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 [BIND_MOUNT
] = "bind",
214 [BIND_MOUNT_RECURSIVE
] = "rbind",
215 [PRIVATE_TMP
] = "private-tmp",
216 [PRIVATE_DEV
] = "private-dev",
217 [BIND_DEV
] = "bind-dev",
218 [EMPTY_DIR
] = "empty",
219 [PRIVATE_SYSFS
] = "private-sysfs",
220 [BIND_SYSFS
] = "bind-sysfs",
222 [READONLY
] = "read-only",
223 [READWRITE
] = "read-write",
225 [MOUNT_IMAGES
] = "mount-images",
226 [READWRITE_IMPLICIT
] = "rw-implicit",
229 [MQUEUEFS
] = "mqueuefs",
232 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(mount_mode
, MountMode
);
234 static const char *mount_entry_path(const MountEntry
*p
) {
237 /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that,
238 * otherwise the stack/static ->path field is returned. */
240 return p
->path_malloc
?: p
->path_const
;
243 static const char *mount_entry_unprefixed_path(const MountEntry
*p
) {
246 /* Returns the unprefixed path (ie: before prefix_where_needed() ran), if any */
248 return p
->unprefixed_path_malloc
?: p
->unprefixed_path_const
?: mount_entry_path(p
);
251 static void mount_entry_consume_prefix(MountEntry
*p
, char *new_path
) {
253 assert(p
->path_malloc
|| p
->path_const
);
256 /* Saves current path in unprefixed_ variable, and takes over new_path */
258 free_and_replace(p
->unprefixed_path_malloc
, p
->path_malloc
);
259 /* If we didn't have a path on the heap, then it's a static one */
260 if (!p
->unprefixed_path_malloc
)
261 p
->unprefixed_path_const
= p
->path_const
;
262 p
->path_malloc
= new_path
;
263 p
->has_prefix
= true;
266 static bool mount_entry_read_only(const MountEntry
*p
) {
269 return p
->read_only
|| IN_SET(p
->mode
, READONLY
, INACCESSIBLE
, PRIVATE_TMP_READONLY
);
272 static bool mount_entry_noexec(const MountEntry
*p
) {
275 return p
->noexec
|| IN_SET(p
->mode
, NOEXEC
, INACCESSIBLE
, PRIVATE_SYSFS
, BIND_SYSFS
, PROCFS
);
278 static bool mount_entry_exec(const MountEntry
*p
) {
281 return p
->exec
|| p
->mode
== EXEC
;
284 static const char *mount_entry_source(const MountEntry
*p
) {
287 return p
->source_malloc
?: p
->source_const
;
290 static const char *mount_entry_options(const MountEntry
*p
) {
293 return p
->options_malloc
?: p
->options_const
;
296 static void mount_entry_done(MountEntry
*p
) {
299 p
->path_malloc
= mfree(p
->path_malloc
);
300 p
->unprefixed_path_malloc
= mfree(p
->unprefixed_path_malloc
);
301 p
->source_malloc
= mfree(p
->source_malloc
);
302 p
->options_malloc
= mfree(p
->options_malloc
);
303 p
->image_options
= mount_options_free_all(p
->image_options
);
306 static int append_access_mounts(MountEntry
**p
, char **strv
, MountMode mode
, bool forcibly_require_prefix
) {
309 /* Adds a list of user-supplied READWRITE/READWRITE_IMPLICIT/READONLY/INACCESSIBLE entries */
311 STRV_FOREACH(i
, strv
) {
312 bool ignore
= false, needs_prefix
= false;
315 /* Look for any prefixes */
316 if (startswith(e
, "-")) {
320 if (startswith(e
, "+")) {
325 if (!path_is_absolute(e
))
326 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
327 "Path is not absolute: %s", e
);
329 *((*p
)++) = (MountEntry
) {
333 .has_prefix
= !needs_prefix
&& !forcibly_require_prefix
,
340 static int append_empty_dir_mounts(MountEntry
**p
, char **strv
) {
343 /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the
344 * "/private/" boundary directories for DynamicUser=1. */
346 STRV_FOREACH(i
, strv
) {
348 *((*p
)++) = (MountEntry
) {
353 .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
,
354 .flags
= MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
,
361 static int append_bind_mounts(MountEntry
**p
, const BindMount
*binds
, size_t n
) {
364 for (size_t i
= 0; i
< n
; i
++) {
365 const BindMount
*b
= binds
+ i
;
367 *((*p
)++) = (MountEntry
) {
368 .path_const
= b
->destination
,
369 .mode
= b
->recursive
? BIND_MOUNT_RECURSIVE
: BIND_MOUNT
,
370 .read_only
= b
->read_only
,
372 .source_const
= b
->source
,
373 .ignore
= b
->ignore_enoent
,
380 static int append_mount_images(MountEntry
**p
, const MountImage
*mount_images
, size_t n
) {
383 for (size_t i
= 0; i
< n
; i
++) {
384 const MountImage
*m
= mount_images
+ i
;
386 *((*p
)++) = (MountEntry
) {
387 .path_const
= m
->destination
,
388 .mode
= MOUNT_IMAGES
,
389 .source_const
= m
->source
,
390 .image_options
= m
->mount_options
,
391 .ignore
= m
->ignore_enoent
,
398 static int append_extensions(
401 const char *extension_dir
,
403 const MountImage
*mount_images
,
405 char **extension_directories
) {
407 _cleanup_strv_free_
char **overlays
= NULL
;
410 if (n
== 0 && strv_isempty(extension_directories
))
414 assert(extension_dir
);
416 /* Prepare a list of overlays, that will have as each element a string suitable for being
417 * passed as a lowerdir= parameter, so start with the hierarchy on the root.
418 * The overlays vector will have the same number of elements and will correspond to the
419 * hierarchies vector, so they can be iterated upon together. */
420 STRV_FOREACH(hierarchy
, hierarchies
) {
421 _cleanup_free_
char *prefixed_hierarchy
= NULL
;
423 prefixed_hierarchy
= path_join(root
, *hierarchy
);
424 if (!prefixed_hierarchy
)
427 r
= strv_consume(&overlays
, TAKE_PTR(prefixed_hierarchy
));
432 /* First, prepare a mount for each image, but these won't be visible to the unit, instead
433 * they will be mounted in our propagate directory, and used as a source for the overlay. */
434 for (size_t i
= 0; i
< n
; i
++) {
435 _cleanup_free_
char *mount_point
= NULL
;
436 const MountImage
*m
= mount_images
+ i
;
438 r
= asprintf(&mount_point
, "%s/%zu", extension_dir
, i
);
442 for (size_t j
= 0; hierarchies
&& hierarchies
[j
]; ++j
) {
443 _cleanup_free_
char *prefixed_hierarchy
= NULL
, *escaped
= NULL
, *lowerdir
= NULL
;
445 prefixed_hierarchy
= path_join(mount_point
, hierarchies
[j
]);
446 if (!prefixed_hierarchy
)
449 escaped
= shell_escape(prefixed_hierarchy
, ",:");
453 /* Note that lowerdir= parameters are in 'reverse' order, so the
454 * top-most directory in the overlay comes first in the list. */
455 lowerdir
= strjoin(escaped
, ":", overlays
[j
]);
459 free_and_replace(overlays
[j
], lowerdir
);
462 *((*p
)++) = (MountEntry
) {
463 .path_malloc
= TAKE_PTR(mount_point
),
464 .image_options
= m
->mount_options
,
465 .ignore
= m
->ignore_enoent
,
466 .source_const
= m
->source
,
467 .mode
= EXTENSION_IMAGES
,
472 /* Secondly, extend the lowerdir= parameters with each ExtensionDirectory.
473 * Bind mount them in the same location as the ExtensionImages, so that we
474 * can check that they are valid trees (extension-release.d). */
475 STRV_FOREACH(extension_directory
, extension_directories
) {
476 _cleanup_free_
char *mount_point
= NULL
, *source
= NULL
;
477 const char *e
= *extension_directory
;
478 bool ignore_enoent
= false;
480 /* Pick up the counter where the ExtensionImages left it. */
481 r
= asprintf(&mount_point
, "%s/%zu", extension_dir
, n
++);
485 /* Look for any prefixes */
486 if (startswith(e
, "-")) {
488 ignore_enoent
= true;
490 /* Ignore this for now */
491 if (startswith(e
, "+"))
498 for (size_t j
= 0; hierarchies
&& hierarchies
[j
]; ++j
) {
499 _cleanup_free_
char *prefixed_hierarchy
= NULL
, *escaped
= NULL
, *lowerdir
= NULL
;
501 prefixed_hierarchy
= path_join(mount_point
, hierarchies
[j
]);
502 if (!prefixed_hierarchy
)
505 escaped
= shell_escape(prefixed_hierarchy
, ",:");
509 /* Note that lowerdir= parameters are in 'reverse' order, so the
510 * top-most directory in the overlay comes first in the list. */
511 lowerdir
= strjoin(escaped
, ":", overlays
[j
]);
515 free_and_replace(overlays
[j
], lowerdir
);
518 *((*p
)++) = (MountEntry
) {
519 .path_malloc
= TAKE_PTR(mount_point
),
520 .source_malloc
= TAKE_PTR(source
),
521 .mode
= EXTENSION_DIRECTORIES
,
522 .ignore
= ignore_enoent
,
528 /* Then, for each hierarchy, prepare an overlay with the list of lowerdir= strings
530 for (size_t i
= 0; hierarchies
&& hierarchies
[i
]; ++i
) {
531 _cleanup_free_
char *prefixed_hierarchy
= NULL
;
533 prefixed_hierarchy
= path_join(root
, hierarchies
[i
]);
534 if (!prefixed_hierarchy
)
537 *((*p
)++) = (MountEntry
) {
538 .path_malloc
= TAKE_PTR(prefixed_hierarchy
),
539 .options_malloc
= TAKE_PTR(overlays
[i
]),
540 .mode
= OVERLAY_MOUNT
,
542 .ignore
= true, /* If the source image doesn't set the ignore bit it will fail earlier. */
549 static int append_tmpfs_mounts(MountEntry
**p
, const TemporaryFileSystem
*tmpfs
, size_t n
) {
552 for (size_t i
= 0; i
< n
; i
++) {
553 const TemporaryFileSystem
*t
= tmpfs
+ i
;
554 _cleanup_free_
char *o
= NULL
, *str
= NULL
;
559 if (!path_is_absolute(t
->path
))
560 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
561 "Path is not absolute: %s",
564 str
= strjoin("mode=0755" NESTED_TMPFS_LIMITS
",", t
->options
);
568 r
= mount_option_mangle(str
, MS_NODEV
|MS_STRICTATIME
, &flags
, &o
);
570 return log_debug_errno(r
, "Failed to parse mount option '%s': %m", str
);
572 ro
= flags
& MS_RDONLY
;
576 *((*p
)++) = (MountEntry
) {
577 .path_const
= t
->path
,
580 .options_malloc
= TAKE_PTR(o
),
588 static int append_static_mounts(MountEntry
**p
, const MountEntry
*mounts
, size_t n
, bool ignore_protect
) {
592 /* Adds a list of static pre-defined entries */
594 for (size_t i
= 0; i
< n
; i
++)
595 *((*p
)++) = (MountEntry
) {
596 .path_const
= mount_entry_path(mounts
+i
),
597 .mode
= mounts
[i
].mode
,
598 .ignore
= mounts
[i
].ignore
|| ignore_protect
,
604 static int append_protect_home(MountEntry
**p
, ProtectHome protect_home
, bool ignore_protect
) {
607 switch (protect_home
) {
609 case PROTECT_HOME_NO
:
612 case PROTECT_HOME_READ_ONLY
:
613 return append_static_mounts(p
, protect_home_read_only_table
, ELEMENTSOF(protect_home_read_only_table
), ignore_protect
);
615 case PROTECT_HOME_TMPFS
:
616 return append_static_mounts(p
, protect_home_tmpfs_table
, ELEMENTSOF(protect_home_tmpfs_table
), ignore_protect
);
618 case PROTECT_HOME_YES
:
619 return append_static_mounts(p
, protect_home_yes_table
, ELEMENTSOF(protect_home_yes_table
), ignore_protect
);
622 assert_not_reached();
626 static int append_protect_system(MountEntry
**p
, ProtectSystem protect_system
, bool ignore_protect
) {
629 switch (protect_system
) {
631 case PROTECT_SYSTEM_NO
:
634 case PROTECT_SYSTEM_STRICT
:
635 return append_static_mounts(p
, protect_system_strict_table
, ELEMENTSOF(protect_system_strict_table
), ignore_protect
);
637 case PROTECT_SYSTEM_YES
:
638 return append_static_mounts(p
, protect_system_yes_table
, ELEMENTSOF(protect_system_yes_table
), ignore_protect
);
640 case PROTECT_SYSTEM_FULL
:
641 return append_static_mounts(p
, protect_system_full_table
, ELEMENTSOF(protect_system_full_table
), ignore_protect
);
644 assert_not_reached();
648 static int mount_path_compare(const MountEntry
*a
, const MountEntry
*b
) {
651 /* ExtensionImages/Directories will be used by other mounts as a base, so sort them first
652 * regardless of the prefix - they are set up in the propagate directory anyway */
653 d
= -CMP(a
->mode
== EXTENSION_IMAGES
, b
->mode
== EXTENSION_IMAGES
);
656 d
= -CMP(a
->mode
== EXTENSION_DIRECTORIES
, b
->mode
== EXTENSION_DIRECTORIES
);
660 /* If the paths are not equal, then order prefixes first */
661 d
= path_compare(mount_entry_path(a
), mount_entry_path(b
));
665 /* If the paths are equal, check the mode */
666 return CMP((int) a
->mode
, (int) b
->mode
);
669 static int prefix_where_needed(MountEntry
*m
, size_t n
, const char *root_directory
) {
670 /* Prefixes all paths in the bind mount table with the root directory if the entry needs that. */
674 for (size_t i
= 0; i
< n
; i
++) {
680 s
= path_join(root_directory
, mount_entry_path(m
+i
));
684 mount_entry_consume_prefix(&m
[i
], s
);
690 static void drop_duplicates(MountEntry
*m
, size_t *n
) {
691 MountEntry
*f
, *t
, *previous
;
696 /* Drops duplicate entries. Expects that the array is properly ordered already. */
698 for (f
= m
, t
= m
, previous
= NULL
; f
< m
+ *n
; f
++) {
700 /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare()
701 * above. Note that we only drop duplicates that haven't been mounted yet. */
703 path_equal(mount_entry_path(f
), mount_entry_path(previous
)) &&
704 !f
->applied
&& !previous
->applied
) {
705 log_debug("%s (%s) is duplicate.", mount_entry_path(f
), mount_mode_to_string(f
->mode
));
706 /* Propagate the flags to the remaining entry */
707 previous
->read_only
= previous
->read_only
|| mount_entry_read_only(f
);
708 previous
->noexec
= previous
->noexec
|| mount_entry_noexec(f
);
709 previous
->exec
= previous
->exec
|| mount_entry_exec(f
);
722 static void drop_inaccessible(MountEntry
*m
, size_t *n
) {
724 const char *clear
= NULL
;
729 /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly
730 * ordered already. */
732 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
734 /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop
735 * it, as inaccessible paths really should drop the entire subtree. */
736 if (clear
&& path_startswith(mount_entry_path(f
), clear
)) {
737 log_debug("%s is masked by %s.", mount_entry_path(f
), clear
);
742 clear
= f
->mode
== INACCESSIBLE
? mount_entry_path(f
) : NULL
;
751 static void drop_nop(MountEntry
*m
, size_t *n
) {
757 /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the
758 * list is ordered by prefixes. */
760 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
762 /* Only suppress such subtrees for READONLY, READWRITE and READWRITE_IMPLICIT entries */
763 if (IN_SET(f
->mode
, READONLY
, READWRITE
, READWRITE_IMPLICIT
)) {
764 MountEntry
*found
= NULL
;
766 /* Now let's find the first parent of the entry we are looking at. */
767 for (MountEntry
*p
= PTR_SUB1(t
, m
); p
; p
= PTR_SUB1(p
, m
))
768 if (path_startswith(mount_entry_path(f
), mount_entry_path(p
))) {
773 /* We found it, let's see if it's the same mode, if so, we can drop this entry */
774 if (found
&& found
->mode
== f
->mode
) {
775 log_debug("%s (%s) is made redundant by %s (%s)",
776 mount_entry_path(f
), mount_mode_to_string(f
->mode
),
777 mount_entry_path(found
), mount_mode_to_string(found
->mode
));
790 static void drop_outside_root(const char *root_directory
, MountEntry
*m
, size_t *n
) {
800 /* Drops all mounts that are outside of the root directory. */
802 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
804 /* ExtensionImages/Directories bases are opened in /run/systemd/unit-extensions on the host */
805 if (!IN_SET(f
->mode
, EXTENSION_IMAGES
, EXTENSION_DIRECTORIES
) && !path_startswith(mount_entry_path(f
), root_directory
)) {
806 log_debug("%s is outside of root directory.", mount_entry_path(f
));
818 static int clone_device_node(
820 const char *temporary_mount
,
821 bool *make_devnode
) {
823 _cleanup_free_
char *sl
= NULL
;
824 const char *dn
, *bn
, *t
;
828 if (stat(d
, &st
) < 0) {
829 if (errno
== ENOENT
) {
830 log_debug_errno(errno
, "Device node '%s' to clone does not exist, ignoring.", d
);
834 return log_debug_errno(errno
, "Failed to stat() device node '%s' to clone, ignoring: %m", d
);
837 if (!S_ISBLK(st
.st_mode
) &&
838 !S_ISCHR(st
.st_mode
))
839 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
840 "Device node '%s' to clone is not a device node, ignoring.",
843 dn
= strjoina(temporary_mount
, d
);
845 /* First, try to create device node properly */
847 mac_selinux_create_file_prepare(d
, st
.st_mode
);
848 r
= mknod(dn
, st
.st_mode
, st
.st_rdev
);
849 mac_selinux_create_file_clear();
853 return log_debug_errno(errno
, "mknod failed for %s: %m", d
);
855 /* This didn't work, let's not try this again for the next iterations. */
856 *make_devnode
= false;
859 /* We're about to fall back to bind-mounting the device node. So create a dummy bind-mount target.
860 * Do not prepare device-node SELinux label (see issue 13762) */
861 r
= mknod(dn
, S_IFREG
, 0);
862 if (r
< 0 && errno
!= EEXIST
)
863 return log_debug_errno(errno
, "mknod() fallback failed for '%s': %m", d
);
865 /* Fallback to bind-mounting: The assumption here is that all used device nodes carry standard
866 * properties. Specifically, the devices nodes we bind-mount should either be owned by root:root or
867 * root:tty (e.g. /dev/tty, /dev/ptmx) and should not carry ACLs. */
868 r
= mount_nofollow_verbose(LOG_DEBUG
, d
, dn
, NULL
, MS_BIND
, NULL
);
873 bn
= path_startswith(d
, "/dev/");
877 /* Create symlinks like /dev/char/1:9 → ../urandom */
878 if (asprintf(&sl
, "%s/dev/%s/" DEVNUM_FORMAT_STR
,
880 S_ISCHR(st
.st_mode
) ? "char" : "block",
881 DEVNUM_FORMAT_VAL(st
.st_rdev
)) < 0)
884 (void) mkdir_parents(sl
, 0755);
886 t
= strjoina("../", bn
);
887 if (symlink(t
, sl
) < 0)
888 log_debug_errno(errno
, "Failed to symlink '%s' to '%s', ignoring: %m", t
, sl
);
893 static int mount_private_dev(MountEntry
*m
) {
894 static const char devnodes
[] =
902 char temporary_mount
[] = "/tmp/namespace-dev-XXXXXX";
903 const char *dev
= NULL
, *devpts
= NULL
, *devshm
= NULL
, *devhugepages
= NULL
, *devmqueue
= NULL
, *devlog
= NULL
, *devptmx
= NULL
;
904 bool can_mknod
= true;
909 if (!mkdtemp(temporary_mount
))
910 return log_debug_errno(errno
, "Failed to create temporary directory '%s': %m", temporary_mount
);
912 dev
= strjoina(temporary_mount
, "/dev");
913 (void) mkdir(dev
, 0755);
914 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", dev
, "tmpfs", DEV_MOUNT_OPTIONS
, "mode=0755" TMPFS_LIMITS_PRIVATE_DEV
);
918 r
= label_fix_full(AT_FDCWD
, dev
, "/dev", 0);
920 log_debug_errno(r
, "Failed to fix label of '%s' as /dev: %m", dev
);
924 devpts
= strjoina(temporary_mount
, "/dev/pts");
925 (void) mkdir(devpts
, 0755);
926 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev/pts", devpts
, NULL
, MS_BIND
, NULL
);
930 /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx.
931 * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible.
932 * Thus, in that case make a clone.
933 * In nspawn and other containers it will be a symlink, in that case make it a symlink. */
934 r
= is_symlink("/dev/ptmx");
936 log_debug_errno(r
, "Failed to detect whether /dev/ptmx is a symlink or not: %m");
939 devptmx
= strjoina(temporary_mount
, "/dev/ptmx");
940 if (symlink("pts/ptmx", devptmx
) < 0) {
941 r
= log_debug_errno(errno
, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx
);
945 r
= clone_device_node("/dev/ptmx", temporary_mount
, &can_mknod
);
950 devshm
= strjoina(temporary_mount
, "/dev/shm");
951 (void) mkdir(devshm
, 0755);
952 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev/shm", devshm
, NULL
, MS_BIND
, NULL
);
956 devmqueue
= strjoina(temporary_mount
, "/dev/mqueue");
957 (void) mkdir(devmqueue
, 0755);
958 (void) mount_nofollow_verbose(LOG_DEBUG
, "/dev/mqueue", devmqueue
, NULL
, MS_BIND
, NULL
);
960 devhugepages
= strjoina(temporary_mount
, "/dev/hugepages");
961 (void) mkdir(devhugepages
, 0755);
962 (void) mount_nofollow_verbose(LOG_DEBUG
, "/dev/hugepages", devhugepages
, NULL
, MS_BIND
, NULL
);
964 devlog
= strjoina(temporary_mount
, "/dev/log");
965 if (symlink("/run/systemd/journal/dev-log", devlog
) < 0)
966 log_debug_errno(errno
, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog
);
968 NULSTR_FOREACH(d
, devnodes
) {
969 r
= clone_device_node(d
, temporary_mount
, &can_mknod
);
970 /* ENXIO means the *source* is not a device file, skip creation in that case */
971 if (r
< 0 && r
!= -ENXIO
)
975 r
= dev_setup(temporary_mount
, UID_INVALID
, GID_INVALID
);
977 log_debug_errno(r
, "Failed to set up basic device tree at '%s', ignoring: %m", temporary_mount
);
979 /* Create the /dev directory if missing. It is more likely to be missing when the service is started
980 * with RootDirectory. This is consistent with mount units creating the mount points when missing. */
981 (void) mkdir_p_label(mount_entry_path(m
), 0755);
983 /* Unmount everything in old /dev */
984 r
= umount_recursive(mount_entry_path(m
), 0);
986 log_debug_errno(r
, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m
));
988 r
= mount_nofollow_verbose(LOG_DEBUG
, dev
, mount_entry_path(m
), NULL
, MS_MOVE
, NULL
);
993 (void) rmdir(temporary_mount
);
999 (void) umount_verbose(LOG_DEBUG
, devpts
, UMOUNT_NOFOLLOW
);
1002 (void) umount_verbose(LOG_DEBUG
, devshm
, UMOUNT_NOFOLLOW
);
1005 (void) umount_verbose(LOG_DEBUG
, devhugepages
, UMOUNT_NOFOLLOW
);
1008 (void) umount_verbose(LOG_DEBUG
, devmqueue
, UMOUNT_NOFOLLOW
);
1010 (void) umount_verbose(LOG_DEBUG
, dev
, UMOUNT_NOFOLLOW
);
1012 (void) rmdir(temporary_mount
);
1017 static int mount_bind_dev(const MountEntry
*m
) {
1022 /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the
1023 * service's /dev. This is only used when RootDirectory= is set. */
1025 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1027 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
1029 return log_debug_errno(r
, "Unable to determine whether /dev is already mounted: %m");
1030 if (r
> 0) /* make this a NOP if /dev is already a mount point */
1033 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
);
1040 static int mount_private_sysfs(const MountEntry
*m
) {
1041 const char *p
= mount_entry_path(ASSERT_PTR(m
));
1044 (void) mkdir_p_label(p
, 0755);
1046 r
= remount_sysfs(p
);
1047 if (r
< 0 && (ERRNO_IS_PRIVILEGE(r
) || ERRNO_IS_NOT_SUPPORTED(r
))) {
1048 /* Running with an unprivileged user (PrivateUsers=yes), or the kernel seems old. Falling
1049 * back to bind mount the host's version so that we get all child mounts of it, too. */
1051 log_debug_errno(r
, "Failed to remount sysfs on %s, falling back to bind mount: %m", p
);
1053 (void) umount_recursive(p
, 0);
1055 r
= mount_nofollow_verbose(LOG_DEBUG
, "/sys", p
, NULL
, MS_BIND
|MS_REC
, NULL
);
1058 return log_debug_errno(r
, "Failed to remount sysfs on %s: %m", p
);
1063 static int mount_bind_sysfs(const MountEntry
*m
) {
1068 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1070 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
1072 return log_debug_errno(r
, "Unable to determine whether /sys is already mounted: %m");
1073 if (r
> 0) /* make this a NOP if /sys is already a mount point */
1076 /* Bind mount the host's version so that we get all child mounts of it, too. */
1077 r
= mount_nofollow_verbose(LOG_DEBUG
, "/sys", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
);
1084 static int mount_procfs(const MountEntry
*m
, const NamespaceInfo
*ns_info
) {
1085 _cleanup_free_
char *opts
= NULL
;
1086 const char *entry_path
;
1092 if (ns_info
->protect_proc
!= PROTECT_PROC_DEFAULT
||
1093 ns_info
->proc_subset
!= PROC_SUBSET_ALL
) {
1095 /* Starting with kernel 5.8 procfs' hidepid= logic is truly per-instance (previously it
1096 * pretended to be per-instance but actually was per-namespace), hence let's make use of it
1097 * if requested. To make sure this logic succeeds only on kernels where hidepid= is
1098 * per-instance, we'll exclusively use the textual value for hidepid=, since support was
1099 * added in the same commit: if it's supported it is thus also per-instance. */
1101 const char *hpv
= ns_info
->protect_proc
== PROTECT_PROC_DEFAULT
?
1103 protect_proc_to_string(ns_info
->protect_proc
);
1105 /* hidepid= support was added in 5.8, so we can use fsconfig()/fsopen() (which were added in
1106 * 5.2) to check if hidepid= is supported. This avoids a noisy dmesg log by the kernel when
1107 * trying to use hidepid= on systems where it isn't supported. The same applies for subset=.
1108 * fsopen()/fsconfig() was also backported on some distros which allows us to detect
1109 * hidepid=/subset= support in even more scenarios. */
1111 if (mount_option_supported("proc", "hidepid", hpv
) != 0) {
1112 opts
= strjoin("hidepid=", hpv
);
1117 if (ns_info
->proc_subset
== PROC_SUBSET_PID
&&
1118 mount_option_supported("proc", "subset", "pid") != 0)
1119 if (!strextend_with_separator(&opts
, ",", "subset=pid"))
1123 entry_path
= mount_entry_path(m
);
1124 (void) mkdir_p_label(entry_path
, 0755);
1126 /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in
1127 * one. i.e we don't reuse existing mounts here under any condition, we want a new instance owned by
1128 * our user namespace and with our hidepid= settings applied. Hence, let's get rid of everything
1129 * mounted on /proc/ first. */
1131 n
= umount_recursive(entry_path
, 0);
1133 r
= mount_nofollow_verbose(LOG_DEBUG
, "proc", entry_path
, "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, opts
);
1134 if (r
== -EINVAL
&& opts
)
1135 /* If this failed with EINVAL then this likely means the textual hidepid= stuff is
1136 * not supported by the kernel, and thus the per-instance hidepid= neither, which
1137 * means we really don't want to use it, since it would affect our host's /proc
1138 * mount. Hence let's gracefully fallback to a classic, unrestricted version. */
1139 r
= mount_nofollow_verbose(LOG_DEBUG
, "proc", entry_path
, "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
1141 /* When we do not have enough privileges to mount /proc, fallback to use existing /proc. */
1144 /* /proc or some of sub-mounts are umounted in the above. Refuse incomplete tree.
1145 * Propagate the original error code returned by mount() in the above. */
1148 r
= path_is_mount_point(entry_path
, NULL
, 0);
1150 return log_debug_errno(r
, "Unable to determine whether /proc is already mounted: %m");
1152 /* We lack permissions to mount a new instance of /proc, and it is not already
1153 * mounted. But we can access the host's, so as a final fallback bind-mount it to
1154 * the destination, as most likely we are inside a user manager in an unprivileged
1155 * user namespace. */
1156 r
= mount_nofollow_verbose(LOG_DEBUG
, "/proc", entry_path
, NULL
, MS_BIND
|MS_REC
, NULL
);
1163 /* We mounted a new instance now. Let's bind mount the children over now. This matters for
1164 * nspawn where a bunch of files are overmounted, in particular the boot id */
1165 (void) bind_mount_submounts("/proc", entry_path
);
1170 static int mount_tmpfs(const MountEntry
*m
) {
1171 const char *entry_path
, *inner_path
;
1176 entry_path
= mount_entry_path(m
);
1177 inner_path
= mount_entry_unprefixed_path(m
);
1179 /* First, get rid of everything that is below if there is anything. Then, overmount with our new
1182 (void) mkdir_p_label(entry_path
, 0755);
1183 (void) umount_recursive(entry_path
, 0);
1185 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", entry_path
, "tmpfs", m
->flags
, mount_entry_options(m
));
1189 r
= label_fix_full(AT_FDCWD
, entry_path
, inner_path
, 0);
1191 return log_debug_errno(r
, "Failed to fix label of '%s' as '%s': %m", entry_path
, inner_path
);
1196 static int mount_run(const MountEntry
*m
) {
1201 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
1202 if (r
< 0 && r
!= -ENOENT
)
1203 return log_debug_errno(r
, "Unable to determine whether /run is already mounted: %m");
1204 if (r
> 0) /* make this a NOP if /run is already a mount point */
1207 return mount_tmpfs(m
);
1210 static int mount_mqueuefs(const MountEntry
*m
) {
1212 const char *entry_path
;
1216 entry_path
= mount_entry_path(m
);
1218 (void) mkdir_p_label(entry_path
, 0755);
1219 (void) umount_recursive(entry_path
, 0);
1221 r
= mount_nofollow_verbose(LOG_DEBUG
, "mqueue", entry_path
, "mqueue", m
->flags
, mount_entry_options(m
));
1228 static int mount_image(
1229 const MountEntry
*m
,
1230 const char *root_directory
,
1231 const ImagePolicy
*image_policy
) {
1233 _cleanup_free_
char *host_os_release_id
= NULL
, *host_os_release_version_id
= NULL
,
1234 *host_os_release_sysext_level
= NULL
;
1239 if (m
->mode
== EXTENSION_IMAGES
) {
1240 r
= parse_os_release(
1241 empty_to_root(root_directory
),
1242 "ID", &host_os_release_id
,
1243 "VERSION_ID", &host_os_release_version_id
,
1244 "SYSEXT_LEVEL", &host_os_release_sysext_level
,
1247 return log_debug_errno(r
, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory
));
1248 if (isempty(host_os_release_id
))
1249 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
));
1252 r
= verity_dissect_and_mount(
1254 mount_entry_source(m
),
1255 mount_entry_path(m
),
1259 host_os_release_version_id
,
1260 host_os_release_sysext_level
,
1262 if (r
== -ENOENT
&& m
->ignore
)
1264 if (r
== -ESTALE
&& host_os_release_id
)
1265 return log_error_errno(r
,
1266 "Failed to mount image %s, extension-release metadata does not match the lower layer's: ID=%s%s%s%s%s",
1267 mount_entry_source(m
),
1269 host_os_release_version_id
? " VERSION_ID=" : "",
1270 strempty(host_os_release_version_id
),
1271 host_os_release_sysext_level
? " SYSEXT_LEVEL=" : "",
1272 strempty(host_os_release_sysext_level
));
1274 return log_debug_errno(r
, "Failed to mount image %s on %s: %m", mount_entry_source(m
), mount_entry_path(m
));
1279 static int mount_overlay(const MountEntry
*m
) {
1280 const char *options
;
1285 options
= strjoina("lowerdir=", mount_entry_options(m
));
1287 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1289 r
= mount_nofollow_verbose(LOG_DEBUG
, "overlay", mount_entry_path(m
), "overlay", MS_RDONLY
, options
);
1290 if (r
== -ENOENT
&& m
->ignore
)
1298 static int follow_symlink(
1299 const char *root_directory
,
1302 _cleanup_free_
char *target
= NULL
;
1305 /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we
1306 * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at
1307 * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the
1308 * end and already have a fully normalized name. */
1310 r
= chase(mount_entry_path(m
), root_directory
, CHASE_STEP
|CHASE_NONEXISTENT
, &target
, NULL
);
1312 return log_debug_errno(r
, "Failed to chase symlinks '%s': %m", mount_entry_path(m
));
1313 if (r
> 0) /* Reached the end, nothing more to resolve */
1316 if (m
->n_followed
>= CHASE_MAX
) /* put a boundary on things */
1317 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
1318 "Symlink loop on '%s'.",
1319 mount_entry_path(m
));
1321 log_debug("Followed mount entry path symlink %s %s %s.",
1322 mount_entry_path(m
), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT
), target
);
1324 mount_entry_consume_prefix(m
, TAKE_PTR(target
));
1331 static int apply_one_mount(
1332 const char *root_directory
,
1334 const ImagePolicy
*mount_image_policy
,
1335 const ImagePolicy
*extension_image_policy
,
1336 const NamespaceInfo
*ns_info
) {
1338 _cleanup_free_
char *inaccessible
= NULL
;
1339 bool rbind
= true, make
= false;
1346 log_debug("Applying namespace mount on %s", mount_entry_path(m
));
1350 case INACCESSIBLE
: {
1351 _cleanup_free_
char *tmp
= NULL
;
1352 const char *runtime_dir
;
1355 /* First, get rid of everything that is below if there
1356 * is anything... Then, overmount it with an
1357 * inaccessible path. */
1358 (void) umount_recursive(mount_entry_path(m
), 0);
1360 if (lstat(mount_entry_path(m
), &target
) < 0) {
1361 if (errno
== ENOENT
&& m
->ignore
)
1364 return log_debug_errno(errno
, "Failed to lstat() %s to determine what to mount over it: %m",
1365 mount_entry_path(m
));
1369 runtime_dir
= "/run";
1371 if (asprintf(&tmp
, "/run/user/" UID_FMT
, geteuid()) < 0)
1377 r
= mode_to_inaccessible_node(runtime_dir
, target
.st_mode
, &inaccessible
);
1379 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
1380 "File type not supported for inaccessible mounts. Note that symlinks are not allowed");
1381 what
= inaccessible
;
1387 case READWRITE_IMPLICIT
:
1390 r
= path_is_mount_point(mount_entry_path(m
), root_directory
, 0);
1391 if (r
== -ENOENT
&& m
->ignore
)
1394 return log_debug_errno(r
, "Failed to determine whether %s is already a mount point: %m",
1395 mount_entry_path(m
));
1396 if (r
> 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY
1397 * and MS_NOEXEC bits for the mount point if needed. */
1399 /* This isn't a mount point yet, let's make it one. */
1400 what
= mount_entry_path(m
);
1403 case EXTENSION_DIRECTORIES
: {
1404 _cleanup_free_
char *host_os_release_id
= NULL
, *host_os_release_version_id
= NULL
,
1405 *host_os_release_sysext_level
= NULL
, *extension_name
= NULL
;
1406 _cleanup_strv_free_
char **extension_release
= NULL
;
1408 r
= path_extract_filename(mount_entry_source(m
), &extension_name
);
1410 return log_debug_errno(r
, "Failed to extract extension name from %s: %m", mount_entry_source(m
));
1412 r
= parse_os_release(
1413 empty_to_root(root_directory
),
1414 "ID", &host_os_release_id
,
1415 "VERSION_ID", &host_os_release_version_id
,
1416 "SYSEXT_LEVEL", &host_os_release_sysext_level
,
1419 return log_debug_errno(r
, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory
));
1420 if (isempty(host_os_release_id
))
1421 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
));
1423 r
= load_extension_release_pairs(mount_entry_source(m
), IMAGE_SYSEXT
, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1424 if (r
== -ENOENT
&& m
->ignore
)
1427 return log_debug_errno(r
, "Failed to parse directory %s extension-release metadata: %m", extension_name
);
1429 r
= extension_release_validate(
1432 host_os_release_version_id
,
1433 host_os_release_sysext_level
,
1434 /* host_sysext_scope */ NULL
, /* Leave empty, we need to accept both system and portable */
1438 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE
), "Directory %s extension-release metadata does not match the root's", extension_name
);
1440 return log_debug_errno(r
, "Failed to compare directory %s extension-release metadata with the root's os-release: %m", extension_name
);
1449 case BIND_MOUNT_RECURSIVE
: {
1450 _cleanup_free_
char *chased
= NULL
;
1452 /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note
1453 * that bind mount source paths are always relative to the host root, hence we pass NULL as
1454 * root directory to chase() here. */
1456 r
= chase(mount_entry_source(m
), NULL
, CHASE_TRAIL_SLASH
, &chased
, NULL
);
1457 if (r
== -ENOENT
&& m
->ignore
) {
1458 log_debug_errno(r
, "Path %s does not exist, ignoring.", mount_entry_source(m
));
1462 return log_debug_errno(r
, "Failed to follow symlinks on %s: %m", mount_entry_source(m
));
1464 log_debug("Followed source symlinks %s %s %s.",
1465 mount_entry_source(m
), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT
), chased
);
1467 free_and_replace(m
->source_malloc
, chased
);
1469 what
= mount_entry_source(m
);
1476 return mount_tmpfs(m
);
1479 case PRIVATE_TMP_READONLY
:
1480 what
= mount_entry_source(m
);
1485 return mount_private_dev(m
);
1488 return mount_bind_dev(m
);
1491 return mount_private_sysfs(m
);
1494 return mount_bind_sysfs(m
);
1497 return mount_procfs(m
, ns_info
);
1500 return mount_run(m
);
1503 return mount_mqueuefs(m
);
1506 return mount_image(m
, NULL
, mount_image_policy
);
1508 case EXTENSION_IMAGES
:
1509 return mount_image(m
, root_directory
, extension_image_policy
);
1512 return mount_overlay(m
);
1515 assert_not_reached();
1520 r
= mount_nofollow_verbose(LOG_DEBUG
, what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
);
1522 bool try_again
= false;
1524 if (r
== -ENOENT
&& make
) {
1527 /* Hmm, either the source or the destination are missing. Let's see if we can create
1528 the destination, then try again. */
1530 (void) mkdir_parents(mount_entry_path(m
), 0755);
1532 q
= make_mount_point_inode_from_path(what
, mount_entry_path(m
), 0755);
1533 if (q
< 0 && q
!= -EEXIST
)
1534 log_error_errno(q
, "Failed to create destination mount point node '%s': %m",
1535 mount_entry_path(m
));
1541 r
= mount_nofollow_verbose(LOG_DEBUG
, what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
);
1543 return log_error_errno(r
, "Failed to mount %s to %s: %m", what
, mount_entry_path(m
));
1546 log_debug("Successfully mounted %s to %s", what
, mount_entry_path(m
));
1550 static int make_read_only(const MountEntry
*m
, char **deny_list
, FILE *proc_self_mountinfo
) {
1551 unsigned long new_flags
= 0, flags_mask
= 0;
1556 assert(proc_self_mountinfo
);
1558 if (mount_entry_read_only(m
) || m
->mode
== PRIVATE_DEV
) {
1559 new_flags
|= MS_RDONLY
;
1560 flags_mask
|= MS_RDONLY
;
1564 new_flags
|= MS_NOSUID
;
1565 flags_mask
|= MS_NOSUID
;
1568 if (flags_mask
== 0) /* No Change? */
1571 /* We generally apply these changes recursively, except for /dev, and the cases we know there's
1572 * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the
1573 * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace
1574 * and running Linux <= 4.17. */
1576 mount_entry_read_only(m
) &&
1577 !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1579 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, deny_list
, proc_self_mountinfo
);
1581 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, proc_self_mountinfo
);
1583 /* Note that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked
1584 * read-only already stays this way. This improves compatibility with container managers, where we
1585 * won't attempt to undo read-only mounts already applied. */
1587 if (r
== -ENOENT
&& m
->ignore
)
1590 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1591 submounts
? " and its submounts" : "");
1595 static int make_noexec(const MountEntry
*m
, char **deny_list
, FILE *proc_self_mountinfo
) {
1596 unsigned long new_flags
= 0, flags_mask
= 0;
1601 assert(proc_self_mountinfo
);
1603 if (mount_entry_noexec(m
)) {
1604 new_flags
|= MS_NOEXEC
;
1605 flags_mask
|= MS_NOEXEC
;
1606 } else if (mount_entry_exec(m
)) {
1607 new_flags
&= ~MS_NOEXEC
;
1608 flags_mask
|= MS_NOEXEC
;
1611 if (flags_mask
== 0) /* No Change? */
1614 submounts
= !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1617 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, deny_list
, proc_self_mountinfo
);
1619 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, proc_self_mountinfo
);
1621 if (r
== -ENOENT
&& m
->ignore
)
1624 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1625 submounts
? " and its submounts" : "");
1629 static int make_nosuid(const MountEntry
*m
, FILE *proc_self_mountinfo
) {
1634 assert(proc_self_mountinfo
);
1636 submounts
= !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1639 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), MS_NOSUID
, MS_NOSUID
, NULL
, proc_self_mountinfo
);
1641 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), MS_NOSUID
, MS_NOSUID
, proc_self_mountinfo
);
1642 if (r
== -ENOENT
&& m
->ignore
)
1645 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1646 submounts
? " and its submounts" : "");
1650 static bool namespace_info_mount_apivfs(const NamespaceInfo
*ns_info
) {
1654 * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
1655 * since to protect the API VFS mounts, they need to be around in the
1659 return ns_info
->mount_apivfs
||
1660 ns_info
->protect_control_groups
||
1661 ns_info
->protect_kernel_tunables
||
1662 ns_info
->protect_proc
!= PROTECT_PROC_DEFAULT
||
1663 ns_info
->proc_subset
!= PROC_SUBSET_ALL
;
1666 static size_t namespace_calculate_mounts(
1667 const NamespaceInfo
*ns_info
,
1668 char** read_write_paths
,
1669 char** read_only_paths
,
1670 char** inaccessible_paths
,
1672 char** no_exec_paths
,
1673 char** empty_directories
,
1674 size_t n_bind_mounts
,
1675 size_t n_temporary_filesystems
,
1676 size_t n_mount_images
,
1677 size_t n_extension_images
,
1678 size_t n_extension_directories
,
1679 size_t n_hierarchies
,
1680 const char* tmp_dir
,
1681 const char* var_tmp_dir
,
1682 const char *creds_path
,
1683 const char* log_namespace
,
1684 bool setup_propagate
,
1685 const char* notify_socket
,
1686 const char* host_os_release
) {
1688 size_t protect_home_cnt
;
1689 size_t protect_system_cnt
=
1690 (ns_info
->protect_system
== PROTECT_SYSTEM_STRICT
?
1691 ELEMENTSOF(protect_system_strict_table
) :
1692 ((ns_info
->protect_system
== PROTECT_SYSTEM_FULL
) ?
1693 ELEMENTSOF(protect_system_full_table
) :
1694 ((ns_info
->protect_system
== PROTECT_SYSTEM_YES
) ?
1695 ELEMENTSOF(protect_system_yes_table
) : 0)));
1698 (ns_info
->protect_home
== PROTECT_HOME_YES
?
1699 ELEMENTSOF(protect_home_yes_table
) :
1700 ((ns_info
->protect_home
== PROTECT_HOME_READ_ONLY
) ?
1701 ELEMENTSOF(protect_home_read_only_table
) :
1702 ((ns_info
->protect_home
== PROTECT_HOME_TMPFS
) ?
1703 ELEMENTSOF(protect_home_tmpfs_table
) : 0)));
1705 return !!tmp_dir
+ !!var_tmp_dir
+
1706 strv_length(read_write_paths
) +
1707 strv_length(read_only_paths
) +
1708 strv_length(inaccessible_paths
) +
1709 strv_length(exec_paths
) +
1710 strv_length(no_exec_paths
) +
1711 strv_length(empty_directories
) +
1714 (n_extension_images
> 0 || n_extension_directories
> 0 ? /* Mount each image and directory plus an overlay per hierarchy */
1715 n_hierarchies
+ n_extension_images
+ n_extension_directories
: 0) +
1716 n_temporary_filesystems
+
1717 ns_info
->private_dev
+
1718 (ns_info
->protect_kernel_tunables
?
1719 ELEMENTSOF(protect_kernel_tunables_proc_table
) + ELEMENTSOF(protect_kernel_tunables_sys_table
) : 0) +
1720 (ns_info
->protect_kernel_modules
? ELEMENTSOF(protect_kernel_modules_table
) : 0) +
1721 (ns_info
->protect_kernel_logs
?
1722 ELEMENTSOF(protect_kernel_logs_proc_table
) + ELEMENTSOF(protect_kernel_logs_dev_table
) : 0) +
1723 (ns_info
->protect_control_groups
? 1 : 0) +
1724 protect_home_cnt
+ protect_system_cnt
+
1725 (ns_info
->protect_hostname
? 2 : 0) +
1726 (namespace_info_mount_apivfs(ns_info
) ? ELEMENTSOF(apivfs_table
) : 0) +
1727 (creds_path
? 2 : 1) +
1729 setup_propagate
+ /* /run/systemd/incoming */
1732 ns_info
->private_network
+ /* /sys */
1733 ns_info
->private_ipc
; /* /dev/mqueue */
1736 /* Walk all mount entries and dropping any unused mounts. This affects all
1738 * - that are implicitly protected by a path that has been rendered inaccessible
1739 * - whose immediate parent requests the same protection mode as the mount itself
1740 * - that are outside of the relevant root directory
1741 * - which are duplicates
1743 static void drop_unused_mounts(const char *root_directory
, MountEntry
*mounts
, size_t *n_mounts
) {
1744 assert(root_directory
);
1746 assert(mounts
|| *n_mounts
== 0);
1748 typesafe_qsort(mounts
, *n_mounts
, mount_path_compare
);
1750 drop_duplicates(mounts
, n_mounts
);
1751 drop_outside_root(root_directory
, mounts
, n_mounts
);
1752 drop_inaccessible(mounts
, n_mounts
);
1753 drop_nop(mounts
, n_mounts
);
1756 static int create_symlinks_from_tuples(const char *root
, char **strv_symlinks
) {
1759 STRV_FOREACH_PAIR(src
, dst
, strv_symlinks
) {
1760 _cleanup_free_
char *src_abs
= NULL
, *dst_abs
= NULL
;
1762 src_abs
= path_join(root
, *src
);
1763 dst_abs
= path_join(root
, *dst
);
1764 if (!src_abs
|| !dst_abs
)
1767 r
= mkdir_parents_label(dst_abs
, 0755);
1771 r
= symlink_idempotent(src_abs
, dst_abs
, true);
1779 static int apply_mounts(
1781 const ImagePolicy
*mount_image_policy
,
1782 const ImagePolicy
*extension_image_policy
,
1783 const NamespaceInfo
*ns_info
,
1786 char **exec_dir_symlinks
,
1787 char **error_path
) {
1789 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
1790 _cleanup_free_
char **deny_list
= NULL
;
1793 if (n_mounts
== 0) /* Shortcut: nothing to do */
1800 /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of
1801 * /proc. For example, this is the case with the option: 'InaccessiblePaths=/proc'. */
1802 proc_self_mountinfo
= fopen("/proc/self/mountinfo", "re");
1803 if (!proc_self_mountinfo
) {
1807 *error_path
= strdup("/proc/self/mountinfo");
1809 return log_debug_errno(r
, "Failed to open /proc/self/mountinfo: %m");
1812 /* First round, establish all mounts we need */
1816 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1821 /* ExtensionImages/Directories are first opened in the propagate directory, not in the root_directory */
1822 r
= follow_symlink(!IN_SET(m
->mode
, EXTENSION_IMAGES
, EXTENSION_DIRECTORIES
) ? root
: NULL
, m
);
1824 if (error_path
&& mount_entry_path(m
))
1825 *error_path
= strdup(mount_entry_path(m
));
1829 /* We hit a symlinked mount point. The entry got rewritten and might
1830 * point to a very different place now. Let's normalize the changed
1831 * list, and start from the beginning. After all to mount the entry
1832 * at the new location we might need some other mounts first */
1837 r
= apply_one_mount(root
, m
, mount_image_policy
, extension_image_policy
, ns_info
);
1839 if (error_path
&& mount_entry_path(m
))
1840 *error_path
= strdup(mount_entry_path(m
));
1850 drop_unused_mounts(root
, mounts
, n_mounts
);
1853 /* Now that all filesystems have been set up, but before the
1854 * read-only switches are flipped, create the exec dirs symlinks.
1855 * Note that when /var/lib is not empty/tmpfs, these symlinks will already
1856 * exist, which means this will be a no-op. */
1857 r
= create_symlinks_from_tuples(root
, exec_dir_symlinks
);
1859 return log_debug_errno(r
, "Failed to set up ExecDirectories symlinks inside mount namespace: %m");
1861 /* Create a deny list we can pass to bind_mount_recursive() */
1862 deny_list
= new(char*, (*n_mounts
)+1);
1865 for (size_t j
= 0; j
< *n_mounts
; j
++)
1866 deny_list
[j
] = (char*) mount_entry_path(mounts
+j
);
1867 deny_list
[*n_mounts
] = NULL
;
1869 /* Second round, flip the ro bits if necessary. */
1870 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1871 r
= make_read_only(m
, deny_list
, proc_self_mountinfo
);
1873 if (error_path
&& mount_entry_path(m
))
1874 *error_path
= strdup(mount_entry_path(m
));
1879 /* Third round, flip the noexec bits with a simplified deny list. */
1880 for (size_t j
= 0; j
< *n_mounts
; j
++)
1881 if (IN_SET((mounts
+j
)->mode
, EXEC
, NOEXEC
))
1882 deny_list
[j
] = (char*) mount_entry_path(mounts
+j
);
1883 deny_list
[*n_mounts
] = NULL
;
1885 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1886 r
= make_noexec(m
, deny_list
, proc_self_mountinfo
);
1888 if (error_path
&& mount_entry_path(m
))
1889 *error_path
= strdup(mount_entry_path(m
));
1894 /* Fourth round, flip the nosuid bits without a deny list. */
1895 if (ns_info
->mount_nosuid
)
1896 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1897 r
= make_nosuid(m
, proc_self_mountinfo
);
1899 if (error_path
&& mount_entry_path(m
))
1900 *error_path
= strdup(mount_entry_path(m
));
1908 static bool root_read_only(
1909 char **read_only_paths
,
1910 ProtectSystem protect_system
) {
1912 /* Determine whether the root directory is going to be read-only given the configured settings. */
1914 if (protect_system
== PROTECT_SYSTEM_STRICT
)
1917 if (prefixed_path_strv_contains(read_only_paths
, "/"))
1923 static bool home_read_only(
1924 char** read_only_paths
,
1925 char** inaccessible_paths
,
1926 char** empty_directories
,
1927 const BindMount
*bind_mounts
,
1928 size_t n_bind_mounts
,
1929 const TemporaryFileSystem
*temporary_filesystems
,
1930 size_t n_temporary_filesystems
,
1931 ProtectHome protect_home
) {
1933 /* Determine whether the /home directory is going to be read-only given the configured settings. Yes,
1934 * this is a bit sloppy, since we don't bother checking for cases where / is affected by multiple
1937 if (protect_home
!= PROTECT_HOME_NO
)
1940 if (prefixed_path_strv_contains(read_only_paths
, "/home") ||
1941 prefixed_path_strv_contains(inaccessible_paths
, "/home") ||
1942 prefixed_path_strv_contains(empty_directories
, "/home"))
1945 for (size_t i
= 0; i
< n_temporary_filesystems
; i
++)
1946 if (path_equal(temporary_filesystems
[i
].path
, "/home"))
1949 /* If /home is overmounted with some dir from the host it's not writable. */
1950 for (size_t i
= 0; i
< n_bind_mounts
; i
++)
1951 if (path_equal(bind_mounts
[i
].destination
, "/home"))
1957 int setup_namespace(
1958 const char* root_directory
,
1959 const char* root_image
,
1960 const MountOptions
*root_image_mount_options
,
1961 const ImagePolicy
*root_image_policy
,
1962 const NamespaceInfo
*ns_info
,
1963 char** read_write_paths
,
1964 char** read_only_paths
,
1965 char** inaccessible_paths
,
1967 char** no_exec_paths
,
1968 char** empty_directories
,
1969 char** exec_dir_symlinks
,
1970 const BindMount
*bind_mounts
,
1971 size_t n_bind_mounts
,
1972 const TemporaryFileSystem
*temporary_filesystems
,
1973 size_t n_temporary_filesystems
,
1974 const MountImage
*mount_images
,
1975 size_t n_mount_images
,
1976 const ImagePolicy
*mount_image_policy
,
1977 const char* tmp_dir
,
1978 const char* var_tmp_dir
,
1979 const char *creds_path
,
1980 const char *log_namespace
,
1981 unsigned long mount_propagation_flag
,
1982 VeritySettings
*verity
,
1983 const MountImage
*extension_images
,
1984 size_t n_extension_images
,
1985 const ImagePolicy
*extension_image_policy
,
1986 char **extension_directories
,
1987 const char *propagate_dir
,
1988 const char *incoming_dir
,
1989 const char *extension_dir
,
1990 const char *notify_socket
,
1991 const char *host_os_release
,
1992 char **error_path
) {
1994 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
1995 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
1996 _cleanup_strv_free_
char **hierarchies
= NULL
;
1997 MountEntry
*m
= NULL
, *mounts
= NULL
;
1998 bool require_prefix
= false, setup_propagate
= false;
2000 DissectImageFlags dissect_image_flags
=
2001 DISSECT_IMAGE_GENERIC_ROOT
|
2002 DISSECT_IMAGE_REQUIRE_ROOT
|
2003 DISSECT_IMAGE_DISCARD_ON_LOOP
|
2004 DISSECT_IMAGE_RELAX_VAR_CHECK
|
2005 DISSECT_IMAGE_FSCK
|
2006 DISSECT_IMAGE_USR_NO_ROOT
|
2007 DISSECT_IMAGE_GROWFS
|
2008 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
2009 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
2015 /* Make sure that all mknod(), mkdir() calls we do are unaffected by the umask, and the access modes
2016 * we configure take effect */
2017 BLOCK_WITH_UMASK(0000);
2019 if (!isempty(propagate_dir
) && !isempty(incoming_dir
))
2020 setup_propagate
= true;
2022 if (mount_propagation_flag
== 0)
2023 mount_propagation_flag
= MS_SHARED
;
2026 /* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */
2027 if (root_read_only(read_only_paths
,
2028 ns_info
->protect_system
) &&
2029 home_read_only(read_only_paths
, inaccessible_paths
, empty_directories
,
2030 bind_mounts
, n_bind_mounts
, temporary_filesystems
, n_temporary_filesystems
,
2031 ns_info
->protect_home
) &&
2032 strv_isempty(read_write_paths
))
2033 dissect_image_flags
|= DISSECT_IMAGE_READ_ONLY
;
2035 SET_FLAG(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
, verity
&& verity
->data_path
);
2037 r
= loop_device_make_by_path(
2039 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_DEVICE_READ_ONLY
) ? O_RDONLY
: -1 /* < 0 means writable if possible, read-only as fallback */,
2040 /* sector_size= */ UINT32_MAX
,
2041 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
2045 return log_debug_errno(r
, "Failed to create loop device for root image: %m");
2047 r
= dissect_loop_device(
2050 root_image_mount_options
,
2052 dissect_image_flags
,
2055 return log_debug_errno(r
, "Failed to dissect image: %m");
2057 r
= dissected_image_load_verity_sig_partition(
2064 r
= dissected_image_decrypt(
2068 dissect_image_flags
);
2070 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
2074 root
= root_directory
;
2076 /* /run/systemd should have been created by PID 1 early on already, but in some cases, like
2077 * when running tests (test-execute), it might not have been created yet so let's make sure
2078 * we create it if it doesn't already exist. */
2079 (void) mkdir_p_label("/run/systemd", 0755);
2081 /* Always create the mount namespace in a temporary directory, instead of operating directly
2082 * in the root. The temporary directory prevents any mounts from being potentially obscured
2083 * my other mounts we already applied. We use the same mount point for all images, which is
2084 * safe, since they all live in their own namespaces after all, and hence won't see each
2085 * other. (Note: this directory is also created by PID 1 early on, we create it here for
2086 * similar reasons as /run/systemd/ first.) */
2087 root
= "/run/systemd/mount-rootfs";
2088 (void) mkdir_label(root
, 0555);
2090 require_prefix
= true;
2093 if (n_extension_images
> 0 || !strv_isempty(extension_directories
)) {
2094 r
= parse_env_extension_hierarchies(&hierarchies
, "SYSTEMD_SYSEXT_HIERARCHIES");
2099 n_mounts
= namespace_calculate_mounts(
2108 n_temporary_filesystems
,
2111 strv_length(extension_directories
),
2112 strv_length(hierarchies
),
2113 tmp_dir
, var_tmp_dir
,
2121 m
= mounts
= new0(MountEntry
, n_mounts
);
2125 r
= append_access_mounts(&m
, read_write_paths
, READWRITE
, require_prefix
);
2129 r
= append_access_mounts(&m
, read_only_paths
, READONLY
, require_prefix
);
2133 r
= append_access_mounts(&m
, inaccessible_paths
, INACCESSIBLE
, require_prefix
);
2137 r
= append_access_mounts(&m
, exec_paths
, EXEC
, require_prefix
);
2141 r
= append_access_mounts(&m
, no_exec_paths
, NOEXEC
, require_prefix
);
2145 r
= append_empty_dir_mounts(&m
, empty_directories
);
2149 r
= append_bind_mounts(&m
, bind_mounts
, n_bind_mounts
);
2153 r
= append_tmpfs_mounts(&m
, temporary_filesystems
, n_temporary_filesystems
);
2158 bool ro
= streq(tmp_dir
, RUN_SYSTEMD_EMPTY
);
2160 *(m
++) = (MountEntry
) {
2161 .path_const
= "/tmp",
2162 .mode
= ro
? PRIVATE_TMP_READONLY
: PRIVATE_TMP
,
2163 .source_const
= tmp_dir
,
2168 bool ro
= streq(var_tmp_dir
, RUN_SYSTEMD_EMPTY
);
2170 *(m
++) = (MountEntry
) {
2171 .path_const
= "/var/tmp",
2172 .mode
= ro
? PRIVATE_TMP_READONLY
: PRIVATE_TMP
,
2173 .source_const
= var_tmp_dir
,
2177 r
= append_mount_images(&m
, mount_images
, n_mount_images
);
2181 r
= append_extensions(&m
, root
, extension_dir
, hierarchies
, extension_images
, n_extension_images
, extension_directories
);
2185 if (ns_info
->private_dev
)
2186 *(m
++) = (MountEntry
) {
2187 .path_const
= "/dev",
2188 .mode
= PRIVATE_DEV
,
2189 .flags
= DEV_MOUNT_OPTIONS
,
2192 /* In case /proc is successfully mounted with pid tree subset only (ProcSubset=pid), the
2193 protective mounts to non-pid /proc paths would fail. But the pid only option may have
2194 failed gracefully, so let's try the mounts but it's not fatal if they don't succeed. */
2195 bool ignore_protect_proc
= ns_info
->ignore_protect_paths
|| ns_info
->proc_subset
== PROC_SUBSET_PID
;
2196 if (ns_info
->protect_kernel_tunables
) {
2197 r
= append_static_mounts(&m
,
2198 protect_kernel_tunables_proc_table
,
2199 ELEMENTSOF(protect_kernel_tunables_proc_table
),
2200 ignore_protect_proc
);
2204 r
= append_static_mounts(&m
,
2205 protect_kernel_tunables_sys_table
,
2206 ELEMENTSOF(protect_kernel_tunables_sys_table
),
2207 ns_info
->ignore_protect_paths
);
2212 if (ns_info
->protect_kernel_modules
) {
2213 r
= append_static_mounts(&m
,
2214 protect_kernel_modules_table
,
2215 ELEMENTSOF(protect_kernel_modules_table
),
2216 ns_info
->ignore_protect_paths
);
2221 if (ns_info
->protect_kernel_logs
) {
2222 r
= append_static_mounts(&m
,
2223 protect_kernel_logs_proc_table
,
2224 ELEMENTSOF(protect_kernel_logs_proc_table
),
2225 ignore_protect_proc
);
2229 r
= append_static_mounts(&m
,
2230 protect_kernel_logs_dev_table
,
2231 ELEMENTSOF(protect_kernel_logs_dev_table
),
2232 ns_info
->ignore_protect_paths
);
2237 if (ns_info
->protect_control_groups
)
2238 *(m
++) = (MountEntry
) {
2239 .path_const
= "/sys/fs/cgroup",
2243 r
= append_protect_home(&m
, ns_info
->protect_home
, ns_info
->ignore_protect_paths
);
2247 r
= append_protect_system(&m
, ns_info
->protect_system
, false);
2251 if (namespace_info_mount_apivfs(ns_info
)) {
2252 r
= append_static_mounts(&m
,
2254 ELEMENTSOF(apivfs_table
),
2255 ns_info
->ignore_protect_paths
);
2260 /* Note, if proc is mounted with subset=pid then neither of the
2261 * two paths will exist, i.e. they are implicitly protected by
2262 * the mount option. */
2263 if (ns_info
->protect_hostname
) {
2264 *(m
++) = (MountEntry
) {
2265 .path_const
= "/proc/sys/kernel/hostname",
2267 .ignore
= ignore_protect_proc
,
2269 *(m
++) = (MountEntry
) {
2270 .path_const
= "/proc/sys/kernel/domainname",
2272 .ignore
= ignore_protect_proc
,
2276 if (ns_info
->private_network
)
2277 *(m
++) = (MountEntry
) {
2278 .path_const
= "/sys",
2279 .mode
= PRIVATE_SYSFS
,
2282 if (ns_info
->private_ipc
)
2283 *(m
++) = (MountEntry
) {
2284 .path_const
= "/dev/mqueue",
2286 .flags
= MS_NOSUID
| MS_NODEV
| MS_NOEXEC
| MS_RELATIME
,
2290 /* If our service has a credentials store configured, then bind that one in, but hide
2291 * everything else. */
2293 *(m
++) = (MountEntry
) {
2294 .path_const
= "/run/credentials",
2297 .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
,
2298 .flags
= MS_NODEV
|MS_STRICTATIME
|MS_NOSUID
|MS_NOEXEC
,
2301 *(m
++) = (MountEntry
) {
2302 .path_const
= creds_path
,
2305 .source_const
= creds_path
,
2309 /* If our service has no credentials store configured, then make the whole
2310 * credentials tree inaccessible wholesale. */
2312 *(m
++) = (MountEntry
) {
2313 .path_const
= "/run/credentials",
2314 .mode
= INACCESSIBLE
,
2319 if (log_namespace
) {
2320 _cleanup_free_
char *q
= NULL
;
2322 q
= strjoin("/run/systemd/journal.", log_namespace
);
2328 *(m
++) = (MountEntry
) {
2329 .path_const
= "/run/systemd/journal",
2330 .mode
= BIND_MOUNT_RECURSIVE
,
2332 .source_malloc
= TAKE_PTR(q
),
2336 /* Will be used to add bind mounts at runtime */
2337 if (setup_propagate
)
2338 *(m
++) = (MountEntry
) {
2339 .source_const
= propagate_dir
,
2340 .path_const
= incoming_dir
,
2346 *(m
++) = (MountEntry
) {
2347 .path_const
= notify_socket
,
2348 .source_const
= notify_socket
,
2353 if (host_os_release
)
2354 *(m
++) = (MountEntry
) {
2355 .path_const
= "/run/host/os-release",
2356 .source_const
= host_os_release
,
2359 .ignore
= true, /* Live copy, don't hard-fail if it goes missing */
2362 assert(mounts
+ n_mounts
== m
);
2364 /* Prepend the root directory where that's necessary */
2365 r
= prefix_where_needed(mounts
, n_mounts
, root
);
2369 drop_unused_mounts(root
, mounts
, &n_mounts
);
2372 /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */
2374 if (unshare(CLONE_NEWNS
) < 0) {
2375 r
= log_debug_errno(errno
, "Failed to unshare the mount namespace: %m");
2376 if (ERRNO_IS_PRIVILEGE(r
) ||
2377 ERRNO_IS_NOT_SUPPORTED(r
))
2378 /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter
2379 * in place that doesn't allow us to create namespaces (or a missing cap), then
2380 * propagate a recognizable error back, which the caller can use to detect this case
2381 * (and only this) and optionally continue without namespacing applied. */
2387 /* Create the source directory to allow runtime propagation of mounts */
2388 if (setup_propagate
)
2389 (void) mkdir_p(propagate_dir
, 0600);
2391 if (n_extension_images
> 0 || !strv_isempty(extension_directories
))
2392 /* ExtensionImages/Directories mountpoint directories will be created while parsing the
2393 * mounts to create, so have the parent ready */
2394 (void) mkdir_p(extension_dir
, 0600);
2396 /* Remount / as SLAVE so that nothing now mounted in the namespace
2397 * shows up in the parent */
2398 if (mount(NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
) < 0) {
2399 r
= log_debug_errno(errno
, "Failed to remount '/' as SLAVE: %m");
2404 /* A root image is specified, mount it to the right place */
2405 r
= dissected_image_mount(dissected_image
, root
, UID_INVALID
, UID_INVALID
, dissect_image_flags
);
2407 log_debug_errno(r
, "Failed to mount root image: %m");
2411 /* Now release the block device lock, so that udevd is free to call BLKRRPART on the device
2413 r
= loop_device_flock(loop_device
, LOCK_UN
);
2415 log_debug_errno(r
, "Failed to release lock on loopback block device: %m");
2419 r
= dissected_image_relinquish(dissected_image
);
2421 log_debug_errno(r
, "Failed to relinquish dissected image: %m");
2425 } else if (root_directory
) {
2427 /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
2428 r
= path_is_mount_point(root
, NULL
, AT_SYMLINK_FOLLOW
);
2430 log_debug_errno(r
, "Failed to detect that %s is a mount point or not: %m", root
);
2434 r
= mount_nofollow_verbose(LOG_DEBUG
, root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2440 /* Let's mount the main root directory to the root directory to use */
2441 r
= mount_nofollow_verbose(LOG_DEBUG
, "/", root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2446 /* Try to set up the new root directory before mounting anything else there. */
2447 if (root_image
|| root_directory
)
2448 (void) base_filesystem_create(root
, UID_INVALID
, GID_INVALID
);
2450 /* Now make the magic happen */
2451 r
= apply_mounts(root
, mount_image_policy
, extension_image_policy
, ns_info
, mounts
, &n_mounts
, exec_dir_symlinks
, error_path
);
2455 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
2456 r
= mount_switch_root(root
, /* mount_propagation_flag = */ 0);
2457 if (r
== -EINVAL
&& root_directory
) {
2458 /* If we are using root_directory and we don't have privileges (ie: user manager in a user
2459 * namespace) and the root_directory is already a mount point in the parent namespace,
2460 * MS_MOVE will fail as we don't have permission to change it (with EINVAL rather than
2461 * EPERM). Attempt to bind-mount it over itself (like we do above if it's not already a
2462 * mount point) and try again. */
2463 r
= mount_nofollow_verbose(LOG_DEBUG
, root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2466 r
= mount_switch_root(root
, /* mount_propagation_flag = */ 0);
2469 log_debug_errno(r
, "Failed to mount root with MS_MOVE: %m");
2473 /* Remount / as the desired mode. Note that this will not reestablish propagation from our side to
2474 * the host, since what's disconnected is disconnected. */
2475 if (mount(NULL
, "/", NULL
, mount_propagation_flag
| MS_REC
, NULL
) < 0) {
2476 r
= log_debug_errno(errno
, "Failed to remount '/' with desired mount flags: %m");
2480 /* bind_mount_in_namespace() will MS_MOVE into that directory, and that's only
2481 * supported for non-shared mounts. This needs to happen after remounting / or it will fail. */
2482 if (setup_propagate
) {
2483 r
= mount(NULL
, incoming_dir
, NULL
, MS_SLAVE
, NULL
);
2485 log_error_errno(r
, "Failed to remount %s with MS_SLAVE: %m", incoming_dir
);
2494 for (m
= mounts
; m
< mounts
+ n_mounts
; m
++)
2495 mount_entry_done(m
);
2502 void bind_mount_free_many(BindMount
*b
, size_t n
) {
2503 assert(b
|| n
== 0);
2505 for (size_t i
= 0; i
< n
; i
++) {
2507 free(b
[i
].destination
);
2513 int bind_mount_add(BindMount
**b
, size_t *n
, const BindMount
*item
) {
2514 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
2521 s
= strdup(item
->source
);
2525 d
= strdup(item
->destination
);
2529 c
= reallocarray(*b
, *n
+ 1, sizeof(BindMount
));
2535 c
[(*n
) ++] = (BindMount
) {
2536 .source
= TAKE_PTR(s
),
2537 .destination
= TAKE_PTR(d
),
2538 .read_only
= item
->read_only
,
2539 .nosuid
= item
->nosuid
,
2540 .recursive
= item
->recursive
,
2541 .ignore_enoent
= item
->ignore_enoent
,
2547 MountImage
* mount_image_free_many(MountImage
*m
, size_t *n
) {
2549 assert(m
|| *n
== 0);
2551 for (size_t i
= 0; i
< *n
; i
++) {
2553 free(m
[i
].destination
);
2554 mount_options_free_all(m
[i
].mount_options
);
2562 int mount_image_add(MountImage
**m
, size_t *n
, const MountImage
*item
) {
2563 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
2564 _cleanup_(mount_options_free_allp
) MountOptions
*options
= NULL
;
2571 s
= strdup(item
->source
);
2575 if (item
->destination
) {
2576 d
= strdup(item
->destination
);
2581 LIST_FOREACH(mount_options
, i
, item
->mount_options
) {
2582 _cleanup_(mount_options_free_allp
) MountOptions
*o
= NULL
;
2584 o
= new(MountOptions
, 1);
2588 *o
= (MountOptions
) {
2589 .partition_designator
= i
->partition_designator
,
2590 .options
= strdup(i
->options
),
2595 LIST_APPEND(mount_options
, options
, TAKE_PTR(o
));
2598 c
= reallocarray(*m
, *n
+ 1, sizeof(MountImage
));
2604 c
[(*n
) ++] = (MountImage
) {
2605 .source
= TAKE_PTR(s
),
2606 .destination
= TAKE_PTR(d
),
2607 .mount_options
= TAKE_PTR(options
),
2608 .ignore_enoent
= item
->ignore_enoent
,
2615 void temporary_filesystem_free_many(TemporaryFileSystem
*t
, size_t n
) {
2616 assert(t
|| n
== 0);
2618 for (size_t i
= 0; i
< n
; i
++) {
2626 int temporary_filesystem_add(
2627 TemporaryFileSystem
**t
,
2630 const char *options
) {
2632 _cleanup_free_
char *p
= NULL
, *o
= NULL
;
2633 TemporaryFileSystem
*c
;
2643 if (!isempty(options
)) {
2644 o
= strdup(options
);
2649 c
= reallocarray(*t
, *n
+ 1, sizeof(TemporaryFileSystem
));
2655 c
[(*n
) ++] = (TemporaryFileSystem
) {
2656 .path
= TAKE_PTR(p
),
2657 .options
= TAKE_PTR(o
),
2663 static int make_tmp_prefix(const char *prefix
) {
2664 _cleanup_free_
char *t
= NULL
;
2665 _cleanup_close_
int fd
= -EBADF
;
2668 /* Don't do anything unless we know the dir is actually missing */
2669 r
= access(prefix
, F_OK
);
2672 if (errno
!= ENOENT
)
2676 r
= mkdir_parents(prefix
, 0755);
2680 r
= tempfn_random(prefix
, NULL
, &t
);
2684 /* umask will corrupt this access mode, but that doesn't matter, we need to call chmod() anyway for
2685 * the suid bit, below. */
2686 fd
= open_mkdir_at(AT_FDCWD
, t
, O_EXCL
|O_CLOEXEC
, 0777);
2690 r
= RET_NERRNO(fchmod(fd
, 01777));
2696 r
= RET_NERRNO(rename(t
, prefix
));
2699 return r
== -EEXIST
? 0 : r
; /* it's fine if someone else created the dir by now */
2706 static int setup_one_tmp_dir(const char *id
, const char *prefix
, char **path
, char **tmp_path
) {
2707 _cleanup_free_
char *x
= NULL
;
2708 _cleanup_free_
char *y
= NULL
;
2717 /* We include the boot id in the directory so that after a
2718 * reboot we can easily identify obsolete directories. */
2720 r
= sd_id128_get_boot(&boot_id
);
2724 x
= strjoin(prefix
, "/systemd-private-", SD_ID128_TO_STRING(boot_id
), "-", id
, "-XXXXXX");
2728 r
= make_tmp_prefix(prefix
);
2734 if (errno
== EROFS
|| ERRNO_IS_DISK_SPACE(errno
))
2741 y
= strjoin(x
, "/tmp");
2746 if (mkdir(y
, 0777 | S_ISVTX
) < 0)
2749 r
= label_fix_full(AT_FDCWD
, y
, prefix
, 0);
2754 *tmp_path
= TAKE_PTR(y
);
2756 /* Trouble: we failed to create the directory. Instead of failing, let's simulate /tmp being
2757 * read-only. This way the service will get the EROFS result as if it was writing to the real
2760 r
= mkdir_p(RUN_SYSTEMD_EMPTY
, 0500);
2764 r
= free_and_strdup(&x
, RUN_SYSTEMD_EMPTY
);
2769 *path
= TAKE_PTR(x
);
2773 int setup_tmp_dirs(const char *id
, char **tmp_dir
, char **var_tmp_dir
) {
2774 _cleanup_(namespace_cleanup_tmpdirp
) char *a
= NULL
;
2775 _cleanup_(rmdir_and_freep
) char *a_tmp
= NULL
;
2781 assert(var_tmp_dir
);
2783 r
= setup_one_tmp_dir(id
, "/tmp", &a
, &a_tmp
);
2787 r
= setup_one_tmp_dir(id
, "/var/tmp", &b
, NULL
);
2791 a_tmp
= mfree(a_tmp
); /* avoid rmdir */
2792 *tmp_dir
= TAKE_PTR(a
);
2793 *var_tmp_dir
= TAKE_PTR(b
);
2798 int setup_shareable_ns(int ns_storage_socket
[static 2], unsigned long nsflag
) {
2799 _cleanup_close_
int ns
= -EBADF
;
2801 const char *ns_name
, *ns_path
;
2803 assert(ns_storage_socket
);
2804 assert(ns_storage_socket
[0] >= 0);
2805 assert(ns_storage_socket
[1] >= 0);
2807 ns_name
= namespace_single_flag_to_string(nsflag
);
2810 /* We use the passed socketpair as a storage buffer for our
2811 * namespace reference fd. Whatever process runs this first
2812 * shall create a new namespace, all others should just join
2813 * it. To serialize that we use a file lock on the socket
2816 * It's a bit crazy, but hey, works great! */
2818 r
= posix_lock(ns_storage_socket
[0], LOCK_EX
);
2822 CLEANUP_POSIX_UNLOCK(ns_storage_socket
[0]);
2824 ns
= receive_one_fd(ns_storage_socket
[0], MSG_PEEK
|MSG_DONTWAIT
);
2826 /* Yay, found something, so let's join the namespace */
2827 r
= RET_NERRNO(setns(ns
, nsflag
));
2837 /* Nothing stored yet, so let's create a new namespace. */
2839 if (unshare(nsflag
) < 0)
2842 (void) loopback_setup();
2844 ns_path
= strjoina("/proc/self/ns/", ns_name
);
2845 ns
= open(ns_path
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
2849 r
= send_one_fd(ns_storage_socket
[1], ns
, MSG_DONTWAIT
);
2856 int open_shareable_ns_path(int ns_storage_socket
[static 2], const char *path
, unsigned long nsflag
) {
2857 _cleanup_close_
int ns
= -EBADF
;
2860 assert(ns_storage_socket
);
2861 assert(ns_storage_socket
[0] >= 0);
2862 assert(ns_storage_socket
[1] >= 0);
2865 /* If the storage socket doesn't contain a ns fd yet, open one via the file system and store it in
2866 * it. This is supposed to be called ahead of time, i.e. before setup_shareable_ns() which will
2867 * allocate a new anonymous ns if needed. */
2869 r
= posix_lock(ns_storage_socket
[0], LOCK_EX
);
2873 CLEANUP_POSIX_UNLOCK(ns_storage_socket
[0]);
2875 ns
= receive_one_fd(ns_storage_socket
[0], MSG_PEEK
|MSG_DONTWAIT
);
2881 /* Nothing stored yet. Open the file from the file system. */
2883 ns
= open(path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
2887 r
= fd_is_ns(ns
, nsflag
);
2890 if (r
< 0 && r
!= -EUCLEAN
) /* EUCLEAN: we don't know */
2893 r
= send_one_fd(ns_storage_socket
[1], ns
, MSG_DONTWAIT
);
2900 bool ns_type_supported(NamespaceType type
) {
2901 const char *t
, *ns_proc
;
2903 t
= namespace_type_to_string(type
);
2904 if (!t
) /* Don't know how to translate this? Then it's not supported */
2907 ns_proc
= strjoina("/proc/self/ns/", t
);
2908 return access(ns_proc
, F_OK
) == 0;
2911 static const char *const protect_home_table
[_PROTECT_HOME_MAX
] = {
2912 [PROTECT_HOME_NO
] = "no",
2913 [PROTECT_HOME_YES
] = "yes",
2914 [PROTECT_HOME_READ_ONLY
] = "read-only",
2915 [PROTECT_HOME_TMPFS
] = "tmpfs",
2918 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home
, ProtectHome
, PROTECT_HOME_YES
);
2920 static const char *const protect_system_table
[_PROTECT_SYSTEM_MAX
] = {
2921 [PROTECT_SYSTEM_NO
] = "no",
2922 [PROTECT_SYSTEM_YES
] = "yes",
2923 [PROTECT_SYSTEM_FULL
] = "full",
2924 [PROTECT_SYSTEM_STRICT
] = "strict",
2927 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system
, ProtectSystem
, PROTECT_SYSTEM_YES
);
2929 static const char* const namespace_type_table
[] = {
2930 [NAMESPACE_MOUNT
] = "mnt",
2931 [NAMESPACE_CGROUP
] = "cgroup",
2932 [NAMESPACE_UTS
] = "uts",
2933 [NAMESPACE_IPC
] = "ipc",
2934 [NAMESPACE_USER
] = "user",
2935 [NAMESPACE_PID
] = "pid",
2936 [NAMESPACE_NET
] = "net",
2937 [NAMESPACE_TIME
] = "time",
2940 DEFINE_STRING_TABLE_LOOKUP(namespace_type
, NamespaceType
);
2942 static const char* const protect_proc_table
[_PROTECT_PROC_MAX
] = {
2943 [PROTECT_PROC_DEFAULT
] = "default",
2944 [PROTECT_PROC_NOACCESS
] = "noaccess",
2945 [PROTECT_PROC_INVISIBLE
] = "invisible",
2946 [PROTECT_PROC_PTRACEABLE
] = "ptraceable",
2949 DEFINE_STRING_TABLE_LOOKUP(protect_proc
, ProtectProc
);
2951 static const char* const proc_subset_table
[_PROC_SUBSET_MAX
] = {
2952 [PROC_SUBSET_ALL
] = "all",
2953 [PROC_SUBSET_PID
] = "pid",
2956 DEFINE_STRING_TABLE_LOOKUP(proc_subset
, ProcSubset
);