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 /* Helper struct for naming simplicity and reusability */
233 static const struct {
234 const char *level_env
;
235 const char *level_env_print
;
236 } image_class_info
[_IMAGE_CLASS_MAX
] = {
238 .level_env
= "SYSEXT_LEVEL",
239 .level_env_print
= " SYSEXT_LEVEL=",
242 .level_env
= "CONFEXT_LEVEL",
243 .level_env_print
= " CONFEXT_LEVEL=",
247 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(mount_mode
, MountMode
);
249 static const char *mount_entry_path(const MountEntry
*p
) {
252 /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that,
253 * otherwise the stack/static ->path field is returned. */
255 return p
->path_malloc
?: p
->path_const
;
258 static const char *mount_entry_unprefixed_path(const MountEntry
*p
) {
261 /* Returns the unprefixed path (ie: before prefix_where_needed() ran), if any */
263 return p
->unprefixed_path_malloc
?: p
->unprefixed_path_const
?: mount_entry_path(p
);
266 static void mount_entry_consume_prefix(MountEntry
*p
, char *new_path
) {
268 assert(p
->path_malloc
|| p
->path_const
);
271 /* Saves current path in unprefixed_ variable, and takes over new_path */
273 free_and_replace(p
->unprefixed_path_malloc
, p
->path_malloc
);
274 /* If we didn't have a path on the heap, then it's a static one */
275 if (!p
->unprefixed_path_malloc
)
276 p
->unprefixed_path_const
= p
->path_const
;
277 p
->path_malloc
= new_path
;
278 p
->has_prefix
= true;
281 static bool mount_entry_read_only(const MountEntry
*p
) {
284 return p
->read_only
|| IN_SET(p
->mode
, READONLY
, INACCESSIBLE
, PRIVATE_TMP_READONLY
);
287 static bool mount_entry_noexec(const MountEntry
*p
) {
290 return p
->noexec
|| IN_SET(p
->mode
, NOEXEC
, INACCESSIBLE
, PRIVATE_SYSFS
, BIND_SYSFS
, PROCFS
);
293 static bool mount_entry_exec(const MountEntry
*p
) {
296 return p
->exec
|| p
->mode
== EXEC
;
299 static const char *mount_entry_source(const MountEntry
*p
) {
302 return p
->source_malloc
?: p
->source_const
;
305 static const char *mount_entry_options(const MountEntry
*p
) {
308 return p
->options_malloc
?: p
->options_const
;
311 static void mount_entry_done(MountEntry
*p
) {
314 p
->path_malloc
= mfree(p
->path_malloc
);
315 p
->unprefixed_path_malloc
= mfree(p
->unprefixed_path_malloc
);
316 p
->source_malloc
= mfree(p
->source_malloc
);
317 p
->options_malloc
= mfree(p
->options_malloc
);
318 p
->image_options
= mount_options_free_all(p
->image_options
);
321 static int append_access_mounts(MountEntry
**p
, char **strv
, MountMode mode
, bool forcibly_require_prefix
) {
324 /* Adds a list of user-supplied READWRITE/READWRITE_IMPLICIT/READONLY/INACCESSIBLE entries */
326 STRV_FOREACH(i
, strv
) {
327 bool ignore
= false, needs_prefix
= false;
330 /* Look for any prefixes */
331 if (startswith(e
, "-")) {
335 if (startswith(e
, "+")) {
340 if (!path_is_absolute(e
))
341 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
342 "Path is not absolute: %s", e
);
344 *((*p
)++) = (MountEntry
) {
348 .has_prefix
= !needs_prefix
&& !forcibly_require_prefix
,
355 static int append_empty_dir_mounts(MountEntry
**p
, char **strv
) {
358 /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the
359 * "/private/" boundary directories for DynamicUser=1. */
361 STRV_FOREACH(i
, strv
) {
363 *((*p
)++) = (MountEntry
) {
368 .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
,
369 .flags
= MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
,
376 static int append_bind_mounts(MountEntry
**p
, const BindMount
*binds
, size_t n
) {
379 for (size_t i
= 0; i
< n
; i
++) {
380 const BindMount
*b
= binds
+ i
;
382 *((*p
)++) = (MountEntry
) {
383 .path_const
= b
->destination
,
384 .mode
= b
->recursive
? BIND_MOUNT_RECURSIVE
: BIND_MOUNT
,
385 .read_only
= b
->read_only
,
387 .source_const
= b
->source
,
388 .ignore
= b
->ignore_enoent
,
395 static int append_mount_images(MountEntry
**p
, const MountImage
*mount_images
, size_t n
) {
398 for (size_t i
= 0; i
< n
; i
++) {
399 const MountImage
*m
= mount_images
+ i
;
401 *((*p
)++) = (MountEntry
) {
402 .path_const
= m
->destination
,
403 .mode
= MOUNT_IMAGES
,
404 .source_const
= m
->source
,
405 .image_options
= m
->mount_options
,
406 .ignore
= m
->ignore_enoent
,
413 static int append_extensions(
416 const char *extension_dir
,
418 const MountImage
*mount_images
,
420 char **extension_directories
) {
422 _cleanup_strv_free_
char **overlays
= NULL
;
425 if (n
== 0 && strv_isempty(extension_directories
))
429 assert(extension_dir
);
431 /* Prepare a list of overlays, that will have as each element a string suitable for being
432 * passed as a lowerdir= parameter, so start with the hierarchy on the root.
433 * The overlays vector will have the same number of elements and will correspond to the
434 * hierarchies vector, so they can be iterated upon together. */
435 STRV_FOREACH(hierarchy
, hierarchies
) {
436 _cleanup_free_
char *prefixed_hierarchy
= NULL
;
438 prefixed_hierarchy
= path_join(root
, *hierarchy
);
439 if (!prefixed_hierarchy
)
442 r
= strv_consume(&overlays
, TAKE_PTR(prefixed_hierarchy
));
447 /* First, prepare a mount for each image, but these won't be visible to the unit, instead
448 * they will be mounted in our propagate directory, and used as a source for the overlay. */
449 for (size_t i
= 0; i
< n
; i
++) {
450 _cleanup_free_
char *mount_point
= NULL
;
451 const MountImage
*m
= mount_images
+ i
;
453 r
= asprintf(&mount_point
, "%s/%zu", extension_dir
, i
);
457 for (size_t j
= 0; hierarchies
&& hierarchies
[j
]; ++j
) {
458 _cleanup_free_
char *prefixed_hierarchy
= NULL
, *escaped
= NULL
, *lowerdir
= NULL
;
460 prefixed_hierarchy
= path_join(mount_point
, hierarchies
[j
]);
461 if (!prefixed_hierarchy
)
464 escaped
= shell_escape(prefixed_hierarchy
, ",:");
468 /* Note that lowerdir= parameters are in 'reverse' order, so the
469 * top-most directory in the overlay comes first in the list. */
470 lowerdir
= strjoin(escaped
, ":", overlays
[j
]);
474 free_and_replace(overlays
[j
], lowerdir
);
477 *((*p
)++) = (MountEntry
) {
478 .path_malloc
= TAKE_PTR(mount_point
),
479 .image_options
= m
->mount_options
,
480 .ignore
= m
->ignore_enoent
,
481 .source_const
= m
->source
,
482 .mode
= EXTENSION_IMAGES
,
487 /* Secondly, extend the lowerdir= parameters with each ExtensionDirectory.
488 * Bind mount them in the same location as the ExtensionImages, so that we
489 * can check that they are valid trees (extension-release.d). */
490 STRV_FOREACH(extension_directory
, extension_directories
) {
491 _cleanup_free_
char *mount_point
= NULL
, *source
= NULL
;
492 const char *e
= *extension_directory
;
493 bool ignore_enoent
= false;
495 /* Pick up the counter where the ExtensionImages left it. */
496 r
= asprintf(&mount_point
, "%s/%zu", extension_dir
, n
++);
500 /* Look for any prefixes */
501 if (startswith(e
, "-")) {
503 ignore_enoent
= true;
505 /* Ignore this for now */
506 if (startswith(e
, "+"))
513 for (size_t j
= 0; hierarchies
&& hierarchies
[j
]; ++j
) {
514 _cleanup_free_
char *prefixed_hierarchy
= NULL
, *escaped
= NULL
, *lowerdir
= NULL
;
516 prefixed_hierarchy
= path_join(mount_point
, hierarchies
[j
]);
517 if (!prefixed_hierarchy
)
520 escaped
= shell_escape(prefixed_hierarchy
, ",:");
524 /* Note that lowerdir= parameters are in 'reverse' order, so the
525 * top-most directory in the overlay comes first in the list. */
526 lowerdir
= strjoin(escaped
, ":", overlays
[j
]);
530 free_and_replace(overlays
[j
], lowerdir
);
533 *((*p
)++) = (MountEntry
) {
534 .path_malloc
= TAKE_PTR(mount_point
),
535 .source_malloc
= TAKE_PTR(source
),
536 .mode
= EXTENSION_DIRECTORIES
,
537 .ignore
= ignore_enoent
,
543 /* Then, for each hierarchy, prepare an overlay with the list of lowerdir= strings
545 for (size_t i
= 0; hierarchies
&& hierarchies
[i
]; ++i
) {
546 _cleanup_free_
char *prefixed_hierarchy
= NULL
;
548 prefixed_hierarchy
= path_join(root
, hierarchies
[i
]);
549 if (!prefixed_hierarchy
)
552 *((*p
)++) = (MountEntry
) {
553 .path_malloc
= TAKE_PTR(prefixed_hierarchy
),
554 .options_malloc
= TAKE_PTR(overlays
[i
]),
555 .mode
= OVERLAY_MOUNT
,
557 .ignore
= true, /* If the source image doesn't set the ignore bit it will fail earlier. */
564 static int append_tmpfs_mounts(MountEntry
**p
, const TemporaryFileSystem
*tmpfs
, size_t n
) {
567 for (size_t i
= 0; i
< n
; i
++) {
568 const TemporaryFileSystem
*t
= tmpfs
+ i
;
569 _cleanup_free_
char *o
= NULL
, *str
= NULL
;
574 if (!path_is_absolute(t
->path
))
575 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
576 "Path is not absolute: %s",
579 str
= strjoin("mode=0755" NESTED_TMPFS_LIMITS
",", t
->options
);
583 r
= mount_option_mangle(str
, MS_NODEV
|MS_STRICTATIME
, &flags
, &o
);
585 return log_debug_errno(r
, "Failed to parse mount option '%s': %m", str
);
587 ro
= flags
& MS_RDONLY
;
591 *((*p
)++) = (MountEntry
) {
592 .path_const
= t
->path
,
595 .options_malloc
= TAKE_PTR(o
),
603 static int append_static_mounts(MountEntry
**p
, const MountEntry
*mounts
, size_t n
, bool ignore_protect
) {
607 /* Adds a list of static pre-defined entries */
609 for (size_t i
= 0; i
< n
; i
++)
610 *((*p
)++) = (MountEntry
) {
611 .path_const
= mount_entry_path(mounts
+i
),
612 .mode
= mounts
[i
].mode
,
613 .ignore
= mounts
[i
].ignore
|| ignore_protect
,
619 static int append_protect_home(MountEntry
**p
, ProtectHome protect_home
, bool ignore_protect
) {
622 switch (protect_home
) {
624 case PROTECT_HOME_NO
:
627 case PROTECT_HOME_READ_ONLY
:
628 return append_static_mounts(p
, protect_home_read_only_table
, ELEMENTSOF(protect_home_read_only_table
), ignore_protect
);
630 case PROTECT_HOME_TMPFS
:
631 return append_static_mounts(p
, protect_home_tmpfs_table
, ELEMENTSOF(protect_home_tmpfs_table
), ignore_protect
);
633 case PROTECT_HOME_YES
:
634 return append_static_mounts(p
, protect_home_yes_table
, ELEMENTSOF(protect_home_yes_table
), ignore_protect
);
637 assert_not_reached();
641 static int append_protect_system(MountEntry
**p
, ProtectSystem protect_system
, bool ignore_protect
) {
644 switch (protect_system
) {
646 case PROTECT_SYSTEM_NO
:
649 case PROTECT_SYSTEM_STRICT
:
650 return append_static_mounts(p
, protect_system_strict_table
, ELEMENTSOF(protect_system_strict_table
), ignore_protect
);
652 case PROTECT_SYSTEM_YES
:
653 return append_static_mounts(p
, protect_system_yes_table
, ELEMENTSOF(protect_system_yes_table
), ignore_protect
);
655 case PROTECT_SYSTEM_FULL
:
656 return append_static_mounts(p
, protect_system_full_table
, ELEMENTSOF(protect_system_full_table
), ignore_protect
);
659 assert_not_reached();
663 static int mount_path_compare(const MountEntry
*a
, const MountEntry
*b
) {
666 /* ExtensionImages/Directories will be used by other mounts as a base, so sort them first
667 * regardless of the prefix - they are set up in the propagate directory anyway */
668 d
= -CMP(a
->mode
== EXTENSION_IMAGES
, b
->mode
== EXTENSION_IMAGES
);
671 d
= -CMP(a
->mode
== EXTENSION_DIRECTORIES
, b
->mode
== EXTENSION_DIRECTORIES
);
675 /* If the paths are not equal, then order prefixes first */
676 d
= path_compare(mount_entry_path(a
), mount_entry_path(b
));
680 /* If the paths are equal, check the mode */
681 return CMP((int) a
->mode
, (int) b
->mode
);
684 static int prefix_where_needed(MountEntry
*m
, size_t n
, const char *root_directory
) {
685 /* Prefixes all paths in the bind mount table with the root directory if the entry needs that. */
689 for (size_t i
= 0; i
< n
; i
++) {
695 s
= path_join(root_directory
, mount_entry_path(m
+i
));
699 mount_entry_consume_prefix(&m
[i
], s
);
705 static void drop_duplicates(MountEntry
*m
, size_t *n
) {
706 MountEntry
*f
, *t
, *previous
;
711 /* Drops duplicate entries. Expects that the array is properly ordered already. */
713 for (f
= m
, t
= m
, previous
= NULL
; f
< m
+ *n
; f
++) {
715 /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare()
716 * above. Note that we only drop duplicates that haven't been mounted yet. */
718 path_equal(mount_entry_path(f
), mount_entry_path(previous
)) &&
719 !f
->applied
&& !previous
->applied
) {
720 log_debug("%s (%s) is duplicate.", mount_entry_path(f
), mount_mode_to_string(f
->mode
));
721 /* Propagate the flags to the remaining entry */
722 previous
->read_only
= previous
->read_only
|| mount_entry_read_only(f
);
723 previous
->noexec
= previous
->noexec
|| mount_entry_noexec(f
);
724 previous
->exec
= previous
->exec
|| mount_entry_exec(f
);
737 static void drop_inaccessible(MountEntry
*m
, size_t *n
) {
739 const char *clear
= NULL
;
744 /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly
745 * ordered already. */
747 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
749 /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop
750 * it, as inaccessible paths really should drop the entire subtree. */
751 if (clear
&& path_startswith(mount_entry_path(f
), clear
)) {
752 log_debug("%s is masked by %s.", mount_entry_path(f
), clear
);
757 clear
= f
->mode
== INACCESSIBLE
? mount_entry_path(f
) : NULL
;
766 static void drop_nop(MountEntry
*m
, size_t *n
) {
772 /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the
773 * list is ordered by prefixes. */
775 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
777 /* Only suppress such subtrees for READONLY, READWRITE and READWRITE_IMPLICIT entries */
778 if (IN_SET(f
->mode
, READONLY
, READWRITE
, READWRITE_IMPLICIT
)) {
779 MountEntry
*found
= NULL
;
781 /* Now let's find the first parent of the entry we are looking at. */
782 for (MountEntry
*p
= PTR_SUB1(t
, m
); p
; p
= PTR_SUB1(p
, m
))
783 if (path_startswith(mount_entry_path(f
), mount_entry_path(p
))) {
788 /* We found it, let's see if it's the same mode, if so, we can drop this entry */
789 if (found
&& found
->mode
== f
->mode
) {
790 log_debug("%s (%s) is made redundant by %s (%s)",
791 mount_entry_path(f
), mount_mode_to_string(f
->mode
),
792 mount_entry_path(found
), mount_mode_to_string(found
->mode
));
805 static void drop_outside_root(const char *root_directory
, MountEntry
*m
, size_t *n
) {
815 /* Drops all mounts that are outside of the root directory. */
817 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
819 /* ExtensionImages/Directories bases are opened in /run/systemd/unit-extensions on the host */
820 if (!IN_SET(f
->mode
, EXTENSION_IMAGES
, EXTENSION_DIRECTORIES
) && !path_startswith(mount_entry_path(f
), root_directory
)) {
821 log_debug("%s is outside of root directory.", mount_entry_path(f
));
833 static int clone_device_node(
835 const char *temporary_mount
,
836 bool *make_devnode
) {
838 _cleanup_free_
char *sl
= NULL
;
839 const char *dn
, *bn
, *t
;
843 if (stat(d
, &st
) < 0) {
844 if (errno
== ENOENT
) {
845 log_debug_errno(errno
, "Device node '%s' to clone does not exist, ignoring.", d
);
849 return log_debug_errno(errno
, "Failed to stat() device node '%s' to clone, ignoring: %m", d
);
852 if (!S_ISBLK(st
.st_mode
) &&
853 !S_ISCHR(st
.st_mode
))
854 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
855 "Device node '%s' to clone is not a device node, ignoring.",
858 dn
= strjoina(temporary_mount
, d
);
860 /* First, try to create device node properly */
862 mac_selinux_create_file_prepare(d
, st
.st_mode
);
863 r
= mknod(dn
, st
.st_mode
, st
.st_rdev
);
864 mac_selinux_create_file_clear();
868 return log_debug_errno(errno
, "mknod failed for %s: %m", d
);
870 /* This didn't work, let's not try this again for the next iterations. */
871 *make_devnode
= false;
874 /* We're about to fall back to bind-mounting the device node. So create a dummy bind-mount target.
875 * Do not prepare device-node SELinux label (see issue 13762) */
876 r
= mknod(dn
, S_IFREG
, 0);
877 if (r
< 0 && errno
!= EEXIST
)
878 return log_debug_errno(errno
, "mknod() fallback failed for '%s': %m", d
);
880 /* Fallback to bind-mounting: The assumption here is that all used device nodes carry standard
881 * properties. Specifically, the devices nodes we bind-mount should either be owned by root:root or
882 * root:tty (e.g. /dev/tty, /dev/ptmx) and should not carry ACLs. */
883 r
= mount_nofollow_verbose(LOG_DEBUG
, d
, dn
, NULL
, MS_BIND
, NULL
);
888 bn
= path_startswith(d
, "/dev/");
892 /* Create symlinks like /dev/char/1:9 → ../urandom */
893 if (asprintf(&sl
, "%s/dev/%s/" DEVNUM_FORMAT_STR
,
895 S_ISCHR(st
.st_mode
) ? "char" : "block",
896 DEVNUM_FORMAT_VAL(st
.st_rdev
)) < 0)
899 (void) mkdir_parents(sl
, 0755);
901 t
= strjoina("../", bn
);
902 if (symlink(t
, sl
) < 0)
903 log_debug_errno(errno
, "Failed to symlink '%s' to '%s', ignoring: %m", t
, sl
);
908 static int mount_private_dev(MountEntry
*m
) {
909 static const char devnodes
[] =
917 char temporary_mount
[] = "/tmp/namespace-dev-XXXXXX";
918 const char *dev
= NULL
, *devpts
= NULL
, *devshm
= NULL
, *devhugepages
= NULL
, *devmqueue
= NULL
, *devlog
= NULL
, *devptmx
= NULL
;
919 bool can_mknod
= true;
924 if (!mkdtemp(temporary_mount
))
925 return log_debug_errno(errno
, "Failed to create temporary directory '%s': %m", temporary_mount
);
927 dev
= strjoina(temporary_mount
, "/dev");
928 (void) mkdir(dev
, 0755);
929 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", dev
, "tmpfs", DEV_MOUNT_OPTIONS
, "mode=0755" TMPFS_LIMITS_PRIVATE_DEV
);
933 r
= label_fix_full(AT_FDCWD
, dev
, "/dev", 0);
935 log_debug_errno(r
, "Failed to fix label of '%s' as /dev: %m", dev
);
939 devpts
= strjoina(temporary_mount
, "/dev/pts");
940 (void) mkdir(devpts
, 0755);
941 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev/pts", devpts
, NULL
, MS_BIND
, NULL
);
945 /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx.
946 * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible.
947 * Thus, in that case make a clone.
948 * In nspawn and other containers it will be a symlink, in that case make it a symlink. */
949 r
= is_symlink("/dev/ptmx");
951 log_debug_errno(r
, "Failed to detect whether /dev/ptmx is a symlink or not: %m");
954 devptmx
= strjoina(temporary_mount
, "/dev/ptmx");
955 if (symlink("pts/ptmx", devptmx
) < 0) {
956 r
= log_debug_errno(errno
, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx
);
960 r
= clone_device_node("/dev/ptmx", temporary_mount
, &can_mknod
);
965 devshm
= strjoina(temporary_mount
, "/dev/shm");
966 (void) mkdir(devshm
, 0755);
967 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev/shm", devshm
, NULL
, MS_BIND
, NULL
);
971 devmqueue
= strjoina(temporary_mount
, "/dev/mqueue");
972 (void) mkdir(devmqueue
, 0755);
973 (void) mount_nofollow_verbose(LOG_DEBUG
, "/dev/mqueue", devmqueue
, NULL
, MS_BIND
, NULL
);
975 devhugepages
= strjoina(temporary_mount
, "/dev/hugepages");
976 (void) mkdir(devhugepages
, 0755);
977 (void) mount_nofollow_verbose(LOG_DEBUG
, "/dev/hugepages", devhugepages
, NULL
, MS_BIND
, NULL
);
979 devlog
= strjoina(temporary_mount
, "/dev/log");
980 if (symlink("/run/systemd/journal/dev-log", devlog
) < 0)
981 log_debug_errno(errno
, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog
);
983 NULSTR_FOREACH(d
, devnodes
) {
984 r
= clone_device_node(d
, temporary_mount
, &can_mknod
);
985 /* ENXIO means the *source* is not a device file, skip creation in that case */
986 if (r
< 0 && r
!= -ENXIO
)
990 r
= dev_setup(temporary_mount
, UID_INVALID
, GID_INVALID
);
992 log_debug_errno(r
, "Failed to set up basic device tree at '%s', ignoring: %m", temporary_mount
);
994 /* Create the /dev directory if missing. It is more likely to be missing when the service is started
995 * with RootDirectory. This is consistent with mount units creating the mount points when missing. */
996 (void) mkdir_p_label(mount_entry_path(m
), 0755);
998 /* Unmount everything in old /dev */
999 r
= umount_recursive(mount_entry_path(m
), 0);
1001 log_debug_errno(r
, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m
));
1003 r
= mount_nofollow_verbose(LOG_DEBUG
, dev
, mount_entry_path(m
), NULL
, MS_MOVE
, NULL
);
1008 (void) rmdir(temporary_mount
);
1014 (void) umount_verbose(LOG_DEBUG
, devpts
, UMOUNT_NOFOLLOW
);
1017 (void) umount_verbose(LOG_DEBUG
, devshm
, UMOUNT_NOFOLLOW
);
1020 (void) umount_verbose(LOG_DEBUG
, devhugepages
, UMOUNT_NOFOLLOW
);
1023 (void) umount_verbose(LOG_DEBUG
, devmqueue
, UMOUNT_NOFOLLOW
);
1025 (void) umount_verbose(LOG_DEBUG
, dev
, UMOUNT_NOFOLLOW
);
1027 (void) rmdir(temporary_mount
);
1032 static int mount_bind_dev(const MountEntry
*m
) {
1037 /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the
1038 * service's /dev. This is only used when RootDirectory= is set. */
1040 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1042 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
1044 return log_debug_errno(r
, "Unable to determine whether /dev is already mounted: %m");
1045 if (r
> 0) /* make this a NOP if /dev is already a mount point */
1048 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
);
1055 static int mount_private_sysfs(const MountEntry
*m
) {
1056 const char *p
= mount_entry_path(ASSERT_PTR(m
));
1059 (void) mkdir_p_label(p
, 0755);
1061 r
= remount_sysfs(p
);
1062 if (r
< 0 && (ERRNO_IS_PRIVILEGE(r
) || ERRNO_IS_NOT_SUPPORTED(r
))) {
1063 /* Running with an unprivileged user (PrivateUsers=yes), or the kernel seems old. Falling
1064 * back to bind mount the host's version so that we get all child mounts of it, too. */
1066 log_debug_errno(r
, "Failed to remount sysfs on %s, falling back to bind mount: %m", p
);
1068 (void) umount_recursive(p
, 0);
1070 r
= mount_nofollow_verbose(LOG_DEBUG
, "/sys", p
, NULL
, MS_BIND
|MS_REC
, NULL
);
1073 return log_debug_errno(r
, "Failed to remount sysfs on %s: %m", p
);
1078 static int mount_bind_sysfs(const MountEntry
*m
) {
1083 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1085 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
1087 return log_debug_errno(r
, "Unable to determine whether /sys is already mounted: %m");
1088 if (r
> 0) /* make this a NOP if /sys is already a mount point */
1091 /* Bind mount the host's version so that we get all child mounts of it, too. */
1092 r
= mount_nofollow_verbose(LOG_DEBUG
, "/sys", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
);
1099 static int mount_procfs(const MountEntry
*m
, const NamespaceInfo
*ns_info
) {
1100 _cleanup_free_
char *opts
= NULL
;
1101 const char *entry_path
;
1107 if (ns_info
->protect_proc
!= PROTECT_PROC_DEFAULT
||
1108 ns_info
->proc_subset
!= PROC_SUBSET_ALL
) {
1110 /* Starting with kernel 5.8 procfs' hidepid= logic is truly per-instance (previously it
1111 * pretended to be per-instance but actually was per-namespace), hence let's make use of it
1112 * if requested. To make sure this logic succeeds only on kernels where hidepid= is
1113 * per-instance, we'll exclusively use the textual value for hidepid=, since support was
1114 * added in the same commit: if it's supported it is thus also per-instance. */
1116 const char *hpv
= ns_info
->protect_proc
== PROTECT_PROC_DEFAULT
?
1118 protect_proc_to_string(ns_info
->protect_proc
);
1120 /* hidepid= support was added in 5.8, so we can use fsconfig()/fsopen() (which were added in
1121 * 5.2) to check if hidepid= is supported. This avoids a noisy dmesg log by the kernel when
1122 * trying to use hidepid= on systems where it isn't supported. The same applies for subset=.
1123 * fsopen()/fsconfig() was also backported on some distros which allows us to detect
1124 * hidepid=/subset= support in even more scenarios. */
1126 if (mount_option_supported("proc", "hidepid", hpv
) != 0) {
1127 opts
= strjoin("hidepid=", hpv
);
1132 if (ns_info
->proc_subset
== PROC_SUBSET_PID
&&
1133 mount_option_supported("proc", "subset", "pid") != 0)
1134 if (!strextend_with_separator(&opts
, ",", "subset=pid"))
1138 entry_path
= mount_entry_path(m
);
1139 (void) mkdir_p_label(entry_path
, 0755);
1141 /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in
1142 * one. i.e we don't reuse existing mounts here under any condition, we want a new instance owned by
1143 * our user namespace and with our hidepid= settings applied. Hence, let's get rid of everything
1144 * mounted on /proc/ first. */
1146 n
= umount_recursive(entry_path
, 0);
1148 r
= mount_nofollow_verbose(LOG_DEBUG
, "proc", entry_path
, "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, opts
);
1149 if (r
== -EINVAL
&& opts
)
1150 /* If this failed with EINVAL then this likely means the textual hidepid= stuff is
1151 * not supported by the kernel, and thus the per-instance hidepid= neither, which
1152 * means we really don't want to use it, since it would affect our host's /proc
1153 * mount. Hence let's gracefully fallback to a classic, unrestricted version. */
1154 r
= mount_nofollow_verbose(LOG_DEBUG
, "proc", entry_path
, "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
1156 /* When we do not have enough privileges to mount /proc, fallback to use existing /proc. */
1159 /* /proc or some of sub-mounts are umounted in the above. Refuse incomplete tree.
1160 * Propagate the original error code returned by mount() in the above. */
1163 r
= path_is_mount_point(entry_path
, NULL
, 0);
1165 return log_debug_errno(r
, "Unable to determine whether /proc is already mounted: %m");
1167 /* We lack permissions to mount a new instance of /proc, and it is not already
1168 * mounted. But we can access the host's, so as a final fallback bind-mount it to
1169 * the destination, as most likely we are inside a user manager in an unprivileged
1170 * user namespace. */
1171 r
= mount_nofollow_verbose(LOG_DEBUG
, "/proc", entry_path
, NULL
, MS_BIND
|MS_REC
, NULL
);
1178 /* We mounted a new instance now. Let's bind mount the children over now. This matters for
1179 * nspawn where a bunch of files are overmounted, in particular the boot id */
1180 (void) bind_mount_submounts("/proc", entry_path
);
1185 static int mount_tmpfs(const MountEntry
*m
) {
1186 const char *entry_path
, *inner_path
;
1191 entry_path
= mount_entry_path(m
);
1192 inner_path
= mount_entry_unprefixed_path(m
);
1194 /* First, get rid of everything that is below if there is anything. Then, overmount with our new
1197 (void) mkdir_p_label(entry_path
, 0755);
1198 (void) umount_recursive(entry_path
, 0);
1200 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", entry_path
, "tmpfs", m
->flags
, mount_entry_options(m
));
1204 r
= label_fix_full(AT_FDCWD
, entry_path
, inner_path
, 0);
1206 return log_debug_errno(r
, "Failed to fix label of '%s' as '%s': %m", entry_path
, inner_path
);
1211 static int mount_run(const MountEntry
*m
) {
1216 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
1217 if (r
< 0 && r
!= -ENOENT
)
1218 return log_debug_errno(r
, "Unable to determine whether /run is already mounted: %m");
1219 if (r
> 0) /* make this a NOP if /run is already a mount point */
1222 return mount_tmpfs(m
);
1225 static int mount_mqueuefs(const MountEntry
*m
) {
1227 const char *entry_path
;
1231 entry_path
= mount_entry_path(m
);
1233 (void) mkdir_p_label(entry_path
, 0755);
1234 (void) umount_recursive(entry_path
, 0);
1236 r
= mount_nofollow_verbose(LOG_DEBUG
, "mqueue", entry_path
, "mqueue", m
->flags
, mount_entry_options(m
));
1243 static int mount_image(
1244 const MountEntry
*m
,
1245 const char *root_directory
,
1246 const ImagePolicy
*image_policy
) {
1248 _cleanup_free_
char *host_os_release_id
= NULL
, *host_os_release_version_id
= NULL
,
1249 *host_os_release_level
= NULL
, *extension_name
= NULL
;
1250 _cleanup_strv_free_
char **extension_release
= NULL
;
1251 ImageClass
class = IMAGE_SYSEXT
;
1256 r
= path_extract_filename(mount_entry_source(m
), &extension_name
);
1258 return log_debug_errno(r
, "Failed to extract extension name from %s: %m", mount_entry_source(m
));
1260 if (m
->mode
== EXTENSION_IMAGES
) {
1261 r
= load_extension_release_pairs(mount_entry_source(m
), IMAGE_SYSEXT
, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1263 r
= load_extension_release_pairs(mount_entry_source(m
), IMAGE_CONFEXT
, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1265 class = IMAGE_CONFEXT
;
1270 r
= parse_os_release(
1271 empty_to_root(root_directory
),
1272 "ID", &host_os_release_id
,
1273 "VERSION_ID", &host_os_release_version_id
,
1274 image_class_info
[class].level_env
, &host_os_release_level
,
1277 return log_debug_errno(r
, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory
));
1278 if (isempty(host_os_release_id
))
1279 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
));
1282 r
= verity_dissect_and_mount(
1284 mount_entry_source(m
),
1285 mount_entry_path(m
),
1289 host_os_release_version_id
,
1290 host_os_release_level
,
1292 if (r
== -ENOENT
&& m
->ignore
)
1294 if (r
== -ESTALE
&& host_os_release_id
)
1295 return log_error_errno(r
,
1296 "Failed to mount image %s, extension-release metadata does not match the lower layer's: ID=%s%s%s%s%s",
1297 mount_entry_source(m
),
1299 host_os_release_version_id
? " VERSION_ID=" : "",
1300 strempty(host_os_release_version_id
),
1301 host_os_release_level
? image_class_info
[class].level_env_print
: "",
1302 strempty(host_os_release_level
));
1304 return log_debug_errno(r
, "Failed to mount image %s on %s: %m", mount_entry_source(m
), mount_entry_path(m
));
1309 static int mount_overlay(const MountEntry
*m
) {
1310 const char *options
;
1315 options
= strjoina("lowerdir=", mount_entry_options(m
));
1317 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1319 r
= mount_nofollow_verbose(LOG_DEBUG
, "overlay", mount_entry_path(m
), "overlay", MS_RDONLY
, options
);
1320 if (r
== -ENOENT
&& m
->ignore
)
1328 static int follow_symlink(
1329 const char *root_directory
,
1332 _cleanup_free_
char *target
= NULL
;
1335 /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we
1336 * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at
1337 * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the
1338 * end and already have a fully normalized name. */
1340 r
= chase(mount_entry_path(m
), root_directory
, CHASE_STEP
|CHASE_NONEXISTENT
, &target
, NULL
);
1342 return log_debug_errno(r
, "Failed to chase symlinks '%s': %m", mount_entry_path(m
));
1343 if (r
> 0) /* Reached the end, nothing more to resolve */
1346 if (m
->n_followed
>= CHASE_MAX
) /* put a boundary on things */
1347 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
1348 "Symlink loop on '%s'.",
1349 mount_entry_path(m
));
1351 log_debug("Followed mount entry path symlink %s %s %s.",
1352 mount_entry_path(m
), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT
), target
);
1354 mount_entry_consume_prefix(m
, TAKE_PTR(target
));
1361 static int apply_one_mount(
1362 const char *root_directory
,
1364 const ImagePolicy
*mount_image_policy
,
1365 const ImagePolicy
*extension_image_policy
,
1366 const NamespaceInfo
*ns_info
) {
1368 _cleanup_free_
char *inaccessible
= NULL
;
1369 bool rbind
= true, make
= false;
1376 log_debug("Applying namespace mount on %s", mount_entry_path(m
));
1380 case INACCESSIBLE
: {
1381 _cleanup_free_
char *tmp
= NULL
;
1382 const char *runtime_dir
;
1385 /* First, get rid of everything that is below if there
1386 * is anything... Then, overmount it with an
1387 * inaccessible path. */
1388 (void) umount_recursive(mount_entry_path(m
), 0);
1390 if (lstat(mount_entry_path(m
), &target
) < 0) {
1391 if (errno
== ENOENT
&& m
->ignore
)
1394 return log_debug_errno(errno
, "Failed to lstat() %s to determine what to mount over it: %m",
1395 mount_entry_path(m
));
1399 runtime_dir
= "/run";
1401 if (asprintf(&tmp
, "/run/user/" UID_FMT
, geteuid()) < 0)
1407 r
= mode_to_inaccessible_node(runtime_dir
, target
.st_mode
, &inaccessible
);
1409 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
1410 "File type not supported for inaccessible mounts. Note that symlinks are not allowed");
1411 what
= inaccessible
;
1417 case READWRITE_IMPLICIT
:
1420 r
= path_is_mount_point(mount_entry_path(m
), root_directory
, 0);
1421 if (r
== -ENOENT
&& m
->ignore
)
1424 return log_debug_errno(r
, "Failed to determine whether %s is already a mount point: %m",
1425 mount_entry_path(m
));
1426 if (r
> 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY
1427 * and MS_NOEXEC bits for the mount point if needed. */
1429 /* This isn't a mount point yet, let's make it one. */
1430 what
= mount_entry_path(m
);
1433 case EXTENSION_DIRECTORIES
: {
1434 _cleanup_free_
char *host_os_release_id
= NULL
, *host_os_release_version_id
= NULL
,
1435 *host_os_release_level
= NULL
, *extension_name
= NULL
;
1436 _cleanup_strv_free_
char **extension_release
= NULL
;
1437 ImageClass
class = IMAGE_SYSEXT
;
1439 r
= path_extract_filename(mount_entry_source(m
), &extension_name
);
1441 return log_debug_errno(r
, "Failed to extract extension name from %s: %m", mount_entry_source(m
));
1443 r
= load_extension_release_pairs(mount_entry_source(m
), IMAGE_SYSEXT
, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1445 r
= load_extension_release_pairs(mount_entry_source(m
), IMAGE_CONFEXT
, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1447 class = IMAGE_CONFEXT
;
1452 r
= parse_os_release(
1453 empty_to_root(root_directory
),
1454 "ID", &host_os_release_id
,
1455 "VERSION_ID", &host_os_release_version_id
,
1456 image_class_info
[class].level_env
, &host_os_release_level
,
1459 return log_debug_errno(r
, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory
));
1460 if (isempty(host_os_release_id
))
1461 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
));
1463 r
= load_extension_release_pairs(mount_entry_source(m
), class, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1464 if (r
== -ENOENT
&& m
->ignore
)
1467 return log_debug_errno(r
, "Failed to parse directory %s extension-release metadata: %m", extension_name
);
1469 r
= extension_release_validate(
1472 host_os_release_version_id
,
1473 host_os_release_level
,
1474 /* host_extension_scope */ NULL
, /* Leave empty, we need to accept both system and portable */
1478 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE
), "Directory %s extension-release metadata does not match the root's", extension_name
);
1480 return log_debug_errno(r
, "Failed to compare directory %s extension-release metadata with the root's os-release: %m", extension_name
);
1489 case BIND_MOUNT_RECURSIVE
: {
1490 _cleanup_free_
char *chased
= NULL
;
1492 /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note
1493 * that bind mount source paths are always relative to the host root, hence we pass NULL as
1494 * root directory to chase() here. */
1496 r
= chase(mount_entry_source(m
), NULL
, CHASE_TRAIL_SLASH
, &chased
, NULL
);
1497 if (r
== -ENOENT
&& m
->ignore
) {
1498 log_debug_errno(r
, "Path %s does not exist, ignoring.", mount_entry_source(m
));
1502 return log_debug_errno(r
, "Failed to follow symlinks on %s: %m", mount_entry_source(m
));
1504 log_debug("Followed source symlinks %s %s %s.",
1505 mount_entry_source(m
), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT
), chased
);
1507 free_and_replace(m
->source_malloc
, chased
);
1509 what
= mount_entry_source(m
);
1516 return mount_tmpfs(m
);
1519 case PRIVATE_TMP_READONLY
:
1520 what
= mount_entry_source(m
);
1525 return mount_private_dev(m
);
1528 return mount_bind_dev(m
);
1531 return mount_private_sysfs(m
);
1534 return mount_bind_sysfs(m
);
1537 return mount_procfs(m
, ns_info
);
1540 return mount_run(m
);
1543 return mount_mqueuefs(m
);
1546 return mount_image(m
, NULL
, mount_image_policy
);
1548 case EXTENSION_IMAGES
:
1549 return mount_image(m
, root_directory
, extension_image_policy
);
1552 return mount_overlay(m
);
1555 assert_not_reached();
1560 r
= mount_nofollow_verbose(LOG_DEBUG
, what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
);
1562 bool try_again
= false;
1564 if (r
== -ENOENT
&& make
) {
1567 /* Hmm, either the source or the destination are missing. Let's see if we can create
1568 the destination, then try again. */
1570 (void) mkdir_parents(mount_entry_path(m
), 0755);
1572 q
= make_mount_point_inode_from_path(what
, mount_entry_path(m
), 0755);
1573 if (q
< 0 && q
!= -EEXIST
)
1574 log_error_errno(q
, "Failed to create destination mount point node '%s': %m",
1575 mount_entry_path(m
));
1581 r
= mount_nofollow_verbose(LOG_DEBUG
, what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
);
1583 return log_error_errno(r
, "Failed to mount %s to %s: %m", what
, mount_entry_path(m
));
1586 log_debug("Successfully mounted %s to %s", what
, mount_entry_path(m
));
1590 static int make_read_only(const MountEntry
*m
, char **deny_list
, FILE *proc_self_mountinfo
) {
1591 unsigned long new_flags
= 0, flags_mask
= 0;
1596 assert(proc_self_mountinfo
);
1598 if (mount_entry_read_only(m
) || m
->mode
== PRIVATE_DEV
) {
1599 new_flags
|= MS_RDONLY
;
1600 flags_mask
|= MS_RDONLY
;
1604 new_flags
|= MS_NOSUID
;
1605 flags_mask
|= MS_NOSUID
;
1608 if (flags_mask
== 0) /* No Change? */
1611 /* We generally apply these changes recursively, except for /dev, and the cases we know there's
1612 * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the
1613 * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace
1614 * and running Linux <= 4.17. */
1616 mount_entry_read_only(m
) &&
1617 !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1619 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, deny_list
, proc_self_mountinfo
);
1621 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, proc_self_mountinfo
);
1623 /* Note that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked
1624 * read-only already stays this way. This improves compatibility with container managers, where we
1625 * won't attempt to undo read-only mounts already applied. */
1627 if (r
== -ENOENT
&& m
->ignore
)
1630 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1631 submounts
? " and its submounts" : "");
1635 static int make_noexec(const MountEntry
*m
, char **deny_list
, FILE *proc_self_mountinfo
) {
1636 unsigned long new_flags
= 0, flags_mask
= 0;
1641 assert(proc_self_mountinfo
);
1643 if (mount_entry_noexec(m
)) {
1644 new_flags
|= MS_NOEXEC
;
1645 flags_mask
|= MS_NOEXEC
;
1646 } else if (mount_entry_exec(m
)) {
1647 new_flags
&= ~MS_NOEXEC
;
1648 flags_mask
|= MS_NOEXEC
;
1651 if (flags_mask
== 0) /* No Change? */
1654 submounts
= !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1657 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, deny_list
, proc_self_mountinfo
);
1659 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, proc_self_mountinfo
);
1661 if (r
== -ENOENT
&& m
->ignore
)
1664 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1665 submounts
? " and its submounts" : "");
1669 static int make_nosuid(const MountEntry
*m
, FILE *proc_self_mountinfo
) {
1674 assert(proc_self_mountinfo
);
1676 submounts
= !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1679 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), MS_NOSUID
, MS_NOSUID
, NULL
, proc_self_mountinfo
);
1681 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), MS_NOSUID
, MS_NOSUID
, proc_self_mountinfo
);
1682 if (r
== -ENOENT
&& m
->ignore
)
1685 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1686 submounts
? " and its submounts" : "");
1690 static bool namespace_info_mount_apivfs(const NamespaceInfo
*ns_info
) {
1694 * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
1695 * since to protect the API VFS mounts, they need to be around in the
1699 return ns_info
->mount_apivfs
||
1700 ns_info
->protect_control_groups
||
1701 ns_info
->protect_kernel_tunables
||
1702 ns_info
->protect_proc
!= PROTECT_PROC_DEFAULT
||
1703 ns_info
->proc_subset
!= PROC_SUBSET_ALL
;
1706 static size_t namespace_calculate_mounts(
1707 const NamespaceInfo
*ns_info
,
1708 char** read_write_paths
,
1709 char** read_only_paths
,
1710 char** inaccessible_paths
,
1712 char** no_exec_paths
,
1713 char** empty_directories
,
1714 size_t n_bind_mounts
,
1715 size_t n_temporary_filesystems
,
1716 size_t n_mount_images
,
1717 size_t n_extension_images
,
1718 size_t n_extension_directories
,
1719 size_t n_hierarchies
,
1720 const char* tmp_dir
,
1721 const char* var_tmp_dir
,
1722 const char *creds_path
,
1723 const char* log_namespace
,
1724 bool setup_propagate
,
1725 const char* notify_socket
,
1726 const char* host_os_release
) {
1728 size_t protect_home_cnt
;
1729 size_t protect_system_cnt
=
1730 (ns_info
->protect_system
== PROTECT_SYSTEM_STRICT
?
1731 ELEMENTSOF(protect_system_strict_table
) :
1732 ((ns_info
->protect_system
== PROTECT_SYSTEM_FULL
) ?
1733 ELEMENTSOF(protect_system_full_table
) :
1734 ((ns_info
->protect_system
== PROTECT_SYSTEM_YES
) ?
1735 ELEMENTSOF(protect_system_yes_table
) : 0)));
1738 (ns_info
->protect_home
== PROTECT_HOME_YES
?
1739 ELEMENTSOF(protect_home_yes_table
) :
1740 ((ns_info
->protect_home
== PROTECT_HOME_READ_ONLY
) ?
1741 ELEMENTSOF(protect_home_read_only_table
) :
1742 ((ns_info
->protect_home
== PROTECT_HOME_TMPFS
) ?
1743 ELEMENTSOF(protect_home_tmpfs_table
) : 0)));
1745 return !!tmp_dir
+ !!var_tmp_dir
+
1746 strv_length(read_write_paths
) +
1747 strv_length(read_only_paths
) +
1748 strv_length(inaccessible_paths
) +
1749 strv_length(exec_paths
) +
1750 strv_length(no_exec_paths
) +
1751 strv_length(empty_directories
) +
1754 (n_extension_images
> 0 || n_extension_directories
> 0 ? /* Mount each image and directory plus an overlay per hierarchy */
1755 n_hierarchies
+ n_extension_images
+ n_extension_directories
: 0) +
1756 n_temporary_filesystems
+
1757 ns_info
->private_dev
+
1758 (ns_info
->protect_kernel_tunables
?
1759 ELEMENTSOF(protect_kernel_tunables_proc_table
) + ELEMENTSOF(protect_kernel_tunables_sys_table
) : 0) +
1760 (ns_info
->protect_kernel_modules
? ELEMENTSOF(protect_kernel_modules_table
) : 0) +
1761 (ns_info
->protect_kernel_logs
?
1762 ELEMENTSOF(protect_kernel_logs_proc_table
) + ELEMENTSOF(protect_kernel_logs_dev_table
) : 0) +
1763 (ns_info
->protect_control_groups
? 1 : 0) +
1764 protect_home_cnt
+ protect_system_cnt
+
1765 (ns_info
->protect_hostname
? 2 : 0) +
1766 (namespace_info_mount_apivfs(ns_info
) ? ELEMENTSOF(apivfs_table
) : 0) +
1767 (creds_path
? 2 : 1) +
1769 setup_propagate
+ /* /run/systemd/incoming */
1772 ns_info
->private_network
+ /* /sys */
1773 ns_info
->private_ipc
; /* /dev/mqueue */
1776 /* Walk all mount entries and dropping any unused mounts. This affects all
1778 * - that are implicitly protected by a path that has been rendered inaccessible
1779 * - whose immediate parent requests the same protection mode as the mount itself
1780 * - that are outside of the relevant root directory
1781 * - which are duplicates
1783 static void drop_unused_mounts(const char *root_directory
, MountEntry
*mounts
, size_t *n_mounts
) {
1784 assert(root_directory
);
1786 assert(mounts
|| *n_mounts
== 0);
1788 typesafe_qsort(mounts
, *n_mounts
, mount_path_compare
);
1790 drop_duplicates(mounts
, n_mounts
);
1791 drop_outside_root(root_directory
, mounts
, n_mounts
);
1792 drop_inaccessible(mounts
, n_mounts
);
1793 drop_nop(mounts
, n_mounts
);
1796 static int create_symlinks_from_tuples(const char *root
, char **strv_symlinks
) {
1799 STRV_FOREACH_PAIR(src
, dst
, strv_symlinks
) {
1800 _cleanup_free_
char *src_abs
= NULL
, *dst_abs
= NULL
;
1802 src_abs
= path_join(root
, *src
);
1803 dst_abs
= path_join(root
, *dst
);
1804 if (!src_abs
|| !dst_abs
)
1807 r
= mkdir_parents_label(dst_abs
, 0755);
1811 r
= symlink_idempotent(src_abs
, dst_abs
, true);
1819 static int apply_mounts(
1821 const ImagePolicy
*mount_image_policy
,
1822 const ImagePolicy
*extension_image_policy
,
1823 const NamespaceInfo
*ns_info
,
1827 char **error_path
) {
1829 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
1830 _cleanup_free_
char **deny_list
= NULL
;
1833 if (n_mounts
== 0) /* Shortcut: nothing to do */
1840 /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of
1841 * /proc. For example, this is the case with the option: 'InaccessiblePaths=/proc'. */
1842 proc_self_mountinfo
= fopen("/proc/self/mountinfo", "re");
1843 if (!proc_self_mountinfo
) {
1847 *error_path
= strdup("/proc/self/mountinfo");
1849 return log_debug_errno(r
, "Failed to open /proc/self/mountinfo: %m");
1852 /* First round, establish all mounts we need */
1856 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1861 /* ExtensionImages/Directories are first opened in the propagate directory, not in the root_directory */
1862 r
= follow_symlink(!IN_SET(m
->mode
, EXTENSION_IMAGES
, EXTENSION_DIRECTORIES
) ? root
: NULL
, m
);
1864 if (error_path
&& mount_entry_path(m
))
1865 *error_path
= strdup(mount_entry_path(m
));
1869 /* We hit a symlinked mount point. The entry got rewritten and might
1870 * point to a very different place now. Let's normalize the changed
1871 * list, and start from the beginning. After all to mount the entry
1872 * at the new location we might need some other mounts first */
1877 r
= apply_one_mount(root
, m
, mount_image_policy
, extension_image_policy
, ns_info
);
1879 if (error_path
&& mount_entry_path(m
))
1880 *error_path
= strdup(mount_entry_path(m
));
1890 drop_unused_mounts(root
, mounts
, n_mounts
);
1893 /* Now that all filesystems have been set up, but before the
1894 * read-only switches are flipped, create the exec dirs and other symlinks.
1895 * Note that when /var/lib is not empty/tmpfs, these symlinks will already
1896 * exist, which means this will be a no-op. */
1897 r
= create_symlinks_from_tuples(root
, symlinks
);
1899 return log_debug_errno(r
, "Failed to set up symlinks inside mount namespace: %m");
1901 /* Create a deny list we can pass to bind_mount_recursive() */
1902 deny_list
= new(char*, (*n_mounts
)+1);
1905 for (size_t j
= 0; j
< *n_mounts
; j
++)
1906 deny_list
[j
] = (char*) mount_entry_path(mounts
+j
);
1907 deny_list
[*n_mounts
] = NULL
;
1909 /* Second round, flip the ro bits if necessary. */
1910 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1911 r
= make_read_only(m
, deny_list
, proc_self_mountinfo
);
1913 if (error_path
&& mount_entry_path(m
))
1914 *error_path
= strdup(mount_entry_path(m
));
1919 /* Third round, flip the noexec bits with a simplified deny list. */
1920 for (size_t j
= 0; j
< *n_mounts
; j
++)
1921 if (IN_SET((mounts
+j
)->mode
, EXEC
, NOEXEC
))
1922 deny_list
[j
] = (char*) mount_entry_path(mounts
+j
);
1923 deny_list
[*n_mounts
] = NULL
;
1925 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1926 r
= make_noexec(m
, deny_list
, proc_self_mountinfo
);
1928 if (error_path
&& mount_entry_path(m
))
1929 *error_path
= strdup(mount_entry_path(m
));
1934 /* Fourth round, flip the nosuid bits without a deny list. */
1935 if (ns_info
->mount_nosuid
)
1936 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1937 r
= make_nosuid(m
, proc_self_mountinfo
);
1939 if (error_path
&& mount_entry_path(m
))
1940 *error_path
= strdup(mount_entry_path(m
));
1948 static bool root_read_only(
1949 char **read_only_paths
,
1950 ProtectSystem protect_system
) {
1952 /* Determine whether the root directory is going to be read-only given the configured settings. */
1954 if (protect_system
== PROTECT_SYSTEM_STRICT
)
1957 if (prefixed_path_strv_contains(read_only_paths
, "/"))
1963 static bool home_read_only(
1964 char** read_only_paths
,
1965 char** inaccessible_paths
,
1966 char** empty_directories
,
1967 const BindMount
*bind_mounts
,
1968 size_t n_bind_mounts
,
1969 const TemporaryFileSystem
*temporary_filesystems
,
1970 size_t n_temporary_filesystems
,
1971 ProtectHome protect_home
) {
1973 /* Determine whether the /home directory is going to be read-only given the configured settings. Yes,
1974 * this is a bit sloppy, since we don't bother checking for cases where / is affected by multiple
1977 if (protect_home
!= PROTECT_HOME_NO
)
1980 if (prefixed_path_strv_contains(read_only_paths
, "/home") ||
1981 prefixed_path_strv_contains(inaccessible_paths
, "/home") ||
1982 prefixed_path_strv_contains(empty_directories
, "/home"))
1985 for (size_t i
= 0; i
< n_temporary_filesystems
; i
++)
1986 if (path_equal(temporary_filesystems
[i
].path
, "/home"))
1989 /* If /home is overmounted with some dir from the host it's not writable. */
1990 for (size_t i
= 0; i
< n_bind_mounts
; i
++)
1991 if (path_equal(bind_mounts
[i
].destination
, "/home"))
1997 int setup_namespace(
1998 const char* root_directory
,
1999 const char* root_image
,
2000 const MountOptions
*root_image_mount_options
,
2001 const ImagePolicy
*root_image_policy
,
2002 const NamespaceInfo
*ns_info
,
2003 char** read_write_paths
,
2004 char** read_only_paths
,
2005 char** inaccessible_paths
,
2007 char** no_exec_paths
,
2008 char** empty_directories
,
2010 const BindMount
*bind_mounts
,
2011 size_t n_bind_mounts
,
2012 const TemporaryFileSystem
*temporary_filesystems
,
2013 size_t n_temporary_filesystems
,
2014 const MountImage
*mount_images
,
2015 size_t n_mount_images
,
2016 const ImagePolicy
*mount_image_policy
,
2017 const char* tmp_dir
,
2018 const char* var_tmp_dir
,
2019 const char *creds_path
,
2020 const char *log_namespace
,
2021 unsigned long mount_propagation_flag
,
2022 VeritySettings
*verity
,
2023 const MountImage
*extension_images
,
2024 size_t n_extension_images
,
2025 const ImagePolicy
*extension_image_policy
,
2026 char **extension_directories
,
2027 const char *propagate_dir
,
2028 const char *incoming_dir
,
2029 const char *extension_dir
,
2030 const char *notify_socket
,
2031 const char *host_os_release_stage
,
2032 char **error_path
) {
2034 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
2035 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
2036 _cleanup_strv_free_
char **hierarchies
= NULL
;
2037 MountEntry
*m
= NULL
, *mounts
= NULL
;
2038 bool require_prefix
= false, setup_propagate
= false;
2040 DissectImageFlags dissect_image_flags
=
2041 DISSECT_IMAGE_GENERIC_ROOT
|
2042 DISSECT_IMAGE_REQUIRE_ROOT
|
2043 DISSECT_IMAGE_DISCARD_ON_LOOP
|
2044 DISSECT_IMAGE_RELAX_VAR_CHECK
|
2045 DISSECT_IMAGE_FSCK
|
2046 DISSECT_IMAGE_USR_NO_ROOT
|
2047 DISSECT_IMAGE_GROWFS
|
2048 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
2049 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
2055 /* Make sure that all mknod(), mkdir() calls we do are unaffected by the umask, and the access modes
2056 * we configure take effect */
2057 BLOCK_WITH_UMASK(0000);
2059 if (!isempty(propagate_dir
) && !isempty(incoming_dir
))
2060 setup_propagate
= true;
2062 if (mount_propagation_flag
== 0)
2063 mount_propagation_flag
= MS_SHARED
;
2066 /* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */
2067 if (root_read_only(read_only_paths
,
2068 ns_info
->protect_system
) &&
2069 home_read_only(read_only_paths
, inaccessible_paths
, empty_directories
,
2070 bind_mounts
, n_bind_mounts
, temporary_filesystems
, n_temporary_filesystems
,
2071 ns_info
->protect_home
) &&
2072 strv_isempty(read_write_paths
))
2073 dissect_image_flags
|= DISSECT_IMAGE_READ_ONLY
;
2075 SET_FLAG(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
, verity
&& verity
->data_path
);
2077 r
= loop_device_make_by_path(
2079 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_DEVICE_READ_ONLY
) ? O_RDONLY
: -1 /* < 0 means writable if possible, read-only as fallback */,
2080 /* sector_size= */ UINT32_MAX
,
2081 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
2085 return log_debug_errno(r
, "Failed to create loop device for root image: %m");
2087 r
= dissect_loop_device(
2090 root_image_mount_options
,
2092 dissect_image_flags
,
2095 return log_debug_errno(r
, "Failed to dissect image: %m");
2097 r
= dissected_image_load_verity_sig_partition(
2104 r
= dissected_image_decrypt(
2108 dissect_image_flags
);
2110 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
2114 root
= root_directory
;
2116 /* /run/systemd should have been created by PID 1 early on already, but in some cases, like
2117 * when running tests (test-execute), it might not have been created yet so let's make sure
2118 * we create it if it doesn't already exist. */
2119 (void) mkdir_p_label("/run/systemd", 0755);
2121 /* Always create the mount namespace in a temporary directory, instead of operating directly
2122 * in the root. The temporary directory prevents any mounts from being potentially obscured
2123 * my other mounts we already applied. We use the same mount point for all images, which is
2124 * safe, since they all live in their own namespaces after all, and hence won't see each
2125 * other. (Note: this directory is also created by PID 1 early on, we create it here for
2126 * similar reasons as /run/systemd/ first.) */
2127 root
= "/run/systemd/mount-rootfs";
2128 (void) mkdir_label(root
, 0555);
2130 require_prefix
= true;
2133 if (n_extension_images
> 0 || !strv_isempty(extension_directories
)) {
2134 /* Hierarchy population needs to be done for sysext and confext extension images */
2135 r
= parse_env_extension_hierarchies(&hierarchies
, "SYSTEMD_SYSEXT_AND_CONFEXT_HIERARCHIES");
2140 n_mounts
= namespace_calculate_mounts(
2149 n_temporary_filesystems
,
2152 strv_length(extension_directories
),
2153 strv_length(hierarchies
),
2154 tmp_dir
, var_tmp_dir
,
2159 host_os_release_stage
);
2162 m
= mounts
= new0(MountEntry
, n_mounts
);
2166 r
= append_access_mounts(&m
, read_write_paths
, READWRITE
, require_prefix
);
2170 r
= append_access_mounts(&m
, read_only_paths
, READONLY
, require_prefix
);
2174 r
= append_access_mounts(&m
, inaccessible_paths
, INACCESSIBLE
, require_prefix
);
2178 r
= append_access_mounts(&m
, exec_paths
, EXEC
, require_prefix
);
2182 r
= append_access_mounts(&m
, no_exec_paths
, NOEXEC
, require_prefix
);
2186 r
= append_empty_dir_mounts(&m
, empty_directories
);
2190 r
= append_bind_mounts(&m
, bind_mounts
, n_bind_mounts
);
2194 r
= append_tmpfs_mounts(&m
, temporary_filesystems
, n_temporary_filesystems
);
2199 bool ro
= streq(tmp_dir
, RUN_SYSTEMD_EMPTY
);
2201 *(m
++) = (MountEntry
) {
2202 .path_const
= "/tmp",
2203 .mode
= ro
? PRIVATE_TMP_READONLY
: PRIVATE_TMP
,
2204 .source_const
= tmp_dir
,
2209 bool ro
= streq(var_tmp_dir
, RUN_SYSTEMD_EMPTY
);
2211 *(m
++) = (MountEntry
) {
2212 .path_const
= "/var/tmp",
2213 .mode
= ro
? PRIVATE_TMP_READONLY
: PRIVATE_TMP
,
2214 .source_const
= var_tmp_dir
,
2218 r
= append_mount_images(&m
, mount_images
, n_mount_images
);
2222 r
= append_extensions(&m
, root
, extension_dir
, hierarchies
, extension_images
, n_extension_images
, extension_directories
);
2226 if (ns_info
->private_dev
)
2227 *(m
++) = (MountEntry
) {
2228 .path_const
= "/dev",
2229 .mode
= PRIVATE_DEV
,
2230 .flags
= DEV_MOUNT_OPTIONS
,
2233 /* In case /proc is successfully mounted with pid tree subset only (ProcSubset=pid), the
2234 protective mounts to non-pid /proc paths would fail. But the pid only option may have
2235 failed gracefully, so let's try the mounts but it's not fatal if they don't succeed. */
2236 bool ignore_protect_proc
= ns_info
->ignore_protect_paths
|| ns_info
->proc_subset
== PROC_SUBSET_PID
;
2237 if (ns_info
->protect_kernel_tunables
) {
2238 r
= append_static_mounts(&m
,
2239 protect_kernel_tunables_proc_table
,
2240 ELEMENTSOF(protect_kernel_tunables_proc_table
),
2241 ignore_protect_proc
);
2245 r
= append_static_mounts(&m
,
2246 protect_kernel_tunables_sys_table
,
2247 ELEMENTSOF(protect_kernel_tunables_sys_table
),
2248 ns_info
->ignore_protect_paths
);
2253 if (ns_info
->protect_kernel_modules
) {
2254 r
= append_static_mounts(&m
,
2255 protect_kernel_modules_table
,
2256 ELEMENTSOF(protect_kernel_modules_table
),
2257 ns_info
->ignore_protect_paths
);
2262 if (ns_info
->protect_kernel_logs
) {
2263 r
= append_static_mounts(&m
,
2264 protect_kernel_logs_proc_table
,
2265 ELEMENTSOF(protect_kernel_logs_proc_table
),
2266 ignore_protect_proc
);
2270 r
= append_static_mounts(&m
,
2271 protect_kernel_logs_dev_table
,
2272 ELEMENTSOF(protect_kernel_logs_dev_table
),
2273 ns_info
->ignore_protect_paths
);
2278 if (ns_info
->protect_control_groups
)
2279 *(m
++) = (MountEntry
) {
2280 .path_const
= "/sys/fs/cgroup",
2284 r
= append_protect_home(&m
, ns_info
->protect_home
, ns_info
->ignore_protect_paths
);
2288 r
= append_protect_system(&m
, ns_info
->protect_system
, false);
2292 if (namespace_info_mount_apivfs(ns_info
)) {
2293 r
= append_static_mounts(&m
,
2295 ELEMENTSOF(apivfs_table
),
2296 ns_info
->ignore_protect_paths
);
2301 /* Note, if proc is mounted with subset=pid then neither of the
2302 * two paths will exist, i.e. they are implicitly protected by
2303 * the mount option. */
2304 if (ns_info
->protect_hostname
) {
2305 *(m
++) = (MountEntry
) {
2306 .path_const
= "/proc/sys/kernel/hostname",
2308 .ignore
= ignore_protect_proc
,
2310 *(m
++) = (MountEntry
) {
2311 .path_const
= "/proc/sys/kernel/domainname",
2313 .ignore
= ignore_protect_proc
,
2317 if (ns_info
->private_network
)
2318 *(m
++) = (MountEntry
) {
2319 .path_const
= "/sys",
2320 .mode
= PRIVATE_SYSFS
,
2323 if (ns_info
->private_ipc
)
2324 *(m
++) = (MountEntry
) {
2325 .path_const
= "/dev/mqueue",
2327 .flags
= MS_NOSUID
| MS_NODEV
| MS_NOEXEC
| MS_RELATIME
,
2331 /* If our service has a credentials store configured, then bind that one in, but hide
2332 * everything else. */
2334 *(m
++) = (MountEntry
) {
2335 .path_const
= "/run/credentials",
2338 .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
,
2339 .flags
= MS_NODEV
|MS_STRICTATIME
|MS_NOSUID
|MS_NOEXEC
,
2342 *(m
++) = (MountEntry
) {
2343 .path_const
= creds_path
,
2346 .source_const
= creds_path
,
2350 /* If our service has no credentials store configured, then make the whole
2351 * credentials tree inaccessible wholesale. */
2353 *(m
++) = (MountEntry
) {
2354 .path_const
= "/run/credentials",
2355 .mode
= INACCESSIBLE
,
2360 if (log_namespace
) {
2361 _cleanup_free_
char *q
= NULL
;
2363 q
= strjoin("/run/systemd/journal.", log_namespace
);
2369 *(m
++) = (MountEntry
) {
2370 .path_const
= "/run/systemd/journal",
2371 .mode
= BIND_MOUNT_RECURSIVE
,
2373 .source_malloc
= TAKE_PTR(q
),
2377 /* Will be used to add bind mounts at runtime */
2378 if (setup_propagate
)
2379 *(m
++) = (MountEntry
) {
2380 .source_const
= propagate_dir
,
2381 .path_const
= incoming_dir
,
2387 *(m
++) = (MountEntry
) {
2388 .path_const
= notify_socket
,
2389 .source_const
= notify_socket
,
2394 if (host_os_release_stage
)
2395 *(m
++) = (MountEntry
) {
2396 .path_const
= "/run/host/.os-release-stage/",
2397 .source_const
= host_os_release_stage
,
2400 .ignore
= true, /* Live copy, don't hard-fail if it goes missing */
2403 assert(mounts
+ n_mounts
== m
);
2405 /* Prepend the root directory where that's necessary */
2406 r
= prefix_where_needed(mounts
, n_mounts
, root
);
2410 drop_unused_mounts(root
, mounts
, &n_mounts
);
2413 /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */
2415 if (unshare(CLONE_NEWNS
) < 0) {
2416 r
= log_debug_errno(errno
, "Failed to unshare the mount namespace: %m");
2417 if (ERRNO_IS_PRIVILEGE(r
) ||
2418 ERRNO_IS_NOT_SUPPORTED(r
))
2419 /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter
2420 * in place that doesn't allow us to create namespaces (or a missing cap), then
2421 * propagate a recognizable error back, which the caller can use to detect this case
2422 * (and only this) and optionally continue without namespacing applied. */
2428 /* Create the source directory to allow runtime propagation of mounts */
2429 if (setup_propagate
)
2430 (void) mkdir_p(propagate_dir
, 0600);
2432 if (n_extension_images
> 0 || !strv_isempty(extension_directories
))
2433 /* ExtensionImages/Directories mountpoint directories will be created while parsing the
2434 * mounts to create, so have the parent ready */
2435 (void) mkdir_p(extension_dir
, 0600);
2437 /* Remount / as SLAVE so that nothing now mounted in the namespace
2438 * shows up in the parent */
2439 if (mount(NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
) < 0) {
2440 r
= log_debug_errno(errno
, "Failed to remount '/' as SLAVE: %m");
2445 /* A root image is specified, mount it to the right place */
2446 r
= dissected_image_mount(dissected_image
, root
, UID_INVALID
, UID_INVALID
, dissect_image_flags
);
2448 log_debug_errno(r
, "Failed to mount root image: %m");
2452 /* Now release the block device lock, so that udevd is free to call BLKRRPART on the device
2454 r
= loop_device_flock(loop_device
, LOCK_UN
);
2456 log_debug_errno(r
, "Failed to release lock on loopback block device: %m");
2460 r
= dissected_image_relinquish(dissected_image
);
2462 log_debug_errno(r
, "Failed to relinquish dissected image: %m");
2466 } else if (root_directory
) {
2468 /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
2469 r
= path_is_mount_point(root
, NULL
, AT_SYMLINK_FOLLOW
);
2471 log_debug_errno(r
, "Failed to detect that %s is a mount point or not: %m", root
);
2475 r
= mount_nofollow_verbose(LOG_DEBUG
, root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2481 /* Let's mount the main root directory to the root directory to use */
2482 r
= mount_nofollow_verbose(LOG_DEBUG
, "/", root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2487 /* Try to set up the new root directory before mounting anything else there. */
2488 if (root_image
|| root_directory
)
2489 (void) base_filesystem_create(root
, UID_INVALID
, GID_INVALID
);
2491 /* Now make the magic happen */
2492 r
= apply_mounts(root
, mount_image_policy
, extension_image_policy
, ns_info
, mounts
, &n_mounts
, symlinks
, error_path
);
2496 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
2497 r
= mount_switch_root(root
, /* mount_propagation_flag = */ 0);
2498 if (r
== -EINVAL
&& root_directory
) {
2499 /* If we are using root_directory and we don't have privileges (ie: user manager in a user
2500 * namespace) and the root_directory is already a mount point in the parent namespace,
2501 * MS_MOVE will fail as we don't have permission to change it (with EINVAL rather than
2502 * EPERM). Attempt to bind-mount it over itself (like we do above if it's not already a
2503 * mount point) and try again. */
2504 r
= mount_nofollow_verbose(LOG_DEBUG
, root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2507 r
= mount_switch_root(root
, /* mount_propagation_flag = */ 0);
2510 log_debug_errno(r
, "Failed to mount root with MS_MOVE: %m");
2514 /* Remount / as the desired mode. Note that this will not reestablish propagation from our side to
2515 * the host, since what's disconnected is disconnected. */
2516 if (mount(NULL
, "/", NULL
, mount_propagation_flag
| MS_REC
, NULL
) < 0) {
2517 r
= log_debug_errno(errno
, "Failed to remount '/' with desired mount flags: %m");
2521 /* bind_mount_in_namespace() will MS_MOVE into that directory, and that's only
2522 * supported for non-shared mounts. This needs to happen after remounting / or it will fail. */
2523 if (setup_propagate
) {
2524 r
= mount(NULL
, incoming_dir
, NULL
, MS_SLAVE
, NULL
);
2526 log_error_errno(r
, "Failed to remount %s with MS_SLAVE: %m", incoming_dir
);
2535 for (m
= mounts
; m
< mounts
+ n_mounts
; m
++)
2536 mount_entry_done(m
);
2543 void bind_mount_free_many(BindMount
*b
, size_t n
) {
2544 assert(b
|| n
== 0);
2546 for (size_t i
= 0; i
< n
; i
++) {
2548 free(b
[i
].destination
);
2554 int bind_mount_add(BindMount
**b
, size_t *n
, const BindMount
*item
) {
2555 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
2562 s
= strdup(item
->source
);
2566 d
= strdup(item
->destination
);
2570 c
= reallocarray(*b
, *n
+ 1, sizeof(BindMount
));
2576 c
[(*n
) ++] = (BindMount
) {
2577 .source
= TAKE_PTR(s
),
2578 .destination
= TAKE_PTR(d
),
2579 .read_only
= item
->read_only
,
2580 .nosuid
= item
->nosuid
,
2581 .recursive
= item
->recursive
,
2582 .ignore_enoent
= item
->ignore_enoent
,
2588 MountImage
* mount_image_free_many(MountImage
*m
, size_t *n
) {
2590 assert(m
|| *n
== 0);
2592 for (size_t i
= 0; i
< *n
; i
++) {
2594 free(m
[i
].destination
);
2595 mount_options_free_all(m
[i
].mount_options
);
2603 int mount_image_add(MountImage
**m
, size_t *n
, const MountImage
*item
) {
2604 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
2605 _cleanup_(mount_options_free_allp
) MountOptions
*options
= NULL
;
2612 s
= strdup(item
->source
);
2616 if (item
->destination
) {
2617 d
= strdup(item
->destination
);
2622 LIST_FOREACH(mount_options
, i
, item
->mount_options
) {
2623 _cleanup_(mount_options_free_allp
) MountOptions
*o
= NULL
;
2625 o
= new(MountOptions
, 1);
2629 *o
= (MountOptions
) {
2630 .partition_designator
= i
->partition_designator
,
2631 .options
= strdup(i
->options
),
2636 LIST_APPEND(mount_options
, options
, TAKE_PTR(o
));
2639 c
= reallocarray(*m
, *n
+ 1, sizeof(MountImage
));
2645 c
[(*n
) ++] = (MountImage
) {
2646 .source
= TAKE_PTR(s
),
2647 .destination
= TAKE_PTR(d
),
2648 .mount_options
= TAKE_PTR(options
),
2649 .ignore_enoent
= item
->ignore_enoent
,
2656 void temporary_filesystem_free_many(TemporaryFileSystem
*t
, size_t n
) {
2657 assert(t
|| n
== 0);
2659 for (size_t i
= 0; i
< n
; i
++) {
2667 int temporary_filesystem_add(
2668 TemporaryFileSystem
**t
,
2671 const char *options
) {
2673 _cleanup_free_
char *p
= NULL
, *o
= NULL
;
2674 TemporaryFileSystem
*c
;
2684 if (!isempty(options
)) {
2685 o
= strdup(options
);
2690 c
= reallocarray(*t
, *n
+ 1, sizeof(TemporaryFileSystem
));
2696 c
[(*n
) ++] = (TemporaryFileSystem
) {
2697 .path
= TAKE_PTR(p
),
2698 .options
= TAKE_PTR(o
),
2704 static int make_tmp_prefix(const char *prefix
) {
2705 _cleanup_free_
char *t
= NULL
;
2706 _cleanup_close_
int fd
= -EBADF
;
2709 /* Don't do anything unless we know the dir is actually missing */
2710 r
= access(prefix
, F_OK
);
2713 if (errno
!= ENOENT
)
2717 r
= mkdir_parents(prefix
, 0755);
2721 r
= tempfn_random(prefix
, NULL
, &t
);
2725 /* umask will corrupt this access mode, but that doesn't matter, we need to call chmod() anyway for
2726 * the suid bit, below. */
2727 fd
= open_mkdir_at(AT_FDCWD
, t
, O_EXCL
|O_CLOEXEC
, 0777);
2731 r
= RET_NERRNO(fchmod(fd
, 01777));
2737 r
= RET_NERRNO(rename(t
, prefix
));
2740 return r
== -EEXIST
? 0 : r
; /* it's fine if someone else created the dir by now */
2747 static int setup_one_tmp_dir(const char *id
, const char *prefix
, char **path
, char **tmp_path
) {
2748 _cleanup_free_
char *x
= NULL
;
2749 _cleanup_free_
char *y
= NULL
;
2758 /* We include the boot id in the directory so that after a
2759 * reboot we can easily identify obsolete directories. */
2761 r
= sd_id128_get_boot(&boot_id
);
2765 x
= strjoin(prefix
, "/systemd-private-", SD_ID128_TO_STRING(boot_id
), "-", id
, "-XXXXXX");
2769 r
= make_tmp_prefix(prefix
);
2775 if (errno
== EROFS
|| ERRNO_IS_DISK_SPACE(errno
))
2782 y
= strjoin(x
, "/tmp");
2787 if (mkdir(y
, 0777 | S_ISVTX
) < 0)
2790 r
= label_fix_full(AT_FDCWD
, y
, prefix
, 0);
2795 *tmp_path
= TAKE_PTR(y
);
2797 /* Trouble: we failed to create the directory. Instead of failing, let's simulate /tmp being
2798 * read-only. This way the service will get the EROFS result as if it was writing to the real
2801 r
= mkdir_p(RUN_SYSTEMD_EMPTY
, 0500);
2805 r
= free_and_strdup(&x
, RUN_SYSTEMD_EMPTY
);
2810 *path
= TAKE_PTR(x
);
2814 int setup_tmp_dirs(const char *id
, char **tmp_dir
, char **var_tmp_dir
) {
2815 _cleanup_(namespace_cleanup_tmpdirp
) char *a
= NULL
;
2816 _cleanup_(rmdir_and_freep
) char *a_tmp
= NULL
;
2822 assert(var_tmp_dir
);
2824 r
= setup_one_tmp_dir(id
, "/tmp", &a
, &a_tmp
);
2828 r
= setup_one_tmp_dir(id
, "/var/tmp", &b
, NULL
);
2832 a_tmp
= mfree(a_tmp
); /* avoid rmdir */
2833 *tmp_dir
= TAKE_PTR(a
);
2834 *var_tmp_dir
= TAKE_PTR(b
);
2839 int setup_shareable_ns(int ns_storage_socket
[static 2], unsigned long nsflag
) {
2840 _cleanup_close_
int ns
= -EBADF
;
2842 const char *ns_name
, *ns_path
;
2844 assert(ns_storage_socket
);
2845 assert(ns_storage_socket
[0] >= 0);
2846 assert(ns_storage_socket
[1] >= 0);
2848 ns_name
= namespace_single_flag_to_string(nsflag
);
2851 /* We use the passed socketpair as a storage buffer for our
2852 * namespace reference fd. Whatever process runs this first
2853 * shall create a new namespace, all others should just join
2854 * it. To serialize that we use a file lock on the socket
2857 * It's a bit crazy, but hey, works great! */
2859 r
= posix_lock(ns_storage_socket
[0], LOCK_EX
);
2863 CLEANUP_POSIX_UNLOCK(ns_storage_socket
[0]);
2865 ns
= receive_one_fd(ns_storage_socket
[0], MSG_PEEK
|MSG_DONTWAIT
);
2867 /* Yay, found something, so let's join the namespace */
2868 r
= RET_NERRNO(setns(ns
, nsflag
));
2878 /* Nothing stored yet, so let's create a new namespace. */
2880 if (unshare(nsflag
) < 0)
2883 (void) loopback_setup();
2885 ns_path
= strjoina("/proc/self/ns/", ns_name
);
2886 ns
= open(ns_path
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
2890 r
= send_one_fd(ns_storage_socket
[1], ns
, MSG_DONTWAIT
);
2897 int open_shareable_ns_path(int ns_storage_socket
[static 2], const char *path
, unsigned long nsflag
) {
2898 _cleanup_close_
int ns
= -EBADF
;
2901 assert(ns_storage_socket
);
2902 assert(ns_storage_socket
[0] >= 0);
2903 assert(ns_storage_socket
[1] >= 0);
2906 /* If the storage socket doesn't contain a ns fd yet, open one via the file system and store it in
2907 * it. This is supposed to be called ahead of time, i.e. before setup_shareable_ns() which will
2908 * allocate a new anonymous ns if needed. */
2910 r
= posix_lock(ns_storage_socket
[0], LOCK_EX
);
2914 CLEANUP_POSIX_UNLOCK(ns_storage_socket
[0]);
2916 ns
= receive_one_fd(ns_storage_socket
[0], MSG_PEEK
|MSG_DONTWAIT
);
2922 /* Nothing stored yet. Open the file from the file system. */
2924 ns
= open(path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
2928 r
= fd_is_ns(ns
, nsflag
);
2931 if (r
< 0 && r
!= -EUCLEAN
) /* EUCLEAN: we don't know */
2934 r
= send_one_fd(ns_storage_socket
[1], ns
, MSG_DONTWAIT
);
2941 bool ns_type_supported(NamespaceType type
) {
2942 const char *t
, *ns_proc
;
2944 t
= namespace_type_to_string(type
);
2945 if (!t
) /* Don't know how to translate this? Then it's not supported */
2948 ns_proc
= strjoina("/proc/self/ns/", t
);
2949 return access(ns_proc
, F_OK
) == 0;
2952 static const char *const protect_home_table
[_PROTECT_HOME_MAX
] = {
2953 [PROTECT_HOME_NO
] = "no",
2954 [PROTECT_HOME_YES
] = "yes",
2955 [PROTECT_HOME_READ_ONLY
] = "read-only",
2956 [PROTECT_HOME_TMPFS
] = "tmpfs",
2959 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home
, ProtectHome
, PROTECT_HOME_YES
);
2961 static const char *const protect_system_table
[_PROTECT_SYSTEM_MAX
] = {
2962 [PROTECT_SYSTEM_NO
] = "no",
2963 [PROTECT_SYSTEM_YES
] = "yes",
2964 [PROTECT_SYSTEM_FULL
] = "full",
2965 [PROTECT_SYSTEM_STRICT
] = "strict",
2968 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system
, ProtectSystem
, PROTECT_SYSTEM_YES
);
2970 static const char* const namespace_type_table
[] = {
2971 [NAMESPACE_MOUNT
] = "mnt",
2972 [NAMESPACE_CGROUP
] = "cgroup",
2973 [NAMESPACE_UTS
] = "uts",
2974 [NAMESPACE_IPC
] = "ipc",
2975 [NAMESPACE_USER
] = "user",
2976 [NAMESPACE_PID
] = "pid",
2977 [NAMESPACE_NET
] = "net",
2978 [NAMESPACE_TIME
] = "time",
2981 DEFINE_STRING_TABLE_LOOKUP(namespace_type
, NamespaceType
);
2983 static const char* const protect_proc_table
[_PROTECT_PROC_MAX
] = {
2984 [PROTECT_PROC_DEFAULT
] = "default",
2985 [PROTECT_PROC_NOACCESS
] = "noaccess",
2986 [PROTECT_PROC_INVISIBLE
] = "invisible",
2987 [PROTECT_PROC_PTRACEABLE
] = "ptraceable",
2990 DEFINE_STRING_TABLE_LOOKUP(protect_proc
, ProtectProc
);
2992 static const char* const proc_subset_table
[_PROC_SUBSET_MAX
] = {
2993 [PROC_SUBSET_ALL
] = "all",
2994 [PROC_SUBSET_PID
] = "pid",
2997 DEFINE_STRING_TABLE_LOOKUP(proc_subset
, ProcSubset
);