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 2nd lowest priority. */
78 MKDIR
, /* Should have the lowest priority. */
82 typedef struct MountEntry
{
83 const char *path_const
; /* Memory allocated on stack or static */
85 bool ignore
:1; /* Ignore if path does not exist? */
86 bool has_prefix
:1; /* Already is prefixed by the root dir? */
87 bool read_only
:1; /* Shall this mount point be read-only? */
88 bool nosuid
:1; /* Shall set MS_NOSUID on the mount itself */
89 bool noexec
:1; /* Shall set MS_NOEXEC on the mount itself */
90 bool exec
:1; /* Shall clear MS_NOEXEC on the mount itself */
91 bool applied
:1; /* Already applied */
92 char *path_malloc
; /* Use this instead of 'path_const' if we had to allocate memory */
93 const char *unprefixed_path_const
; /* If the path was amended with a prefix, these will save the original */
94 char *unprefixed_path_malloc
;
95 const char *source_const
; /* The source path, for bind mounts or images */
97 const char *options_const
;/* Mount options for tmpfs */
99 unsigned long flags
; /* Mount flags used by EMPTY_DIR and TMPFS. Do not include MS_RDONLY here, but please use read_only. */
101 LIST_HEAD(MountOptions
, image_options
);
104 /* 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
105 * something there already. These mounts are hence overridden by any other explicitly configured mounts. */
106 static const MountEntry apivfs_table
[] = {
107 { "/proc", PROCFS
, false },
108 { "/dev", BIND_DEV
, false },
109 { "/sys", BIND_SYSFS
, false },
110 { "/run", RUN
, false, .options_const
= "mode=0755" TMPFS_LIMITS_RUN
, .flags
= MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
},
113 /* ProtectKernelTunables= option and the related filesystem APIs */
114 static const MountEntry protect_kernel_tunables_proc_table
[] = {
115 { "/proc/acpi", READONLY
, true },
116 { "/proc/apm", READONLY
, true }, /* Obsolete API, there's no point in permitting access to this, ever */
117 { "/proc/asound", READONLY
, true },
118 { "/proc/bus", READONLY
, true },
119 { "/proc/fs", READONLY
, true },
120 { "/proc/irq", READONLY
, true },
121 { "/proc/kallsyms", INACCESSIBLE
, true },
122 { "/proc/kcore", INACCESSIBLE
, true },
123 { "/proc/latency_stats", READONLY
, true },
124 { "/proc/mtrr", READONLY
, true },
125 { "/proc/scsi", READONLY
, true },
126 { "/proc/sys", READONLY
, true },
127 { "/proc/sysrq-trigger", READONLY
, true },
128 { "/proc/timer_stats", READONLY
, true },
131 static const MountEntry protect_kernel_tunables_sys_table
[] = {
132 { "/sys", READONLY
, false },
133 { "/sys/fs/bpf", READONLY
, true },
134 { "/sys/fs/cgroup", READWRITE_IMPLICIT
, false }, /* READONLY is set by ProtectControlGroups= option */
135 { "/sys/fs/selinux", READWRITE_IMPLICIT
, true },
136 { "/sys/kernel/debug", READONLY
, true },
137 { "/sys/kernel/tracing", READONLY
, true },
140 /* ProtectKernelModules= option */
141 static const MountEntry protect_kernel_modules_table
[] = {
142 { "/usr/lib/modules", INACCESSIBLE
, true },
145 /* ProtectKernelLogs= option */
146 static const MountEntry protect_kernel_logs_proc_table
[] = {
147 { "/proc/kmsg", INACCESSIBLE
, true },
150 static const MountEntry protect_kernel_logs_dev_table
[] = {
151 { "/dev/kmsg", INACCESSIBLE
, true },
155 * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of
156 * system should be protected by ProtectSystem=
158 static const MountEntry protect_home_read_only_table
[] = {
159 { "/home", READONLY
, true },
160 { "/run/user", READONLY
, true },
161 { "/root", READONLY
, true },
164 /* ProtectHome=tmpfs table */
165 static const MountEntry protect_home_tmpfs_table
[] = {
166 { "/home", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
167 { "/run/user", TMPFS
, true, .read_only
= true, .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
168 { "/root", TMPFS
, true, .read_only
= true, .options_const
= "mode=0700" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
171 /* ProtectHome=yes table */
172 static const MountEntry protect_home_yes_table
[] = {
173 { "/home", INACCESSIBLE
, true },
174 { "/run/user", INACCESSIBLE
, true },
175 { "/root", INACCESSIBLE
, true },
178 /* ProtectSystem=yes table */
179 static const MountEntry protect_system_yes_table
[] = {
180 { "/usr", READONLY
, false },
181 { "/boot", READONLY
, true },
182 { "/efi", READONLY
, true },
185 /* ProtectSystem=full includes ProtectSystem=yes */
186 static const MountEntry protect_system_full_table
[] = {
187 { "/usr", READONLY
, false },
188 { "/boot", READONLY
, true },
189 { "/efi", READONLY
, true },
190 { "/etc", READONLY
, false },
194 * ProtectSystem=strict table. In this strict mode, we mount everything
195 * read-only, except for /proc, /dev, /sys which are the kernel API VFS,
196 * which are left writable, but PrivateDevices= + ProtectKernelTunables=
197 * protect those, and these options should be fully orthogonal.
198 * (And of course /home and friends are also left writable, as ProtectHome=
199 * shall manage those, orthogonally).
201 static const MountEntry protect_system_strict_table
[] = {
202 { "/", READONLY
, false },
203 { "/proc", READWRITE_IMPLICIT
, false }, /* ProtectKernelTunables= */
204 { "/sys", READWRITE_IMPLICIT
, false }, /* ProtectKernelTunables= */
205 { "/dev", READWRITE_IMPLICIT
, false }, /* PrivateDevices= */
206 { "/home", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
207 { "/run/user", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
208 { "/root", READWRITE_IMPLICIT
, true }, /* ProtectHome= */
211 static const char * const mount_mode_table
[_MOUNT_MODE_MAX
] = {
212 [INACCESSIBLE
] = "inaccessible",
213 [OVERLAY_MOUNT
] = "overlay",
214 [MOUNT_IMAGES
] = "mount-images",
215 [BIND_MOUNT
] = "bind",
216 [BIND_MOUNT_RECURSIVE
] = "rbind",
217 [PRIVATE_TMP
] = "private-tmp",
218 [PRIVATE_TMP_READONLY
] = "private-tmp-read-only",
219 [PRIVATE_DEV
] = "private-dev",
220 [BIND_DEV
] = "bind-dev",
221 [EMPTY_DIR
] = "empty",
222 [PRIVATE_SYSFS
] = "private-sysfs",
223 [BIND_SYSFS
] = "bind-sysfs",
225 [READONLY
] = "read-only",
226 [READWRITE
] = "read-write",
231 [EXTENSION_DIRECTORIES
] = "extension-directories",
232 [EXTENSION_IMAGES
] = "extension-images",
233 [MQUEUEFS
] = "mqueuefs",
234 [READWRITE_IMPLICIT
] = "read-write-implicit",
238 /* Helper struct for naming simplicity and reusability */
239 static const struct {
240 const char *level_env
;
241 const char *level_env_print
;
242 } image_class_info
[_IMAGE_CLASS_MAX
] = {
244 .level_env
= "SYSEXT_LEVEL",
245 .level_env_print
= " SYSEXT_LEVEL=",
248 .level_env
= "CONFEXT_LEVEL",
249 .level_env_print
= " CONFEXT_LEVEL=",
253 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(mount_mode
, MountMode
);
255 static const char *mount_entry_path(const MountEntry
*p
) {
258 /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that,
259 * otherwise the stack/static ->path field is returned. */
261 return p
->path_malloc
?: p
->path_const
;
264 static const char *mount_entry_unprefixed_path(const MountEntry
*p
) {
267 /* Returns the unprefixed path (ie: before prefix_where_needed() ran), if any */
269 return p
->unprefixed_path_malloc
?: p
->unprefixed_path_const
?: mount_entry_path(p
);
272 static void mount_entry_consume_prefix(MountEntry
*p
, char *new_path
) {
274 assert(p
->path_malloc
|| p
->path_const
);
277 /* Saves current path in unprefixed_ variable, and takes over new_path */
279 free_and_replace(p
->unprefixed_path_malloc
, p
->path_malloc
);
280 /* If we didn't have a path on the heap, then it's a static one */
281 if (!p
->unprefixed_path_malloc
)
282 p
->unprefixed_path_const
= p
->path_const
;
283 p
->path_malloc
= new_path
;
284 p
->has_prefix
= true;
287 static bool mount_entry_read_only(const MountEntry
*p
) {
290 return p
->read_only
|| IN_SET(p
->mode
, READONLY
, INACCESSIBLE
, PRIVATE_TMP_READONLY
);
293 static bool mount_entry_noexec(const MountEntry
*p
) {
296 return p
->noexec
|| IN_SET(p
->mode
, NOEXEC
, INACCESSIBLE
, PRIVATE_SYSFS
, BIND_SYSFS
, PROCFS
);
299 static bool mount_entry_exec(const MountEntry
*p
) {
302 return p
->exec
|| p
->mode
== EXEC
;
305 static const char *mount_entry_source(const MountEntry
*p
) {
308 return p
->source_malloc
?: p
->source_const
;
311 static const char *mount_entry_options(const MountEntry
*p
) {
314 return p
->options_malloc
?: p
->options_const
;
317 static void mount_entry_done(MountEntry
*p
) {
320 p
->path_malloc
= mfree(p
->path_malloc
);
321 p
->unprefixed_path_malloc
= mfree(p
->unprefixed_path_malloc
);
322 p
->source_malloc
= mfree(p
->source_malloc
);
323 p
->options_malloc
= mfree(p
->options_malloc
);
324 p
->image_options
= mount_options_free_all(p
->image_options
);
327 static int append_access_mounts(MountEntry
**p
, char **strv
, MountMode mode
, bool forcibly_require_prefix
) {
330 /* Adds a list of user-supplied READWRITE/READWRITE_IMPLICIT/READONLY/INACCESSIBLE entries */
332 STRV_FOREACH(i
, strv
) {
333 bool ignore
= false, needs_prefix
= false;
336 /* Look for any prefixes */
337 if (startswith(e
, "-")) {
341 if (startswith(e
, "+")) {
346 if (!path_is_absolute(e
))
347 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
348 "Path is not absolute: %s", e
);
350 *((*p
)++) = (MountEntry
) {
354 .has_prefix
= !needs_prefix
&& !forcibly_require_prefix
,
361 static int append_empty_dir_mounts(MountEntry
**p
, char **strv
) {
364 /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the
365 * "/private/" boundary directories for DynamicUser=1. */
367 STRV_FOREACH(i
, strv
) {
369 *((*p
)++) = (MountEntry
) {
374 .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
,
375 .flags
= MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
,
382 static int append_bind_mounts(MountEntry
**p
, const BindMount
*binds
, size_t n
) {
385 for (size_t i
= 0; i
< n
; i
++) {
386 const BindMount
*b
= binds
+ i
;
388 *((*p
)++) = (MountEntry
) {
389 .path_const
= b
->destination
,
390 .mode
= b
->recursive
? BIND_MOUNT_RECURSIVE
: BIND_MOUNT
,
391 .read_only
= b
->read_only
,
393 .source_const
= b
->source
,
394 .ignore
= b
->ignore_enoent
,
401 static int append_mount_images(MountEntry
**p
, const MountImage
*mount_images
, size_t n
) {
404 for (size_t i
= 0; i
< n
; i
++) {
405 const MountImage
*m
= mount_images
+ i
;
407 *((*p
)++) = (MountEntry
) {
408 .path_const
= m
->destination
,
409 .mode
= MOUNT_IMAGES
,
410 .source_const
= m
->source
,
411 .image_options
= m
->mount_options
,
412 .ignore
= m
->ignore_enoent
,
419 static int append_extensions(
422 const char *extension_dir
,
424 const MountImage
*mount_images
,
426 char **extension_directories
) {
428 _cleanup_strv_free_
char **overlays
= NULL
;
431 if (n
== 0 && strv_isempty(extension_directories
))
435 assert(extension_dir
);
437 /* Prepare a list of overlays, that will have as each element a string suitable for being
438 * passed as a lowerdir= parameter, so start with the hierarchy on the root.
439 * The overlays vector will have the same number of elements and will correspond to the
440 * hierarchies vector, so they can be iterated upon together. */
441 STRV_FOREACH(hierarchy
, hierarchies
) {
442 _cleanup_free_
char *prefixed_hierarchy
= NULL
;
444 prefixed_hierarchy
= path_join(root
, *hierarchy
);
445 if (!prefixed_hierarchy
)
448 r
= strv_consume(&overlays
, TAKE_PTR(prefixed_hierarchy
));
453 /* First, prepare a mount for each image, but these won't be visible to the unit, instead
454 * they will be mounted in our propagate directory, and used as a source for the overlay. */
455 for (size_t i
= 0; i
< n
; i
++) {
456 _cleanup_free_
char *mount_point
= NULL
;
457 const MountImage
*m
= mount_images
+ i
;
459 r
= asprintf(&mount_point
, "%s/%zu", extension_dir
, i
);
463 for (size_t j
= 0; hierarchies
&& hierarchies
[j
]; ++j
) {
464 _cleanup_free_
char *prefixed_hierarchy
= NULL
, *escaped
= NULL
, *lowerdir
= NULL
;
466 prefixed_hierarchy
= path_join(mount_point
, hierarchies
[j
]);
467 if (!prefixed_hierarchy
)
470 escaped
= shell_escape(prefixed_hierarchy
, ",:");
474 /* Note that lowerdir= parameters are in 'reverse' order, so the
475 * top-most directory in the overlay comes first in the list. */
476 lowerdir
= strjoin(escaped
, ":", overlays
[j
]);
480 free_and_replace(overlays
[j
], lowerdir
);
483 *((*p
)++) = (MountEntry
) {
484 .path_malloc
= TAKE_PTR(mount_point
),
485 .image_options
= m
->mount_options
,
486 .ignore
= m
->ignore_enoent
,
487 .source_const
= m
->source
,
488 .mode
= EXTENSION_IMAGES
,
493 /* Secondly, extend the lowerdir= parameters with each ExtensionDirectory.
494 * Bind mount them in the same location as the ExtensionImages, so that we
495 * can check that they are valid trees (extension-release.d). */
496 STRV_FOREACH(extension_directory
, extension_directories
) {
497 _cleanup_free_
char *mount_point
= NULL
, *source
= NULL
;
498 const char *e
= *extension_directory
;
499 bool ignore_enoent
= false;
501 /* Pick up the counter where the ExtensionImages left it. */
502 r
= asprintf(&mount_point
, "%s/%zu", extension_dir
, n
++);
506 /* Look for any prefixes */
507 if (startswith(e
, "-")) {
509 ignore_enoent
= true;
511 /* Ignore this for now */
512 if (startswith(e
, "+"))
519 for (size_t j
= 0; hierarchies
&& hierarchies
[j
]; ++j
) {
520 _cleanup_free_
char *prefixed_hierarchy
= NULL
, *escaped
= NULL
, *lowerdir
= NULL
;
522 prefixed_hierarchy
= path_join(mount_point
, hierarchies
[j
]);
523 if (!prefixed_hierarchy
)
526 escaped
= shell_escape(prefixed_hierarchy
, ",:");
530 /* Note that lowerdir= parameters are in 'reverse' order, so the
531 * top-most directory in the overlay comes first in the list. */
532 lowerdir
= strjoin(escaped
, ":", overlays
[j
]);
536 free_and_replace(overlays
[j
], lowerdir
);
539 *((*p
)++) = (MountEntry
) {
540 .path_malloc
= TAKE_PTR(mount_point
),
541 .source_malloc
= TAKE_PTR(source
),
542 .mode
= EXTENSION_DIRECTORIES
,
543 .ignore
= ignore_enoent
,
549 /* Then, for each hierarchy, prepare an overlay with the list of lowerdir= strings
551 for (size_t i
= 0; hierarchies
&& hierarchies
[i
]; ++i
) {
552 _cleanup_free_
char *prefixed_hierarchy
= NULL
;
554 prefixed_hierarchy
= path_join(root
, hierarchies
[i
]);
555 if (!prefixed_hierarchy
)
558 *((*p
)++) = (MountEntry
) {
559 .path_malloc
= TAKE_PTR(prefixed_hierarchy
),
560 .options_malloc
= TAKE_PTR(overlays
[i
]),
561 .mode
= OVERLAY_MOUNT
,
563 .ignore
= true, /* If the source image doesn't set the ignore bit it will fail earlier. */
570 static int append_tmpfs_mounts(MountEntry
**p
, const TemporaryFileSystem
*tmpfs
, size_t n
) {
573 for (size_t i
= 0; i
< n
; i
++) {
574 const TemporaryFileSystem
*t
= tmpfs
+ i
;
575 _cleanup_free_
char *o
= NULL
, *str
= NULL
;
580 if (!path_is_absolute(t
->path
))
581 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
582 "Path is not absolute: %s",
585 str
= strjoin("mode=0755" NESTED_TMPFS_LIMITS
",", t
->options
);
589 r
= mount_option_mangle(str
, MS_NODEV
|MS_STRICTATIME
, &flags
, &o
);
591 return log_debug_errno(r
, "Failed to parse mount option '%s': %m", str
);
593 ro
= flags
& MS_RDONLY
;
597 *((*p
)++) = (MountEntry
) {
598 .path_const
= t
->path
,
601 .options_malloc
= TAKE_PTR(o
),
609 static int append_static_mounts(MountEntry
**p
, const MountEntry
*mounts
, size_t n
, bool ignore_protect
) {
613 /* Adds a list of static pre-defined entries */
615 for (size_t i
= 0; i
< n
; i
++)
616 *((*p
)++) = (MountEntry
) {
617 .path_const
= mount_entry_path(mounts
+i
),
618 .mode
= mounts
[i
].mode
,
619 .ignore
= mounts
[i
].ignore
|| ignore_protect
,
625 static int append_protect_home(MountEntry
**p
, ProtectHome protect_home
, bool ignore_protect
) {
628 switch (protect_home
) {
630 case PROTECT_HOME_NO
:
633 case PROTECT_HOME_READ_ONLY
:
634 return append_static_mounts(p
, protect_home_read_only_table
, ELEMENTSOF(protect_home_read_only_table
), ignore_protect
);
636 case PROTECT_HOME_TMPFS
:
637 return append_static_mounts(p
, protect_home_tmpfs_table
, ELEMENTSOF(protect_home_tmpfs_table
), ignore_protect
);
639 case PROTECT_HOME_YES
:
640 return append_static_mounts(p
, protect_home_yes_table
, ELEMENTSOF(protect_home_yes_table
), ignore_protect
);
643 assert_not_reached();
647 static int append_protect_system(MountEntry
**p
, ProtectSystem protect_system
, bool ignore_protect
) {
650 switch (protect_system
) {
652 case PROTECT_SYSTEM_NO
:
655 case PROTECT_SYSTEM_STRICT
:
656 return append_static_mounts(p
, protect_system_strict_table
, ELEMENTSOF(protect_system_strict_table
), ignore_protect
);
658 case PROTECT_SYSTEM_YES
:
659 return append_static_mounts(p
, protect_system_yes_table
, ELEMENTSOF(protect_system_yes_table
), ignore_protect
);
661 case PROTECT_SYSTEM_FULL
:
662 return append_static_mounts(p
, protect_system_full_table
, ELEMENTSOF(protect_system_full_table
), ignore_protect
);
665 assert_not_reached();
669 static int mount_path_compare(const MountEntry
*a
, const MountEntry
*b
) {
672 /* ExtensionImages/Directories will be used by other mounts as a base, so sort them first
673 * regardless of the prefix - they are set up in the propagate directory anyway */
674 d
= -CMP(a
->mode
== EXTENSION_IMAGES
, b
->mode
== EXTENSION_IMAGES
);
677 d
= -CMP(a
->mode
== EXTENSION_DIRECTORIES
, b
->mode
== EXTENSION_DIRECTORIES
);
681 /* If the paths are not equal, then order prefixes first */
682 d
= path_compare(mount_entry_path(a
), mount_entry_path(b
));
686 /* If the paths are equal, check the mode */
687 return CMP((int) a
->mode
, (int) b
->mode
);
690 static int prefix_where_needed(MountEntry
*m
, size_t n
, const char *root_directory
) {
691 /* Prefixes all paths in the bind mount table with the root directory if the entry needs that. */
695 for (size_t i
= 0; i
< n
; i
++) {
701 s
= path_join(root_directory
, mount_entry_path(m
+i
));
705 mount_entry_consume_prefix(&m
[i
], s
);
711 static void drop_duplicates(MountEntry
*m
, size_t *n
) {
712 MountEntry
*f
, *t
, *previous
;
717 /* Drops duplicate entries. Expects that the array is properly ordered already. */
719 for (f
= m
, t
= m
, previous
= NULL
; f
< m
+ *n
; f
++) {
721 /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare()
722 * above. Note that we only drop duplicates that haven't been mounted yet. */
724 path_equal(mount_entry_path(f
), mount_entry_path(previous
)) &&
725 !f
->applied
&& !previous
->applied
) {
726 log_debug("%s (%s) is duplicate.", mount_entry_path(f
), mount_mode_to_string(f
->mode
));
727 /* Propagate the flags to the remaining entry */
728 previous
->read_only
= previous
->read_only
|| mount_entry_read_only(f
);
729 previous
->noexec
= previous
->noexec
|| mount_entry_noexec(f
);
730 previous
->exec
= previous
->exec
|| mount_entry_exec(f
);
743 static void drop_inaccessible(MountEntry
*m
, size_t *n
) {
745 const char *clear
= NULL
;
750 /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly
751 * ordered already. */
753 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
755 /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop
756 * it, as inaccessible paths really should drop the entire subtree. */
757 if (clear
&& path_startswith(mount_entry_path(f
), clear
)) {
758 log_debug("%s is masked by %s.", mount_entry_path(f
), clear
);
763 clear
= f
->mode
== INACCESSIBLE
? mount_entry_path(f
) : NULL
;
772 static void drop_nop(MountEntry
*m
, size_t *n
) {
778 /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the
779 * list is ordered by prefixes. */
781 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
783 /* Only suppress such subtrees for READONLY, READWRITE and READWRITE_IMPLICIT entries */
784 if (IN_SET(f
->mode
, READONLY
, READWRITE
, READWRITE_IMPLICIT
)) {
785 MountEntry
*found
= NULL
;
787 /* Now let's find the first parent of the entry we are looking at. */
788 for (MountEntry
*p
= PTR_SUB1(t
, m
); p
; p
= PTR_SUB1(p
, m
))
789 if (path_startswith(mount_entry_path(f
), mount_entry_path(p
))) {
794 /* We found it, let's see if it's the same mode, if so, we can drop this entry */
795 if (found
&& found
->mode
== f
->mode
) {
796 log_debug("%s (%s) is made redundant by %s (%s)",
797 mount_entry_path(f
), mount_mode_to_string(f
->mode
),
798 mount_entry_path(found
), mount_mode_to_string(found
->mode
));
811 static void drop_outside_root(const char *root_directory
, MountEntry
*m
, size_t *n
) {
821 /* Drops all mounts that are outside of the root directory. */
823 for (f
= m
, t
= m
; f
< m
+ *n
; f
++) {
825 /* ExtensionImages/Directories bases are opened in /run/systemd/unit-extensions on the host */
826 if (!IN_SET(f
->mode
, EXTENSION_IMAGES
, EXTENSION_DIRECTORIES
) && !path_startswith(mount_entry_path(f
), root_directory
)) {
827 log_debug("%s is outside of root directory.", mount_entry_path(f
));
839 static int clone_device_node(
841 const char *temporary_mount
,
842 bool *make_devnode
) {
844 _cleanup_free_
char *sl
= NULL
;
845 const char *dn
, *bn
, *t
;
849 if (stat(d
, &st
) < 0) {
850 if (errno
== ENOENT
) {
851 log_debug_errno(errno
, "Device node '%s' to clone does not exist, ignoring.", d
);
855 return log_debug_errno(errno
, "Failed to stat() device node '%s' to clone, ignoring: %m", d
);
858 if (!S_ISBLK(st
.st_mode
) &&
859 !S_ISCHR(st
.st_mode
))
860 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
861 "Device node '%s' to clone is not a device node, ignoring.",
864 dn
= strjoina(temporary_mount
, d
);
866 /* First, try to create device node properly */
868 mac_selinux_create_file_prepare(d
, st
.st_mode
);
869 r
= mknod(dn
, st
.st_mode
, st
.st_rdev
);
870 mac_selinux_create_file_clear();
874 return log_debug_errno(errno
, "mknod failed for %s: %m", d
);
876 /* This didn't work, let's not try this again for the next iterations. */
877 *make_devnode
= false;
880 /* We're about to fall back to bind-mounting the device node. So create a dummy bind-mount target.
881 * Do not prepare device-node SELinux label (see issue 13762) */
882 r
= mknod(dn
, S_IFREG
, 0);
883 if (r
< 0 && errno
!= EEXIST
)
884 return log_debug_errno(errno
, "mknod() fallback failed for '%s': %m", d
);
886 /* Fallback to bind-mounting: The assumption here is that all used device nodes carry standard
887 * properties. Specifically, the devices nodes we bind-mount should either be owned by root:root or
888 * root:tty (e.g. /dev/tty, /dev/ptmx) and should not carry ACLs. */
889 r
= mount_nofollow_verbose(LOG_DEBUG
, d
, dn
, NULL
, MS_BIND
, NULL
);
894 bn
= path_startswith(d
, "/dev/");
898 /* Create symlinks like /dev/char/1:9 → ../urandom */
899 if (asprintf(&sl
, "%s/dev/%s/" DEVNUM_FORMAT_STR
,
901 S_ISCHR(st
.st_mode
) ? "char" : "block",
902 DEVNUM_FORMAT_VAL(st
.st_rdev
)) < 0)
905 (void) mkdir_parents(sl
, 0755);
907 t
= strjoina("../", bn
);
908 if (symlink(t
, sl
) < 0)
909 log_debug_errno(errno
, "Failed to symlink '%s' to '%s', ignoring: %m", t
, sl
);
914 static int mount_private_dev(MountEntry
*m
) {
915 static const char devnodes
[] =
923 char temporary_mount
[] = "/tmp/namespace-dev-XXXXXX";
924 const char *dev
= NULL
, *devpts
= NULL
, *devshm
= NULL
, *devhugepages
= NULL
, *devmqueue
= NULL
, *devlog
= NULL
, *devptmx
= NULL
;
925 bool can_mknod
= true;
930 if (!mkdtemp(temporary_mount
))
931 return log_debug_errno(errno
, "Failed to create temporary directory '%s': %m", temporary_mount
);
933 dev
= strjoina(temporary_mount
, "/dev");
934 (void) mkdir(dev
, 0755);
935 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", dev
, "tmpfs", DEV_MOUNT_OPTIONS
, "mode=0755" TMPFS_LIMITS_PRIVATE_DEV
);
939 r
= label_fix_full(AT_FDCWD
, dev
, "/dev", 0);
941 log_debug_errno(r
, "Failed to fix label of '%s' as /dev: %m", dev
);
945 devpts
= strjoina(temporary_mount
, "/dev/pts");
946 (void) mkdir(devpts
, 0755);
947 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev/pts", devpts
, NULL
, MS_BIND
, NULL
);
951 /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx.
952 * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible.
953 * Thus, in that case make a clone.
954 * In nspawn and other containers it will be a symlink, in that case make it a symlink. */
955 r
= is_symlink("/dev/ptmx");
957 log_debug_errno(r
, "Failed to detect whether /dev/ptmx is a symlink or not: %m");
960 devptmx
= strjoina(temporary_mount
, "/dev/ptmx");
961 if (symlink("pts/ptmx", devptmx
) < 0) {
962 r
= log_debug_errno(errno
, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx
);
966 r
= clone_device_node("/dev/ptmx", temporary_mount
, &can_mknod
);
971 devshm
= strjoina(temporary_mount
, "/dev/shm");
972 (void) mkdir(devshm
, 0755);
973 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev/shm", devshm
, NULL
, MS_BIND
, NULL
);
977 devmqueue
= strjoina(temporary_mount
, "/dev/mqueue");
978 (void) mkdir(devmqueue
, 0755);
979 (void) mount_nofollow_verbose(LOG_DEBUG
, "/dev/mqueue", devmqueue
, NULL
, MS_BIND
, NULL
);
981 devhugepages
= strjoina(temporary_mount
, "/dev/hugepages");
982 (void) mkdir(devhugepages
, 0755);
983 (void) mount_nofollow_verbose(LOG_DEBUG
, "/dev/hugepages", devhugepages
, NULL
, MS_BIND
, NULL
);
985 devlog
= strjoina(temporary_mount
, "/dev/log");
986 if (symlink("/run/systemd/journal/dev-log", devlog
) < 0)
987 log_debug_errno(errno
, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog
);
989 NULSTR_FOREACH(d
, devnodes
) {
990 r
= clone_device_node(d
, temporary_mount
, &can_mknod
);
991 /* ENXIO means the *source* is not a device file, skip creation in that case */
992 if (r
< 0 && r
!= -ENXIO
)
996 r
= dev_setup(temporary_mount
, UID_INVALID
, GID_INVALID
);
998 log_debug_errno(r
, "Failed to set up basic device tree at '%s', ignoring: %m", temporary_mount
);
1000 /* Create the /dev directory if missing. It is more likely to be missing when the service is started
1001 * with RootDirectory. This is consistent with mount units creating the mount points when missing. */
1002 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1004 /* Unmount everything in old /dev */
1005 r
= umount_recursive(mount_entry_path(m
), 0);
1007 log_debug_errno(r
, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m
));
1009 r
= mount_nofollow_verbose(LOG_DEBUG
, dev
, mount_entry_path(m
), NULL
, MS_MOVE
, NULL
);
1014 (void) rmdir(temporary_mount
);
1020 (void) umount_verbose(LOG_DEBUG
, devpts
, UMOUNT_NOFOLLOW
);
1023 (void) umount_verbose(LOG_DEBUG
, devshm
, UMOUNT_NOFOLLOW
);
1026 (void) umount_verbose(LOG_DEBUG
, devhugepages
, UMOUNT_NOFOLLOW
);
1029 (void) umount_verbose(LOG_DEBUG
, devmqueue
, UMOUNT_NOFOLLOW
);
1031 (void) umount_verbose(LOG_DEBUG
, dev
, UMOUNT_NOFOLLOW
);
1033 (void) rmdir(temporary_mount
);
1038 static int mount_bind_dev(const MountEntry
*m
) {
1043 /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the
1044 * service's /dev. This is only used when RootDirectory= is set. */
1046 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1048 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
1050 return log_debug_errno(r
, "Unable to determine whether /dev is already mounted: %m");
1051 if (r
> 0) /* make this a NOP if /dev is already a mount point */
1054 return mount_nofollow_verbose(LOG_DEBUG
, "/dev", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
);
1057 static int mount_bind_sysfs(const MountEntry
*m
) {
1062 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1064 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
1066 return log_debug_errno(r
, "Unable to determine whether /sys is already mounted: %m");
1067 if (r
> 0) /* make this a NOP if /sys is already a mount point */
1070 /* Bind mount the host's version so that we get all child mounts of it, too. */
1071 return mount_nofollow_verbose(LOG_DEBUG
, "/sys", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
);
1074 static int mount_private_sysfs(const MountEntry
*m
) {
1075 const char *entry_path
= mount_entry_path(ASSERT_PTR(m
));
1078 (void) mkdir_p_label(entry_path
, 0755);
1080 n
= umount_recursive(entry_path
, 0);
1082 r
= mount_nofollow_verbose(LOG_DEBUG
, "sysfs", entry_path
, "sysfs", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
1083 if (ERRNO_IS_NEG_PRIVILEGE(r
)) {
1084 /* When we do not have enough privileges to mount sysfs, fall back to use existing /sys. */
1087 /* /sys or some of sub-mounts are umounted in the above. Refuse incomplete tree.
1088 * Propagate the original error code returned by mount() in the above. */
1091 return mount_bind_sysfs(m
);
1096 /* We mounted a new instance now. Let's bind mount the children over now. */
1097 (void) bind_mount_submounts("/sys", entry_path
);
1101 static int mount_procfs(const MountEntry
*m
, const NamespaceInfo
*ns_info
) {
1102 _cleanup_free_
char *opts
= NULL
;
1103 const char *entry_path
;
1109 if (ns_info
->protect_proc
!= PROTECT_PROC_DEFAULT
||
1110 ns_info
->proc_subset
!= PROC_SUBSET_ALL
) {
1112 /* Starting with kernel 5.8 procfs' hidepid= logic is truly per-instance (previously it
1113 * pretended to be per-instance but actually was per-namespace), hence let's make use of it
1114 * if requested. To make sure this logic succeeds only on kernels where hidepid= is
1115 * per-instance, we'll exclusively use the textual value for hidepid=, since support was
1116 * added in the same commit: if it's supported it is thus also per-instance. */
1118 const char *hpv
= ns_info
->protect_proc
== PROTECT_PROC_DEFAULT
?
1120 protect_proc_to_string(ns_info
->protect_proc
);
1122 /* hidepid= support was added in 5.8, so we can use fsconfig()/fsopen() (which were added in
1123 * 5.2) to check if hidepid= is supported. This avoids a noisy dmesg log by the kernel when
1124 * trying to use hidepid= on systems where it isn't supported. The same applies for subset=.
1125 * fsopen()/fsconfig() was also backported on some distros which allows us to detect
1126 * hidepid=/subset= support in even more scenarios. */
1128 if (mount_option_supported("proc", "hidepid", hpv
) != 0) {
1129 opts
= strjoin("hidepid=", hpv
);
1134 if (ns_info
->proc_subset
== PROC_SUBSET_PID
&&
1135 mount_option_supported("proc", "subset", "pid") != 0)
1136 if (!strextend_with_separator(&opts
, ",", "subset=pid"))
1140 entry_path
= mount_entry_path(m
);
1141 (void) mkdir_p_label(entry_path
, 0755);
1143 /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in
1144 * one. i.e we don't reuse existing mounts here under any condition, we want a new instance owned by
1145 * our user namespace and with our hidepid= settings applied. Hence, let's get rid of everything
1146 * mounted on /proc/ first. */
1148 n
= umount_recursive(entry_path
, 0);
1150 r
= mount_nofollow_verbose(LOG_DEBUG
, "proc", entry_path
, "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, opts
);
1151 if (r
== -EINVAL
&& opts
)
1152 /* If this failed with EINVAL then this likely means the textual hidepid= stuff is
1153 * not supported by the kernel, and thus the per-instance hidepid= neither, which
1154 * means we really don't want to use it, since it would affect our host's /proc
1155 * mount. Hence let's gracefully fallback to a classic, unrestricted version. */
1156 r
= mount_nofollow_verbose(LOG_DEBUG
, "proc", entry_path
, "proc", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, NULL
);
1157 if (ERRNO_IS_NEG_PRIVILEGE(r
)) {
1158 /* When we do not have enough privileges to mount /proc, fall back to use existing /proc. */
1161 /* /proc or some of sub-mounts are umounted in the above. Refuse incomplete tree.
1162 * Propagate the original error code returned by mount() in the above. */
1165 r
= path_is_mount_point(entry_path
, NULL
, 0);
1167 return log_debug_errno(r
, "Unable to determine whether /proc is already mounted: %m");
1171 /* We lack permissions to mount a new instance of /proc, and it is not already mounted. But
1172 * we can access the host's, so as a final fallback bind-mount it to the destination, as most
1173 * likely we are inside a user manager in an unprivileged user namespace. */
1174 return mount_nofollow_verbose(LOG_DEBUG
, "/proc", entry_path
, NULL
, MS_BIND
|MS_REC
, NULL
);
1179 /* We mounted a new instance now. Let's bind mount the children over now. This matters for nspawn
1180 * where a bunch of files are overmounted, in particular the boot id */
1181 (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
)
1326 static int follow_symlink(
1327 const char *root_directory
,
1330 _cleanup_free_
char *target
= NULL
;
1333 /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we
1334 * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at
1335 * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the
1336 * end and already have a fully normalized name. */
1338 r
= chase(mount_entry_path(m
), root_directory
, CHASE_STEP
|CHASE_NONEXISTENT
, &target
, NULL
);
1340 return log_debug_errno(r
, "Failed to chase symlinks '%s': %m", mount_entry_path(m
));
1341 if (r
> 0) /* Reached the end, nothing more to resolve */
1344 if (m
->n_followed
>= CHASE_MAX
) /* put a boundary on things */
1345 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
1346 "Symlink loop on '%s'.",
1347 mount_entry_path(m
));
1349 log_debug("Followed mount entry path symlink %s %s %s.",
1350 mount_entry_path(m
), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT
), target
);
1352 mount_entry_consume_prefix(m
, TAKE_PTR(target
));
1359 static int apply_one_mount(
1360 const char *root_directory
,
1362 const ImagePolicy
*mount_image_policy
,
1363 const ImagePolicy
*extension_image_policy
,
1364 const NamespaceInfo
*ns_info
) {
1366 _cleanup_free_
char *inaccessible
= NULL
;
1367 bool rbind
= true, make
= false;
1374 log_debug("Applying namespace mount on %s", mount_entry_path(m
));
1378 case INACCESSIBLE
: {
1379 _cleanup_free_
char *tmp
= NULL
;
1380 const char *runtime_dir
;
1383 /* First, get rid of everything that is below if there
1384 * is anything... Then, overmount it with an
1385 * inaccessible path. */
1386 (void) umount_recursive(mount_entry_path(m
), 0);
1388 if (lstat(mount_entry_path(m
), &target
) < 0) {
1389 if (errno
== ENOENT
&& m
->ignore
)
1392 return log_debug_errno(errno
, "Failed to lstat() %s to determine what to mount over it: %m",
1393 mount_entry_path(m
));
1397 runtime_dir
= "/run";
1399 if (asprintf(&tmp
, "/run/user/" UID_FMT
, geteuid()) < 0)
1405 r
= mode_to_inaccessible_node(runtime_dir
, target
.st_mode
, &inaccessible
);
1407 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
1408 "File type not supported for inaccessible mounts. Note that symlinks are not allowed");
1409 what
= inaccessible
;
1415 case READWRITE_IMPLICIT
:
1418 r
= path_is_mount_point(mount_entry_path(m
), root_directory
, 0);
1419 if (r
== -ENOENT
&& m
->ignore
)
1422 return log_debug_errno(r
, "Failed to determine whether %s is already a mount point: %m",
1423 mount_entry_path(m
));
1424 if (r
> 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY
1425 * and MS_NOEXEC bits for the mount point if needed. */
1427 /* This isn't a mount point yet, let's make it one. */
1428 what
= mount_entry_path(m
);
1431 case EXTENSION_DIRECTORIES
: {
1432 _cleanup_free_
char *host_os_release_id
= NULL
, *host_os_release_version_id
= NULL
,
1433 *host_os_release_level
= NULL
, *extension_name
= NULL
;
1434 _cleanup_strv_free_
char **extension_release
= NULL
;
1435 ImageClass
class = IMAGE_SYSEXT
;
1437 r
= path_extract_filename(mount_entry_source(m
), &extension_name
);
1439 return log_debug_errno(r
, "Failed to extract extension name from %s: %m", mount_entry_source(m
));
1441 r
= load_extension_release_pairs(mount_entry_source(m
), IMAGE_SYSEXT
, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1443 r
= load_extension_release_pairs(mount_entry_source(m
), IMAGE_CONFEXT
, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1445 class = IMAGE_CONFEXT
;
1450 r
= parse_os_release(
1451 empty_to_root(root_directory
),
1452 "ID", &host_os_release_id
,
1453 "VERSION_ID", &host_os_release_version_id
,
1454 image_class_info
[class].level_env
, &host_os_release_level
,
1457 return log_debug_errno(r
, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory
));
1458 if (isempty(host_os_release_id
))
1459 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
));
1461 r
= load_extension_release_pairs(mount_entry_source(m
), class, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1462 if (r
== -ENOENT
&& m
->ignore
)
1465 return log_debug_errno(r
, "Failed to parse directory %s extension-release metadata: %m", extension_name
);
1467 r
= extension_release_validate(
1470 host_os_release_version_id
,
1471 host_os_release_level
,
1472 /* host_extension_scope */ NULL
, /* Leave empty, we need to accept both system and portable */
1476 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE
), "Directory %s extension-release metadata does not match the root's", extension_name
);
1478 return log_debug_errno(r
, "Failed to compare directory %s extension-release metadata with the root's os-release: %m", extension_name
);
1487 case BIND_MOUNT_RECURSIVE
: {
1488 _cleanup_free_
char *chased
= NULL
;
1490 /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note
1491 * that bind mount source paths are always relative to the host root, hence we pass NULL as
1492 * root directory to chase() here. */
1494 r
= chase(mount_entry_source(m
), NULL
, CHASE_TRAIL_SLASH
, &chased
, NULL
);
1495 if (r
== -ENOENT
&& m
->ignore
) {
1496 log_debug_errno(r
, "Path %s does not exist, ignoring.", mount_entry_source(m
));
1500 return log_debug_errno(r
, "Failed to follow symlinks on %s: %m", mount_entry_source(m
));
1502 log_debug("Followed source symlinks %s %s %s.",
1503 mount_entry_source(m
), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT
), chased
);
1505 free_and_replace(m
->source_malloc
, chased
);
1507 what
= mount_entry_source(m
);
1514 return mount_tmpfs(m
);
1517 case PRIVATE_TMP_READONLY
:
1518 what
= mount_entry_source(m
);
1523 return mount_private_dev(m
);
1526 return mount_bind_dev(m
);
1529 return mount_private_sysfs(m
);
1532 return mount_bind_sysfs(m
);
1535 return mount_procfs(m
, ns_info
);
1538 return mount_run(m
);
1541 return mount_mqueuefs(m
);
1544 return mount_image(m
, NULL
, mount_image_policy
);
1546 case EXTENSION_IMAGES
:
1547 return mount_image(m
, root_directory
, extension_image_policy
);
1550 return mount_overlay(m
);
1553 r
= mkdir_p_label(mount_entry_path(m
), 0755);
1559 assert_not_reached();
1564 r
= mount_nofollow_verbose(LOG_DEBUG
, what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
);
1566 bool try_again
= false;
1568 if (r
== -ENOENT
&& make
) {
1571 /* Hmm, either the source or the destination are missing. Let's see if we can create
1572 the destination, then try again. */
1574 (void) mkdir_parents(mount_entry_path(m
), 0755);
1576 q
= make_mount_point_inode_from_path(what
, mount_entry_path(m
), 0755);
1577 if (q
< 0 && q
!= -EEXIST
)
1578 log_error_errno(q
, "Failed to create destination mount point node '%s': %m",
1579 mount_entry_path(m
));
1585 r
= mount_nofollow_verbose(LOG_DEBUG
, what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
);
1587 return log_error_errno(r
, "Failed to mount %s to %s: %m", what
, mount_entry_path(m
));
1590 log_debug("Successfully mounted %s to %s", what
, mount_entry_path(m
));
1594 static int make_read_only(const MountEntry
*m
, char **deny_list
, FILE *proc_self_mountinfo
) {
1595 unsigned long new_flags
= 0, flags_mask
= 0;
1600 assert(proc_self_mountinfo
);
1602 if (mount_entry_read_only(m
) || m
->mode
== PRIVATE_DEV
) {
1603 new_flags
|= MS_RDONLY
;
1604 flags_mask
|= MS_RDONLY
;
1608 new_flags
|= MS_NOSUID
;
1609 flags_mask
|= MS_NOSUID
;
1612 if (flags_mask
== 0) /* No Change? */
1615 /* We generally apply these changes recursively, except for /dev, and the cases we know there's
1616 * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the
1617 * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace
1618 * and running Linux <= 4.17. */
1620 mount_entry_read_only(m
) &&
1621 !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1623 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, deny_list
, proc_self_mountinfo
);
1625 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, proc_self_mountinfo
);
1627 /* Note that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked
1628 * read-only already stays this way. This improves compatibility with container managers, where we
1629 * won't attempt to undo read-only mounts already applied. */
1631 if (r
== -ENOENT
&& m
->ignore
)
1634 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1635 submounts
? " and its submounts" : "");
1639 static int make_noexec(const MountEntry
*m
, char **deny_list
, FILE *proc_self_mountinfo
) {
1640 unsigned long new_flags
= 0, flags_mask
= 0;
1645 assert(proc_self_mountinfo
);
1647 if (mount_entry_noexec(m
)) {
1648 new_flags
|= MS_NOEXEC
;
1649 flags_mask
|= MS_NOEXEC
;
1650 } else if (mount_entry_exec(m
)) {
1651 new_flags
&= ~MS_NOEXEC
;
1652 flags_mask
|= MS_NOEXEC
;
1655 if (flags_mask
== 0) /* No Change? */
1658 submounts
= !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1661 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, deny_list
, proc_self_mountinfo
);
1663 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, proc_self_mountinfo
);
1665 if (r
== -ENOENT
&& m
->ignore
)
1668 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1669 submounts
? " and its submounts" : "");
1673 static int make_nosuid(const MountEntry
*m
, FILE *proc_self_mountinfo
) {
1678 assert(proc_self_mountinfo
);
1680 submounts
= !IN_SET(m
->mode
, EMPTY_DIR
, TMPFS
);
1683 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), MS_NOSUID
, MS_NOSUID
, NULL
, proc_self_mountinfo
);
1685 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), MS_NOSUID
, MS_NOSUID
, proc_self_mountinfo
);
1686 if (r
== -ENOENT
&& m
->ignore
)
1689 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1690 submounts
? " and its submounts" : "");
1694 static bool namespace_info_mount_apivfs(const NamespaceInfo
*ns_info
) {
1698 * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
1699 * since to protect the API VFS mounts, they need to be around in the
1703 return ns_info
->mount_apivfs
||
1704 ns_info
->protect_control_groups
||
1705 ns_info
->protect_kernel_tunables
||
1706 ns_info
->protect_proc
!= PROTECT_PROC_DEFAULT
||
1707 ns_info
->proc_subset
!= PROC_SUBSET_ALL
;
1710 static size_t namespace_calculate_mounts(
1711 const NamespaceInfo
*ns_info
,
1712 char** read_write_paths
,
1713 char** read_only_paths
,
1714 char** inaccessible_paths
,
1716 char** no_exec_paths
,
1717 char** empty_directories
,
1718 size_t n_bind_mounts
,
1719 size_t n_temporary_filesystems
,
1720 size_t n_mount_images
,
1721 size_t n_extension_images
,
1722 size_t n_extension_directories
,
1723 size_t n_hierarchies
,
1724 const char* tmp_dir
,
1725 const char* var_tmp_dir
,
1726 const char *creds_path
,
1727 const char* log_namespace
,
1728 bool setup_propagate
,
1729 const char* notify_socket
,
1730 const char* host_os_release
) {
1732 size_t protect_home_cnt
;
1733 size_t protect_system_cnt
=
1734 (ns_info
->protect_system
== PROTECT_SYSTEM_STRICT
?
1735 ELEMENTSOF(protect_system_strict_table
) :
1736 ((ns_info
->protect_system
== PROTECT_SYSTEM_FULL
) ?
1737 ELEMENTSOF(protect_system_full_table
) :
1738 ((ns_info
->protect_system
== PROTECT_SYSTEM_YES
) ?
1739 ELEMENTSOF(protect_system_yes_table
) : 0)));
1742 (ns_info
->protect_home
== PROTECT_HOME_YES
?
1743 ELEMENTSOF(protect_home_yes_table
) :
1744 ((ns_info
->protect_home
== PROTECT_HOME_READ_ONLY
) ?
1745 ELEMENTSOF(protect_home_read_only_table
) :
1746 ((ns_info
->protect_home
== PROTECT_HOME_TMPFS
) ?
1747 ELEMENTSOF(protect_home_tmpfs_table
) : 0)));
1749 return !!tmp_dir
+ !!var_tmp_dir
+
1750 strv_length(read_write_paths
) +
1751 strv_length(read_only_paths
) +
1752 strv_length(inaccessible_paths
) +
1753 strv_length(exec_paths
) +
1754 strv_length(no_exec_paths
) +
1755 strv_length(empty_directories
) +
1758 (n_extension_images
> 0 || n_extension_directories
> 0 ? /* Mount each image and directory plus an overlay per hierarchy */
1759 n_hierarchies
+ n_extension_images
+ n_extension_directories
: 0) +
1760 n_temporary_filesystems
+
1761 ns_info
->private_dev
+
1762 (ns_info
->protect_kernel_tunables
?
1763 ELEMENTSOF(protect_kernel_tunables_proc_table
) + ELEMENTSOF(protect_kernel_tunables_sys_table
) : 0) +
1764 (ns_info
->protect_kernel_modules
? ELEMENTSOF(protect_kernel_modules_table
) : 0) +
1765 (ns_info
->protect_kernel_logs
?
1766 ELEMENTSOF(protect_kernel_logs_proc_table
) + ELEMENTSOF(protect_kernel_logs_dev_table
) : 0) +
1767 (ns_info
->protect_control_groups
? 1 : 0) +
1768 protect_home_cnt
+ protect_system_cnt
+
1769 (ns_info
->protect_hostname
? 2 : 0) +
1770 (namespace_info_mount_apivfs(ns_info
) ? ELEMENTSOF(apivfs_table
) : 0) +
1771 (creds_path
? 2 : 1) +
1773 setup_propagate
+ /* /run/systemd/incoming */
1776 ns_info
->private_network
+ /* /sys */
1777 ns_info
->private_ipc
; /* /dev/mqueue */
1780 /* Walk all mount entries and dropping any unused mounts. This affects all
1782 * - that are implicitly protected by a path that has been rendered inaccessible
1783 * - whose immediate parent requests the same protection mode as the mount itself
1784 * - that are outside of the relevant root directory
1785 * - which are duplicates
1787 static void drop_unused_mounts(const char *root_directory
, MountEntry
*mounts
, size_t *n_mounts
) {
1788 assert(root_directory
);
1790 assert(mounts
|| *n_mounts
== 0);
1792 typesafe_qsort(mounts
, *n_mounts
, mount_path_compare
);
1794 drop_duplicates(mounts
, n_mounts
);
1795 drop_outside_root(root_directory
, mounts
, n_mounts
);
1796 drop_inaccessible(mounts
, n_mounts
);
1797 drop_nop(mounts
, n_mounts
);
1800 static int create_symlinks_from_tuples(const char *root
, char **strv_symlinks
) {
1803 STRV_FOREACH_PAIR(src
, dst
, strv_symlinks
) {
1804 _cleanup_free_
char *src_abs
= NULL
, *dst_abs
= NULL
;
1806 src_abs
= path_join(root
, *src
);
1807 dst_abs
= path_join(root
, *dst
);
1808 if (!src_abs
|| !dst_abs
)
1811 r
= mkdir_parents_label(dst_abs
, 0755);
1815 r
= symlink_idempotent(src_abs
, dst_abs
, true);
1823 static int apply_mounts(
1825 const ImagePolicy
*mount_image_policy
,
1826 const ImagePolicy
*extension_image_policy
,
1827 const NamespaceInfo
*ns_info
,
1831 char **error_path
) {
1833 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
1834 _cleanup_free_
char **deny_list
= NULL
;
1837 if (n_mounts
== 0) /* Shortcut: nothing to do */
1844 /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of
1845 * /proc. For example, this is the case with the option: 'InaccessiblePaths=/proc'. */
1846 proc_self_mountinfo
= fopen("/proc/self/mountinfo", "re");
1847 if (!proc_self_mountinfo
) {
1851 *error_path
= strdup("/proc/self/mountinfo");
1853 return log_debug_errno(r
, "Failed to open /proc/self/mountinfo: %m");
1856 /* First round, establish all mounts we need */
1860 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1865 /* ExtensionImages/Directories are first opened in the propagate directory, not in the root_directory */
1866 r
= follow_symlink(!IN_SET(m
->mode
, EXTENSION_IMAGES
, EXTENSION_DIRECTORIES
) ? root
: NULL
, m
);
1868 if (error_path
&& mount_entry_path(m
))
1869 *error_path
= strdup(mount_entry_path(m
));
1873 /* We hit a symlinked mount point. The entry got rewritten and might
1874 * point to a very different place now. Let's normalize the changed
1875 * list, and start from the beginning. After all to mount the entry
1876 * at the new location we might need some other mounts first */
1881 r
= apply_one_mount(root
, m
, mount_image_policy
, extension_image_policy
, ns_info
);
1883 if (error_path
&& mount_entry_path(m
))
1884 *error_path
= strdup(mount_entry_path(m
));
1894 drop_unused_mounts(root
, mounts
, n_mounts
);
1897 /* Now that all filesystems have been set up, but before the
1898 * read-only switches are flipped, create the exec dirs and other symlinks.
1899 * Note that when /var/lib is not empty/tmpfs, these symlinks will already
1900 * exist, which means this will be a no-op. */
1901 r
= create_symlinks_from_tuples(root
, symlinks
);
1903 return log_debug_errno(r
, "Failed to set up symlinks inside mount namespace: %m");
1905 /* Create a deny list we can pass to bind_mount_recursive() */
1906 deny_list
= new(char*, (*n_mounts
)+1);
1909 for (size_t j
= 0; j
< *n_mounts
; j
++)
1910 deny_list
[j
] = (char*) mount_entry_path(mounts
+j
);
1911 deny_list
[*n_mounts
] = NULL
;
1913 /* Second round, flip the ro bits if necessary. */
1914 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1915 r
= make_read_only(m
, deny_list
, proc_self_mountinfo
);
1917 if (error_path
&& mount_entry_path(m
))
1918 *error_path
= strdup(mount_entry_path(m
));
1923 /* Third round, flip the noexec bits with a simplified deny list. */
1924 for (size_t j
= 0; j
< *n_mounts
; j
++)
1925 if (IN_SET((mounts
+j
)->mode
, EXEC
, NOEXEC
))
1926 deny_list
[j
] = (char*) mount_entry_path(mounts
+j
);
1927 deny_list
[*n_mounts
] = NULL
;
1929 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1930 r
= make_noexec(m
, deny_list
, proc_self_mountinfo
);
1932 if (error_path
&& mount_entry_path(m
))
1933 *error_path
= strdup(mount_entry_path(m
));
1938 /* Fourth round, flip the nosuid bits without a deny list. */
1939 if (ns_info
->mount_nosuid
)
1940 for (MountEntry
*m
= mounts
; m
< mounts
+ *n_mounts
; ++m
) {
1941 r
= make_nosuid(m
, proc_self_mountinfo
);
1943 if (error_path
&& mount_entry_path(m
))
1944 *error_path
= strdup(mount_entry_path(m
));
1952 static bool root_read_only(
1953 char **read_only_paths
,
1954 ProtectSystem protect_system
) {
1956 /* Determine whether the root directory is going to be read-only given the configured settings. */
1958 if (protect_system
== PROTECT_SYSTEM_STRICT
)
1961 if (prefixed_path_strv_contains(read_only_paths
, "/"))
1967 static bool home_read_only(
1968 char** read_only_paths
,
1969 char** inaccessible_paths
,
1970 char** empty_directories
,
1971 const BindMount
*bind_mounts
,
1972 size_t n_bind_mounts
,
1973 const TemporaryFileSystem
*temporary_filesystems
,
1974 size_t n_temporary_filesystems
,
1975 ProtectHome protect_home
) {
1977 /* Determine whether the /home directory is going to be read-only given the configured settings. Yes,
1978 * this is a bit sloppy, since we don't bother checking for cases where / is affected by multiple
1981 if (protect_home
!= PROTECT_HOME_NO
)
1984 if (prefixed_path_strv_contains(read_only_paths
, "/home") ||
1985 prefixed_path_strv_contains(inaccessible_paths
, "/home") ||
1986 prefixed_path_strv_contains(empty_directories
, "/home"))
1989 for (size_t i
= 0; i
< n_temporary_filesystems
; i
++)
1990 if (path_equal(temporary_filesystems
[i
].path
, "/home"))
1993 /* If /home is overmounted with some dir from the host it's not writable. */
1994 for (size_t i
= 0; i
< n_bind_mounts
; i
++)
1995 if (path_equal(bind_mounts
[i
].destination
, "/home"))
2001 int setup_namespace(
2002 const char* root_directory
,
2003 const char* root_image
,
2004 const MountOptions
*root_image_mount_options
,
2005 const ImagePolicy
*root_image_policy
,
2006 const NamespaceInfo
*ns_info
,
2007 char** read_write_paths
,
2008 char** read_only_paths
,
2009 char** inaccessible_paths
,
2011 char** no_exec_paths
,
2012 char** empty_directories
,
2014 const BindMount
*bind_mounts
,
2015 size_t n_bind_mounts
,
2016 const TemporaryFileSystem
*temporary_filesystems
,
2017 size_t n_temporary_filesystems
,
2018 const MountImage
*mount_images
,
2019 size_t n_mount_images
,
2020 const ImagePolicy
*mount_image_policy
,
2021 const char* tmp_dir
,
2022 const char* var_tmp_dir
,
2023 const char *creds_path
,
2025 const char *log_namespace
,
2026 unsigned long mount_propagation_flag
,
2027 VeritySettings
*verity
,
2028 const MountImage
*extension_images
,
2029 size_t n_extension_images
,
2030 const ImagePolicy
*extension_image_policy
,
2031 char **extension_directories
,
2032 const char *propagate_dir
,
2033 const char *incoming_dir
,
2034 const char *extension_dir
,
2035 const char *notify_socket
,
2036 const char *host_os_release_stage
,
2037 char **error_path
) {
2039 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
2040 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
2041 _cleanup_strv_free_
char **hierarchies
= NULL
;
2042 MountEntry
*m
= NULL
, *mounts
= NULL
;
2043 bool require_prefix
= false, setup_propagate
= false;
2045 DissectImageFlags dissect_image_flags
=
2046 DISSECT_IMAGE_GENERIC_ROOT
|
2047 DISSECT_IMAGE_REQUIRE_ROOT
|
2048 DISSECT_IMAGE_DISCARD_ON_LOOP
|
2049 DISSECT_IMAGE_RELAX_VAR_CHECK
|
2050 DISSECT_IMAGE_FSCK
|
2051 DISSECT_IMAGE_USR_NO_ROOT
|
2052 DISSECT_IMAGE_GROWFS
|
2053 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
2054 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
2060 /* Make sure that all mknod(), mkdir() calls we do are unaffected by the umask, and the access modes
2061 * we configure take effect */
2062 BLOCK_WITH_UMASK(0000);
2064 if (!isempty(propagate_dir
) && !isempty(incoming_dir
))
2065 setup_propagate
= true;
2067 if (mount_propagation_flag
== 0)
2068 mount_propagation_flag
= MS_SHARED
;
2071 /* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */
2072 if (root_read_only(read_only_paths
,
2073 ns_info
->protect_system
) &&
2074 home_read_only(read_only_paths
, inaccessible_paths
, empty_directories
,
2075 bind_mounts
, n_bind_mounts
, temporary_filesystems
, n_temporary_filesystems
,
2076 ns_info
->protect_home
) &&
2077 strv_isempty(read_write_paths
))
2078 dissect_image_flags
|= DISSECT_IMAGE_READ_ONLY
;
2080 SET_FLAG(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
, verity
&& verity
->data_path
);
2082 r
= loop_device_make_by_path(
2084 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_DEVICE_READ_ONLY
) ? O_RDONLY
: -1 /* < 0 means writable if possible, read-only as fallback */,
2085 /* sector_size= */ UINT32_MAX
,
2086 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
2090 return log_debug_errno(r
, "Failed to create loop device for root image: %m");
2092 r
= dissect_loop_device(
2095 root_image_mount_options
,
2097 dissect_image_flags
,
2100 return log_debug_errno(r
, "Failed to dissect image: %m");
2102 r
= dissected_image_load_verity_sig_partition(
2109 r
= dissected_image_decrypt(
2113 dissect_image_flags
);
2115 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
2119 root
= root_directory
;
2121 /* /run/systemd should have been created by PID 1 early on already, but in some cases, like
2122 * when running tests (test-execute), it might not have been created yet so let's make sure
2123 * we create it if it doesn't already exist. */
2124 (void) mkdir_p_label("/run/systemd", 0755);
2126 /* Always create the mount namespace in a temporary directory, instead of operating directly
2127 * in the root. The temporary directory prevents any mounts from being potentially obscured
2128 * my other mounts we already applied. We use the same mount point for all images, which is
2129 * safe, since they all live in their own namespaces after all, and hence won't see each
2130 * other. (Note: this directory is also created by PID 1 early on, we create it here for
2131 * similar reasons as /run/systemd/ first.) */
2132 root
= "/run/systemd/mount-rootfs";
2133 (void) mkdir_label(root
, 0555);
2135 require_prefix
= true;
2138 if (n_extension_images
> 0 || !strv_isempty(extension_directories
)) {
2139 /* Hierarchy population needs to be done for sysext and confext extension images */
2140 r
= parse_env_extension_hierarchies(&hierarchies
, "SYSTEMD_SYSEXT_AND_CONFEXT_HIERARCHIES");
2145 n_mounts
= namespace_calculate_mounts(
2154 n_temporary_filesystems
,
2157 strv_length(extension_directories
),
2158 strv_length(hierarchies
),
2159 tmp_dir
, var_tmp_dir
,
2164 host_os_release_stage
);
2167 m
= mounts
= new0(MountEntry
, n_mounts
);
2171 r
= append_access_mounts(&m
, read_write_paths
, READWRITE
, require_prefix
);
2175 r
= append_access_mounts(&m
, read_only_paths
, READONLY
, require_prefix
);
2179 r
= append_access_mounts(&m
, inaccessible_paths
, INACCESSIBLE
, require_prefix
);
2183 r
= append_access_mounts(&m
, exec_paths
, EXEC
, require_prefix
);
2187 r
= append_access_mounts(&m
, no_exec_paths
, NOEXEC
, require_prefix
);
2191 r
= append_empty_dir_mounts(&m
, empty_directories
);
2195 r
= append_bind_mounts(&m
, bind_mounts
, n_bind_mounts
);
2199 r
= append_tmpfs_mounts(&m
, temporary_filesystems
, n_temporary_filesystems
);
2204 bool ro
= streq(tmp_dir
, RUN_SYSTEMD_EMPTY
);
2206 *(m
++) = (MountEntry
) {
2207 .path_const
= "/tmp",
2208 .mode
= ro
? PRIVATE_TMP_READONLY
: PRIVATE_TMP
,
2209 .source_const
= tmp_dir
,
2214 bool ro
= streq(var_tmp_dir
, RUN_SYSTEMD_EMPTY
);
2216 *(m
++) = (MountEntry
) {
2217 .path_const
= "/var/tmp",
2218 .mode
= ro
? PRIVATE_TMP_READONLY
: PRIVATE_TMP
,
2219 .source_const
= var_tmp_dir
,
2223 r
= append_mount_images(&m
, mount_images
, n_mount_images
);
2227 r
= append_extensions(&m
, root
, extension_dir
, hierarchies
, extension_images
, n_extension_images
, extension_directories
);
2231 if (ns_info
->private_dev
)
2232 *(m
++) = (MountEntry
) {
2233 .path_const
= "/dev",
2234 .mode
= PRIVATE_DEV
,
2235 .flags
= DEV_MOUNT_OPTIONS
,
2238 /* In case /proc is successfully mounted with pid tree subset only (ProcSubset=pid), the
2239 protective mounts to non-pid /proc paths would fail. But the pid only option may have
2240 failed gracefully, so let's try the mounts but it's not fatal if they don't succeed. */
2241 bool ignore_protect_proc
= ns_info
->ignore_protect_paths
|| ns_info
->proc_subset
== PROC_SUBSET_PID
;
2242 if (ns_info
->protect_kernel_tunables
) {
2243 r
= append_static_mounts(&m
,
2244 protect_kernel_tunables_proc_table
,
2245 ELEMENTSOF(protect_kernel_tunables_proc_table
),
2246 ignore_protect_proc
);
2250 r
= append_static_mounts(&m
,
2251 protect_kernel_tunables_sys_table
,
2252 ELEMENTSOF(protect_kernel_tunables_sys_table
),
2253 ns_info
->ignore_protect_paths
);
2258 if (ns_info
->protect_kernel_modules
) {
2259 r
= append_static_mounts(&m
,
2260 protect_kernel_modules_table
,
2261 ELEMENTSOF(protect_kernel_modules_table
),
2262 ns_info
->ignore_protect_paths
);
2267 if (ns_info
->protect_kernel_logs
) {
2268 r
= append_static_mounts(&m
,
2269 protect_kernel_logs_proc_table
,
2270 ELEMENTSOF(protect_kernel_logs_proc_table
),
2271 ignore_protect_proc
);
2275 r
= append_static_mounts(&m
,
2276 protect_kernel_logs_dev_table
,
2277 ELEMENTSOF(protect_kernel_logs_dev_table
),
2278 ns_info
->ignore_protect_paths
);
2283 if (ns_info
->protect_control_groups
)
2284 *(m
++) = (MountEntry
) {
2285 .path_const
= "/sys/fs/cgroup",
2289 r
= append_protect_home(&m
, ns_info
->protect_home
, ns_info
->ignore_protect_paths
);
2293 r
= append_protect_system(&m
, ns_info
->protect_system
, false);
2297 if (namespace_info_mount_apivfs(ns_info
)) {
2298 r
= append_static_mounts(&m
,
2300 ELEMENTSOF(apivfs_table
),
2301 ns_info
->ignore_protect_paths
);
2306 /* Note, if proc is mounted with subset=pid then neither of the
2307 * two paths will exist, i.e. they are implicitly protected by
2308 * the mount option. */
2309 if (ns_info
->protect_hostname
) {
2310 *(m
++) = (MountEntry
) {
2311 .path_const
= "/proc/sys/kernel/hostname",
2313 .ignore
= ignore_protect_proc
,
2315 *(m
++) = (MountEntry
) {
2316 .path_const
= "/proc/sys/kernel/domainname",
2318 .ignore
= ignore_protect_proc
,
2322 if (ns_info
->private_network
)
2323 *(m
++) = (MountEntry
) {
2324 .path_const
= "/sys",
2325 .mode
= PRIVATE_SYSFS
,
2328 if (ns_info
->private_ipc
)
2329 *(m
++) = (MountEntry
) {
2330 .path_const
= "/dev/mqueue",
2332 .flags
= MS_NOSUID
| MS_NODEV
| MS_NOEXEC
| MS_RELATIME
,
2336 /* If our service has a credentials store configured, then bind that one in, but hide
2337 * everything else. */
2339 *(m
++) = (MountEntry
) {
2340 .path_const
= "/run/credentials",
2343 .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
,
2344 .flags
= MS_NODEV
|MS_STRICTATIME
|MS_NOSUID
|MS_NOEXEC
,
2347 /* If we have mount fd for credentials directory, then it will be mounted after
2348 * namespace is set up. So, here we only create the mount point. */
2351 *(m
++) = (MountEntry
) {
2352 .path_const
= creds_path
,
2355 .source_const
= creds_path
,
2359 *(m
++) = (MountEntry
) {
2360 .path_const
= creds_path
,
2364 /* If our service has no credentials store configured, then make the whole
2365 * credentials tree inaccessible wholesale. */
2367 *(m
++) = (MountEntry
) {
2368 .path_const
= "/run/credentials",
2369 .mode
= INACCESSIBLE
,
2374 if (log_namespace
) {
2375 _cleanup_free_
char *q
= NULL
;
2377 q
= strjoin("/run/systemd/journal.", log_namespace
);
2383 *(m
++) = (MountEntry
) {
2384 .path_const
= "/run/systemd/journal",
2385 .mode
= BIND_MOUNT_RECURSIVE
,
2387 .source_malloc
= TAKE_PTR(q
),
2391 /* Will be used to add bind mounts at runtime */
2392 if (setup_propagate
)
2393 *(m
++) = (MountEntry
) {
2394 .source_const
= propagate_dir
,
2395 .path_const
= incoming_dir
,
2401 *(m
++) = (MountEntry
) {
2402 .path_const
= notify_socket
,
2403 .source_const
= notify_socket
,
2408 if (host_os_release_stage
)
2409 *(m
++) = (MountEntry
) {
2410 .path_const
= "/run/host/.os-release-stage/",
2411 .source_const
= host_os_release_stage
,
2414 .ignore
= true, /* Live copy, don't hard-fail if it goes missing */
2417 assert(mounts
+ n_mounts
== m
);
2419 /* Prepend the root directory where that's necessary */
2420 r
= prefix_where_needed(mounts
, n_mounts
, root
);
2424 drop_unused_mounts(root
, mounts
, &n_mounts
);
2427 /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */
2429 if (unshare(CLONE_NEWNS
) < 0) {
2430 r
= log_debug_errno(errno
, "Failed to unshare the mount namespace: %m");
2431 if (ERRNO_IS_PRIVILEGE(r
) ||
2432 ERRNO_IS_NOT_SUPPORTED(r
))
2433 /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter
2434 * in place that doesn't allow us to create namespaces (or a missing cap), then
2435 * propagate a recognizable error back, which the caller can use to detect this case
2436 * (and only this) and optionally continue without namespacing applied. */
2442 /* Create the source directory to allow runtime propagation of mounts */
2443 if (setup_propagate
)
2444 (void) mkdir_p(propagate_dir
, 0600);
2446 if (n_extension_images
> 0 || !strv_isempty(extension_directories
))
2447 /* ExtensionImages/Directories mountpoint directories will be created while parsing the
2448 * mounts to create, so have the parent ready */
2449 (void) mkdir_p(extension_dir
, 0600);
2451 /* Remount / as SLAVE so that nothing now mounted in the namespace
2452 * shows up in the parent */
2453 if (mount(NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
) < 0) {
2454 r
= log_debug_errno(errno
, "Failed to remount '/' as SLAVE: %m");
2459 /* A root image is specified, mount it to the right place */
2460 r
= dissected_image_mount(dissected_image
, root
, UID_INVALID
, UID_INVALID
, dissect_image_flags
);
2462 log_debug_errno(r
, "Failed to mount root image: %m");
2466 /* Now release the block device lock, so that udevd is free to call BLKRRPART on the device
2468 r
= loop_device_flock(loop_device
, LOCK_UN
);
2470 log_debug_errno(r
, "Failed to release lock on loopback block device: %m");
2474 r
= dissected_image_relinquish(dissected_image
);
2476 log_debug_errno(r
, "Failed to relinquish dissected image: %m");
2480 } else if (root_directory
) {
2482 /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
2483 r
= path_is_mount_point(root
, NULL
, AT_SYMLINK_FOLLOW
);
2485 log_debug_errno(r
, "Failed to detect that %s is a mount point or not: %m", root
);
2489 r
= mount_nofollow_verbose(LOG_DEBUG
, root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2495 /* Let's mount the main root directory to the root directory to use */
2496 r
= mount_nofollow_verbose(LOG_DEBUG
, "/", root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2501 /* Try to set up the new root directory before mounting anything else there. */
2502 if (root_image
|| root_directory
)
2503 (void) base_filesystem_create(root
, UID_INVALID
, GID_INVALID
);
2505 /* Now make the magic happen */
2506 r
= apply_mounts(root
, mount_image_policy
, extension_image_policy
, ns_info
, mounts
, &n_mounts
, symlinks
, error_path
);
2510 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
2511 r
= mount_switch_root(root
, /* mount_propagation_flag = */ 0);
2512 if (r
== -EINVAL
&& root_directory
) {
2513 /* If we are using root_directory and we don't have privileges (ie: user manager in a user
2514 * namespace) and the root_directory is already a mount point in the parent namespace,
2515 * MS_MOVE will fail as we don't have permission to change it (with EINVAL rather than
2516 * EPERM). Attempt to bind-mount it over itself (like we do above if it's not already a
2517 * mount point) and try again. */
2518 r
= mount_nofollow_verbose(LOG_DEBUG
, root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2521 r
= mount_switch_root(root
, /* mount_propagation_flag = */ 0);
2524 log_debug_errno(r
, "Failed to mount root with MS_MOVE: %m");
2528 /* Remount / as the desired mode. Note that this will not reestablish propagation from our side to
2529 * the host, since what's disconnected is disconnected. */
2530 if (mount(NULL
, "/", NULL
, mount_propagation_flag
| MS_REC
, NULL
) < 0) {
2531 r
= log_debug_errno(errno
, "Failed to remount '/' with desired mount flags: %m");
2535 /* bind_mount_in_namespace() will MS_MOVE into that directory, and that's only
2536 * supported for non-shared mounts. This needs to happen after remounting / or it will fail. */
2537 if (setup_propagate
) {
2538 r
= mount(NULL
, incoming_dir
, NULL
, MS_SLAVE
, NULL
);
2540 log_error_errno(r
, "Failed to remount %s with MS_SLAVE: %m", incoming_dir
);
2549 for (m
= mounts
; m
< mounts
+ n_mounts
; m
++)
2550 mount_entry_done(m
);
2557 void bind_mount_free_many(BindMount
*b
, size_t n
) {
2558 assert(b
|| n
== 0);
2560 for (size_t i
= 0; i
< n
; i
++) {
2562 free(b
[i
].destination
);
2568 int bind_mount_add(BindMount
**b
, size_t *n
, const BindMount
*item
) {
2569 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
2576 s
= strdup(item
->source
);
2580 d
= strdup(item
->destination
);
2584 c
= reallocarray(*b
, *n
+ 1, sizeof(BindMount
));
2590 c
[(*n
) ++] = (BindMount
) {
2591 .source
= TAKE_PTR(s
),
2592 .destination
= TAKE_PTR(d
),
2593 .read_only
= item
->read_only
,
2594 .nosuid
= item
->nosuid
,
2595 .recursive
= item
->recursive
,
2596 .ignore_enoent
= item
->ignore_enoent
,
2602 MountImage
* mount_image_free_many(MountImage
*m
, size_t *n
) {
2604 assert(m
|| *n
== 0);
2606 for (size_t i
= 0; i
< *n
; i
++) {
2608 free(m
[i
].destination
);
2609 mount_options_free_all(m
[i
].mount_options
);
2617 int mount_image_add(MountImage
**m
, size_t *n
, const MountImage
*item
) {
2618 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
2619 _cleanup_(mount_options_free_allp
) MountOptions
*options
= NULL
;
2626 s
= strdup(item
->source
);
2630 if (item
->destination
) {
2631 d
= strdup(item
->destination
);
2636 LIST_FOREACH(mount_options
, i
, item
->mount_options
) {
2637 _cleanup_(mount_options_free_allp
) MountOptions
*o
= NULL
;
2639 o
= new(MountOptions
, 1);
2643 *o
= (MountOptions
) {
2644 .partition_designator
= i
->partition_designator
,
2645 .options
= strdup(i
->options
),
2650 LIST_APPEND(mount_options
, options
, TAKE_PTR(o
));
2653 c
= reallocarray(*m
, *n
+ 1, sizeof(MountImage
));
2659 c
[(*n
) ++] = (MountImage
) {
2660 .source
= TAKE_PTR(s
),
2661 .destination
= TAKE_PTR(d
),
2662 .mount_options
= TAKE_PTR(options
),
2663 .ignore_enoent
= item
->ignore_enoent
,
2670 void temporary_filesystem_free_many(TemporaryFileSystem
*t
, size_t n
) {
2671 assert(t
|| n
== 0);
2673 for (size_t i
= 0; i
< n
; i
++) {
2681 int temporary_filesystem_add(
2682 TemporaryFileSystem
**t
,
2685 const char *options
) {
2687 _cleanup_free_
char *p
= NULL
, *o
= NULL
;
2688 TemporaryFileSystem
*c
;
2698 if (!isempty(options
)) {
2699 o
= strdup(options
);
2704 c
= reallocarray(*t
, *n
+ 1, sizeof(TemporaryFileSystem
));
2710 c
[(*n
) ++] = (TemporaryFileSystem
) {
2711 .path
= TAKE_PTR(p
),
2712 .options
= TAKE_PTR(o
),
2718 static int make_tmp_prefix(const char *prefix
) {
2719 _cleanup_free_
char *t
= NULL
;
2720 _cleanup_close_
int fd
= -EBADF
;
2723 /* Don't do anything unless we know the dir is actually missing */
2724 r
= access(prefix
, F_OK
);
2727 if (errno
!= ENOENT
)
2731 r
= mkdir_parents(prefix
, 0755);
2735 r
= tempfn_random(prefix
, NULL
, &t
);
2739 /* umask will corrupt this access mode, but that doesn't matter, we need to call chmod() anyway for
2740 * the suid bit, below. */
2741 fd
= open_mkdir_at(AT_FDCWD
, t
, O_EXCL
|O_CLOEXEC
, 0777);
2745 r
= RET_NERRNO(fchmod(fd
, 01777));
2751 r
= RET_NERRNO(rename(t
, prefix
));
2754 return r
== -EEXIST
? 0 : r
; /* it's fine if someone else created the dir by now */
2761 static int setup_one_tmp_dir(const char *id
, const char *prefix
, char **path
, char **tmp_path
) {
2762 _cleanup_free_
char *x
= NULL
;
2763 _cleanup_free_
char *y
= NULL
;
2772 /* We include the boot id in the directory so that after a
2773 * reboot we can easily identify obsolete directories. */
2775 r
= sd_id128_get_boot(&boot_id
);
2779 x
= strjoin(prefix
, "/systemd-private-", SD_ID128_TO_STRING(boot_id
), "-", id
, "-XXXXXX");
2783 r
= make_tmp_prefix(prefix
);
2789 if (errno
== EROFS
|| ERRNO_IS_DISK_SPACE(errno
))
2796 y
= strjoin(x
, "/tmp");
2801 if (mkdir(y
, 0777 | S_ISVTX
) < 0)
2804 r
= label_fix_full(AT_FDCWD
, y
, prefix
, 0);
2809 *tmp_path
= TAKE_PTR(y
);
2811 /* Trouble: we failed to create the directory. Instead of failing, let's simulate /tmp being
2812 * read-only. This way the service will get the EROFS result as if it was writing to the real
2815 r
= mkdir_p(RUN_SYSTEMD_EMPTY
, 0500);
2819 r
= free_and_strdup(&x
, RUN_SYSTEMD_EMPTY
);
2824 *path
= TAKE_PTR(x
);
2828 int setup_tmp_dirs(const char *id
, char **tmp_dir
, char **var_tmp_dir
) {
2829 _cleanup_(namespace_cleanup_tmpdirp
) char *a
= NULL
;
2830 _cleanup_(rmdir_and_freep
) char *a_tmp
= NULL
;
2836 assert(var_tmp_dir
);
2838 r
= setup_one_tmp_dir(id
, "/tmp", &a
, &a_tmp
);
2842 r
= setup_one_tmp_dir(id
, "/var/tmp", &b
, NULL
);
2846 a_tmp
= mfree(a_tmp
); /* avoid rmdir */
2847 *tmp_dir
= TAKE_PTR(a
);
2848 *var_tmp_dir
= TAKE_PTR(b
);
2853 int setup_shareable_ns(int ns_storage_socket
[static 2], unsigned long nsflag
) {
2854 _cleanup_close_
int ns
= -EBADF
;
2856 const char *ns_name
, *ns_path
;
2858 assert(ns_storage_socket
);
2859 assert(ns_storage_socket
[0] >= 0);
2860 assert(ns_storage_socket
[1] >= 0);
2862 ns_name
= namespace_single_flag_to_string(nsflag
);
2865 /* We use the passed socketpair as a storage buffer for our
2866 * namespace reference fd. Whatever process runs this first
2867 * shall create a new namespace, all others should just join
2868 * it. To serialize that we use a file lock on the socket
2871 * It's a bit crazy, but hey, works great! */
2873 r
= posix_lock(ns_storage_socket
[0], LOCK_EX
);
2877 CLEANUP_POSIX_UNLOCK(ns_storage_socket
[0]);
2879 ns
= receive_one_fd(ns_storage_socket
[0], MSG_PEEK
|MSG_DONTWAIT
);
2881 /* Yay, found something, so let's join the namespace */
2882 r
= RET_NERRNO(setns(ns
, nsflag
));
2892 /* Nothing stored yet, so let's create a new namespace. */
2894 if (unshare(nsflag
) < 0)
2897 (void) loopback_setup();
2899 ns_path
= strjoina("/proc/self/ns/", ns_name
);
2900 ns
= open(ns_path
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
2904 r
= send_one_fd(ns_storage_socket
[1], ns
, MSG_DONTWAIT
);
2911 int open_shareable_ns_path(int ns_storage_socket
[static 2], const char *path
, unsigned long nsflag
) {
2912 _cleanup_close_
int ns
= -EBADF
;
2915 assert(ns_storage_socket
);
2916 assert(ns_storage_socket
[0] >= 0);
2917 assert(ns_storage_socket
[1] >= 0);
2920 /* If the storage socket doesn't contain a ns fd yet, open one via the file system and store it in
2921 * it. This is supposed to be called ahead of time, i.e. before setup_shareable_ns() which will
2922 * allocate a new anonymous ns if needed. */
2924 r
= posix_lock(ns_storage_socket
[0], LOCK_EX
);
2928 CLEANUP_POSIX_UNLOCK(ns_storage_socket
[0]);
2930 ns
= receive_one_fd(ns_storage_socket
[0], MSG_PEEK
|MSG_DONTWAIT
);
2936 /* Nothing stored yet. Open the file from the file system. */
2938 ns
= open(path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
2942 r
= fd_is_ns(ns
, nsflag
);
2945 if (r
< 0 && r
!= -EUCLEAN
) /* EUCLEAN: we don't know */
2948 r
= send_one_fd(ns_storage_socket
[1], ns
, MSG_DONTWAIT
);
2955 bool ns_type_supported(NamespaceType type
) {
2956 const char *t
, *ns_proc
;
2958 t
= namespace_type_to_string(type
);
2959 if (!t
) /* Don't know how to translate this? Then it's not supported */
2962 ns_proc
= strjoina("/proc/self/ns/", t
);
2963 return access(ns_proc
, F_OK
) == 0;
2966 static const char *const protect_home_table
[_PROTECT_HOME_MAX
] = {
2967 [PROTECT_HOME_NO
] = "no",
2968 [PROTECT_HOME_YES
] = "yes",
2969 [PROTECT_HOME_READ_ONLY
] = "read-only",
2970 [PROTECT_HOME_TMPFS
] = "tmpfs",
2973 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home
, ProtectHome
, PROTECT_HOME_YES
);
2975 static const char *const protect_system_table
[_PROTECT_SYSTEM_MAX
] = {
2976 [PROTECT_SYSTEM_NO
] = "no",
2977 [PROTECT_SYSTEM_YES
] = "yes",
2978 [PROTECT_SYSTEM_FULL
] = "full",
2979 [PROTECT_SYSTEM_STRICT
] = "strict",
2982 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system
, ProtectSystem
, PROTECT_SYSTEM_YES
);
2984 static const char* const namespace_type_table
[] = {
2985 [NAMESPACE_MOUNT
] = "mnt",
2986 [NAMESPACE_CGROUP
] = "cgroup",
2987 [NAMESPACE_UTS
] = "uts",
2988 [NAMESPACE_IPC
] = "ipc",
2989 [NAMESPACE_USER
] = "user",
2990 [NAMESPACE_PID
] = "pid",
2991 [NAMESPACE_NET
] = "net",
2992 [NAMESPACE_TIME
] = "time",
2995 DEFINE_STRING_TABLE_LOOKUP(namespace_type
, NamespaceType
);
2997 static const char* const protect_proc_table
[_PROTECT_PROC_MAX
] = {
2998 [PROTECT_PROC_DEFAULT
] = "default",
2999 [PROTECT_PROC_NOACCESS
] = "noaccess",
3000 [PROTECT_PROC_INVISIBLE
] = "invisible",
3001 [PROTECT_PROC_PTRACEABLE
] = "ptraceable",
3004 DEFINE_STRING_TABLE_LOOKUP(protect_proc
, ProtectProc
);
3006 static const char* const proc_subset_table
[_PROC_SUBSET_MAX
] = {
3007 [PROC_SUBSET_ALL
] = "all",
3008 [PROC_SUBSET_PID
] = "pid",
3011 DEFINE_STRING_TABLE_LOOKUP(proc_subset
, ProcSubset
);