used, let's mount /sys and /proc into the it, but only as a fallback if the user hasn't mounted
80 * something there already. These mounts are hence overridden by any other explicitly configured mounts. */
81 static const MountEntry apivfs_table
[] = {
82 { "/proc", PROCFS
, false },
83 { "/dev", BIND_DEV
, false },
84 { "/sys", SYSFS
, false },
87 /* ProtectKernelTunables= option and the related filesystem APIs */
88 static const MountEntry protect_kernel_tunables_table
[] = {
89 { "/proc/acpi", READONLY
, true },
90 { "/proc/apm", READONLY
, true }, /* Obsolete API, there's no point in permitting access to this, ever */
91 { "/proc/asound", READONLY
, true },
92 { "/proc/bus", READONLY
, true },
93 { "/proc/fs", READONLY
, true },
94 { "/proc/irq", READONLY
, true },
95 { "/proc/kallsyms", INACCESSIBLE
, true },
96 { "/proc/kcore", INACCESSIBLE
, true },
97 { "/proc/latency_stats", READONLY
, true },
98 { "/proc/mtrr", READONLY
, true },
99 { "/proc/scsi", READONLY
, true },
100 { "/proc/sys", READONLY
, true },
101 { "/proc/sysrq-trigger", READONLY
, true },
102 { "/proc/timer_stats", READONLY
, true },
103 { "/sys", READONLY
, false },
104 { "/sys/fs/bpf", READONLY
, true },
105 { "/sys/fs/cgroup", READWRITE_IMPLICIT
, false }, /* READONLY is set by ProtectControlGroups= option */
106 { "/sys/fs/selinux", READWRITE_IMPLICIT
, true },
107 { "/sys/kernel/debug", READONLY
, true },
108 { "/sys/kernel/tracing", READONLY
, true },
111 /* ProtectKernelModules= option */
112 static const MountEntry protect_kernel_modules_table
[] = {
114 { "/lib/modules", INACCESSIBLE
, true },
116 { "/usr/lib/modules", INACCESSIBLE
, true },
119 /* ProtectKernelLogs= option */
120 static const MountEntry protect_kernel_logs_table
[] = {
121 { "/proc/kmsg", INACCESSIBLE
, true },
122 { "/dev/kmsg", INACCESSIBLE
, true },
126 * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of
127 * system should be protected by ProtectSystem=
129 static const MountEntry protect_home_read_only_table
[] = {
130 { "/home", READONLY
, true },
131 { "/run/user", READONLY
, true },
132 { "/root", READONLY
, true },
135 /* ProtectHome=tmpfs table */
136 static const MountEntry protect_home_tmpfs_table
[] = {
137 { "/home", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
138 { "/run/user", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
139 { "/root", TMPFS
, true, .read_only
= true, .options_const
= "mode=0700" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
142 /* ProtectHome=yes table */
143 static const MountEntry protect_home_yes_table
[] = {
144 { "/home", INACCESSIBLE
, true },
145 { "/run/user", INACCESSIBLE
, true },
146 { "/root", INACCESSIBLE
, true },
149 /* ProtectSystem=yes table */
150 static const MountEntry protect_system_yes_table
[] = {
151 { "/usr", READONLY
, false },
152 { "/boot", READONLY
, true },
153 { "/efi", READONLY
, true },
155 { "/lib", READONLY
, true },
156 { "/lib64", READONLY
, true },
157 { "/bin", READONLY
, true },
159 { "/sbin", READONLY
, true },
164 /* ProtectSystem=full includes ProtectSystem=yes */
165 static const MountEntry protect_system_full_table
[] = {
166 { "/usr", READONLY
, false },
167 { "/boot", READONLY
, true },
168 { "/efi", READONLY
, true },
169 { "/etc", READONLY
, false },
171 { "/lib", READONLY
, true },
172 { "/lib64", READONLY
, true },
173 { "/bin", READONLY
, true },
175 { "/sbin", READONLY
, true },
181 * ProtectSystem=strict table. In this strict mode, we mount everything
182 * read-only, except for /proc, /dev, /sys which are the kernel API VFS,
183 * which are left writable, but PrivateDevices= + ProtectKernelTunables=
184 * protect those, and these options should be fully orthogonal.
185 * (And of course /home and friends are also left writable, as ProtectHome=
186 * shall manage those, orthogonally).
188 static const MountEntry protect_system_strict_table
[] = {
189 { "/", READONLY
, false },
190 { "/proc", READWRITE_IMPLICIT
, false }, /* ProtectKernelTunables= */
191 { "/sys", READWRITE_IMPLICIT
, false }, /* ProtectKernelTunables= */
192 { "/dev", READWRITE_IMPLICIT
, false }, /* PrivateDevices= */
193 { "/home", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
194 { "/run/user", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
195 { "/root", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
198 static const char * const mount_mode_table
[_MOUNT_MODE_MAX
] = {
199 [INACCESSIBLE
] = "inaccessible",
200 [BIND_MOUNT
] = "bind",
201 [BIND_MOUNT_RECURSIVE
] = "rbind",
202 [PRIVATE_TMP
] = "private-tmp",
203 [PRIVATE_DEV
] = "private-dev",
204 [BIND_DEV
] = "bind-dev",
205 [EMPTY_DIR
] = "empty",
208 [READONLY
] = "read-only",
209 [READWRITE
] = "read-write",
211 [MOUNT_IMAGES
] = "mount-images",
212 [READWRITE_IMPLICIT
] = "rw-implicit",
215 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(mount_mode
, MountMode
);
217 static const char *mount_entry_path(const MountEntry
*p
) {
220 /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that,
221 * otherwise the stack/static ->path field is returned. */
223 return p
->path_malloc
?: p
->path_const
;
226 static bool mount_entry_read_only(const MountEntry
*p
) {
229 return p
->read_only
|| IN_SET(p
->mode
, READONLY
, INACCESSIBLE
, PRIVATE_TMP_READONLY
);
232 static const char *mount_entry_source(const MountEntry
*p
) {
235 return p
->source_malloc
?: p
->source_const
;
238 static const char *mount_entry_options(const MountEntry
*p
) {
241 return p
->options_malloc
?: p
->options_const
;
244 static void mount_entry_done(MountEntry
*p
) {
247 p
->path_malloc
= mfree(p
->path_malloc
);
248 p
->source_malloc
= mfree(p
->source_malloc
);
249 p
->options_malloc
= mfree(p
->options_malloc
);
250 p
->image_options
= mount_options_free_all(p
->image_options
);
253 static int append_access_mounts(MountEntry
**p
, char **strv
, MountMode mode
, bool forcibly_require_prefix
) {
258 /* Adds a list of user-supplied READWRITE/READWRITE_IMPLICIT/READONLY/INACCESSIBLE entries */
260 STRV_FOREACH(i
, strv
) {
261 bool ignore
= false, needs_prefix
= false;
264 /* Look for any prefixes */
265 if (startswith(e
, "-")) {
269 if (startswith(e
, "+")) {
274 if (!path_is_absolute(e
))
275 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
276 "Path is not absolute: %s", e
);
278 *((*p
)++) = (MountEntry
) {
282 .has_prefix
= !needs_prefix
&& !forcibly_require_prefix
,
289 static int append_empty_dir_mounts(MountEntry
**p
, char **strv
) {
294 /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the
295 * "/private/" boundary directories for DynamicUser=1. */
297 STRV_FOREACH(i
, strv
) {
299 *((*p
)++) = (MountEntry
) {
304 .options_const
= "mode=755" TMPFS_LIMITS_EMPTY_OR_ALMOST
,
305 .flags
= MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
,
312 static int append_bind_mounts(MountEntry
**p
, const BindMount
*binds
, size_t n
) {
317 for (i
= 0; i
< n
; i
++) {
318 const BindMount
*b
= binds
+ i
;
320 *((*p
)++) = (MountEntry
) {
321 .path_const
= b
->destination
,
322 .mode
= b
->recursive
? BIND_MOUNT_RECURSIVE
: BIND_MOUNT
,
323 .read_only
= b
->read_only
,
325 .source_const
= b
->source
,
326 .ignore
= b
->ignore_enoent
,
333 static int append_mount_images(MountEntry
**p
, const MountImage
*mount_images
, size_t n
) {
336 for (size_t i
= 0; i
< n
; i
++) {
337 const MountImage
*m
= mount_images
+ i
;
339 *((*p
)++) = (MountEntry
) {
340 .path_const
= m
->destination
,
341 .mode
= MOUNT_IMAGES
,
342 .source_const
= m
->source
,
343 .image_options
= m
->mount_options
,
344 .ignore
= m
->ignore_enoent
,
351 static int append_tmpfs_mounts(MountEntry
**p
, const TemporaryFileSystem
*tmpfs
, size_t n
) {
354 for (size_t i
= 0; i
< n
; i
++) {
355 const TemporaryFileSystem
*t
= tmpfs
+ i
;
356 _cleanup_free_
char *o
= NULL
, *str
= NULL
;
361 if (!path_is_absolute(t
->path
))
362 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
363 "Path is not absolute: %s",
366 str
= strjoin("mode=0755" NESTED_TMPFS_LIMITS
",", t
->options
);
370 r
= mount_option_mangle(str
, MS_NODEV
|MS_STRICTATIME
, &flags
, &o
);
372 return log_debug_errno(r
, "Failed to parse mount option '%s': %m", str
);
374 ro
= flags
& MS_RDONLY
;
378 *((*p
)++) = (MountEntry
) {
379 .path_const
= t
->path
,
382 .options_malloc
= TAKE_PTR(o
),
390 static int append_static_mounts(MountEntry
**p
, const MountEntry
*mounts
, size_t n
, bool ignore_protect
) {
396 /* Adds a list of static pre-defined entries */
398 for (i
= 0; i
< n
; i
++)
399 *((*p
)++) = (MountEntry
) {
400 .path_const
= mount_entry_path(mounts
+i
),
401 .mode
= mounts
[i
].mode
,
402 .ignore
= mounts
[i
].ignore
|| ignore_protect
,
408 static int append_protect_home(MountEntry
**p
, ProtectHome protect_home
, bool ignore_protect
) {
411 switch (protect_home
) {
413 case PROTECT_HOME_NO
:
416 case PROTECT_HOME_READ_ONLY
:
417 return append_static_mounts(p
, protect_home_read_only_table
, ELEMENTSOF(protect_home_read_only_table
), ignore_protect
);
419 case PROTECT_HOME_TMPFS
:
420 return append_static_mounts(p
, protect_home_tmpfs_table
, ELEMENTSOF(protect_home_tmpfs_table
), ignore_protect
);
422 case PROTECT_HOME_YES
:
423 return append_static_mounts(p
, protect_home_yes_table
, ELEMENTSOF(protect_home_yes_table
), ignore_protect
);
426 assert_not_reached("Unexpected ProtectHome= value");
430 static int append_protect_system(MountEntry
**p
, ProtectSystem protect_system
, bool ignore_protect
) {
433 switch (protect_system
) {
435 case PROTECT_SYSTEM_NO
:
438 case PROTECT_SYSTEM_STRICT
:
439 return append_static_mounts(p
, protect_system_strict_table
, ELEMENTSOF(protect_system_strict_table
), ignore_protect
);
441 case PROTECT_SYSTEM_YES
:
442 return append_static_mounts(p
, protect_system_yes_table
, ELEMENTSOF(protect_system_yes_table
), ignore_protect
);
444 case PROTECT_SYSTEM_FULL
:
445 return append_static_mounts(p
, protect_system_full_table
, ELEMENTSOF(protect_system_full_table
), ignore_protect
);
448 assert_not_reached("Unexpected ProtectSystem= value");
452 static int mount_path_compare(const MountEntry
*a
, const MountEntry
*b
) {
455 /* If the paths are not equal, then order prefixes first */
456 d
= path_compare(mount_entry_path(a
), mount_entry_path(b
));
460 /* If the paths are equal, check the mode */
461 return CMP((int) a
->mode
, (int) b
->mode
);
464 static int prefix_where_needed(MountEntry
*m
, size_t n
, const char *root_directory
) {
467 /* Prefixes all paths in the bind mount table with the root directory if the entry needs that. */
469 for (i
= 0; i
< n
; i
++) {
475 s
= path_join(root_directory
, mount_entry_path(m
+i
));
479 free_and_replace(m
[i
].path_malloc
, s
);
480 m
[i
].has_prefix
= true;
486 static void drop_duplicates(MountEntry
*m
, size_t *n
) {
487 MountEntry
*f
, *t
, *previous
;
492 /* Drops duplicate entries. Expects that the array is properly ordered already. */
494 for (f
= m
, t
= m
, previous
= NULL
; f
< m
+ *n
; f
++) {
496 /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare()
497 * above. Note that we only drop duplicates that haven't been mounted yet. */
499 path_equal(mount_entry_path(f
), mount_entry_path(previous
)) &&
500 !f
->applied
&& !previous
->applied
) {
501 log_debug("%s (%s) is duplicate.", mount_entry_path(f
), mount_mode_to_string(f
->mode
));
502 previous
->read_only
= previous
->read_only
|| mount_entry_read_only(f
); /* Propagate the read-only flag to the remaining entry */
515 static void drop_inaccessible(MountEntry
*m
, size_t *n
) {
517 const char *clear
= NULL
;
522 /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly
523 * ordered already. */
525 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
527 /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop
528 * it, as inaccessible paths really should drop the entire subtree. */
529 if (clear
&& path_startswith(mount_entry_path(f
), clear
)) {
530 log_debug("%s is masked by %s.", mount_entry_path(f
), clear
);
535 clear
= f
->mode
== INACCESSIBLE
? mount_entry_path(f
) : NULL
;
544 static void drop_nop(MountEntry
*m
, size_t *n
) {
550 /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the
551 * list is ordered by prefixes. */
553 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
555 /* Only suppress such subtrees for READONLY, READWRITE and READWRITE_IMPLICIT entries */
556 if (IN_SET(f
->mode
, READONLY
, READWRITE
, READWRITE_IMPLICIT
)) {
560 /* Now let's find the first parent of the entry we are looking at. */
561 for (p
= t
-1; p
>= m
; p
--) {
562 if (path_startswith(mount_entry_path(f
), mount_entry_path(p
))) {
568 /* We found it, let's see if it's the same mode, if so, we can drop this entry */
569 if (found
&& p
->mode
== f
->mode
) {
570 log_debug("%s (%s) is made redundant by %s (%s)",
571 mount_entry_path(f
), mount_mode_to_string(f
->mode
),
572 mount_entry_path(p
), mount_mode_to_string(p
->mode
));
585 static void drop_outside_root(const char *root_directory
, MountEntry
*m
, size_t *n
) {
595 /* Drops all mounts that are outside of the root directory. */
597 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
599 if (!path_startswith(mount_entry_path(f
), root_directory
)) {
600 log_debug("%s is outside of root directory.", mount_entry_path(f
));
612 static int clone_device_node(
614 const char *temporary_mount
,
615 bool *make_devnode
) {
617 _cleanup_free_
char *sl
= NULL
;
618 const char *dn
, *bn
, *t
;
622 if (stat(d
, &st
) < 0) {
623 if (errno
== ENOENT
) {
624 log_debug_errno(errno
, "Device node '%s' to clone does not exist, ignoring.", d
);
628 return log_debug_errno(errno
, "Failed to stat() device node '%s' to clone, ignoring: %m", d
);
631 if (!S_ISBLK(st
.st_mode
) &&
632 !S_ISCHR(st
.st_mode
))
633 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
634 "Device node '%s' to clone is not a device node, ignoring.",
637 dn
= strjoina(temporary_mount
, d
);
639 /* First, try to create device node properly */
641 mac_selinux_create_file_prepare(d
, st
.st_mode
);
642 r
= mknod(dn
, st
.st_mode
, st
.st_rdev
);
643 mac_selinux_create_file_clear();
647 return log_debug_errno(errno
, "mknod failed for %s: %m", d
);
649 /* This didn't work, let's not try this again for the next iterations. */
650 *make_devnode
= false;
653 /* We're about to fall back to bind-mounting the device
654 * node. So create a dummy bind-mount target.
655 * Do not prepare device-node SELinux label (see issue 13762) */
656 r
= mknod(dn
, S_IFREG
, 0);
657 if (r
< 0 && errno
!= EEXIST
)
658 return log_debug_errno(errno
, "mknod() fallback failed for '%s': %m", d
);
660 /* Fallback to bind-mounting:
661 * The assumption here is that all used device nodes carry standard
662 * properties. Specifically, the devices nodes we bind-mount should
663 * either be owned by root:root or root:tty (e.g. /dev/tty, /dev/ptmx)
664 * and should not carry ACLs. */
665 if (mount(d
, dn
, NULL
, MS_BIND
, NULL
) < 0)
666 return log_debug_errno(errno
, "Bind mounting failed for '%s': %m", d
);
669 bn
= path_startswith(d
, "/dev/");
673 /* Create symlinks like /dev/char/1:9 → ../urandom */
674 if (asprintf(&sl
, "%s/dev/%s/%u:%u",
676 S_ISCHR(st
.st_mode
) ? "char" : "block",
677 major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
680 (void) mkdir_parents(sl
, 0755);
682 t
= strjoina("../", bn
);
683 if (symlink(t
, sl
) < 0)
684 log_debug_errno(errno
, "Failed to symlink '%s' to '%s', ignoring: %m", t
, sl
);
689 static int mount_private_dev(MountEntry
*m
) {
690 static const char devnodes
[] =
698 char temporary_mount
[] = "/tmp/namespace-dev-XXXXXX";
699 const char *d
, *dev
= NULL
, *devpts
= NULL
, *devshm
= NULL
, *devhugepages
= NULL
, *devmqueue
= NULL
, *devlog
= NULL
, *devptmx
= NULL
;
700 bool can_mknod
= true;
701 _cleanup_umask_ mode_t u
;
708 if (!mkdtemp(temporary_mount
))
709 return log_debug_errno(errno
, "Failed to create temporary directory '%s': %m", temporary_mount
);
711 dev
= strjoina(temporary_mount
, "/dev");
712 (void) mkdir(dev
, 0755);
713 if (mount("tmpfs", dev
, "tmpfs", DEV_MOUNT_OPTIONS
, "mode=755" TMPFS_LIMITS_DEV
) < 0) {
714 r
= log_debug_errno(errno
, "Failed to mount tmpfs on '%s': %m", dev
);
717 r
= label_fix_container(dev
, "/dev", 0);
719 log_debug_errno(errno
, "Failed to fix label of '%s' as /dev: %m", dev
);
723 devpts
= strjoina(temporary_mount
, "/dev/pts");
724 (void) mkdir(devpts
, 0755);
725 if (mount("/dev/pts", devpts
, NULL
, MS_BIND
, NULL
) < 0) {
726 r
= log_debug_errno(errno
, "Failed to bind mount /dev/pts on '%s': %m", devpts
);
730 /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx.
731 * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible.
732 * Thus, in that case make a clone.
733 * In nspawn and other containers it will be a symlink, in that case make it a symlink. */
734 r
= is_symlink("/dev/ptmx");
736 log_debug_errno(r
, "Failed to detect whether /dev/ptmx is a symlink or not: %m");
739 devptmx
= strjoina(temporary_mount
, "/dev/ptmx");
740 if (symlink("pts/ptmx", devptmx
) < 0) {
741 r
= log_debug_errno(errno
, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx
);
745 r
= clone_device_node("/dev/ptmx", temporary_mount
, &can_mknod
);
750 devshm
= strjoina(temporary_mount
, "/dev/shm");
751 (void) mkdir(devshm
, 0755);
752 r
= mount("/dev/shm", devshm
, NULL
, MS_BIND
, NULL
);
754 r
= log_debug_errno(errno
, "Failed to bind mount /dev/shm on '%s': %m", devshm
);
758 devmqueue
= strjoina(temporary_mount
, "/dev/mqueue");
759 (void) mkdir(devmqueue
, 0755);
760 if (mount("/dev/mqueue", devmqueue
, NULL
, MS_BIND
, NULL
) < 0)
761 log_debug_errno(errno
, "Failed to bind mount /dev/mqueue on '%s', ignoring: %m", devmqueue
);
763 devhugepages
= strjoina(temporary_mount
, "/dev/hugepages");
764 (void) mkdir(devhugepages
, 0755);
765 if (mount("/dev/hugepages", devhugepages
, NULL
, MS_BIND
, NULL
) < 0)
766 log_debug_errno(errno
, "Failed to bind mount /dev/hugepages on '%s', ignoring: %m", devhugepages
);
768 devlog
= strjoina(temporary_mount
, "/dev/log");
769 if (symlink("/run/systemd/journal/dev-log", devlog
) < 0)
770 log_debug_errno(errno
, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog
);
772 NULSTR_FOREACH(d
, devnodes
) {
773 r
= clone_device_node(d
, temporary_mount
, &can_mknod
);
774 /* ENXIO means the *source* is not a device file, skip creation in that case */
775 if (r
< 0 && r
!= -ENXIO
)
779 r
= dev_setup(temporary_mount
, UID_INVALID
, GID_INVALID
);
781 log_debug_errno(r
, "Failed to set up basic device tree at '%s', ignoring: %m", temporary_mount
);
783 /* Create the /dev directory if missing. It is more likely to be
784 * missing when the service is started with RootDirectory. This is
785 * consistent with mount units creating the mount points when missing.
787 (void) mkdir_p_label(mount_entry_path(m
), 0755);
789 /* Unmount everything in old /dev */
790 r
= umount_recursive(mount_entry_path(m
), 0);
792 log_debug_errno(r
, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m
));
794 if (mount(dev
, mount_entry_path(m
), NULL
, MS_MOVE
, NULL
) < 0) {
795 r
= log_debug_errno(errno
, "Failed to move mount point '%s' to '%s': %m", dev
, mount_entry_path(m
));
800 (void) rmdir(temporary_mount
);
806 (void) umount(devpts
);
809 (void) umount(devshm
);
812 (void) umount(devhugepages
);
815 (void) umount(devmqueue
);
819 (void) rmdir(temporary_mount
);
824 static int mount_bind_dev(const MountEntry
*m
) {
829 /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the service's
830 * /dev. This is only used when RootDirectory= is set. */
832 (void) mkdir_p_label(mount_entry_path(m
), 0755);
834 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
836 return log_debug_errno(r
, "Unable to determine whether /dev is already mounted: %m");
837 if (r
> 0) /* make this a NOP if /dev is already a mount point */
840 if (mount("/dev", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
841 return log_debug_errno(errno
, "Failed to bind mount %s: %m", mount_entry_path(m
));
846 static int mount_sysfs(const MountEntry
*m
) {
851 (void) mkdir_p_label(mount_entry_path(m
), 0755);
853 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
855 return log_debug_errno(r
, "Unable to determine whether /sys is already mounted: %m");
856 if (r
> 0) /* make this a NOP if /sys is already a mount point */
859 /* Bind mount the host's version so that we get all child mounts of it, too. */
860 if (mount("/sys", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
861 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
866 static int mount_procfs(const MountEntry
*m
, const NamespaceInfo
*ns_info
) {
867 const char *entry_path
;
872 entry_path
= mount_entry_path(m
);
874 /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in
875 * one. i.e we don't reuse existing mounts here under any condition, we want a new instance owned by
876 * our user namespace and with our hidepid= settings applied. Hence, let's get rid of everything
877 * mounted on /proc/ first. */
879 (void) mkdir_p_label(entry_path
, 0755);
880 (void) umount_recursive(entry_path
, 0);
882 if (ns_info
->protect_proc
!= PROTECT_PROC_DEFAULT
||
883 ns_info
->proc_subset
!= PROC_SUBSET_ALL
) {
884 _cleanup_free_
char *opts
= NULL
;
886 /* Starting with kernel 5.8 procfs' hidepid= logic is truly per-instance (previously it
887 * pretended to be per-instance but actually was per-namespace), hence let's make use of it
888 * if requested. To make sure this logic succeeds only on kernels where hidepid= is
889 * per-instance, we'll exclusively use the textual value for hidepid=, since support was
890 * added in the same commit: if it's supported it is thus also per-instance. */
892 opts
= strjoin("hidepid=",
893 ns_info
->protect_proc
== PROTECT_PROC_DEFAULT
? "off" :
894 protect_proc_to_string(ns_info
->protect_proc
),
895 ns_info
->proc_subset
== PROC_SUBSET_PID
? ",subset=pid" : "");
899 if (mount("proc", entry_path
, "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, opts
) < 0) {
901 return log_debug_errno(errno
, "Failed to mount %s (options=%s): %m", mount_entry_path(m
), opts
);
903 /* If this failed with EINVAL then this likely means the textual hidepid= stuff is
904 * not supported by the kernel, and thus the per-instance hidepid= neither, which
905 * means we really don't want to use it, since it would affect our host's /proc
906 * mount. Hence let's gracefully fallback to a classic, unrestricted version. */
911 if (mount("proc", entry_path
, "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
) < 0)
912 return log_debug_errno(errno
, "Failed to mount %s (no options): %m", mount_entry_path(m
));
917 static int mount_tmpfs(const MountEntry
*m
) {
918 const char *entry_path
, *inner_path
;
923 entry_path
= mount_entry_path(m
);
924 inner_path
= m
->path_const
;
926 /* First, get rid of everything that is below if there is anything. Then, overmount with our new tmpfs */
928 (void) mkdir_p_label(entry_path
, 0755);
929 (void) umount_recursive(entry_path
, 0);
931 if (mount("tmpfs", entry_path
, "tmpfs", m
->flags
, mount_entry_options(m
)) < 0)
932 return log_debug_errno(errno
, "Failed to mount %s: %m", entry_path
);
934 r
= label_fix_container(entry_path
, inner_path
, 0);
936 return log_debug_errno(r
, "Failed to fix label of '%s' as '%s': %m", entry_path
, inner_path
);
941 static int mount_images(const MountEntry
*m
) {
942 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
943 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
944 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
945 _cleanup_(verity_settings_done
) VeritySettings verity
= {};
946 DissectImageFlags dissect_image_flags
;
951 r
= verity_settings_load(&verity
, mount_entry_source(m
), NULL
, NULL
);
953 return log_debug_errno(r
, "Failed to load root hash: %m");
955 dissect_image_flags
=
956 (m
->read_only
? DISSECT_IMAGE_READ_ONLY
: 0) |
957 (verity
.data_path
? DISSECT_IMAGE_NO_PARTITION_TABLE
: 0);
959 r
= loop_device_make_by_path(
960 mount_entry_source(m
),
961 m
->read_only
? O_RDONLY
: -1 /* < 0 means writable if possible, read-only as fallback */,
962 verity
.data_path
? 0 : LO_FLAGS_PARTSCAN
,
965 return log_debug_errno(r
, "Failed to create loop device for image: %m");
973 /* No partition table? Might be a single-filesystem image, try again */
974 if (!verity
.data_path
&& r
== -ENOPKG
)
979 dissect_image_flags
|DISSECT_IMAGE_NO_PARTITION_TABLE
,
982 return log_debug_errno(r
, "Failed to dissect image: %m");
984 r
= dissected_image_decrypt(
991 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
993 r
= mkdir_p_label(mount_entry_path(m
), 0755);
995 return log_debug_errno(r
, "Failed to create destination directory %s: %m", mount_entry_path(m
));
996 r
= umount_recursive(mount_entry_path(m
), 0);
998 return log_debug_errno(r
, "Failed to umount under destination directory %s: %m", mount_entry_path(m
));
1000 r
= dissected_image_mount(dissected_image
, mount_entry_path(m
), UID_INVALID
, dissect_image_flags
);
1002 return log_debug_errno(r
, "Failed to mount image: %m");
1004 if (decrypted_image
) {
1005 r
= decrypted_image_relinquish(decrypted_image
);
1007 return log_debug_errno(r
, "Failed to relinquish decrypted image: %m");
1010 loop_device_relinquish(loop_device
);
1015 static int follow_symlink(
1016 const char *root_directory
,
1019 _cleanup_free_
char *target
= NULL
;
1022 /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we
1023 * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at
1024 * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the
1025 * end and already have a fully normalized name. */
1027 r
= chase_symlinks(mount_entry_path(m
), root_directory
, CHASE_STEP
|CHASE_NONEXISTENT
, &target
, NULL
);
1029 return log_debug_errno(r
, "Failed to chase symlinks '%s': %m", mount_entry_path(m
));
1030 if (r
> 0) /* Reached the end, nothing more to resolve */
1033 if (m
->n_followed
>= CHASE_SYMLINKS_MAX
) /* put a boundary on things */
1034 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
1035 "Symlink loop on '%s'.",
1036 mount_entry_path(m
));
1038 log_debug("Followed mount entry path symlink %s → %s.", mount_entry_path(m
), target
);
1040 free_and_replace(m
->path_malloc
, target
);
1041 m
->has_prefix
= true;
1048 static int apply_mount(
1049 const char *root_directory
,
1051 const NamespaceInfo
*ns_info
) {
1053 _cleanup_free_
char *inaccessible
= NULL
;
1054 bool rbind
= true, make
= false;
1061 log_debug("Applying namespace mount on %s", mount_entry_path(m
));
1065 case INACCESSIBLE
: {
1066 _cleanup_free_
char *tmp
= NULL
;
1067 const char *runtime_dir
;
1070 /* First, get rid of everything that is below if there
1071 * is anything... Then, overmount it with an
1072 * inaccessible path. */
1073 (void) umount_recursive(mount_entry_path(m
), 0);
1075 if (lstat(mount_entry_path(m
), &target
) < 0) {
1076 if (errno
== ENOENT
&& m
->ignore
)
1079 return log_debug_errno(errno
, "Failed to lstat() %s to determine what to mount over it: %m",
1080 mount_entry_path(m
));
1084 runtime_dir
= "/run";
1086 if (asprintf(&tmp
, "/run/user/" UID_FMT
, geteuid()) < 0)
1092 r
= mode_to_inaccessible_node(runtime_dir
, target
.st_mode
, &inaccessible
);
1094 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
1095 "File type not supported for inaccessible mounts. Note that symlinks are not allowed");
1096 what
= inaccessible
;
1102 case READWRITE_IMPLICIT
:
1103 r
= path_is_mount_point(mount_entry_path(m
), root_directory
, 0);
1104 if (r
== -ENOENT
&& m
->ignore
)
1107 return log_debug_errno(r
, "Failed to determine whether %s is already a mount point: %m",
1108 mount_entry_path(m
));
1109 if (r
> 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY
1110 * bit for the mount point if needed. */
1112 /* This isn't a mount point yet, let's make it one. */
1113 what
= mount_entry_path(m
);
1120 case BIND_MOUNT_RECURSIVE
: {
1121 _cleanup_free_
char *chased
= NULL
;
1123 /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note
1124 * that bind mount source paths are always relative to the host root, hence we pass NULL as
1125 * root directory to chase_symlinks() here. */
1127 r
= chase_symlinks(mount_entry_source(m
), NULL
, CHASE_TRAIL_SLASH
, &chased
, NULL
);
1128 if (r
== -ENOENT
&& m
->ignore
) {
1129 log_debug_errno(r
, "Path %s does not exist, ignoring.", mount_entry_source(m
));
1133 return log_debug_errno(r
, "Failed to follow symlinks on %s: %m", mount_entry_source(m
));
1135 log_debug("Followed source symlinks %s → %s.", mount_entry_source(m
), chased
);
1137 free_and_replace(m
->source_malloc
, chased
);
1139 what
= mount_entry_source(m
);
1146 return mount_tmpfs(m
);
1149 case PRIVATE_TMP_READONLY
:
1150 what
= mount_entry_source(m
);
1155 return mount_private_dev(m
);
1158 return mount_bind_dev(m
);
1161 return mount_sysfs(m
);
1164 return mount_procfs(m
, ns_info
);
1167 return mount_images(m
);
1170 assert_not_reached("Unknown mode");
1175 if (mount(what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
) < 0) {
1176 bool try_again
= false;
1179 if (r
== -ENOENT
&& make
) {
1182 /* Hmm, either the source or the destination are missing. Let's see if we can create
1183 the destination, then try again. */
1185 if (stat(what
, &st
) < 0)
1186 log_error_errno(errno
, "Mount point source '%s' is not accessible: %m", what
);
1190 (void) mkdir_parents(mount_entry_path(m
), 0755);
1192 if (S_ISDIR(st
.st_mode
))
1193 q
= mkdir(mount_entry_path(m
), 0755) < 0 ? -errno
: 0;
1195 q
= touch(mount_entry_path(m
));
1198 log_error_errno(q
, "Failed to create destination mount point node '%s': %m",
1199 mount_entry_path(m
));
1206 if (mount(what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
) < 0)
1213 return log_error_errno(r
, "Failed to mount %s to %s: %m", what
, mount_entry_path(m
));
1216 log_debug("Successfully mounted %s to %s", what
, mount_entry_path(m
));
1220 static int make_read_only(const MountEntry
*m
, char **deny_list
, FILE *proc_self_mountinfo
) {
1221 unsigned long new_flags
= 0, flags_mask
= 0;
1222 bool submounts
= false;
1226 assert(proc_self_mountinfo
);
1228 if (mount_entry_read_only(m
) || m
->mode
== PRIVATE_DEV
) {
1229 new_flags
|= MS_RDONLY
;
1230 flags_mask
|= MS_RDONLY
;
1234 new_flags
|= MS_NOSUID
;
1235 flags_mask
|= MS_NOSUID
;
1238 if (flags_mask
== 0) /* No Change? */
1241 /* We generally apply these changes recursively, except for /dev, and the cases we know there's
1242 * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the
1243 * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace
1244 * and running Linux <= 4.17. */
1246 mount_entry_read_only(m
) &&
1247 !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1249 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, deny_list
, proc_self_mountinfo
);
1251 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, proc_self_mountinfo
);
1253 /* Not that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked
1254 * read-only already stays this way. This improves compatibility with container managers, where we
1255 * won't attempt to undo read-only mounts already applied. */
1257 if (r
== -ENOENT
&& m
->ignore
)
1260 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1261 submounts
? " and its submounts" : "");
1265 static bool namespace_info_mount_apivfs(const NamespaceInfo
*ns_info
) {
1269 * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
1270 * since to protect the API VFS mounts, they need to be around in the
1274 return ns_info
->mount_apivfs
||
1275 ns_info
->protect_control_groups
||
1276 ns_info
->protect_kernel_tunables
||
1277 ns_info
->protect_proc
!= PROTECT_PROC_DEFAULT
||
1278 ns_info
->proc_subset
!= PROC_SUBSET_ALL
;
1281 static size_t namespace_calculate_mounts(
1282 const NamespaceInfo
*ns_info
,
1283 char** read_write_paths
,
1284 char** read_only_paths
,
1285 char** inaccessible_paths
,
1286 char** empty_directories
,
1287 size_t n_bind_mounts
,
1288 size_t n_temporary_filesystems
,
1289 size_t n_mount_images
,
1290 const char* tmp_dir
,
1291 const char* var_tmp_dir
,
1292 const char *creds_path
,
1293 const char* log_namespace
) {
1295 size_t protect_home_cnt
;
1296 size_t protect_system_cnt
=
1297 (ns_info
->protect_system
== PROTECT_SYSTEM_STRICT
?
1298 ELEMENTSOF(protect_system_strict_table
) :
1299 ((ns_info
->protect_system
== PROTECT_SYSTEM_FULL
) ?
1300 ELEMENTSOF(protect_system_full_table
) :
1301 ((ns_info
->protect_system
== PROTECT_SYSTEM_YES
) ?
1302 ELEMENTSOF(protect_system_yes_table
) : 0)));
1305 (ns_info
->protect_home
== PROTECT_HOME_YES
?
1306 ELEMENTSOF(protect_home_yes_table
) :
1307 ((ns_info
->protect_home
== PROTECT_HOME_READ_ONLY
) ?
1308 ELEMENTSOF(protect_home_read_only_table
) :
1309 ((ns_info
->protect_home
== PROTECT_HOME_TMPFS
) ?
1310 ELEMENTSOF(protect_home_tmpfs_table
) : 0)));
1312 return !!tmp_dir
+ !!var_tmp_dir
+
1313 strv_length(read_write_paths
) +
1314 strv_length(read_only_paths
) +
1315 strv_length(inaccessible_paths
) +
1316 strv_length(empty_directories
) +
1319 n_temporary_filesystems
+
1320 ns_info
->private_dev
+
1321 (ns_info
->protect_kernel_tunables
? ELEMENTSOF(protect_kernel_tunables_table
) : 0) +
1322 (ns_info
->protect_kernel_modules
? ELEMENTSOF(protect_kernel_modules_table
) : 0) +
1323 (ns_info
->protect_kernel_logs
? ELEMENTSOF(protect_kernel_logs_table
) : 0) +
1324 (ns_info
->protect_control_groups
? 1 : 0) +
1325 protect_home_cnt
+ protect_system_cnt
+
1326 (ns_info
->protect_hostname
? 2 : 0) +
1327 (namespace_info_mount_apivfs(ns_info
) ? ELEMENTSOF(apivfs_table
) : 0) +
1328 (creds_path
? 2 : 1) +
1332 static void normalize_mounts(const char *root_directory
, MountEntry
*mounts
, size_t *n_mounts
) {
1333 assert(root_directory
);
1335 assert(mounts
|| *n_mounts
== 0);
1337 typesafe_qsort(mounts
, *n_mounts
, mount_path_compare
);
1339 drop_duplicates(mounts
, n_mounts
);
1340 drop_outside_root(root_directory
, mounts
, n_mounts
);
1341 drop_inaccessible(mounts
, n_mounts
);
1342 drop_nop(mounts
, n_mounts
);
1345 static bool root_read_only(
1346 char **read_only_paths
,
1347 ProtectSystem protect_system
) {
1349 /* Determine whether the root directory is going to be read-only given the configured settings. */
1351 if (protect_system
== PROTECT_SYSTEM_STRICT
)
1354 if (prefixed_path_strv_contains(read_only_paths
, "/"))
1360 static bool home_read_only(
1361 char** read_only_paths
,
1362 char** inaccessible_paths
,
1363 char** empty_directories
,
1364 const BindMount
*bind_mounts
,
1365 size_t n_bind_mounts
,
1366 const TemporaryFileSystem
*temporary_filesystems
,
1367 size_t n_temporary_filesystems
,
1368 ProtectHome protect_home
) {
1372 /* Determine whether the /home directory is going to be read-only given the configured settings. Yes,
1373 * this is a bit sloppy, since we don't bother checking for cases where / is affected by multiple
1376 if (protect_home
!= PROTECT_HOME_NO
)
1379 if (prefixed_path_strv_contains(read_only_paths
, "/home") ||
1380 prefixed_path_strv_contains(inaccessible_paths
, "/home") ||
1381 prefixed_path_strv_contains(empty_directories
, "/home"))
1384 for (i
= 0; i
< n_temporary_filesystems
; i
++)
1385 if (path_equal(temporary_filesystems
[i
].path
, "/home"))
1388 /* If /home is overmounted with some dir from the host it's not writable. */
1389 for (i
= 0; i
< n_bind_mounts
; i
++)
1390 if (path_equal(bind_mounts
[i
].destination
, "/home"))
1396 static int verity_settings_prepare(
1397 VeritySettings
*verity
,
1398 const char *root_image
,
1399 const void *root_hash
,
1400 size_t root_hash_size
,
1401 const char *root_hash_path
,
1402 const void *root_hash_sig
,
1403 size_t root_hash_sig_size
,
1404 const char *root_hash_sig_path
,
1405 const char *verity_data_path
) {
1414 d
= memdup(root_hash
, root_hash_size
);
1418 free_and_replace(verity
->root_hash
, d
);
1419 verity
->root_hash_size
= root_hash_size
;
1422 if (root_hash_sig
) {
1425 d
= memdup(root_hash_sig
, root_hash_sig_size
);
1429 free_and_replace(verity
->root_hash_sig
, d
);
1430 verity
->root_hash_sig_size
= root_hash_sig_size
;
1433 if (verity_data_path
) {
1434 r
= free_and_strdup(&verity
->data_path
, verity_data_path
);
1439 r
= verity_settings_load(
1443 root_hash_sig_path
);
1445 return log_debug_errno(r
, "Failed to load root hash: %m");
1450 int setup_namespace(
1451 const char* root_directory
,
1452 const char* root_image
,
1453 const MountOptions
*root_image_options
,
1454 const NamespaceInfo
*ns_info
,
1455 char** read_write_paths
,
1456 char** read_only_paths
,
1457 char** inaccessible_paths
,
1458 char** empty_directories
,
1459 const BindMount
*bind_mounts
,
1460 size_t n_bind_mounts
,
1461 const TemporaryFileSystem
*temporary_filesystems
,
1462 size_t n_temporary_filesystems
,
1463 const MountImage
*mount_images
,
1464 size_t n_mount_images
,
1465 const char* tmp_dir
,
1466 const char* var_tmp_dir
,
1467 const char *creds_path
,
1468 const char *log_namespace
,
1469 unsigned long mount_flags
,
1470 const void *root_hash
,
1471 size_t root_hash_size
,
1472 const char *root_hash_path
,
1473 const void *root_hash_sig
,
1474 size_t root_hash_sig_size
,
1475 const char *root_hash_sig_path
,
1476 const char *verity_data_path
,
1477 DissectImageFlags dissect_image_flags
,
1478 char **error_path
) {
1480 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
1481 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
1482 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
1483 _cleanup_(verity_settings_done
) VeritySettings verity
= {};
1484 MountEntry
*m
= NULL
, *mounts
= NULL
;
1485 bool require_prefix
= false;
1492 if (mount_flags
== 0)
1493 mount_flags
= MS_SHARED
;
1496 dissect_image_flags
|= DISSECT_IMAGE_REQUIRE_ROOT
;
1498 /* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */
1499 if (root_read_only(read_only_paths
,
1500 ns_info
->protect_system
) &&
1501 home_read_only(read_only_paths
, inaccessible_paths
, empty_directories
,
1502 bind_mounts
, n_bind_mounts
, temporary_filesystems
, n_temporary_filesystems
,
1503 ns_info
->protect_home
) &&
1504 strv_isempty(read_write_paths
))
1505 dissect_image_flags
|= DISSECT_IMAGE_READ_ONLY
;
1507 r
= verity_settings_prepare(
1510 root_hash
, root_hash_size
, root_hash_path
,
1511 root_hash_sig
, root_hash_sig_size
, root_hash_sig_path
,
1516 SET_FLAG(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
, verity
.data_path
);
1518 r
= loop_device_make_by_path(
1520 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_READ_ONLY
) ? O_RDONLY
: -1 /* < 0 means writable if possible, read-only as fallback */,
1521 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
1524 return log_debug_errno(r
, "Failed to create loop device for root image: %m");
1530 dissect_image_flags
,
1533 return log_debug_errno(r
, "Failed to dissect image: %m");
1535 r
= dissected_image_decrypt(
1539 dissect_image_flags
,
1542 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
1546 root
= root_directory
;
1548 /* Always create the mount namespace in a temporary directory, instead of operating
1549 * directly in the root. The temporary directory prevents any mounts from being
1550 * potentially obscured my other mounts we already applied.
1551 * We use the same mount point for all images, which is safe, since they all live
1552 * in their own namespaces after all, and hence won't see each other. */
1554 root
= "/run/systemd/unit-root";
1555 (void) mkdir_label(root
, 0700);
1556 require_prefix
= true;
1559 n_mounts
= namespace_calculate_mounts(
1566 n_temporary_filesystems
,
1568 tmp_dir
, var_tmp_dir
,
1573 m
= mounts
= new0(MountEntry
, n_mounts
);
1577 r
= append_access_mounts(&m
, read_write_paths
, READWRITE
, require_prefix
);
1581 r
= append_access_mounts(&m
, read_only_paths
, READONLY
, require_prefix
);
1585 r
= append_access_mounts(&m
, inaccessible_paths
, INACCESSIBLE
, require_prefix
);
1589 r
= append_empty_dir_mounts(&m
, empty_directories
);
1593 r
= append_bind_mounts(&m
, bind_mounts
, n_bind_mounts
);
1597 r
= append_tmpfs_mounts(&m
, temporary_filesystems
, n_temporary_filesystems
);
1602 bool ro
= streq(tmp_dir
, RUN_SYSTEMD_EMPTY
);
1604 *(m
++) = (MountEntry
) {
1605 .path_const
= "/tmp",
1606 .mode
= ro
? PRIVATE_TMP_READONLY
: PRIVATE_TMP
,
1607 .source_const
= tmp_dir
,
1612 bool ro
= streq(var_tmp_dir
, RUN_SYSTEMD_EMPTY
);
1614 *(m
++) = (MountEntry
) {
1615 .path_const
= "/var/tmp",
1616 .mode
= ro
? PRIVATE_TMP_READONLY
: PRIVATE_TMP
,
1617 .source_const
= var_tmp_dir
,
1621 r
= append_mount_images(&m
, mount_images
, n_mount_images
);
1625 if (ns_info
->private_dev
) {
1626 *(m
++) = (MountEntry
) {
1627 .path_const
= "/dev",
1628 .mode
= PRIVATE_DEV
,
1629 .flags
= DEV_MOUNT_OPTIONS
,
1633 if (ns_info
->protect_kernel_tunables
) {
1634 r
= append_static_mounts(&m
,
1635 protect_kernel_tunables_table
,
1636 ELEMENTSOF(protect_kernel_tunables_table
),
1637 ns_info
->ignore_protect_paths
);
1642 if (ns_info
->protect_kernel_modules
) {
1643 r
= append_static_mounts(&m
,
1644 protect_kernel_modules_table
,
1645 ELEMENTSOF(protect_kernel_modules_table
),
1646 ns_info
->ignore_protect_paths
);
1651 if (ns_info
->protect_kernel_logs
) {
1652 r
= append_static_mounts(&m
,
1653 protect_kernel_logs_table
,
1654 ELEMENTSOF(protect_kernel_logs_table
),
1655 ns_info
->ignore_protect_paths
);
1660 if (ns_info
->protect_control_groups
) {
1661 *(m
++) = (MountEntry
) {
1662 .path_const
= "/sys/fs/cgroup",
1667 r
= append_protect_home(&m
, ns_info
->protect_home
, ns_info
->ignore_protect_paths
);
1671 r
= append_protect_system(&m
, ns_info
->protect_system
, false);
1675 if (namespace_info_mount_apivfs(ns_info
)) {
1676 r
= append_static_mounts(&m
,
1678 ELEMENTSOF(apivfs_table
),
1679 ns_info
->ignore_protect_paths
);
1684 if (ns_info
->protect_hostname
) {
1685 *(m
++) = (MountEntry
) {
1686 .path_const
= "/proc/sys/kernel/hostname",
1689 *(m
++) = (MountEntry
) {
1690 .path_const
= "/proc/sys/kernel/domainname",
1696 /* If our service has a credentials store configured, then bind that one in, but hide
1697 * everything else. */
1699 *(m
++) = (MountEntry
) {
1700 .path_const
= "/run/credentials",
1703 .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
,
1704 .flags
= MS_NODEV
|MS_STRICTATIME
|MS_NOSUID
|MS_NOEXEC
,
1707 *(m
++) = (MountEntry
) {
1708 .path_const
= creds_path
,
1711 .source_const
= creds_path
,
1714 /* If our service has no credentials store configured, then make the whole
1715 * credentials tree inaccessible wholesale. */
1717 *(m
++) = (MountEntry
) {
1718 .path_const
= "/run/credentials",
1719 .mode
= INACCESSIBLE
,
1724 if (log_namespace
) {
1725 _cleanup_free_
char *q
;
1727 q
= strjoin("/run/systemd/journal.", log_namespace
);
1733 *(m
++) = (MountEntry
) {
1734 .path_const
= "/run/systemd/journal",
1735 .mode
= BIND_MOUNT_RECURSIVE
,
1737 .source_malloc
= TAKE_PTR(q
),
1741 assert(mounts
+ n_mounts
== m
);
1743 /* Prepend the root directory where that's necessary */
1744 r
= prefix_where_needed(mounts
, n_mounts
, root
);
1748 normalize_mounts(root
, mounts
, &n_mounts
);
1751 /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */
1753 if (unshare(CLONE_NEWNS
) < 0) {
1754 r
= log_debug_errno(errno
, "Failed to unshare the mount namespace: %m");
1755 if (IN_SET(r
, -EACCES
, -EPERM
, -EOPNOTSUPP
, -ENOSYS
))
1756 /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter
1757 * in place that doesn't allow us to create namespaces (or a missing cap), then
1758 * propagate a recognizable error back, which the caller can use to detect this case
1759 * (and only this) and optionally continue without namespacing applied. */
1765 /* Remount / as SLAVE so that nothing now mounted in the namespace
1766 * shows up in the parent */
1767 if (mount(NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
) < 0) {
1768 r
= log_debug_errno(errno
, "Failed to remount '/' as SLAVE: %m");
1773 /* A root image is specified, mount it to the right place */
1774 r
= dissected_image_mount(dissected_image
, root
, UID_INVALID
, dissect_image_flags
);
1776 log_debug_errno(r
, "Failed to mount root image: %m");
1780 if (decrypted_image
) {
1781 r
= decrypted_image_relinquish(decrypted_image
);
1783 log_debug_errno(r
, "Failed to relinquish decrypted image: %m");
1788 loop_device_relinquish(loop_device
);
1790 } else if (root_directory
) {
1792 /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
1793 r
= path_is_mount_point(root
, NULL
, AT_SYMLINK_FOLLOW
);
1795 log_debug_errno(r
, "Failed to detect that %s is a mount point or not: %m", root
);
1799 if (mount(root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0) {
1800 r
= log_debug_errno(errno
, "Failed to bind mount '%s': %m", root
);
1807 /* Let's mount the main root directory to the root directory to use */
1808 if (mount("/", root
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0) {
1809 r
= log_debug_errno(errno
, "Failed to bind mount '/' on '%s': %m", root
);
1814 /* Try to set up the new root directory before mounting anything else there. */
1815 if (root_image
|| root_directory
)
1816 (void) base_filesystem_create(root
, UID_INVALID
, GID_INVALID
);
1819 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
1820 _cleanup_free_
char **deny_list
= NULL
;
1823 /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of
1824 * /proc. For example, this is the case with the option: 'InaccessiblePaths=/proc'. */
1825 proc_self_mountinfo
= fopen("/proc/self/mountinfo", "re");
1826 if (!proc_self_mountinfo
) {
1827 r
= log_debug_errno(errno
, "Failed to open /proc/self/mountinfo: %m");
1829 *error_path
= strdup("/proc/self/mountinfo");
1833 /* First round, establish all mounts we need */
1837 for (m
= mounts
; m
< mounts
+ n_mounts
; ++m
) {
1842 r
= follow_symlink(root
, m
);
1844 if (error_path
&& mount_entry_path(m
))
1845 *error_path
= strdup(mount_entry_path(m
));
1849 /* We hit a symlinked mount point. The entry got rewritten and might
1850 * point to a very different place now. Let's normalize the changed
1851 * list, and start from the beginning. After all to mount the entry
1852 * at the new location we might need some other mounts first */
1857 r
= apply_mount(root
, m
, ns_info
);
1859 if (error_path
&& mount_entry_path(m
))
1860 *error_path
= strdup(mount_entry_path(m
));
1870 normalize_mounts(root
, mounts
, &n_mounts
);
1873 /* Create a deny list we can pass to bind_mount_recursive() */
1874 deny_list
= new(char*, n_mounts
+1);
1879 for (j
= 0; j
< n_mounts
; j
++)
1880 deny_list
[j
] = (char*) mount_entry_path(mounts
+j
);
1881 deny_list
[j
] = NULL
;
1883 /* Second round, flip the ro bits if necessary. */
1884 for (m
= mounts
; m
< mounts
+ n_mounts
; ++m
) {
1885 r
= make_read_only(m
, deny_list
, proc_self_mountinfo
);
1887 if (error_path
&& mount_entry_path(m
))
1888 *error_path
= strdup(mount_entry_path(m
));
1894 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
1895 r
= mount_move_root(root
);
1897 log_debug_errno(r
, "Failed to mount root with MS_MOVE: %m");
1901 /* Remount / as the desired mode. Note that this will not
1902 * reestablish propagation from our side to the host, since
1903 * what's disconnected is disconnected. */
1904 if (mount(NULL
, "/", NULL
, mount_flags
| MS_REC
, NULL
) < 0) {
1905 r
= log_debug_errno(errno
, "Failed to remount '/' with desired mount flags: %m");
1913 for (m
= mounts
; m
< mounts
+ n_mounts
; m
++)
1914 mount_entry_done(m
);
1921 void bind_mount_free_many(BindMount
*b
, size_t n
) {
1924 assert(b
|| n
== 0);
1926 for (i
= 0; i
< n
; i
++) {
1928 free(b
[i
].destination
);
1934 int bind_mount_add(BindMount
**b
, size_t *n
, const BindMount
*item
) {
1935 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
1942 s
= strdup(item
->source
);
1946 d
= strdup(item
->destination
);
1950 c
= reallocarray(*b
, *n
+ 1, sizeof(BindMount
));
1956 c
[(*n
) ++] = (BindMount
) {
1957 .source
= TAKE_PTR(s
),
1958 .destination
= TAKE_PTR(d
),
1959 .read_only
= item
->read_only
,
1960 .nosuid
= item
->nosuid
,
1961 .recursive
= item
->recursive
,
1962 .ignore_enoent
= item
->ignore_enoent
,
1968 MountImage
* mount_image_free_many(MountImage
*m
, size_t *n
) {
1972 assert(m
|| *n
== 0);
1974 for (i
= 0; i
< *n
; i
++) {
1976 free(m
[i
].destination
);
1977 mount_options_free_all(m
[i
].mount_options
);
1985 int mount_image_add(MountImage
**m
, size_t *n
, const MountImage
*item
) {
1986 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
1987 _cleanup_(mount_options_free_allp
) MountOptions
*options
= NULL
;
1995 s
= strdup(item
->source
);
1999 d
= strdup(item
->destination
);
2003 LIST_FOREACH(mount_options
, i
, item
->mount_options
) {
2004 _cleanup_(mount_options_free_allp
) MountOptions
*o
;
2006 o
= new(MountOptions
, 1);
2010 *o
= (MountOptions
) {
2011 .partition_designator
= i
->partition_designator
,
2012 .options
= strdup(i
->options
),
2017 LIST_APPEND(mount_options
, options
, TAKE_PTR(o
));
2020 c
= reallocarray(*m
, *n
+ 1, sizeof(MountImage
));
2026 c
[(*n
) ++] = (MountImage
) {
2027 .source
= TAKE_PTR(s
),
2028 .destination
= TAKE_PTR(d
),
2029 .mount_options
= TAKE_PTR(options
),
2030 .ignore_enoent
= item
->ignore_enoent
,
2036 void temporary_filesystem_free_many(TemporaryFileSystem
*t
, size_t n
) {
2039 assert(t
|| n
== 0);
2041 for (i
= 0; i
< n
; i
++) {
2049 int temporary_filesystem_add(
2050 TemporaryFileSystem
**t
,
2053 const char *options
) {
2055 _cleanup_free_
char *p
= NULL
, *o
= NULL
;
2056 TemporaryFileSystem
*c
;
2066 if (!isempty(options
)) {
2067 o
= strdup(options
);
2072 c
= reallocarray(*t
, *n
+ 1, sizeof(TemporaryFileSystem
));
2078 c
[(*n
) ++] = (TemporaryFileSystem
) {
2079 .path
= TAKE_PTR(p
),
2080 .options
= TAKE_PTR(o
),
2086 static int make_tmp_prefix(const char *prefix
) {
2087 _cleanup_free_
char *t
= NULL
;
2090 /* Don't do anything unless we know the dir is actually missing */
2091 r
= access(prefix
, F_OK
);
2094 if (errno
!= ENOENT
)
2097 r
= mkdir_parents(prefix
, 0755);
2101 r
= tempfn_random(prefix
, NULL
, &t
);
2105 if (mkdir(t
, 0777) < 0)
2108 if (chmod(t
, 01777) < 0) {
2114 if (rename(t
, prefix
) < 0) {
2117 return r
== -EEXIST
? 0 : r
; /* it's fine if someone else created the dir by now */
2124 static int make_tmp_subdir(const char *parent
, char **ret
) {
2125 _cleanup_free_
char *y
= NULL
;
2127 y
= path_join(parent
, "/tmp");
2131 RUN_WITH_UMASK(0000) {
2132 if (mkdir(y
, 0777 | S_ISVTX
) < 0)
2141 static int setup_one_tmp_dir(const char *id
, const char *prefix
, char **path
, char **tmp_path
) {
2142 _cleanup_free_
char *x
= NULL
;
2143 char bid
[SD_ID128_STRING_MAX
];
2152 /* We include the boot id in the directory so that after a
2153 * reboot we can easily identify obsolete directories. */
2155 r
= sd_id128_get_boot(&boot_id
);
2159 x
= strjoin(prefix
, "/systemd-private-", sd_id128_to_string(boot_id
, bid
), "-", id
, "-XXXXXX");
2163 r
= make_tmp_prefix(prefix
);
2167 RUN_WITH_UMASK(0077)
2169 if (errno
== EROFS
|| ERRNO_IS_DISK_SPACE(errno
))
2176 r
= make_tmp_subdir(x
, tmp_path
);
2180 /* Trouble: we failed to create the directory. Instead of failing, let's simulate /tmp being
2181 * read-only. This way the service will get the EROFS result as if it was writing to the real
2183 r
= mkdir_p(RUN_SYSTEMD_EMPTY
, 0500);
2187 r
= free_and_strdup(&x
, RUN_SYSTEMD_EMPTY
);
2192 *path
= TAKE_PTR(x
);
2196 int setup_tmp_dirs(const char *id
, char **tmp_dir
, char **var_tmp_dir
) {
2197 _cleanup_(namespace_cleanup_tmpdirp
) char *a
= NULL
;
2198 _cleanup_(rmdir_and_freep
) char *a_tmp
= NULL
;
2204 assert(var_tmp_dir
);
2206 r
= setup_one_tmp_dir(id
, "/tmp", &a
, &a_tmp
);
2210 r
= setup_one_tmp_dir(id
, "/var/tmp", &b
, NULL
);
2214 a_tmp
= mfree(a_tmp
); /* avoid rmdir */
2215 *tmp_dir
= TAKE_PTR(a
);
2216 *var_tmp_dir
= TAKE_PTR(b
);
2221 int setup_netns(const int netns_storage_socket
[static 2]) {
2222 _cleanup_close_
int netns
= -1;
2225 assert(netns_storage_socket
);
2226 assert(netns_storage_socket
[0] >= 0);
2227 assert(netns_storage_socket
[1] >= 0);
2229 /* We use the passed socketpair as a storage buffer for our
2230 * namespace reference fd. Whatever process runs this first
2231 * shall create a new namespace, all others should just join
2232 * it. To serialize that we use a file lock on the socket
2235 * It's a bit crazy, but hey, works great! */
2237 if (lockf(netns_storage_socket
[0], F_LOCK
, 0) < 0)
2240 netns
= receive_one_fd(netns_storage_socket
[0], MSG_DONTWAIT
);
2241 if (netns
== -EAGAIN
) {
2242 /* Nothing stored yet, so let's create a new namespace. */
2244 if (unshare(CLONE_NEWNET
) < 0) {
2249 (void) loopback_setup();
2251 netns
= open("/proc/self/ns/net", O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
2259 } else if (netns
< 0) {
2264 /* Yay, found something, so let's join the namespace */
2265 if (setns(netns
, CLONE_NEWNET
) < 0) {
2273 q
= send_one_fd(netns_storage_socket
[1], netns
, MSG_DONTWAIT
);
2280 (void) lockf(netns_storage_socket
[0], F_ULOCK
, 0);
2284 int open_netns_path(const int netns_storage_socket
[static 2], const char *path
) {
2285 _cleanup_close_
int netns
= -1;
2288 assert(netns_storage_socket
);
2289 assert(netns_storage_socket
[0] >= 0);
2290 assert(netns_storage_socket
[1] >= 0);
2293 /* If the storage socket doesn't contain a netns fd yet, open one via the file system and store it in
2294 * it. This is supposed to be called ahead of time, i.e. before setup_netns() which will allocate a
2295 * new anonymous netns if needed. */
2297 if (lockf(netns_storage_socket
[0], F_LOCK
, 0) < 0)
2300 netns
= receive_one_fd(netns_storage_socket
[0], MSG_DONTWAIT
);
2301 if (netns
== -EAGAIN
) {
2302 /* Nothing stored yet. Open the file from the file system. */
2304 netns
= open(path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
2310 r
= fd_is_network_ns(netns
);
2311 if (r
== 0) { /* Not a netns? Refuse early. */
2315 if (r
< 0 && r
!= -EUCLEAN
) /* EUCLEAN: we don't know */
2320 } else if (netns
< 0) {
2324 r
= 0; /* Already allocated */
2326 q
= send_one_fd(netns_storage_socket
[1], netns
, MSG_DONTWAIT
);
2333 (void) lockf(netns_storage_socket
[0], F_ULOCK
, 0);
2337 bool ns_type_supported(NamespaceType type
) {
2338 const char *t
, *ns_proc
;
2340 t
= namespace_type_to_string(type
);
2341 if (!t
) /* Don't know how to translate this? Then it's not supported */
2344 ns_proc
= strjoina("/proc/self/ns/", t
);
2345 return access(ns_proc
, F_OK
) == 0;
2348 static const char *const protect_home_table
[_PROTECT_HOME_MAX
] = {
2349 [PROTECT_HOME_NO
] = "no",
2350 [PROTECT_HOME_YES
] = "yes",
2351 [PROTECT_HOME_READ_ONLY
] = "read-only",
2352 [PROTECT_HOME_TMPFS
] = "tmpfs",
2355 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home
, ProtectHome
, PROTECT_HOME_YES
);
2357 static const char *const protect_system_table
[_PROTECT_SYSTEM_MAX
] = {
2358 [PROTECT_SYSTEM_NO
] = "no",
2359 [PROTECT_SYSTEM_YES
] = "yes",
2360 [PROTECT_SYSTEM_FULL
] = "full",
2361 [PROTECT_SYSTEM_STRICT
] = "strict",
2364 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system
, ProtectSystem
, PROTECT_SYSTEM_YES
);
2366 static const char* const namespace_type_table
[] = {
2367 [NAMESPACE_MOUNT
] = "mnt",
2368 [NAMESPACE_CGROUP
] = "cgroup",
2369 [NAMESPACE_UTS
] = "uts",
2370 [NAMESPACE_IPC
] = "ipc",
2371 [NAMESPACE_USER
] = "user",
2372 [NAMESPACE_PID
] = "pid",
2373 [NAMESPACE_NET
] = "net",
2376 DEFINE_STRING_TABLE_LOOKUP(namespace_type
, NamespaceType
);
2378 static const char* const protect_proc_table
[_PROTECT_PROC_MAX
] = {
2379 [PROTECT_PROC_DEFAULT
] = "default",
2380 [PROTECT_PROC_NOACCESS
] = "noaccess",
2381 [PROTECT_PROC_INVISIBLE
] = "invisible",
2382 [PROTECT_PROC_PTRACEABLE
] = "ptraceable",
2385 DEFINE_STRING_TABLE_LOOKUP(protect_proc
, ProtectProc
);
2387 static const char* const proc_subset_table
[_PROC_SUBSET_MAX
] = {
2388 [PROC_SUBSET_ALL
] = "all",
2389 [PROC_SUBSET_PID
] = "pid",
2392 DEFINE_STRING_TABLE_LOOKUP(proc_subset
, ProcSubset
);