1 /* SPDX-License-Identifier: LGPL-2.1+ */
12 #include "alloc-util.h"
13 #include "base-filesystem.h"
14 #include "dev-setup.h"
18 #include "loop-util.h"
19 #include "loopback-setup.h"
22 #include "mount-util.h"
23 #include "namespace.h"
24 #include "path-util.h"
25 #include "selinux-util.h"
26 #include "socket-util.h"
27 #include "stat-util.h"
28 #include "string-table.h"
29 #include "string-util.h"
31 #include "umask-util.h"
32 #include "user-util.h"
35 #define DEV_MOUNT_OPTIONS (MS_NOSUID|MS_STRICTATIME|MS_NOEXEC)
37 typedef enum MountMode
{
38 /* This is ordered by priority! */
53 typedef struct MountEntry
{
54 const char *path_const
; /* Memory allocated on stack or static */
56 bool ignore
:1; /* Ignore if path does not exist? */
57 bool has_prefix
:1; /* Already is prefixed by the root dir? */
58 bool read_only
:1; /* Shall this mount point be read-only? */
59 bool applied
:1; /* Already applied */
60 char *path_malloc
; /* Use this instead of 'path_const' if we had to allocate memory */
61 const char *source_const
; /* The source path, for bind mounts */
63 const char *options_const
;/* Mount options for tmpfs */
65 unsigned long flags
; /* Mount flags used by EMPTY_DIR and TMPFS. Do not include MS_RDONLY here, but please use read_only. */
69 /* If MountAPIVFS= is used, let's mount /sys and /proc into the it, but only as a fallback if the user hasn't mounted
70 * something there already. These mounts are hence overridden by any other explicitly configured mounts. */
71 static const MountEntry apivfs_table
[] = {
72 { "/proc", PROCFS
, false },
73 { "/dev", BIND_DEV
, false },
74 { "/sys", SYSFS
, false },
77 /* ProtectKernelTunables= option and the related filesystem APIs */
78 static const MountEntry protect_kernel_tunables_table
[] = {
79 { "/proc/acpi", READONLY
, true },
80 { "/proc/apm", READONLY
, true }, /* Obsolete API, there's no point in permitting access to this, ever */
81 { "/proc/asound", READONLY
, true },
82 { "/proc/bus", READONLY
, true },
83 { "/proc/fs", READONLY
, true },
84 { "/proc/irq", READONLY
, true },
85 { "/proc/kallsyms", INACCESSIBLE
, true },
86 { "/proc/kcore", INACCESSIBLE
, true },
87 { "/proc/latency_stats", READONLY
, true },
88 { "/proc/mtrr", READONLY
, true },
89 { "/proc/scsi", READONLY
, true },
90 { "/proc/sys", READONLY
, false },
91 { "/proc/sysrq-trigger", READONLY
, true },
92 { "/proc/timer_stats", READONLY
, true },
93 { "/sys", READONLY
, false },
94 { "/sys/fs/bpf", READONLY
, true },
95 { "/sys/fs/cgroup", READWRITE
, false }, /* READONLY is set by ProtectControlGroups= option */
96 { "/sys/fs/selinux", READWRITE
, true },
97 { "/sys/kernel/debug", READONLY
, true },
98 { "/sys/kernel/tracing", READONLY
, true },
101 /* ProtectKernelModules= option */
102 static const MountEntry protect_kernel_modules_table
[] = {
104 { "/lib/modules", INACCESSIBLE
, true },
106 { "/usr/lib/modules", INACCESSIBLE
, true },
110 * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of
111 * system should be protected by ProtectSystem=
113 static const MountEntry protect_home_read_only_table
[] = {
114 { "/home", READONLY
, true },
115 { "/run/user", READONLY
, true },
116 { "/root", READONLY
, true },
119 /* ProtectHome=tmpfs table */
120 static const MountEntry protect_home_tmpfs_table
[] = {
121 { "/home", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755", .flags
= MS_NODEV
|MS_STRICTATIME
},
122 { "/run/user", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755", .flags
= MS_NODEV
|MS_STRICTATIME
},
123 { "/root", TMPFS
, true, .read_only
= true, .options_const
= "mode=0700", .flags
= MS_NODEV
|MS_STRICTATIME
},
126 /* ProtectHome=yes table */
127 static const MountEntry protect_home_yes_table
[] = {
128 { "/home", INACCESSIBLE
, true },
129 { "/run/user", INACCESSIBLE
, true },
130 { "/root", INACCESSIBLE
, true },
133 /* ProtectSystem=yes table */
134 static const MountEntry protect_system_yes_table
[] = {
135 { "/usr", READONLY
, false },
136 { "/boot", READONLY
, true },
137 { "/efi", READONLY
, true },
139 { "/lib", READONLY
, true },
140 { "/lib64", READONLY
, true },
141 { "/bin", READONLY
, true },
143 { "/sbin", READONLY
, true },
148 /* ProtectSystem=full includes ProtectSystem=yes */
149 static const MountEntry protect_system_full_table
[] = {
150 { "/usr", READONLY
, false },
151 { "/boot", READONLY
, true },
152 { "/efi", READONLY
, true },
153 { "/etc", READONLY
, false },
155 { "/lib", READONLY
, true },
156 { "/lib64", READONLY
, true },
157 { "/bin", READONLY
, true },
159 { "/sbin", READONLY
, true },
165 * ProtectSystem=strict table. In this strict mode, we mount everything
166 * read-only, except for /proc, /dev, /sys which are the kernel API VFS,
167 * which are left writable, but PrivateDevices= + ProtectKernelTunables=
168 * protect those, and these options should be fully orthogonal.
169 * (And of course /home and friends are also left writable, as ProtectHome=
170 * shall manage those, orthogonally).
172 static const MountEntry protect_system_strict_table
[] = {
173 { "/", READONLY
, false },
174 { "/proc", READWRITE
, false }, /* ProtectKernelTunables= */
175 { "/sys", READWRITE
, false }, /* ProtectKernelTunables= */
176 { "/dev", READWRITE
, false }, /* PrivateDevices= */
177 { "/home", READWRITE
, true }, /* ProtectHome= */
178 { "/run/user", READWRITE
, true }, /* ProtectHome= */
179 { "/root", READWRITE
, true }, /* ProtectHome= */
182 static const char *mount_entry_path(const MountEntry
*p
) {
185 /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that,
186 * otherwise the stack/static ->path field is returned. */
188 return p
->path_malloc
?: p
->path_const
;
191 static bool mount_entry_read_only(const MountEntry
*p
) {
194 return p
->read_only
|| IN_SET(p
->mode
, READONLY
, INACCESSIBLE
);
197 static const char *mount_entry_source(const MountEntry
*p
) {
200 return p
->source_malloc
?: p
->source_const
;
203 static const char *mount_entry_options(const MountEntry
*p
) {
206 return p
->options_malloc
?: p
->options_const
;
209 static void mount_entry_done(MountEntry
*p
) {
212 p
->path_malloc
= mfree(p
->path_malloc
);
213 p
->source_malloc
= mfree(p
->source_malloc
);
214 p
->options_malloc
= mfree(p
->options_malloc
);
217 static int append_access_mounts(MountEntry
**p
, char **strv
, MountMode mode
, bool forcibly_require_prefix
) {
222 /* Adds a list of user-supplied READWRITE/READONLY/INACCESSIBLE entries */
224 STRV_FOREACH(i
, strv
) {
225 bool ignore
= false, needs_prefix
= false;
228 /* Look for any prefixes */
229 if (startswith(e
, "-")) {
233 if (startswith(e
, "+")) {
238 if (!path_is_absolute(e
))
241 *((*p
)++) = (MountEntry
) {
245 .has_prefix
= !needs_prefix
&& !forcibly_require_prefix
,
252 static int append_empty_dir_mounts(MountEntry
**p
, char **strv
) {
257 /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the
258 * "/private/" boundary directories for DynamicUser=1. */
260 STRV_FOREACH(i
, strv
) {
262 *((*p
)++) = (MountEntry
) {
268 .options_const
= "mode=755",
269 .flags
= MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
,
276 static int append_bind_mounts(MountEntry
**p
, const BindMount
*binds
, size_t n
) {
281 for (i
= 0; i
< n
; i
++) {
282 const BindMount
*b
= binds
+ i
;
284 *((*p
)++) = (MountEntry
) {
285 .path_const
= b
->destination
,
286 .mode
= b
->recursive
? BIND_MOUNT_RECURSIVE
: BIND_MOUNT
,
287 .read_only
= b
->read_only
,
288 .source_const
= b
->source
,
289 .ignore
= b
->ignore_enoent
,
296 static int append_tmpfs_mounts(MountEntry
**p
, const TemporaryFileSystem
*tmpfs
, size_t n
) {
302 for (i
= 0; i
< n
; i
++) {
303 const TemporaryFileSystem
*t
= tmpfs
+ i
;
304 _cleanup_free_
char *o
= NULL
, *str
= NULL
;
305 unsigned long flags
= MS_NODEV
|MS_STRICTATIME
;
308 if (!path_is_absolute(t
->path
))
311 if (!isempty(t
->options
)) {
312 str
= strjoin("mode=0755,", t
->options
);
316 r
= mount_option_mangle(str
, MS_NODEV
|MS_STRICTATIME
, &flags
, &o
);
320 ro
= flags
& MS_RDONLY
;
325 *((*p
)++) = (MountEntry
) {
326 .path_const
= t
->path
,
339 static int append_static_mounts(MountEntry
**p
, const MountEntry
*mounts
, size_t n
, bool ignore_protect
) {
345 /* Adds a list of static pre-defined entries */
347 for (i
= 0; i
< n
; i
++)
348 *((*p
)++) = (MountEntry
) {
349 .path_const
= mount_entry_path(mounts
+i
),
350 .mode
= mounts
[i
].mode
,
351 .ignore
= mounts
[i
].ignore
|| ignore_protect
,
357 static int append_protect_home(MountEntry
**p
, ProtectHome protect_home
, bool ignore_protect
) {
360 switch (protect_home
) {
362 case PROTECT_HOME_NO
:
365 case PROTECT_HOME_READ_ONLY
:
366 return append_static_mounts(p
, protect_home_read_only_table
, ELEMENTSOF(protect_home_read_only_table
), ignore_protect
);
368 case PROTECT_HOME_TMPFS
:
369 return append_static_mounts(p
, protect_home_tmpfs_table
, ELEMENTSOF(protect_home_tmpfs_table
), ignore_protect
);
371 case PROTECT_HOME_YES
:
372 return append_static_mounts(p
, protect_home_yes_table
, ELEMENTSOF(protect_home_yes_table
), ignore_protect
);
375 assert_not_reached("Unexpected ProtectHome= value");
379 static int append_protect_system(MountEntry
**p
, ProtectSystem protect_system
, bool ignore_protect
) {
382 switch (protect_system
) {
384 case PROTECT_SYSTEM_NO
:
387 case PROTECT_SYSTEM_STRICT
:
388 return append_static_mounts(p
, protect_system_strict_table
, ELEMENTSOF(protect_system_strict_table
), ignore_protect
);
390 case PROTECT_SYSTEM_YES
:
391 return append_static_mounts(p
, protect_system_yes_table
, ELEMENTSOF(protect_system_yes_table
), ignore_protect
);
393 case PROTECT_SYSTEM_FULL
:
394 return append_static_mounts(p
, protect_system_full_table
, ELEMENTSOF(protect_system_full_table
), ignore_protect
);
397 assert_not_reached("Unexpected ProtectSystem= value");
401 static int mount_path_compare(const void *a
, const void *b
) {
402 const MountEntry
*p
= a
, *q
= b
;
405 /* If the paths are not equal, then order prefixes first */
406 d
= path_compare(mount_entry_path(p
), mount_entry_path(q
));
410 /* If the paths are equal, check the mode */
411 if (p
->mode
< q
->mode
)
413 if (p
->mode
> q
->mode
)
419 static int prefix_where_needed(MountEntry
*m
, size_t n
, const char *root_directory
) {
422 /* Prefixes all paths in the bind mount table with the root directory if it is specified and the entry needs
428 for (i
= 0; i
< n
; i
++) {
434 s
= prefix_root(root_directory
, mount_entry_path(m
+i
));
438 free_and_replace(m
[i
].path_malloc
, s
);
439 m
[i
].has_prefix
= true;
445 static void drop_duplicates(MountEntry
*m
, size_t *n
) {
446 MountEntry
*f
, *t
, *previous
;
451 /* Drops duplicate entries. Expects that the array is properly ordered already. */
453 for (f
= m
, t
= m
, previous
= NULL
; f
< m
+ *n
; f
++) {
455 /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare()
456 * above. Note that we only drop duplicates that haven't been mounted yet. */
458 path_equal(mount_entry_path(f
), mount_entry_path(previous
)) &&
459 !f
->applied
&& !previous
->applied
) {
460 log_debug("%s is duplicate.", mount_entry_path(f
));
461 previous
->read_only
= previous
->read_only
|| mount_entry_read_only(f
); /* Propagate the read-only flag to the remaining entry */
474 static void drop_inaccessible(MountEntry
*m
, size_t *n
) {
476 const char *clear
= NULL
;
481 /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly
482 * ordered already. */
484 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
486 /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop
487 * it, as inaccessible paths really should drop the entire subtree. */
488 if (clear
&& path_startswith(mount_entry_path(f
), clear
)) {
489 log_debug("%s is masked by %s.", mount_entry_path(f
), clear
);
494 clear
= f
->mode
== INACCESSIBLE
? mount_entry_path(f
) : NULL
;
503 static void drop_nop(MountEntry
*m
, size_t *n
) {
509 /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the
510 * list is ordered by prefixes. */
512 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
514 /* Only suppress such subtrees for READONLY and READWRITE entries */
515 if (IN_SET(f
->mode
, READONLY
, READWRITE
)) {
519 /* Now let's find the first parent of the entry we are looking at. */
520 for (p
= t
-1; p
>= m
; p
--) {
521 if (path_startswith(mount_entry_path(f
), mount_entry_path(p
))) {
527 /* We found it, let's see if it's the same mode, if so, we can drop this entry */
528 if (found
&& p
->mode
== f
->mode
) {
529 log_debug("%s is redundant by %s", mount_entry_path(f
), mount_entry_path(p
));
542 static void drop_outside_root(const char *root_directory
, MountEntry
*m
, size_t *n
) {
552 /* Drops all mounts that are outside of the root directory. */
554 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
556 if (!path_startswith(mount_entry_path(f
), root_directory
)) {
557 log_debug("%s is outside of root directory.", mount_entry_path(f
));
569 static int clone_device_node(const char *d
, const char *temporary_mount
, bool *make_devnode
) {
574 if (stat(d
, &st
) < 0) {
580 if (!S_ISBLK(st
.st_mode
) &&
581 !S_ISCHR(st
.st_mode
))
587 dn
= strjoina(temporary_mount
, d
);
590 mac_selinux_create_file_prepare(d
, st
.st_mode
);
591 r
= mknod(dn
, st
.st_mode
, st
.st_rdev
);
592 mac_selinux_create_file_clear();
597 return log_debug_errno(errno
, "mknod failed for %s: %m", d
);
599 *make_devnode
= false;
602 /* We're about to fallback to bind-mounting the device
603 * node. So create a dummy bind-mount target. */
604 mac_selinux_create_file_prepare(d
, 0);
605 r
= mknod(dn
, S_IFREG
, 0);
606 mac_selinux_create_file_clear();
608 if (r
< 0 && errno
!= EEXIST
)
609 return log_debug_errno(errno
, "mknod fallback failed for %s: %m", d
);
611 /* Fallback to bind-mounting:
612 * The assumption here is that all used device nodes carry standard
613 * properties. Specifically, the devices nodes we bind-mount should
614 * either be owned by root:root or root:tty (e.g. /dev/tty, /dev/ptmx)
615 * and should not carry ACLs. */
616 if (mount(d
, dn
, NULL
, MS_BIND
, NULL
) < 0)
617 return log_debug_errno(errno
, "mount failed for %s: %m", d
);
622 static int mount_private_dev(MountEntry
*m
) {
623 static const char devnodes
[] =
631 char temporary_mount
[] = "/tmp/namespace-dev-XXXXXX";
632 const char *d
, *dev
= NULL
, *devpts
= NULL
, *devshm
= NULL
, *devhugepages
= NULL
, *devmqueue
= NULL
, *devlog
= NULL
, *devptmx
= NULL
;
633 bool can_mknod
= true;
634 _cleanup_umask_ mode_t u
;
641 if (!mkdtemp(temporary_mount
))
644 dev
= strjoina(temporary_mount
, "/dev");
645 (void) mkdir(dev
, 0755);
646 if (mount("tmpfs", dev
, "tmpfs", DEV_MOUNT_OPTIONS
, "mode=755") < 0) {
651 devpts
= strjoina(temporary_mount
, "/dev/pts");
652 (void) mkdir(devpts
, 0755);
653 if (mount("/dev/pts", devpts
, NULL
, MS_BIND
, NULL
) < 0) {
658 /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx
659 * when /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible
660 * thus, in that case make a clone
662 * in nspawn and other containers it will be a symlink, in that case make it a symlink
664 r
= is_symlink("/dev/ptmx");
668 devptmx
= strjoina(temporary_mount
, "/dev/ptmx");
669 if (symlink("pts/ptmx", devptmx
) < 0) {
674 r
= clone_device_node("/dev/ptmx", temporary_mount
, &can_mknod
);
679 devshm
= strjoina(temporary_mount
, "/dev/shm");
680 (void) mkdir(devshm
, 0755);
681 r
= mount("/dev/shm", devshm
, NULL
, MS_BIND
, NULL
);
687 devmqueue
= strjoina(temporary_mount
, "/dev/mqueue");
688 (void) mkdir(devmqueue
, 0755);
689 (void) mount("/dev/mqueue", devmqueue
, NULL
, MS_BIND
, NULL
);
691 devhugepages
= strjoina(temporary_mount
, "/dev/hugepages");
692 (void) mkdir(devhugepages
, 0755);
693 (void) mount("/dev/hugepages", devhugepages
, NULL
, MS_BIND
, NULL
);
695 devlog
= strjoina(temporary_mount
, "/dev/log");
696 (void) symlink("/run/systemd/journal/dev-log", devlog
);
698 NULSTR_FOREACH(d
, devnodes
) {
699 r
= clone_device_node(d
, temporary_mount
, &can_mknod
);
700 /* ENXIO means the the *source* is not a device file, skip creation in that case */
701 if (r
< 0 && r
!= -ENXIO
)
705 dev_setup(temporary_mount
, UID_INVALID
, GID_INVALID
);
707 /* Create the /dev directory if missing. It is more likely to be
708 * missing when the service is started with RootDirectory. This is
709 * consistent with mount units creating the mount points when missing.
711 (void) mkdir_p_label(mount_entry_path(m
), 0755);
713 /* Unmount everything in old /dev */
714 umount_recursive(mount_entry_path(m
), 0);
715 if (mount(dev
, mount_entry_path(m
), NULL
, MS_MOVE
, NULL
) < 0) {
721 rmdir(temporary_mount
);
733 umount(devhugepages
);
740 rmdir(temporary_mount
);
745 static int mount_bind_dev(const MountEntry
*m
) {
750 /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the service's
751 * /dev. This is only used when RootDirectory= is set. */
753 (void) mkdir_p_label(mount_entry_path(m
), 0755);
755 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
757 return log_debug_errno(r
, "Unable to determine whether /dev is already mounted: %m");
758 if (r
> 0) /* make this a NOP if /dev is already a mount point */
761 if (mount("/dev", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
762 return log_debug_errno(errno
, "Failed to bind mount %s: %m", mount_entry_path(m
));
767 static int mount_sysfs(const MountEntry
*m
) {
772 (void) mkdir_p_label(mount_entry_path(m
), 0755);
774 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
776 return log_debug_errno(r
, "Unable to determine whether /sys is already mounted: %m");
777 if (r
> 0) /* make this a NOP if /sys is already a mount point */
780 /* Bind mount the host's version so that we get all child mounts of it, too. */
781 if (mount("/sys", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
) < 0)
782 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
787 static int mount_procfs(const MountEntry
*m
) {
792 (void) mkdir_p_label(mount_entry_path(m
), 0755);
794 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
796 return log_debug_errno(r
, "Unable to determine whether /proc is already mounted: %m");
797 if (r
> 0) /* make this a NOP if /proc is already a mount point */
800 /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in one */
801 if (mount("proc", mount_entry_path(m
), "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
) < 0)
802 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
807 static int mount_tmpfs(const MountEntry
*m
) {
810 /* First, get rid of everything that is below if there is anything. Then, overmount with our new tmpfs */
812 (void) mkdir_p_label(mount_entry_path(m
), 0755);
813 (void) umount_recursive(mount_entry_path(m
), 0);
815 if (mount("tmpfs", mount_entry_path(m
), "tmpfs", m
->flags
, mount_entry_options(m
)) < 0)
816 return log_debug_errno(errno
, "Failed to mount %s: %m", mount_entry_path(m
));
821 static int follow_symlink(
822 const char *root_directory
,
825 _cleanup_free_
char *target
= NULL
;
828 /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we
829 * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at
830 * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the
831 * end and already have a fully normalized name. */
833 r
= chase_symlinks(mount_entry_path(m
), root_directory
, CHASE_STEP
|CHASE_NONEXISTENT
, &target
);
835 return log_debug_errno(r
, "Failed to chase symlinks '%s': %m", mount_entry_path(m
));
836 if (r
> 0) /* Reached the end, nothing more to resolve */
839 if (m
->n_followed
>= CHASE_SYMLINKS_MAX
) { /* put a boundary on things */
840 log_debug("Symlink loop on '%s'.", mount_entry_path(m
));
844 log_debug("Followed mount entry path symlink %s → %s.", mount_entry_path(m
), target
);
846 free_and_replace(m
->path_malloc
, target
);
847 m
->has_prefix
= true;
854 static int apply_mount(
855 const char *root_directory
,
858 bool rbind
= true, make
= false;
864 log_debug("Applying namespace mount on %s", mount_entry_path(m
));
871 /* First, get rid of everything that is below if there
872 * is anything... Then, overmount it with an
873 * inaccessible path. */
874 (void) umount_recursive(mount_entry_path(m
), 0);
876 if (lstat(mount_entry_path(m
), &target
) < 0) {
877 if (errno
== ENOENT
&& m
->ignore
)
880 return log_debug_errno(errno
, "Failed to lstat() %s to determine what to mount over it: %m", mount_entry_path(m
));
883 what
= mode_to_inaccessible_node(target
.st_mode
);
885 log_debug("File type not supported for inaccessible mounts. Note that symlinks are not allowed");
893 r
= path_is_mount_point(mount_entry_path(m
), root_directory
, 0);
894 if (r
== -ENOENT
&& m
->ignore
)
897 return log_debug_errno(r
, "Failed to determine whether %s is already a mount point: %m", mount_entry_path(m
));
898 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. */
900 /* This isn't a mount point yet, let's make it one. */
901 what
= mount_entry_path(m
);
908 case BIND_MOUNT_RECURSIVE
: {
909 _cleanup_free_
char *chased
= NULL
;
911 /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note that bind
912 * mount source paths are always relative to the host root, hence we pass NULL as root directory to
913 * chase_symlinks() here. */
915 r
= chase_symlinks(mount_entry_source(m
), NULL
, CHASE_TRAIL_SLASH
, &chased
);
916 if (r
== -ENOENT
&& m
->ignore
) {
917 log_debug_errno(r
, "Path %s does not exist, ignoring.", mount_entry_source(m
));
921 return log_debug_errno(r
, "Failed to follow symlinks on %s: %m", mount_entry_source(m
));
923 log_debug("Followed source symlinks %s → %s.", mount_entry_source(m
), chased
);
925 free_and_replace(m
->source_malloc
, chased
);
927 what
= mount_entry_source(m
);
934 return mount_tmpfs(m
);
937 what
= mount_entry_source(m
);
942 return mount_private_dev(m
);
945 return mount_bind_dev(m
);
948 return mount_sysfs(m
);
951 return mount_procfs(m
);
954 assert_not_reached("Unknown mode");
959 if (mount(what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
) < 0) {
960 bool try_again
= false;
963 if (r
== -ENOENT
&& make
) {
966 /* Hmm, either the source or the destination are missing. Let's see if we can create the destination, then try again */
968 if (stat(what
, &st
) < 0)
969 log_debug_errno(errno
, "Mount point source '%s' is not accessible: %m", what
);
973 (void) mkdir_parents(mount_entry_path(m
), 0755);
975 if (S_ISDIR(st
.st_mode
))
976 q
= mkdir(mount_entry_path(m
), 0755) < 0 ? -errno
: 0;
978 q
= touch(mount_entry_path(m
));
981 log_debug_errno(q
, "Failed to create destination mount point node '%s': %m", mount_entry_path(m
));
988 if (mount(what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
) < 0)
995 return log_debug_errno(r
, "Failed to mount %s to %s: %m", what
, mount_entry_path(m
));
998 log_debug("Successfully mounted %s to %s", what
, mount_entry_path(m
));
1002 static int make_read_only(const MountEntry
*m
, char **blacklist
, FILE *proc_self_mountinfo
) {
1006 assert(proc_self_mountinfo
);
1008 if (mount_entry_read_only(m
)) {
1009 if (IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
)) {
1010 /* Make superblock readonly */
1011 if (mount(NULL
, mount_entry_path(m
), NULL
, MS_REMOUNT
| MS_RDONLY
| m
->flags
, mount_entry_options(m
)) < 0)
1014 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), true, blacklist
, proc_self_mountinfo
);
1015 } else if (m
->mode
== PRIVATE_DEV
) {
1016 /* Superblock can be readonly but the submounts can't */
1017 if (mount(NULL
, mount_entry_path(m
), NULL
, MS_REMOUNT
|DEV_MOUNT_OPTIONS
|MS_RDONLY
, NULL
) < 0)
1022 /* Not that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked read-only
1023 * already stays this way. This improves compatibility with container managers, where we won't attempt to undo
1024 * read-only mounts already applied. */
1026 if (r
== -ENOENT
&& m
->ignore
)
1032 static bool namespace_info_mount_apivfs(const NamespaceInfo
*ns_info
) {
1036 * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
1037 * since to protect the API VFS mounts, they need to be around in the
1041 return ns_info
->mount_apivfs
||
1042 ns_info
->protect_control_groups
||
1043 ns_info
->protect_kernel_tunables
;
1046 static size_t namespace_calculate_mounts(
1047 const NamespaceInfo
*ns_info
,
1048 char** read_write_paths
,
1049 char** read_only_paths
,
1050 char** inaccessible_paths
,
1051 char** empty_directories
,
1052 size_t n_bind_mounts
,
1053 size_t n_temporary_filesystems
,
1054 const char* tmp_dir
,
1055 const char* var_tmp_dir
,
1056 ProtectHome protect_home
,
1057 ProtectSystem protect_system
) {
1059 size_t protect_home_cnt
;
1060 size_t protect_system_cnt
=
1061 (protect_system
== PROTECT_SYSTEM_STRICT
?
1062 ELEMENTSOF(protect_system_strict_table
) :
1063 ((protect_system
== PROTECT_SYSTEM_FULL
) ?
1064 ELEMENTSOF(protect_system_full_table
) :
1065 ((protect_system
== PROTECT_SYSTEM_YES
) ?
1066 ELEMENTSOF(protect_system_yes_table
) : 0)));
1069 (protect_home
== PROTECT_HOME_YES
?
1070 ELEMENTSOF(protect_home_yes_table
) :
1071 ((protect_home
== PROTECT_HOME_READ_ONLY
) ?
1072 ELEMENTSOF(protect_home_read_only_table
) :
1073 ((protect_home
== PROTECT_HOME_TMPFS
) ?
1074 ELEMENTSOF(protect_home_tmpfs_table
) : 0)));
1076 return !!tmp_dir
+ !!var_tmp_dir
+
1077 strv_length(read_write_paths
) +
1078 strv_length(read_only_paths
) +
1079 strv_length(inaccessible_paths
) +
1080 strv_length(empty_directories
) +
1082 n_temporary_filesystems
+
1083 ns_info
->private_dev
+
1084 (ns_info
->protect_kernel_tunables
? ELEMENTSOF(protect_kernel_tunables_table
) : 0) +
1085 (ns_info
->protect_control_groups
? 1 : 0) +
1086 (ns_info
->protect_kernel_modules
? ELEMENTSOF(protect_kernel_modules_table
) : 0) +
1087 protect_home_cnt
+ protect_system_cnt
+
1088 (namespace_info_mount_apivfs(ns_info
) ? ELEMENTSOF(apivfs_table
) : 0);
1091 static void normalize_mounts(const char *root_directory
, MountEntry
*mounts
, size_t *n_mounts
) {
1092 assert(root_directory
);
1094 assert(mounts
|| *n_mounts
== 0);
1096 qsort_safe(mounts
, *n_mounts
, sizeof(MountEntry
), mount_path_compare
);
1098 drop_duplicates(mounts
, n_mounts
);
1099 drop_outside_root(root_directory
, mounts
, n_mounts
);
1100 drop_inaccessible(mounts
, n_mounts
);
1101 drop_nop(mounts
, n_mounts
);
1104 int setup_namespace(
1105 const char* root_directory
,
1106 const char* root_image
,
1107 const NamespaceInfo
*ns_info
,
1108 char** read_write_paths
,
1109 char** read_only_paths
,
1110 char** inaccessible_paths
,
1111 char** empty_directories
,
1112 const BindMount
*bind_mounts
,
1113 size_t n_bind_mounts
,
1114 const TemporaryFileSystem
*temporary_filesystems
,
1115 size_t n_temporary_filesystems
,
1116 const char* tmp_dir
,
1117 const char* var_tmp_dir
,
1118 ProtectHome protect_home
,
1119 ProtectSystem protect_system
,
1120 unsigned long mount_flags
,
1121 DissectImageFlags dissect_image_flags
) {
1123 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
1124 _cleanup_(decrypted_image_unrefp
) DecryptedImage
*decrypted_image
= NULL
;
1125 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
1126 _cleanup_free_
void *root_hash
= NULL
;
1127 MountEntry
*m
, *mounts
= NULL
;
1128 size_t n_mounts
, root_hash_size
= 0;
1129 bool require_prefix
= false;
1135 if (mount_flags
== 0)
1136 mount_flags
= MS_SHARED
;
1139 dissect_image_flags
|= DISSECT_IMAGE_REQUIRE_ROOT
;
1141 if (protect_system
== PROTECT_SYSTEM_STRICT
&&
1142 protect_home
!= PROTECT_HOME_NO
&&
1143 strv_isempty(read_write_paths
))
1144 dissect_image_flags
|= DISSECT_IMAGE_READ_ONLY
;
1146 r
= loop_device_make_by_path(root_image
,
1147 dissect_image_flags
& DISSECT_IMAGE_READ_ONLY
? O_RDONLY
: O_RDWR
,
1152 r
= root_hash_load(root_image
, &root_hash
, &root_hash_size
);
1156 r
= dissect_image(loop_device
->fd
, root_hash
, root_hash_size
, dissect_image_flags
, &dissected_image
);
1160 r
= dissected_image_decrypt(dissected_image
, NULL
, root_hash
, root_hash_size
, dissect_image_flags
, &decrypted_image
);
1166 root
= root_directory
;
1168 /* Always create the mount namespace in a temporary directory, instead of operating
1169 * directly in the root. The temporary directory prevents any mounts from being
1170 * potentially obscured my other mounts we already applied.
1171 * We use the same mount point for all images, which is safe, since they all live
1172 * in their own namespaces after all, and hence won't see each other. */
1174 root
= "/run/systemd/unit-root";
1175 (void) mkdir_label(root
, 0700);
1176 require_prefix
= true;
1179 n_mounts
= namespace_calculate_mounts(
1186 n_temporary_filesystems
,
1187 tmp_dir
, var_tmp_dir
,
1188 protect_home
, protect_system
);
1191 m
= mounts
= (MountEntry
*) alloca0(n_mounts
* sizeof(MountEntry
));
1192 r
= append_access_mounts(&m
, read_write_paths
, READWRITE
, require_prefix
);
1196 r
= append_access_mounts(&m
, read_only_paths
, READONLY
, require_prefix
);
1200 r
= append_access_mounts(&m
, inaccessible_paths
, INACCESSIBLE
, require_prefix
);
1204 r
= append_empty_dir_mounts(&m
, empty_directories
);
1208 r
= append_bind_mounts(&m
, bind_mounts
, n_bind_mounts
);
1212 r
= append_tmpfs_mounts(&m
, temporary_filesystems
, n_temporary_filesystems
);
1217 *(m
++) = (MountEntry
) {
1218 .path_const
= "/tmp",
1219 .mode
= PRIVATE_TMP
,
1220 .source_const
= tmp_dir
,
1225 *(m
++) = (MountEntry
) {
1226 .path_const
= "/var/tmp",
1227 .mode
= PRIVATE_TMP
,
1228 .source_const
= var_tmp_dir
,
1232 if (ns_info
->private_dev
) {
1233 *(m
++) = (MountEntry
) {
1234 .path_const
= "/dev",
1235 .mode
= PRIVATE_DEV
,
1239 if (ns_info
->protect_kernel_tunables
) {
1240 r
= append_static_mounts(&m
, protect_kernel_tunables_table
, ELEMENTSOF(protect_kernel_tunables_table
), ns_info
->ignore_protect_paths
);
1245 if (ns_info
->protect_kernel_modules
) {
1246 r
= append_static_mounts(&m
, protect_kernel_modules_table
, ELEMENTSOF(protect_kernel_modules_table
), ns_info
->ignore_protect_paths
);
1251 if (ns_info
->protect_control_groups
) {
1252 *(m
++) = (MountEntry
) {
1253 .path_const
= "/sys/fs/cgroup",
1258 r
= append_protect_home(&m
, protect_home
, ns_info
->ignore_protect_paths
);
1262 r
= append_protect_system(&m
, protect_system
, false);
1266 if (namespace_info_mount_apivfs(ns_info
)) {
1267 r
= append_static_mounts(&m
, apivfs_table
, ELEMENTSOF(apivfs_table
), ns_info
->ignore_protect_paths
);
1272 assert(mounts
+ n_mounts
== m
);
1274 /* Prepend the root directory where that's necessary */
1275 r
= prefix_where_needed(mounts
, n_mounts
, root
);
1279 normalize_mounts(root_directory
, mounts
, &n_mounts
);
1282 if (unshare(CLONE_NEWNS
) < 0) {
1287 /* Remount / as SLAVE so that nothing now mounted in the namespace
1288 * shows up in the parent */
1289 if (mount(NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
) < 0) {
1295 /* A root image is specified, mount it to the right place */
1296 r
= dissected_image_mount(dissected_image
, root
, UID_INVALID
, dissect_image_flags
);
1300 if (decrypted_image
) {
1301 r
= decrypted_image_relinquish(decrypted_image
);
1306 loop_device_relinquish(loop_device
);
1308 } else if (root_directory
) {
1310 /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
1311 r
= path_is_mount_point(root
, NULL
, AT_SYMLINK_FOLLOW
);
1315 if (mount(root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0) {
1323 /* Let's mount the main root directory to the root directory to use */
1324 if (mount("/", root
, NULL
, MS_BIND
|MS_REC
, NULL
) < 0) {
1330 /* Try to set up the new root directory before mounting anything else there. */
1331 if (root_image
|| root_directory
)
1332 (void) base_filesystem_create(root
, UID_INVALID
, GID_INVALID
);
1335 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
1339 /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of /proc.
1340 * For example, this is the case with the option: 'InaccessiblePaths=/proc' */
1341 proc_self_mountinfo
= fopen("/proc/self/mountinfo", "re");
1342 if (!proc_self_mountinfo
) {
1347 /* First round, establish all mounts we need */
1351 for (m
= mounts
; m
< mounts
+ n_mounts
; ++m
) {
1356 r
= follow_symlink(root
, m
);
1360 /* We hit a symlinked mount point. The entry got rewritten and might point to a
1361 * very different place now. Let's normalize the changed list, and start from
1362 * the beginning. After all to mount the entry at the new location we might
1363 * need some other mounts first */
1368 r
= apply_mount(root
, m
);
1378 normalize_mounts(root_directory
, mounts
, &n_mounts
);
1381 /* Create a blacklist we can pass to bind_mount_recursive() */
1382 blacklist
= newa(char*, n_mounts
+1);
1383 for (j
= 0; j
< n_mounts
; j
++)
1384 blacklist
[j
] = (char*) mount_entry_path(mounts
+j
);
1385 blacklist
[j
] = NULL
;
1387 /* Second round, flip the ro bits if necessary. */
1388 for (m
= mounts
; m
< mounts
+ n_mounts
; ++m
) {
1389 r
= make_read_only(m
, blacklist
, proc_self_mountinfo
);
1395 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
1396 r
= mount_move_root(root
);
1400 /* Remount / as the desired mode. Note that this will not
1401 * reestablish propagation from our side to the host, since
1402 * what's disconnected is disconnected. */
1403 if (mount(NULL
, "/", NULL
, mount_flags
| MS_REC
, NULL
) < 0) {
1411 for (m
= mounts
; m
< mounts
+ n_mounts
; m
++)
1412 mount_entry_done(m
);
1417 void bind_mount_free_many(BindMount
*b
, size_t n
) {
1420 assert(b
|| n
== 0);
1422 for (i
= 0; i
< n
; i
++) {
1424 free(b
[i
].destination
);
1430 int bind_mount_add(BindMount
**b
, size_t *n
, const BindMount
*item
) {
1431 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
1438 s
= strdup(item
->source
);
1442 d
= strdup(item
->destination
);
1446 c
= reallocarray(*b
, *n
+ 1, sizeof(BindMount
));
1452 c
[(*n
) ++] = (BindMount
) {
1453 .source
= TAKE_PTR(s
),
1454 .destination
= TAKE_PTR(d
),
1455 .read_only
= item
->read_only
,
1456 .recursive
= item
->recursive
,
1457 .ignore_enoent
= item
->ignore_enoent
,
1463 void temporary_filesystem_free_many(TemporaryFileSystem
*t
, size_t n
) {
1466 assert(t
|| n
== 0);
1468 for (i
= 0; i
< n
; i
++) {
1476 int temporary_filesystem_add(
1477 TemporaryFileSystem
**t
,
1480 const char *options
) {
1482 _cleanup_free_
char *p
= NULL
, *o
= NULL
;
1483 TemporaryFileSystem
*c
;
1493 if (!isempty(options
)) {
1494 o
= strdup(options
);
1499 c
= reallocarray(*t
, *n
+ 1, sizeof(TemporaryFileSystem
));
1505 c
[(*n
) ++] = (TemporaryFileSystem
) {
1506 .path
= TAKE_PTR(p
),
1507 .options
= TAKE_PTR(o
),
1513 static int setup_one_tmp_dir(const char *id
, const char *prefix
, char **path
) {
1514 _cleanup_free_
char *x
= NULL
;
1515 char bid
[SD_ID128_STRING_MAX
];
1523 /* We include the boot id in the directory so that after a
1524 * reboot we can easily identify obsolete directories. */
1526 r
= sd_id128_get_boot(&boot_id
);
1530 x
= strjoin(prefix
, "/systemd-private-", sd_id128_to_string(boot_id
, bid
), "-", id
, "-XXXXXX");
1534 RUN_WITH_UMASK(0077)
1538 RUN_WITH_UMASK(0000) {
1541 y
= strjoina(x
, "/tmp");
1543 if (mkdir(y
, 0777 | S_ISVTX
) < 0)
1547 *path
= TAKE_PTR(x
);
1552 int setup_tmp_dirs(const char *id
, char **tmp_dir
, char **var_tmp_dir
) {
1558 assert(var_tmp_dir
);
1560 r
= setup_one_tmp_dir(id
, "/tmp", &a
);
1564 r
= setup_one_tmp_dir(id
, "/var/tmp", &b
);
1568 t
= strjoina(a
, "/tmp");
1582 int setup_netns(int netns_storage_socket
[2]) {
1583 _cleanup_close_
int netns
= -1;
1586 assert(netns_storage_socket
);
1587 assert(netns_storage_socket
[0] >= 0);
1588 assert(netns_storage_socket
[1] >= 0);
1590 /* We use the passed socketpair as a storage buffer for our
1591 * namespace reference fd. Whatever process runs this first
1592 * shall create a new namespace, all others should just join
1593 * it. To serialize that we use a file lock on the socket
1596 * It's a bit crazy, but hey, works great! */
1598 if (lockf(netns_storage_socket
[0], F_LOCK
, 0) < 0)
1601 netns
= receive_one_fd(netns_storage_socket
[0], MSG_DONTWAIT
);
1602 if (netns
== -EAGAIN
) {
1603 /* Nothing stored yet, so let's create a new namespace */
1605 if (unshare(CLONE_NEWNET
) < 0) {
1612 netns
= open("/proc/self/ns/net", O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
1620 } else if (netns
< 0) {
1625 /* Yay, found something, so let's join the namespace */
1626 if (setns(netns
, CLONE_NEWNET
) < 0) {
1634 q
= send_one_fd(netns_storage_socket
[1], netns
, MSG_DONTWAIT
);
1641 (void) lockf(netns_storage_socket
[0], F_ULOCK
, 0);
1645 bool ns_type_supported(NamespaceType type
) {
1646 const char *t
, *ns_proc
;
1648 t
= namespace_type_to_string(type
);
1649 if (!t
) /* Don't know how to translate this? Then it's not supported */
1652 ns_proc
= strjoina("/proc/self/ns/", t
);
1653 return access(ns_proc
, F_OK
) == 0;
1656 static const char *const protect_home_table
[_PROTECT_HOME_MAX
] = {
1657 [PROTECT_HOME_NO
] = "no",
1658 [PROTECT_HOME_YES
] = "yes",
1659 [PROTECT_HOME_READ_ONLY
] = "read-only",
1660 [PROTECT_HOME_TMPFS
] = "tmpfs",
1663 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home
, ProtectHome
, PROTECT_HOME_YES
);
1665 static const char *const protect_system_table
[_PROTECT_SYSTEM_MAX
] = {
1666 [PROTECT_SYSTEM_NO
] = "no",
1667 [PROTECT_SYSTEM_YES
] = "yes",
1668 [PROTECT_SYSTEM_FULL
] = "full",
1669 [PROTECT_SYSTEM_STRICT
] = "strict",
1672 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system
, ProtectSystem
, PROTECT_SYSTEM_YES
);
1674 static const char* const namespace_type_table
[] = {
1675 [NAMESPACE_MOUNT
] = "mnt",
1676 [NAMESPACE_CGROUP
] = "cgroup",
1677 [NAMESPACE_UTS
] = "uts",
1678 [NAMESPACE_IPC
] = "ipc",
1679 [NAMESPACE_USER
] = "user",
1680 [NAMESPACE_PID
] = "pid",
1681 [NAMESPACE_NET
] = "net",
1684 DEFINE_STRING_TABLE_LOOKUP(namespace_type
, NamespaceType
);