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! */
54 READWRITE_IMPLICIT
, /* Should have the lowest priority. */
58 typedef struct MountEntry
{
59 const char *path_const
; /* Memory allocated on stack or static */
61 bool ignore
:1; /* Ignore if path does not exist? */
62 bool has_prefix
:1; /* Already is prefixed by the root dir? */
63 bool read_only
:1; /* Shall this mount point be read-only? */
64 bool nosuid
:1; /* Shall set MS_NOSUID on the mount itself */
65 bool applied
:1; /* Already applied */
66 char *path_malloc
; /* Use this instead of 'path_const' if we had to allocate memory */
67 const char *source_const
; /* The source path, for bind mounts */
69 const char *options_const
;/* Mount options for tmpfs */
71 unsigned long flags
; /* Mount flags used by EMPTY_DIR and TMPFS. Do not include MS_RDONLY here, but please use read_only. */
75 /* If MountAPIVFS= is used, let's mount /sys and /proc into the it, but only as a fallback if the user hasn't mounted
76 * something there already. These mounts are hence overridden by any other explicitly configured mounts. */
77 static const MountEntry apivfs_table
[] = {
78 { "/proc", PROCFS
, false },
79 { "/dev", BIND_DEV
, false },
80 { "/sys", SYSFS
, false },
83 /* ProtectKernelTunables= option and the related filesystem APIs */
84 static const MountEntry protect_kernel_tunables_table
[] = {
85 { "/proc/acpi", READONLY
, true },
86 { "/proc/apm", READONLY
, true }, /* Obsolete API, there's no point in permitting access to this, ever */
87 { "/proc/asound", READONLY
, true },
88 { "/proc/bus", READONLY
, true },
89 { "/proc/fs", READONLY
, true },
90 { "/proc/irq", READONLY
, true },
91 { "/proc/kallsyms", INACCESSIBLE
, true },
92 { "/proc/kcore", INACCESSIBLE
, true },
93 { "/proc/latency_stats", READONLY
, true },
94 { "/proc/mtrr", READONLY
, true },
95 { "/proc/scsi", READONLY
, true },
96 { "/proc/sys", READONLY
, false },
97 { "/proc/sysrq-trigger", READONLY
, true },
98 { "/proc/timer_stats", READONLY
, true },
99 { "/sys", READONLY
, false },
100 { "/sys/fs/bpf", READONLY
, true },
101 { "/sys/fs/cgroup", READWRITE_IMPLICIT
, false }, /* READONLY is set by ProtectControlGroups= option */
102 { "/sys/fs/selinux", READWRITE_IMPLICIT
, true },
103 { "/sys/kernel/debug", READONLY
, true },
104 { "/sys/kernel/tracing", READONLY
, true },
107 /* ProtectKernelModules= option */
108 static const MountEntry protect_kernel_modules_table
[] = {
110 { "/lib/modules", INACCESSIBLE
, true },
112 { "/usr/lib/modules", INACCESSIBLE
, true },
115 /* ProtectKernelLogs= option */
116 static const MountEntry protect_kernel_logs_table
[] = {
117 { "/proc/kmsg", INACCESSIBLE
, true },
118 { "/dev/kmsg", INACCESSIBLE
, true },
122 * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of
123 * system should be protected by ProtectSystem=
125 static const MountEntry protect_home_read_only_table
[] = {
126 { "/home", READONLY
, true },
127 { "/run/user", READONLY
, true },
128 { "/root", READONLY
, true },
131 /* ProtectHome=tmpfs table */
132 static const MountEntry protect_home_tmpfs_table
[] = {
133 { "/home", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755", .flags
= MS_NODEV
|MS_STRICTATIME
},
134 { "/run/user", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755", .flags
= MS_NODEV
|MS_STRICTATIME
},
135 { "/root", TMPFS
, true, .read_only
= true, .options_const
= "mode=0700", .flags
= MS_NODEV
|MS_STRICTATIME
},
138 /* ProtectHome=yes table */
139 static const MountEntry protect_home_yes_table
[] = {
140 { "/home", INACCESSIBLE
, true },
141 { "/run/user", INACCESSIBLE
, true },
142 { "/root", INACCESSIBLE
, true },
145 /* ProtectSystem=yes table */
146 static const MountEntry protect_system_yes_table
[] = {
147 { "/usr", READONLY
, false },
148 { "/boot", READONLY
, true },
149 { "/efi", READONLY
, true },
151 { "/lib", READONLY
, true },
152 { "/lib64", READONLY
, true },
153 { "/bin", READONLY
, true },
155 { "/sbin", READONLY
, true },
160 /* ProtectSystem=full includes ProtectSystem=yes */
161 static const MountEntry protect_system_full_table
[] = {
162 { "/usr", READONLY
, false },
163 { "/boot", READONLY
, true },
164 { "/efi", READONLY
, true },
165 { "/etc", READONLY
, false },
167 { "/lib", READONLY
, true },
168 { "/lib64", READONLY
, true },
169 { "/bin", READONLY
, true },
171 { "/sbin", READONLY
, true },
177 * ProtectSystem=strict table. In this strict mode, we mount everything
178 * read-only, except for /proc, /dev, /sys which are the kernel API VFS,
179 * which are left writable, but PrivateDevices= + ProtectKernelTunables=
180 * protect those, and these options should be fully orthogonal.
181 * (And of course /home and friends are also left writable, as ProtectHome=
182 * shall manage those, orthogonally).
184 static const MountEntry protect_system_strict_table
[] = {
185 { "/", READONLY
, false },
186 { "/proc", READWRITE_IMPLICIT
, false }, /* ProtectKernelTunables= */
187 { "/sys", READWRITE_IMPLICIT
, false }, /* ProtectKernelTunables= */
188 { "/dev", READWRITE_IMPLICIT
, false }, /* PrivateDevices= */
189 { "/home", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
190 { "/run/user", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
191 { "/root", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
194 static const char * const mount_mode_table
[_MOUNT_MODE_MAX
] = {
195 [INACCESSIBLE
] = "inaccessible",
196 [BIND_MOUNT
] = "bind",
197 [BIND_MOUNT_RECURSIVE
] = "rbind",
198 [PRIVATE_TMP
] = "private-tmp",
199 [PRIVATE_DEV
] = "private-dev",
200 [BIND_DEV
] = "bind-dev",
201 [EMPTY_DIR
] = "empty",
204 [READONLY
] = "read-only",
205 [READWRITE
] = "read-write",
207 [READWRITE_IMPLICIT
] = "rw-implicit",
210 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(mount_mode
, MountMode
);
212 static const char *mount_entry_path(const MountEntry
*p
) {
215 /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that,
216 * otherwise the stack/static ->path field is returned. */
218 return p
->path_malloc
?: p
->path_const
;
221 static bool mount_entry_read_only(const MountEntry
*p
) {
224 return p
->read_only
|| IN_SET(p
->mode
, READONLY
, INACCESSIBLE
);
227 static const char *mount_entry_source(const MountEntry
*p
) {
230 return p
->source_malloc
?: p
->source_const
;
233 static const char *mount_entry_options(const MountEntry
*p
) {
236 return p
->options_malloc
?: p
->options_const
;
239 static void mount_entry_done(MountEntry
*p
) {
242 p
->path_malloc
= mfree(p
->path_malloc
);
243 p
->source_malloc
= mfree(p
->source_malloc
);
244 p
->options_malloc
= mfree(p
->options_malloc
);
247 static int append_access_mounts(MountEntry
**p
, char **strv
, MountMode mode
, bool forcibly_require_prefix
) {
252 /* Adds a list of user-supplied READWRITE/READWRITE_IMPLICIT/READONLY/INACCESSIBLE entries */
254 STRV_FOREACH(i
, strv
) {
255 bool ignore
= false, needs_prefix
= false;
258 /* Look for any prefixes */
259 if (startswith(e
, "-")) {
263 if (startswith(e
, "+")) {
268 if (!path_is_absolute(e
))
269 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
270 "Path is not absolute: %s", e
);
272 *((*p
)++) = (MountEntry
) {
276 .has_prefix
= !needs_prefix
&& !forcibly_require_prefix
,
283 static int append_empty_dir_mounts(MountEntry
**p
, char **strv
) {
288 /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the
289 * "/private/" boundary directories for DynamicUser=1. */
291 STRV_FOREACH(i
, strv
) {
293 *((*p
)++) = (MountEntry
) {
298 .options_const
= "mode=755",
299 .flags
= MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
,
306 static int append_bind_mounts(MountEntry
**p
, const BindMount
*binds
, size_t n
) {
311 for (i
= 0; i
< n
; i
++) {
312 const BindMount
*b
= binds
+ i
;
314 *((*p
)++) = (MountEntry
) {
315 .path_const
= b
->destination
,
316 .mode
= b
->recursive
? BIND_MOUNT_RECURSIVE
: BIND_MOUNT
,
317 .read_only
= b
->read_only
,
319 .source_const
= b
->source
,
320 .ignore
= b
->ignore_enoent
,
327 static int append_tmpfs_mounts(MountEntry
**p
, const TemporaryFileSystem
*tmpfs
, size_t n
) {
333 for (i
= 0; i
< n
; i
++) {
334 const TemporaryFileSystem
*t
= tmpfs
+ i
;
335 _cleanup_free_
char *o
= NULL
, *str
= NULL
;
339 if (!path_is_absolute(t
->path
))
340 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
341 "Path is not absolute: %s",
344 str
= strjoin("mode=0755,", t
->options
);
348 r
= mount_option_mangle(str
, MS_NODEV
|MS_STRICTATIME
, &flags
, &o
);
350 return log_debug_errno(r
, "Failed to parse mount option '%s': %m", str
);
352 ro
= flags
& MS_RDONLY
;
356 *((*p
)++) = (MountEntry
) {
357 .path_const
= t
->path
,
360 .options_malloc
= TAKE_PTR(o
),
368 static int append_static_mounts(MountEntry
**p
, const MountEntry
*mounts
, size_t n
, bool ignore_protect
) {
374 /* Adds a list of static pre-defined entries */
376 for (i
= 0; i
< n
; i
++)
377 *((*p
)++) = (MountEntry
) {
378 .path_const
= mount_entry_path(mounts
+i
),
379 .mode
= mounts
[i
].mode
,
380 .ignore
= mounts
[i
].ignore
|| ignore_protect
,
386 static int append_protect_home(MountEntry
**p
, ProtectHome protect_home
, bool ignore_protect
) {
389 switch (protect_home
) {
391 case PROTECT_HOME_NO
:
394 case PROTECT_HOME_READ_ONLY
:
395 return append_static_mounts(p
, protect_home_read_only_table
, ELEMENTSOF(protect_home_read_only_table
), ignore_protect
);
397 case PROTECT_HOME_TMPFS
:
398 return append_static_mounts(p
, protect_home_tmpfs_table
, ELEMENTSOF(protect_home_tmpfs_table
), ignore_protect
);
400 case PROTECT_HOME_YES
:
401 return append_static_mounts(p
, protect_home_yes_table
, ELEMENTSOF(protect_home_yes_table
), ignore_protect
);
404 assert_not_reached("Unexpected ProtectHome= value");
408 static int append_protect_system(MountEntry
**p
, ProtectSystem protect_system
, bool ignore_protect
) {
411 switch (protect_system
) {
413 case PROTECT_SYSTEM_NO
:
416 case PROTECT_SYSTEM_STRICT
:
417 return append_static_mounts(p
, protect_system_strict_table
, ELEMENTSOF(protect_system_strict_table
), ignore_protect
);
419 case PROTECT_SYSTEM_YES
:
420 return append_static_mounts(p
, protect_system_yes_table
, ELEMENTSOF(protect_system_yes_table
), ignore_protect
);
422 case PROTECT_SYSTEM_FULL
:
423 return append_static_mounts(p
, protect_system_full_table
, ELEMENTSOF(protect_system_full_table
), ignore_protect
);
426 assert_not_reached("Unexpected ProtectSystem= value");
430 static int mount_path_compare(const MountEntry
*a
, const MountEntry
*b
) {
433 /* If the paths are not equal, then order prefixes first */
434 d
= path_compare(mount_entry_path(a
), mount_entry_path(b
));
438 /* If the paths are equal, check the mode */
439 return CMP((int) a
->mode
, (int) b
->mode
);
442 static int prefix_where_needed(MountEntry
*m
, size_t n
, const char *root_directory
) {
445 /* Prefixes all paths in the bind mount table with the root directory if the entry needs that. */
447 for (i
= 0; i
< n
; i
++) {
453 s
= path_join(root_directory
, mount_entry_path(m
+i
));
457 free_and_replace(m
[i
].path_malloc
, s
);
458 m
[i
].has_prefix
= true;
464 static void drop_duplicates(MountEntry
*m
, size_t *n
) {
465 MountEntry
*f
, *t
, *previous
;
470 /* Drops duplicate entries. Expects that the array is properly ordered already. */
472 for (f
= m
, t
= m
, previous
= NULL
; f
< m
+ *n
; f
++) {
474 /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare()
475 * above. Note that we only drop duplicates that haven't been mounted yet. */
477 path_equal(mount_entry_path(f
), mount_entry_path(previous
)) &&
478 !f
->applied
&& !previous
->applied
) {
479 log_debug("%s (%s) is duplicate.", mount_entry_path(f
), mount_mode_to_string(f
->mode
));
480 previous
->read_only
= previous
->read_only
|| mount_entry_read_only(f
); /* Propagate the read-only flag to the remaining entry */
493 static void drop_inaccessible(MountEntry
*m
, size_t *n
) {
495 const char *clear
= NULL
;
500 /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly
501 * ordered already. */
503 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
505 /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop
506 * it, as inaccessible paths really should drop the entire subtree. */
507 if (clear
&& path_startswith(mount_entry_path(f
), clear
)) {
508 log_debug("%s is masked by %s.", mount_entry_path(f
), clear
);
513 clear
= f
->mode
== INACCESSIBLE
? mount_entry_path(f
) : NULL
;
522 static void drop_nop(MountEntry
*m
, size_t *n
) {
528 /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the
529 * list is ordered by prefixes. */
531 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
533 /* Only suppress such subtrees for READONLY, READWRITE and READWRITE_IMPLICIT entries */
534 if (IN_SET(f
->mode
, READONLY
, READWRITE
, READWRITE_IMPLICIT
)) {
538 /* Now let's find the first parent of the entry we are looking at. */
539 for (p
= t
-1; p
>= m
; p
--) {
540 if (path_startswith(mount_entry_path(f
), mount_entry_path(p
))) {
546 /* We found it, let's see if it's the same mode, if so, we can drop this entry */
547 if (found
&& p
->mode
== f
->mode
) {
548 log_debug("%s (%s) is made redundant by %s (%s)",
549 mount_entry_path(f
), mount_mode_to_string(f
->mode
),
550 mount_entry_path(p
), mount_mode_to_string(p
->mode
));
563 static void drop_outside_root(const char *root_directory
, MountEntry
*m
, size_t *n
) {
573 /* Drops all mounts that are outside of the root directory. */
575 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
577 if (!path_startswith(mount_entry_path(f
), root_directory
)) {
578 log_debug("%s is outside of root directory.", mount_entry_path(f
));
590 static int clone_device_node(
592 const char *temporary_mount
,
593 bool *make_devnode
) {
595 _cleanup_free_
char *sl
= NULL
;
596 const char *dn
, *bn
, *t
;
600 if (stat(d
, &st
) < 0) {
601 if (errno
== ENOENT
) {
602 log_debug_errno(errno
, "Device node '%s' to clone does not exist, ignoring.", d
);
606 return log_debug_errno(errno
, "Failed to stat() device node '%s' to clone, ignoring: %m", d
);
609 if (!S_ISBLK(st
.st_mode
) &&
610 !S_ISCHR(st
.st_mode
))
611 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
612 "Device node '%s' to clone is not a device node, ignoring.",
615 dn
= strjoina(temporary_mount
, d
);
617 /* First, try to create device node properly */
619 mac_selinux_create_file_prepare(d
, st
.st_mode
);
620 r
= mknod(dn
, st
.st_mode
, st
.st_rdev
);
621 mac_selinux_create_file_clear();
625 return log_debug_errno(errno
, "mknod failed for %s: %m", d
);
627 /* This didn't work, let's not try this again for the next iterations. */
628 *make_devnode
= false;
631 /* We're about to fallback to bind-mounting the device
632 * node. So create a dummy bind-mount target. */
633 mac_selinux_create_file_prepare(d
, 0);
634 r
= mknod(dn
, S_IFREG
, 0);
635 mac_selinux_create_file_clear();
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", temporary_mount
, S_ISCHR(st
.st_mode
) ? "char" : "block", major(st
.st_rdev
), minor(st
.st_rdev
)) < 0)
656 (void) mkdir_parents(sl
, 0755);
658 t
= strjoina("../", bn
);
660 if (symlink(t
, sl
) < 0)
661 log_debug_errno(errno
, "Failed to symlink '%s' to '%s', ignoring: %m", t
, sl
);
666 static int mount_private_dev(MountEntry
*m
) {
667 static const char devnodes
[] =
675 char temporary_mount
[] = "/tmp/namespace-dev-XXXXXX";
676 const char *d
, *dev
= NULL
, *devpts
= NULL
, *devshm
= NULL
, *devhugepages
= NULL
, *devmqueue
= NULL
, *devlog
= NULL
, *devptmx
= NULL
;
677 bool can_mknod
= true;
678 _cleanup_umask_ mode_t u
;
685 if (!mkdtemp(temporary_mount
))
686 return log_debug_errno(errno
, "Failed to create temporary directory '%s': %m", temporary_mount
);
688 dev
= strjoina(temporary_mount
, "/dev");
689 (void) mkdir(dev
, 0755);
690 if (mount("tmpfs", dev
, "tmpfs", DEV_MOUNT_OPTIONS
, "mode=755") < 0) {
691 r
= log_debug_errno(errno
, "Failed to mount tmpfs on '%s': %m", dev
);
695 devpts
= strjoina(temporary_mount
, "/dev/pts");
696 (void) mkdir(devpts
, 0755);
697 if (mount("/dev/pts", devpts
, NULL
, MS_BIND
, NULL
) < 0) {
698 r
= log_debug_errno(errno
, "Failed to bind mount /dev/pts on '%s': %m", devpts
);
702 /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx.
703 * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible.
704 * Thus, in that case make a clone.
705 * In nspawn and other containers it will be a symlink, in that case make it a symlink. */
706 r
= is_symlink("/dev/ptmx");
708 log_debug_errno(r
, "Failed to detect whether /dev/ptmx is a symlink or not: %m");
711 devptmx
= strjoina(temporary_mount
, "/dev/ptmx");
712 if (symlink("pts/ptmx", devptmx
) < 0) {
713 r
= log_debug_errno(errno
, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx
);
717 r
= clone_device_node("/dev/ptmx", temporary_mount
, &can_mknod
);
722 devshm
= strjoina(temporary_mount
, "/dev/shm");
723 (void) mkdir(devshm
, 0755);
724 r
= mount("/dev/shm", devshm
, NULL
, MS_BIND
, NULL
);
726 r
= log_debug_errno(errno
, "Failed to bind mount /dev/shm on '%s': %m", devshm
);
730 devmqueue
= strjoina(temporary_mount
, "/dev/mqueue");
731 (void) mkdir(devmqueue
, 0755);
732 if (mount("/dev/mqueue", devmqueue
, NULL
, MS_BIND
, NULL
) < 0)
733 log_debug_errno(errno
, "Failed to bind mount /dev/mqueue on '%s', ignoring: %m", devmqueue
);
735 devhugepages
= strjoina(temporary_mount
, "/dev/hugepages");
736 (void) mkdir(devhugepages
, 0755);
737 if (mount("/dev/hugepages", devhugepages
, NULL
, MS_BIND
, NULL
) < 0)
738 log_debug_errno(errno
, "Failed to bind mount /dev/hugepages on '%s', ignoring: %m", devhugepages
);
740 devlog
= strjoina(temporary_mount
, "/dev/log");
741 if (symlink("/run/systemd/journal/dev-log", devlog
) < 0)
742 log_debug_errno(errno
, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog
);
744 NULSTR_FOREACH(d
, devnodes
) {
745 r
= clone_device_node(d
, temporary_mount
, &can_mknod
);
746 /* ENXIO means the the *source* is not a device file, skip creation in that case */
747 if (r
< 0 && r
!= -ENXIO
)
751 r
= dev_setup(temporary_mount
, UID_INVALID
, GID_INVALID
);
753 log_debug_errno(r
, "Failed to setup basic device tree at '%s', ignoring: %m", temporary_mount
);
755 /* Create the /dev directory if missing. It is more likely to be
756 * missing when the service is started with RootDirectory. This is
757 * consistent with mount units creating the mount points when missing.
759 (void) mkdir_p_label(mount_entry_path(m
), 0755);
761 /* Unmount everything in old /dev */
762 r
= umount_recursive(mount_entry_path(m
), 0);
764 log_debug_errno(r
, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m
));
766 if (mount(dev
, mount_entry_path(m
), NULL
, MS_MOVE
, NULL
) < 0) {
767 r
= log_debug_errno(errno
, "Failed to move mount point '%s' to '%s': %m", dev
, mount_entry_path(m
));
772 (void) rmdir(temporary_mount
);
778 (void) umount(devpts
);
781 (void) umount(devshm
);
784 (void) umount(devhugepages
);
787 (void) umount(devmqueue
);
791 (void) rmdir(temporary_mount
);
796 static int mount_bind_dev(const MountEntry
*m
) {
801 /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the service's
802 * /dev. This is only used when RootDirectory= is set. */
804 (void) mkdir_p_label(mount_entry_path(m
), 0755);
806 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
808 return log_debug_errno(r
, "Unable to determine whether /dev is already mounted: %m");
809 if (r
> 0) /* make this a NOP if /dev is already a mount point */
812 if (mount("/dev", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
813 return log_debug_errno(errno
, "Failed to bind mount %s: %m", mount_entry_path(m
));
818 static int mount_sysfs(const MountEntry
*m
) {
823 (void) mkdir_p_label(mount_entry_path(m
), 0755);
825 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
827 return log_debug_errno(r
, "Unable to determine whether /sys is already mounted: %m");
828 if (r
> 0) /* make this a NOP if /sys is already a mount point */
831 /* Bind mount the host's version so that we get all child mounts of it, too. */
832 if (mount("/sys", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
833 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
838 static int mount_procfs(const MountEntry
*m
) {
843 (void) mkdir_p_label(mount_entry_path(m
), 0755);
845 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
847 return log_debug_errno(r
, "Unable to determine whether /proc is already mounted: %m");
848 if (r
> 0) /* make this a NOP if /proc is already a mount point */
851 /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in one */
852 if (mount("proc", mount_entry_path(m
), "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
) < 0)
853 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
858 static int mount_tmpfs(const MountEntry
*m
) {
861 /* First, get rid of everything that is below if there is anything. Then, overmount with our new tmpfs */
863 (void) mkdir_p_label(mount_entry_path(m
), 0755);
864 (void) umount_recursive(mount_entry_path(m
), 0);
866 if (mount("tmpfs", mount_entry_path(m
), "tmpfs", m
->flags
, mount_entry_options(m
)) < 0)
867 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
872 static int follow_symlink(
873 const char *root_directory
,
876 _cleanup_free_
char *target
= NULL
;
879 /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we
880 * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at
881 * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the
882 * end and already have a fully normalized name. */
884 r
= chase_symlinks(mount_entry_path(m
), root_directory
, CHASE_STEP
|CHASE_NONEXISTENT
, &target
, NULL
);
886 return log_debug_errno(r
, "Failed to chase symlinks '%s': %m", mount_entry_path(m
));
887 if (r
> 0) /* Reached the end, nothing more to resolve */
890 if (m
->n_followed
>= CHASE_SYMLINKS_MAX
) /* put a boundary on things */
891 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
892 "Symlink loop on '%s'.",
893 mount_entry_path(m
));
895 log_debug("Followed mount entry path symlink %s → %s.", mount_entry_path(m
), target
);
897 free_and_replace(m
->path_malloc
, target
);
898 m
->has_prefix
= true;
905 static int apply_mount(
906 const char *root_directory
,
909 _cleanup_free_
char *inaccessible
= NULL
;
910 bool rbind
= true, make
= false;
916 log_debug("Applying namespace mount on %s", mount_entry_path(m
));
921 _cleanup_free_
char *tmp
= NULL
;
922 const char *runtime_dir
;
925 /* First, get rid of everything that is below if there
926 * is anything... Then, overmount it with an
927 * inaccessible path. */
928 (void) umount_recursive(mount_entry_path(m
), 0);
930 if (lstat(mount_entry_path(m
), &target
) < 0) {
931 if (errno
== ENOENT
&& m
->ignore
)
934 return log_debug_errno(errno
, "Failed to lstat() %s to determine what to mount over it: %m", mount_entry_path(m
));
938 runtime_dir
= "/run/systemd";
940 if (asprintf(&tmp
, "/run/user/"UID_FMT
, geteuid()) < 0)
946 r
= mode_to_inaccessible_node(runtime_dir
, target
.st_mode
, &inaccessible
);
948 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
949 "File type not supported for inaccessible mounts. Note that symlinks are not allowed");
956 case READWRITE_IMPLICIT
:
957 r
= path_is_mount_point(mount_entry_path(m
), root_directory
, 0);
958 if (r
== -ENOENT
&& m
->ignore
)
961 return log_debug_errno(r
, "Failed to determine whether %s is already a mount point: %m", mount_entry_path(m
));
962 if (r
> 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY bit for the mount point if needed. */
964 /* This isn't a mount point yet, let's make it one. */
965 what
= mount_entry_path(m
);
972 case BIND_MOUNT_RECURSIVE
: {
973 _cleanup_free_
char *chased
= NULL
;
975 /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note that bind
976 * mount source paths are always relative to the host root, hence we pass NULL as root directory to
977 * chase_symlinks() here. */
979 r
= chase_symlinks(mount_entry_source(m
), NULL
, CHASE_TRAIL_SLASH
, &chased
, NULL
);
980 if (r
== -ENOENT
&& m
->ignore
) {
981 log_debug_errno(r
, "Path %s does not exist, ignoring.", mount_entry_source(m
));
985 return log_debug_errno(r
, "Failed to follow symlinks on %s: %m", mount_entry_source(m
));
987 log_debug("Followed source symlinks %s → %s.", mount_entry_source(m
), chased
);
989 free_and_replace(m
->source_malloc
, chased
);
991 what
= mount_entry_source(m
);
998 return mount_tmpfs(m
);
1001 what
= mount_entry_source(m
);
1006 return mount_private_dev(m
);
1009 return mount_bind_dev(m
);
1012 return mount_sysfs(m
);
1015 return mount_procfs(m
);
1018 assert_not_reached("Unknown mode");
1023 if (mount(what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
) < 0) {
1024 bool try_again
= false;
1027 if (r
== -ENOENT
&& make
) {
1030 /* Hmm, either the source or the destination are missing. Let's see if we can create the destination, then try again */
1032 if (stat(what
, &st
) < 0)
1033 log_debug_errno(errno
, "Mount point source '%s' is not accessible: %m", what
);
1037 (void) mkdir_parents(mount_entry_path(m
), 0755);
1039 if (S_ISDIR(st
.st_mode
))
1040 q
= mkdir(mount_entry_path(m
), 0755) < 0 ? -errno
: 0;
1042 q
= touch(mount_entry_path(m
));
1045 log_debug_errno(q
, "Failed to create destination mount point node '%s': %m", mount_entry_path(m
));
1052 if (mount(what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
) < 0)
1059 return log_debug_errno(r
, "Failed to mount %s to %s: %m", what
, mount_entry_path(m
));
1062 log_debug("Successfully mounted %s to %s", what
, mount_entry_path(m
));
1066 /* Change per-mount flags on an existing mount */
1067 static int bind_remount_one(const char *path
, unsigned long orig_flags
, unsigned long new_flags
, unsigned long flags_mask
) {
1068 if (mount(NULL
, path
, NULL
, (orig_flags
& ~flags_mask
) | MS_REMOUNT
| MS_BIND
| new_flags
, NULL
) < 0)
1074 static int make_read_only(const MountEntry
*m
, char **blacklist
, FILE *proc_self_mountinfo
) {
1075 unsigned long new_flags
= 0, flags_mask
= 0;
1076 bool submounts
= false;
1080 assert(proc_self_mountinfo
);
1082 if (mount_entry_read_only(m
) || m
->mode
== PRIVATE_DEV
) {
1083 new_flags
|= MS_RDONLY
;
1084 flags_mask
|= MS_RDONLY
;
1088 new_flags
|= MS_NOSUID
;
1089 flags_mask
|= MS_NOSUID
;
1092 if (flags_mask
== 0) /* No Change? */
1095 /* We generally apply these changes recursively, except for /dev, and the cases we know there's
1096 * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the
1097 * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace
1098 * and running Linux <= 4.17. */
1100 mount_entry_read_only(m
) &&
1101 !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1103 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, blacklist
, proc_self_mountinfo
);
1105 r
= bind_remount_one(mount_entry_path(m
), m
->flags
, new_flags
, flags_mask
);
1107 /* Not that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked
1108 * read-only already stays this way. This improves compatibility with container managers, where we
1109 * won't attempt to undo read-only mounts already applied. */
1111 if (r
== -ENOENT
&& m
->ignore
)
1114 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1115 submounts
? " and its submounts" : "");
1119 static bool namespace_info_mount_apivfs(const NamespaceInfo
*ns_info
) {
1123 * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
1124 * since to protect the API VFS mounts, they need to be around in the
1128 return ns_info
->mount_apivfs
||
1129 ns_info
->protect_control_groups
||
1130 ns_info
->protect_kernel_tunables
;
1133 static size_t namespace_calculate_mounts(
1134 const NamespaceInfo
*ns_info
,
1135 char** read_write_paths
,
1136 char** read_only_paths
,
1137 char** inaccessible_paths
,
1138 char** empty_directories
,
1139 size_t n_bind_mounts
,
1140 size_t n_temporary_filesystems
,
1141 const char* tmp_dir
,
1142 const char* var_tmp_dir
,
1143 ProtectHome protect_home
,
1144 ProtectSystem protect_system
) {
1146 size_t protect_home_cnt
;
1147 size_t protect_system_cnt
=
1148 (protect_system
== PROTECT_SYSTEM_STRICT
?
1149 ELEMENTSOF(protect_system_strict_table
) :
1150 ((protect_system
== PROTECT_SYSTEM_FULL
) ?
1151 ELEMENTSOF(protect_system_full_table
) :
1152 ((protect_system
== PROTECT_SYSTEM_YES
) ?
1153 ELEMENTSOF(protect_system_yes_table
) : 0)));
1156 (protect_home
== PROTECT_HOME_YES
?
1157 ELEMENTSOF(protect_home_yes_table
) :
1158 ((protect_home
== PROTECT_HOME_READ_ONLY
) ?
1159 ELEMENTSOF(protect_home_read_only_table
) :
1160 ((protect_home
== PROTECT_HOME_TMPFS
) ?
1161 ELEMENTSOF(protect_home_tmpfs_table
) : 0)));
1163 return !!tmp_dir
+ !!var_tmp_dir
+
1164 strv_length(read_write_paths
) +
1165 strv_length(read_only_paths
) +
1166 strv_length(inaccessible_paths
) +
1167 strv_length(empty_directories
) +
1169 n_temporary_filesystems
+
1170 ns_info
->private_dev
+
1171 (ns_info
->protect_kernel_tunables
? ELEMENTSOF(protect_kernel_tunables_table
) : 0) +
1172 (ns_info
->protect_kernel_modules
? ELEMENTSOF(protect_kernel_modules_table
) : 0) +
1173 (ns_info
->protect_kernel_logs
? ELEMENTSOF(protect_kernel_logs_table
) : 0) +
1174 (ns_info
->protect_control_groups
? 1 : 0) +
1175 protect_home_cnt
+ protect_system_cnt
+
1176 (ns_info
->protect_hostname
? 2 : 0) +
1177 (namespace_info_mount_apivfs(ns_info
) ? ELEMENTSOF(apivfs_table
) : 0);
1180 static void normalize_mounts(const char *root_directory
, MountEntry
*mounts
, size_t *n_mounts
) {
1181 assert(root_directory
);
1183 assert(mounts
|| *n_mounts
== 0);
1185 typesafe_qsort(mounts
, *n_mounts
, mount_path_compare
);
1187 drop_duplicates(mounts
, n_mounts
);
1188 drop_outside_root(root_directory
, mounts
, n_mounts
);
1189 drop_inaccessible(mounts
, n_mounts
);
1190 drop_nop(mounts
, n_mounts
);
1193 static bool root_read_only(
1194 char **read_only_paths
,
1195 ProtectSystem protect_system
) {
1197 /* Determine whether the root directory is going to be read-only given the configured settings. */
1199 if (protect_system
== PROTECT_SYSTEM_STRICT
)
1202 if (path_strv_contains(read_only_paths
, "/"))
1208 static bool home_read_only(
1209 char** read_only_paths
,
1210 char** inaccessible_paths
,
1211 char** empty_directories
,
1212 const BindMount
*bind_mounts
,
1213 size_t n_bind_mounts
,
1214 const TemporaryFileSystem
*temporary_filesystems
,
1215 size_t n_temporary_filesystems
,
1216 ProtectHome protect_home
) {
1220 /* Determine whether the /home directory is going to be read-only given the configured settings. Yes,
1221 * this is a bit sloppy, since we don't bother checking for cases where / is affected by multiple
1224 if (protect_home
!= PROTECT_HOME_NO
)
1227 if (path_strv_contains(read_only_paths
, "/home") ||
1228 path_strv_contains(inaccessible_paths
, "/home") ||
1229 path_strv_contains(empty_directories
, "/home"))
1232 for (i
= 0; i
< n_temporary_filesystems
; i
++)
1233 if (path_equal(temporary_filesystems
[i
].path
, "/home"))
1236 /* If /home is overmounted with some dir from the host it's not writable. */
1237 for (i
= 0; i
< n_bind_mounts
; i
++)
1238 if (path_equal(bind_mounts
[i
].destination
, "/home"))
1244 int setup_namespace(
1245 const char* root_directory
,
1246 const char* root_image
,
1247 const NamespaceInfo
*ns_info
,
1248 char** read_write_paths
,
1249 char** read_only_paths
,
1250 char** inaccessible_paths
,
1251 char** empty_directories
,
1252 const BindMount
*bind_mounts
,
1253 size_t n_bind_mounts
,
1254 const TemporaryFileSystem
*temporary_filesystems
,
1255 size_t n_temporary_filesystems
,
1256 const char* tmp_dir
,
1257 const char* var_tmp_dir
,
1258 ProtectHome protect_home
,
1259 ProtectSystem protect_system
,
1260 unsigned long mount_flags
,
1261 DissectImageFlags dissect_image_flags
,
1262 char **error_path
) {
1264 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
1265 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
1266 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
1267 _cleanup_free_
void *root_hash
= NULL
;
1268 MountEntry
*m
= NULL
, *mounts
= NULL
;
1269 size_t n_mounts
, root_hash_size
= 0;
1270 bool require_prefix
= false;
1276 if (mount_flags
== 0)
1277 mount_flags
= MS_SHARED
;
1280 dissect_image_flags
|= DISSECT_IMAGE_REQUIRE_ROOT
;
1282 /* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */
1283 if (root_read_only(read_only_paths
,
1285 home_read_only(read_only_paths
, inaccessible_paths
, empty_directories
,
1286 bind_mounts
, n_bind_mounts
, temporary_filesystems
, n_temporary_filesystems
,
1288 strv_isempty(read_write_paths
))
1289 dissect_image_flags
|= DISSECT_IMAGE_READ_ONLY
;
1291 r
= loop_device_make_by_path(root_image
,
1292 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_READ_ONLY
) ? O_RDONLY
: -1 /* < 0 means writable if possible, read-only as fallback */,
1296 return log_debug_errno(r
, "Failed to create loop device for root image: %m");
1298 r
= root_hash_load(root_image
, &root_hash
, &root_hash_size
);
1300 return log_debug_errno(r
, "Failed to load root hash: %m");
1302 r
= dissect_image(loop_device
->fd
, root_hash
, root_hash_size
, dissect_image_flags
, &dissected_image
);
1304 return log_debug_errno(r
, "Failed to dissect image: %m");
1306 r
= dissected_image_decrypt(dissected_image
, NULL
, root_hash
, root_hash_size
, dissect_image_flags
, &decrypted_image
);
1308 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
1312 root
= root_directory
;
1314 /* Always create the mount namespace in a temporary directory, instead of operating
1315 * directly in the root. The temporary directory prevents any mounts from being
1316 * potentially obscured my other mounts we already applied.
1317 * We use the same mount point for all images, which is safe, since they all live
1318 * in their own namespaces after all, and hence won't see each other. */
1320 root
= "/run/systemd/unit-root";
1321 (void) mkdir_label(root
, 0700);
1322 require_prefix
= true;
1325 n_mounts
= namespace_calculate_mounts(
1332 n_temporary_filesystems
,
1333 tmp_dir
, var_tmp_dir
,
1334 protect_home
, protect_system
);
1337 m
= mounts
= new0(MountEntry
, n_mounts
);
1341 r
= append_access_mounts(&m
, read_write_paths
, READWRITE
, require_prefix
);
1345 r
= append_access_mounts(&m
, read_only_paths
, READONLY
, require_prefix
);
1349 r
= append_access_mounts(&m
, inaccessible_paths
, INACCESSIBLE
, require_prefix
);
1353 r
= append_empty_dir_mounts(&m
, empty_directories
);
1357 r
= append_bind_mounts(&m
, bind_mounts
, n_bind_mounts
);
1361 r
= append_tmpfs_mounts(&m
, temporary_filesystems
, n_temporary_filesystems
);
1366 *(m
++) = (MountEntry
) {
1367 .path_const
= "/tmp",
1368 .mode
= PRIVATE_TMP
,
1369 .source_const
= tmp_dir
,
1374 *(m
++) = (MountEntry
) {
1375 .path_const
= "/var/tmp",
1376 .mode
= PRIVATE_TMP
,
1377 .source_const
= var_tmp_dir
,
1381 if (ns_info
->private_dev
) {
1382 *(m
++) = (MountEntry
) {
1383 .path_const
= "/dev",
1384 .mode
= PRIVATE_DEV
,
1385 .flags
= DEV_MOUNT_OPTIONS
,
1389 if (ns_info
->protect_kernel_tunables
) {
1390 r
= append_static_mounts(&m
, protect_kernel_tunables_table
, ELEMENTSOF(protect_kernel_tunables_table
), ns_info
->ignore_protect_paths
);
1395 if (ns_info
->protect_kernel_modules
) {
1396 r
= append_static_mounts(&m
, protect_kernel_modules_table
, ELEMENTSOF(protect_kernel_modules_table
), ns_info
->ignore_protect_paths
);
1401 if (ns_info
->protect_kernel_logs
) {
1402 r
= append_static_mounts(&m
, protect_kernel_logs_table
, ELEMENTSOF(protect_kernel_logs_table
), ns_info
->ignore_protect_paths
);
1407 if (ns_info
->protect_control_groups
) {
1408 *(m
++) = (MountEntry
) {
1409 .path_const
= "/sys/fs/cgroup",
1414 r
= append_protect_home(&m
, protect_home
, ns_info
->ignore_protect_paths
);
1418 r
= append_protect_system(&m
, protect_system
, false);
1422 if (namespace_info_mount_apivfs(ns_info
)) {
1423 r
= append_static_mounts(&m
, apivfs_table
, ELEMENTSOF(apivfs_table
), ns_info
->ignore_protect_paths
);
1428 if (ns_info
->protect_hostname
) {
1429 *(m
++) = (MountEntry
) {
1430 .path_const
= "/proc/sys/kernel/hostname",
1433 *(m
++) = (MountEntry
) {
1434 .path_const
= "/proc/sys/kernel/domainname",
1439 assert(mounts
+ n_mounts
== m
);
1441 /* Prepend the root directory where that's necessary */
1442 r
= prefix_where_needed(mounts
, n_mounts
, root
);
1446 normalize_mounts(root
, mounts
, &n_mounts
);
1449 /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */
1451 if (unshare(CLONE_NEWNS
) < 0) {
1452 r
= log_debug_errno(errno
, "Failed to unshare the mount namespace: %m");
1453 if (IN_SET(r
, -EACCES
, -EPERM
, -EOPNOTSUPP
, -ENOSYS
))
1454 /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter in place
1455 * that doesn't allow us to create namespaces (or a missing cap), then propagate a recognizable
1456 * error back, which the caller can use to detect this case (and only this) and optionally
1457 * continue without namespacing applied. */
1463 /* Remount / as SLAVE so that nothing now mounted in the namespace
1464 * shows up in the parent */
1465 if (mount(NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
) < 0) {
1466 r
= log_debug_errno(errno
, "Failed to remount '/' as SLAVE: %m");
1471 /* A root image is specified, mount it to the right place */
1472 r
= dissected_image_mount(dissected_image
, root
, UID_INVALID
, dissect_image_flags
);
1474 log_debug_errno(r
, "Failed to mount root image: %m");
1478 if (decrypted_image
) {
1479 r
= decrypted_image_relinquish(decrypted_image
);
1481 log_debug_errno(r
, "Failed to relinquish decrypted image: %m");
1486 loop_device_relinquish(loop_device
);
1488 } else if (root_directory
) {
1490 /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
1491 r
= path_is_mount_point(root
, NULL
, AT_SYMLINK_FOLLOW
);
1493 log_debug_errno(r
, "Failed to detect that %s is a mount point or not: %m", root
);
1497 if (mount(root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0) {
1498 r
= log_debug_errno(errno
, "Failed to bind mount '%s': %m", root
);
1505 /* Let's mount the main root directory to the root directory to use */
1506 if (mount("/", root
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0) {
1507 r
= log_debug_errno(errno
, "Failed to bind mount '/' on '%s': %m", root
);
1512 /* Try to set up the new root directory before mounting anything else there. */
1513 if (root_image
|| root_directory
)
1514 (void) base_filesystem_create(root
, UID_INVALID
, GID_INVALID
);
1517 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
1518 _cleanup_free_
char **blacklist
= NULL
;
1521 /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of /proc.
1522 * For example, this is the case with the option: 'InaccessiblePaths=/proc' */
1523 proc_self_mountinfo
= fopen("/proc/self/mountinfo", "re");
1524 if (!proc_self_mountinfo
) {
1525 r
= log_debug_errno(errno
, "Failed to open /proc/self/mountinfo: %m");
1527 *error_path
= strdup("/proc/self/mountinfo");
1531 /* First round, establish all mounts we need */
1535 for (m
= mounts
; m
< mounts
+ n_mounts
; ++m
) {
1540 r
= follow_symlink(root
, m
);
1542 if (error_path
&& mount_entry_path(m
))
1543 *error_path
= strdup(mount_entry_path(m
));
1547 /* We hit a symlinked mount point. The entry got rewritten and might point to a
1548 * very different place now. Let's normalize the changed list, and start from
1549 * the beginning. After all to mount the entry at the new location we might
1550 * need some other mounts first */
1555 r
= apply_mount(root
, m
);
1557 if (error_path
&& mount_entry_path(m
))
1558 *error_path
= strdup(mount_entry_path(m
));
1568 normalize_mounts(root
, mounts
, &n_mounts
);
1571 /* Create a blacklist we can pass to bind_mount_recursive() */
1572 blacklist
= new(char*, n_mounts
+1);
1577 for (j
= 0; j
< n_mounts
; j
++)
1578 blacklist
[j
] = (char*) mount_entry_path(mounts
+j
);
1579 blacklist
[j
] = NULL
;
1581 /* Second round, flip the ro bits if necessary. */
1582 for (m
= mounts
; m
< mounts
+ n_mounts
; ++m
) {
1583 r
= make_read_only(m
, blacklist
, proc_self_mountinfo
);
1585 if (error_path
&& mount_entry_path(m
))
1586 *error_path
= strdup(mount_entry_path(m
));
1592 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
1593 r
= mount_move_root(root
);
1595 log_debug_errno(r
, "Failed to mount root with MS_MOVE: %m");
1599 /* Remount / as the desired mode. Note that this will not
1600 * reestablish propagation from our side to the host, since
1601 * what's disconnected is disconnected. */
1602 if (mount(NULL
, "/", NULL
, mount_flags
| MS_REC
, NULL
) < 0) {
1603 r
= log_debug_errno(errno
, "Failed to remount '/' with desired mount flags: %m");
1610 for (m
= mounts
; m
< mounts
+ n_mounts
; m
++)
1611 mount_entry_done(m
);
1618 void bind_mount_free_many(BindMount
*b
, size_t n
) {
1621 assert(b
|| n
== 0);
1623 for (i
= 0; i
< n
; i
++) {
1625 free(b
[i
].destination
);
1631 int bind_mount_add(BindMount
**b
, size_t *n
, const BindMount
*item
) {
1632 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
1639 s
= strdup(item
->source
);
1643 d
= strdup(item
->destination
);
1647 c
= reallocarray(*b
, *n
+ 1, sizeof(BindMount
));
1653 c
[(*n
) ++] = (BindMount
) {
1654 .source
= TAKE_PTR(s
),
1655 .destination
= TAKE_PTR(d
),
1656 .read_only
= item
->read_only
,
1657 .nosuid
= item
->nosuid
,
1658 .recursive
= item
->recursive
,
1659 .ignore_enoent
= item
->ignore_enoent
,
1665 void temporary_filesystem_free_many(TemporaryFileSystem
*t
, size_t n
) {
1668 assert(t
|| n
== 0);
1670 for (i
= 0; i
< n
; i
++) {
1678 int temporary_filesystem_add(
1679 TemporaryFileSystem
**t
,
1682 const char *options
) {
1684 _cleanup_free_
char *p
= NULL
, *o
= NULL
;
1685 TemporaryFileSystem
*c
;
1695 if (!isempty(options
)) {
1696 o
= strdup(options
);
1701 c
= reallocarray(*t
, *n
+ 1, sizeof(TemporaryFileSystem
));
1707 c
[(*n
) ++] = (TemporaryFileSystem
) {
1708 .path
= TAKE_PTR(p
),
1709 .options
= TAKE_PTR(o
),
1715 static int make_tmp_prefix(const char *prefix
) {
1716 _cleanup_free_
char *t
= NULL
;
1719 /* Don't do anything unless we know the dir is actually missing */
1720 r
= access(prefix
, F_OK
);
1723 if (errno
!= ENOENT
)
1726 r
= mkdir_parents(prefix
, 0755);
1730 r
= tempfn_random(prefix
, NULL
, &t
);
1734 if (mkdir(t
, 0777) < 0)
1737 if (chmod(t
, 01777) < 0) {
1743 if (rename(t
, prefix
) < 0) {
1746 return r
== -EEXIST
? 0 : r
; /* it's fine if someone else created the dir by now */
1753 static int setup_one_tmp_dir(const char *id
, const char *prefix
, char **path
) {
1754 _cleanup_free_
char *x
= NULL
;
1755 char bid
[SD_ID128_STRING_MAX
];
1763 /* We include the boot id in the directory so that after a
1764 * reboot we can easily identify obsolete directories. */
1766 r
= sd_id128_get_boot(&boot_id
);
1770 x
= strjoin(prefix
, "/systemd-private-", sd_id128_to_string(boot_id
, bid
), "-", id
, "-XXXXXX");
1774 r
= make_tmp_prefix(prefix
);
1778 RUN_WITH_UMASK(0077)
1782 RUN_WITH_UMASK(0000) {
1785 y
= strjoina(x
, "/tmp");
1787 if (mkdir(y
, 0777 | S_ISVTX
) < 0)
1791 *path
= TAKE_PTR(x
);
1796 int setup_tmp_dirs(const char *id
, char **tmp_dir
, char **var_tmp_dir
) {
1802 assert(var_tmp_dir
);
1804 r
= setup_one_tmp_dir(id
, "/tmp", &a
);
1808 r
= setup_one_tmp_dir(id
, "/var/tmp", &b
);
1812 t
= strjoina(a
, "/tmp");
1826 int setup_netns(const int netns_storage_socket
[static 2]) {
1827 _cleanup_close_
int netns
= -1;
1830 assert(netns_storage_socket
);
1831 assert(netns_storage_socket
[0] >= 0);
1832 assert(netns_storage_socket
[1] >= 0);
1834 /* We use the passed socketpair as a storage buffer for our
1835 * namespace reference fd. Whatever process runs this first
1836 * shall create a new namespace, all others should just join
1837 * it. To serialize that we use a file lock on the socket
1840 * It's a bit crazy, but hey, works great! */
1842 if (lockf(netns_storage_socket
[0], F_LOCK
, 0) < 0)
1845 netns
= receive_one_fd(netns_storage_socket
[0], MSG_DONTWAIT
);
1846 if (netns
== -EAGAIN
) {
1847 /* Nothing stored yet, so let's create a new namespace. */
1849 if (unshare(CLONE_NEWNET
) < 0) {
1854 (void) loopback_setup();
1856 netns
= open("/proc/self/ns/net", O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
1864 } else if (netns
< 0) {
1869 /* Yay, found something, so let's join the namespace */
1870 if (setns(netns
, CLONE_NEWNET
) < 0) {
1878 q
= send_one_fd(netns_storage_socket
[1], netns
, MSG_DONTWAIT
);
1885 (void) lockf(netns_storage_socket
[0], F_ULOCK
, 0);
1889 int open_netns_path(const int netns_storage_socket
[static 2], const char *path
) {
1890 _cleanup_close_
int netns
= -1;
1893 assert(netns_storage_socket
);
1894 assert(netns_storage_socket
[0] >= 0);
1895 assert(netns_storage_socket
[1] >= 0);
1898 /* If the storage socket doesn't contain a netns fd yet, open one via the file system and store it in
1899 * it. This is supposed to be called ahead of time, i.e. before setup_netns() which will allocate a
1900 * new anonymous netns if needed. */
1902 if (lockf(netns_storage_socket
[0], F_LOCK
, 0) < 0)
1905 netns
= receive_one_fd(netns_storage_socket
[0], MSG_DONTWAIT
);
1906 if (netns
== -EAGAIN
) {
1907 /* Nothing stored yet. Open the file from the file system. */
1909 netns
= open(path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
1915 r
= fd_is_network_ns(netns
);
1916 if (r
== 0) { /* Not a netns? Refuse early. */
1920 if (r
< 0 && r
!= -EUCLEAN
) /* EUCLEAN: we don't know */
1925 } else if (netns
< 0) {
1929 r
= 0; /* Already allocated */
1931 q
= send_one_fd(netns_storage_socket
[1], netns
, MSG_DONTWAIT
);
1938 (void) lockf(netns_storage_socket
[0], F_ULOCK
, 0);
1942 bool ns_type_supported(NamespaceType type
) {
1943 const char *t
, *ns_proc
;
1945 t
= namespace_type_to_string(type
);
1946 if (!t
) /* Don't know how to translate this? Then it's not supported */
1949 ns_proc
= strjoina("/proc/self/ns/", t
);
1950 return access(ns_proc
, F_OK
) == 0;
1953 static const char *const protect_home_table
[_PROTECT_HOME_MAX
] = {
1954 [PROTECT_HOME_NO
] = "no",
1955 [PROTECT_HOME_YES
] = "yes",
1956 [PROTECT_HOME_READ_ONLY
] = "read-only",
1957 [PROTECT_HOME_TMPFS
] = "tmpfs",
1960 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home
, ProtectHome
, PROTECT_HOME_YES
);
1962 static const char *const protect_system_table
[_PROTECT_SYSTEM_MAX
] = {
1963 [PROTECT_SYSTEM_NO
] = "no",
1964 [PROTECT_SYSTEM_YES
] = "yes",
1965 [PROTECT_SYSTEM_FULL
] = "full",
1966 [PROTECT_SYSTEM_STRICT
] = "strict",
1969 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system
, ProtectSystem
, PROTECT_SYSTEM_YES
);
1971 static const char* const namespace_type_table
[] = {
1972 [NAMESPACE_MOUNT
] = "mnt",
1973 [NAMESPACE_CGROUP
] = "cgroup",
1974 [NAMESPACE_UTS
] = "uts",
1975 [NAMESPACE_IPC
] = "ipc",
1976 [NAMESPACE_USER
] = "user",
1977 [NAMESPACE_PID
] = "pid",
1978 [NAMESPACE_NET
] = "net",
1981 DEFINE_STRING_TABLE_LOOKUP(namespace_type
, NamespaceType
);