1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
4 #include <linux/loop.h>
14 #include "alloc-util.h"
15 #include "base-filesystem.h"
17 #include "dev-setup.h"
18 #include "devnum-util.h"
21 #include "extension-util.h"
23 #include "format-util.h"
24 #include "glyph-util.h"
25 #include "label-util.h"
27 #include "lock-util.h"
28 #include "loop-util.h"
29 #include "loopback-setup.h"
30 #include "missing_syscall.h"
31 #include "mkdir-label.h"
32 #include "mount-util.h"
33 #include "mountpoint-util.h"
34 #include "namespace-util.h"
35 #include "namespace.h"
37 #include "nulstr-util.h"
39 #include "path-util.h"
40 #include "selinux-util.h"
41 #include "socket-util.h"
42 #include "sort-util.h"
43 #include "stat-util.h"
44 #include "string-table.h"
45 #include "string-util.h"
47 #include "tmpfile-util.h"
48 #include "umask-util.h"
49 #include "user-util.h"
51 #define DEV_MOUNT_OPTIONS (MS_NOSUID|MS_STRICTATIME|MS_NOEXEC)
53 typedef enum MountMode
{
54 /* This is ordered by priority! */
74 EXTENSION_DIRECTORIES
, /* Bind-mounted outside the root directory, and used by subsequent mounts */
75 EXTENSION_IMAGES
, /* Mounted outside the root directory, and used by subsequent mounts */
77 READWRITE_IMPLICIT
, /* Should have the lowest priority. */
81 typedef struct MountEntry
{
82 const char *path_const
; /* Memory allocated on stack or static */
84 bool ignore
:1; /* Ignore if path does not exist? */
85 bool has_prefix
:1; /* Already is prefixed by the root dir? */
86 bool read_only
:1; /* Shall this mount point be read-only? */
87 bool nosuid
:1; /* Shall set MS_NOSUID on the mount itself */
88 bool noexec
:1; /* Shall set MS_NOEXEC on the mount itself */
89 bool exec
:1; /* Shall clear MS_NOEXEC on the mount itself */
90 bool applied
:1; /* Already applied */
91 char *path_malloc
; /* Use this instead of 'path_const' if we had to allocate memory */
92 const char *unprefixed_path_const
; /* If the path was amended with a prefix, these will save the original */
93 char *unprefixed_path_malloc
;
94 const char *source_const
; /* The source path, for bind mounts or images */
96 const char *options_const
;/* Mount options for tmpfs */
98 unsigned long flags
; /* Mount flags used by EMPTY_DIR and TMPFS. Do not include MS_RDONLY here, but please use read_only. */
100 LIST_HEAD(MountOptions
, image_options
);
103 /* If MountAPIVFS= is used, let's mount /sys, /proc, /dev and /run into the it, but only as a fallback if the user hasn't mounted
104 * something there already. These mounts are hence overridden by any other explicitly configured mounts. */
105 static const MountEntry apivfs_table
[] = {
106 { "/proc", PROCFS
, false },
107 { "/dev", BIND_DEV
, false },
108 { "/sys", BIND_SYSFS
, false },
109 { "/run", RUN
, false, .options_const
= "mode=0755" TMPFS_LIMITS_RUN
, .flags
= MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
},
112 /* ProtectKernelTunables= option and the related filesystem APIs */
113 static const MountEntry protect_kernel_tunables_proc_table
[] = {
114 { "/proc/acpi", READONLY
, true },
115 { "/proc/apm", READONLY
, true }, /* Obsolete API, there's no point in permitting access to this, ever */
116 { "/proc/asound", READONLY
, true },
117 { "/proc/bus", READONLY
, true },
118 { "/proc/fs", READONLY
, true },
119 { "/proc/irq", READONLY
, true },
120 { "/proc/kallsyms", INACCESSIBLE
, true },
121 { "/proc/kcore", INACCESSIBLE
, true },
122 { "/proc/latency_stats", READONLY
, true },
123 { "/proc/mtrr", READONLY
, true },
124 { "/proc/scsi", READONLY
, true },
125 { "/proc/sys", READONLY
, true },
126 { "/proc/sysrq-trigger", READONLY
, true },
127 { "/proc/timer_stats", READONLY
, true },
130 static const MountEntry protect_kernel_tunables_sys_table
[] = {
131 { "/sys", READONLY
, false },
132 { "/sys/fs/bpf", READONLY
, true },
133 { "/sys/fs/cgroup", READWRITE_IMPLICIT
, false }, /* READONLY is set by ProtectControlGroups= option */
134 { "/sys/fs/selinux", READWRITE_IMPLICIT
, true },
135 { "/sys/kernel/debug", READONLY
, true },
136 { "/sys/kernel/tracing", READONLY
, true },
139 /* ProtectKernelModules= option */
140 static const MountEntry protect_kernel_modules_table
[] = {
141 { "/usr/lib/modules", INACCESSIBLE
, true },
144 /* ProtectKernelLogs= option */
145 static const MountEntry protect_kernel_logs_proc_table
[] = {
146 { "/proc/kmsg", INACCESSIBLE
, true },
149 static const MountEntry protect_kernel_logs_dev_table
[] = {
150 { "/dev/kmsg", INACCESSIBLE
, true },
154 * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of
155 * system should be protected by ProtectSystem=
157 static const MountEntry protect_home_read_only_table
[] = {
158 { "/home", READONLY
, true },
159 { "/run/user", READONLY
, true },
160 { "/root", READONLY
, true },
163 /* ProtectHome=tmpfs table */
164 static const MountEntry protect_home_tmpfs_table
[] = {
165 { "/home", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
166 { "/run/user", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
167 { "/root", TMPFS
, true, .read_only
= true, .options_const
= "mode=0700" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
170 /* ProtectHome=yes table */
171 static const MountEntry protect_home_yes_table
[] = {
172 { "/home", INACCESSIBLE
, true },
173 { "/run/user", INACCESSIBLE
, true },
174 { "/root", INACCESSIBLE
, true },
177 /* ProtectSystem=yes table */
178 static const MountEntry protect_system_yes_table
[] = {
179 { "/usr", READONLY
, false },
180 { "/boot", READONLY
, true },
181 { "/efi", READONLY
, true },
184 /* ProtectSystem=full includes ProtectSystem=yes */
185 static const MountEntry protect_system_full_table
[] = {
186 { "/usr", READONLY
, false },
187 { "/boot", READONLY
, true },
188 { "/efi", READONLY
, true },
189 { "/etc", READONLY
, false },
193 * ProtectSystem=strict table. In this strict mode, we mount everything
194 * read-only, except for /proc, /dev, /sys which are the kernel API VFS,
195 * which are left writable, but PrivateDevices= + ProtectKernelTunables=
196 * protect those, and these options should be fully orthogonal.
197 * (And of course /home and friends are also left writable, as ProtectHome=
198 * shall manage those, orthogonally).
200 static const MountEntry protect_system_strict_table
[] = {
201 { "/", READONLY
, false },
202 { "/proc", READWRITE_IMPLICIT
, false }, /* ProtectKernelTunables= */
203 { "/sys", READWRITE_IMPLICIT
, false }, /* ProtectKernelTunables= */
204 { "/dev", READWRITE_IMPLICIT
, false }, /* PrivateDevices= */
205 { "/home", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
206 { "/run/user", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
207 { "/root", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
210 static const char * const mount_mode_table
[_MOUNT_MODE_MAX
] = {
211 [INACCESSIBLE
] = "inaccessible",
212 [OVERLAY_MOUNT
] = "overlay",
213 [MOUNT_IMAGES
] = "mount-images",
214 [BIND_MOUNT
] = "bind",
215 [BIND_MOUNT_RECURSIVE
] = "rbind",
216 [PRIVATE_TMP
] = "private-tmp",
217 [PRIVATE_TMP_READONLY
] = "private-tmp-read-only",
218 [PRIVATE_DEV
] = "private-dev",
219 [BIND_DEV
] = "bind-dev",
220 [EMPTY_DIR
] = "empty",
221 [PRIVATE_SYSFS
] = "private-sysfs",
222 [BIND_SYSFS
] = "bind-sysfs",
224 [READONLY
] = "read-only",
225 [READWRITE
] = "read-write",
230 [EXTENSION_DIRECTORIES
] = "extension-directories",
231 [EXTENSION_IMAGES
] = "extension-images",
232 [MQUEUEFS
] = "mqueuefs",
233 [READWRITE_IMPLICIT
] = "read-write-implicit",
236 /* Helper struct for naming simplicity and reusability */
237 static const struct {
238 const char *level_env
;
239 const char *level_env_print
;
240 } image_class_info
[_IMAGE_CLASS_MAX
] = {
242 .level_env
= "SYSEXT_LEVEL",
243 .level_env_print
= " SYSEXT_LEVEL=",
246 .level_env
= "CONFEXT_LEVEL",
247 .level_env_print
= " CONFEXT_LEVEL=",
251 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(mount_mode
, MountMode
);
253 static const char *mount_entry_path(const MountEntry
*p
) {
256 /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that,
257 * otherwise the stack/static ->path field is returned. */
259 return p
->path_malloc
?: p
->path_const
;
262 static const char *mount_entry_unprefixed_path(const MountEntry
*p
) {
265 /* Returns the unprefixed path (ie: before prefix_where_needed() ran), if any */
267 return p
->unprefixed_path_malloc
?: p
->unprefixed_path_const
?: mount_entry_path(p
);
270 static void mount_entry_consume_prefix(MountEntry
*p
, char *new_path
) {
272 assert(p
->path_malloc
|| p
->path_const
);
275 /* Saves current path in unprefixed_ variable, and takes over new_path */
277 free_and_replace(p
->unprefixed_path_malloc
, p
->path_malloc
);
278 /* If we didn't have a path on the heap, then it's a static one */
279 if (!p
->unprefixed_path_malloc
)
280 p
->unprefixed_path_const
= p
->path_const
;
281 p
->path_malloc
= new_path
;
282 p
->has_prefix
= true;
285 static bool mount_entry_read_only(const MountEntry
*p
) {
288 return p
->read_only
|| IN_SET(p
->mode
, READONLY
, INACCESSIBLE
, PRIVATE_TMP_READONLY
);
291 static bool mount_entry_noexec(const MountEntry
*p
) {
294 return p
->noexec
|| IN_SET(p
->mode
, NOEXEC
, INACCESSIBLE
, PRIVATE_SYSFS
, BIND_SYSFS
, PROCFS
);
297 static bool mount_entry_exec(const MountEntry
*p
) {
300 return p
->exec
|| p
->mode
== EXEC
;
303 static const char *mount_entry_source(const MountEntry
*p
) {
306 return p
->source_malloc
?: p
->source_const
;
309 static const char *mount_entry_options(const MountEntry
*p
) {
312 return p
->options_malloc
?: p
->options_const
;
315 static void mount_entry_done(MountEntry
*p
) {
318 p
->path_malloc
= mfree(p
->path_malloc
);
319 p
->unprefixed_path_malloc
= mfree(p
->unprefixed_path_malloc
);
320 p
->source_malloc
= mfree(p
->source_malloc
);
321 p
->options_malloc
= mfree(p
->options_malloc
);
322 p
->image_options
= mount_options_free_all(p
->image_options
);
325 static int append_access_mounts(MountEntry
**p
, char **strv
, MountMode mode
, bool forcibly_require_prefix
) {
328 /* Adds a list of user-supplied READWRITE/READWRITE_IMPLICIT/READONLY/INACCESSIBLE entries */
330 STRV_FOREACH(i
, strv
) {
331 bool ignore
= false, needs_prefix
= false;
334 /* Look for any prefixes */
335 if (startswith(e
, "-")) {
339 if (startswith(e
, "+")) {
344 if (!path_is_absolute(e
))
345 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
346 "Path is not absolute: %s", e
);
348 *((*p
)++) = (MountEntry
) {
352 .has_prefix
= !needs_prefix
&& !forcibly_require_prefix
,
359 static int append_empty_dir_mounts(MountEntry
**p
, char **strv
) {
362 /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the
363 * "/private/" boundary directories for DynamicUser=1. */
365 STRV_FOREACH(i
, strv
) {
367 *((*p
)++) = (MountEntry
) {
372 .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
,
373 .flags
= MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
,
380 static int append_bind_mounts(MountEntry
**p
, const BindMount
*binds
, size_t n
) {
383 for (size_t i
= 0; i
< n
; i
++) {
384 const BindMount
*b
= binds
+ i
;
386 *((*p
)++) = (MountEntry
) {
387 .path_const
= b
->destination
,
388 .mode
= b
->recursive
? BIND_MOUNT_RECURSIVE
: BIND_MOUNT
,
389 .read_only
= b
->read_only
,
391 .source_const
= b
->source
,
392 .ignore
= b
->ignore_enoent
,
399 static int append_mount_images(MountEntry
**p
, const MountImage
*mount_images
, size_t n
) {
402 for (size_t i
= 0; i
< n
; i
++) {
403 const MountImage
*m
= mount_images
+ i
;
405 *((*p
)++) = (MountEntry
) {
406 .path_const
= m
->destination
,
407 .mode
= MOUNT_IMAGES
,
408 .source_const
= m
->source
,
409 .image_options
= m
->mount_options
,
410 .ignore
= m
->ignore_enoent
,
417 static int append_extensions(
420 const char *extension_dir
,
422 const MountImage
*mount_images
,
424 char **extension_directories
) {
426 _cleanup_strv_free_
char **overlays
= NULL
;
429 if (n
== 0 && strv_isempty(extension_directories
))
433 assert(extension_dir
);
435 /* Prepare a list of overlays, that will have as each element a string suitable for being
436 * passed as a lowerdir= parameter, so start with the hierarchy on the root.
437 * The overlays vector will have the same number of elements and will correspond to the
438 * hierarchies vector, so they can be iterated upon together. */
439 STRV_FOREACH(hierarchy
, hierarchies
) {
440 _cleanup_free_
char *prefixed_hierarchy
= NULL
;
442 prefixed_hierarchy
= path_join(root
, *hierarchy
);
443 if (!prefixed_hierarchy
)
446 r
= strv_consume(&overlays
, TAKE_PTR(prefixed_hierarchy
));
451 /* First, prepare a mount for each image, but these won't be visible to the unit, instead
452 * they will be mounted in our propagate directory, and used as a source for the overlay. */
453 for (size_t i
= 0; i
< n
; i
++) {
454 _cleanup_free_
char *mount_point
= NULL
;
455 const MountImage
*m
= mount_images
+ i
;
457 r
= asprintf(&mount_point
, "%s/%zu", extension_dir
, i
);
461 for (size_t j
= 0; hierarchies
&& hierarchies
[j
]; ++j
) {
462 _cleanup_free_
char *prefixed_hierarchy
= NULL
, *escaped
= NULL
, *lowerdir
= NULL
;
464 prefixed_hierarchy
= path_join(mount_point
, hierarchies
[j
]);
465 if (!prefixed_hierarchy
)
468 escaped
= shell_escape(prefixed_hierarchy
, ",:");
472 /* Note that lowerdir= parameters are in 'reverse' order, so the
473 * top-most directory in the overlay comes first in the list. */
474 lowerdir
= strjoin(escaped
, ":", overlays
[j
]);
478 free_and_replace(overlays
[j
], lowerdir
);
481 *((*p
)++) = (MountEntry
) {
482 .path_malloc
= TAKE_PTR(mount_point
),
483 .image_options
= m
->mount_options
,
484 .ignore
= m
->ignore_enoent
,
485 .source_const
= m
->source
,
486 .mode
= EXTENSION_IMAGES
,
491 /* Secondly, extend the lowerdir= parameters with each ExtensionDirectory.
492 * Bind mount them in the same location as the ExtensionImages, so that we
493 * can check that they are valid trees (extension-release.d). */
494 STRV_FOREACH(extension_directory
, extension_directories
) {
495 _cleanup_free_
char *mount_point
= NULL
, *source
= NULL
;
496 const char *e
= *extension_directory
;
497 bool ignore_enoent
= false;
499 /* Pick up the counter where the ExtensionImages left it. */
500 r
= asprintf(&mount_point
, "%s/%zu", extension_dir
, n
++);
504 /* Look for any prefixes */
505 if (startswith(e
, "-")) {
507 ignore_enoent
= true;
509 /* Ignore this for now */
510 if (startswith(e
, "+"))
517 for (size_t j
= 0; hierarchies
&& hierarchies
[j
]; ++j
) {
518 _cleanup_free_
char *prefixed_hierarchy
= NULL
, *escaped
= NULL
, *lowerdir
= NULL
;
520 prefixed_hierarchy
= path_join(mount_point
, hierarchies
[j
]);
521 if (!prefixed_hierarchy
)
524 escaped
= shell_escape(prefixed_hierarchy
, ",:");
528 /* Note that lowerdir= parameters are in 'reverse' order, so the
529 * top-most directory in the overlay comes first in the list. */
530 lowerdir
= strjoin(escaped
, ":", overlays
[j
]);
534 free_and_replace(overlays
[j
], lowerdir
);
537 *((*p
)++) = (MountEntry
) {
538 .path_malloc
= TAKE_PTR(mount_point
),
539 .source_malloc
= TAKE_PTR(source
),
540 .mode
= EXTENSION_DIRECTORIES
,
541 .ignore
= ignore_enoent
,
547 /* Then, for each hierarchy, prepare an overlay with the list of lowerdir= strings
549 for (size_t i
= 0; hierarchies
&& hierarchies
[i
]; ++i
) {
550 _cleanup_free_
char *prefixed_hierarchy
= NULL
;
552 prefixed_hierarchy
= path_join(root
, hierarchies
[i
]);
553 if (!prefixed_hierarchy
)
556 *((*p
)++) = (MountEntry
) {
557 .path_malloc
= TAKE_PTR(prefixed_hierarchy
),
558 .options_malloc
= TAKE_PTR(overlays
[i
]),
559 .mode
= OVERLAY_MOUNT
,
561 .ignore
= true, /* If the source image doesn't set the ignore bit it will fail earlier. */
568 static int append_tmpfs_mounts(MountEntry
**p
, const TemporaryFileSystem
*tmpfs
, size_t n
) {
571 for (size_t i
= 0; i
< n
; i
++) {
572 const TemporaryFileSystem
*t
= tmpfs
+ i
;
573 _cleanup_free_
char *o
= NULL
, *str
= NULL
;
578 if (!path_is_absolute(t
->path
))
579 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
580 "Path is not absolute: %s",
583 str
= strjoin("mode=0755" NESTED_TMPFS_LIMITS
",", t
->options
);
587 r
= mount_option_mangle(str
, MS_NODEV
|MS_STRICTATIME
, &flags
, &o
);
589 return log_debug_errno(r
, "Failed to parse mount option '%s': %m", str
);
591 ro
= flags
& MS_RDONLY
;
595 *((*p
)++) = (MountEntry
) {
596 .path_const
= t
->path
,
599 .options_malloc
= TAKE_PTR(o
),
607 static int append_static_mounts(MountEntry
**p
, const MountEntry
*mounts
, size_t n
, bool ignore_protect
) {
611 /* Adds a list of static pre-defined entries */
613 for (size_t i
= 0; i
< n
; i
++)
614 *((*p
)++) = (MountEntry
) {
615 .path_const
= mount_entry_path(mounts
+i
),
616 .mode
= mounts
[i
].mode
,
617 .ignore
= mounts
[i
].ignore
|| ignore_protect
,
623 static int append_protect_home(MountEntry
**p
, ProtectHome protect_home
, bool ignore_protect
) {
626 switch (protect_home
) {
628 case PROTECT_HOME_NO
:
631 case PROTECT_HOME_READ_ONLY
:
632 return append_static_mounts(p
, protect_home_read_only_table
, ELEMENTSOF(protect_home_read_only_table
), ignore_protect
);
634 case PROTECT_HOME_TMPFS
:
635 return append_static_mounts(p
, protect_home_tmpfs_table
, ELEMENTSOF(protect_home_tmpfs_table
), ignore_protect
);
637 case PROTECT_HOME_YES
:
638 return append_static_mounts(p
, protect_home_yes_table
, ELEMENTSOF(protect_home_yes_table
), ignore_protect
);
641 assert_not_reached();
645 static int append_protect_system(MountEntry
**p
, ProtectSystem protect_system
, bool ignore_protect
) {
648 switch (protect_system
) {
650 case PROTECT_SYSTEM_NO
:
653 case PROTECT_SYSTEM_STRICT
:
654 return append_static_mounts(p
, protect_system_strict_table
, ELEMENTSOF(protect_system_strict_table
), ignore_protect
);
656 case PROTECT_SYSTEM_YES
:
657 return append_static_mounts(p
, protect_system_yes_table
, ELEMENTSOF(protect_system_yes_table
), ignore_protect
);
659 case PROTECT_SYSTEM_FULL
:
660 return append_static_mounts(p
, protect_system_full_table
, ELEMENTSOF(protect_system_full_table
), ignore_protect
);
663 assert_not_reached();
667 static int mount_path_compare(const MountEntry
*a
, const MountEntry
*b
) {
670 /* ExtensionImages/Directories will be used by other mounts as a base, so sort them first
671 * regardless of the prefix - they are set up in the propagate directory anyway */
672 d
= -CMP(a
->mode
== EXTENSION_IMAGES
, b
->mode
== EXTENSION_IMAGES
);
675 d
= -CMP(a
->mode
== EXTENSION_DIRECTORIES
, b
->mode
== EXTENSION_DIRECTORIES
);
679 /* If the paths are not equal, then order prefixes first */
680 d
= path_compare(mount_entry_path(a
), mount_entry_path(b
));
684 /* If the paths are equal, check the mode */
685 return CMP((int) a
->mode
, (int) b
->mode
);
688 static int prefix_where_needed(MountEntry
*m
, size_t n
, const char *root_directory
) {
689 /* Prefixes all paths in the bind mount table with the root directory if the entry needs that. */
693 for (size_t i
= 0; i
< n
; i
++) {
699 s
= path_join(root_directory
, mount_entry_path(m
+i
));
703 mount_entry_consume_prefix(&m
[i
], s
);
709 static void drop_duplicates(MountEntry
*m
, size_t *n
) {
710 MountEntry
*f
, *t
, *previous
;
715 /* Drops duplicate entries. Expects that the array is properly ordered already. */
717 for (f
= m
, t
= m
, previous
= NULL
; f
< m
+ *n
; f
++) {
719 /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare()
720 * above. Note that we only drop duplicates that haven't been mounted yet. */
722 path_equal(mount_entry_path(f
), mount_entry_path(previous
)) &&
723 !f
->applied
&& !previous
->applied
) {
724 log_debug("%s (%s) is duplicate.", mount_entry_path(f
), mount_mode_to_string(f
->mode
));
725 /* Propagate the flags to the remaining entry */
726 previous
->read_only
= previous
->read_only
|| mount_entry_read_only(f
);
727 previous
->noexec
= previous
->noexec
|| mount_entry_noexec(f
);
728 previous
->exec
= previous
->exec
|| mount_entry_exec(f
);
741 static void drop_inaccessible(MountEntry
*m
, size_t *n
) {
743 const char *clear
= NULL
;
748 /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly
749 * ordered already. */
751 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
753 /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop
754 * it, as inaccessible paths really should drop the entire subtree. */
755 if (clear
&& path_startswith(mount_entry_path(f
), clear
)) {
756 log_debug("%s is masked by %s.", mount_entry_path(f
), clear
);
761 clear
= f
->mode
== INACCESSIBLE
? mount_entry_path(f
) : NULL
;
770 static void drop_nop(MountEntry
*m
, size_t *n
) {
776 /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the
777 * list is ordered by prefixes. */
779 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
781 /* Only suppress such subtrees for READONLY, READWRITE and READWRITE_IMPLICIT entries */
782 if (IN_SET(f
->mode
, READONLY
, READWRITE
, READWRITE_IMPLICIT
)) {
783 MountEntry
*found
= NULL
;
785 /* Now let's find the first parent of the entry we are looking at. */
786 for (MountEntry
*p
= PTR_SUB1(t
, m
); p
; p
= PTR_SUB1(p
, m
))
787 if (path_startswith(mount_entry_path(f
), mount_entry_path(p
))) {
792 /* We found it, let's see if it's the same mode, if so, we can drop this entry */
793 if (found
&& found
->mode
== f
->mode
) {
794 log_debug("%s (%s) is made redundant by %s (%s)",
795 mount_entry_path(f
), mount_mode_to_string(f
->mode
),
796 mount_entry_path(found
), mount_mode_to_string(found
->mode
));
809 static void drop_outside_root(const char *root_directory
, MountEntry
*m
, size_t *n
) {
819 /* Drops all mounts that are outside of the root directory. */
821 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
823 /* ExtensionImages/Directories bases are opened in /run/systemd/unit-extensions on the host */
824 if (!IN_SET(f
->mode
, EXTENSION_IMAGES
, EXTENSION_DIRECTORIES
) && !path_startswith(mount_entry_path(f
), root_directory
)) {
825 log_debug("%s is outside of root directory.", mount_entry_path(f
));
837 static int clone_device_node(
839 const char *temporary_mount
,
840 bool *make_devnode
) {
842 _cleanup_free_
char *sl
= NULL
;
843 const char *dn
, *bn
, *t
;
847 if (stat(d
, &st
) < 0) {
848 if (errno
== ENOENT
) {
849 log_debug_errno(errno
, "Device node '%s' to clone does not exist, ignoring.", d
);
853 return log_debug_errno(errno
, "Failed to stat() device node '%s' to clone, ignoring: %m", d
);
856 if (!S_ISBLK(st
.st_mode
) &&
857 !S_ISCHR(st
.st_mode
))
858 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
859 "Device node '%s' to clone is not a device node, ignoring.",
862 dn
= strjoina(temporary_mount
, d
);
864 /* First, try to create device node properly */
866 mac_selinux_create_file_prepare(d
, st
.st_mode
);
867 r
= mknod(dn
, st
.st_mode
, st
.st_rdev
);
868 mac_selinux_create_file_clear();
872 return log_debug_errno(errno
, "mknod failed for %s: %m", d
);
874 /* This didn't work, let's not try this again for the next iterations. */
875 *make_devnode
= false;
878 /* We're about to fall back to bind-mounting the device node. So create a dummy bind-mount target.
879 * Do not prepare device-node SELinux label (see issue 13762) */
880 r
= mknod(dn
, S_IFREG
, 0);
881 if (r
< 0 && errno
!= EEXIST
)
882 return log_debug_errno(errno
, "mknod() fallback failed for '%s': %m", d
);
884 /* Fallback to bind-mounting: The assumption here is that all used device nodes carry standard
885 * properties. Specifically, the devices nodes we bind-mount should either be owned by root:root or
886 * root:tty (e.g. /dev/tty, /dev/ptmx) and should not carry ACLs. */
887 r
= mount_nofollow_verbose(LOG_DEBUG
, d
, dn
, NULL
, MS_BIND
, NULL
);
892 bn
= path_startswith(d
, "/dev/");
896 /* Create symlinks like /dev/char/1:9 → ../urandom */
897 if (asprintf(&sl
, "%s/dev/%s/" DEVNUM_FORMAT_STR
,
899 S_ISCHR(st
.st_mode
) ? "char" : "block",
900 DEVNUM_FORMAT_VAL(st
.st_rdev
)) < 0)
903 (void) mkdir_parents(sl
, 0755);
905 t
= strjoina("../", bn
);
906 if (symlink(t
, sl
) < 0)
907 log_debug_errno(errno
, "Failed to symlink '%s' to '%s', ignoring: %m", t
, sl
);
912 static char *settle_runtime_dir(RuntimeScope scope
) {
915 if (scope
!= RUNTIME_SCOPE_USER
)
916 return strdup("/run/");
918 if (asprintf(&runtime_dir
, "/run/user/" UID_FMT
, geteuid()) < 0)
924 static int mount_private_dev(MountEntry
*m
, RuntimeScope scope
) {
925 static const char devnodes
[] =
933 _cleanup_free_
char *runtime_dir
= NULL
, *temporary_mount
= NULL
;
934 const char *dev
= NULL
, *devpts
= NULL
, *devshm
= NULL
, *devhugepages
= NULL
, *devmqueue
= NULL
, *devlog
= NULL
, *devptmx
= NULL
;
935 bool can_mknod
= true;
940 runtime_dir
= settle_runtime_dir(scope
);
942 return log_oom_debug();
944 temporary_mount
= path_join(runtime_dir
, "systemd/namespace-dev-XXXXXX");
945 if (!temporary_mount
)
946 return log_oom_debug();
948 if (!mkdtemp(temporary_mount
))
949 return log_debug_errno(errno
, "Failed to create temporary directory '%s': %m", temporary_mount
);
951 dev
= strjoina(temporary_mount
, "/dev");
952 (void) mkdir(dev
, 0755);
953 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", dev
, "tmpfs", DEV_MOUNT_OPTIONS
, "mode=0755" TMPFS_LIMITS_PRIVATE_DEV
);
957 r
= label_fix_full(AT_FDCWD
, dev
, "/dev", 0);
959 log_debug_errno(r
, "Failed to fix label of '%s' as /dev: %m", dev
);
963 devpts
= strjoina(temporary_mount
, "/dev/pts");
964 (void) mkdir(devpts
, 0755);
965 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev/pts", devpts
, NULL
, MS_BIND
, NULL
);
969 /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx.
970 * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible.
971 * Thus, in that case make a clone.
972 * In nspawn and other containers it will be a symlink, in that case make it a symlink. */
973 r
= is_symlink("/dev/ptmx");
975 log_debug_errno(r
, "Failed to detect whether /dev/ptmx is a symlink or not: %m");
978 devptmx
= strjoina(temporary_mount
, "/dev/ptmx");
979 if (symlink("pts/ptmx", devptmx
) < 0) {
980 r
= log_debug_errno(errno
, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx
);
984 r
= clone_device_node("/dev/ptmx", temporary_mount
, &can_mknod
);
989 devshm
= strjoina(temporary_mount
, "/dev/shm");
990 (void) mkdir(devshm
, 0755);
991 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev/shm", devshm
, NULL
, MS_BIND
, NULL
);
995 devmqueue
= strjoina(temporary_mount
, "/dev/mqueue");
996 (void) mkdir(devmqueue
, 0755);
997 (void) mount_nofollow_verbose(LOG_DEBUG
, "/dev/mqueue", devmqueue
, NULL
, MS_BIND
, NULL
);
999 devhugepages
= strjoina(temporary_mount
, "/dev/hugepages");
1000 (void) mkdir(devhugepages
, 0755);
1001 (void) mount_nofollow_verbose(LOG_DEBUG
, "/dev/hugepages", devhugepages
, NULL
, MS_BIND
, NULL
);
1003 devlog
= strjoina(temporary_mount
, "/dev/log");
1004 if (symlink("/run/systemd/journal/dev-log", devlog
) < 0)
1005 log_debug_errno(errno
, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog
);
1007 NULSTR_FOREACH(d
, devnodes
) {
1008 r
= clone_device_node(d
, temporary_mount
, &can_mknod
);
1009 /* ENXIO means the *source* is not a device file, skip creation in that case */
1010 if (r
< 0 && r
!= -ENXIO
)
1014 r
= dev_setup(temporary_mount
, UID_INVALID
, GID_INVALID
);
1016 log_debug_errno(r
, "Failed to set up basic device tree at '%s', ignoring: %m", temporary_mount
);
1018 /* Make the bind mount read-only. */
1019 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, dev
, NULL
, MS_REMOUNT
|MS_BIND
|MS_RDONLY
, NULL
);
1023 /* Create the /dev directory if missing. It is more likely to be missing when the service is started
1024 * with RootDirectory. This is consistent with mount units creating the mount points when missing. */
1025 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1027 /* Unmount everything in old /dev */
1028 r
= umount_recursive(mount_entry_path(m
), 0);
1030 log_debug_errno(r
, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m
));
1032 r
= mount_nofollow_verbose(LOG_DEBUG
, dev
, mount_entry_path(m
), NULL
, MS_MOVE
, NULL
);
1037 (void) rmdir(temporary_mount
);
1043 (void) umount_verbose(LOG_DEBUG
, devpts
, UMOUNT_NOFOLLOW
);
1046 (void) umount_verbose(LOG_DEBUG
, devshm
, UMOUNT_NOFOLLOW
);
1049 (void) umount_verbose(LOG_DEBUG
, devhugepages
, UMOUNT_NOFOLLOW
);
1052 (void) umount_verbose(LOG_DEBUG
, devmqueue
, UMOUNT_NOFOLLOW
);
1054 (void) umount_verbose(LOG_DEBUG
, dev
, UMOUNT_NOFOLLOW
);
1056 (void) rmdir(temporary_mount
);
1061 static int mount_bind_dev(const MountEntry
*m
) {
1066 /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the
1067 * service's /dev. This is only used when RootDirectory= is set. */
1069 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1071 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
1073 return log_debug_errno(r
, "Unable to determine whether /dev is already mounted: %m");
1074 if (r
> 0) /* make this a NOP if /dev is already a mount point */
1077 return mount_nofollow_verbose(LOG_DEBUG
, "/dev", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
);
1080 static int mount_bind_sysfs(const MountEntry
*m
) {
1085 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1087 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
1089 return log_debug_errno(r
, "Unable to determine whether /sys is already mounted: %m");
1090 if (r
> 0) /* make this a NOP if /sys is already a mount point */
1093 /* Bind mount the host's version so that we get all child mounts of it, too. */
1094 return mount_nofollow_verbose(LOG_DEBUG
, "/sys", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
);
1097 static int mount_private_sysfs(const MountEntry
*m
) {
1098 const char *entry_path
= mount_entry_path(ASSERT_PTR(m
));
1101 (void) mkdir_p_label(entry_path
, 0755);
1103 n
= umount_recursive(entry_path
, 0);
1105 r
= mount_nofollow_verbose(LOG_DEBUG
, "sysfs", entry_path
, "sysfs", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
1106 if (ERRNO_IS_NEG_PRIVILEGE(r
)) {
1107 /* When we do not have enough privileges to mount sysfs, fall back to use existing /sys. */
1110 /* /sys or some of sub-mounts are umounted in the above. Refuse incomplete tree.
1111 * Propagate the original error code returned by mount() in the above. */
1114 return mount_bind_sysfs(m
);
1119 /* We mounted a new instance now. Let's bind mount the children over now. */
1120 (void) bind_mount_submounts("/sys", entry_path
);
1124 static int mount_procfs(const MountEntry
*m
, const NamespaceInfo
*ns_info
) {
1125 _cleanup_free_
char *opts
= NULL
;
1126 const char *entry_path
;
1132 if (ns_info
->protect_proc
!= PROTECT_PROC_DEFAULT
||
1133 ns_info
->proc_subset
!= PROC_SUBSET_ALL
) {
1135 /* Starting with kernel 5.8 procfs' hidepid= logic is truly per-instance (previously it
1136 * pretended to be per-instance but actually was per-namespace), hence let's make use of it
1137 * if requested. To make sure this logic succeeds only on kernels where hidepid= is
1138 * per-instance, we'll exclusively use the textual value for hidepid=, since support was
1139 * added in the same commit: if it's supported it is thus also per-instance. */
1141 const char *hpv
= ns_info
->protect_proc
== PROTECT_PROC_DEFAULT
?
1143 protect_proc_to_string(ns_info
->protect_proc
);
1145 /* hidepid= support was added in 5.8, so we can use fsconfig()/fsopen() (which were added in
1146 * 5.2) to check if hidepid= is supported. This avoids a noisy dmesg log by the kernel when
1147 * trying to use hidepid= on systems where it isn't supported. The same applies for subset=.
1148 * fsopen()/fsconfig() was also backported on some distros which allows us to detect
1149 * hidepid=/subset= support in even more scenarios. */
1151 if (mount_option_supported("proc", "hidepid", hpv
) != 0) {
1152 opts
= strjoin("hidepid=", hpv
);
1157 if (ns_info
->proc_subset
== PROC_SUBSET_PID
&&
1158 mount_option_supported("proc", "subset", "pid") != 0)
1159 if (!strextend_with_separator(&opts
, ",", "subset=pid"))
1163 entry_path
= mount_entry_path(m
);
1164 (void) mkdir_p_label(entry_path
, 0755);
1166 /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in
1167 * one. i.e we don't reuse existing mounts here under any condition, we want a new instance owned by
1168 * our user namespace and with our hidepid= settings applied. Hence, let's get rid of everything
1169 * mounted on /proc/ first. */
1171 n
= umount_recursive(entry_path
, 0);
1173 r
= mount_nofollow_verbose(LOG_DEBUG
, "proc", entry_path
, "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, opts
);
1174 if (r
== -EINVAL
&& opts
)
1175 /* If this failed with EINVAL then this likely means the textual hidepid= stuff is
1176 * not supported by the kernel, and thus the per-instance hidepid= neither, which
1177 * means we really don't want to use it, since it would affect our host's /proc
1178 * mount. Hence let's gracefully fallback to a classic, unrestricted version. */
1179 r
= mount_nofollow_verbose(LOG_DEBUG
, "proc", entry_path
, "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
1180 if (ERRNO_IS_NEG_PRIVILEGE(r
)) {
1181 /* When we do not have enough privileges to mount /proc, fall back to use existing /proc. */
1184 /* /proc or some of sub-mounts are umounted in the above. Refuse incomplete tree.
1185 * Propagate the original error code returned by mount() in the above. */
1188 r
= path_is_mount_point(entry_path
, NULL
, 0);
1190 return log_debug_errno(r
, "Unable to determine whether /proc is already mounted: %m");
1194 /* We lack permissions to mount a new instance of /proc, and it is not already mounted. But
1195 * we can access the host's, so as a final fallback bind-mount it to the destination, as most
1196 * likely we are inside a user manager in an unprivileged user namespace. */
1197 return mount_nofollow_verbose(LOG_DEBUG
, "/proc", entry_path
, NULL
, MS_BIND
|MS_REC
, NULL
);
1202 /* We mounted a new instance now. Let's bind mount the children over now. This matters for nspawn
1203 * where a bunch of files are overmounted, in particular the boot id */
1204 (void) bind_mount_submounts("/proc", entry_path
);
1208 static int mount_tmpfs(const MountEntry
*m
) {
1209 const char *entry_path
, *inner_path
;
1214 entry_path
= mount_entry_path(m
);
1215 inner_path
= mount_entry_unprefixed_path(m
);
1217 /* First, get rid of everything that is below if there is anything. Then, overmount with our new
1220 (void) mkdir_p_label(entry_path
, 0755);
1221 (void) umount_recursive(entry_path
, 0);
1223 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", entry_path
, "tmpfs", m
->flags
, mount_entry_options(m
));
1227 r
= label_fix_full(AT_FDCWD
, entry_path
, inner_path
, 0);
1229 return log_debug_errno(r
, "Failed to fix label of '%s' as '%s': %m", entry_path
, inner_path
);
1234 static int mount_run(const MountEntry
*m
) {
1239 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
1240 if (r
< 0 && r
!= -ENOENT
)
1241 return log_debug_errno(r
, "Unable to determine whether /run is already mounted: %m");
1242 if (r
> 0) /* make this a NOP if /run is already a mount point */
1245 return mount_tmpfs(m
);
1248 static int mount_mqueuefs(const MountEntry
*m
) {
1250 const char *entry_path
;
1254 entry_path
= mount_entry_path(m
);
1256 (void) mkdir_p_label(entry_path
, 0755);
1257 (void) umount_recursive(entry_path
, 0);
1259 r
= mount_nofollow_verbose(LOG_DEBUG
, "mqueue", entry_path
, "mqueue", m
->flags
, mount_entry_options(m
));
1266 static int mount_image(
1267 const MountEntry
*m
,
1268 const char *root_directory
,
1269 const ImagePolicy
*image_policy
) {
1271 _cleanup_free_
char *host_os_release_id
= NULL
, *host_os_release_version_id
= NULL
,
1272 *host_os_release_level
= NULL
, *extension_name
= NULL
;
1273 _cleanup_strv_free_
char **extension_release
= NULL
;
1274 ImageClass
class = IMAGE_SYSEXT
;
1279 r
= path_extract_filename(mount_entry_source(m
), &extension_name
);
1281 return log_debug_errno(r
, "Failed to extract extension name from %s: %m", mount_entry_source(m
));
1283 if (m
->mode
== EXTENSION_IMAGES
) {
1284 r
= load_extension_release_pairs(mount_entry_source(m
), IMAGE_SYSEXT
, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1286 r
= load_extension_release_pairs(mount_entry_source(m
), IMAGE_CONFEXT
, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1288 class = IMAGE_CONFEXT
;
1293 r
= parse_os_release(
1294 empty_to_root(root_directory
),
1295 "ID", &host_os_release_id
,
1296 "VERSION_ID", &host_os_release_version_id
,
1297 image_class_info
[class].level_env
, &host_os_release_level
,
1300 return log_debug_errno(r
, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory
));
1301 if (isempty(host_os_release_id
))
1302 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'ID' field not found or empty in 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory
));
1305 r
= verity_dissect_and_mount(
1307 mount_entry_source(m
),
1308 mount_entry_path(m
),
1312 host_os_release_version_id
,
1313 host_os_release_level
,
1314 /* required_sysext_scope= */ NULL
,
1315 /* ret_image= */ NULL
);
1316 if (r
== -ENOENT
&& m
->ignore
)
1318 if (r
== -ESTALE
&& host_os_release_id
)
1319 return log_error_errno(r
,
1320 "Failed to mount image %s, extension-release metadata does not match the lower layer's: ID=%s%s%s%s%s",
1321 mount_entry_source(m
),
1323 host_os_release_version_id
? " VERSION_ID=" : "",
1324 strempty(host_os_release_version_id
),
1325 host_os_release_level
? image_class_info
[class].level_env_print
: "",
1326 strempty(host_os_release_level
));
1328 return log_debug_errno(r
, "Failed to mount image %s on %s: %m", mount_entry_source(m
), mount_entry_path(m
));
1333 static int mount_overlay(const MountEntry
*m
) {
1334 const char *options
;
1339 options
= strjoina("lowerdir=", mount_entry_options(m
));
1341 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1343 r
= mount_nofollow_verbose(LOG_DEBUG
, "overlay", mount_entry_path(m
), "overlay", MS_RDONLY
, options
);
1344 if (r
== -ENOENT
&& m
->ignore
)
1350 static int follow_symlink(
1351 const char *root_directory
,
1354 _cleanup_free_
char *target
= NULL
;
1357 /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we
1358 * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at
1359 * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the
1360 * end and already have a fully normalized name. */
1362 r
= chase(mount_entry_path(m
), root_directory
, CHASE_STEP
|CHASE_NONEXISTENT
, &target
, NULL
);
1364 return log_debug_errno(r
, "Failed to chase symlinks '%s': %m", mount_entry_path(m
));
1365 if (r
> 0) /* Reached the end, nothing more to resolve */
1368 if (m
->n_followed
>= CHASE_MAX
) /* put a boundary on things */
1369 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
1370 "Symlink loop on '%s'.",
1371 mount_entry_path(m
));
1373 log_debug("Followed mount entry path symlink %s %s %s.",
1374 mount_entry_path(m
), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT
), target
);
1376 mount_entry_consume_prefix(m
, TAKE_PTR(target
));
1383 static int apply_one_mount(
1384 const char *root_directory
,
1386 const ImagePolicy
*mount_image_policy
,
1387 const ImagePolicy
*extension_image_policy
,
1388 const NamespaceInfo
*ns_info
,
1389 RuntimeScope scope
) {
1391 _cleanup_free_
char *inaccessible
= NULL
;
1392 bool rbind
= true, make
= false;
1399 log_debug("Applying namespace mount on %s", mount_entry_path(m
));
1403 case INACCESSIBLE
: {
1404 _cleanup_free_
char *runtime_dir
= NULL
;
1407 /* First, get rid of everything that is below if there
1408 * is anything... Then, overmount it with an
1409 * inaccessible path. */
1410 (void) umount_recursive(mount_entry_path(m
), 0);
1412 if (lstat(mount_entry_path(m
), &target
) < 0) {
1413 if (errno
== ENOENT
&& m
->ignore
)
1416 return log_debug_errno(errno
, "Failed to lstat() %s to determine what to mount over it: %m",
1417 mount_entry_path(m
));
1420 /* We don't pass the literal runtime scope through here but one based purely on our UID. This
1421 * means that the root user's --user services will use the host's inaccessible inodes rather
1422 * then root's private ones. This is preferable since it means device nodes that are
1423 * overmounted to make them inaccessible will be overmounted with a device node, rather than
1424 * an AF_UNIX socket inode. */
1425 runtime_dir
= settle_runtime_dir(geteuid() == 0 ? RUNTIME_SCOPE_SYSTEM
: RUNTIME_SCOPE_USER
);
1427 return log_oom_debug();
1429 r
= mode_to_inaccessible_node(runtime_dir
, target
.st_mode
, &inaccessible
);
1431 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
1432 "File type not supported for inaccessible mounts. Note that symlinks are not allowed");
1433 what
= inaccessible
;
1439 case READWRITE_IMPLICIT
:
1442 r
= path_is_mount_point(mount_entry_path(m
), root_directory
, 0);
1443 if (r
== -ENOENT
&& m
->ignore
)
1446 return log_debug_errno(r
, "Failed to determine whether %s is already a mount point: %m",
1447 mount_entry_path(m
));
1448 if (r
> 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY
1449 * and MS_NOEXEC bits for the mount point if needed. */
1451 /* This isn't a mount point yet, let's make it one. */
1452 what
= mount_entry_path(m
);
1455 case EXTENSION_DIRECTORIES
: {
1456 _cleanup_free_
char *host_os_release_id
= NULL
, *host_os_release_version_id
= NULL
,
1457 *host_os_release_level
= NULL
, *extension_name
= NULL
;
1458 _cleanup_strv_free_
char **extension_release
= NULL
;
1459 ImageClass
class = IMAGE_SYSEXT
;
1461 r
= path_extract_filename(mount_entry_source(m
), &extension_name
);
1463 return log_debug_errno(r
, "Failed to extract extension name from %s: %m", mount_entry_source(m
));
1465 r
= load_extension_release_pairs(mount_entry_source(m
), IMAGE_SYSEXT
, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1467 r
= load_extension_release_pairs(mount_entry_source(m
), IMAGE_CONFEXT
, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1469 class = IMAGE_CONFEXT
;
1474 r
= parse_os_release(
1475 empty_to_root(root_directory
),
1476 "ID", &host_os_release_id
,
1477 "VERSION_ID", &host_os_release_version_id
,
1478 image_class_info
[class].level_env
, &host_os_release_level
,
1481 return log_debug_errno(r
, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory
));
1482 if (isempty(host_os_release_id
))
1483 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "'ID' field not found or empty in 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory
));
1485 r
= load_extension_release_pairs(mount_entry_source(m
), class, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1486 if (r
== -ENOENT
&& m
->ignore
)
1489 return log_debug_errno(r
, "Failed to parse directory %s extension-release metadata: %m", extension_name
);
1491 r
= extension_release_validate(
1494 host_os_release_version_id
,
1495 host_os_release_level
,
1496 /* host_extension_scope */ NULL
, /* Leave empty, we need to accept both system and portable */
1500 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE
), "Directory %s extension-release metadata does not match the root's", extension_name
);
1502 return log_debug_errno(r
, "Failed to compare directory %s extension-release metadata with the root's os-release: %m", extension_name
);
1511 case BIND_MOUNT_RECURSIVE
: {
1512 _cleanup_free_
char *chased
= NULL
;
1514 /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note
1515 * that bind mount source paths are always relative to the host root, hence we pass NULL as
1516 * root directory to chase() here. */
1518 r
= chase(mount_entry_source(m
), NULL
, CHASE_TRAIL_SLASH
, &chased
, NULL
);
1519 if (r
== -ENOENT
&& m
->ignore
) {
1520 log_debug_errno(r
, "Path %s does not exist, ignoring.", mount_entry_source(m
));
1524 return log_debug_errno(r
, "Failed to follow symlinks on %s: %m", mount_entry_source(m
));
1526 log_debug("Followed source symlinks %s %s %s.",
1527 mount_entry_source(m
), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT
), chased
);
1529 free_and_replace(m
->source_malloc
, chased
);
1531 what
= mount_entry_source(m
);
1538 return mount_tmpfs(m
);
1541 case PRIVATE_TMP_READONLY
:
1542 what
= mount_entry_source(m
);
1547 return mount_private_dev(m
, scope
);
1550 return mount_bind_dev(m
);
1553 return mount_private_sysfs(m
);
1556 return mount_bind_sysfs(m
);
1559 return mount_procfs(m
, ns_info
);
1562 return mount_run(m
);
1565 return mount_mqueuefs(m
);
1568 return mount_image(m
, NULL
, mount_image_policy
);
1570 case EXTENSION_IMAGES
:
1571 return mount_image(m
, root_directory
, extension_image_policy
);
1574 return mount_overlay(m
);
1577 assert_not_reached();
1582 r
= mount_nofollow_verbose(LOG_DEBUG
, what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
);
1584 bool try_again
= false;
1586 if (r
== -ENOENT
&& make
) {
1589 /* Hmm, either the source or the destination are missing. Let's see if we can create
1590 the destination, then try again. */
1592 (void) mkdir_parents(mount_entry_path(m
), 0755);
1594 q
= make_mount_point_inode_from_path(what
, mount_entry_path(m
), 0755);
1595 if (q
< 0 && q
!= -EEXIST
)
1596 log_error_errno(q
, "Failed to create destination mount point node '%s': %m",
1597 mount_entry_path(m
));
1603 r
= mount_nofollow_verbose(LOG_DEBUG
, what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
);
1605 return log_error_errno(r
, "Failed to mount %s to %s: %m", what
, mount_entry_path(m
));
1608 log_debug("Successfully mounted %s to %s", what
, mount_entry_path(m
));
1612 static int make_read_only(const MountEntry
*m
, char **deny_list
, FILE *proc_self_mountinfo
) {
1613 unsigned long new_flags
= 0, flags_mask
= 0;
1618 assert(proc_self_mountinfo
);
1620 if (mount_entry_read_only(m
) || m
->mode
== PRIVATE_DEV
) {
1621 new_flags
|= MS_RDONLY
;
1622 flags_mask
|= MS_RDONLY
;
1626 new_flags
|= MS_NOSUID
;
1627 flags_mask
|= MS_NOSUID
;
1630 if (flags_mask
== 0) /* No Change? */
1633 /* We generally apply these changes recursively, except for /dev, and the cases we know there's
1634 * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the
1635 * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace
1636 * and running Linux <= 4.17. */
1638 mount_entry_read_only(m
) &&
1639 !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1641 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, deny_list
, proc_self_mountinfo
);
1643 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, proc_self_mountinfo
);
1645 /* Note that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked
1646 * read-only already stays this way. This improves compatibility with container managers, where we
1647 * won't attempt to undo read-only mounts already applied. */
1649 if (r
== -ENOENT
&& m
->ignore
)
1652 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1653 submounts
? " and its submounts" : "");
1657 static int make_noexec(const MountEntry
*m
, char **deny_list
, FILE *proc_self_mountinfo
) {
1658 unsigned long new_flags
= 0, flags_mask
= 0;
1663 assert(proc_self_mountinfo
);
1665 if (mount_entry_noexec(m
)) {
1666 new_flags
|= MS_NOEXEC
;
1667 flags_mask
|= MS_NOEXEC
;
1668 } else if (mount_entry_exec(m
)) {
1669 new_flags
&= ~MS_NOEXEC
;
1670 flags_mask
|= MS_NOEXEC
;
1673 if (flags_mask
== 0) /* No Change? */
1676 submounts
= !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1679 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, deny_list
, proc_self_mountinfo
);
1681 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, proc_self_mountinfo
);
1683 if (r
== -ENOENT
&& m
->ignore
)
1686 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1687 submounts
? " and its submounts" : "");
1691 static int make_nosuid(const MountEntry
*m
, FILE *proc_self_mountinfo
) {
1696 assert(proc_self_mountinfo
);
1698 submounts
= !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1701 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), MS_NOSUID
, MS_NOSUID
, NULL
, proc_self_mountinfo
);
1703 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), MS_NOSUID
, MS_NOSUID
, proc_self_mountinfo
);
1704 if (r
== -ENOENT
&& m
->ignore
)
1707 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1708 submounts
? " and its submounts" : "");
1712 static bool namespace_info_mount_apivfs(const NamespaceInfo
*ns_info
) {
1716 * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
1717 * since to protect the API VFS mounts, they need to be around in the
1721 return ns_info
->mount_apivfs
||
1722 ns_info
->protect_control_groups
||
1723 ns_info
->protect_kernel_tunables
||
1724 ns_info
->protect_proc
!= PROTECT_PROC_DEFAULT
||
1725 ns_info
->proc_subset
!= PROC_SUBSET_ALL
;
1728 static size_t namespace_calculate_mounts(
1729 const NamespaceInfo
*ns_info
,
1730 char** read_write_paths
,
1731 char** read_only_paths
,
1732 char** inaccessible_paths
,
1734 char** no_exec_paths
,
1735 char** empty_directories
,
1736 size_t n_bind_mounts
,
1737 size_t n_temporary_filesystems
,
1738 size_t n_mount_images
,
1739 size_t n_extension_images
,
1740 size_t n_extension_directories
,
1741 size_t n_hierarchies
,
1742 const char* tmp_dir
,
1743 const char* var_tmp_dir
,
1744 const char *creds_path
,
1745 const char* log_namespace
,
1746 bool setup_propagate
,
1747 const char* notify_socket
,
1748 const char* host_os_release
) {
1750 size_t protect_home_cnt
;
1751 size_t protect_system_cnt
=
1752 (ns_info
->protect_system
== PROTECT_SYSTEM_STRICT
?
1753 ELEMENTSOF(protect_system_strict_table
) :
1754 ((ns_info
->protect_system
== PROTECT_SYSTEM_FULL
) ?
1755 ELEMENTSOF(protect_system_full_table
) :
1756 ((ns_info
->protect_system
== PROTECT_SYSTEM_YES
) ?
1757 ELEMENTSOF(protect_system_yes_table
) : 0)));
1760 (ns_info
->protect_home
== PROTECT_HOME_YES
?
1761 ELEMENTSOF(protect_home_yes_table
) :
1762 ((ns_info
->protect_home
== PROTECT_HOME_READ_ONLY
) ?
1763 ELEMENTSOF(protect_home_read_only_table
) :
1764 ((ns_info
->protect_home
== PROTECT_HOME_TMPFS
) ?
1765 ELEMENTSOF(protect_home_tmpfs_table
) : 0)));
1767 return !!tmp_dir
+ !!var_tmp_dir
+
1768 strv_length(read_write_paths
) +
1769 strv_length(read_only_paths
) +
1770 strv_length(inaccessible_paths
) +
1771 strv_length(exec_paths
) +
1772 strv_length(no_exec_paths
) +
1773 strv_length(empty_directories
) +
1776 (n_extension_images
> 0 || n_extension_directories
> 0 ? /* Mount each image and directory plus an overlay per hierarchy */
1777 n_hierarchies
+ n_extension_images
+ n_extension_directories
: 0) +
1778 n_temporary_filesystems
+
1779 ns_info
->private_dev
+
1780 (ns_info
->protect_kernel_tunables
?
1781 ELEMENTSOF(protect_kernel_tunables_proc_table
) + ELEMENTSOF(protect_kernel_tunables_sys_table
) : 0) +
1782 (ns_info
->protect_kernel_modules
? ELEMENTSOF(protect_kernel_modules_table
) : 0) +
1783 (ns_info
->protect_kernel_logs
?
1784 ELEMENTSOF(protect_kernel_logs_proc_table
) + ELEMENTSOF(protect_kernel_logs_dev_table
) : 0) +
1785 (ns_info
->protect_control_groups
? 1 : 0) +
1786 protect_home_cnt
+ protect_system_cnt
+
1787 (ns_info
->protect_hostname
? 2 : 0) +
1788 (namespace_info_mount_apivfs(ns_info
) ? ELEMENTSOF(apivfs_table
) : 0) +
1789 (creds_path
? 2 : 1) +
1791 setup_propagate
+ /* /run/systemd/incoming */
1794 ns_info
->private_network
+ /* /sys */
1795 ns_info
->private_ipc
; /* /dev/mqueue */
1798 /* Walk all mount entries and dropping any unused mounts. This affects all
1800 * - that are implicitly protected by a path that has been rendered inaccessible
1801 * - whose immediate parent requests the same protection mode as the mount itself
1802 * - that are outside of the relevant root directory
1803 * - which are duplicates
1805 static void drop_unused_mounts(const char *root_directory
, MountEntry
*mounts
, size_t *n_mounts
) {
1806 assert(root_directory
);
1808 assert(mounts
|| *n_mounts
== 0);
1810 typesafe_qsort(mounts
, *n_mounts
, mount_path_compare
);
1812 drop_duplicates(mounts
, n_mounts
);
1813 drop_outside_root(root_directory
, mounts
, n_mounts
);
1814 drop_inaccessible(mounts
, n_mounts
);
1815 drop_nop(mounts
, n_mounts
);
1818 static int create_symlinks_from_tuples(const char *root
, char **strv_symlinks
) {
1821 STRV_FOREACH_PAIR(src
, dst
, strv_symlinks
) {
1822 _cleanup_free_
char *src_abs
= NULL
, *dst_abs
= NULL
;
1824 src_abs
= path_join(root
, *src
);
1825 dst_abs
= path_join(root
, *dst
);
1826 if (!src_abs
|| !dst_abs
)
1829 r
= mkdir_parents_label(dst_abs
, 0755);
1833 r
= symlink_idempotent(src_abs
, dst_abs
, true);
1841 static void mount_entry_path_debug_string(const char *root
, MountEntry
*m
, char **error_path
) {
1844 /* Create a string suitable for debugging logs, stripping for example the local working directory.
1845 * For example, with a BindPaths=/var/bar that does not exist on the host:
1848 * foo.service: Failed to set up mount namespacing: /run/systemd/unit-root/var/bar: No such file or directory
1850 * foo.service: Failed to set up mount namespacing: /var/bar: No such file or directory
1852 * Note that this is an error path, so no OOM check is done on purpose. */
1857 if (!mount_entry_path(m
)) {
1863 const char *e
= startswith(mount_entry_path(m
), root
);
1865 *error_path
= strdup(e
);
1870 *error_path
= strdup(mount_entry_path(m
));
1874 static int apply_mounts(
1876 const ImagePolicy
*mount_image_policy
,
1877 const ImagePolicy
*extension_image_policy
,
1878 const NamespaceInfo
*ns_info
,
1883 char **error_path
) {
1885 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
1886 _cleanup_free_
char **deny_list
= NULL
;
1889 if (n_mounts
== 0) /* Shortcut: nothing to do */
1896 /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of
1897 * /proc. For example, this is the case with the option: 'InaccessiblePaths=/proc'. */
1898 proc_self_mountinfo
= fopen("/proc/self/mountinfo", "re");
1899 if (!proc_self_mountinfo
) {
1903 *error_path
= strdup("/proc/self/mountinfo");
1905 return log_debug_errno(r
, "Failed to open /proc/self/mountinfo: %m");
1908 /* First round, establish all mounts we need */
1912 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1917 /* ExtensionImages/Directories are first opened in the propagate directory, not in the root_directory */
1918 r
= follow_symlink(!IN_SET(m
->mode
, EXTENSION_IMAGES
, EXTENSION_DIRECTORIES
) ? root
: NULL
, m
);
1920 mount_entry_path_debug_string(root
, m
, error_path
);
1924 /* We hit a symlinked mount point. The entry got rewritten and might
1925 * point to a very different place now. Let's normalize the changed
1926 * list, and start from the beginning. After all to mount the entry
1927 * at the new location we might need some other mounts first */
1932 r
= apply_one_mount(root
, m
, mount_image_policy
, extension_image_policy
, ns_info
, scope
);
1934 mount_entry_path_debug_string(root
, m
, error_path
);
1944 drop_unused_mounts(root
, mounts
, n_mounts
);
1947 /* Now that all filesystems have been set up, but before the
1948 * read-only switches are flipped, create the exec dirs and other symlinks.
1949 * Note that when /var/lib is not empty/tmpfs, these symlinks will already
1950 * exist, which means this will be a no-op. */
1951 r
= create_symlinks_from_tuples(root
, symlinks
);
1953 return log_debug_errno(r
, "Failed to set up symlinks inside mount namespace: %m");
1955 /* Create a deny list we can pass to bind_mount_recursive() */
1956 deny_list
= new(char*, (*n_mounts
)+1);
1959 for (size_t j
= 0; j
< *n_mounts
; j
++)
1960 deny_list
[j
] = (char*) mount_entry_path(mounts
+j
);
1961 deny_list
[*n_mounts
] = NULL
;
1963 /* Second round, flip the ro bits if necessary. */
1964 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1965 r
= make_read_only(m
, deny_list
, proc_self_mountinfo
);
1967 mount_entry_path_debug_string(root
, m
, error_path
);
1972 /* Third round, flip the noexec bits with a simplified deny list. */
1973 for (size_t j
= 0; j
< *n_mounts
; j
++)
1974 if (IN_SET((mounts
+j
)->mode
, EXEC
, NOEXEC
))
1975 deny_list
[j
] = (char*) mount_entry_path(mounts
+j
);
1976 deny_list
[*n_mounts
] = NULL
;
1978 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1979 r
= make_noexec(m
, deny_list
, proc_self_mountinfo
);
1981 mount_entry_path_debug_string(root
, m
, error_path
);
1986 /* Fourth round, flip the nosuid bits without a deny list. */
1987 if (ns_info
->mount_nosuid
)
1988 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1989 r
= make_nosuid(m
, proc_self_mountinfo
);
1991 mount_entry_path_debug_string(root
, m
, error_path
);
1999 static bool root_read_only(
2000 char **read_only_paths
,
2001 ProtectSystem protect_system
) {
2003 /* Determine whether the root directory is going to be read-only given the configured settings. */
2005 if (protect_system
== PROTECT_SYSTEM_STRICT
)
2008 if (prefixed_path_strv_contains(read_only_paths
, "/"))
2014 static bool home_read_only(
2015 char** read_only_paths
,
2016 char** inaccessible_paths
,
2017 char** empty_directories
,
2018 const BindMount
*bind_mounts
,
2019 size_t n_bind_mounts
,
2020 const TemporaryFileSystem
*temporary_filesystems
,
2021 size_t n_temporary_filesystems
,
2022 ProtectHome protect_home
) {
2024 /* Determine whether the /home directory is going to be read-only given the configured settings. Yes,
2025 * this is a bit sloppy, since we don't bother checking for cases where / is affected by multiple
2028 if (protect_home
!= PROTECT_HOME_NO
)
2031 if (prefixed_path_strv_contains(read_only_paths
, "/home") ||
2032 prefixed_path_strv_contains(inaccessible_paths
, "/home") ||
2033 prefixed_path_strv_contains(empty_directories
, "/home"))
2036 for (size_t i
= 0; i
< n_temporary_filesystems
; i
++)
2037 if (path_equal(temporary_filesystems
[i
].path
, "/home"))
2040 /* If /home is overmounted with some dir from the host it's not writable. */
2041 for (size_t i
= 0; i
< n_bind_mounts
; i
++)
2042 if (path_equal(bind_mounts
[i
].destination
, "/home"))
2048 int setup_namespace(
2049 const char* root_directory
,
2050 const char* root_image
,
2051 const MountOptions
*root_image_mount_options
,
2052 const ImagePolicy
*root_image_policy
,
2053 const NamespaceInfo
*ns_info
,
2054 char** read_write_paths
,
2055 char** read_only_paths
,
2056 char** inaccessible_paths
,
2058 char** no_exec_paths
,
2059 char** empty_directories
,
2061 const BindMount
*bind_mounts
,
2062 size_t n_bind_mounts
,
2063 const TemporaryFileSystem
*temporary_filesystems
,
2064 size_t n_temporary_filesystems
,
2065 const MountImage
*mount_images
,
2066 size_t n_mount_images
,
2067 const ImagePolicy
*mount_image_policy
,
2068 const char* tmp_dir
,
2069 const char* var_tmp_dir
,
2070 const char *creds_path
,
2071 const char *log_namespace
,
2072 unsigned long mount_propagation_flag
,
2073 VeritySettings
*verity
,
2074 const MountImage
*extension_images
,
2075 size_t n_extension_images
,
2076 const ImagePolicy
*extension_image_policy
,
2077 char **extension_directories
,
2078 const char *propagate_dir
,
2079 const char *incoming_dir
,
2080 const char *extension_dir
,
2081 const char *notify_socket
,
2082 const char *host_os_release_stage
,
2084 char **error_path
) {
2086 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
2087 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
2088 _cleanup_strv_free_
char **hierarchies
= NULL
;
2089 MountEntry
*m
= NULL
, *mounts
= NULL
;
2090 bool require_prefix
= false, setup_propagate
= false;
2092 DissectImageFlags dissect_image_flags
=
2093 DISSECT_IMAGE_GENERIC_ROOT
|
2094 DISSECT_IMAGE_REQUIRE_ROOT
|
2095 DISSECT_IMAGE_DISCARD_ON_LOOP
|
2096 DISSECT_IMAGE_RELAX_VAR_CHECK
|
2097 DISSECT_IMAGE_FSCK
|
2098 DISSECT_IMAGE_USR_NO_ROOT
|
2099 DISSECT_IMAGE_GROWFS
|
2100 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
2101 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
2107 /* Make sure that all mknod(), mkdir() calls we do are unaffected by the umask, and the access modes
2108 * we configure take effect */
2109 BLOCK_WITH_UMASK(0000);
2111 if (!isempty(propagate_dir
) && !isempty(incoming_dir
))
2112 setup_propagate
= true;
2114 if (mount_propagation_flag
== 0)
2115 mount_propagation_flag
= MS_SHARED
;
2118 /* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */
2119 if (root_read_only(read_only_paths
,
2120 ns_info
->protect_system
) &&
2121 home_read_only(read_only_paths
, inaccessible_paths
, empty_directories
,
2122 bind_mounts
, n_bind_mounts
, temporary_filesystems
, n_temporary_filesystems
,
2123 ns_info
->protect_home
) &&
2124 strv_isempty(read_write_paths
))
2125 dissect_image_flags
|= DISSECT_IMAGE_READ_ONLY
;
2127 SET_FLAG(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
, verity
&& verity
->data_path
);
2129 r
= loop_device_make_by_path(
2131 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_DEVICE_READ_ONLY
) ? O_RDONLY
: -1 /* < 0 means writable if possible, read-only as fallback */,
2132 /* sector_size= */ UINT32_MAX
,
2133 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
2137 return log_debug_errno(r
, "Failed to create loop device for root image: %m");
2139 r
= dissect_loop_device(
2142 root_image_mount_options
,
2144 dissect_image_flags
,
2147 return log_debug_errno(r
, "Failed to dissect image: %m");
2149 r
= dissected_image_load_verity_sig_partition(
2156 r
= dissected_image_decrypt(
2160 dissect_image_flags
);
2162 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
2166 root
= root_directory
;
2168 /* /run/systemd should have been created by PID 1 early on already, but in some cases, like
2169 * when running tests (test-execute), it might not have been created yet so let's make sure
2170 * we create it if it doesn't already exist. */
2171 (void) mkdir_p_label("/run/systemd", 0755);
2173 /* Always create the mount namespace in a temporary directory, instead of operating directly
2174 * in the root. The temporary directory prevents any mounts from being potentially obscured
2175 * my other mounts we already applied. We use the same mount point for all images, which is
2176 * safe, since they all live in their own namespaces after all, and hence won't see each
2177 * other. (Note: this directory is also created by PID 1 early on, we create it here for
2178 * similar reasons as /run/systemd/ first.) */
2179 root
= "/run/systemd/mount-rootfs";
2180 (void) mkdir_label(root
, 0555);
2182 require_prefix
= true;
2185 if (n_extension_images
> 0 || !strv_isempty(extension_directories
)) {
2186 /* Hierarchy population needs to be done for sysext and confext extension images */
2187 r
= parse_env_extension_hierarchies(&hierarchies
, "SYSTEMD_SYSEXT_AND_CONFEXT_HIERARCHIES");
2192 n_mounts
= namespace_calculate_mounts(
2201 n_temporary_filesystems
,
2204 strv_length(extension_directories
),
2205 strv_length(hierarchies
),
2206 tmp_dir
, var_tmp_dir
,
2211 host_os_release_stage
);
2214 m
= mounts
= new0(MountEntry
, n_mounts
);
2218 r
= append_access_mounts(&m
, read_write_paths
, READWRITE
, require_prefix
);
2222 r
= append_access_mounts(&m
, read_only_paths
, READONLY
, require_prefix
);
2226 r
= append_access_mounts(&m
, inaccessible_paths
, INACCESSIBLE
, require_prefix
);
2230 r
= append_access_mounts(&m
, exec_paths
, EXEC
, require_prefix
);
2234 r
= append_access_mounts(&m
, no_exec_paths
, NOEXEC
, require_prefix
);
2238 r
= append_empty_dir_mounts(&m
, empty_directories
);
2242 r
= append_bind_mounts(&m
, bind_mounts
, n_bind_mounts
);
2246 r
= append_tmpfs_mounts(&m
, temporary_filesystems
, n_temporary_filesystems
);
2251 bool ro
= streq(tmp_dir
, RUN_SYSTEMD_EMPTY
);
2253 *(m
++) = (MountEntry
) {
2254 .path_const
= "/tmp",
2255 .mode
= ro
? PRIVATE_TMP_READONLY
: PRIVATE_TMP
,
2256 .source_const
= tmp_dir
,
2261 bool ro
= streq(var_tmp_dir
, RUN_SYSTEMD_EMPTY
);
2263 *(m
++) = (MountEntry
) {
2264 .path_const
= "/var/tmp",
2265 .mode
= ro
? PRIVATE_TMP_READONLY
: PRIVATE_TMP
,
2266 .source_const
= var_tmp_dir
,
2270 r
= append_mount_images(&m
, mount_images
, n_mount_images
);
2274 r
= append_extensions(&m
, root
, extension_dir
, hierarchies
, extension_images
, n_extension_images
, extension_directories
);
2278 if (ns_info
->private_dev
)
2279 *(m
++) = (MountEntry
) {
2280 .path_const
= "/dev",
2281 .mode
= PRIVATE_DEV
,
2282 .flags
= DEV_MOUNT_OPTIONS
,
2285 /* In case /proc is successfully mounted with pid tree subset only (ProcSubset=pid), the
2286 protective mounts to non-pid /proc paths would fail. But the pid only option may have
2287 failed gracefully, so let's try the mounts but it's not fatal if they don't succeed. */
2288 bool ignore_protect_proc
= ns_info
->ignore_protect_paths
|| ns_info
->proc_subset
== PROC_SUBSET_PID
;
2289 if (ns_info
->protect_kernel_tunables
) {
2290 r
= append_static_mounts(&m
,
2291 protect_kernel_tunables_proc_table
,
2292 ELEMENTSOF(protect_kernel_tunables_proc_table
),
2293 ignore_protect_proc
);
2297 r
= append_static_mounts(&m
,
2298 protect_kernel_tunables_sys_table
,
2299 ELEMENTSOF(protect_kernel_tunables_sys_table
),
2300 ns_info
->ignore_protect_paths
);
2305 if (ns_info
->protect_kernel_modules
) {
2306 r
= append_static_mounts(&m
,
2307 protect_kernel_modules_table
,
2308 ELEMENTSOF(protect_kernel_modules_table
),
2309 ns_info
->ignore_protect_paths
);
2314 if (ns_info
->protect_kernel_logs
) {
2315 r
= append_static_mounts(&m
,
2316 protect_kernel_logs_proc_table
,
2317 ELEMENTSOF(protect_kernel_logs_proc_table
),
2318 ignore_protect_proc
);
2322 r
= append_static_mounts(&m
,
2323 protect_kernel_logs_dev_table
,
2324 ELEMENTSOF(protect_kernel_logs_dev_table
),
2325 ns_info
->ignore_protect_paths
);
2330 if (ns_info
->protect_control_groups
)
2331 *(m
++) = (MountEntry
) {
2332 .path_const
= "/sys/fs/cgroup",
2336 r
= append_protect_home(&m
, ns_info
->protect_home
, ns_info
->ignore_protect_paths
);
2340 r
= append_protect_system(&m
, ns_info
->protect_system
, false);
2344 if (namespace_info_mount_apivfs(ns_info
)) {
2345 r
= append_static_mounts(&m
,
2347 ELEMENTSOF(apivfs_table
),
2348 ns_info
->ignore_protect_paths
);
2353 /* Note, if proc is mounted with subset=pid then neither of the
2354 * two paths will exist, i.e. they are implicitly protected by
2355 * the mount option. */
2356 if (ns_info
->protect_hostname
) {
2357 *(m
++) = (MountEntry
) {
2358 .path_const
= "/proc/sys/kernel/hostname",
2360 .ignore
= ignore_protect_proc
,
2362 *(m
++) = (MountEntry
) {
2363 .path_const
= "/proc/sys/kernel/domainname",
2365 .ignore
= ignore_protect_proc
,
2369 if (ns_info
->private_network
)
2370 *(m
++) = (MountEntry
) {
2371 .path_const
= "/sys",
2372 .mode
= PRIVATE_SYSFS
,
2375 if (ns_info
->private_ipc
)
2376 *(m
++) = (MountEntry
) {
2377 .path_const
= "/dev/mqueue",
2379 .flags
= MS_NOSUID
| MS_NODEV
| MS_NOEXEC
| MS_RELATIME
,
2383 /* If our service has a credentials store configured, then bind that one in, but hide
2384 * everything else. */
2386 *(m
++) = (MountEntry
) {
2387 .path_const
= "/run/credentials",
2390 .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
,
2391 .flags
= MS_NODEV
|MS_STRICTATIME
|MS_NOSUID
|MS_NOEXEC
,
2394 *(m
++) = (MountEntry
) {
2395 .path_const
= creds_path
,
2398 .source_const
= creds_path
,
2402 /* If our service has no credentials store configured, then make the whole
2403 * credentials tree inaccessible wholesale. */
2405 *(m
++) = (MountEntry
) {
2406 .path_const
= "/run/credentials",
2407 .mode
= INACCESSIBLE
,
2412 if (log_namespace
) {
2413 _cleanup_free_
char *q
= NULL
;
2415 q
= strjoin("/run/systemd/journal.", log_namespace
);
2421 *(m
++) = (MountEntry
) {
2422 .path_const
= "/run/systemd/journal",
2423 .mode
= BIND_MOUNT_RECURSIVE
,
2425 .source_malloc
= TAKE_PTR(q
),
2429 /* Will be used to add bind mounts at runtime */
2430 if (setup_propagate
)
2431 *(m
++) = (MountEntry
) {
2432 .source_const
= propagate_dir
,
2433 .path_const
= incoming_dir
,
2439 *(m
++) = (MountEntry
) {
2440 .path_const
= notify_socket
,
2441 .source_const
= notify_socket
,
2446 if (host_os_release_stage
)
2447 *(m
++) = (MountEntry
) {
2448 .path_const
= "/run/host/.os-release-stage/",
2449 .source_const
= host_os_release_stage
,
2452 .ignore
= true, /* Live copy, don't hard-fail if it goes missing */
2455 assert(mounts
+ n_mounts
== m
);
2457 /* Prepend the root directory where that's necessary */
2458 r
= prefix_where_needed(mounts
, n_mounts
, root
);
2462 drop_unused_mounts(root
, mounts
, &n_mounts
);
2465 /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */
2467 if (unshare(CLONE_NEWNS
) < 0) {
2468 r
= log_debug_errno(errno
, "Failed to unshare the mount namespace: %m");
2469 if (ERRNO_IS_PRIVILEGE(r
) ||
2470 ERRNO_IS_NOT_SUPPORTED(r
))
2471 /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter
2472 * in place that doesn't allow us to create namespaces (or a missing cap), then
2473 * propagate a recognizable error back, which the caller can use to detect this case
2474 * (and only this) and optionally continue without namespacing applied. */
2480 /* Create the source directory to allow runtime propagation of mounts */
2481 if (setup_propagate
)
2482 (void) mkdir_p(propagate_dir
, 0600);
2484 if (n_extension_images
> 0 || !strv_isempty(extension_directories
))
2485 /* ExtensionImages/Directories mountpoint directories will be created while parsing the
2486 * mounts to create, so have the parent ready */
2487 (void) mkdir_p(extension_dir
, 0600);
2489 /* Remount / as SLAVE so that nothing now mounted in the namespace
2490 * shows up in the parent */
2491 if (mount(NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
) < 0) {
2492 r
= log_debug_errno(errno
, "Failed to remount '/' as SLAVE: %m");
2497 /* A root image is specified, mount it to the right place */
2498 r
= dissected_image_mount(
2501 /* uid_shift= */ UID_INVALID
,
2502 /* uid_range= */ UID_INVALID
,
2503 /* userns_fd= */ -EBADF
,
2504 dissect_image_flags
);
2506 log_debug_errno(r
, "Failed to mount root image: %m");
2510 /* Now release the block device lock, so that udevd is free to call BLKRRPART on the device
2512 r
= loop_device_flock(loop_device
, LOCK_UN
);
2514 log_debug_errno(r
, "Failed to release lock on loopback block device: %m");
2518 r
= dissected_image_relinquish(dissected_image
);
2520 log_debug_errno(r
, "Failed to relinquish dissected image: %m");
2524 } else if (root_directory
) {
2526 /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
2527 r
= path_is_mount_point(root
, NULL
, AT_SYMLINK_FOLLOW
);
2529 log_debug_errno(r
, "Failed to detect that %s is a mount point or not: %m", root
);
2533 r
= mount_nofollow_verbose(LOG_DEBUG
, root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2539 /* Let's mount the main root directory to the root directory to use */
2540 r
= mount_nofollow_verbose(LOG_DEBUG
, "/", root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2545 /* Try to set up the new root directory before mounting anything else there. */
2546 if (root_image
|| root_directory
)
2547 (void) base_filesystem_create(root
, UID_INVALID
, GID_INVALID
);
2549 /* Now make the magic happen */
2550 r
= apply_mounts(root
,
2552 extension_image_policy
,
2561 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
2562 r
= mount_switch_root(root
, /* mount_propagation_flag = */ 0);
2563 if (r
== -EINVAL
&& root_directory
) {
2564 /* If we are using root_directory and we don't have privileges (ie: user manager in a user
2565 * namespace) and the root_directory is already a mount point in the parent namespace,
2566 * MS_MOVE will fail as we don't have permission to change it (with EINVAL rather than
2567 * EPERM). Attempt to bind-mount it over itself (like we do above if it's not already a
2568 * mount point) and try again. */
2569 r
= mount_nofollow_verbose(LOG_DEBUG
, root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2572 r
= mount_switch_root(root
, /* mount_propagation_flag = */ 0);
2575 log_debug_errno(r
, "Failed to mount root with MS_MOVE: %m");
2579 /* Remount / as the desired mode. Note that this will not reestablish propagation from our side to
2580 * the host, since what's disconnected is disconnected. */
2581 if (mount(NULL
, "/", NULL
, mount_propagation_flag
| MS_REC
, NULL
) < 0) {
2582 r
= log_debug_errno(errno
, "Failed to remount '/' with desired mount flags: %m");
2586 /* bind_mount_in_namespace() will MS_MOVE into that directory, and that's only
2587 * supported for non-shared mounts. This needs to happen after remounting / or it will fail. */
2588 if (setup_propagate
) {
2589 r
= mount(NULL
, incoming_dir
, NULL
, MS_SLAVE
, NULL
);
2591 log_error_errno(r
, "Failed to remount %s with MS_SLAVE: %m", incoming_dir
);
2600 for (m
= mounts
; m
< mounts
+ n_mounts
; m
++)
2601 mount_entry_done(m
);
2608 void bind_mount_free_many(BindMount
*b
, size_t n
) {
2609 assert(b
|| n
== 0);
2611 for (size_t i
= 0; i
< n
; i
++) {
2613 free(b
[i
].destination
);
2619 int bind_mount_add(BindMount
**b
, size_t *n
, const BindMount
*item
) {
2620 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
2627 s
= strdup(item
->source
);
2631 d
= strdup(item
->destination
);
2635 c
= reallocarray(*b
, *n
+ 1, sizeof(BindMount
));
2641 c
[(*n
) ++] = (BindMount
) {
2642 .source
= TAKE_PTR(s
),
2643 .destination
= TAKE_PTR(d
),
2644 .read_only
= item
->read_only
,
2645 .nosuid
= item
->nosuid
,
2646 .recursive
= item
->recursive
,
2647 .ignore_enoent
= item
->ignore_enoent
,
2653 MountImage
* mount_image_free_many(MountImage
*m
, size_t *n
) {
2655 assert(m
|| *n
== 0);
2657 for (size_t i
= 0; i
< *n
; i
++) {
2659 free(m
[i
].destination
);
2660 mount_options_free_all(m
[i
].mount_options
);
2668 int mount_image_add(MountImage
**m
, size_t *n
, const MountImage
*item
) {
2669 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
2670 _cleanup_(mount_options_free_allp
) MountOptions
*options
= NULL
;
2677 s
= strdup(item
->source
);
2681 if (item
->destination
) {
2682 d
= strdup(item
->destination
);
2687 LIST_FOREACH(mount_options
, i
, item
->mount_options
) {
2688 _cleanup_(mount_options_free_allp
) MountOptions
*o
= NULL
;
2690 o
= new(MountOptions
, 1);
2694 *o
= (MountOptions
) {
2695 .partition_designator
= i
->partition_designator
,
2696 .options
= strdup(i
->options
),
2701 LIST_APPEND(mount_options
, options
, TAKE_PTR(o
));
2704 c
= reallocarray(*m
, *n
+ 1, sizeof(MountImage
));
2710 c
[(*n
) ++] = (MountImage
) {
2711 .source
= TAKE_PTR(s
),
2712 .destination
= TAKE_PTR(d
),
2713 .mount_options
= TAKE_PTR(options
),
2714 .ignore_enoent
= item
->ignore_enoent
,
2721 void temporary_filesystem_free_many(TemporaryFileSystem
*t
, size_t n
) {
2722 assert(t
|| n
== 0);
2724 for (size_t i
= 0; i
< n
; i
++) {
2732 int temporary_filesystem_add(
2733 TemporaryFileSystem
**t
,
2736 const char *options
) {
2738 _cleanup_free_
char *p
= NULL
, *o
= NULL
;
2739 TemporaryFileSystem
*c
;
2749 if (!isempty(options
)) {
2750 o
= strdup(options
);
2755 c
= reallocarray(*t
, *n
+ 1, sizeof(TemporaryFileSystem
));
2761 c
[(*n
) ++] = (TemporaryFileSystem
) {
2762 .path
= TAKE_PTR(p
),
2763 .options
= TAKE_PTR(o
),
2769 static int make_tmp_prefix(const char *prefix
) {
2770 _cleanup_free_
char *t
= NULL
;
2771 _cleanup_close_
int fd
= -EBADF
;
2774 /* Don't do anything unless we know the dir is actually missing */
2775 r
= access(prefix
, F_OK
);
2778 if (errno
!= ENOENT
)
2782 r
= mkdir_parents(prefix
, 0755);
2786 r
= tempfn_random(prefix
, NULL
, &t
);
2790 /* umask will corrupt this access mode, but that doesn't matter, we need to call chmod() anyway for
2791 * the suid bit, below. */
2792 fd
= open_mkdir_at(AT_FDCWD
, t
, O_EXCL
|O_CLOEXEC
, 0777);
2796 r
= RET_NERRNO(fchmod(fd
, 01777));
2802 r
= RET_NERRNO(rename(t
, prefix
));
2805 return r
== -EEXIST
? 0 : r
; /* it's fine if someone else created the dir by now */
2812 static int setup_one_tmp_dir(const char *id
, const char *prefix
, char **path
, char **tmp_path
) {
2813 _cleanup_free_
char *x
= NULL
;
2814 _cleanup_free_
char *y
= NULL
;
2823 /* We include the boot id in the directory so that after a
2824 * reboot we can easily identify obsolete directories. */
2826 r
= sd_id128_get_boot(&boot_id
);
2830 x
= strjoin(prefix
, "/systemd-private-", SD_ID128_TO_STRING(boot_id
), "-", id
, "-XXXXXX");
2834 r
= make_tmp_prefix(prefix
);
2840 if (errno
== EROFS
|| ERRNO_IS_DISK_SPACE(errno
))
2847 y
= strjoin(x
, "/tmp");
2852 if (mkdir(y
, 0777 | S_ISVTX
) < 0)
2855 r
= label_fix_full(AT_FDCWD
, y
, prefix
, 0);
2860 *tmp_path
= TAKE_PTR(y
);
2862 /* Trouble: we failed to create the directory. Instead of failing, let's simulate /tmp being
2863 * read-only. This way the service will get the EROFS result as if it was writing to the real
2866 r
= mkdir_p(RUN_SYSTEMD_EMPTY
, 0500);
2870 r
= free_and_strdup(&x
, RUN_SYSTEMD_EMPTY
);
2875 *path
= TAKE_PTR(x
);
2879 int setup_tmp_dirs(const char *id
, char **tmp_dir
, char **var_tmp_dir
) {
2880 _cleanup_(namespace_cleanup_tmpdirp
) char *a
= NULL
;
2881 _cleanup_(rmdir_and_freep
) char *a_tmp
= NULL
;
2887 assert(var_tmp_dir
);
2889 r
= setup_one_tmp_dir(id
, "/tmp", &a
, &a_tmp
);
2893 r
= setup_one_tmp_dir(id
, "/var/tmp", &b
, NULL
);
2897 a_tmp
= mfree(a_tmp
); /* avoid rmdir */
2898 *tmp_dir
= TAKE_PTR(a
);
2899 *var_tmp_dir
= TAKE_PTR(b
);
2904 int setup_shareable_ns(int ns_storage_socket
[static 2], unsigned long nsflag
) {
2905 _cleanup_close_
int ns
= -EBADF
;
2907 const char *ns_name
, *ns_path
;
2909 assert(ns_storage_socket
);
2910 assert(ns_storage_socket
[0] >= 0);
2911 assert(ns_storage_socket
[1] >= 0);
2913 ns_name
= namespace_single_flag_to_string(nsflag
);
2916 /* We use the passed socketpair as a storage buffer for our
2917 * namespace reference fd. Whatever process runs this first
2918 * shall create a new namespace, all others should just join
2919 * it. To serialize that we use a file lock on the socket
2922 * It's a bit crazy, but hey, works great! */
2924 r
= posix_lock(ns_storage_socket
[0], LOCK_EX
);
2928 CLEANUP_POSIX_UNLOCK(ns_storage_socket
[0]);
2930 ns
= receive_one_fd(ns_storage_socket
[0], MSG_PEEK
|MSG_DONTWAIT
);
2932 /* Yay, found something, so let's join the namespace */
2933 r
= RET_NERRNO(setns(ns
, nsflag
));
2943 /* Nothing stored yet, so let's create a new namespace. */
2945 if (unshare(nsflag
) < 0)
2948 (void) loopback_setup();
2950 ns_path
= strjoina("/proc/self/ns/", ns_name
);
2951 ns
= open(ns_path
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
2955 r
= send_one_fd(ns_storage_socket
[1], ns
, MSG_DONTWAIT
);
2962 int open_shareable_ns_path(int ns_storage_socket
[static 2], const char *path
, unsigned long nsflag
) {
2963 _cleanup_close_
int ns
= -EBADF
;
2966 assert(ns_storage_socket
);
2967 assert(ns_storage_socket
[0] >= 0);
2968 assert(ns_storage_socket
[1] >= 0);
2971 /* If the storage socket doesn't contain a ns fd yet, open one via the file system and store it in
2972 * it. This is supposed to be called ahead of time, i.e. before setup_shareable_ns() which will
2973 * allocate a new anonymous ns if needed. */
2975 r
= posix_lock(ns_storage_socket
[0], LOCK_EX
);
2979 CLEANUP_POSIX_UNLOCK(ns_storage_socket
[0]);
2981 ns
= receive_one_fd(ns_storage_socket
[0], MSG_PEEK
|MSG_DONTWAIT
);
2987 /* Nothing stored yet. Open the file from the file system. */
2989 ns
= open(path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
2993 r
= fd_is_ns(ns
, nsflag
);
2996 if (r
< 0 && r
!= -EUCLEAN
) /* EUCLEAN: we don't know */
2999 r
= send_one_fd(ns_storage_socket
[1], ns
, MSG_DONTWAIT
);
3006 bool ns_type_supported(NamespaceType type
) {
3007 const char *t
, *ns_proc
;
3009 t
= namespace_type_to_string(type
);
3010 if (!t
) /* Don't know how to translate this? Then it's not supported */
3013 ns_proc
= strjoina("/proc/self/ns/", t
);
3014 return access(ns_proc
, F_OK
) == 0;
3017 static const char *const protect_home_table
[_PROTECT_HOME_MAX
] = {
3018 [PROTECT_HOME_NO
] = "no",
3019 [PROTECT_HOME_YES
] = "yes",
3020 [PROTECT_HOME_READ_ONLY
] = "read-only",
3021 [PROTECT_HOME_TMPFS
] = "tmpfs",
3024 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home
, ProtectHome
, PROTECT_HOME_YES
);
3026 static const char *const protect_system_table
[_PROTECT_SYSTEM_MAX
] = {
3027 [PROTECT_SYSTEM_NO
] = "no",
3028 [PROTECT_SYSTEM_YES
] = "yes",
3029 [PROTECT_SYSTEM_FULL
] = "full",
3030 [PROTECT_SYSTEM_STRICT
] = "strict",
3033 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system
, ProtectSystem
, PROTECT_SYSTEM_YES
);
3035 static const char* const namespace_type_table
[] = {
3036 [NAMESPACE_MOUNT
] = "mnt",
3037 [NAMESPACE_CGROUP
] = "cgroup",
3038 [NAMESPACE_UTS
] = "uts",
3039 [NAMESPACE_IPC
] = "ipc",
3040 [NAMESPACE_USER
] = "user",
3041 [NAMESPACE_PID
] = "pid",
3042 [NAMESPACE_NET
] = "net",
3043 [NAMESPACE_TIME
] = "time",
3046 DEFINE_STRING_TABLE_LOOKUP(namespace_type
, NamespaceType
);
3048 static const char* const protect_proc_table
[_PROTECT_PROC_MAX
] = {
3049 [PROTECT_PROC_DEFAULT
] = "default",
3050 [PROTECT_PROC_NOACCESS
] = "noaccess",
3051 [PROTECT_PROC_INVISIBLE
] = "invisible",
3052 [PROTECT_PROC_PTRACEABLE
] = "ptraceable",
3055 DEFINE_STRING_TABLE_LOOKUP(protect_proc
, ProtectProc
);
3057 static const char* const proc_subset_table
[_PROC_SUBSET_MAX
] = {
3058 [PROC_SUBSET_ALL
] = "all",
3059 [PROC_SUBSET_PID
] = "pid",
3062 DEFINE_STRING_TABLE_LOOKUP(proc_subset
, ProcSubset
);