1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
4 #include <linux/loop.h>
14 #include "alloc-util.h"
15 #include "base-filesystem.h"
16 #include "chase-symlinks.h"
17 #include "dev-setup.h"
18 #include "devnum-util.h"
21 #include "extension-release.h"
23 #include "format-util.h"
24 #include "glyph-util.h"
27 #include "loop-util.h"
28 #include "loopback-setup.h"
29 #include "missing_syscall.h"
30 #include "mkdir-label.h"
31 #include "mount-util.h"
32 #include "mountpoint-util.h"
33 #include "namespace-util.h"
34 #include "namespace.h"
36 #include "nulstr-util.h"
38 #include "path-util.h"
39 #include "selinux-util.h"
40 #include "socket-util.h"
41 #include "sort-util.h"
42 #include "stat-util.h"
43 #include "string-table.h"
44 #include "string-util.h"
46 #include "tmpfile-util.h"
47 #include "umask-util.h"
48 #include "user-util.h"
50 #define DEV_MOUNT_OPTIONS (MS_NOSUID|MS_STRICTATIME|MS_NOEXEC)
52 typedef enum MountMode
{
53 /* This is ordered by priority! */
72 EXTENSION_DIRECTORIES
, /* Bind-mounted outside the root directory, and used by subsequent mounts */
73 EXTENSION_IMAGES
, /* Mounted outside the root directory, and used by subsequent mounts */
75 READWRITE_IMPLICIT
, /* Should have the lowest priority. */
79 typedef struct MountEntry
{
80 const char *path_const
; /* Memory allocated on stack or static */
82 bool ignore
:1; /* Ignore if path does not exist? */
83 bool has_prefix
:1; /* Already is prefixed by the root dir? */
84 bool read_only
:1; /* Shall this mount point be read-only? */
85 bool nosuid
:1; /* Shall set MS_NOSUID on the mount itself */
86 bool noexec
:1; /* Shall set MS_NOEXEC on the mount itself */
87 bool exec
:1; /* Shall clear MS_NOEXEC on the mount itself */
88 bool applied
:1; /* Already applied */
89 char *path_malloc
; /* Use this instead of 'path_const' if we had to allocate memory */
90 const char *unprefixed_path_const
; /* If the path was amended with a prefix, these will save the original */
91 char *unprefixed_path_malloc
;
92 const char *source_const
; /* The source path, for bind mounts or images */
94 const char *options_const
;/* Mount options for tmpfs */
96 unsigned long flags
; /* Mount flags used by EMPTY_DIR and TMPFS. Do not include MS_RDONLY here, but please use read_only. */
98 LIST_HEAD(MountOptions
, image_options
);
101 /* 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
102 * something there already. These mounts are hence overridden by any other explicitly configured mounts. */
103 static const MountEntry apivfs_table
[] = {
104 { "/proc", PROCFS
, false },
105 { "/dev", BIND_DEV
, false },
106 { "/sys", SYSFS
, false },
107 { "/run", RUN
, false, .options_const
= "mode=0755" TMPFS_LIMITS_RUN
, .flags
= MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
},
110 /* ProtectKernelTunables= option and the related filesystem APIs */
111 static const MountEntry protect_kernel_tunables_proc_table
[] = {
112 { "/proc/acpi", READONLY
, true },
113 { "/proc/apm", READONLY
, true }, /* Obsolete API, there's no point in permitting access to this, ever */
114 { "/proc/asound", READONLY
, true },
115 { "/proc/bus", READONLY
, true },
116 { "/proc/fs", READONLY
, true },
117 { "/proc/irq", READONLY
, true },
118 { "/proc/kallsyms", INACCESSIBLE
, true },
119 { "/proc/kcore", INACCESSIBLE
, true },
120 { "/proc/latency_stats", READONLY
, true },
121 { "/proc/mtrr", READONLY
, true },
122 { "/proc/scsi", READONLY
, true },
123 { "/proc/sys", READONLY
, true },
124 { "/proc/sysrq-trigger", READONLY
, true },
125 { "/proc/timer_stats", READONLY
, true },
128 static const MountEntry protect_kernel_tunables_sys_table
[] = {
129 { "/sys", READONLY
, false },
130 { "/sys/fs/bpf", READONLY
, true },
131 { "/sys/fs/cgroup", READWRITE_IMPLICIT
, false }, /* READONLY is set by ProtectControlGroups= option */
132 { "/sys/fs/selinux", READWRITE_IMPLICIT
, true },
133 { "/sys/kernel/debug", READONLY
, true },
134 { "/sys/kernel/tracing", READONLY
, true },
137 /* ProtectKernelModules= option */
138 static const MountEntry protect_kernel_modules_table
[] = {
140 { "/lib/modules", INACCESSIBLE
, true },
142 { "/usr/lib/modules", INACCESSIBLE
, true },
145 /* ProtectKernelLogs= option */
146 static const MountEntry protect_kernel_logs_proc_table
[] = {
147 { "/proc/kmsg", INACCESSIBLE
, true },
150 static const MountEntry protect_kernel_logs_dev_table
[] = {
151 { "/dev/kmsg", INACCESSIBLE
, true },
155 * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of
156 * system should be protected by ProtectSystem=
158 static const MountEntry protect_home_read_only_table
[] = {
159 { "/home", READONLY
, true },
160 { "/run/user", READONLY
, true },
161 { "/root", READONLY
, true },
164 /* ProtectHome=tmpfs table */
165 static const MountEntry protect_home_tmpfs_table
[] = {
166 { "/home", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
167 { "/run/user", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
168 { "/root", TMPFS
, true, .read_only
= true, .options_const
= "mode=0700" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
171 /* ProtectHome=yes table */
172 static const MountEntry protect_home_yes_table
[] = {
173 { "/home", INACCESSIBLE
, true },
174 { "/run/user", INACCESSIBLE
, true },
175 { "/root", INACCESSIBLE
, true },
178 /* ProtectSystem=yes table */
179 static const MountEntry protect_system_yes_table
[] = {
180 { "/usr", READONLY
, false },
181 { "/boot", READONLY
, true },
182 { "/efi", READONLY
, true },
184 { "/lib", READONLY
, true },
185 { "/lib64", READONLY
, true },
186 { "/bin", READONLY
, true },
188 { "/sbin", READONLY
, true },
193 /* ProtectSystem=full includes ProtectSystem=yes */
194 static const MountEntry protect_system_full_table
[] = {
195 { "/usr", READONLY
, false },
196 { "/boot", READONLY
, true },
197 { "/efi", READONLY
, true },
198 { "/etc", READONLY
, false },
200 { "/lib", READONLY
, true },
201 { "/lib64", READONLY
, true },
202 { "/bin", READONLY
, true },
204 { "/sbin", READONLY
, true },
210 * ProtectSystem=strict table. In this strict mode, we mount everything
211 * read-only, except for /proc, /dev, /sys which are the kernel API VFS,
212 * which are left writable, but PrivateDevices= + ProtectKernelTunables=
213 * protect those, and these options should be fully orthogonal.
214 * (And of course /home and friends are also left writable, as ProtectHome=
215 * shall manage those, orthogonally).
217 static const MountEntry protect_system_strict_table
[] = {
218 { "/", READONLY
, false },
219 { "/proc", READWRITE_IMPLICIT
, false }, /* ProtectKernelTunables= */
220 { "/sys", READWRITE_IMPLICIT
, false }, /* ProtectKernelTunables= */
221 { "/dev", READWRITE_IMPLICIT
, false }, /* PrivateDevices= */
222 { "/home", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
223 { "/run/user", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
224 { "/root", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
227 static const char * const mount_mode_table
[_MOUNT_MODE_MAX
] = {
228 [INACCESSIBLE
] = "inaccessible",
229 [OVERLAY_MOUNT
] = "overlay",
230 [BIND_MOUNT
] = "bind",
231 [BIND_MOUNT_RECURSIVE
] = "rbind",
232 [PRIVATE_TMP
] = "private-tmp",
233 [PRIVATE_DEV
] = "private-dev",
234 [BIND_DEV
] = "bind-dev",
235 [EMPTY_DIR
] = "empty",
238 [READONLY
] = "read-only",
239 [READWRITE
] = "read-write",
241 [MOUNT_IMAGES
] = "mount-images",
242 [READWRITE_IMPLICIT
] = "rw-implicit",
245 [MQUEUEFS
] = "mqueuefs",
248 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(mount_mode
, MountMode
);
250 static const char *mount_entry_path(const MountEntry
*p
) {
253 /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that,
254 * otherwise the stack/static ->path field is returned. */
256 return p
->path_malloc
?: p
->path_const
;
259 static const char *mount_entry_unprefixed_path(const MountEntry
*p
) {
262 /* Returns the unprefixed path (ie: before prefix_where_needed() ran), if any */
264 return p
->unprefixed_path_malloc
?: p
->unprefixed_path_const
?: mount_entry_path(p
);
267 static void mount_entry_consume_prefix(MountEntry
*p
, char *new_path
) {
269 assert(p
->path_malloc
|| p
->path_const
);
272 /* Saves current path in unprefixed_ variable, and takes over new_path */
274 free_and_replace(p
->unprefixed_path_malloc
, p
->path_malloc
);
275 /* If we didn't have a path on the heap, then it's a static one */
276 if (!p
->unprefixed_path_malloc
)
277 p
->unprefixed_path_const
= p
->path_const
;
278 p
->path_malloc
= new_path
;
279 p
->has_prefix
= true;
282 static bool mount_entry_read_only(const MountEntry
*p
) {
285 return p
->read_only
|| IN_SET(p
->mode
, READONLY
, INACCESSIBLE
, PRIVATE_TMP_READONLY
);
288 static bool mount_entry_noexec(const MountEntry
*p
) {
291 return p
->noexec
|| IN_SET(p
->mode
, NOEXEC
, INACCESSIBLE
, SYSFS
, PROCFS
);
294 static bool mount_entry_exec(const MountEntry
*p
) {
297 return p
->exec
|| p
->mode
== EXEC
;
300 static const char *mount_entry_source(const MountEntry
*p
) {
303 return p
->source_malloc
?: p
->source_const
;
306 static const char *mount_entry_options(const MountEntry
*p
) {
309 return p
->options_malloc
?: p
->options_const
;
312 static void mount_entry_done(MountEntry
*p
) {
315 p
->path_malloc
= mfree(p
->path_malloc
);
316 p
->unprefixed_path_malloc
= mfree(p
->unprefixed_path_malloc
);
317 p
->source_malloc
= mfree(p
->source_malloc
);
318 p
->options_malloc
= mfree(p
->options_malloc
);
319 p
->image_options
= mount_options_free_all(p
->image_options
);
322 static int append_access_mounts(MountEntry
**p
, char **strv
, MountMode mode
, bool forcibly_require_prefix
) {
325 /* Adds a list of user-supplied READWRITE/READWRITE_IMPLICIT/READONLY/INACCESSIBLE entries */
327 STRV_FOREACH(i
, strv
) {
328 bool ignore
= false, needs_prefix
= false;
331 /* Look for any prefixes */
332 if (startswith(e
, "-")) {
336 if (startswith(e
, "+")) {
341 if (!path_is_absolute(e
))
342 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
343 "Path is not absolute: %s", e
);
345 *((*p
)++) = (MountEntry
) {
349 .has_prefix
= !needs_prefix
&& !forcibly_require_prefix
,
356 static int append_empty_dir_mounts(MountEntry
**p
, char **strv
) {
359 /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the
360 * "/private/" boundary directories for DynamicUser=1. */
362 STRV_FOREACH(i
, strv
) {
364 *((*p
)++) = (MountEntry
) {
369 .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
,
370 .flags
= MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
,
377 static int append_bind_mounts(MountEntry
**p
, const BindMount
*binds
, size_t n
) {
380 for (size_t i
= 0; i
< n
; i
++) {
381 const BindMount
*b
= binds
+ i
;
383 *((*p
)++) = (MountEntry
) {
384 .path_const
= b
->destination
,
385 .mode
= b
->recursive
? BIND_MOUNT_RECURSIVE
: BIND_MOUNT
,
386 .read_only
= b
->read_only
,
388 .source_const
= b
->source
,
389 .ignore
= b
->ignore_enoent
,
396 static int append_mount_images(MountEntry
**p
, const MountImage
*mount_images
, size_t n
) {
399 for (size_t i
= 0; i
< n
; i
++) {
400 const MountImage
*m
= mount_images
+ i
;
402 *((*p
)++) = (MountEntry
) {
403 .path_const
= m
->destination
,
404 .mode
= MOUNT_IMAGES
,
405 .source_const
= m
->source
,
406 .image_options
= m
->mount_options
,
407 .ignore
= m
->ignore_enoent
,
414 static int append_extensions(
417 const char *extension_dir
,
419 const MountImage
*mount_images
,
421 char **extension_directories
) {
423 _cleanup_strv_free_
char **overlays
= NULL
;
426 if (n
== 0 && strv_isempty(extension_directories
))
430 assert(extension_dir
);
432 /* Prepare a list of overlays, that will have as each element a string suitable for being
433 * passed as a lowerdir= parameter, so start with the hierarchy on the root.
434 * The overlays vector will have the same number of elements and will correspond to the
435 * hierarchies vector, so they can be iterated upon together. */
436 STRV_FOREACH(hierarchy
, hierarchies
) {
437 _cleanup_free_
char *prefixed_hierarchy
= NULL
;
439 prefixed_hierarchy
= path_join(root
, *hierarchy
);
440 if (!prefixed_hierarchy
)
443 r
= strv_consume(&overlays
, TAKE_PTR(prefixed_hierarchy
));
448 /* First, prepare a mount for each image, but these won't be visible to the unit, instead
449 * they will be mounted in our propagate directory, and used as a source for the overlay. */
450 for (size_t i
= 0; i
< n
; i
++) {
451 _cleanup_free_
char *mount_point
= NULL
;
452 const MountImage
*m
= mount_images
+ i
;
454 r
= asprintf(&mount_point
, "%s/%zu", extension_dir
, i
);
458 for (size_t j
= 0; hierarchies
&& hierarchies
[j
]; ++j
) {
459 _cleanup_free_
char *prefixed_hierarchy
= NULL
, *escaped
= NULL
, *lowerdir
= NULL
;
461 prefixed_hierarchy
= path_join(mount_point
, hierarchies
[j
]);
462 if (!prefixed_hierarchy
)
465 escaped
= shell_escape(prefixed_hierarchy
, ",:");
469 /* Note that lowerdir= parameters are in 'reverse' order, so the
470 * top-most directory in the overlay comes first in the list. */
471 lowerdir
= strjoin(escaped
, ":", overlays
[j
]);
475 free_and_replace(overlays
[j
], lowerdir
);
478 *((*p
)++) = (MountEntry
) {
479 .path_malloc
= TAKE_PTR(mount_point
),
480 .image_options
= m
->mount_options
,
481 .ignore
= m
->ignore_enoent
,
482 .source_const
= m
->source
,
483 .mode
= EXTENSION_IMAGES
,
488 /* Secondly, extend the lowerdir= parameters with each ExtensionDirectory.
489 * Bind mount them in the same location as the ExtensionImages, so that we
490 * can check that they are valid trees (extension-release.d). */
491 STRV_FOREACH(extension_directory
, extension_directories
) {
492 _cleanup_free_
char *mount_point
= NULL
, *source
= NULL
;
493 const char *e
= *extension_directory
;
494 bool ignore_enoent
= false;
496 /* Pick up the counter where the ExtensionImages left it. */
497 r
= asprintf(&mount_point
, "%s/%zu", extension_dir
, n
++);
501 /* Look for any prefixes */
502 if (startswith(e
, "-")) {
504 ignore_enoent
= true;
506 /* Ignore this for now */
507 if (startswith(e
, "+"))
514 for (size_t j
= 0; hierarchies
&& hierarchies
[j
]; ++j
) {
515 _cleanup_free_
char *prefixed_hierarchy
= NULL
, *escaped
= NULL
, *lowerdir
= NULL
;
517 prefixed_hierarchy
= path_join(mount_point
, hierarchies
[j
]);
518 if (!prefixed_hierarchy
)
521 escaped
= shell_escape(prefixed_hierarchy
, ",:");
525 /* Note that lowerdir= parameters are in 'reverse' order, so the
526 * top-most directory in the overlay comes first in the list. */
527 lowerdir
= strjoin(escaped
, ":", overlays
[j
]);
531 free_and_replace(overlays
[j
], lowerdir
);
534 *((*p
)++) = (MountEntry
) {
535 .path_malloc
= TAKE_PTR(mount_point
),
536 .source_const
= TAKE_PTR(source
),
537 .mode
= EXTENSION_DIRECTORIES
,
538 .ignore
= ignore_enoent
,
544 /* Then, for each hierarchy, prepare an overlay with the list of lowerdir= strings
546 for (size_t i
= 0; hierarchies
&& hierarchies
[i
]; ++i
) {
547 _cleanup_free_
char *prefixed_hierarchy
= NULL
;
549 prefixed_hierarchy
= path_join(root
, hierarchies
[i
]);
550 if (!prefixed_hierarchy
)
553 *((*p
)++) = (MountEntry
) {
554 .path_malloc
= TAKE_PTR(prefixed_hierarchy
),
555 .options_malloc
= TAKE_PTR(overlays
[i
]),
556 .mode
= OVERLAY_MOUNT
,
558 .ignore
= true, /* If the source image doesn't set the ignore bit it will fail earlier. */
565 static int append_tmpfs_mounts(MountEntry
**p
, const TemporaryFileSystem
*tmpfs
, size_t n
) {
568 for (size_t i
= 0; i
< n
; i
++) {
569 const TemporaryFileSystem
*t
= tmpfs
+ i
;
570 _cleanup_free_
char *o
= NULL
, *str
= NULL
;
575 if (!path_is_absolute(t
->path
))
576 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
577 "Path is not absolute: %s",
580 str
= strjoin("mode=0755" NESTED_TMPFS_LIMITS
",", t
->options
);
584 r
= mount_option_mangle(str
, MS_NODEV
|MS_STRICTATIME
, &flags
, &o
);
586 return log_debug_errno(r
, "Failed to parse mount option '%s': %m", str
);
588 ro
= flags
& MS_RDONLY
;
592 *((*p
)++) = (MountEntry
) {
593 .path_const
= t
->path
,
596 .options_malloc
= TAKE_PTR(o
),
604 static int append_static_mounts(MountEntry
**p
, const MountEntry
*mounts
, size_t n
, bool ignore_protect
) {
608 /* Adds a list of static pre-defined entries */
610 for (size_t i
= 0; i
< n
; i
++)
611 *((*p
)++) = (MountEntry
) {
612 .path_const
= mount_entry_path(mounts
+i
),
613 .mode
= mounts
[i
].mode
,
614 .ignore
= mounts
[i
].ignore
|| ignore_protect
,
620 static int append_protect_home(MountEntry
**p
, ProtectHome protect_home
, bool ignore_protect
) {
623 switch (protect_home
) {
625 case PROTECT_HOME_NO
:
628 case PROTECT_HOME_READ_ONLY
:
629 return append_static_mounts(p
, protect_home_read_only_table
, ELEMENTSOF(protect_home_read_only_table
), ignore_protect
);
631 case PROTECT_HOME_TMPFS
:
632 return append_static_mounts(p
, protect_home_tmpfs_table
, ELEMENTSOF(protect_home_tmpfs_table
), ignore_protect
);
634 case PROTECT_HOME_YES
:
635 return append_static_mounts(p
, protect_home_yes_table
, ELEMENTSOF(protect_home_yes_table
), ignore_protect
);
638 assert_not_reached();
642 static int append_protect_system(MountEntry
**p
, ProtectSystem protect_system
, bool ignore_protect
) {
645 switch (protect_system
) {
647 case PROTECT_SYSTEM_NO
:
650 case PROTECT_SYSTEM_STRICT
:
651 return append_static_mounts(p
, protect_system_strict_table
, ELEMENTSOF(protect_system_strict_table
), ignore_protect
);
653 case PROTECT_SYSTEM_YES
:
654 return append_static_mounts(p
, protect_system_yes_table
, ELEMENTSOF(protect_system_yes_table
), ignore_protect
);
656 case PROTECT_SYSTEM_FULL
:
657 return append_static_mounts(p
, protect_system_full_table
, ELEMENTSOF(protect_system_full_table
), ignore_protect
);
660 assert_not_reached();
664 static int mount_path_compare(const MountEntry
*a
, const MountEntry
*b
) {
667 /* ExtensionImages/Directories will be used by other mounts as a base, so sort them first
668 * regardless of the prefix - they are set up in the propagate directory anyway */
669 d
= -CMP(a
->mode
== EXTENSION_IMAGES
, b
->mode
== EXTENSION_IMAGES
);
672 d
= -CMP(a
->mode
== EXTENSION_DIRECTORIES
, b
->mode
== EXTENSION_DIRECTORIES
);
676 /* If the paths are not equal, then order prefixes first */
677 d
= path_compare(mount_entry_path(a
), mount_entry_path(b
));
681 /* If the paths are equal, check the mode */
682 return CMP((int) a
->mode
, (int) b
->mode
);
685 static int prefix_where_needed(MountEntry
*m
, size_t n
, const char *root_directory
) {
686 /* Prefixes all paths in the bind mount table with the root directory if the entry needs that. */
690 for (size_t i
= 0; i
< n
; i
++) {
696 s
= path_join(root_directory
, mount_entry_path(m
+i
));
700 mount_entry_consume_prefix(&m
[i
], s
);
706 static void drop_duplicates(MountEntry
*m
, size_t *n
) {
707 MountEntry
*f
, *t
, *previous
;
712 /* Drops duplicate entries. Expects that the array is properly ordered already. */
714 for (f
= m
, t
= m
, previous
= NULL
; f
< m
+ *n
; f
++) {
716 /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare()
717 * above. Note that we only drop duplicates that haven't been mounted yet. */
719 path_equal(mount_entry_path(f
), mount_entry_path(previous
)) &&
720 !f
->applied
&& !previous
->applied
) {
721 log_debug("%s (%s) is duplicate.", mount_entry_path(f
), mount_mode_to_string(f
->mode
));
722 /* Propagate the flags to the remaining entry */
723 previous
->read_only
= previous
->read_only
|| mount_entry_read_only(f
);
724 previous
->noexec
= previous
->noexec
|| mount_entry_noexec(f
);
725 previous
->exec
= previous
->exec
|| mount_entry_exec(f
);
738 static void drop_inaccessible(MountEntry
*m
, size_t *n
) {
740 const char *clear
= NULL
;
745 /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly
746 * ordered already. */
748 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
750 /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop
751 * it, as inaccessible paths really should drop the entire subtree. */
752 if (clear
&& path_startswith(mount_entry_path(f
), clear
)) {
753 log_debug("%s is masked by %s.", mount_entry_path(f
), clear
);
758 clear
= f
->mode
== INACCESSIBLE
? mount_entry_path(f
) : NULL
;
767 static void drop_nop(MountEntry
*m
, size_t *n
) {
773 /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the
774 * list is ordered by prefixes. */
776 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
778 /* Only suppress such subtrees for READONLY, READWRITE and READWRITE_IMPLICIT entries */
779 if (IN_SET(f
->mode
, READONLY
, READWRITE
, READWRITE_IMPLICIT
)) {
780 MountEntry
*found
= NULL
;
782 /* Now let's find the first parent of the entry we are looking at. */
783 for (MountEntry
*p
= PTR_SUB1(t
, m
); p
; p
= PTR_SUB1(p
, m
))
784 if (path_startswith(mount_entry_path(f
), mount_entry_path(p
))) {
789 /* We found it, let's see if it's the same mode, if so, we can drop this entry */
790 if (found
&& found
->mode
== f
->mode
) {
791 log_debug("%s (%s) is made redundant by %s (%s)",
792 mount_entry_path(f
), mount_mode_to_string(f
->mode
),
793 mount_entry_path(found
), mount_mode_to_string(found
->mode
));
806 static void drop_outside_root(const char *root_directory
, MountEntry
*m
, size_t *n
) {
816 /* Drops all mounts that are outside of the root directory. */
818 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
820 /* ExtensionImages/Directories bases are opened in /run/systemd/unit-extensions on the host */
821 if (!IN_SET(f
->mode
, EXTENSION_IMAGES
, EXTENSION_DIRECTORIES
) && !path_startswith(mount_entry_path(f
), root_directory
)) {
822 log_debug("%s is outside of root directory.", mount_entry_path(f
));
834 static int clone_device_node(
836 const char *temporary_mount
,
837 bool *make_devnode
) {
839 _cleanup_free_
char *sl
= NULL
;
840 const char *dn
, *bn
, *t
;
844 if (stat(d
, &st
) < 0) {
845 if (errno
== ENOENT
) {
846 log_debug_errno(errno
, "Device node '%s' to clone does not exist, ignoring.", d
);
850 return log_debug_errno(errno
, "Failed to stat() device node '%s' to clone, ignoring: %m", d
);
853 if (!S_ISBLK(st
.st_mode
) &&
854 !S_ISCHR(st
.st_mode
))
855 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
856 "Device node '%s' to clone is not a device node, ignoring.",
859 dn
= strjoina(temporary_mount
, d
);
861 /* First, try to create device node properly */
863 mac_selinux_create_file_prepare(d
, st
.st_mode
);
864 r
= mknod(dn
, st
.st_mode
, st
.st_rdev
);
865 mac_selinux_create_file_clear();
869 return log_debug_errno(errno
, "mknod failed for %s: %m", d
);
871 /* This didn't work, let's not try this again for the next iterations. */
872 *make_devnode
= false;
875 /* We're about to fall back to bind-mounting the device node. So create a dummy bind-mount target.
876 * Do not prepare device-node SELinux label (see issue 13762) */
877 r
= mknod(dn
, S_IFREG
, 0);
878 if (r
< 0 && errno
!= EEXIST
)
879 return log_debug_errno(errno
, "mknod() fallback failed for '%s': %m", d
);
881 /* Fallback to bind-mounting: The assumption here is that all used device nodes carry standard
882 * properties. Specifically, the devices nodes we bind-mount should either be owned by root:root or
883 * root:tty (e.g. /dev/tty, /dev/ptmx) and should not carry ACLs. */
884 r
= mount_nofollow_verbose(LOG_DEBUG
, d
, dn
, NULL
, MS_BIND
, NULL
);
889 bn
= path_startswith(d
, "/dev/");
893 /* Create symlinks like /dev/char/1:9 → ../urandom */
894 if (asprintf(&sl
, "%s/dev/%s/" DEVNUM_FORMAT_STR
,
896 S_ISCHR(st
.st_mode
) ? "char" : "block",
897 DEVNUM_FORMAT_VAL(st
.st_rdev
)) < 0)
900 (void) mkdir_parents(sl
, 0755);
902 t
= strjoina("../", bn
);
903 if (symlink(t
, sl
) < 0)
904 log_debug_errno(errno
, "Failed to symlink '%s' to '%s', ignoring: %m", t
, sl
);
909 static int mount_private_dev(MountEntry
*m
) {
910 static const char devnodes
[] =
918 char temporary_mount
[] = "/tmp/namespace-dev-XXXXXX";
919 const char *dev
= NULL
, *devpts
= NULL
, *devshm
= NULL
, *devhugepages
= NULL
, *devmqueue
= NULL
, *devlog
= NULL
, *devptmx
= NULL
;
920 bool can_mknod
= true;
925 if (!mkdtemp(temporary_mount
))
926 return log_debug_errno(errno
, "Failed to create temporary directory '%s': %m", temporary_mount
);
928 dev
= strjoina(temporary_mount
, "/dev");
929 (void) mkdir(dev
, 0755);
930 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", dev
, "tmpfs", DEV_MOUNT_OPTIONS
, "mode=0755" TMPFS_LIMITS_PRIVATE_DEV
);
934 r
= label_fix_full(AT_FDCWD
, dev
, "/dev", 0);
936 log_debug_errno(r
, "Failed to fix label of '%s' as /dev: %m", dev
);
940 devpts
= strjoina(temporary_mount
, "/dev/pts");
941 (void) mkdir(devpts
, 0755);
942 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev/pts", devpts
, NULL
, MS_BIND
, NULL
);
946 /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx.
947 * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible.
948 * Thus, in that case make a clone.
949 * In nspawn and other containers it will be a symlink, in that case make it a symlink. */
950 r
= is_symlink("/dev/ptmx");
952 log_debug_errno(r
, "Failed to detect whether /dev/ptmx is a symlink or not: %m");
955 devptmx
= strjoina(temporary_mount
, "/dev/ptmx");
956 if (symlink("pts/ptmx", devptmx
) < 0) {
957 r
= log_debug_errno(errno
, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx
);
961 r
= clone_device_node("/dev/ptmx", temporary_mount
, &can_mknod
);
966 devshm
= strjoina(temporary_mount
, "/dev/shm");
967 (void) mkdir(devshm
, 0755);
968 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev/shm", devshm
, NULL
, MS_BIND
, NULL
);
972 devmqueue
= strjoina(temporary_mount
, "/dev/mqueue");
973 (void) mkdir(devmqueue
, 0755);
974 (void) mount_nofollow_verbose(LOG_DEBUG
, "/dev/mqueue", devmqueue
, NULL
, MS_BIND
, NULL
);
976 devhugepages
= strjoina(temporary_mount
, "/dev/hugepages");
977 (void) mkdir(devhugepages
, 0755);
978 (void) mount_nofollow_verbose(LOG_DEBUG
, "/dev/hugepages", devhugepages
, NULL
, MS_BIND
, NULL
);
980 devlog
= strjoina(temporary_mount
, "/dev/log");
981 if (symlink("/run/systemd/journal/dev-log", devlog
) < 0)
982 log_debug_errno(errno
, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog
);
984 NULSTR_FOREACH(d
, devnodes
) {
985 r
= clone_device_node(d
, temporary_mount
, &can_mknod
);
986 /* ENXIO means the *source* is not a device file, skip creation in that case */
987 if (r
< 0 && r
!= -ENXIO
)
991 r
= dev_setup(temporary_mount
, UID_INVALID
, GID_INVALID
);
993 log_debug_errno(r
, "Failed to set up basic device tree at '%s', ignoring: %m", temporary_mount
);
995 /* Create the /dev directory if missing. It is more likely to be missing when the service is started
996 * with RootDirectory. This is consistent with mount units creating the mount points when missing. */
997 (void) mkdir_p_label(mount_entry_path(m
), 0755);
999 /* Unmount everything in old /dev */
1000 r
= umount_recursive(mount_entry_path(m
), 0);
1002 log_debug_errno(r
, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m
));
1004 r
= mount_nofollow_verbose(LOG_DEBUG
, dev
, mount_entry_path(m
), NULL
, MS_MOVE
, NULL
);
1009 (void) rmdir(temporary_mount
);
1015 (void) umount_verbose(LOG_DEBUG
, devpts
, UMOUNT_NOFOLLOW
);
1018 (void) umount_verbose(LOG_DEBUG
, devshm
, UMOUNT_NOFOLLOW
);
1021 (void) umount_verbose(LOG_DEBUG
, devhugepages
, UMOUNT_NOFOLLOW
);
1024 (void) umount_verbose(LOG_DEBUG
, devmqueue
, UMOUNT_NOFOLLOW
);
1026 (void) umount_verbose(LOG_DEBUG
, dev
, UMOUNT_NOFOLLOW
);
1028 (void) rmdir(temporary_mount
);
1033 static int mount_bind_dev(const MountEntry
*m
) {
1038 /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the
1039 * service's /dev. This is only used when RootDirectory= is set. */
1041 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1043 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
1045 return log_debug_errno(r
, "Unable to determine whether /dev is already mounted: %m");
1046 if (r
> 0) /* make this a NOP if /dev is already a mount point */
1049 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
);
1056 static int mount_sysfs(const MountEntry
*m
) {
1061 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1063 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
1065 return log_debug_errno(r
, "Unable to determine whether /sys is already mounted: %m");
1066 if (r
> 0) /* make this a NOP if /sys is already a mount point */
1069 /* Bind mount the host's version so that we get all child mounts of it, too. */
1070 r
= mount_nofollow_verbose(LOG_DEBUG
, "/sys", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
);
1077 static bool mount_option_supported(const char *fstype
, const char *key
, const char *value
) {
1078 _cleanup_close_
int fd
= -EBADF
;
1081 /* This function assumes support by default. Only if the fsconfig() call fails with -EINVAL/-EOPNOTSUPP
1082 * will it report that the option/value is not supported. */
1084 fd
= fsopen(fstype
, FSOPEN_CLOEXEC
);
1086 if (errno
!= ENOSYS
)
1087 log_debug_errno(errno
, "Failed to open superblock context for '%s': %m", fstype
);
1088 return true; /* If fsopen() fails for whatever reason, assume the value is supported. */
1091 r
= fsconfig(fd
, FSCONFIG_SET_STRING
, key
, value
, 0);
1092 if (r
< 0 && !IN_SET(errno
, EINVAL
, EOPNOTSUPP
, ENOSYS
))
1093 log_debug_errno(errno
, "Failed to set '%s=%s' on '%s' superblock context: %m", key
, value
, fstype
);
1095 return r
>= 0 || !IN_SET(errno
, EINVAL
, EOPNOTSUPP
);
1098 static int mount_procfs(const MountEntry
*m
, const NamespaceInfo
*ns_info
) {
1099 _cleanup_free_
char *opts
= NULL
;
1100 const char *entry_path
;
1106 if (ns_info
->protect_proc
!= PROTECT_PROC_DEFAULT
||
1107 ns_info
->proc_subset
!= PROC_SUBSET_ALL
) {
1109 /* Starting with kernel 5.8 procfs' hidepid= logic is truly per-instance (previously it
1110 * pretended to be per-instance but actually was per-namespace), hence let's make use of it
1111 * if requested. To make sure this logic succeeds only on kernels where hidepid= is
1112 * per-instance, we'll exclusively use the textual value for hidepid=, since support was
1113 * added in the same commit: if it's supported it is thus also per-instance. */
1115 const char *hpv
= ns_info
->protect_proc
== PROTECT_PROC_DEFAULT
?
1117 protect_proc_to_string(ns_info
->protect_proc
);
1119 /* hidepid= support was added in 5.8, so we can use fsconfig()/fsopen() (which were added in
1120 * 5.2) to check if hidepid= is supported. This avoids a noisy dmesg log by the kernel when
1121 * trying to use hidepid= on systems where it isn't supported. The same applies for subset=.
1122 * fsopen()/fsconfig() was also backported on some distros which allows us to detect
1123 * hidepid=/subset= support in even more scenarios. */
1125 if (mount_option_supported("proc", "hidepid", hpv
)) {
1126 opts
= strjoin("hidepid=", hpv
);
1131 if (ns_info
->proc_subset
== PROC_SUBSET_PID
&& mount_option_supported("proc", "subset", "pid"))
1132 if (!strextend_with_separator(&opts
, ",", "subset=pid"))
1136 entry_path
= mount_entry_path(m
);
1137 (void) mkdir_p_label(entry_path
, 0755);
1139 /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in
1140 * one. i.e we don't reuse existing mounts here under any condition, we want a new instance owned by
1141 * our user namespace and with our hidepid= settings applied. Hence, let's get rid of everything
1142 * mounted on /proc/ first. */
1144 n
= umount_recursive(entry_path
, 0);
1146 r
= mount_nofollow_verbose(LOG_DEBUG
, "proc", entry_path
, "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, opts
);
1147 if (r
== -EINVAL
&& opts
)
1148 /* If this failed with EINVAL then this likely means the textual hidepid= stuff is
1149 * not supported by the kernel, and thus the per-instance hidepid= neither, which
1150 * means we really don't want to use it, since it would affect our host's /proc
1151 * mount. Hence let's gracefully fallback to a classic, unrestricted version. */
1152 r
= mount_nofollow_verbose(LOG_DEBUG
, "proc", entry_path
, "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
1154 /* When we do not have enough privileges to mount /proc, fallback to use existing /proc. */
1157 /* /proc or some of sub-mounts are umounted in the above. Refuse incomplete tree.
1158 * Propagate the original error code returned by mount() in the above. */
1161 r
= path_is_mount_point(entry_path
, NULL
, 0);
1163 return log_debug_errno(r
, "Unable to determine whether /proc is already mounted: %m");
1165 /* We lack permissions to mount a new instance of /proc, and it is not already
1166 * mounted. But we can access the host's, so as a final fallback bind-mount it to
1167 * the destination, as most likely we are inside a user manager in an unprivileged
1168 * user namespace. */
1169 r
= mount_nofollow_verbose(LOG_DEBUG
, "/proc", entry_path
, NULL
, MS_BIND
|MS_REC
, NULL
);
1179 static int mount_tmpfs(const MountEntry
*m
) {
1180 const char *entry_path
, *inner_path
;
1185 entry_path
= mount_entry_path(m
);
1186 inner_path
= mount_entry_unprefixed_path(m
);
1188 /* First, get rid of everything that is below if there is anything. Then, overmount with our new
1191 (void) mkdir_p_label(entry_path
, 0755);
1192 (void) umount_recursive(entry_path
, 0);
1194 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", entry_path
, "tmpfs", m
->flags
, mount_entry_options(m
));
1198 r
= label_fix_full(AT_FDCWD
, entry_path
, inner_path
, 0);
1200 return log_debug_errno(r
, "Failed to fix label of '%s' as '%s': %m", entry_path
, inner_path
);
1205 static int mount_run(const MountEntry
*m
) {
1210 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
1211 if (r
< 0 && r
!= -ENOENT
)
1212 return log_debug_errno(r
, "Unable to determine whether /run is already mounted: %m");
1213 if (r
> 0) /* make this a NOP if /run is already a mount point */
1216 return mount_tmpfs(m
);
1219 static int mount_mqueuefs(const MountEntry
*m
) {
1221 const char *entry_path
;
1225 entry_path
= mount_entry_path(m
);
1227 (void) mkdir_p_label(entry_path
, 0755);
1228 (void) umount_recursive(entry_path
, 0);
1230 r
= mount_nofollow_verbose(LOG_DEBUG
, "mqueue", entry_path
, "mqueue", m
->flags
, mount_entry_options(m
));
1237 static int mount_image(const MountEntry
*m
, const char *root_directory
) {
1239 _cleanup_free_
char *host_os_release_id
= NULL
, *host_os_release_version_id
= NULL
,
1240 *host_os_release_sysext_level
= NULL
;
1245 if (m
->mode
== EXTENSION_IMAGES
) {
1246 r
= parse_os_release(
1247 empty_to_root(root_directory
),
1248 "ID", &host_os_release_id
,
1249 "VERSION_ID", &host_os_release_version_id
,
1250 "SYSEXT_LEVEL", &host_os_release_sysext_level
,
1253 return log_debug_errno(r
, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory
));
1254 if (isempty(host_os_release_id
))
1255 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
));
1258 r
= verity_dissect_and_mount(
1259 /* src_fd= */ -1, mount_entry_source(m
), mount_entry_path(m
), m
->image_options
,
1260 host_os_release_id
, host_os_release_version_id
, host_os_release_sysext_level
, NULL
);
1261 if (r
== -ENOENT
&& m
->ignore
)
1263 if (r
== -ESTALE
&& host_os_release_id
)
1264 return log_error_errno(r
,
1265 "Failed to mount image %s, extension-release metadata does not match the lower layer's: ID=%s%s%s%s%s",
1266 mount_entry_source(m
),
1268 host_os_release_version_id
? " VERSION_ID=" : "",
1269 strempty(host_os_release_version_id
),
1270 host_os_release_sysext_level
? " SYSEXT_LEVEL=" : "",
1271 strempty(host_os_release_sysext_level
));
1273 return log_debug_errno(r
, "Failed to mount image %s on %s: %m", mount_entry_source(m
), mount_entry_path(m
));
1278 static int mount_overlay(const MountEntry
*m
) {
1279 const char *options
;
1284 options
= strjoina("lowerdir=", mount_entry_options(m
));
1286 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1288 r
= mount_nofollow_verbose(LOG_DEBUG
, "overlay", mount_entry_path(m
), "overlay", MS_RDONLY
, options
);
1289 if (r
== -ENOENT
&& m
->ignore
)
1297 static int follow_symlink(
1298 const char *root_directory
,
1301 _cleanup_free_
char *target
= NULL
;
1304 /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we
1305 * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at
1306 * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the
1307 * end and already have a fully normalized name. */
1309 r
= chase_symlinks(mount_entry_path(m
), root_directory
, CHASE_STEP
|CHASE_NONEXISTENT
, &target
, NULL
);
1311 return log_debug_errno(r
, "Failed to chase symlinks '%s': %m", mount_entry_path(m
));
1312 if (r
> 0) /* Reached the end, nothing more to resolve */
1315 if (m
->n_followed
>= CHASE_SYMLINKS_MAX
) /* put a boundary on things */
1316 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
1317 "Symlink loop on '%s'.",
1318 mount_entry_path(m
));
1320 log_debug("Followed mount entry path symlink %s %s %s.",
1321 mount_entry_path(m
), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT
), target
);
1323 mount_entry_consume_prefix(m
, TAKE_PTR(target
));
1330 static int apply_one_mount(
1331 const char *root_directory
,
1333 const NamespaceInfo
*ns_info
) {
1335 _cleanup_free_
char *inaccessible
= NULL
;
1336 bool rbind
= true, make
= false;
1343 log_debug("Applying namespace mount on %s", mount_entry_path(m
));
1347 case INACCESSIBLE
: {
1348 _cleanup_free_
char *tmp
= NULL
;
1349 const char *runtime_dir
;
1352 /* First, get rid of everything that is below if there
1353 * is anything... Then, overmount it with an
1354 * inaccessible path. */
1355 (void) umount_recursive(mount_entry_path(m
), 0);
1357 if (lstat(mount_entry_path(m
), &target
) < 0) {
1358 if (errno
== ENOENT
&& m
->ignore
)
1361 return log_debug_errno(errno
, "Failed to lstat() %s to determine what to mount over it: %m",
1362 mount_entry_path(m
));
1366 runtime_dir
= "/run";
1368 if (asprintf(&tmp
, "/run/user/" UID_FMT
, geteuid()) < 0)
1374 r
= mode_to_inaccessible_node(runtime_dir
, target
.st_mode
, &inaccessible
);
1376 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
1377 "File type not supported for inaccessible mounts. Note that symlinks are not allowed");
1378 what
= inaccessible
;
1384 case READWRITE_IMPLICIT
:
1387 r
= path_is_mount_point(mount_entry_path(m
), root_directory
, 0);
1388 if (r
== -ENOENT
&& m
->ignore
)
1391 return log_debug_errno(r
, "Failed to determine whether %s is already a mount point: %m",
1392 mount_entry_path(m
));
1393 if (r
> 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY
1394 * and MS_NOEXEC bits for the mount point if needed. */
1396 /* This isn't a mount point yet, let's make it one. */
1397 what
= mount_entry_path(m
);
1400 case EXTENSION_DIRECTORIES
: {
1401 _cleanup_free_
char *host_os_release_id
= NULL
, *host_os_release_version_id
= NULL
,
1402 *host_os_release_sysext_level
= NULL
, *extension_name
= NULL
;
1403 _cleanup_strv_free_
char **extension_release
= NULL
;
1405 r
= path_extract_filename(mount_entry_source(m
), &extension_name
);
1407 return log_debug_errno(r
, "Failed to extract extension name from %s: %m", mount_entry_source(m
));
1409 r
= parse_os_release(
1410 empty_to_root(root_directory
),
1411 "ID", &host_os_release_id
,
1412 "VERSION_ID", &host_os_release_version_id
,
1413 "SYSEXT_LEVEL", &host_os_release_sysext_level
,
1416 return log_debug_errno(r
, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory
));
1417 if (isempty(host_os_release_id
))
1418 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
));
1420 r
= load_extension_release_pairs(mount_entry_source(m
), extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1421 if (r
== -ENOENT
&& m
->ignore
)
1424 return log_debug_errno(r
, "Failed to parse directory %s extension-release metadata: %m", extension_name
);
1426 r
= extension_release_validate(
1429 host_os_release_version_id
,
1430 host_os_release_sysext_level
,
1431 /* host_sysext_scope */ NULL
, /* Leave empty, we need to accept both system and portable */
1434 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE
), "Directory %s extension-release metadata does not match the root's", extension_name
);
1436 return log_debug_errno(r
, "Failed to compare directory %s extension-release metadata with the root's os-release: %m", extension_name
);
1445 case BIND_MOUNT_RECURSIVE
: {
1446 _cleanup_free_
char *chased
= NULL
;
1448 /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note
1449 * that bind mount source paths are always relative to the host root, hence we pass NULL as
1450 * root directory to chase_symlinks() here. */
1452 r
= chase_symlinks(mount_entry_source(m
), NULL
, CHASE_TRAIL_SLASH
, &chased
, NULL
);
1453 if (r
== -ENOENT
&& m
->ignore
) {
1454 log_debug_errno(r
, "Path %s does not exist, ignoring.", mount_entry_source(m
));
1458 return log_debug_errno(r
, "Failed to follow symlinks on %s: %m", mount_entry_source(m
));
1460 log_debug("Followed source symlinks %s %s %s.",
1461 mount_entry_source(m
), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT
), chased
);
1463 free_and_replace(m
->source_malloc
, chased
);
1465 what
= mount_entry_source(m
);
1472 return mount_tmpfs(m
);
1475 case PRIVATE_TMP_READONLY
:
1476 what
= mount_entry_source(m
);
1481 return mount_private_dev(m
);
1484 return mount_bind_dev(m
);
1487 return mount_sysfs(m
);
1490 return mount_procfs(m
, ns_info
);
1493 return mount_run(m
);
1496 return mount_mqueuefs(m
);
1499 return mount_image(m
, NULL
);
1501 case EXTENSION_IMAGES
:
1502 return mount_image(m
, root_directory
);
1505 return mount_overlay(m
);
1508 assert_not_reached();
1513 r
= mount_nofollow_verbose(LOG_DEBUG
, what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
);
1515 bool try_again
= false;
1517 if (r
== -ENOENT
&& make
) {
1520 /* Hmm, either the source or the destination are missing. Let's see if we can create
1521 the destination, then try again. */
1523 (void) mkdir_parents(mount_entry_path(m
), 0755);
1525 q
= make_mount_point_inode_from_path(what
, mount_entry_path(m
), 0755);
1526 if (q
< 0 && q
!= -EEXIST
)
1527 log_error_errno(q
, "Failed to create destination mount point node '%s': %m",
1528 mount_entry_path(m
));
1534 r
= mount_nofollow_verbose(LOG_DEBUG
, what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
);
1536 return log_error_errno(r
, "Failed to mount %s to %s: %m", what
, mount_entry_path(m
));
1539 log_debug("Successfully mounted %s to %s", what
, mount_entry_path(m
));
1543 static int make_read_only(const MountEntry
*m
, char **deny_list
, FILE *proc_self_mountinfo
) {
1544 unsigned long new_flags
= 0, flags_mask
= 0;
1549 assert(proc_self_mountinfo
);
1551 if (mount_entry_read_only(m
) || m
->mode
== PRIVATE_DEV
) {
1552 new_flags
|= MS_RDONLY
;
1553 flags_mask
|= MS_RDONLY
;
1557 new_flags
|= MS_NOSUID
;
1558 flags_mask
|= MS_NOSUID
;
1561 if (flags_mask
== 0) /* No Change? */
1564 /* We generally apply these changes recursively, except for /dev, and the cases we know there's
1565 * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the
1566 * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace
1567 * and running Linux <= 4.17. */
1569 mount_entry_read_only(m
) &&
1570 !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1572 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, deny_list
, proc_self_mountinfo
);
1574 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, proc_self_mountinfo
);
1576 /* Note that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked
1577 * read-only already stays this way. This improves compatibility with container managers, where we
1578 * won't attempt to undo read-only mounts already applied. */
1580 if (r
== -ENOENT
&& m
->ignore
)
1583 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1584 submounts
? " and its submounts" : "");
1588 static int make_noexec(const MountEntry
*m
, char **deny_list
, FILE *proc_self_mountinfo
) {
1589 unsigned long new_flags
= 0, flags_mask
= 0;
1594 assert(proc_self_mountinfo
);
1596 if (mount_entry_noexec(m
)) {
1597 new_flags
|= MS_NOEXEC
;
1598 flags_mask
|= MS_NOEXEC
;
1599 } else if (mount_entry_exec(m
)) {
1600 new_flags
&= ~MS_NOEXEC
;
1601 flags_mask
|= MS_NOEXEC
;
1604 if (flags_mask
== 0) /* No Change? */
1607 submounts
= !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1610 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, deny_list
, proc_self_mountinfo
);
1612 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, proc_self_mountinfo
);
1614 if (r
== -ENOENT
&& m
->ignore
)
1617 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1618 submounts
? " and its submounts" : "");
1622 static int make_nosuid(const MountEntry
*m
, FILE *proc_self_mountinfo
) {
1627 assert(proc_self_mountinfo
);
1629 submounts
= !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1632 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), MS_NOSUID
, MS_NOSUID
, NULL
, proc_self_mountinfo
);
1634 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), MS_NOSUID
, MS_NOSUID
, proc_self_mountinfo
);
1635 if (r
== -ENOENT
&& m
->ignore
)
1638 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1639 submounts
? " and its submounts" : "");
1643 static bool namespace_info_mount_apivfs(const NamespaceInfo
*ns_info
) {
1647 * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
1648 * since to protect the API VFS mounts, they need to be around in the
1652 return ns_info
->mount_apivfs
||
1653 ns_info
->protect_control_groups
||
1654 ns_info
->protect_kernel_tunables
||
1655 ns_info
->protect_proc
!= PROTECT_PROC_DEFAULT
||
1656 ns_info
->proc_subset
!= PROC_SUBSET_ALL
;
1659 static size_t namespace_calculate_mounts(
1660 const NamespaceInfo
*ns_info
,
1661 char** read_write_paths
,
1662 char** read_only_paths
,
1663 char** inaccessible_paths
,
1665 char** no_exec_paths
,
1666 char** empty_directories
,
1667 size_t n_bind_mounts
,
1668 size_t n_temporary_filesystems
,
1669 size_t n_mount_images
,
1670 size_t n_extension_images
,
1671 size_t n_extension_directories
,
1672 size_t n_hierarchies
,
1673 const char* tmp_dir
,
1674 const char* var_tmp_dir
,
1675 const char *creds_path
,
1676 const char* log_namespace
,
1677 bool setup_propagate
,
1678 const char* notify_socket
) {
1680 size_t protect_home_cnt
;
1681 size_t protect_system_cnt
=
1682 (ns_info
->protect_system
== PROTECT_SYSTEM_STRICT
?
1683 ELEMENTSOF(protect_system_strict_table
) :
1684 ((ns_info
->protect_system
== PROTECT_SYSTEM_FULL
) ?
1685 ELEMENTSOF(protect_system_full_table
) :
1686 ((ns_info
->protect_system
== PROTECT_SYSTEM_YES
) ?
1687 ELEMENTSOF(protect_system_yes_table
) : 0)));
1690 (ns_info
->protect_home
== PROTECT_HOME_YES
?
1691 ELEMENTSOF(protect_home_yes_table
) :
1692 ((ns_info
->protect_home
== PROTECT_HOME_READ_ONLY
) ?
1693 ELEMENTSOF(protect_home_read_only_table
) :
1694 ((ns_info
->protect_home
== PROTECT_HOME_TMPFS
) ?
1695 ELEMENTSOF(protect_home_tmpfs_table
) : 0)));
1697 return !!tmp_dir
+ !!var_tmp_dir
+
1698 strv_length(read_write_paths
) +
1699 strv_length(read_only_paths
) +
1700 strv_length(inaccessible_paths
) +
1701 strv_length(exec_paths
) +
1702 strv_length(no_exec_paths
) +
1703 strv_length(empty_directories
) +
1706 (n_extension_images
> 0 || n_extension_directories
> 0 ? /* Mount each image and directory plus an overlay per hierarchy */
1707 n_hierarchies
+ n_extension_images
+ n_extension_directories
: 0) +
1708 n_temporary_filesystems
+
1709 ns_info
->private_dev
+
1710 (ns_info
->protect_kernel_tunables
?
1711 ELEMENTSOF(protect_kernel_tunables_proc_table
) + ELEMENTSOF(protect_kernel_tunables_sys_table
) : 0) +
1712 (ns_info
->protect_kernel_modules
? ELEMENTSOF(protect_kernel_modules_table
) : 0) +
1713 (ns_info
->protect_kernel_logs
?
1714 ELEMENTSOF(protect_kernel_logs_proc_table
) + ELEMENTSOF(protect_kernel_logs_dev_table
) : 0) +
1715 (ns_info
->protect_control_groups
? 1 : 0) +
1716 protect_home_cnt
+ protect_system_cnt
+
1717 (ns_info
->protect_hostname
? 2 : 0) +
1718 (namespace_info_mount_apivfs(ns_info
) ? ELEMENTSOF(apivfs_table
) : 0) +
1719 (creds_path
? 2 : 1) +
1721 setup_propagate
+ /* /run/systemd/incoming */
1723 ns_info
->private_ipc
; /* /dev/mqueue */
1726 /* Walk all mount entries and dropping any unused mounts. This affects all
1728 * - that are implicitly protected by a path that has been rendered inaccessible
1729 * - whose immediate parent requests the same protection mode as the mount itself
1730 * - that are outside of the relevant root directory
1731 * - which are duplicates
1733 static void drop_unused_mounts(const char *root_directory
, MountEntry
*mounts
, size_t *n_mounts
) {
1734 assert(root_directory
);
1736 assert(mounts
|| *n_mounts
== 0);
1738 typesafe_qsort(mounts
, *n_mounts
, mount_path_compare
);
1740 drop_duplicates(mounts
, n_mounts
);
1741 drop_outside_root(root_directory
, mounts
, n_mounts
);
1742 drop_inaccessible(mounts
, n_mounts
);
1743 drop_nop(mounts
, n_mounts
);
1746 static int create_symlinks_from_tuples(const char *root
, char **strv_symlinks
) {
1749 STRV_FOREACH_PAIR(src
, dst
, strv_symlinks
) {
1750 _cleanup_free_
char *src_abs
= NULL
, *dst_abs
= NULL
;
1752 src_abs
= path_join(root
, *src
);
1753 dst_abs
= path_join(root
, *dst
);
1754 if (!src_abs
|| !dst_abs
)
1757 r
= mkdir_parents_label(dst_abs
, 0755);
1761 r
= symlink_idempotent(src_abs
, dst_abs
, true);
1769 static int apply_mounts(
1771 const NamespaceInfo
*ns_info
,
1774 char **exec_dir_symlinks
,
1775 char **error_path
) {
1777 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
1778 _cleanup_free_
char **deny_list
= NULL
;
1781 if (n_mounts
== 0) /* Shortcut: nothing to do */
1788 /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of
1789 * /proc. For example, this is the case with the option: 'InaccessiblePaths=/proc'. */
1790 proc_self_mountinfo
= fopen("/proc/self/mountinfo", "re");
1791 if (!proc_self_mountinfo
) {
1795 *error_path
= strdup("/proc/self/mountinfo");
1797 return log_debug_errno(r
, "Failed to open /proc/self/mountinfo: %m");
1800 /* First round, establish all mounts we need */
1804 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1809 /* ExtensionImages/Directories are first opened in the propagate directory, not in the root_directory */
1810 r
= follow_symlink(!IN_SET(m
->mode
, EXTENSION_IMAGES
, EXTENSION_DIRECTORIES
) ? root
: NULL
, m
);
1812 if (error_path
&& mount_entry_path(m
))
1813 *error_path
= strdup(mount_entry_path(m
));
1817 /* We hit a symlinked mount point. The entry got rewritten and might
1818 * point to a very different place now. Let's normalize the changed
1819 * list, and start from the beginning. After all to mount the entry
1820 * at the new location we might need some other mounts first */
1825 r
= apply_one_mount(root
, m
, ns_info
);
1827 if (error_path
&& mount_entry_path(m
))
1828 *error_path
= strdup(mount_entry_path(m
));
1838 drop_unused_mounts(root
, mounts
, n_mounts
);
1841 /* Now that all filesystems have been set up, but before the
1842 * read-only switches are flipped, create the exec dirs symlinks.
1843 * Note that when /var/lib is not empty/tmpfs, these symlinks will already
1844 * exist, which means this will be a no-op. */
1845 r
= create_symlinks_from_tuples(root
, exec_dir_symlinks
);
1847 return log_debug_errno(r
, "Failed to set up ExecDirectories symlinks inside mount namespace: %m");
1849 /* Create a deny list we can pass to bind_mount_recursive() */
1850 deny_list
= new(char*, (*n_mounts
)+1);
1853 for (size_t j
= 0; j
< *n_mounts
; j
++)
1854 deny_list
[j
] = (char*) mount_entry_path(mounts
+j
);
1855 deny_list
[*n_mounts
] = NULL
;
1857 /* Second round, flip the ro bits if necessary. */
1858 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1859 r
= make_read_only(m
, deny_list
, proc_self_mountinfo
);
1861 if (error_path
&& mount_entry_path(m
))
1862 *error_path
= strdup(mount_entry_path(m
));
1867 /* Third round, flip the noexec bits with a simplified deny list. */
1868 for (size_t j
= 0; j
< *n_mounts
; j
++)
1869 if (IN_SET((mounts
+j
)->mode
, EXEC
, NOEXEC
))
1870 deny_list
[j
] = (char*) mount_entry_path(mounts
+j
);
1871 deny_list
[*n_mounts
] = NULL
;
1873 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1874 r
= make_noexec(m
, deny_list
, proc_self_mountinfo
);
1876 if (error_path
&& mount_entry_path(m
))
1877 *error_path
= strdup(mount_entry_path(m
));
1882 /* Fourth round, flip the nosuid bits without a deny list. */
1883 if (ns_info
->mount_nosuid
)
1884 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1885 r
= make_nosuid(m
, proc_self_mountinfo
);
1887 if (error_path
&& mount_entry_path(m
))
1888 *error_path
= strdup(mount_entry_path(m
));
1896 static bool root_read_only(
1897 char **read_only_paths
,
1898 ProtectSystem protect_system
) {
1900 /* Determine whether the root directory is going to be read-only given the configured settings. */
1902 if (protect_system
== PROTECT_SYSTEM_STRICT
)
1905 if (prefixed_path_strv_contains(read_only_paths
, "/"))
1911 static bool home_read_only(
1912 char** read_only_paths
,
1913 char** inaccessible_paths
,
1914 char** empty_directories
,
1915 const BindMount
*bind_mounts
,
1916 size_t n_bind_mounts
,
1917 const TemporaryFileSystem
*temporary_filesystems
,
1918 size_t n_temporary_filesystems
,
1919 ProtectHome protect_home
) {
1921 /* Determine whether the /home directory is going to be read-only given the configured settings. Yes,
1922 * this is a bit sloppy, since we don't bother checking for cases where / is affected by multiple
1925 if (protect_home
!= PROTECT_HOME_NO
)
1928 if (prefixed_path_strv_contains(read_only_paths
, "/home") ||
1929 prefixed_path_strv_contains(inaccessible_paths
, "/home") ||
1930 prefixed_path_strv_contains(empty_directories
, "/home"))
1933 for (size_t i
= 0; i
< n_temporary_filesystems
; i
++)
1934 if (path_equal(temporary_filesystems
[i
].path
, "/home"))
1937 /* If /home is overmounted with some dir from the host it's not writable. */
1938 for (size_t i
= 0; i
< n_bind_mounts
; i
++)
1939 if (path_equal(bind_mounts
[i
].destination
, "/home"))
1945 static int verity_settings_prepare(
1946 VeritySettings
*verity
,
1947 const char *root_image
,
1948 const void *root_hash
,
1949 size_t root_hash_size
,
1950 const char *root_hash_path
,
1951 const void *root_hash_sig
,
1952 size_t root_hash_sig_size
,
1953 const char *root_hash_sig_path
,
1954 const char *verity_data_path
) {
1963 d
= memdup(root_hash
, root_hash_size
);
1967 free_and_replace(verity
->root_hash
, d
);
1968 verity
->root_hash_size
= root_hash_size
;
1969 verity
->designator
= PARTITION_ROOT
;
1972 if (root_hash_sig
) {
1975 d
= memdup(root_hash_sig
, root_hash_sig_size
);
1979 free_and_replace(verity
->root_hash_sig
, d
);
1980 verity
->root_hash_sig_size
= root_hash_sig_size
;
1981 verity
->designator
= PARTITION_ROOT
;
1984 if (verity_data_path
) {
1985 r
= free_and_strdup(&verity
->data_path
, verity_data_path
);
1990 r
= verity_settings_load(
1994 root_hash_sig_path
);
1996 return log_debug_errno(r
, "Failed to load root hash: %m");
2001 int setup_namespace(
2002 const char* root_directory
,
2003 const char* root_image
,
2004 const MountOptions
*root_image_options
,
2005 const NamespaceInfo
*ns_info
,
2006 char** read_write_paths
,
2007 char** read_only_paths
,
2008 char** inaccessible_paths
,
2010 char** no_exec_paths
,
2011 char** empty_directories
,
2012 char** exec_dir_symlinks
,
2013 const BindMount
*bind_mounts
,
2014 size_t n_bind_mounts
,
2015 const TemporaryFileSystem
*temporary_filesystems
,
2016 size_t n_temporary_filesystems
,
2017 const MountImage
*mount_images
,
2018 size_t n_mount_images
,
2019 const char* tmp_dir
,
2020 const char* var_tmp_dir
,
2021 const char *creds_path
,
2022 const char *log_namespace
,
2023 unsigned long mount_flags
,
2024 const void *root_hash
,
2025 size_t root_hash_size
,
2026 const char *root_hash_path
,
2027 const void *root_hash_sig
,
2028 size_t root_hash_sig_size
,
2029 const char *root_hash_sig_path
,
2030 const char *verity_data_path
,
2031 const MountImage
*extension_images
,
2032 size_t n_extension_images
,
2033 char **extension_directories
,
2034 const char *propagate_dir
,
2035 const char *incoming_dir
,
2036 const char *extension_dir
,
2037 const char *notify_socket
,
2038 char **error_path
) {
2040 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
2041 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
2042 _cleanup_(verity_settings_done
) VeritySettings verity
= VERITY_SETTINGS_DEFAULT
;
2043 _cleanup_strv_free_
char **hierarchies
= NULL
;
2044 MountEntry
*m
= NULL
, *mounts
= NULL
;
2045 bool require_prefix
= false, setup_propagate
= false;
2047 DissectImageFlags dissect_image_flags
=
2048 DISSECT_IMAGE_GENERIC_ROOT
|
2049 DISSECT_IMAGE_REQUIRE_ROOT
|
2050 DISSECT_IMAGE_DISCARD_ON_LOOP
|
2051 DISSECT_IMAGE_RELAX_VAR_CHECK
|
2052 DISSECT_IMAGE_FSCK
|
2053 DISSECT_IMAGE_USR_NO_ROOT
|
2054 DISSECT_IMAGE_GROWFS
|
2055 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
2056 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
2062 /* Make sure that all mknod(), mkdir() calls we do are unaffected by the umask, and the access modes
2063 * we configure take effect */
2064 BLOCK_WITH_UMASK(0000);
2066 if (!isempty(propagate_dir
) && !isempty(incoming_dir
))
2067 setup_propagate
= true;
2069 if (mount_flags
== 0)
2070 mount_flags
= MS_SHARED
;
2073 /* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */
2074 if (root_read_only(read_only_paths
,
2075 ns_info
->protect_system
) &&
2076 home_read_only(read_only_paths
, inaccessible_paths
, empty_directories
,
2077 bind_mounts
, n_bind_mounts
, temporary_filesystems
, n_temporary_filesystems
,
2078 ns_info
->protect_home
) &&
2079 strv_isempty(read_write_paths
))
2080 dissect_image_flags
|= DISSECT_IMAGE_READ_ONLY
;
2082 r
= verity_settings_prepare(
2085 root_hash
, root_hash_size
, root_hash_path
,
2086 root_hash_sig
, root_hash_sig_size
, root_hash_sig_path
,
2091 SET_FLAG(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
, verity
.data_path
);
2093 r
= loop_device_make_by_path(
2095 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_DEVICE_READ_ONLY
) ? O_RDONLY
: -1 /* < 0 means writable if possible, read-only as fallback */,
2096 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
2100 return log_debug_errno(r
, "Failed to create loop device for root image: %m");
2102 r
= dissect_loop_device(
2106 dissect_image_flags
,
2109 return log_debug_errno(r
, "Failed to dissect image: %m");
2111 r
= dissected_image_load_verity_sig_partition(
2118 r
= dissected_image_decrypt(
2122 dissect_image_flags
);
2124 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
2128 root
= root_directory
;
2130 /* /run/systemd should have been created by PID 1 early on already, but in some cases, like
2131 * when running tests (test-execute), it might not have been created yet so let's make sure
2132 * we create it if it doesn't already exist. */
2133 (void) mkdir_p_label("/run/systemd", 0755);
2135 /* Always create the mount namespace in a temporary directory, instead of operating directly
2136 * in the root. The temporary directory prevents any mounts from being potentially obscured
2137 * my other mounts we already applied. We use the same mount point for all images, which is
2138 * safe, since they all live in their own namespaces after all, and hence won't see each
2141 root
= "/run/systemd/unit-root";
2142 (void) mkdir_label(root
, 0700);
2143 require_prefix
= true;
2146 if (n_extension_images
> 0 || !strv_isempty(extension_directories
)) {
2147 r
= parse_env_extension_hierarchies(&hierarchies
);
2152 n_mounts
= namespace_calculate_mounts(
2161 n_temporary_filesystems
,
2164 strv_length(extension_directories
),
2165 strv_length(hierarchies
),
2166 tmp_dir
, var_tmp_dir
,
2173 m
= mounts
= new0(MountEntry
, n_mounts
);
2177 r
= append_access_mounts(&m
, read_write_paths
, READWRITE
, require_prefix
);
2181 r
= append_access_mounts(&m
, read_only_paths
, READONLY
, require_prefix
);
2185 r
= append_access_mounts(&m
, inaccessible_paths
, INACCESSIBLE
, require_prefix
);
2189 r
= append_access_mounts(&m
, exec_paths
, EXEC
, require_prefix
);
2193 r
= append_access_mounts(&m
, no_exec_paths
, NOEXEC
, require_prefix
);
2197 r
= append_empty_dir_mounts(&m
, empty_directories
);
2201 r
= append_bind_mounts(&m
, bind_mounts
, n_bind_mounts
);
2205 r
= append_tmpfs_mounts(&m
, temporary_filesystems
, n_temporary_filesystems
);
2210 bool ro
= streq(tmp_dir
, RUN_SYSTEMD_EMPTY
);
2212 *(m
++) = (MountEntry
) {
2213 .path_const
= "/tmp",
2214 .mode
= ro
? PRIVATE_TMP_READONLY
: PRIVATE_TMP
,
2215 .source_const
= tmp_dir
,
2220 bool ro
= streq(var_tmp_dir
, RUN_SYSTEMD_EMPTY
);
2222 *(m
++) = (MountEntry
) {
2223 .path_const
= "/var/tmp",
2224 .mode
= ro
? PRIVATE_TMP_READONLY
: PRIVATE_TMP
,
2225 .source_const
= var_tmp_dir
,
2229 r
= append_mount_images(&m
, mount_images
, n_mount_images
);
2233 r
= append_extensions(&m
, root
, extension_dir
, hierarchies
, extension_images
, n_extension_images
, extension_directories
);
2237 if (ns_info
->private_dev
)
2238 *(m
++) = (MountEntry
) {
2239 .path_const
= "/dev",
2240 .mode
= PRIVATE_DEV
,
2241 .flags
= DEV_MOUNT_OPTIONS
,
2244 /* In case /proc is successfully mounted with pid tree subset only (ProcSubset=pid), the
2245 protective mounts to non-pid /proc paths would fail. But the pid only option may have
2246 failed gracefully, so let's try the mounts but it's not fatal if they don't succeed. */
2247 bool ignore_protect_proc
= ns_info
->ignore_protect_paths
|| ns_info
->proc_subset
== PROC_SUBSET_PID
;
2248 if (ns_info
->protect_kernel_tunables
) {
2249 r
= append_static_mounts(&m
,
2250 protect_kernel_tunables_proc_table
,
2251 ELEMENTSOF(protect_kernel_tunables_proc_table
),
2252 ignore_protect_proc
);
2256 r
= append_static_mounts(&m
,
2257 protect_kernel_tunables_sys_table
,
2258 ELEMENTSOF(protect_kernel_tunables_sys_table
),
2259 ns_info
->ignore_protect_paths
);
2264 if (ns_info
->protect_kernel_modules
) {
2265 r
= append_static_mounts(&m
,
2266 protect_kernel_modules_table
,
2267 ELEMENTSOF(protect_kernel_modules_table
),
2268 ns_info
->ignore_protect_paths
);
2273 if (ns_info
->protect_kernel_logs
) {
2274 r
= append_static_mounts(&m
,
2275 protect_kernel_logs_proc_table
,
2276 ELEMENTSOF(protect_kernel_logs_proc_table
),
2277 ignore_protect_proc
);
2281 r
= append_static_mounts(&m
,
2282 protect_kernel_logs_dev_table
,
2283 ELEMENTSOF(protect_kernel_logs_dev_table
),
2284 ns_info
->ignore_protect_paths
);
2289 if (ns_info
->protect_control_groups
)
2290 *(m
++) = (MountEntry
) {
2291 .path_const
= "/sys/fs/cgroup",
2295 r
= append_protect_home(&m
, ns_info
->protect_home
, ns_info
->ignore_protect_paths
);
2299 r
= append_protect_system(&m
, ns_info
->protect_system
, false);
2303 if (namespace_info_mount_apivfs(ns_info
)) {
2304 r
= append_static_mounts(&m
,
2306 ELEMENTSOF(apivfs_table
),
2307 ns_info
->ignore_protect_paths
);
2312 /* Note, if proc is mounted with subset=pid then neither of the
2313 * two paths will exist, i.e. they are implicitly protected by
2314 * the mount option. */
2315 if (ns_info
->protect_hostname
) {
2316 *(m
++) = (MountEntry
) {
2317 .path_const
= "/proc/sys/kernel/hostname",
2319 .ignore
= ignore_protect_proc
,
2321 *(m
++) = (MountEntry
) {
2322 .path_const
= "/proc/sys/kernel/domainname",
2324 .ignore
= ignore_protect_proc
,
2328 if (ns_info
->private_ipc
)
2329 *(m
++) = (MountEntry
) {
2330 .path_const
= "/dev/mqueue",
2332 .flags
= MS_NOSUID
| MS_NODEV
| MS_NOEXEC
| MS_RELATIME
,
2336 /* If our service has a credentials store configured, then bind that one in, but hide
2337 * everything else. */
2339 *(m
++) = (MountEntry
) {
2340 .path_const
= "/run/credentials",
2343 .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
,
2344 .flags
= MS_NODEV
|MS_STRICTATIME
|MS_NOSUID
|MS_NOEXEC
,
2347 *(m
++) = (MountEntry
) {
2348 .path_const
= creds_path
,
2351 .source_const
= creds_path
,
2354 /* If our service has no credentials store configured, then make the whole
2355 * credentials tree inaccessible wholesale. */
2357 *(m
++) = (MountEntry
) {
2358 .path_const
= "/run/credentials",
2359 .mode
= INACCESSIBLE
,
2364 if (log_namespace
) {
2365 _cleanup_free_
char *q
= NULL
;
2367 q
= strjoin("/run/systemd/journal.", log_namespace
);
2373 *(m
++) = (MountEntry
) {
2374 .path_const
= "/run/systemd/journal",
2375 .mode
= BIND_MOUNT_RECURSIVE
,
2377 .source_malloc
= TAKE_PTR(q
),
2381 /* Will be used to add bind mounts at runtime */
2382 if (setup_propagate
)
2383 *(m
++) = (MountEntry
) {
2384 .source_const
= propagate_dir
,
2385 .path_const
= incoming_dir
,
2391 *(m
++) = (MountEntry
) {
2392 .path_const
= notify_socket
,
2393 .source_const
= notify_socket
,
2398 assert(mounts
+ n_mounts
== m
);
2400 /* Prepend the root directory where that's necessary */
2401 r
= prefix_where_needed(mounts
, n_mounts
, root
);
2405 drop_unused_mounts(root
, mounts
, &n_mounts
);
2408 /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */
2410 if (unshare(CLONE_NEWNS
) < 0) {
2411 r
= log_debug_errno(errno
, "Failed to unshare the mount namespace: %m");
2412 if (IN_SET(r
, -EACCES
, -EPERM
, -EOPNOTSUPP
, -ENOSYS
))
2413 /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter
2414 * in place that doesn't allow us to create namespaces (or a missing cap), then
2415 * propagate a recognizable error back, which the caller can use to detect this case
2416 * (and only this) and optionally continue without namespacing applied. */
2422 /* Create the source directory to allow runtime propagation of mounts */
2423 if (setup_propagate
)
2424 (void) mkdir_p(propagate_dir
, 0600);
2426 if (n_extension_images
> 0 || !strv_isempty(extension_directories
))
2427 /* ExtensionImages/Directories mountpoint directories will be created while parsing the
2428 * mounts to create, so have the parent ready */
2429 (void) mkdir_p(extension_dir
, 0600);
2431 /* Remount / as SLAVE so that nothing now mounted in the namespace
2432 * shows up in the parent */
2433 if (mount(NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
) < 0) {
2434 r
= log_debug_errno(errno
, "Failed to remount '/' as SLAVE: %m");
2439 /* A root image is specified, mount it to the right place */
2440 r
= dissected_image_mount(dissected_image
, root
, UID_INVALID
, UID_INVALID
, dissect_image_flags
);
2442 log_debug_errno(r
, "Failed to mount root image: %m");
2446 /* Now release the block device lock, so that udevd is free to call BLKRRPART on the device
2448 r
= loop_device_flock(loop_device
, LOCK_UN
);
2450 log_debug_errno(r
, "Failed to release lock on loopback block device: %m");
2454 r
= dissected_image_relinquish(dissected_image
);
2456 log_debug_errno(r
, "Failed to relinquish dissected image: %m");
2460 } else if (root_directory
) {
2462 /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
2463 r
= path_is_mount_point(root
, NULL
, AT_SYMLINK_FOLLOW
);
2465 log_debug_errno(r
, "Failed to detect that %s is a mount point or not: %m", root
);
2469 r
= mount_nofollow_verbose(LOG_DEBUG
, root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2475 /* Let's mount the main root directory to the root directory to use */
2476 r
= mount_nofollow_verbose(LOG_DEBUG
, "/", root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2481 /* Try to set up the new root directory before mounting anything else there. */
2482 if (root_image
|| root_directory
)
2483 (void) base_filesystem_create(root
, UID_INVALID
, GID_INVALID
);
2485 /* Now make the magic happen */
2486 r
= apply_mounts(root
, ns_info
, mounts
, &n_mounts
, exec_dir_symlinks
, error_path
);
2490 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
2491 r
= mount_switch_root(root
, /* mount_propagation_flag = */ 0);
2492 if (r
== -EINVAL
&& root_directory
) {
2493 /* If we are using root_directory and we don't have privileges (ie: user manager in a user
2494 * namespace) and the root_directory is already a mount point in the parent namespace,
2495 * MS_MOVE will fail as we don't have permission to change it (with EINVAL rather than
2496 * EPERM). Attempt to bind-mount it over itself (like we do above if it's not already a
2497 * mount point) and try again. */
2498 r
= mount_nofollow_verbose(LOG_DEBUG
, root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2501 r
= mount_switch_root(root
, /* mount_propagation_flag = */ 0);
2504 log_debug_errno(r
, "Failed to mount root with MS_MOVE: %m");
2508 /* Remount / as the desired mode. Note that this will not
2509 * reestablish propagation from our side to the host, since
2510 * what's disconnected is disconnected. */
2511 if (mount(NULL
, "/", NULL
, mount_flags
| MS_REC
, NULL
) < 0) {
2512 r
= log_debug_errno(errno
, "Failed to remount '/' with desired mount flags: %m");
2516 /* bind_mount_in_namespace() will MS_MOVE into that directory, and that's only
2517 * supported for non-shared mounts. This needs to happen after remounting / or it will fail. */
2518 if (setup_propagate
) {
2519 r
= mount(NULL
, incoming_dir
, NULL
, MS_SLAVE
, NULL
);
2521 log_error_errno(r
, "Failed to remount %s with MS_SLAVE: %m", incoming_dir
);
2530 for (m
= mounts
; m
< mounts
+ n_mounts
; m
++)
2531 mount_entry_done(m
);
2538 void bind_mount_free_many(BindMount
*b
, size_t n
) {
2539 assert(b
|| n
== 0);
2541 for (size_t i
= 0; i
< n
; i
++) {
2543 free(b
[i
].destination
);
2549 int bind_mount_add(BindMount
**b
, size_t *n
, const BindMount
*item
) {
2550 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
2557 s
= strdup(item
->source
);
2561 d
= strdup(item
->destination
);
2565 c
= reallocarray(*b
, *n
+ 1, sizeof(BindMount
));
2571 c
[(*n
) ++] = (BindMount
) {
2572 .source
= TAKE_PTR(s
),
2573 .destination
= TAKE_PTR(d
),
2574 .read_only
= item
->read_only
,
2575 .nosuid
= item
->nosuid
,
2576 .recursive
= item
->recursive
,
2577 .ignore_enoent
= item
->ignore_enoent
,
2583 MountImage
* mount_image_free_many(MountImage
*m
, size_t *n
) {
2585 assert(m
|| *n
== 0);
2587 for (size_t i
= 0; i
< *n
; i
++) {
2589 free(m
[i
].destination
);
2590 mount_options_free_all(m
[i
].mount_options
);
2598 int mount_image_add(MountImage
**m
, size_t *n
, const MountImage
*item
) {
2599 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
2600 _cleanup_(mount_options_free_allp
) MountOptions
*options
= NULL
;
2607 s
= strdup(item
->source
);
2611 if (item
->destination
) {
2612 d
= strdup(item
->destination
);
2617 LIST_FOREACH(mount_options
, i
, item
->mount_options
) {
2618 _cleanup_(mount_options_free_allp
) MountOptions
*o
= NULL
;
2620 o
= new(MountOptions
, 1);
2624 *o
= (MountOptions
) {
2625 .partition_designator
= i
->partition_designator
,
2626 .options
= strdup(i
->options
),
2631 LIST_APPEND(mount_options
, options
, TAKE_PTR(o
));
2634 c
= reallocarray(*m
, *n
+ 1, sizeof(MountImage
));
2640 c
[(*n
) ++] = (MountImage
) {
2641 .source
= TAKE_PTR(s
),
2642 .destination
= TAKE_PTR(d
),
2643 .mount_options
= TAKE_PTR(options
),
2644 .ignore_enoent
= item
->ignore_enoent
,
2651 void temporary_filesystem_free_many(TemporaryFileSystem
*t
, size_t n
) {
2652 assert(t
|| n
== 0);
2654 for (size_t i
= 0; i
< n
; i
++) {
2662 int temporary_filesystem_add(
2663 TemporaryFileSystem
**t
,
2666 const char *options
) {
2668 _cleanup_free_
char *p
= NULL
, *o
= NULL
;
2669 TemporaryFileSystem
*c
;
2679 if (!isempty(options
)) {
2680 o
= strdup(options
);
2685 c
= reallocarray(*t
, *n
+ 1, sizeof(TemporaryFileSystem
));
2691 c
[(*n
) ++] = (TemporaryFileSystem
) {
2692 .path
= TAKE_PTR(p
),
2693 .options
= TAKE_PTR(o
),
2699 static int make_tmp_prefix(const char *prefix
) {
2700 _cleanup_free_
char *t
= NULL
;
2701 _cleanup_close_
int fd
= -EBADF
;
2704 /* Don't do anything unless we know the dir is actually missing */
2705 r
= access(prefix
, F_OK
);
2708 if (errno
!= ENOENT
)
2712 r
= mkdir_parents(prefix
, 0755);
2716 r
= tempfn_random(prefix
, NULL
, &t
);
2720 /* umask will corrupt this access mode, but that doesn't matter, we need to call chmod() anyway for
2721 * the suid bit, below. */
2722 fd
= open_mkdir_at(AT_FDCWD
, t
, O_EXCL
|O_CLOEXEC
, 0777);
2726 r
= RET_NERRNO(fchmod(fd
, 01777));
2732 r
= RET_NERRNO(rename(t
, prefix
));
2735 return r
== -EEXIST
? 0 : r
; /* it's fine if someone else created the dir by now */
2742 static int setup_one_tmp_dir(const char *id
, const char *prefix
, char **path
, char **tmp_path
) {
2743 _cleanup_free_
char *x
= NULL
;
2744 _cleanup_free_
char *y
= NULL
;
2753 /* We include the boot id in the directory so that after a
2754 * reboot we can easily identify obsolete directories. */
2756 r
= sd_id128_get_boot(&boot_id
);
2760 x
= strjoin(prefix
, "/systemd-private-", SD_ID128_TO_STRING(boot_id
), "-", id
, "-XXXXXX");
2764 r
= make_tmp_prefix(prefix
);
2770 if (errno
== EROFS
|| ERRNO_IS_DISK_SPACE(errno
))
2777 y
= strjoin(x
, "/tmp");
2782 if (mkdir(y
, 0777 | S_ISVTX
) < 0)
2785 r
= label_fix_full(AT_FDCWD
, y
, prefix
, 0);
2790 *tmp_path
= TAKE_PTR(y
);
2792 /* Trouble: we failed to create the directory. Instead of failing, let's simulate /tmp being
2793 * read-only. This way the service will get the EROFS result as if it was writing to the real
2796 r
= mkdir_p(RUN_SYSTEMD_EMPTY
, 0500);
2800 r
= free_and_strdup(&x
, RUN_SYSTEMD_EMPTY
);
2805 *path
= TAKE_PTR(x
);
2809 int setup_tmp_dirs(const char *id
, char **tmp_dir
, char **var_tmp_dir
) {
2810 _cleanup_(namespace_cleanup_tmpdirp
) char *a
= NULL
;
2811 _cleanup_(rmdir_and_freep
) char *a_tmp
= NULL
;
2817 assert(var_tmp_dir
);
2819 r
= setup_one_tmp_dir(id
, "/tmp", &a
, &a_tmp
);
2823 r
= setup_one_tmp_dir(id
, "/var/tmp", &b
, NULL
);
2827 a_tmp
= mfree(a_tmp
); /* avoid rmdir */
2828 *tmp_dir
= TAKE_PTR(a
);
2829 *var_tmp_dir
= TAKE_PTR(b
);
2834 int setup_shareable_ns(const int ns_storage_socket
[static 2], unsigned long nsflag
) {
2835 _cleanup_close_
int ns
= -1;
2837 const char *ns_name
, *ns_path
;
2839 assert(ns_storage_socket
);
2840 assert(ns_storage_socket
[0] >= 0);
2841 assert(ns_storage_socket
[1] >= 0);
2843 ns_name
= namespace_single_flag_to_string(nsflag
);
2846 /* We use the passed socketpair as a storage buffer for our
2847 * namespace reference fd. Whatever process runs this first
2848 * shall create a new namespace, all others should just join
2849 * it. To serialize that we use a file lock on the socket
2852 * It's a bit crazy, but hey, works great! */
2854 if (lockf(ns_storage_socket
[0], F_LOCK
, 0) < 0)
2857 ns
= receive_one_fd(ns_storage_socket
[0], MSG_DONTWAIT
);
2858 if (ns
== -EAGAIN
) {
2859 /* Nothing stored yet, so let's create a new namespace. */
2861 if (unshare(nsflag
) < 0) {
2866 (void) loopback_setup();
2868 ns_path
= strjoina("/proc/self/ns/", ns_name
);
2869 ns
= open(ns_path
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
2877 } else if (ns
< 0) {
2882 /* Yay, found something, so let's join the namespace */
2883 if (setns(ns
, nsflag
) < 0) {
2891 q
= send_one_fd(ns_storage_socket
[1], ns
, MSG_DONTWAIT
);
2898 (void) lockf(ns_storage_socket
[0], F_ULOCK
, 0);
2902 int open_shareable_ns_path(const int ns_storage_socket
[static 2], const char *path
, unsigned long nsflag
) {
2903 _cleanup_close_
int ns
= -1;
2906 assert(ns_storage_socket
);
2907 assert(ns_storage_socket
[0] >= 0);
2908 assert(ns_storage_socket
[1] >= 0);
2911 /* If the storage socket doesn't contain a ns fd yet, open one via the file system and store it in
2912 * it. This is supposed to be called ahead of time, i.e. before setup_shareable_ns() which will
2913 * allocate a new anonymous ns if needed. */
2915 if (lockf(ns_storage_socket
[0], F_LOCK
, 0) < 0)
2918 ns
= receive_one_fd(ns_storage_socket
[0], MSG_DONTWAIT
);
2919 if (ns
== -EAGAIN
) {
2920 /* Nothing stored yet. Open the file from the file system. */
2922 ns
= open(path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
2928 r
= fd_is_ns(ns
, nsflag
);
2929 if (r
== 0) { /* Not a ns of our type? Refuse early. */
2933 if (r
< 0 && r
!= -EUCLEAN
) /* EUCLEAN: we don't know */
2938 } else if (ns
< 0) {
2942 r
= 0; /* Already allocated */
2944 q
= send_one_fd(ns_storage_socket
[1], ns
, MSG_DONTWAIT
);
2951 (void) lockf(ns_storage_socket
[0], F_ULOCK
, 0);
2955 bool ns_type_supported(NamespaceType type
) {
2956 const char *t
, *ns_proc
;
2958 t
= namespace_type_to_string(type
);
2959 if (!t
) /* Don't know how to translate this? Then it's not supported */
2962 ns_proc
= strjoina("/proc/self/ns/", t
);
2963 return access(ns_proc
, F_OK
) == 0;
2966 static const char *const protect_home_table
[_PROTECT_HOME_MAX
] = {
2967 [PROTECT_HOME_NO
] = "no",
2968 [PROTECT_HOME_YES
] = "yes",
2969 [PROTECT_HOME_READ_ONLY
] = "read-only",
2970 [PROTECT_HOME_TMPFS
] = "tmpfs",
2973 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home
, ProtectHome
, PROTECT_HOME_YES
);
2975 static const char *const protect_system_table
[_PROTECT_SYSTEM_MAX
] = {
2976 [PROTECT_SYSTEM_NO
] = "no",
2977 [PROTECT_SYSTEM_YES
] = "yes",
2978 [PROTECT_SYSTEM_FULL
] = "full",
2979 [PROTECT_SYSTEM_STRICT
] = "strict",
2982 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system
, ProtectSystem
, PROTECT_SYSTEM_YES
);
2984 static const char* const namespace_type_table
[] = {
2985 [NAMESPACE_MOUNT
] = "mnt",
2986 [NAMESPACE_CGROUP
] = "cgroup",
2987 [NAMESPACE_UTS
] = "uts",
2988 [NAMESPACE_IPC
] = "ipc",
2989 [NAMESPACE_USER
] = "user",
2990 [NAMESPACE_PID
] = "pid",
2991 [NAMESPACE_NET
] = "net",
2992 [NAMESPACE_TIME
] = "time",
2995 DEFINE_STRING_TABLE_LOOKUP(namespace_type
, NamespaceType
);
2997 static const char* const protect_proc_table
[_PROTECT_PROC_MAX
] = {
2998 [PROTECT_PROC_DEFAULT
] = "default",
2999 [PROTECT_PROC_NOACCESS
] = "noaccess",
3000 [PROTECT_PROC_INVISIBLE
] = "invisible",
3001 [PROTECT_PROC_PTRACEABLE
] = "ptraceable",
3004 DEFINE_STRING_TABLE_LOOKUP(protect_proc
, ProtectProc
);
3006 static const char* const proc_subset_table
[_PROC_SUBSET_MAX
] = {
3007 [PROC_SUBSET_ALL
] = "all",
3008 [PROC_SUBSET_PID
] = "pid",
3011 DEFINE_STRING_TABLE_LOOKUP(proc_subset
, ProcSubset
);