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"
18 #include "loop-util.h"
19 #include "loopback-setup.h"
21 #include "mount-util.h"
22 #include "mountpoint-util.h"
23 #include "namespace-util.h"
24 #include "namespace.h"
25 #include "nulstr-util.h"
26 #include "path-util.h"
27 #include "selinux-util.h"
28 #include "socket-util.h"
29 #include "sort-util.h"
30 #include "stat-util.h"
31 #include "string-table.h"
32 #include "string-util.h"
34 #include "tmpfile-util.h"
35 #include "umask-util.h"
36 #include "user-util.h"
38 #define DEV_MOUNT_OPTIONS (MS_NOSUID|MS_STRICTATIME|MS_NOEXEC)
40 typedef enum MountMode
{
41 /* This is ordered by priority! */
55 READWRITE_IMPLICIT
, /* Should have the lowest priority. */
59 typedef struct MountEntry
{
60 const char *path_const
; /* Memory allocated on stack or static */
62 bool ignore
:1; /* Ignore if path does not exist? */
63 bool has_prefix
:1; /* Already is prefixed by the root dir? */
64 bool read_only
:1; /* Shall this mount point be read-only? */
65 bool nosuid
:1; /* Shall set MS_NOSUID on the mount itself */
66 bool applied
:1; /* Already applied */
67 char *path_malloc
; /* Use this instead of 'path_const' if we had to allocate memory */
68 const char *source_const
; /* The source path, for bind mounts */
70 const char *options_const
;/* Mount options for tmpfs */
72 unsigned long flags
; /* Mount flags used by EMPTY_DIR and TMPFS. Do not include MS_RDONLY here, but please use read_only. */
76 /* If MountAPIVFS= is used, let's mount /sys and /proc into the it, but only as a fallback if the user hasn't mounted
77 * something there already. These mounts are hence overridden by any other explicitly configured mounts. */
78 static const MountEntry apivfs_table
[] = {
79 { "/proc", PROCFS
, false },
80 { "/dev", BIND_DEV
, false },
81 { "/sys", SYSFS
, false },
84 /* ProtectKernelTunables= option and the related filesystem APIs */
85 static const MountEntry protect_kernel_tunables_table
[] = {
86 { "/proc/acpi", READONLY
, true },
87 { "/proc/apm", READONLY
, true }, /* Obsolete API, there's no point in permitting access to this, ever */
88 { "/proc/asound", READONLY
, true },
89 { "/proc/bus", READONLY
, true },
90 { "/proc/fs", READONLY
, true },
91 { "/proc/irq", READONLY
, true },
92 { "/proc/kallsyms", INACCESSIBLE
, true },
93 { "/proc/kcore", INACCESSIBLE
, true },
94 { "/proc/latency_stats", READONLY
, true },
95 { "/proc/mtrr", READONLY
, true },
96 { "/proc/scsi", READONLY
, true },
97 { "/proc/sys", READONLY
, false },
98 { "/proc/sysrq-trigger", READONLY
, true },
99 { "/proc/timer_stats", READONLY
, true },
100 { "/sys", READONLY
, false },
101 { "/sys/fs/bpf", READONLY
, true },
102 { "/sys/fs/cgroup", READWRITE_IMPLICIT
, false }, /* READONLY is set by ProtectControlGroups= option */
103 { "/sys/fs/selinux", READWRITE_IMPLICIT
, true },
104 { "/sys/kernel/debug", READONLY
, true },
105 { "/sys/kernel/tracing", READONLY
, true },
108 /* ProtectKernelModules= option */
109 static const MountEntry protect_kernel_modules_table
[] = {
111 { "/lib/modules", INACCESSIBLE
, true },
113 { "/usr/lib/modules", INACCESSIBLE
, true },
116 /* ProtectKernelLogs= option */
117 static const MountEntry protect_kernel_logs_table
[] = {
118 { "/proc/kmsg", INACCESSIBLE
, true },
119 { "/dev/kmsg", INACCESSIBLE
, true },
123 * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of
124 * system should be protected by ProtectSystem=
126 static const MountEntry protect_home_read_only_table
[] = {
127 { "/home", READONLY
, true },
128 { "/run/user", READONLY
, true },
129 { "/root", READONLY
, true },
132 /* ProtectHome=tmpfs table */
133 static const MountEntry protect_home_tmpfs_table
[] = {
134 { "/home", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
135 { "/run/user", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
136 { "/root", TMPFS
, true, .read_only
= true, .options_const
= "mode=0700" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
139 /* ProtectHome=yes table */
140 static const MountEntry protect_home_yes_table
[] = {
141 { "/home", INACCESSIBLE
, true },
142 { "/run/user", INACCESSIBLE
, true },
143 { "/root", INACCESSIBLE
, true },
146 /* ProtectSystem=yes table */
147 static const MountEntry protect_system_yes_table
[] = {
148 { "/usr", READONLY
, false },
149 { "/boot", READONLY
, true },
150 { "/efi", READONLY
, true },
152 { "/lib", READONLY
, true },
153 { "/lib64", READONLY
, true },
154 { "/bin", READONLY
, true },
156 { "/sbin", READONLY
, true },
161 /* ProtectSystem=full includes ProtectSystem=yes */
162 static const MountEntry protect_system_full_table
[] = {
163 { "/usr", READONLY
, false },
164 { "/boot", READONLY
, true },
165 { "/efi", READONLY
, true },
166 { "/etc", READONLY
, false },
168 { "/lib", READONLY
, true },
169 { "/lib64", READONLY
, true },
170 { "/bin", READONLY
, true },
172 { "/sbin", READONLY
, true },
178 * ProtectSystem=strict table. In this strict mode, we mount everything
179 * read-only, except for /proc, /dev, /sys which are the kernel API VFS,
180 * which are left writable, but PrivateDevices= + ProtectKernelTunables=
181 * protect those, and these options should be fully orthogonal.
182 * (And of course /home and friends are also left writable, as ProtectHome=
183 * shall manage those, orthogonally).
185 static const MountEntry protect_system_strict_table
[] = {
186 { "/", READONLY
, false },
187 { "/proc", READWRITE_IMPLICIT
, false }, /* ProtectKernelTunables= */
188 { "/sys", READWRITE_IMPLICIT
, false }, /* ProtectKernelTunables= */
189 { "/dev", READWRITE_IMPLICIT
, false }, /* PrivateDevices= */
190 { "/home", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
191 { "/run/user", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
192 { "/root", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
195 static const char * const mount_mode_table
[_MOUNT_MODE_MAX
] = {
196 [INACCESSIBLE
] = "inaccessible",
197 [BIND_MOUNT
] = "bind",
198 [BIND_MOUNT_RECURSIVE
] = "rbind",
199 [PRIVATE_TMP
] = "private-tmp",
200 [PRIVATE_DEV
] = "private-dev",
201 [BIND_DEV
] = "bind-dev",
202 [EMPTY_DIR
] = "empty",
205 [READONLY
] = "read-only",
206 [READWRITE
] = "read-write",
208 [READWRITE_IMPLICIT
] = "rw-implicit",
211 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(mount_mode
, MountMode
);
213 static const char *mount_entry_path(const MountEntry
*p
) {
216 /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that,
217 * otherwise the stack/static ->path field is returned. */
219 return p
->path_malloc
?: p
->path_const
;
222 static bool mount_entry_read_only(const MountEntry
*p
) {
225 return p
->read_only
|| IN_SET(p
->mode
, READONLY
, INACCESSIBLE
, PRIVATE_TMP_READONLY
);
228 static const char *mount_entry_source(const MountEntry
*p
) {
231 return p
->source_malloc
?: p
->source_const
;
234 static const char *mount_entry_options(const MountEntry
*p
) {
237 return p
->options_malloc
?: p
->options_const
;
240 static void mount_entry_done(MountEntry
*p
) {
243 p
->path_malloc
= mfree(p
->path_malloc
);
244 p
->source_malloc
= mfree(p
->source_malloc
);
245 p
->options_malloc
= mfree(p
->options_malloc
);
248 static int append_access_mounts(MountEntry
**p
, char **strv
, MountMode mode
, bool forcibly_require_prefix
) {
253 /* Adds a list of user-supplied READWRITE/READWRITE_IMPLICIT/READONLY/INACCESSIBLE entries */
255 STRV_FOREACH(i
, strv
) {
256 bool ignore
= false, needs_prefix
= false;
259 /* Look for any prefixes */
260 if (startswith(e
, "-")) {
264 if (startswith(e
, "+")) {
269 if (!path_is_absolute(e
))
270 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
271 "Path is not absolute: %s", e
);
273 *((*p
)++) = (MountEntry
) {
277 .has_prefix
= !needs_prefix
&& !forcibly_require_prefix
,
284 static int append_empty_dir_mounts(MountEntry
**p
, char **strv
) {
289 /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the
290 * "/private/" boundary directories for DynamicUser=1. */
292 STRV_FOREACH(i
, strv
) {
294 *((*p
)++) = (MountEntry
) {
299 .options_const
= "mode=755" TMPFS_LIMITS_EMPTY_OR_ALMOST
,
300 .flags
= MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
,
307 static int append_bind_mounts(MountEntry
**p
, const BindMount
*binds
, size_t n
) {
312 for (i
= 0; i
< n
; i
++) {
313 const BindMount
*b
= binds
+ i
;
315 *((*p
)++) = (MountEntry
) {
316 .path_const
= b
->destination
,
317 .mode
= b
->recursive
? BIND_MOUNT_RECURSIVE
: BIND_MOUNT
,
318 .read_only
= b
->read_only
,
320 .source_const
= b
->source
,
321 .ignore
= b
->ignore_enoent
,
328 static int append_tmpfs_mounts(MountEntry
**p
, const TemporaryFileSystem
*tmpfs
, size_t n
) {
334 for (i
= 0; i
< n
; i
++) {
335 const TemporaryFileSystem
*t
= tmpfs
+ i
;
336 _cleanup_free_
char *o
= NULL
, *str
= NULL
;
340 if (!path_is_absolute(t
->path
))
341 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
342 "Path is not absolute: %s",
345 str
= strjoin("mode=0755" TMPFS_LIMITS_TEMPORARY_FS
",", t
->options
);
349 r
= mount_option_mangle(str
, MS_NODEV
|MS_STRICTATIME
, &flags
, &o
);
351 return log_debug_errno(r
, "Failed to parse mount option '%s': %m", str
);
353 ro
= flags
& MS_RDONLY
;
357 *((*p
)++) = (MountEntry
) {
358 .path_const
= t
->path
,
361 .options_malloc
= TAKE_PTR(o
),
369 static int append_static_mounts(MountEntry
**p
, const MountEntry
*mounts
, size_t n
, bool ignore_protect
) {
375 /* Adds a list of static pre-defined entries */
377 for (i
= 0; i
< n
; i
++)
378 *((*p
)++) = (MountEntry
) {
379 .path_const
= mount_entry_path(mounts
+i
),
380 .mode
= mounts
[i
].mode
,
381 .ignore
= mounts
[i
].ignore
|| ignore_protect
,
387 static int append_protect_home(MountEntry
**p
, ProtectHome protect_home
, bool ignore_protect
) {
390 switch (protect_home
) {
392 case PROTECT_HOME_NO
:
395 case PROTECT_HOME_READ_ONLY
:
396 return append_static_mounts(p
, protect_home_read_only_table
, ELEMENTSOF(protect_home_read_only_table
), ignore_protect
);
398 case PROTECT_HOME_TMPFS
:
399 return append_static_mounts(p
, protect_home_tmpfs_table
, ELEMENTSOF(protect_home_tmpfs_table
), ignore_protect
);
401 case PROTECT_HOME_YES
:
402 return append_static_mounts(p
, protect_home_yes_table
, ELEMENTSOF(protect_home_yes_table
), ignore_protect
);
405 assert_not_reached("Unexpected ProtectHome= value");
409 static int append_protect_system(MountEntry
**p
, ProtectSystem protect_system
, bool ignore_protect
) {
412 switch (protect_system
) {
414 case PROTECT_SYSTEM_NO
:
417 case PROTECT_SYSTEM_STRICT
:
418 return append_static_mounts(p
, protect_system_strict_table
, ELEMENTSOF(protect_system_strict_table
), ignore_protect
);
420 case PROTECT_SYSTEM_YES
:
421 return append_static_mounts(p
, protect_system_yes_table
, ELEMENTSOF(protect_system_yes_table
), ignore_protect
);
423 case PROTECT_SYSTEM_FULL
:
424 return append_static_mounts(p
, protect_system_full_table
, ELEMENTSOF(protect_system_full_table
), ignore_protect
);
427 assert_not_reached("Unexpected ProtectSystem= value");
431 static int mount_path_compare(const MountEntry
*a
, const MountEntry
*b
) {
434 /* If the paths are not equal, then order prefixes first */
435 d
= path_compare(mount_entry_path(a
), mount_entry_path(b
));
439 /* If the paths are equal, check the mode */
440 return CMP((int) a
->mode
, (int) b
->mode
);
443 static int prefix_where_needed(MountEntry
*m
, size_t n
, const char *root_directory
) {
446 /* Prefixes all paths in the bind mount table with the root directory if the entry needs that. */
448 for (i
= 0; i
< n
; i
++) {
454 s
= path_join(root_directory
, mount_entry_path(m
+i
));
458 free_and_replace(m
[i
].path_malloc
, s
);
459 m
[i
].has_prefix
= true;
465 static void drop_duplicates(MountEntry
*m
, size_t *n
) {
466 MountEntry
*f
, *t
, *previous
;
471 /* Drops duplicate entries. Expects that the array is properly ordered already. */
473 for (f
= m
, t
= m
, previous
= NULL
; f
< m
+ *n
; f
++) {
475 /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare()
476 * above. Note that we only drop duplicates that haven't been mounted yet. */
478 path_equal(mount_entry_path(f
), mount_entry_path(previous
)) &&
479 !f
->applied
&& !previous
->applied
) {
480 log_debug("%s (%s) is duplicate.", mount_entry_path(f
), mount_mode_to_string(f
->mode
));
481 previous
->read_only
= previous
->read_only
|| mount_entry_read_only(f
); /* Propagate the read-only flag to the remaining entry */
494 static void drop_inaccessible(MountEntry
*m
, size_t *n
) {
496 const char *clear
= NULL
;
501 /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly
502 * ordered already. */
504 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
506 /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop
507 * it, as inaccessible paths really should drop the entire subtree. */
508 if (clear
&& path_startswith(mount_entry_path(f
), clear
)) {
509 log_debug("%s is masked by %s.", mount_entry_path(f
), clear
);
514 clear
= f
->mode
== INACCESSIBLE
? mount_entry_path(f
) : NULL
;
523 static void drop_nop(MountEntry
*m
, size_t *n
) {
529 /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the
530 * list is ordered by prefixes. */
532 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
534 /* Only suppress such subtrees for READONLY, READWRITE and READWRITE_IMPLICIT entries */
535 if (IN_SET(f
->mode
, READONLY
, READWRITE
, READWRITE_IMPLICIT
)) {
539 /* Now let's find the first parent of the entry we are looking at. */
540 for (p
= t
-1; p
>= m
; p
--) {
541 if (path_startswith(mount_entry_path(f
), mount_entry_path(p
))) {
547 /* We found it, let's see if it's the same mode, if so, we can drop this entry */
548 if (found
&& p
->mode
== f
->mode
) {
549 log_debug("%s (%s) is made redundant by %s (%s)",
550 mount_entry_path(f
), mount_mode_to_string(f
->mode
),
551 mount_entry_path(p
), mount_mode_to_string(p
->mode
));
564 static void drop_outside_root(const char *root_directory
, MountEntry
*m
, size_t *n
) {
574 /* Drops all mounts that are outside of the root directory. */
576 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
578 if (!path_startswith(mount_entry_path(f
), root_directory
)) {
579 log_debug("%s is outside of root directory.", mount_entry_path(f
));
591 static int clone_device_node(
593 const char *temporary_mount
,
594 bool *make_devnode
) {
596 _cleanup_free_
char *sl
= NULL
;
597 const char *dn
, *bn
, *t
;
601 if (stat(d
, &st
) < 0) {
602 if (errno
== ENOENT
) {
603 log_debug_errno(errno
, "Device node '%s' to clone does not exist, ignoring.", d
);
607 return log_debug_errno(errno
, "Failed to stat() device node '%s' to clone, ignoring: %m", d
);
610 if (!S_ISBLK(st
.st_mode
) &&
611 !S_ISCHR(st
.st_mode
))
612 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
613 "Device node '%s' to clone is not a device node, ignoring.",
616 dn
= strjoina(temporary_mount
, d
);
618 /* First, try to create device node properly */
620 mac_selinux_create_file_prepare(d
, st
.st_mode
);
621 r
= mknod(dn
, st
.st_mode
, st
.st_rdev
);
622 mac_selinux_create_file_clear();
626 return log_debug_errno(errno
, "mknod failed for %s: %m", d
);
628 /* This didn't work, let's not try this again for the next iterations. */
629 *make_devnode
= false;
632 /* We're about to fallback to bind-mounting the device
633 * node. So create a dummy bind-mount target.
634 * Do not prepare device-node SELinux label (see issue 13762) */
635 r
= mknod(dn
, S_IFREG
, 0);
636 if (r
< 0 && errno
!= EEXIST
)
637 return log_debug_errno(errno
, "mknod() fallback failed for '%s': %m", d
);
639 /* Fallback to bind-mounting:
640 * The assumption here is that all used device nodes carry standard
641 * properties. Specifically, the devices nodes we bind-mount should
642 * either be owned by root:root or root:tty (e.g. /dev/tty, /dev/ptmx)
643 * and should not carry ACLs. */
644 if (mount(d
, dn
, NULL
, MS_BIND
, NULL
) < 0)
645 return log_debug_errno(errno
, "Bind mounting failed for '%s': %m", d
);
648 bn
= path_startswith(d
, "/dev/");
652 /* Create symlinks like /dev/char/1:9 → ../urandom */
653 if (asprintf(&sl
, "%s/dev/%s/%u:%u",
655 S_ISCHR(st
.st_mode
) ? "char" : "block",
656 major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
659 (void) mkdir_parents(sl
, 0755);
661 t
= strjoina("../", bn
);
662 if (symlink(t
, sl
) < 0)
663 log_debug_errno(errno
, "Failed to symlink '%s' to '%s', ignoring: %m", t
, sl
);
668 static int mount_private_dev(MountEntry
*m
) {
669 static const char devnodes
[] =
677 char temporary_mount
[] = "/tmp/namespace-dev-XXXXXX";
678 const char *d
, *dev
= NULL
, *devpts
= NULL
, *devshm
= NULL
, *devhugepages
= NULL
, *devmqueue
= NULL
, *devlog
= NULL
, *devptmx
= NULL
;
679 bool can_mknod
= true;
680 _cleanup_umask_ mode_t u
;
687 if (!mkdtemp(temporary_mount
))
688 return log_debug_errno(errno
, "Failed to create temporary directory '%s': %m", temporary_mount
);
690 dev
= strjoina(temporary_mount
, "/dev");
691 (void) mkdir(dev
, 0755);
692 if (mount("tmpfs", dev
, "tmpfs", DEV_MOUNT_OPTIONS
, "mode=755" TMPFS_LIMITS_DEV
) < 0) {
693 r
= log_debug_errno(errno
, "Failed to mount tmpfs on '%s': %m", dev
);
696 r
= label_fix_container(dev
, "/dev", 0);
698 log_debug_errno(errno
, "Failed to fix label of '%s' as /dev: %m", dev
);
702 devpts
= strjoina(temporary_mount
, "/dev/pts");
703 (void) mkdir(devpts
, 0755);
704 if (mount("/dev/pts", devpts
, NULL
, MS_BIND
, NULL
) < 0) {
705 r
= log_debug_errno(errno
, "Failed to bind mount /dev/pts on '%s': %m", devpts
);
709 /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx.
710 * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible.
711 * Thus, in that case make a clone.
712 * In nspawn and other containers it will be a symlink, in that case make it a symlink. */
713 r
= is_symlink("/dev/ptmx");
715 log_debug_errno(r
, "Failed to detect whether /dev/ptmx is a symlink or not: %m");
718 devptmx
= strjoina(temporary_mount
, "/dev/ptmx");
719 if (symlink("pts/ptmx", devptmx
) < 0) {
720 r
= log_debug_errno(errno
, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx
);
724 r
= clone_device_node("/dev/ptmx", temporary_mount
, &can_mknod
);
729 devshm
= strjoina(temporary_mount
, "/dev/shm");
730 (void) mkdir(devshm
, 0755);
731 r
= mount("/dev/shm", devshm
, NULL
, MS_BIND
, NULL
);
733 r
= log_debug_errno(errno
, "Failed to bind mount /dev/shm on '%s': %m", devshm
);
737 devmqueue
= strjoina(temporary_mount
, "/dev/mqueue");
738 (void) mkdir(devmqueue
, 0755);
739 if (mount("/dev/mqueue", devmqueue
, NULL
, MS_BIND
, NULL
) < 0)
740 log_debug_errno(errno
, "Failed to bind mount /dev/mqueue on '%s', ignoring: %m", devmqueue
);
742 devhugepages
= strjoina(temporary_mount
, "/dev/hugepages");
743 (void) mkdir(devhugepages
, 0755);
744 if (mount("/dev/hugepages", devhugepages
, NULL
, MS_BIND
, NULL
) < 0)
745 log_debug_errno(errno
, "Failed to bind mount /dev/hugepages on '%s', ignoring: %m", devhugepages
);
747 devlog
= strjoina(temporary_mount
, "/dev/log");
748 if (symlink("/run/systemd/journal/dev-log", devlog
) < 0)
749 log_debug_errno(errno
, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog
);
751 NULSTR_FOREACH(d
, devnodes
) {
752 r
= clone_device_node(d
, temporary_mount
, &can_mknod
);
753 /* ENXIO means the *source* is not a device file, skip creation in that case */
754 if (r
< 0 && r
!= -ENXIO
)
758 r
= dev_setup(temporary_mount
, UID_INVALID
, GID_INVALID
);
760 log_debug_errno(r
, "Failed to set up basic device tree at '%s', ignoring: %m", temporary_mount
);
762 /* Create the /dev directory if missing. It is more likely to be
763 * missing when the service is started with RootDirectory. This is
764 * consistent with mount units creating the mount points when missing.
766 (void) mkdir_p_label(mount_entry_path(m
), 0755);
768 /* Unmount everything in old /dev */
769 r
= umount_recursive(mount_entry_path(m
), 0);
771 log_debug_errno(r
, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m
));
773 if (mount(dev
, mount_entry_path(m
), NULL
, MS_MOVE
, NULL
) < 0) {
774 r
= log_debug_errno(errno
, "Failed to move mount point '%s' to '%s': %m", dev
, mount_entry_path(m
));
779 (void) rmdir(temporary_mount
);
785 (void) umount(devpts
);
788 (void) umount(devshm
);
791 (void) umount(devhugepages
);
794 (void) umount(devmqueue
);
798 (void) rmdir(temporary_mount
);
803 static int mount_bind_dev(const MountEntry
*m
) {
808 /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the service's
809 * /dev. This is only used when RootDirectory= is set. */
811 (void) mkdir_p_label(mount_entry_path(m
), 0755);
813 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
815 return log_debug_errno(r
, "Unable to determine whether /dev is already mounted: %m");
816 if (r
> 0) /* make this a NOP if /dev is already a mount point */
819 if (mount("/dev", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
820 return log_debug_errno(errno
, "Failed to bind mount %s: %m", mount_entry_path(m
));
825 static int mount_sysfs(const MountEntry
*m
) {
830 (void) mkdir_p_label(mount_entry_path(m
), 0755);
832 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
834 return log_debug_errno(r
, "Unable to determine whether /sys is already mounted: %m");
835 if (r
> 0) /* make this a NOP if /sys is already a mount point */
838 /* Bind mount the host's version so that we get all child mounts of it, too. */
839 if (mount("/sys", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
840 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
845 static int mount_procfs(const MountEntry
*m
) {
850 (void) mkdir_p_label(mount_entry_path(m
), 0755);
852 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
854 return log_debug_errno(r
, "Unable to determine whether /proc is already mounted: %m");
855 if (r
> 0) /* make this a NOP if /proc is already a mount point */
858 /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in one */
859 if (mount("proc", mount_entry_path(m
), "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
) < 0)
860 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
865 static int mount_tmpfs(const MountEntry
*m
) {
868 /* First, get rid of everything that is below if there is anything. Then, overmount with our new tmpfs */
870 (void) mkdir_p_label(mount_entry_path(m
), 0755);
871 (void) umount_recursive(mount_entry_path(m
), 0);
873 if (mount("tmpfs", mount_entry_path(m
), "tmpfs", m
->flags
, mount_entry_options(m
)) < 0)
874 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
879 static int follow_symlink(
880 const char *root_directory
,
883 _cleanup_free_
char *target
= NULL
;
886 /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we
887 * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at
888 * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the
889 * end and already have a fully normalized name. */
891 r
= chase_symlinks(mount_entry_path(m
), root_directory
, CHASE_STEP
|CHASE_NONEXISTENT
, &target
, NULL
);
893 return log_debug_errno(r
, "Failed to chase symlinks '%s': %m", mount_entry_path(m
));
894 if (r
> 0) /* Reached the end, nothing more to resolve */
897 if (m
->n_followed
>= CHASE_SYMLINKS_MAX
) /* put a boundary on things */
898 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
899 "Symlink loop on '%s'.",
900 mount_entry_path(m
));
902 log_debug("Followed mount entry path symlink %s → %s.", mount_entry_path(m
), target
);
904 free_and_replace(m
->path_malloc
, target
);
905 m
->has_prefix
= true;
912 static int apply_mount(
913 const char *root_directory
,
916 _cleanup_free_
char *inaccessible
= NULL
;
917 bool rbind
= true, make
= false;
923 log_debug("Applying namespace mount on %s", mount_entry_path(m
));
928 _cleanup_free_
char *tmp
= NULL
;
929 const char *runtime_dir
;
932 /* First, get rid of everything that is below if there
933 * is anything... Then, overmount it with an
934 * inaccessible path. */
935 (void) umount_recursive(mount_entry_path(m
), 0);
937 if (lstat(mount_entry_path(m
), &target
) < 0) {
938 if (errno
== ENOENT
&& m
->ignore
)
941 return log_debug_errno(errno
, "Failed to lstat() %s to determine what to mount over it: %m",
942 mount_entry_path(m
));
946 runtime_dir
= "/run";
948 if (asprintf(&tmp
, "/run/user/" UID_FMT
, geteuid()) < 0)
954 r
= mode_to_inaccessible_node(runtime_dir
, target
.st_mode
, &inaccessible
);
956 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
957 "File type not supported for inaccessible mounts. Note that symlinks are not allowed");
964 case READWRITE_IMPLICIT
:
965 r
= path_is_mount_point(mount_entry_path(m
), root_directory
, 0);
966 if (r
== -ENOENT
&& m
->ignore
)
969 return log_debug_errno(r
, "Failed to determine whether %s is already a mount point: %m",
970 mount_entry_path(m
));
971 if (r
> 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY
972 * bit for the mount point if needed. */
974 /* This isn't a mount point yet, let's make it one. */
975 what
= mount_entry_path(m
);
982 case BIND_MOUNT_RECURSIVE
: {
983 _cleanup_free_
char *chased
= NULL
;
985 /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note
986 * that bind mount source paths are always relative to the host root, hence we pass NULL as
987 * root directory to chase_symlinks() here. */
989 r
= chase_symlinks(mount_entry_source(m
), NULL
, CHASE_TRAIL_SLASH
, &chased
, NULL
);
990 if (r
== -ENOENT
&& m
->ignore
) {
991 log_debug_errno(r
, "Path %s does not exist, ignoring.", mount_entry_source(m
));
995 return log_debug_errno(r
, "Failed to follow symlinks on %s: %m", mount_entry_source(m
));
997 log_debug("Followed source symlinks %s → %s.", mount_entry_source(m
), chased
);
999 free_and_replace(m
->source_malloc
, chased
);
1001 what
= mount_entry_source(m
);
1008 return mount_tmpfs(m
);
1011 case PRIVATE_TMP_READONLY
:
1012 what
= mount_entry_source(m
);
1017 return mount_private_dev(m
);
1020 return mount_bind_dev(m
);
1023 return mount_sysfs(m
);
1026 return mount_procfs(m
);
1029 assert_not_reached("Unknown mode");
1034 if (mount(what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
) < 0) {
1035 bool try_again
= false;
1038 if (r
== -ENOENT
&& make
) {
1041 /* Hmm, either the source or the destination are missing. Let's see if we can create
1042 the destination, then try again. */
1044 if (stat(what
, &st
) < 0)
1045 log_error_errno(errno
, "Mount point source '%s' is not accessible: %m", what
);
1049 (void) mkdir_parents(mount_entry_path(m
), 0755);
1051 if (S_ISDIR(st
.st_mode
))
1052 q
= mkdir(mount_entry_path(m
), 0755) < 0 ? -errno
: 0;
1054 q
= touch(mount_entry_path(m
));
1057 log_error_errno(q
, "Failed to create destination mount point node '%s': %m",
1058 mount_entry_path(m
));
1065 if (mount(what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
) < 0)
1072 return log_error_errno(r
, "Failed to mount %s to %s: %m", what
, mount_entry_path(m
));
1075 log_debug("Successfully mounted %s to %s", what
, mount_entry_path(m
));
1079 static int make_read_only(const MountEntry
*m
, char **deny_list
, FILE *proc_self_mountinfo
) {
1080 unsigned long new_flags
= 0, flags_mask
= 0;
1081 bool submounts
= false;
1085 assert(proc_self_mountinfo
);
1087 if (mount_entry_read_only(m
) || m
->mode
== PRIVATE_DEV
) {
1088 new_flags
|= MS_RDONLY
;
1089 flags_mask
|= MS_RDONLY
;
1093 new_flags
|= MS_NOSUID
;
1094 flags_mask
|= MS_NOSUID
;
1097 if (flags_mask
== 0) /* No Change? */
1100 /* We generally apply these changes recursively, except for /dev, and the cases we know there's
1101 * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the
1102 * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace
1103 * and running Linux <= 4.17. */
1105 mount_entry_read_only(m
) &&
1106 !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1108 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, deny_list
, proc_self_mountinfo
);
1110 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, proc_self_mountinfo
);
1112 /* Not that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked
1113 * read-only already stays this way. This improves compatibility with container managers, where we
1114 * won't attempt to undo read-only mounts already applied. */
1116 if (r
== -ENOENT
&& m
->ignore
)
1119 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1120 submounts
? " and its submounts" : "");
1124 static bool namespace_info_mount_apivfs(const NamespaceInfo
*ns_info
) {
1128 * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
1129 * since to protect the API VFS mounts, they need to be around in the
1133 return ns_info
->mount_apivfs
||
1134 ns_info
->protect_control_groups
||
1135 ns_info
->protect_kernel_tunables
;
1138 static size_t namespace_calculate_mounts(
1139 const NamespaceInfo
*ns_info
,
1140 char** read_write_paths
,
1141 char** read_only_paths
,
1142 char** inaccessible_paths
,
1143 char** empty_directories
,
1144 size_t n_bind_mounts
,
1145 size_t n_temporary_filesystems
,
1146 const char* tmp_dir
,
1147 const char* var_tmp_dir
,
1148 const char* log_namespace
,
1149 ProtectHome protect_home
,
1150 ProtectSystem protect_system
) {
1152 size_t protect_home_cnt
;
1153 size_t protect_system_cnt
=
1154 (protect_system
== PROTECT_SYSTEM_STRICT
?
1155 ELEMENTSOF(protect_system_strict_table
) :
1156 ((protect_system
== PROTECT_SYSTEM_FULL
) ?
1157 ELEMENTSOF(protect_system_full_table
) :
1158 ((protect_system
== PROTECT_SYSTEM_YES
) ?
1159 ELEMENTSOF(protect_system_yes_table
) : 0)));
1162 (protect_home
== PROTECT_HOME_YES
?
1163 ELEMENTSOF(protect_home_yes_table
) :
1164 ((protect_home
== PROTECT_HOME_READ_ONLY
) ?
1165 ELEMENTSOF(protect_home_read_only_table
) :
1166 ((protect_home
== PROTECT_HOME_TMPFS
) ?
1167 ELEMENTSOF(protect_home_tmpfs_table
) : 0)));
1169 return !!tmp_dir
+ !!var_tmp_dir
+
1170 strv_length(read_write_paths
) +
1171 strv_length(read_only_paths
) +
1172 strv_length(inaccessible_paths
) +
1173 strv_length(empty_directories
) +
1175 n_temporary_filesystems
+
1176 ns_info
->private_dev
+
1177 (ns_info
->protect_kernel_tunables
? ELEMENTSOF(protect_kernel_tunables_table
) : 0) +
1178 (ns_info
->protect_kernel_modules
? ELEMENTSOF(protect_kernel_modules_table
) : 0) +
1179 (ns_info
->protect_kernel_logs
? ELEMENTSOF(protect_kernel_logs_table
) : 0) +
1180 (ns_info
->protect_control_groups
? 1 : 0) +
1181 protect_home_cnt
+ protect_system_cnt
+
1182 (ns_info
->protect_hostname
? 2 : 0) +
1183 (namespace_info_mount_apivfs(ns_info
) ? ELEMENTSOF(apivfs_table
) : 0) +
1187 static void normalize_mounts(const char *root_directory
, MountEntry
*mounts
, size_t *n_mounts
) {
1188 assert(root_directory
);
1190 assert(mounts
|| *n_mounts
== 0);
1192 typesafe_qsort(mounts
, *n_mounts
, mount_path_compare
);
1194 drop_duplicates(mounts
, n_mounts
);
1195 drop_outside_root(root_directory
, mounts
, n_mounts
);
1196 drop_inaccessible(mounts
, n_mounts
);
1197 drop_nop(mounts
, n_mounts
);
1200 static bool root_read_only(
1201 char **read_only_paths
,
1202 ProtectSystem protect_system
) {
1204 /* Determine whether the root directory is going to be read-only given the configured settings. */
1206 if (protect_system
== PROTECT_SYSTEM_STRICT
)
1209 if (prefixed_path_strv_contains(read_only_paths
, "/"))
1215 static bool home_read_only(
1216 char** read_only_paths
,
1217 char** inaccessible_paths
,
1218 char** empty_directories
,
1219 const BindMount
*bind_mounts
,
1220 size_t n_bind_mounts
,
1221 const TemporaryFileSystem
*temporary_filesystems
,
1222 size_t n_temporary_filesystems
,
1223 ProtectHome protect_home
) {
1227 /* Determine whether the /home directory is going to be read-only given the configured settings. Yes,
1228 * this is a bit sloppy, since we don't bother checking for cases where / is affected by multiple
1231 if (protect_home
!= PROTECT_HOME_NO
)
1234 if (prefixed_path_strv_contains(read_only_paths
, "/home") ||
1235 prefixed_path_strv_contains(inaccessible_paths
, "/home") ||
1236 prefixed_path_strv_contains(empty_directories
, "/home"))
1239 for (i
= 0; i
< n_temporary_filesystems
; i
++)
1240 if (path_equal(temporary_filesystems
[i
].path
, "/home"))
1243 /* If /home is overmounted with some dir from the host it's not writable. */
1244 for (i
= 0; i
< n_bind_mounts
; i
++)
1245 if (path_equal(bind_mounts
[i
].destination
, "/home"))
1251 int setup_namespace(
1252 const char* root_directory
,
1253 const char* root_image
,
1254 const NamespaceInfo
*ns_info
,
1255 char** read_write_paths
,
1256 char** read_only_paths
,
1257 char** inaccessible_paths
,
1258 char** empty_directories
,
1259 const BindMount
*bind_mounts
,
1260 size_t n_bind_mounts
,
1261 const TemporaryFileSystem
*temporary_filesystems
,
1262 size_t n_temporary_filesystems
,
1263 const char* tmp_dir
,
1264 const char* var_tmp_dir
,
1265 const char *log_namespace
,
1266 ProtectHome protect_home
,
1267 ProtectSystem protect_system
,
1268 unsigned long mount_flags
,
1269 const void *root_hash
,
1270 size_t root_hash_size
,
1271 const char *root_hash_path
,
1272 const void *root_hash_sig
,
1273 size_t root_hash_sig_size
,
1274 const char *root_hash_sig_path
,
1275 const char *root_verity
,
1276 DissectImageFlags dissect_image_flags
,
1277 char **error_path
) {
1279 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
1280 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
1281 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
1282 _cleanup_free_
void *root_hash_decoded
= NULL
;
1283 _cleanup_free_
char *verity_data
= NULL
, *hash_sig_path
= NULL
;
1284 MountEntry
*m
= NULL
, *mounts
= NULL
;
1286 bool require_prefix
= false;
1292 if (mount_flags
== 0)
1293 mount_flags
= MS_SHARED
;
1296 dissect_image_flags
|= DISSECT_IMAGE_REQUIRE_ROOT
;
1298 /* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */
1299 if (root_read_only(read_only_paths
,
1301 home_read_only(read_only_paths
, inaccessible_paths
, empty_directories
,
1302 bind_mounts
, n_bind_mounts
, temporary_filesystems
, n_temporary_filesystems
,
1304 strv_isempty(read_write_paths
))
1305 dissect_image_flags
|= DISSECT_IMAGE_READ_ONLY
;
1307 r
= loop_device_make_by_path(root_image
,
1308 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_READ_ONLY
) ? O_RDONLY
: -1 /* < 0 means writable if possible, read-only as fallback */,
1312 return log_debug_errno(r
, "Failed to create loop device for root image: %m");
1314 r
= verity_metadata_load(root_image
,
1316 root_hash
? NULL
: &root_hash_decoded
,
1317 root_hash
? NULL
: &root_hash_size
,
1318 root_verity
? NULL
: &verity_data
,
1319 root_hash_sig
|| root_hash_sig_path
? NULL
: &hash_sig_path
);
1321 return log_debug_errno(r
, "Failed to load root hash: %m");
1322 dissect_image_flags
|= root_verity
|| verity_data
? DISSECT_IMAGE_NO_PARTITION_TABLE
: 0;
1324 r
= dissect_image(loop_device
->fd
,
1325 root_hash
?: root_hash_decoded
,
1327 root_verity
?: verity_data
,
1328 dissect_image_flags
,
1331 return log_debug_errno(r
, "Failed to dissect image: %m");
1333 r
= dissected_image_decrypt(dissected_image
,
1335 root_hash
?: root_hash_decoded
,
1337 root_verity
?: verity_data
,
1338 root_hash_sig_path
?: hash_sig_path
,
1341 dissect_image_flags
,
1344 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
1348 root
= root_directory
;
1350 /* Always create the mount namespace in a temporary directory, instead of operating
1351 * directly in the root. The temporary directory prevents any mounts from being
1352 * potentially obscured my other mounts we already applied.
1353 * We use the same mount point for all images, which is safe, since they all live
1354 * in their own namespaces after all, and hence won't see each other. */
1356 root
= "/run/systemd/unit-root";
1357 (void) mkdir_label(root
, 0700);
1358 require_prefix
= true;
1361 n_mounts
= namespace_calculate_mounts(
1368 n_temporary_filesystems
,
1369 tmp_dir
, var_tmp_dir
,
1371 protect_home
, protect_system
);
1374 m
= mounts
= new0(MountEntry
, n_mounts
);
1378 r
= append_access_mounts(&m
, read_write_paths
, READWRITE
, require_prefix
);
1382 r
= append_access_mounts(&m
, read_only_paths
, READONLY
, require_prefix
);
1386 r
= append_access_mounts(&m
, inaccessible_paths
, INACCESSIBLE
, require_prefix
);
1390 r
= append_empty_dir_mounts(&m
, empty_directories
);
1394 r
= append_bind_mounts(&m
, bind_mounts
, n_bind_mounts
);
1398 r
= append_tmpfs_mounts(&m
, temporary_filesystems
, n_temporary_filesystems
);
1403 bool ro
= streq(tmp_dir
, RUN_SYSTEMD_EMPTY
);
1405 *(m
++) = (MountEntry
) {
1406 .path_const
= "/tmp",
1407 .mode
= ro
? PRIVATE_TMP_READONLY
: PRIVATE_TMP
,
1408 .source_const
= tmp_dir
,
1413 bool ro
= streq(var_tmp_dir
, RUN_SYSTEMD_EMPTY
);
1415 *(m
++) = (MountEntry
) {
1416 .path_const
= "/var/tmp",
1417 .mode
= ro
? PRIVATE_TMP_READONLY
: PRIVATE_TMP
,
1418 .source_const
= var_tmp_dir
,
1422 if (ns_info
->private_dev
) {
1423 *(m
++) = (MountEntry
) {
1424 .path_const
= "/dev",
1425 .mode
= PRIVATE_DEV
,
1426 .flags
= DEV_MOUNT_OPTIONS
,
1430 if (ns_info
->protect_kernel_tunables
) {
1431 r
= append_static_mounts(&m
,
1432 protect_kernel_tunables_table
,
1433 ELEMENTSOF(protect_kernel_tunables_table
),
1434 ns_info
->ignore_protect_paths
);
1439 if (ns_info
->protect_kernel_modules
) {
1440 r
= append_static_mounts(&m
,
1441 protect_kernel_modules_table
,
1442 ELEMENTSOF(protect_kernel_modules_table
),
1443 ns_info
->ignore_protect_paths
);
1448 if (ns_info
->protect_kernel_logs
) {
1449 r
= append_static_mounts(&m
,
1450 protect_kernel_logs_table
,
1451 ELEMENTSOF(protect_kernel_logs_table
),
1452 ns_info
->ignore_protect_paths
);
1457 if (ns_info
->protect_control_groups
) {
1458 *(m
++) = (MountEntry
) {
1459 .path_const
= "/sys/fs/cgroup",
1464 r
= append_protect_home(&m
, protect_home
, ns_info
->ignore_protect_paths
);
1468 r
= append_protect_system(&m
, protect_system
, false);
1472 if (namespace_info_mount_apivfs(ns_info
)) {
1473 r
= append_static_mounts(&m
,
1475 ELEMENTSOF(apivfs_table
),
1476 ns_info
->ignore_protect_paths
);
1481 if (ns_info
->protect_hostname
) {
1482 *(m
++) = (MountEntry
) {
1483 .path_const
= "/proc/sys/kernel/hostname",
1486 *(m
++) = (MountEntry
) {
1487 .path_const
= "/proc/sys/kernel/domainname",
1492 if (log_namespace
) {
1493 _cleanup_free_
char *q
;
1495 q
= strjoin("/run/systemd/journal.", log_namespace
);
1501 *(m
++) = (MountEntry
) {
1502 .path_const
= "/run/systemd/journal",
1503 .mode
= BIND_MOUNT_RECURSIVE
,
1505 .source_malloc
= TAKE_PTR(q
),
1509 assert(mounts
+ n_mounts
== m
);
1511 /* Prepend the root directory where that's necessary */
1512 r
= prefix_where_needed(mounts
, n_mounts
, root
);
1516 normalize_mounts(root
, mounts
, &n_mounts
);
1519 /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */
1521 if (unshare(CLONE_NEWNS
) < 0) {
1522 r
= log_debug_errno(errno
, "Failed to unshare the mount namespace: %m");
1523 if (IN_SET(r
, -EACCES
, -EPERM
, -EOPNOTSUPP
, -ENOSYS
))
1524 /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter
1525 * in place that doesn't allow us to create namespaces (or a missing cap), then
1526 * propagate a recognizable error back, which the caller can use to detect this case
1527 * (and only this) and optionally continue without namespacing applied. */
1533 /* Remount / as SLAVE so that nothing now mounted in the namespace
1534 * shows up in the parent */
1535 if (mount(NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
) < 0) {
1536 r
= log_debug_errno(errno
, "Failed to remount '/' as SLAVE: %m");
1541 /* A root image is specified, mount it to the right place */
1542 r
= dissected_image_mount(dissected_image
, root
, UID_INVALID
, dissect_image_flags
);
1544 log_debug_errno(r
, "Failed to mount root image: %m");
1548 if (decrypted_image
) {
1549 r
= decrypted_image_relinquish(decrypted_image
);
1551 log_debug_errno(r
, "Failed to relinquish decrypted image: %m");
1556 loop_device_relinquish(loop_device
);
1558 } else if (root_directory
) {
1560 /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
1561 r
= path_is_mount_point(root
, NULL
, AT_SYMLINK_FOLLOW
);
1563 log_debug_errno(r
, "Failed to detect that %s is a mount point or not: %m", root
);
1567 if (mount(root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0) {
1568 r
= log_debug_errno(errno
, "Failed to bind mount '%s': %m", root
);
1575 /* Let's mount the main root directory to the root directory to use */
1576 if (mount("/", root
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0) {
1577 r
= log_debug_errno(errno
, "Failed to bind mount '/' on '%s': %m", root
);
1582 /* Try to set up the new root directory before mounting anything else there. */
1583 if (root_image
|| root_directory
)
1584 (void) base_filesystem_create(root
, UID_INVALID
, GID_INVALID
);
1587 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
1588 _cleanup_free_
char **deny_list
= NULL
;
1591 /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of
1592 * /proc. For example, this is the case with the option: 'InaccessiblePaths=/proc'. */
1593 proc_self_mountinfo
= fopen("/proc/self/mountinfo", "re");
1594 if (!proc_self_mountinfo
) {
1595 r
= log_debug_errno(errno
, "Failed to open /proc/self/mountinfo: %m");
1597 *error_path
= strdup("/proc/self/mountinfo");
1601 /* First round, establish all mounts we need */
1605 for (m
= mounts
; m
< mounts
+ n_mounts
; ++m
) {
1610 r
= follow_symlink(root
, m
);
1612 if (error_path
&& mount_entry_path(m
))
1613 *error_path
= strdup(mount_entry_path(m
));
1617 /* We hit a symlinked mount point. The entry got rewritten and might
1618 * point to a very different place now. Let's normalize the changed
1619 * list, and start from the beginning. After all to mount the entry
1620 * at the new location we might need some other mounts first */
1625 r
= apply_mount(root
, m
);
1627 if (error_path
&& mount_entry_path(m
))
1628 *error_path
= strdup(mount_entry_path(m
));
1638 normalize_mounts(root
, mounts
, &n_mounts
);
1641 /* Create a deny list we can pass to bind_mount_recursive() */
1642 deny_list
= new(char*, n_mounts
+1);
1647 for (j
= 0; j
< n_mounts
; j
++)
1648 deny_list
[j
] = (char*) mount_entry_path(mounts
+j
);
1649 deny_list
[j
] = NULL
;
1651 /* Second round, flip the ro bits if necessary. */
1652 for (m
= mounts
; m
< mounts
+ n_mounts
; ++m
) {
1653 r
= make_read_only(m
, deny_list
, proc_self_mountinfo
);
1655 if (error_path
&& mount_entry_path(m
))
1656 *error_path
= strdup(mount_entry_path(m
));
1662 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
1663 r
= mount_move_root(root
);
1665 log_debug_errno(r
, "Failed to mount root with MS_MOVE: %m");
1669 /* Remount / as the desired mode. Note that this will not
1670 * reestablish propagation from our side to the host, since
1671 * what's disconnected is disconnected. */
1672 if (mount(NULL
, "/", NULL
, mount_flags
| MS_REC
, NULL
) < 0) {
1673 r
= log_debug_errno(errno
, "Failed to remount '/' with desired mount flags: %m");
1681 for (m
= mounts
; m
< mounts
+ n_mounts
; m
++)
1682 mount_entry_done(m
);
1689 void bind_mount_free_many(BindMount
*b
, size_t n
) {
1692 assert(b
|| n
== 0);
1694 for (i
= 0; i
< n
; i
++) {
1696 free(b
[i
].destination
);
1702 int bind_mount_add(BindMount
**b
, size_t *n
, const BindMount
*item
) {
1703 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
1710 s
= strdup(item
->source
);
1714 d
= strdup(item
->destination
);
1718 c
= reallocarray(*b
, *n
+ 1, sizeof(BindMount
));
1724 c
[(*n
) ++] = (BindMount
) {
1725 .source
= TAKE_PTR(s
),
1726 .destination
= TAKE_PTR(d
),
1727 .read_only
= item
->read_only
,
1728 .nosuid
= item
->nosuid
,
1729 .recursive
= item
->recursive
,
1730 .ignore_enoent
= item
->ignore_enoent
,
1736 void temporary_filesystem_free_many(TemporaryFileSystem
*t
, size_t n
) {
1739 assert(t
|| n
== 0);
1741 for (i
= 0; i
< n
; i
++) {
1749 int temporary_filesystem_add(
1750 TemporaryFileSystem
**t
,
1753 const char *options
) {
1755 _cleanup_free_
char *p
= NULL
, *o
= NULL
;
1756 TemporaryFileSystem
*c
;
1766 if (!isempty(options
)) {
1767 o
= strdup(options
);
1772 c
= reallocarray(*t
, *n
+ 1, sizeof(TemporaryFileSystem
));
1778 c
[(*n
) ++] = (TemporaryFileSystem
) {
1779 .path
= TAKE_PTR(p
),
1780 .options
= TAKE_PTR(o
),
1786 static int make_tmp_prefix(const char *prefix
) {
1787 _cleanup_free_
char *t
= NULL
;
1790 /* Don't do anything unless we know the dir is actually missing */
1791 r
= access(prefix
, F_OK
);
1794 if (errno
!= ENOENT
)
1797 r
= mkdir_parents(prefix
, 0755);
1801 r
= tempfn_random(prefix
, NULL
, &t
);
1805 if (mkdir(t
, 0777) < 0)
1808 if (chmod(t
, 01777) < 0) {
1814 if (rename(t
, prefix
) < 0) {
1817 return r
== -EEXIST
? 0 : r
; /* it's fine if someone else created the dir by now */
1824 static int make_tmp_subdir(const char *parent
, char **ret
) {
1825 _cleanup_free_
char *y
= NULL
;
1827 RUN_WITH_UMASK(0000) {
1828 y
= strjoin(parent
, "/tmp");
1832 if (mkdir(y
, 0777 | S_ISVTX
) < 0)
1841 static int setup_one_tmp_dir(const char *id
, const char *prefix
, char **path
, char **tmp_path
) {
1842 _cleanup_free_
char *x
= NULL
;
1843 char bid
[SD_ID128_STRING_MAX
];
1852 /* We include the boot id in the directory so that after a
1853 * reboot we can easily identify obsolete directories. */
1855 r
= sd_id128_get_boot(&boot_id
);
1859 x
= strjoin(prefix
, "/systemd-private-", sd_id128_to_string(boot_id
, bid
), "-", id
, "-XXXXXX");
1863 r
= make_tmp_prefix(prefix
);
1867 RUN_WITH_UMASK(0077)
1869 if (errno
== EROFS
|| ERRNO_IS_DISK_SPACE(errno
))
1876 r
= make_tmp_subdir(x
, tmp_path
);
1880 /* Trouble: we failed to create the directory. Instead of failing, let's simulate /tmp being
1881 * read-only. This way the service will get the EROFS result as if it was writing to the real
1883 r
= mkdir_p(RUN_SYSTEMD_EMPTY
, 0500);
1887 x
= strdup(RUN_SYSTEMD_EMPTY
);
1892 *path
= TAKE_PTR(x
);
1896 int setup_tmp_dirs(const char *id
, char **tmp_dir
, char **var_tmp_dir
) {
1897 _cleanup_(namespace_cleanup_tmpdirp
) char *a
= NULL
;
1898 _cleanup_(rmdir_and_freep
) char *a_tmp
= NULL
;
1904 assert(var_tmp_dir
);
1906 r
= setup_one_tmp_dir(id
, "/tmp", &a
, &a_tmp
);
1910 r
= setup_one_tmp_dir(id
, "/var/tmp", &b
, NULL
);
1914 a_tmp
= mfree(a_tmp
); /* avoid rmdir */
1915 *tmp_dir
= TAKE_PTR(a
);
1916 *var_tmp_dir
= TAKE_PTR(b
);
1921 int setup_netns(const int netns_storage_socket
[static 2]) {
1922 _cleanup_close_
int netns
= -1;
1925 assert(netns_storage_socket
);
1926 assert(netns_storage_socket
[0] >= 0);
1927 assert(netns_storage_socket
[1] >= 0);
1929 /* We use the passed socketpair as a storage buffer for our
1930 * namespace reference fd. Whatever process runs this first
1931 * shall create a new namespace, all others should just join
1932 * it. To serialize that we use a file lock on the socket
1935 * It's a bit crazy, but hey, works great! */
1937 if (lockf(netns_storage_socket
[0], F_LOCK
, 0) < 0)
1940 netns
= receive_one_fd(netns_storage_socket
[0], MSG_DONTWAIT
);
1941 if (netns
== -EAGAIN
) {
1942 /* Nothing stored yet, so let's create a new namespace. */
1944 if (unshare(CLONE_NEWNET
) < 0) {
1949 (void) loopback_setup();
1951 netns
= open("/proc/self/ns/net", O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
1959 } else if (netns
< 0) {
1964 /* Yay, found something, so let's join the namespace */
1965 if (setns(netns
, CLONE_NEWNET
) < 0) {
1973 q
= send_one_fd(netns_storage_socket
[1], netns
, MSG_DONTWAIT
);
1980 (void) lockf(netns_storage_socket
[0], F_ULOCK
, 0);
1984 int open_netns_path(const int netns_storage_socket
[static 2], const char *path
) {
1985 _cleanup_close_
int netns
= -1;
1988 assert(netns_storage_socket
);
1989 assert(netns_storage_socket
[0] >= 0);
1990 assert(netns_storage_socket
[1] >= 0);
1993 /* If the storage socket doesn't contain a netns fd yet, open one via the file system and store it in
1994 * it. This is supposed to be called ahead of time, i.e. before setup_netns() which will allocate a
1995 * new anonymous netns if needed. */
1997 if (lockf(netns_storage_socket
[0], F_LOCK
, 0) < 0)
2000 netns
= receive_one_fd(netns_storage_socket
[0], MSG_DONTWAIT
);
2001 if (netns
== -EAGAIN
) {
2002 /* Nothing stored yet. Open the file from the file system. */
2004 netns
= open(path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
2010 r
= fd_is_network_ns(netns
);
2011 if (r
== 0) { /* Not a netns? Refuse early. */
2015 if (r
< 0 && r
!= -EUCLEAN
) /* EUCLEAN: we don't know */
2020 } else if (netns
< 0) {
2024 r
= 0; /* Already allocated */
2026 q
= send_one_fd(netns_storage_socket
[1], netns
, MSG_DONTWAIT
);
2033 (void) lockf(netns_storage_socket
[0], F_ULOCK
, 0);
2037 bool ns_type_supported(NamespaceType type
) {
2038 const char *t
, *ns_proc
;
2040 t
= namespace_type_to_string(type
);
2041 if (!t
) /* Don't know how to translate this? Then it's not supported */
2044 ns_proc
= strjoina("/proc/self/ns/", t
);
2045 return access(ns_proc
, F_OK
) == 0;
2048 static const char *const protect_home_table
[_PROTECT_HOME_MAX
] = {
2049 [PROTECT_HOME_NO
] = "no",
2050 [PROTECT_HOME_YES
] = "yes",
2051 [PROTECT_HOME_READ_ONLY
] = "read-only",
2052 [PROTECT_HOME_TMPFS
] = "tmpfs",
2055 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home
, ProtectHome
, PROTECT_HOME_YES
);
2057 static const char *const protect_system_table
[_PROTECT_SYSTEM_MAX
] = {
2058 [PROTECT_SYSTEM_NO
] = "no",
2059 [PROTECT_SYSTEM_YES
] = "yes",
2060 [PROTECT_SYSTEM_FULL
] = "full",
2061 [PROTECT_SYSTEM_STRICT
] = "strict",
2064 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system
, ProtectSystem
, PROTECT_SYSTEM_YES
);
2066 static const char* const namespace_type_table
[] = {
2067 [NAMESPACE_MOUNT
] = "mnt",
2068 [NAMESPACE_CGROUP
] = "cgroup",
2069 [NAMESPACE_UTS
] = "uts",
2070 [NAMESPACE_IPC
] = "ipc",
2071 [NAMESPACE_USER
] = "user",
2072 [NAMESPACE_PID
] = "pid",
2073 [NAMESPACE_NET
] = "net",
2076 DEFINE_STRING_TABLE_LOOKUP(namespace_type
, NamespaceType
);