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! */
61 MOUNT_PRIVATE_TMP_READ_ONLY
,
74 MOUNT_EXTENSION_DIRECTORY
, /* Bind-mounted outside the root directory, and used by subsequent mounts */
75 MOUNT_EXTENSION_IMAGE
, /* Mounted outside the root directory, and used by subsequent mounts */
77 MOUNT_READ_WRITE_IMPLICIT
, /* Should have the lowest priority. */
79 _MOUNT_MODE_INVALID
= -EINVAL
,
82 typedef enum MountEntryState
{
86 _MOUNT_ENTRY_STATE_MAX
,
87 _MOUNT_ENTRY_STATE_INVALID
= -EINVAL
,
90 typedef struct MountEntry
{
91 const char *path_const
; /* Memory allocated on stack or static */
93 bool ignore
:1; /* Ignore if path does not exist? */
94 bool has_prefix
:1; /* Already is prefixed by the root dir? */
95 bool read_only
:1; /* Shall this mount point be read-only? */
96 bool nosuid
:1; /* Shall set MS_NOSUID on the mount itself */
97 bool noexec
:1; /* Shall set MS_NOEXEC on the mount itself */
98 bool exec
:1; /* Shall clear MS_NOEXEC on the mount itself */
99 MountEntryState state
; /* Whether it was already processed or skipped */
100 char *path_malloc
; /* Use this instead of 'path_const' if we had to allocate memory */
101 const char *unprefixed_path_const
; /* If the path was amended with a prefix, these will save the original */
102 char *unprefixed_path_malloc
;
103 const char *source_const
; /* The source path, for bind mounts or images */
105 const char *options_const
;/* Mount options for tmpfs */
106 char *options_malloc
;
107 unsigned long flags
; /* Mount flags used by EMPTY_DIR and TMPFS. Do not include MS_RDONLY here, but please use read_only. */
109 LIST_HEAD(MountOptions
, image_options_const
);
110 char **overlay_layers
;
113 typedef struct MountList
{
118 /* 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
119 * something there already. These mounts are hence overridden by any other explicitly configured mounts. */
120 static const MountEntry apivfs_table
[] = {
121 { "/proc", MOUNT_PROCFS
, false },
122 { "/dev", MOUNT_BIND_DEV
, false },
123 { "/sys", MOUNT_BIND_SYSFS
, false },
124 { "/run", MOUNT_RUN
, false, .options_const
= "mode=0755" TMPFS_LIMITS_RUN
, .flags
= MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
},
127 /* ProtectKernelTunables= option and the related filesystem APIs */
128 static const MountEntry protect_kernel_tunables_proc_table
[] = {
129 { "/proc/acpi", MOUNT_READ_ONLY
, true },
130 { "/proc/apm", MOUNT_READ_ONLY
, true }, /* Obsolete API, there's no point in permitting access to this, ever */
131 { "/proc/asound", MOUNT_READ_ONLY
, true },
132 { "/proc/bus", MOUNT_READ_ONLY
, true },
133 { "/proc/fs", MOUNT_READ_ONLY
, true },
134 { "/proc/irq", MOUNT_READ_ONLY
, true },
135 { "/proc/kallsyms", MOUNT_INACCESSIBLE
, true },
136 { "/proc/kcore", MOUNT_INACCESSIBLE
, true },
137 { "/proc/latency_stats", MOUNT_READ_ONLY
, true },
138 { "/proc/mtrr", MOUNT_READ_ONLY
, true },
139 { "/proc/scsi", MOUNT_READ_ONLY
, true },
140 { "/proc/sys", MOUNT_READ_ONLY
, true },
141 { "/proc/sysrq-trigger", MOUNT_READ_ONLY
, true },
142 { "/proc/timer_stats", MOUNT_READ_ONLY
, true },
145 static const MountEntry protect_kernel_tunables_sys_table
[] = {
146 { "/sys", MOUNT_READ_ONLY
, false },
147 { "/sys/fs/bpf", MOUNT_READ_ONLY
, true },
148 { "/sys/fs/cgroup", MOUNT_READ_WRITE_IMPLICIT
, false }, /* READ_ONLY is set by ProtectControlGroups= option */
149 { "/sys/fs/selinux", MOUNT_READ_WRITE_IMPLICIT
, true },
150 { "/sys/kernel/debug", MOUNT_READ_ONLY
, true },
151 { "/sys/kernel/tracing", MOUNT_READ_ONLY
, true },
154 /* ProtectKernelModules= option */
155 static const MountEntry protect_kernel_modules_table
[] = {
156 { "/usr/lib/modules", MOUNT_INACCESSIBLE
, true },
159 /* ProtectKernelLogs= option */
160 static const MountEntry protect_kernel_logs_proc_table
[] = {
161 { "/proc/kmsg", MOUNT_INACCESSIBLE
, true },
164 static const MountEntry protect_kernel_logs_dev_table
[] = {
165 { "/dev/kmsg", MOUNT_INACCESSIBLE
, true },
169 * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of
170 * system should be protected by ProtectSystem=
172 static const MountEntry protect_home_read_only_table
[] = {
173 { "/home", MOUNT_READ_ONLY
, true },
174 { "/run/user", MOUNT_READ_ONLY
, true },
175 { "/root", MOUNT_READ_ONLY
, true },
178 /* ProtectHome=tmpfs table */
179 static const MountEntry protect_home_tmpfs_table
[] = {
180 { "/home", MOUNT_TMPFS
, true, .read_only
= true, .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
181 { "/run/user", MOUNT_TMPFS
, true, .read_only
= true, .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
182 { "/root", MOUNT_TMPFS
, true, .read_only
= true, .options_const
= "mode=0700" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
185 /* ProtectHome=yes table */
186 static const MountEntry protect_home_yes_table
[] = {
187 { "/home", MOUNT_INACCESSIBLE
, true },
188 { "/run/user", MOUNT_INACCESSIBLE
, true },
189 { "/root", MOUNT_INACCESSIBLE
, true },
192 /* ProtectSystem=yes table */
193 static const MountEntry protect_system_yes_table
[] = {
194 { "/usr", MOUNT_READ_ONLY
, false },
195 { "/boot", MOUNT_READ_ONLY
, true },
196 { "/efi", MOUNT_READ_ONLY
, true },
199 /* ProtectSystem=full includes ProtectSystem=yes */
200 static const MountEntry protect_system_full_table
[] = {
201 { "/usr", MOUNT_READ_ONLY
, false },
202 { "/boot", MOUNT_READ_ONLY
, true },
203 { "/efi", MOUNT_READ_ONLY
, true },
204 { "/etc", MOUNT_READ_ONLY
, false },
207 /* ProtectSystem=strict table. In this strict mode, we mount everything read-only, except for /proc, /dev,
208 * /sys which are the kernel API VFS, which are left writable, but PrivateDevices= + ProtectKernelTunables=
209 * protect those, and these options should be fully orthogonal. (And of course /home and friends are also
210 * left writable, as ProtectHome= shall manage those, orthogonally).
212 static const MountEntry protect_system_strict_table
[] = {
213 { "/", MOUNT_READ_ONLY
, false },
214 { "/proc", MOUNT_READ_WRITE_IMPLICIT
, false }, /* ProtectKernelTunables= */
215 { "/sys", MOUNT_READ_WRITE_IMPLICIT
, false }, /* ProtectKernelTunables= */
216 { "/dev", MOUNT_READ_WRITE_IMPLICIT
, false }, /* PrivateDevices= */
217 { "/home", MOUNT_READ_WRITE_IMPLICIT
, true }, /* ProtectHome= */
218 { "/run/user", MOUNT_READ_WRITE_IMPLICIT
, true }, /* ProtectHome= */
219 { "/root", MOUNT_READ_WRITE_IMPLICIT
, true }, /* ProtectHome= */
222 /* ProtectHostname=yes able */
223 static const MountEntry protect_hostname_table
[] = {
224 { "/proc/sys/kernel/hostname", MOUNT_READ_ONLY
, false },
225 { "/proc/sys/kernel/domainname", MOUNT_READ_ONLY
, false },
228 static const char * const mount_mode_table
[_MOUNT_MODE_MAX
] = {
229 [MOUNT_INACCESSIBLE
] = "inaccessible",
230 [MOUNT_OVERLAY
] = "overlay",
231 [MOUNT_IMAGE
] = "image",
232 [MOUNT_BIND
] = "bind",
233 [MOUNT_BIND_RECURSIVE
] = "bind-recursive",
234 [MOUNT_PRIVATE_TMP
] = "private-tmp",
235 [MOUNT_PRIVATE_TMP_READ_ONLY
] = "private-tmp-read-only",
236 [MOUNT_PRIVATE_DEV
] = "private-dev",
237 [MOUNT_BIND_DEV
] = "bind-dev",
238 [MOUNT_EMPTY_DIR
] = "empty-dir",
239 [MOUNT_PRIVATE_SYSFS
] = "private-sysfs",
240 [MOUNT_BIND_SYSFS
] = "bind-sysfs",
241 [MOUNT_PROCFS
] = "procfs",
242 [MOUNT_READ_ONLY
] = "read-only",
243 [MOUNT_READ_WRITE
] = "read-write",
244 [MOUNT_NOEXEC
] = "noexec",
245 [MOUNT_EXEC
] = "exec",
246 [MOUNT_TMPFS
] = "tmpfs",
248 [MOUNT_EXTENSION_DIRECTORY
] = "extension-directory",
249 [MOUNT_EXTENSION_IMAGE
] = "extension-image",
250 [MOUNT_MQUEUEFS
] = "mqueuefs",
251 [MOUNT_READ_WRITE_IMPLICIT
] = "read-write-implicit",
254 /* Helper struct for naming simplicity and reusability */
255 static const struct {
256 const char *level_env
;
257 const char *level_env_print
;
258 } image_class_info
[_IMAGE_CLASS_MAX
] = {
260 .level_env
= "SYSEXT_LEVEL",
261 .level_env_print
= " SYSEXT_LEVEL=",
264 .level_env
= "CONFEXT_LEVEL",
265 .level_env_print
= " CONFEXT_LEVEL=",
269 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(mount_mode
, MountMode
);
271 static const char *mount_entry_path(const MountEntry
*p
) {
274 /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that,
275 * otherwise the stack/static ->path field is returned. */
277 return p
->path_malloc
?: p
->path_const
;
280 static const char *mount_entry_unprefixed_path(const MountEntry
*p
) {
283 /* Returns the unprefixed path (ie: before prefix_where_needed() ran), if any */
285 return p
->unprefixed_path_malloc
?: p
->unprefixed_path_const
?: mount_entry_path(p
);
288 static void mount_entry_consume_prefix(MountEntry
*p
, char *new_path
) {
290 assert(p
->path_malloc
|| p
->path_const
);
293 /* Saves current path in unprefixed_ variable, and takes over new_path */
295 free_and_replace(p
->unprefixed_path_malloc
, p
->path_malloc
);
296 /* If we didn't have a path on the heap, then it's a static one */
297 if (!p
->unprefixed_path_malloc
)
298 p
->unprefixed_path_const
= p
->path_const
;
299 p
->path_malloc
= new_path
;
300 p
->has_prefix
= true;
303 static bool mount_entry_read_only(const MountEntry
*p
) {
306 return p
->read_only
|| IN_SET(p
->mode
, MOUNT_READ_ONLY
, MOUNT_INACCESSIBLE
, MOUNT_PRIVATE_TMP_READ_ONLY
);
309 static bool mount_entry_noexec(const MountEntry
*p
) {
312 return p
->noexec
|| IN_SET(p
->mode
, MOUNT_NOEXEC
, MOUNT_INACCESSIBLE
, MOUNT_PRIVATE_SYSFS
, MOUNT_BIND_SYSFS
, MOUNT_PROCFS
);
315 static bool mount_entry_exec(const MountEntry
*p
) {
318 return p
->exec
|| p
->mode
== MOUNT_EXEC
;
321 static const char *mount_entry_source(const MountEntry
*p
) {
324 return p
->source_malloc
?: p
->source_const
;
327 static const char *mount_entry_options(const MountEntry
*p
) {
330 return p
->options_malloc
?: p
->options_const
;
333 static void mount_entry_done(MountEntry
*p
) {
336 p
->path_malloc
= mfree(p
->path_malloc
);
337 p
->unprefixed_path_malloc
= mfree(p
->unprefixed_path_malloc
);
338 p
->source_malloc
= mfree(p
->source_malloc
);
339 p
->options_malloc
= mfree(p
->options_malloc
);
340 p
->overlay_layers
= strv_free(p
->overlay_layers
);
343 static void mount_list_done(MountList
*ml
) {
346 FOREACH_ARRAY(m
, ml
->mounts
, ml
->n_mounts
)
349 ml
->mounts
= mfree(ml
->mounts
);
353 static MountEntry
*mount_list_extend(MountList
*ml
) {
356 if (!GREEDY_REALLOC0(ml
->mounts
, ml
->n_mounts
+1))
359 return ml
->mounts
+ ml
->n_mounts
++;
362 static int append_access_mounts(MountList
*ml
, char **strv
, MountMode mode
, bool forcibly_require_prefix
) {
365 /* Adds a list of user-supplied READ_WRITE/READ_WRITE_IMPLICIT/READ_ONLY/INACCESSIBLE entries */
367 STRV_FOREACH(i
, strv
) {
368 bool ignore
= false, needs_prefix
= false;
371 /* Look for any prefixes */
372 if (startswith(e
, "-")) {
376 if (startswith(e
, "+")) {
381 if (!path_is_absolute(e
))
382 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Path is not absolute: %s", e
);
384 MountEntry
*me
= mount_list_extend(ml
);
386 return log_oom_debug();
392 .has_prefix
= !needs_prefix
&& !forcibly_require_prefix
,
399 static int append_empty_dir_mounts(MountList
*ml
, char **strv
) {
402 /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the
403 * "/private/" boundary directories for DynamicUser=1. */
405 STRV_FOREACH(i
, strv
) {
406 MountEntry
*me
= mount_list_extend(ml
);
408 return log_oom_debug();
412 .mode
= MOUNT_EMPTY_DIR
,
415 .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
,
416 .flags
= MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
,
423 static int append_bind_mounts(MountList
*ml
, const BindMount
*binds
, size_t n
) {
425 assert(binds
|| n
== 0);
427 FOREACH_ARRAY(b
, binds
, n
) {
428 MountEntry
*me
= mount_list_extend(ml
);
430 return log_oom_debug();
433 .path_const
= b
->destination
,
434 .mode
= b
->recursive
? MOUNT_BIND_RECURSIVE
: MOUNT_BIND
,
435 .read_only
= b
->read_only
,
437 .source_const
= b
->source
,
438 .ignore
= b
->ignore_enoent
,
445 static int append_mount_images(MountList
*ml
, const MountImage
*mount_images
, size_t n
) {
447 assert(mount_images
|| n
== 0);
449 FOREACH_ARRAY(m
, mount_images
, n
) {
450 MountEntry
*me
= mount_list_extend(ml
);
452 return log_oom_debug();
455 .path_const
= m
->destination
,
457 .source_const
= m
->source
,
458 .image_options_const
= m
->mount_options
,
459 .ignore
= m
->ignore_enoent
,
466 static int append_extensions(
469 const char *extension_dir
,
471 const MountImage
*mount_images
,
473 char **extension_directories
) {
475 char ***overlays
= NULL
;
476 size_t n_overlays
= 0;
481 if (n
== 0 && strv_isempty(extension_directories
))
484 assert(extension_dir
);
486 n_overlays
= strv_length(hierarchies
);
490 /* Prepare a list of overlays, that will have as each element a strv containing all the layers that
491 * will later be concatenated as a lowerdir= parameter for the mount operation.
492 * The overlays vector will have the same number of elements and will correspond to the
493 * hierarchies vector, so they can be iterated upon together. */
494 overlays
= new0(char**, n_overlays
);
498 CLEANUP_ARRAY(overlays
, n_overlays
, strv_free_many
);
500 /* First, prepare a mount for each image, but these won't be visible to the unit, instead
501 * they will be mounted in our propagate directory, and used as a source for the overlay. */
502 for (size_t i
= 0; i
< n
; i
++) {
503 _cleanup_free_
char *mount_point
= NULL
;
504 const MountImage
*m
= mount_images
+ i
;
506 if (asprintf(&mount_point
, "%s/%zu", extension_dir
, i
) < 0)
509 for (size_t j
= 0; hierarchies
&& hierarchies
[j
]; ++j
) {
510 char *prefixed_hierarchy
= path_join(mount_point
, hierarchies
[j
]);
511 if (!prefixed_hierarchy
)
514 r
= strv_consume(&overlays
[j
], TAKE_PTR(prefixed_hierarchy
));
519 MountEntry
*me
= mount_list_extend(ml
);
524 .path_malloc
= TAKE_PTR(mount_point
),
525 .image_options_const
= m
->mount_options
,
526 .ignore
= m
->ignore_enoent
,
527 .source_const
= m
->source
,
528 .mode
= MOUNT_EXTENSION_IMAGE
,
533 /* Secondly, extend the lowerdir= parameters with each ExtensionDirectory.
534 * Bind mount them in the same location as the ExtensionImages, so that we
535 * can check that they are valid trees (extension-release.d). */
536 STRV_FOREACH(extension_directory
, extension_directories
) {
537 _cleanup_free_
char *mount_point
= NULL
, *source
= NULL
;
538 const char *e
= *extension_directory
;
539 bool ignore_enoent
= false;
541 /* Pick up the counter where the ExtensionImages left it. */
542 if (asprintf(&mount_point
, "%s/%zu", extension_dir
, n
++) < 0)
545 /* Look for any prefixes */
546 if (startswith(e
, "-")) {
548 ignore_enoent
= true;
550 /* Ignore this for now */
551 if (startswith(e
, "+"))
558 for (size_t j
= 0; hierarchies
&& hierarchies
[j
]; ++j
) {
559 char *prefixed_hierarchy
= path_join(mount_point
, hierarchies
[j
]);
560 if (!prefixed_hierarchy
)
563 r
= strv_consume(&overlays
[j
], TAKE_PTR(prefixed_hierarchy
));
568 MountEntry
*me
= mount_list_extend(ml
);
573 .path_malloc
= TAKE_PTR(mount_point
),
574 .source_malloc
= TAKE_PTR(source
),
575 .mode
= MOUNT_EXTENSION_DIRECTORY
,
576 .ignore
= ignore_enoent
,
582 /* Then, for each hierarchy, prepare an overlay with the list of lowerdir= strings
584 for (size_t i
= 0; hierarchies
&& hierarchies
[i
]; ++i
) {
585 _cleanup_free_
char *prefixed_hierarchy
= NULL
;
587 prefixed_hierarchy
= path_join(root
, hierarchies
[i
]);
588 if (!prefixed_hierarchy
)
591 MountEntry
*me
= mount_list_extend(ml
);
596 .path_malloc
= TAKE_PTR(prefixed_hierarchy
),
597 .overlay_layers
= TAKE_PTR(overlays
[i
]),
598 .mode
= MOUNT_OVERLAY
,
600 .ignore
= true, /* If the source image doesn't set the ignore bit it will fail earlier. */
607 static int append_tmpfs_mounts(MountList
*ml
, const TemporaryFileSystem
*tmpfs
, size_t n
) {
609 assert(tmpfs
|| n
== 0);
611 FOREACH_ARRAY(t
, tmpfs
, n
) {
612 _cleanup_free_
char *o
= NULL
, *str
= NULL
;
617 if (!path_is_absolute(t
->path
))
618 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Path is not absolute: %s", t
->path
);
620 str
= strjoin("mode=0755" NESTED_TMPFS_LIMITS
",", t
->options
);
624 r
= mount_option_mangle(str
, MS_NODEV
|MS_STRICTATIME
, &flags
, &o
);
626 return log_debug_errno(r
, "Failed to parse mount option '%s': %m", str
);
628 ro
= flags
& MS_RDONLY
;
631 MountEntry
*me
= mount_list_extend(ml
);
633 return log_oom_debug();
636 .path_const
= t
->path
,
639 .options_malloc
= TAKE_PTR(o
),
647 static int append_static_mounts(MountList
*ml
, const MountEntry
*mounts
, size_t n
, bool ignore_protect
) {
649 assert(mounts
|| n
== 0);
651 /* Adds a list of static pre-defined entries */
653 FOREACH_ARRAY(m
, mounts
, n
) {
654 MountEntry
*me
= mount_list_extend(ml
);
656 return log_oom_debug();
659 .path_const
= mount_entry_path(m
),
661 .ignore
= m
->ignore
|| ignore_protect
,
668 static int append_protect_home(MountList
*ml
, ProtectHome protect_home
, bool ignore_protect
) {
671 switch (protect_home
) {
673 case PROTECT_HOME_NO
:
676 case PROTECT_HOME_READ_ONLY
:
677 return append_static_mounts(ml
, protect_home_read_only_table
, ELEMENTSOF(protect_home_read_only_table
), ignore_protect
);
679 case PROTECT_HOME_TMPFS
:
680 return append_static_mounts(ml
, protect_home_tmpfs_table
, ELEMENTSOF(protect_home_tmpfs_table
), ignore_protect
);
682 case PROTECT_HOME_YES
:
683 return append_static_mounts(ml
, protect_home_yes_table
, ELEMENTSOF(protect_home_yes_table
), ignore_protect
);
686 assert_not_reached();
690 static int append_protect_system(MountList
*ml
, ProtectSystem protect_system
, bool ignore_protect
) {
693 switch (protect_system
) {
695 case PROTECT_SYSTEM_NO
:
698 case PROTECT_SYSTEM_STRICT
:
699 return append_static_mounts(ml
, protect_system_strict_table
, ELEMENTSOF(protect_system_strict_table
), ignore_protect
);
701 case PROTECT_SYSTEM_YES
:
702 return append_static_mounts(ml
, protect_system_yes_table
, ELEMENTSOF(protect_system_yes_table
), ignore_protect
);
704 case PROTECT_SYSTEM_FULL
:
705 return append_static_mounts(ml
, protect_system_full_table
, ELEMENTSOF(protect_system_full_table
), ignore_protect
);
708 assert_not_reached();
712 static int mount_path_compare(const MountEntry
*a
, const MountEntry
*b
) {
715 /* ExtensionImages/Directories will be used by other mounts as a base, so sort them first
716 * regardless of the prefix - they are set up in the propagate directory anyway */
717 d
= -CMP(a
->mode
== MOUNT_EXTENSION_IMAGE
, b
->mode
== MOUNT_EXTENSION_IMAGE
);
720 d
= -CMP(a
->mode
== MOUNT_EXTENSION_DIRECTORY
, b
->mode
== MOUNT_EXTENSION_DIRECTORY
);
724 /* If the paths are not equal, then order prefixes first */
725 d
= path_compare(mount_entry_path(a
), mount_entry_path(b
));
729 /* If the paths are equal, check the mode */
730 return CMP((int) a
->mode
, (int) b
->mode
);
733 static int prefix_where_needed(MountList
*ml
, const char *root_directory
) {
734 /* Prefixes all paths in the bind mount table with the root directory if the entry needs that. */
738 FOREACH_ARRAY(me
, ml
->mounts
, ml
->n_mounts
) {
744 s
= path_join(root_directory
, mount_entry_path(me
));
748 mount_entry_consume_prefix(me
, s
);
754 static void drop_duplicates(MountList
*ml
) {
755 MountEntry
*f
, *t
, *previous
;
759 /* Drops duplicate entries. Expects that the array is properly ordered already. */
761 for (f
= ml
->mounts
, t
= ml
->mounts
, previous
= NULL
; f
< ml
->mounts
+ ml
->n_mounts
; f
++) {
763 /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare()
764 * above. Note that we only drop duplicates that haven't been mounted yet. */
766 path_equal(mount_entry_path(f
), mount_entry_path(previous
)) &&
767 f
->state
== MOUNT_PENDING
&& previous
->state
== MOUNT_PENDING
) {
768 log_debug("%s (%s) is duplicate.", mount_entry_path(f
), mount_mode_to_string(f
->mode
));
769 /* Propagate the flags to the remaining entry */
770 previous
->read_only
= previous
->read_only
|| mount_entry_read_only(f
);
771 previous
->noexec
= previous
->noexec
|| mount_entry_noexec(f
);
772 previous
->exec
= previous
->exec
|| mount_entry_exec(f
);
782 ml
->n_mounts
= t
- ml
->mounts
;
785 static void drop_inaccessible(MountList
*ml
) {
787 const char *clear
= NULL
;
791 /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly
792 * ordered already. */
794 for (f
= ml
->mounts
, t
= ml
->mounts
; f
< ml
->mounts
+ ml
->n_mounts
; f
++) {
796 /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop
797 * it, as inaccessible paths really should drop the entire subtree. */
798 if (clear
&& path_startswith(mount_entry_path(f
), clear
)) {
799 log_debug("%s is masked by %s.", mount_entry_path(f
), clear
);
804 clear
= f
->mode
== MOUNT_INACCESSIBLE
? mount_entry_path(f
) : NULL
;
810 ml
->n_mounts
= t
- ml
->mounts
;
813 static void drop_nop(MountList
*ml
) {
818 /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the
819 * list is ordered by prefixes. */
821 for (f
= ml
->mounts
, t
= ml
->mounts
; f
< ml
->mounts
+ ml
->n_mounts
; f
++) {
823 /* Only suppress such subtrees for READ_ONLY, READ_WRITE and READ_WRITE_IMPLICIT entries */
824 if (IN_SET(f
->mode
, MOUNT_READ_ONLY
, MOUNT_READ_WRITE
, MOUNT_READ_WRITE_IMPLICIT
)) {
825 MountEntry
*found
= NULL
;
827 /* Now let's find the first parent of the entry we are looking at. */
828 for (MountEntry
*p
= PTR_SUB1(t
, ml
->mounts
); p
; p
= PTR_SUB1(p
, ml
->mounts
))
829 if (path_startswith(mount_entry_path(f
), mount_entry_path(p
))) {
834 /* We found it, let's see if it's the same mode, if so, we can drop this entry */
835 if (found
&& found
->mode
== f
->mode
) {
836 log_debug("%s (%s) is made redundant by %s (%s)",
837 mount_entry_path(f
), mount_mode_to_string(f
->mode
),
838 mount_entry_path(found
), mount_mode_to_string(found
->mode
));
848 ml
->n_mounts
= t
- ml
->mounts
;
851 static void drop_outside_root(MountList
*ml
, const char *root_directory
) {
860 /* Drops all mounts that are outside of the root directory. */
862 for (f
= ml
->mounts
, t
= ml
->mounts
; f
< ml
->mounts
+ ml
->n_mounts
; f
++) {
864 /* ExtensionImages/Directories bases are opened in /run/systemd/unit-extensions on the host */
865 if (!IN_SET(f
->mode
, MOUNT_EXTENSION_IMAGE
, MOUNT_EXTENSION_DIRECTORY
) && !path_startswith(mount_entry_path(f
), root_directory
)) {
866 log_debug("%s is outside of root directory.", mount_entry_path(f
));
875 ml
->n_mounts
= t
- ml
->mounts
;
878 static int clone_device_node(
880 const char *temporary_mount
,
881 bool *make_devnode
) {
883 _cleanup_free_
char *sl
= NULL
;
884 const char *dn
, *bn
, *t
;
888 if (stat(d
, &st
) < 0) {
889 if (errno
== ENOENT
) {
890 log_debug_errno(errno
, "Device node '%s' to clone does not exist, ignoring.", d
);
894 return log_debug_errno(errno
, "Failed to stat() device node '%s' to clone, ignoring: %m", d
);
897 if (!S_ISBLK(st
.st_mode
) &&
898 !S_ISCHR(st
.st_mode
))
899 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
900 "Device node '%s' to clone is not a device node, ignoring.",
903 dn
= strjoina(temporary_mount
, d
);
905 /* First, try to create device node properly */
907 mac_selinux_create_file_prepare(d
, st
.st_mode
);
908 r
= mknod(dn
, st
.st_mode
, st
.st_rdev
);
909 mac_selinux_create_file_clear();
913 return log_debug_errno(errno
, "mknod failed for %s: %m", d
);
915 /* This didn't work, let's not try this again for the next iterations. */
916 *make_devnode
= false;
919 /* We're about to fall back to bind-mounting the device node. So create a dummy bind-mount target.
920 * Do not prepare device-node SELinux label (see issue 13762) */
921 r
= mknod(dn
, S_IFREG
, 0);
922 if (r
< 0 && errno
!= EEXIST
)
923 return log_debug_errno(errno
, "mknod() fallback failed for '%s': %m", d
);
925 /* Fallback to bind-mounting: The assumption here is that all used device nodes carry standard
926 * properties. Specifically, the devices nodes we bind-mount should either be owned by root:root or
927 * root:tty (e.g. /dev/tty, /dev/ptmx) and should not carry ACLs. */
928 r
= mount_nofollow_verbose(LOG_DEBUG
, d
, dn
, NULL
, MS_BIND
, NULL
);
933 bn
= path_startswith(d
, "/dev/");
937 /* Create symlinks like /dev/char/1:9 → ../urandom */
938 if (asprintf(&sl
, "%s/dev/%s/" DEVNUM_FORMAT_STR
,
940 S_ISCHR(st
.st_mode
) ? "char" : "block",
941 DEVNUM_FORMAT_VAL(st
.st_rdev
)) < 0)
942 return log_oom_debug();
944 (void) mkdir_parents(sl
, 0755);
946 t
= strjoina("../", bn
);
947 if (symlink(t
, sl
) < 0)
948 log_debug_errno(errno
, "Failed to symlink '%s' to '%s', ignoring: %m", t
, sl
);
953 static char *settle_runtime_dir(RuntimeScope scope
) {
956 if (scope
!= RUNTIME_SCOPE_USER
)
957 return strdup("/run/");
959 if (asprintf(&runtime_dir
, "/run/user/" UID_FMT
, geteuid()) < 0)
965 static int create_temporary_mount_point(RuntimeScope scope
, char **ret
) {
966 _cleanup_free_
char *runtime_dir
= NULL
, *temporary_mount
= NULL
;
970 runtime_dir
= settle_runtime_dir(scope
);
972 return log_oom_debug();
974 temporary_mount
= path_join(runtime_dir
, "systemd/namespace-XXXXXX");
975 if (!temporary_mount
)
976 return log_oom_debug();
978 if (!mkdtemp(temporary_mount
))
979 return log_debug_errno(errno
, "Failed to create temporary directory '%s': %m", temporary_mount
);
981 *ret
= TAKE_PTR(temporary_mount
);
985 static int mount_private_dev(MountEntry
*m
, RuntimeScope scope
) {
986 static const char devnodes
[] =
994 _cleanup_free_
char *temporary_mount
= NULL
;
995 const char *dev
= NULL
, *devpts
= NULL
, *devshm
= NULL
, *devhugepages
= NULL
, *devmqueue
= NULL
, *devlog
= NULL
, *devptmx
= NULL
;
996 bool can_mknod
= true;
1001 r
= create_temporary_mount_point(scope
, &temporary_mount
);
1005 dev
= strjoina(temporary_mount
, "/dev");
1006 (void) mkdir(dev
, 0755);
1007 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", dev
, "tmpfs", DEV_MOUNT_OPTIONS
, "mode=0755" TMPFS_LIMITS_PRIVATE_DEV
);
1011 r
= label_fix_full(AT_FDCWD
, dev
, "/dev", 0);
1013 log_debug_errno(r
, "Failed to fix label of '%s' as /dev: %m", dev
);
1017 devpts
= strjoina(temporary_mount
, "/dev/pts");
1018 (void) mkdir(devpts
, 0755);
1019 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev/pts", devpts
, NULL
, MS_BIND
, NULL
);
1023 /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx.
1024 * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible.
1025 * Thus, in that case make a clone.
1026 * In nspawn and other containers it will be a symlink, in that case make it a symlink. */
1027 r
= is_symlink("/dev/ptmx");
1029 log_debug_errno(r
, "Failed to detect whether /dev/ptmx is a symlink or not: %m");
1032 devptmx
= strjoina(temporary_mount
, "/dev/ptmx");
1033 if (symlink("pts/ptmx", devptmx
) < 0) {
1034 r
= log_debug_errno(errno
, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx
);
1038 r
= clone_device_node("/dev/ptmx", temporary_mount
, &can_mknod
);
1043 devshm
= strjoina(temporary_mount
, "/dev/shm");
1044 (void) mkdir(devshm
, 0755);
1045 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev/shm", devshm
, NULL
, MS_BIND
, NULL
);
1049 devmqueue
= strjoina(temporary_mount
, "/dev/mqueue");
1050 (void) mkdir(devmqueue
, 0755);
1051 (void) mount_nofollow_verbose(LOG_DEBUG
, "/dev/mqueue", devmqueue
, NULL
, MS_BIND
, NULL
);
1053 devhugepages
= strjoina(temporary_mount
, "/dev/hugepages");
1054 (void) mkdir(devhugepages
, 0755);
1055 (void) mount_nofollow_verbose(LOG_DEBUG
, "/dev/hugepages", devhugepages
, NULL
, MS_BIND
, NULL
);
1057 devlog
= strjoina(temporary_mount
, "/dev/log");
1058 if (symlink("/run/systemd/journal/dev-log", devlog
) < 0)
1059 log_debug_errno(errno
, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog
);
1061 NULSTR_FOREACH(d
, devnodes
) {
1062 r
= clone_device_node(d
, temporary_mount
, &can_mknod
);
1063 /* ENXIO means the *source* is not a device file, skip creation in that case */
1064 if (r
< 0 && r
!= -ENXIO
)
1068 r
= dev_setup(temporary_mount
, UID_INVALID
, GID_INVALID
);
1070 log_debug_errno(r
, "Failed to set up basic device tree at '%s', ignoring: %m", temporary_mount
);
1072 /* Create the /dev directory if missing. It is more likely to be missing when the service is started
1073 * with RootDirectory. This is consistent with mount units creating the mount points when missing. */
1074 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1076 /* Unmount everything in old /dev */
1077 r
= umount_recursive(mount_entry_path(m
), 0);
1079 log_debug_errno(r
, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m
));
1081 r
= mount_nofollow_verbose(LOG_DEBUG
, dev
, mount_entry_path(m
), NULL
, MS_MOVE
, NULL
);
1086 (void) rmdir(temporary_mount
);
1092 (void) umount_verbose(LOG_DEBUG
, devpts
, UMOUNT_NOFOLLOW
);
1095 (void) umount_verbose(LOG_DEBUG
, devshm
, UMOUNT_NOFOLLOW
);
1098 (void) umount_verbose(LOG_DEBUG
, devhugepages
, UMOUNT_NOFOLLOW
);
1101 (void) umount_verbose(LOG_DEBUG
, devmqueue
, UMOUNT_NOFOLLOW
);
1103 (void) umount_verbose(LOG_DEBUG
, dev
, UMOUNT_NOFOLLOW
);
1105 (void) rmdir(temporary_mount
);
1110 static int mount_bind_dev(const MountEntry
*m
) {
1115 /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the
1116 * service's /dev. This is only used when RootDirectory= is set. */
1118 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1120 r
= path_is_mount_point(mount_entry_path(m
));
1122 return log_debug_errno(r
, "Unable to determine whether /dev is already mounted: %m");
1123 if (r
> 0) /* make this a NOP if /dev is already a mount point */
1126 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
);
1133 static int mount_bind_sysfs(const MountEntry
*m
) {
1138 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1140 r
= path_is_mount_point(mount_entry_path(m
));
1142 return log_debug_errno(r
, "Unable to determine whether /sys is already mounted: %m");
1143 if (r
> 0) /* make this a NOP if /sys is already a mount point */
1146 /* Bind mount the host's version so that we get all child mounts of it, too. */
1147 r
= mount_nofollow_verbose(LOG_DEBUG
, "/sys", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
);
1154 static int mount_private_apivfs(
1156 const char *entry_path
,
1157 const char *bind_source
,
1159 RuntimeScope scope
) {
1161 _cleanup_(rmdir_and_freep
) char *temporary_mount
= NULL
;
1166 assert(bind_source
);
1168 (void) mkdir_p_label(entry_path
, 0755);
1170 /* First, check if we have enough privileges to mount a new instance. Note, a new sysfs instance
1171 * cannot be mounted on an already existing mount. Let's use a temporary place. */
1172 r
= create_temporary_mount_point(scope
, &temporary_mount
);
1176 r
= mount_nofollow_verbose(LOG_DEBUG
, fstype
, temporary_mount
, fstype
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, opts
);
1177 if (r
== -EINVAL
&& opts
)
1178 /* If this failed with EINVAL then this likely means the textual hidepid= stuff for procfs is
1179 * not supported by the kernel, and thus the per-instance hidepid= neither, which means we
1180 * really don't want to use it, since it would affect our host's /proc mount. Hence let's
1181 * gracefully fallback to a classic, unrestricted version. */
1182 r
= mount_nofollow_verbose(LOG_DEBUG
, fstype
, temporary_mount
, fstype
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, /* opts = */ NULL
);
1183 if (ERRNO_IS_NEG_PRIVILEGE(r
)) {
1184 /* When we do not have enough privileges to mount a new instance, fall back to use an
1185 * existing mount. */
1187 r
= path_is_mount_point(entry_path
);
1189 return log_debug_errno(r
, "Unable to determine whether '%s' is already mounted: %m", entry_path
);
1191 return 0; /* Use the current mount as is. */
1193 /* We lack permissions to mount a new instance, and it is not already mounted. But we can
1194 * access the host's, so as a final fallback bind-mount it to the destination, as most likely
1195 * we are inside a user manager in an unprivileged user namespace. */
1196 r
= mount_nofollow_verbose(LOG_DEBUG
, bind_source
, entry_path
, /* fstype = */ NULL
, MS_BIND
|MS_REC
, /* opts = */ NULL
);
1205 /* OK. We have a new mount instance. Let's clear an existing mount and its submounts. */
1206 r
= umount_recursive(entry_path
, /* flags = */ 0);
1208 log_debug_errno(r
, "Failed to unmount directories below '%s', ignoring: %m", entry_path
);
1210 /* Then, move the new mount instance. */
1211 r
= mount_nofollow_verbose(LOG_DEBUG
, temporary_mount
, entry_path
, /* fstype = */ NULL
, MS_MOVE
, /* opts = */ NULL
);
1215 /* We mounted a new instance now. Let's bind mount the children over now. This matters for nspawn
1216 * where a bunch of files are overmounted, in particular the boot id. */
1217 (void) bind_mount_submounts(bind_source
, entry_path
);
1221 static int mount_private_sysfs(const MountEntry
*m
, const NamespaceParameters
*p
) {
1224 return mount_private_apivfs("sysfs", mount_entry_path(m
), "/sys", /* opts = */ NULL
, p
->runtime_scope
);
1227 static int mount_procfs(const MountEntry
*m
, const NamespaceParameters
*p
) {
1228 _cleanup_free_
char *opts
= NULL
;
1233 if (p
->protect_proc
!= PROTECT_PROC_DEFAULT
||
1234 p
->proc_subset
!= PROC_SUBSET_ALL
) {
1236 /* Starting with kernel 5.8 procfs' hidepid= logic is truly per-instance (previously it
1237 * pretended to be per-instance but actually was per-namespace), hence let's make use of it
1238 * if requested. To make sure this logic succeeds only on kernels where hidepid= is
1239 * per-instance, we'll exclusively use the textual value for hidepid=, since support was
1240 * added in the same commit: if it's supported it is thus also per-instance. */
1242 const char *hpv
= p
->protect_proc
== PROTECT_PROC_DEFAULT
?
1244 protect_proc_to_string(p
->protect_proc
);
1246 /* hidepid= support was added in 5.8, so we can use fsconfig()/fsopen() (which were added in
1247 * 5.2) to check if hidepid= is supported. This avoids a noisy dmesg log by the kernel when
1248 * trying to use hidepid= on systems where it isn't supported. The same applies for subset=.
1249 * fsopen()/fsconfig() was also backported on some distros which allows us to detect
1250 * hidepid=/subset= support in even more scenarios. */
1252 if (mount_option_supported("proc", "hidepid", hpv
) != 0) {
1253 opts
= strjoin("hidepid=", hpv
);
1258 if (p
->proc_subset
== PROC_SUBSET_PID
&&
1259 mount_option_supported("proc", "subset", "pid") != 0)
1260 if (!strextend_with_separator(&opts
, ",", "subset=pid"))
1264 /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in
1265 * one. i.e we don't reuse existing mounts here under any condition, we want a new instance owned by
1266 * our user namespace and with our hidepid= settings applied. Hence, let's get rid of everything
1267 * mounted on /proc/ first. */
1268 return mount_private_apivfs("proc", mount_entry_path(m
), "/proc", opts
, p
->runtime_scope
);
1271 static int mount_tmpfs(const MountEntry
*m
) {
1272 const char *entry_path
, *inner_path
;
1277 entry_path
= mount_entry_path(m
);
1278 inner_path
= mount_entry_unprefixed_path(m
);
1280 /* First, get rid of everything that is below if there is anything. Then, overmount with our new
1283 (void) mkdir_p_label(entry_path
, 0755);
1284 (void) umount_recursive(entry_path
, 0);
1286 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", entry_path
, "tmpfs", m
->flags
, mount_entry_options(m
));
1290 r
= label_fix_full(AT_FDCWD
, entry_path
, inner_path
, 0);
1292 return log_debug_errno(r
, "Failed to fix label of '%s' as '%s': %m", entry_path
, inner_path
);
1297 static int mount_run(const MountEntry
*m
) {
1302 r
= path_is_mount_point(mount_entry_path(m
));
1303 if (r
< 0 && r
!= -ENOENT
)
1304 return log_debug_errno(r
, "Unable to determine whether /run is already mounted: %m");
1305 if (r
> 0) /* make this a NOP if /run is already a mount point */
1308 return mount_tmpfs(m
);
1311 static int mount_mqueuefs(const MountEntry
*m
) {
1313 const char *entry_path
;
1317 entry_path
= mount_entry_path(m
);
1319 (void) mkdir_p_label(entry_path
, 0755);
1320 (void) umount_recursive(entry_path
, 0);
1322 r
= mount_nofollow_verbose(LOG_DEBUG
, "mqueue", entry_path
, "mqueue", m
->flags
, mount_entry_options(m
));
1329 static int mount_image(
1330 const MountEntry
*m
,
1331 const char *root_directory
,
1332 const ImagePolicy
*image_policy
) {
1334 _cleanup_free_
char *host_os_release_id
= NULL
, *host_os_release_version_id
= NULL
,
1335 *host_os_release_sysext_level
= NULL
, *host_os_release_confext_level
= NULL
,
1336 *extension_name
= NULL
;
1341 r
= path_extract_filename(mount_entry_source(m
), &extension_name
);
1343 return log_debug_errno(r
, "Failed to extract extension name from %s: %m", mount_entry_source(m
));
1345 if (m
->mode
== MOUNT_EXTENSION_IMAGE
) {
1346 r
= parse_os_release(
1347 empty_to_root(root_directory
),
1348 "ID", &host_os_release_id
,
1349 "VERSION_ID", &host_os_release_version_id
,
1350 image_class_info
[IMAGE_SYSEXT
].level_env
, &host_os_release_sysext_level
,
1351 image_class_info
[IMAGE_CONFEXT
].level_env
, &host_os_release_confext_level
,
1354 return log_debug_errno(r
, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory
));
1355 if (isempty(host_os_release_id
))
1356 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
));
1359 r
= verity_dissect_and_mount(
1361 mount_entry_source(m
),
1362 mount_entry_path(m
),
1363 m
->image_options_const
,
1366 host_os_release_version_id
,
1367 host_os_release_sysext_level
,
1368 host_os_release_confext_level
,
1369 /* required_sysext_scope= */ NULL
,
1370 /* ret_image= */ NULL
);
1371 if (r
== -ENOENT
&& m
->ignore
)
1373 if (r
== -ESTALE
&& host_os_release_id
)
1374 return log_error_errno(r
, // FIXME: this should not be logged ad LOG_ERR, as it will result in duplicate logging.
1375 "Failed to mount image %s, extension-release metadata does not match the lower layer's: ID=%s%s%s%s%s%s%s",
1376 mount_entry_source(m
),
1378 host_os_release_version_id
? " VERSION_ID=" : "",
1379 strempty(host_os_release_version_id
),
1380 host_os_release_sysext_level
? image_class_info
[IMAGE_SYSEXT
].level_env_print
: "",
1381 strempty(host_os_release_sysext_level
),
1382 host_os_release_confext_level
? image_class_info
[IMAGE_CONFEXT
].level_env_print
: "",
1383 strempty(host_os_release_confext_level
));
1385 return log_debug_errno(r
, "Failed to mount image %s on %s: %m", mount_entry_source(m
), mount_entry_path(m
));
1390 static int mount_overlay(const MountEntry
*m
) {
1391 _cleanup_free_
char *options
= NULL
, *layers
= NULL
;
1396 /* Extension hierarchies are optional (e.g.: confext might not have /opt) so check if they actually
1397 * exist in an image before attempting to create an overlay with them, otherwise the mount will
1398 * fail. We can't check before this, as the images will not be mounted until now. */
1400 /* Note that lowerdir= parameters are in 'reverse' order, so the top-most directory in the overlay
1401 * comes first in the list. */
1402 STRV_FOREACH_BACKWARDS(o
, m
->overlay_layers
) {
1403 _cleanup_free_
char *escaped
= NULL
;
1405 r
= is_dir(*o
, /* follow= */ false);
1409 "Failed to check whether overlay layer source path '%s' exists, ignoring: %m",
1414 escaped
= shell_escape(*o
, ",:");
1416 return log_oom_debug();
1418 if (!strextend_with_separator(&layers
, ":", escaped
))
1419 return log_oom_debug();
1423 log_debug("None of the overlays specified in '%s' exist at the source, skipping.",
1424 mount_entry_options(m
));
1425 return 0; /* Only the root is set? Then there's nothing to overlay */
1428 options
= strjoin("lowerdir=", layers
, ":", mount_entry_path(m
)); /* The root goes in last */
1430 return log_oom_debug();
1432 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1434 r
= mount_nofollow_verbose(LOG_DEBUG
, "overlay", mount_entry_path(m
), "overlay", MS_RDONLY
, options
);
1435 if (r
== -ENOENT
&& m
->ignore
)
1443 static int follow_symlink(
1444 const char *root_directory
,
1447 _cleanup_free_
char *target
= NULL
;
1450 /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we
1451 * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at
1452 * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the
1453 * end and already have a fully normalized name. */
1455 r
= chase(mount_entry_path(m
), root_directory
, CHASE_STEP
|CHASE_NONEXISTENT
, &target
, NULL
);
1457 return log_debug_errno(r
, "Failed to chase symlinks '%s': %m", mount_entry_path(m
));
1458 if (r
> 0) /* Reached the end, nothing more to resolve */
1461 if (m
->n_followed
>= CHASE_MAX
) /* put a boundary on things */
1462 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
1463 "Symlink loop on '%s'.",
1464 mount_entry_path(m
));
1466 log_debug("Followed mount entry path symlink %s %s %s.",
1467 mount_entry_path(m
), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT
), target
);
1469 mount_entry_consume_prefix(m
, TAKE_PTR(target
));
1476 static int apply_one_mount(
1477 const char *root_directory
,
1479 const NamespaceParameters
*p
) {
1481 _cleanup_free_
char *inaccessible
= NULL
;
1482 bool rbind
= true, make
= false;
1486 /* Return 1 when the mount should be post-processed (remounted r/o, etc.), 0 otherwise. In most
1487 * cases post-processing is the right thing, the typical exception is when the mount is gracefully
1493 log_debug("Applying namespace mount on %s", mount_entry_path(m
));
1497 case MOUNT_INACCESSIBLE
: {
1498 _cleanup_free_
char *runtime_dir
= NULL
;
1501 /* First, get rid of everything that is below if there
1502 * is anything... Then, overmount it with an
1503 * inaccessible path. */
1504 (void) umount_recursive(mount_entry_path(m
), 0);
1506 if (lstat(mount_entry_path(m
), &target
) < 0) {
1507 if (errno
== ENOENT
&& m
->ignore
)
1510 return log_debug_errno(errno
, "Failed to lstat() %s to determine what to mount over it: %m",
1511 mount_entry_path(m
));
1514 /* We don't pass the literal runtime scope through here but one based purely on our UID. This
1515 * means that the root user's --user services will use the host's inaccessible inodes rather
1516 * then root's private ones. This is preferable since it means device nodes that are
1517 * overmounted to make them inaccessible will be overmounted with a device node, rather than
1518 * an AF_UNIX socket inode. */
1519 runtime_dir
= settle_runtime_dir(geteuid() == 0 ? RUNTIME_SCOPE_SYSTEM
: RUNTIME_SCOPE_USER
);
1521 return log_oom_debug();
1523 r
= mode_to_inaccessible_node(runtime_dir
, target
.st_mode
, &inaccessible
);
1525 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
1526 "File type not supported for inaccessible mounts. Note that symlinks are not allowed");
1527 what
= inaccessible
;
1531 case MOUNT_READ_ONLY
:
1532 case MOUNT_READ_WRITE
:
1533 case MOUNT_READ_WRITE_IMPLICIT
:
1536 r
= path_is_mount_point_full(mount_entry_path(m
), root_directory
, /* flags = */ 0);
1537 if (r
== -ENOENT
&& m
->ignore
)
1540 return log_debug_errno(r
, "Failed to determine whether %s is already a mount point: %m",
1541 mount_entry_path(m
));
1542 if (r
> 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY
1543 * and MS_NOEXEC bits for the mount point if needed. */
1545 /* This isn't a mount point yet, let's make it one. */
1546 what
= mount_entry_path(m
);
1549 case MOUNT_EXTENSION_DIRECTORY
: {
1550 _cleanup_free_
char *host_os_release_id
= NULL
, *host_os_release_version_id
= NULL
,
1551 *host_os_release_level
= NULL
, *extension_name
= NULL
;
1552 _cleanup_strv_free_
char **extension_release
= NULL
;
1553 ImageClass
class = IMAGE_SYSEXT
;
1555 r
= path_extract_filename(mount_entry_source(m
), &extension_name
);
1557 return log_debug_errno(r
, "Failed to extract extension name from %s: %m", mount_entry_source(m
));
1559 r
= load_extension_release_pairs(mount_entry_source(m
), IMAGE_SYSEXT
, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1561 r
= load_extension_release_pairs(mount_entry_source(m
), IMAGE_CONFEXT
, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1563 class = IMAGE_CONFEXT
;
1566 return log_debug_errno(r
, "Failed to acquire 'extension-release' data of extension tree %s: %m", mount_entry_source(m
));
1568 r
= parse_os_release(
1569 empty_to_root(root_directory
),
1570 "ID", &host_os_release_id
,
1571 "VERSION_ID", &host_os_release_version_id
,
1572 image_class_info
[class].level_env
, &host_os_release_level
,
1575 return log_debug_errno(r
, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory
));
1576 if (isempty(host_os_release_id
))
1577 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
));
1579 r
= load_extension_release_pairs(mount_entry_source(m
), class, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1580 if (r
== -ENOENT
&& m
->ignore
)
1583 return log_debug_errno(r
, "Failed to parse directory %s extension-release metadata: %m", extension_name
);
1585 r
= extension_release_validate(
1588 host_os_release_version_id
,
1589 host_os_release_level
,
1590 /* host_extension_scope */ NULL
, /* Leave empty, we need to accept both system and portable */
1594 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE
), "Directory %s extension-release metadata does not match the root's", extension_name
);
1596 return log_debug_errno(r
, "Failed to compare directory %s extension-release metadata with the root's os-release: %m", extension_name
);
1605 case MOUNT_BIND_RECURSIVE
: {
1606 _cleanup_free_
char *chased
= NULL
;
1608 /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note
1609 * that bind mount source paths are always relative to the host root, hence we pass NULL as
1610 * root directory to chase() here. */
1612 r
= chase(mount_entry_source(m
), NULL
, CHASE_TRAIL_SLASH
, &chased
, NULL
);
1613 if (r
== -ENOENT
&& m
->ignore
) {
1614 log_debug_errno(r
, "Path %s does not exist, ignoring.", mount_entry_source(m
));
1618 return log_debug_errno(r
, "Failed to follow symlinks on %s: %m", mount_entry_source(m
));
1620 log_debug("Followed source symlinks %s %s %s.",
1621 mount_entry_source(m
), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT
), chased
);
1623 free_and_replace(m
->source_malloc
, chased
);
1625 what
= mount_entry_source(m
);
1630 case MOUNT_EMPTY_DIR
:
1632 return mount_tmpfs(m
);
1634 case MOUNT_PRIVATE_TMP
:
1635 case MOUNT_PRIVATE_TMP_READ_ONLY
:
1636 what
= mount_entry_source(m
);
1640 case MOUNT_PRIVATE_DEV
:
1641 return mount_private_dev(m
, p
->runtime_scope
);
1643 case MOUNT_BIND_DEV
:
1644 return mount_bind_dev(m
);
1646 case MOUNT_PRIVATE_SYSFS
:
1647 return mount_private_sysfs(m
, p
);
1649 case MOUNT_BIND_SYSFS
:
1650 return mount_bind_sysfs(m
);
1653 return mount_procfs(m
, p
);
1656 return mount_run(m
);
1658 case MOUNT_MQUEUEFS
:
1659 return mount_mqueuefs(m
);
1662 return mount_image(m
, NULL
, p
->mount_image_policy
);
1664 case MOUNT_EXTENSION_IMAGE
:
1665 return mount_image(m
, root_directory
, p
->extension_image_policy
);
1668 return mount_overlay(m
);
1671 assert_not_reached();
1676 r
= mount_nofollow_verbose(LOG_DEBUG
, what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
);
1678 bool try_again
= false;
1680 if (r
== -ENOENT
&& make
) {
1683 /* Hmm, either the source or the destination are missing. Let's see if we can create
1684 the destination, then try again. */
1686 (void) mkdir_parents(mount_entry_path(m
), 0755);
1688 q
= make_mount_point_inode_from_path(what
, mount_entry_path(m
), 0755);
1690 if (q
!= -EEXIST
) // FIXME: this shouldn't be logged at LOG_WARNING, but be bubbled up, and logged there to avoid duplicate logging
1691 log_warning_errno(q
, "Failed to create destination mount point node '%s', ignoring: %m",
1692 mount_entry_path(m
));
1698 r
= mount_nofollow_verbose(LOG_DEBUG
, what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
);
1700 return log_error_errno(r
, "Failed to mount %s to %s: %m", what
, mount_entry_path(m
)); // FIXME: this should not be logged here, but be bubbled up, to avoid duplicate logging
1703 log_debug("Successfully mounted %s to %s", what
, mount_entry_path(m
));
1707 static int make_read_only(const MountEntry
*m
, char **deny_list
, FILE *proc_self_mountinfo
) {
1708 unsigned long new_flags
= 0, flags_mask
= 0;
1713 assert(proc_self_mountinfo
);
1715 if (m
->state
!= MOUNT_APPLIED
)
1718 if (mount_entry_read_only(m
) || m
->mode
== MOUNT_PRIVATE_DEV
) {
1719 new_flags
|= MS_RDONLY
;
1720 flags_mask
|= MS_RDONLY
;
1724 new_flags
|= MS_NOSUID
;
1725 flags_mask
|= MS_NOSUID
;
1728 if (flags_mask
== 0) /* No Change? */
1731 /* We generally apply these changes recursively, except for /dev, and the cases we know there's
1732 * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the
1733 * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace
1734 * and running Linux <= 4.17. */
1736 mount_entry_read_only(m
) &&
1737 !IN_SET(m
->mode
, MOUNT_EMPTY_DIR
, MOUNT_TMPFS
);
1739 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, deny_list
, proc_self_mountinfo
);
1741 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, proc_self_mountinfo
);
1743 /* Note that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked
1744 * read-only already stays this way. This improves compatibility with container managers, where we
1745 * won't attempt to undo read-only mounts already applied. */
1747 if (r
== -ENOENT
&& m
->ignore
)
1750 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1751 submounts
? " and its submounts" : "");
1755 static int make_noexec(const MountEntry
*m
, char **deny_list
, FILE *proc_self_mountinfo
) {
1756 unsigned long new_flags
= 0, flags_mask
= 0;
1761 assert(proc_self_mountinfo
);
1763 if (m
->state
!= MOUNT_APPLIED
)
1766 if (mount_entry_noexec(m
)) {
1767 new_flags
|= MS_NOEXEC
;
1768 flags_mask
|= MS_NOEXEC
;
1769 } else if (mount_entry_exec(m
)) {
1770 new_flags
&= ~MS_NOEXEC
;
1771 flags_mask
|= MS_NOEXEC
;
1774 if (flags_mask
== 0) /* No Change? */
1777 submounts
= !IN_SET(m
->mode
, MOUNT_EMPTY_DIR
, MOUNT_TMPFS
);
1780 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, deny_list
, proc_self_mountinfo
);
1782 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, proc_self_mountinfo
);
1784 if (r
== -ENOENT
&& m
->ignore
)
1787 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1788 submounts
? " and its submounts" : "");
1792 static int make_nosuid(const MountEntry
*m
, FILE *proc_self_mountinfo
) {
1797 assert(proc_self_mountinfo
);
1799 if (m
->state
!= MOUNT_APPLIED
)
1802 submounts
= !IN_SET(m
->mode
, MOUNT_EMPTY_DIR
, MOUNT_TMPFS
);
1804 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), MS_NOSUID
, MS_NOSUID
, NULL
, proc_self_mountinfo
);
1806 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), MS_NOSUID
, MS_NOSUID
, proc_self_mountinfo
);
1807 if (r
== -ENOENT
&& m
->ignore
)
1810 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1811 submounts
? " and its submounts" : "");
1815 static bool namespace_parameters_mount_apivfs(const NamespaceParameters
*p
) {
1819 * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
1820 * since to protect the API VFS mounts, they need to be around in the
1824 return p
->mount_apivfs
||
1825 p
->protect_control_groups
||
1826 p
->protect_kernel_tunables
||
1827 p
->protect_proc
!= PROTECT_PROC_DEFAULT
||
1828 p
->proc_subset
!= PROC_SUBSET_ALL
;
1831 /* Walk all mount entries and dropping any unused mounts. This affects all
1833 * - that are implicitly protected by a path that has been rendered inaccessible
1834 * - whose immediate parent requests the same protection mode as the mount itself
1835 * - that are outside of the relevant root directory
1836 * - which are duplicates
1838 static void drop_unused_mounts(MountList
*ml
, const char *root_directory
) {
1840 assert(root_directory
);
1842 assert(ml
->mounts
|| ml
->n_mounts
== 0);
1844 typesafe_qsort(ml
->mounts
, ml
->n_mounts
, mount_path_compare
);
1846 drop_duplicates(ml
);
1847 drop_outside_root(ml
, root_directory
);
1848 drop_inaccessible(ml
);
1852 static int create_symlinks_from_tuples(const char *root
, char **strv_symlinks
) {
1855 STRV_FOREACH_PAIR(src
, dst
, strv_symlinks
) {
1856 _cleanup_free_
char *src_abs
= NULL
, *dst_abs
= NULL
;
1858 src_abs
= path_join(root
, *src
);
1859 dst_abs
= path_join(root
, *dst
);
1860 if (!src_abs
|| !dst_abs
)
1863 r
= mkdir_parents_label(dst_abs
, 0755);
1865 return log_debug_errno(
1867 "Failed to create parent directory for symlink '%s': %m",
1870 r
= symlink_idempotent(src_abs
, dst_abs
, true);
1872 return log_debug_errno(
1874 "Failed to create symlink from '%s' to '%s': %m",
1882 static void mount_entry_path_debug_string(const char *root
, MountEntry
*m
, char **error_path
) {
1885 /* Create a string suitable for debugging logs, stripping for example the local working directory.
1886 * For example, with a BindPaths=/var/bar that does not exist on the host:
1889 * foo.service: Failed to set up mount namespacing: /run/systemd/unit-root/var/bar: No such file or directory
1891 * foo.service: Failed to set up mount namespacing: /var/bar: No such file or directory
1893 * Note that this is an error path, so no OOM check is done on purpose. */
1898 if (!mount_entry_path(m
)) {
1904 const char *e
= startswith(mount_entry_path(m
), root
);
1906 *error_path
= strdup(e
);
1911 *error_path
= strdup(mount_entry_path(m
));
1915 static int apply_mounts(
1918 const NamespaceParameters
*p
,
1919 char **error_path
) {
1921 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
1922 _cleanup_free_
char **deny_list
= NULL
;
1929 if (ml
->n_mounts
== 0) /* Shortcut: nothing to do */
1932 /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of
1933 * /proc. For example, this is the case with the option: 'InaccessiblePaths=/proc'. */
1934 proc_self_mountinfo
= fopen("/proc/self/mountinfo", "re");
1935 if (!proc_self_mountinfo
) {
1939 *error_path
= strdup("/proc/self/mountinfo");
1941 return log_debug_errno(r
, "Failed to open /proc/self/mountinfo: %m");
1944 /* First round, establish all mounts we need */
1948 FOREACH_ARRAY(m
, ml
->mounts
, ml
->n_mounts
) {
1950 if (m
->state
!= MOUNT_PENDING
)
1953 /* ExtensionImages/Directories are first opened in the propagate directory, not in the root_directory */
1954 r
= follow_symlink(!IN_SET(m
->mode
, MOUNT_EXTENSION_IMAGE
, MOUNT_EXTENSION_DIRECTORY
) ? root
: NULL
, m
);
1956 mount_entry_path_debug_string(root
, m
, error_path
);
1960 /* We hit a symlinked mount point. The entry got rewritten and might
1961 * point to a very different place now. Let's normalize the changed
1962 * list, and start from the beginning. After all to mount the entry
1963 * at the new location we might need some other mounts first */
1968 /* Returns 1 if the mount should be post-processed, 0 otherwise */
1969 r
= apply_one_mount(root
, m
, p
);
1971 mount_entry_path_debug_string(root
, m
, error_path
);
1974 m
->state
= r
== 0 ? MOUNT_SKIPPED
: MOUNT_APPLIED
;
1980 drop_unused_mounts(ml
, root
);
1983 /* Now that all filesystems have been set up, but before the
1984 * read-only switches are flipped, create the exec dirs and other symlinks.
1985 * Note that when /var/lib is not empty/tmpfs, these symlinks will already
1986 * exist, which means this will be a no-op. */
1987 r
= create_symlinks_from_tuples(root
, p
->symlinks
);
1989 return log_debug_errno(r
, "Failed to set up symlinks inside mount namespace: %m");
1991 /* Create a deny list we can pass to bind_mount_recursive() */
1992 deny_list
= new(char*, ml
->n_mounts
+1);
1995 for (size_t j
= 0; j
< ml
->n_mounts
; j
++)
1996 deny_list
[j
] = (char*) mount_entry_path(ml
->mounts
+j
);
1997 deny_list
[ml
->n_mounts
] = NULL
;
1999 /* Second round, flip the ro bits if necessary. */
2000 FOREACH_ARRAY(m
, ml
->mounts
, ml
->n_mounts
) {
2001 r
= make_read_only(m
, deny_list
, proc_self_mountinfo
);
2003 mount_entry_path_debug_string(root
, m
, error_path
);
2008 /* Third round, flip the noexec bits with a simplified deny list. */
2009 for (size_t j
= 0; j
< ml
->n_mounts
; j
++)
2010 if (IN_SET((ml
->mounts
+j
)->mode
, MOUNT_EXEC
, MOUNT_NOEXEC
))
2011 deny_list
[j
] = (char*) mount_entry_path(ml
->mounts
+j
);
2012 deny_list
[ml
->n_mounts
] = NULL
;
2014 FOREACH_ARRAY(m
, ml
->mounts
, ml
->n_mounts
) {
2015 r
= make_noexec(m
, deny_list
, proc_self_mountinfo
);
2017 mount_entry_path_debug_string(root
, m
, error_path
);
2022 /* Fourth round, flip the nosuid bits without a deny list. */
2023 if (p
->mount_nosuid
)
2024 FOREACH_ARRAY(m
, ml
->mounts
, ml
->n_mounts
) {
2025 r
= make_nosuid(m
, proc_self_mountinfo
);
2027 mount_entry_path_debug_string(root
, m
, error_path
);
2035 static bool root_read_only(
2036 char **read_only_paths
,
2037 ProtectSystem protect_system
) {
2039 /* Determine whether the root directory is going to be read-only given the configured settings. */
2041 if (protect_system
== PROTECT_SYSTEM_STRICT
)
2044 if (prefixed_path_strv_contains(read_only_paths
, "/"))
2050 static bool home_read_only(
2051 char** read_only_paths
,
2052 char** inaccessible_paths
,
2053 char** empty_directories
,
2054 const BindMount
*bind_mounts
,
2055 size_t n_bind_mounts
,
2056 const TemporaryFileSystem
*temporary_filesystems
,
2057 size_t n_temporary_filesystems
,
2058 ProtectHome protect_home
) {
2060 /* Determine whether the /home directory is going to be read-only given the configured settings. Yes,
2061 * this is a bit sloppy, since we don't bother checking for cases where / is affected by multiple
2064 if (protect_home
!= PROTECT_HOME_NO
)
2067 if (prefixed_path_strv_contains(read_only_paths
, "/home") ||
2068 prefixed_path_strv_contains(inaccessible_paths
, "/home") ||
2069 prefixed_path_strv_contains(empty_directories
, "/home"))
2072 for (size_t i
= 0; i
< n_temporary_filesystems
; i
++)
2073 if (path_equal(temporary_filesystems
[i
].path
, "/home"))
2076 /* If /home is overmounted with some dir from the host it's not writable. */
2077 for (size_t i
= 0; i
< n_bind_mounts
; i
++)
2078 if (path_equal(bind_mounts
[i
].destination
, "/home"))
2084 int setup_namespace(const NamespaceParameters
*p
, char **error_path
) {
2086 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
2087 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
2088 _cleanup_strv_free_
char **hierarchies
= NULL
;
2089 _cleanup_(mount_list_done
) MountList ml
= {};
2090 bool require_prefix
= false;
2092 DissectImageFlags dissect_image_flags
=
2093 DISSECT_IMAGE_GENERIC_ROOT
|
2094 DISSECT_IMAGE_REQUIRE_ROOT
|
2095 DISSECT_IMAGE_DISCARD_ON_LOOP
|
2096 DISSECT_IMAGE_RELAX_VAR_CHECK
|
2097 DISSECT_IMAGE_FSCK
|
2098 DISSECT_IMAGE_USR_NO_ROOT
|
2099 DISSECT_IMAGE_GROWFS
|
2100 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
2101 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
2106 /* Make sure that all mknod(), mkdir() calls we do are unaffected by the umask, and the access modes
2107 * we configure take effect */
2108 BLOCK_WITH_UMASK(0000);
2110 bool setup_propagate
= !isempty(p
->propagate_dir
) && !isempty(p
->incoming_dir
);
2111 unsigned long mount_propagation_flag
= p
->mount_propagation_flag
!= 0 ? p
->mount_propagation_flag
: MS_SHARED
;
2113 if (p
->root_image
) {
2114 /* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */
2115 if (root_read_only(p
->read_only_paths
,
2116 p
->protect_system
) &&
2117 home_read_only(p
->read_only_paths
, p
->inaccessible_paths
, p
->empty_directories
,
2118 p
->bind_mounts
, p
->n_bind_mounts
, p
->temporary_filesystems
, p
->n_temporary_filesystems
,
2120 strv_isempty(p
->read_write_paths
))
2121 dissect_image_flags
|= DISSECT_IMAGE_READ_ONLY
;
2123 SET_FLAG(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
, p
->verity
&& p
->verity
->data_path
);
2125 r
= loop_device_make_by_path(
2127 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_DEVICE_READ_ONLY
) ? O_RDONLY
: -1 /* < 0 means writable if possible, read-only as fallback */,
2128 /* sector_size= */ UINT32_MAX
,
2129 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
2133 return log_debug_errno(r
, "Failed to create loop device for root image: %m");
2135 r
= dissect_loop_device(
2138 p
->root_image_options
,
2139 p
->root_image_policy
,
2140 dissect_image_flags
,
2143 return log_debug_errno(r
, "Failed to dissect image: %m");
2145 r
= dissected_image_load_verity_sig_partition(
2152 r
= dissected_image_decrypt(
2156 dissect_image_flags
);
2158 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
2161 if (p
->root_directory
)
2162 root
= p
->root_directory
;
2164 /* /run/systemd should have been created by PID 1 early on already, but in some cases, like
2165 * when running tests (test-execute), it might not have been created yet so let's make sure
2166 * we create it if it doesn't already exist. */
2167 (void) mkdir_p_label("/run/systemd", 0755);
2169 /* Always create the mount namespace in a temporary directory, instead of operating directly
2170 * in the root. The temporary directory prevents any mounts from being potentially obscured
2171 * my other mounts we already applied. We use the same mount point for all images, which is
2172 * safe, since they all live in their own namespaces after all, and hence won't see each
2173 * other. (Note: this directory is also created by PID 1 early on, we create it here for
2174 * similar reasons as /run/systemd/ first.) */
2175 root
= "/run/systemd/mount-rootfs";
2176 (void) mkdir_label(root
, 0555);
2178 require_prefix
= true;
2181 if (p
->n_extension_images
> 0 || !strv_isempty(p
->extension_directories
)) {
2182 /* Hierarchy population needs to be done for sysext and confext extension images */
2183 r
= parse_env_extension_hierarchies(&hierarchies
, "SYSTEMD_SYSEXT_AND_CONFEXT_HIERARCHIES");
2188 r
= append_access_mounts(&ml
, p
->read_write_paths
, MOUNT_READ_WRITE
, require_prefix
);
2192 r
= append_access_mounts(&ml
, p
->read_only_paths
, MOUNT_READ_ONLY
, require_prefix
);
2196 r
= append_access_mounts(&ml
, p
->inaccessible_paths
, MOUNT_INACCESSIBLE
, require_prefix
);
2200 r
= append_access_mounts(&ml
, p
->exec_paths
, MOUNT_EXEC
, require_prefix
);
2204 r
= append_access_mounts(&ml
, p
->no_exec_paths
, MOUNT_NOEXEC
, require_prefix
);
2208 r
= append_empty_dir_mounts(&ml
, p
->empty_directories
);
2212 r
= append_bind_mounts(&ml
, p
->bind_mounts
, p
->n_bind_mounts
);
2216 r
= append_tmpfs_mounts(&ml
, p
->temporary_filesystems
, p
->n_temporary_filesystems
);
2221 bool ro
= streq(p
->tmp_dir
, RUN_SYSTEMD_EMPTY
);
2223 MountEntry
*me
= mount_list_extend(&ml
);
2225 return log_oom_debug();
2227 *me
= (MountEntry
) {
2228 .path_const
= "/tmp",
2229 .mode
= ro
? MOUNT_PRIVATE_TMP_READ_ONLY
: MOUNT_PRIVATE_TMP
,
2230 .source_const
= p
->tmp_dir
,
2234 if (p
->var_tmp_dir
) {
2235 bool ro
= streq(p
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
);
2237 MountEntry
*me
= mount_list_extend(&ml
);
2239 return log_oom_debug();
2241 *me
= (MountEntry
) {
2242 .path_const
= "/var/tmp",
2243 .mode
= ro
? MOUNT_PRIVATE_TMP_READ_ONLY
: MOUNT_PRIVATE_TMP
,
2244 .source_const
= p
->var_tmp_dir
,
2248 r
= append_mount_images(&ml
, p
->mount_images
, p
->n_mount_images
);
2252 r
= append_extensions(&ml
, root
, p
->extension_dir
, hierarchies
, p
->extension_images
, p
->n_extension_images
, p
->extension_directories
);
2256 if (p
->private_dev
) {
2257 MountEntry
*me
= mount_list_extend(&ml
);
2259 return log_oom_debug();
2261 *me
= (MountEntry
) {
2262 .path_const
= "/dev",
2263 .mode
= MOUNT_PRIVATE_DEV
,
2264 .flags
= DEV_MOUNT_OPTIONS
,
2268 /* In case /proc is successfully mounted with pid tree subset only (ProcSubset=pid), the protective
2269 mounts to non-pid /proc paths would fail. But the pid only option may have failed gracefully, so
2270 let's try the mounts but it's not fatal if they don't succeed. */
2271 bool ignore_protect_proc
= p
->ignore_protect_paths
|| p
->proc_subset
== PROC_SUBSET_PID
;
2272 if (p
->protect_kernel_tunables
) {
2273 r
= append_static_mounts(&ml
,
2274 protect_kernel_tunables_proc_table
,
2275 ELEMENTSOF(protect_kernel_tunables_proc_table
),
2276 ignore_protect_proc
);
2280 r
= append_static_mounts(&ml
,
2281 protect_kernel_tunables_sys_table
,
2282 ELEMENTSOF(protect_kernel_tunables_sys_table
),
2283 p
->ignore_protect_paths
);
2288 if (p
->protect_kernel_modules
) {
2289 r
= append_static_mounts(&ml
,
2290 protect_kernel_modules_table
,
2291 ELEMENTSOF(protect_kernel_modules_table
),
2292 p
->ignore_protect_paths
);
2297 if (p
->protect_kernel_logs
) {
2298 r
= append_static_mounts(&ml
,
2299 protect_kernel_logs_proc_table
,
2300 ELEMENTSOF(protect_kernel_logs_proc_table
),
2301 ignore_protect_proc
);
2305 r
= append_static_mounts(&ml
,
2306 protect_kernel_logs_dev_table
,
2307 ELEMENTSOF(protect_kernel_logs_dev_table
),
2308 p
->ignore_protect_paths
);
2313 if (p
->protect_control_groups
) {
2314 MountEntry
*me
= mount_list_extend(&ml
);
2316 return log_oom_debug();
2318 *me
= (MountEntry
) {
2319 .path_const
= "/sys/fs/cgroup",
2320 .mode
= MOUNT_READ_ONLY
,
2324 r
= append_protect_home(&ml
, p
->protect_home
, p
->ignore_protect_paths
);
2328 r
= append_protect_system(&ml
, p
->protect_system
, false);
2332 if (namespace_parameters_mount_apivfs(p
)) {
2333 r
= append_static_mounts(&ml
,
2335 ELEMENTSOF(apivfs_table
),
2336 p
->ignore_protect_paths
);
2341 /* Note, if proc is mounted with subset=pid then neither of the two paths will exist, i.e. they are
2342 * implicitly protected by the mount option. */
2343 if (p
->protect_hostname
) {
2344 r
= append_static_mounts(
2346 protect_hostname_table
,
2347 ELEMENTSOF(protect_hostname_table
),
2348 ignore_protect_proc
);
2353 if (p
->private_network
) {
2354 MountEntry
*me
= mount_list_extend(&ml
);
2356 return log_oom_debug();
2358 *me
= (MountEntry
) {
2359 .path_const
= "/sys",
2360 .mode
= MOUNT_PRIVATE_SYSFS
,
2364 if (p
->private_ipc
) {
2365 MountEntry
*me
= mount_list_extend(&ml
);
2367 return log_oom_debug();
2369 *me
= (MountEntry
) {
2370 .path_const
= "/dev/mqueue",
2371 .mode
= MOUNT_MQUEUEFS
,
2372 .flags
= MS_NOSUID
| MS_NODEV
| MS_NOEXEC
| MS_RELATIME
,
2376 if (p
->creds_path
) {
2377 /* If our service has a credentials store configured, then bind that one in, but hide
2378 * everything else. */
2380 MountEntry
*me
= mount_list_extend(&ml
);
2382 return log_oom_debug();
2384 *me
= (MountEntry
) {
2385 .path_const
= "/run/credentials",
2386 .mode
= MOUNT_TMPFS
,
2388 .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
,
2389 .flags
= MS_NODEV
|MS_STRICTATIME
|MS_NOSUID
|MS_NOEXEC
,
2392 me
= mount_list_extend(&ml
);
2394 return log_oom_debug();
2396 *me
= (MountEntry
) {
2397 .path_const
= p
->creds_path
,
2400 .source_const
= p
->creds_path
,
2404 /* If our service has no credentials store configured, then make the whole credentials tree
2405 * inaccessible wholesale. */
2407 MountEntry
*me
= mount_list_extend(&ml
);
2409 return log_oom_debug();
2411 *me
= (MountEntry
) {
2412 .path_const
= "/run/credentials",
2413 .mode
= MOUNT_INACCESSIBLE
,
2418 if (p
->log_namespace
) {
2419 _cleanup_free_
char *q
= NULL
;
2421 q
= strjoin("/run/systemd/journal.", p
->log_namespace
);
2423 return log_oom_debug();
2425 MountEntry
*me
= mount_list_extend(&ml
);
2427 return log_oom_debug();
2429 *me
= (MountEntry
) {
2430 .path_const
= "/run/systemd/journal",
2431 .mode
= MOUNT_BIND_RECURSIVE
,
2433 .source_malloc
= TAKE_PTR(q
),
2437 /* Will be used to add bind mounts at runtime */
2438 if (setup_propagate
) {
2439 MountEntry
*me
= mount_list_extend(&ml
);
2441 return log_oom_debug();
2443 *me
= (MountEntry
) {
2444 .source_const
= p
->propagate_dir
,
2445 .path_const
= p
->incoming_dir
,
2451 if (p
->notify_socket
) {
2452 MountEntry
*me
= mount_list_extend(&ml
);
2454 return log_oom_debug();
2456 *me
= (MountEntry
) {
2457 .path_const
= p
->notify_socket
,
2458 .source_const
= p
->notify_socket
,
2464 if (p
->host_os_release_stage
) {
2465 MountEntry
*me
= mount_list_extend(&ml
);
2467 return log_oom_debug();
2469 *me
= (MountEntry
) {
2470 .path_const
= "/run/host/.os-release-stage/",
2471 .source_const
= p
->host_os_release_stage
,
2474 .ignore
= true, /* Live copy, don't hard-fail if it goes missing */
2478 /* Prepend the root directory where that's necessary */
2479 r
= prefix_where_needed(&ml
, root
);
2483 drop_unused_mounts(&ml
, root
);
2485 /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */
2487 if (unshare(CLONE_NEWNS
) < 0) {
2488 r
= log_debug_errno(errno
, "Failed to unshare the mount namespace: %m");
2490 if (ERRNO_IS_PRIVILEGE(r
) ||
2491 ERRNO_IS_NOT_SUPPORTED(r
))
2492 /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter
2493 * in place that doesn't allow us to create namespaces (or a missing cap), then
2494 * propagate a recognizable error back, which the caller can use to detect this case
2495 * (and only this) and optionally continue without namespacing applied. */
2501 /* Create the source directory to allow runtime propagation of mounts */
2502 if (setup_propagate
)
2503 (void) mkdir_p(p
->propagate_dir
, 0600);
2505 if (p
->n_extension_images
> 0 || !strv_isempty(p
->extension_directories
))
2506 /* ExtensionImages/Directories mountpoint directories will be created while parsing the
2507 * mounts to create, so have the parent ready */
2508 (void) mkdir_p(p
->extension_dir
, 0600);
2510 /* Remount / as SLAVE so that nothing now mounted in the namespace
2511 * shows up in the parent */
2512 if (mount(NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
) < 0)
2513 return log_debug_errno(errno
, "Failed to remount '/' as SLAVE: %m");
2515 if (p
->root_image
) {
2516 /* A root image is specified, mount it to the right place */
2517 r
= dissected_image_mount(
2520 /* uid_shift= */ UID_INVALID
,
2521 /* uid_range= */ UID_INVALID
,
2522 /* userns_fd= */ -EBADF
,
2523 dissect_image_flags
);
2525 return log_debug_errno(r
, "Failed to mount root image: %m");
2527 /* Now release the block device lock, so that udevd is free to call BLKRRPART on the device
2529 r
= loop_device_flock(loop_device
, LOCK_UN
);
2531 return log_debug_errno(r
, "Failed to release lock on loopback block device: %m");
2533 r
= dissected_image_relinquish(dissected_image
);
2535 return log_debug_errno(r
, "Failed to relinquish dissected image: %m");
2537 } else if (p
->root_directory
) {
2539 /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
2540 r
= path_is_mount_point_full(root
, /* root = */ NULL
, AT_SYMLINK_FOLLOW
);
2542 return log_debug_errno(r
, "Failed to detect that %s is a mount point or not: %m", root
);
2544 r
= mount_nofollow_verbose(LOG_DEBUG
, root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2550 /* Let's mount the main root directory to the root directory to use */
2551 r
= mount_nofollow_verbose(LOG_DEBUG
, "/", root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2556 /* Try to set up the new root directory before mounting anything else there. */
2557 if (p
->root_image
|| p
->root_directory
)
2558 (void) base_filesystem_create(root
, UID_INVALID
, GID_INVALID
);
2560 /* Now make the magic happen */
2561 r
= apply_mounts(&ml
, root
, p
, error_path
);
2565 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
2566 r
= mount_switch_root(root
, /* mount_propagation_flag = */ 0);
2567 if (r
== -EINVAL
&& p
->root_directory
) {
2568 /* If we are using root_directory and we don't have privileges (ie: user manager in a user
2569 * namespace) and the root_directory is already a mount point in the parent namespace,
2570 * MS_MOVE will fail as we don't have permission to change it (with EINVAL rather than
2571 * EPERM). Attempt to bind-mount it over itself (like we do above if it's not already a
2572 * mount point) and try again. */
2573 r
= mount_nofollow_verbose(LOG_DEBUG
, root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2576 r
= mount_switch_root(root
, /* mount_propagation_flag = */ 0);
2579 return log_debug_errno(r
, "Failed to mount root with MS_MOVE: %m");
2581 /* Remount / as the desired mode. Note that this will not reestablish propagation from our side to
2582 * the host, since what's disconnected is disconnected. */
2583 if (mount(NULL
, "/", NULL
, mount_propagation_flag
| MS_REC
, NULL
) < 0)
2584 return log_debug_errno(errno
, "Failed to remount '/' with desired mount flags: %m");
2586 /* bind_mount_in_namespace() will MS_MOVE into that directory, and that's only supported for
2587 * non-shared mounts. This needs to happen after remounting / or it will fail. */
2588 if (setup_propagate
&& mount(NULL
, p
->incoming_dir
, NULL
, MS_SLAVE
, NULL
) < 0)
2589 return log_debug_errno(errno
, "Failed to remount %s with MS_SLAVE: %m", p
->incoming_dir
);
2594 void bind_mount_free_many(BindMount
*b
, size_t n
) {
2595 assert(b
|| n
== 0);
2597 for (size_t i
= 0; i
< n
; i
++) {
2599 free(b
[i
].destination
);
2605 int bind_mount_add(BindMount
**b
, size_t *n
, const BindMount
*item
) {
2606 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
2613 s
= strdup(item
->source
);
2617 d
= strdup(item
->destination
);
2621 c
= reallocarray(*b
, *n
+ 1, sizeof(BindMount
));
2627 c
[(*n
)++] = (BindMount
) {
2628 .source
= TAKE_PTR(s
),
2629 .destination
= TAKE_PTR(d
),
2630 .read_only
= item
->read_only
,
2631 .nosuid
= item
->nosuid
,
2632 .recursive
= item
->recursive
,
2633 .ignore_enoent
= item
->ignore_enoent
,
2639 MountImage
* mount_image_free_many(MountImage
*m
, size_t *n
) {
2641 assert(m
|| *n
== 0);
2643 for (size_t i
= 0; i
< *n
; i
++) {
2645 free(m
[i
].destination
);
2646 mount_options_free_all(m
[i
].mount_options
);
2654 int mount_image_add(MountImage
**m
, size_t *n
, const MountImage
*item
) {
2655 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
2656 _cleanup_(mount_options_free_allp
) MountOptions
*options
= NULL
;
2663 s
= strdup(item
->source
);
2667 if (item
->destination
) {
2668 d
= strdup(item
->destination
);
2673 LIST_FOREACH(mount_options
, i
, item
->mount_options
) {
2674 _cleanup_(mount_options_free_allp
) MountOptions
*o
= NULL
;
2676 o
= new(MountOptions
, 1);
2680 *o
= (MountOptions
) {
2681 .partition_designator
= i
->partition_designator
,
2682 .options
= strdup(i
->options
),
2687 LIST_APPEND(mount_options
, options
, TAKE_PTR(o
));
2690 c
= reallocarray(*m
, *n
+ 1, sizeof(MountImage
));
2696 c
[(*n
)++] = (MountImage
) {
2697 .source
= TAKE_PTR(s
),
2698 .destination
= TAKE_PTR(d
),
2699 .mount_options
= TAKE_PTR(options
),
2700 .ignore_enoent
= item
->ignore_enoent
,
2707 void temporary_filesystem_free_many(TemporaryFileSystem
*t
, size_t n
) {
2708 assert(t
|| n
== 0);
2710 for (size_t i
= 0; i
< n
; i
++) {
2718 int temporary_filesystem_add(
2719 TemporaryFileSystem
**t
,
2722 const char *options
) {
2724 _cleanup_free_
char *p
= NULL
, *o
= NULL
;
2725 TemporaryFileSystem
*c
;
2735 if (!isempty(options
)) {
2736 o
= strdup(options
);
2741 c
= reallocarray(*t
, *n
+ 1, sizeof(TemporaryFileSystem
));
2747 c
[(*n
)++] = (TemporaryFileSystem
) {
2748 .path
= TAKE_PTR(p
),
2749 .options
= TAKE_PTR(o
),
2755 static int make_tmp_prefix(const char *prefix
) {
2756 _cleanup_free_
char *t
= NULL
;
2757 _cleanup_close_
int fd
= -EBADF
;
2760 /* Don't do anything unless we know the dir is actually missing */
2761 r
= access(prefix
, F_OK
);
2764 if (errno
!= ENOENT
)
2768 r
= mkdir_parents(prefix
, 0755);
2772 r
= tempfn_random(prefix
, NULL
, &t
);
2776 /* umask will corrupt this access mode, but that doesn't matter, we need to call chmod() anyway for
2777 * the suid bit, below. */
2778 fd
= open_mkdir_at(AT_FDCWD
, t
, O_EXCL
|O_CLOEXEC
, 0777);
2782 r
= RET_NERRNO(fchmod(fd
, 01777));
2788 r
= RET_NERRNO(rename(t
, prefix
));
2791 return r
== -EEXIST
? 0 : r
; /* it's fine if someone else created the dir by now */
2798 static int setup_one_tmp_dir(const char *id
, const char *prefix
, char **path
, char **tmp_path
) {
2799 _cleanup_free_
char *x
= NULL
;
2800 _cleanup_free_
char *y
= NULL
;
2809 /* We include the boot id in the directory so that after a
2810 * reboot we can easily identify obsolete directories. */
2812 r
= sd_id128_get_boot(&boot_id
);
2816 x
= strjoin(prefix
, "/systemd-private-", SD_ID128_TO_STRING(boot_id
), "-", id
, "-XXXXXX");
2820 r
= make_tmp_prefix(prefix
);
2826 if (errno
== EROFS
|| ERRNO_IS_DISK_SPACE(errno
))
2833 y
= strjoin(x
, "/tmp");
2838 if (mkdir(y
, 0777 | S_ISVTX
) < 0)
2841 r
= label_fix_full(AT_FDCWD
, y
, prefix
, 0);
2846 *tmp_path
= TAKE_PTR(y
);
2848 /* Trouble: we failed to create the directory. Instead of failing, let's simulate /tmp being
2849 * read-only. This way the service will get the EROFS result as if it was writing to the real
2852 r
= mkdir_p(RUN_SYSTEMD_EMPTY
, 0500);
2856 r
= free_and_strdup(&x
, RUN_SYSTEMD_EMPTY
);
2861 *path
= TAKE_PTR(x
);
2865 int setup_tmp_dirs(const char *id
, char **tmp_dir
, char **var_tmp_dir
) {
2866 _cleanup_(namespace_cleanup_tmpdirp
) char *a
= NULL
;
2867 _cleanup_(rmdir_and_freep
) char *a_tmp
= NULL
;
2873 assert(var_tmp_dir
);
2875 r
= setup_one_tmp_dir(id
, "/tmp", &a
, &a_tmp
);
2879 r
= setup_one_tmp_dir(id
, "/var/tmp", &b
, NULL
);
2883 a_tmp
= mfree(a_tmp
); /* avoid rmdir */
2884 *tmp_dir
= TAKE_PTR(a
);
2885 *var_tmp_dir
= TAKE_PTR(b
);
2890 int setup_shareable_ns(int ns_storage_socket
[static 2], unsigned long nsflag
) {
2891 _cleanup_close_
int ns
= -EBADF
;
2892 const char *ns_name
, *ns_path
;
2895 assert(ns_storage_socket
);
2896 assert(ns_storage_socket
[0] >= 0);
2897 assert(ns_storage_socket
[1] >= 0);
2899 ns_name
= ASSERT_PTR(namespace_single_flag_to_string(nsflag
));
2901 /* We use the passed socketpair as a storage buffer for our namespace reference fd. Whatever process
2902 * runs this first shall create a new namespace, all others should just join it. To serialize that we
2903 * use a file lock on the socket pair.
2905 * It's a bit crazy, but hey, works great! */
2907 r
= posix_lock(ns_storage_socket
[0], LOCK_EX
);
2911 CLEANUP_POSIX_UNLOCK(ns_storage_socket
[0]);
2913 ns
= receive_one_fd(ns_storage_socket
[0], MSG_PEEK
|MSG_DONTWAIT
);
2915 /* Yay, found something, so let's join the namespace */
2916 r
= RET_NERRNO(setns(ns
, nsflag
));
2926 /* Nothing stored yet, so let's create a new namespace. */
2928 if (unshare(nsflag
) < 0)
2931 if (nsflag
== CLONE_NEWNET
)
2932 (void) loopback_setup();
2934 ns_path
= strjoina("/proc/self/ns/", ns_name
);
2935 ns
= open(ns_path
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
2939 r
= send_one_fd(ns_storage_socket
[1], ns
, MSG_DONTWAIT
);
2946 int open_shareable_ns_path(int ns_storage_socket
[static 2], const char *path
, unsigned long nsflag
) {
2947 _cleanup_close_
int ns
= -EBADF
;
2950 assert(ns_storage_socket
);
2951 assert(ns_storage_socket
[0] >= 0);
2952 assert(ns_storage_socket
[1] >= 0);
2955 /* If the storage socket doesn't contain a ns fd yet, open one via the file system and store it in
2956 * it. This is supposed to be called ahead of time, i.e. before setup_shareable_ns() which will
2957 * allocate a new anonymous ns if needed. */
2959 r
= posix_lock(ns_storage_socket
[0], LOCK_EX
);
2963 CLEANUP_POSIX_UNLOCK(ns_storage_socket
[0]);
2965 ns
= receive_one_fd(ns_storage_socket
[0], MSG_PEEK
|MSG_DONTWAIT
);
2971 /* Nothing stored yet. Open the file from the file system. */
2973 ns
= open(path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
2977 r
= fd_is_ns(ns
, nsflag
);
2980 if (r
< 0 && r
!= -EUCLEAN
) /* EUCLEAN: we don't know */
2983 r
= send_one_fd(ns_storage_socket
[1], ns
, MSG_DONTWAIT
);
2990 bool ns_type_supported(NamespaceType type
) {
2991 const char *t
, *ns_proc
;
2993 t
= namespace_type_to_string(type
);
2994 if (!t
) /* Don't know how to translate this? Then it's not supported */
2997 ns_proc
= strjoina("/proc/self/ns/", t
);
2998 return access(ns_proc
, F_OK
) == 0;
3001 static const char *const protect_home_table
[_PROTECT_HOME_MAX
] = {
3002 [PROTECT_HOME_NO
] = "no",
3003 [PROTECT_HOME_YES
] = "yes",
3004 [PROTECT_HOME_READ_ONLY
] = "read-only",
3005 [PROTECT_HOME_TMPFS
] = "tmpfs",
3008 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home
, ProtectHome
, PROTECT_HOME_YES
);
3010 static const char *const protect_system_table
[_PROTECT_SYSTEM_MAX
] = {
3011 [PROTECT_SYSTEM_NO
] = "no",
3012 [PROTECT_SYSTEM_YES
] = "yes",
3013 [PROTECT_SYSTEM_FULL
] = "full",
3014 [PROTECT_SYSTEM_STRICT
] = "strict",
3017 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system
, ProtectSystem
, PROTECT_SYSTEM_YES
);
3019 static const char* const namespace_type_table
[] = {
3020 [NAMESPACE_MOUNT
] = "mnt",
3021 [NAMESPACE_CGROUP
] = "cgroup",
3022 [NAMESPACE_UTS
] = "uts",
3023 [NAMESPACE_IPC
] = "ipc",
3024 [NAMESPACE_USER
] = "user",
3025 [NAMESPACE_PID
] = "pid",
3026 [NAMESPACE_NET
] = "net",
3027 [NAMESPACE_TIME
] = "time",
3030 DEFINE_STRING_TABLE_LOOKUP(namespace_type
, NamespaceType
);
3032 static const char* const protect_proc_table
[_PROTECT_PROC_MAX
] = {
3033 [PROTECT_PROC_DEFAULT
] = "default",
3034 [PROTECT_PROC_NOACCESS
] = "noaccess",
3035 [PROTECT_PROC_INVISIBLE
] = "invisible",
3036 [PROTECT_PROC_PTRACEABLE
] = "ptraceable",
3039 DEFINE_STRING_TABLE_LOOKUP(protect_proc
, ProtectProc
);
3041 static const char* const proc_subset_table
[_PROC_SUBSET_MAX
] = {
3042 [PROC_SUBSET_ALL
] = "all",
3043 [PROC_SUBSET_PID
] = "pid",
3046 DEFINE_STRING_TABLE_LOOKUP(proc_subset
, ProcSubset
);