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
);
112 typedef struct MountList
{
117 /* 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
118 * something there already. These mounts are hence overridden by any other explicitly configured mounts. */
119 static const MountEntry apivfs_table
[] = {
120 { "/proc", MOUNT_PROCFS
, false },
121 { "/dev", MOUNT_BIND_DEV
, false },
122 { "/sys", MOUNT_BIND_SYSFS
, false },
123 { "/run", MOUNT_RUN
, false, .options_const
= "mode=0755" TMPFS_LIMITS_RUN
, .flags
= MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
},
126 /* ProtectKernelTunables= option and the related filesystem APIs */
127 static const MountEntry protect_kernel_tunables_proc_table
[] = {
128 { "/proc/acpi", MOUNT_READ_ONLY
, true },
129 { "/proc/apm", MOUNT_READ_ONLY
, true }, /* Obsolete API, there's no point in permitting access to this, ever */
130 { "/proc/asound", MOUNT_READ_ONLY
, true },
131 { "/proc/bus", MOUNT_READ_ONLY
, true },
132 { "/proc/fs", MOUNT_READ_ONLY
, true },
133 { "/proc/irq", MOUNT_READ_ONLY
, true },
134 { "/proc/kallsyms", MOUNT_INACCESSIBLE
, true },
135 { "/proc/kcore", MOUNT_INACCESSIBLE
, true },
136 { "/proc/latency_stats", MOUNT_READ_ONLY
, true },
137 { "/proc/mtrr", MOUNT_READ_ONLY
, true },
138 { "/proc/scsi", MOUNT_READ_ONLY
, true },
139 { "/proc/sys", MOUNT_READ_ONLY
, true },
140 { "/proc/sysrq-trigger", MOUNT_READ_ONLY
, true },
141 { "/proc/timer_stats", MOUNT_READ_ONLY
, true },
144 static const MountEntry protect_kernel_tunables_sys_table
[] = {
145 { "/sys", MOUNT_READ_ONLY
, false },
146 { "/sys/fs/bpf", MOUNT_READ_ONLY
, true },
147 { "/sys/fs/cgroup", MOUNT_READ_WRITE_IMPLICIT
, false }, /* READ_ONLY is set by ProtectControlGroups= option */
148 { "/sys/fs/selinux", MOUNT_READ_WRITE_IMPLICIT
, true },
149 { "/sys/kernel/debug", MOUNT_READ_ONLY
, true },
150 { "/sys/kernel/tracing", MOUNT_READ_ONLY
, true },
153 /* ProtectKernelModules= option */
154 static const MountEntry protect_kernel_modules_table
[] = {
155 { "/usr/lib/modules", MOUNT_INACCESSIBLE
, true },
158 /* ProtectKernelLogs= option */
159 static const MountEntry protect_kernel_logs_proc_table
[] = {
160 { "/proc/kmsg", MOUNT_INACCESSIBLE
, true },
163 static const MountEntry protect_kernel_logs_dev_table
[] = {
164 { "/dev/kmsg", MOUNT_INACCESSIBLE
, true },
168 * ProtectHome=read-only table, protect $HOME and $XDG_RUNTIME_DIR and rest of
169 * system should be protected by ProtectSystem=
171 static const MountEntry protect_home_read_only_table
[] = {
172 { "/home", MOUNT_READ_ONLY
, true },
173 { "/run/user", MOUNT_READ_ONLY
, true },
174 { "/root", MOUNT_READ_ONLY
, true },
177 /* ProtectHome=tmpfs table */
178 static const MountEntry protect_home_tmpfs_table
[] = {
179 { "/home", MOUNT_TMPFS
, true, .read_only
= true, .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
180 { "/run/user", MOUNT_TMPFS
, true, .read_only
= true, .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
181 { "/root", MOUNT_TMPFS
, true, .read_only
= true, .options_const
= "mode=0700" TMPFS_LIMITS_EMPTY_OR_ALMOST
, .flags
= MS_NODEV
|MS_STRICTATIME
},
184 /* ProtectHome=yes table */
185 static const MountEntry protect_home_yes_table
[] = {
186 { "/home", MOUNT_INACCESSIBLE
, true },
187 { "/run/user", MOUNT_INACCESSIBLE
, true },
188 { "/root", MOUNT_INACCESSIBLE
, true },
191 /* ProtectSystem=yes table */
192 static const MountEntry protect_system_yes_table
[] = {
193 { "/usr", MOUNT_READ_ONLY
, false },
194 { "/boot", MOUNT_READ_ONLY
, true },
195 { "/efi", MOUNT_READ_ONLY
, true },
198 /* ProtectSystem=full includes ProtectSystem=yes */
199 static const MountEntry protect_system_full_table
[] = {
200 { "/usr", MOUNT_READ_ONLY
, false },
201 { "/boot", MOUNT_READ_ONLY
, true },
202 { "/efi", MOUNT_READ_ONLY
, true },
203 { "/etc", MOUNT_READ_ONLY
, false },
206 /* ProtectSystem=strict table. In this strict mode, we mount everything read-only, except for /proc, /dev,
207 * /sys which are the kernel API VFS, which are left writable, but PrivateDevices= + ProtectKernelTunables=
208 * protect those, and these options should be fully orthogonal. (And of course /home and friends are also
209 * left writable, as ProtectHome= shall manage those, orthogonally).
211 static const MountEntry protect_system_strict_table
[] = {
212 { "/", MOUNT_READ_ONLY
, false },
213 { "/proc", MOUNT_READ_WRITE_IMPLICIT
, false }, /* ProtectKernelTunables= */
214 { "/sys", MOUNT_READ_WRITE_IMPLICIT
, false }, /* ProtectKernelTunables= */
215 { "/dev", MOUNT_READ_WRITE_IMPLICIT
, false }, /* PrivateDevices= */
216 { "/home", MOUNT_READ_WRITE_IMPLICIT
, true }, /* ProtectHome= */
217 { "/run/user", MOUNT_READ_WRITE_IMPLICIT
, true }, /* ProtectHome= */
218 { "/root", MOUNT_READ_WRITE_IMPLICIT
, true }, /* ProtectHome= */
221 /* ProtectHostname=yes able */
222 static const MountEntry protect_hostname_table
[] = {
223 { "/proc/sys/kernel/hostname", MOUNT_READ_ONLY
, false },
224 { "/proc/sys/kernel/domainname", MOUNT_READ_ONLY
, false },
227 static const char * const mount_mode_table
[_MOUNT_MODE_MAX
] = {
228 [MOUNT_INACCESSIBLE
] = "inaccessible",
229 [MOUNT_OVERLAY
] = "overlay",
230 [MOUNT_IMAGE
] = "image",
231 [MOUNT_BIND
] = "bind",
232 [MOUNT_BIND_RECURSIVE
] = "bind-recursive",
233 [MOUNT_PRIVATE_TMP
] = "private-tmp",
234 [MOUNT_PRIVATE_TMP_READ_ONLY
] = "private-tmp-read-only",
235 [MOUNT_PRIVATE_DEV
] = "private-dev",
236 [MOUNT_BIND_DEV
] = "bind-dev",
237 [MOUNT_EMPTY_DIR
] = "empty-dir",
238 [MOUNT_PRIVATE_SYSFS
] = "private-sysfs",
239 [MOUNT_BIND_SYSFS
] = "bind-sysfs",
240 [MOUNT_PROCFS
] = "procfs",
241 [MOUNT_READ_ONLY
] = "read-only",
242 [MOUNT_READ_WRITE
] = "read-write",
243 [MOUNT_NOEXEC
] = "noexec",
244 [MOUNT_EXEC
] = "exec",
245 [MOUNT_TMPFS
] = "tmpfs",
247 [MOUNT_EXTENSION_DIRECTORY
] = "extension-directory",
248 [MOUNT_EXTENSION_IMAGE
] = "extension-image",
249 [MOUNT_MQUEUEFS
] = "mqueuefs",
250 [MOUNT_READ_WRITE_IMPLICIT
] = "read-write-implicit",
253 /* Helper struct for naming simplicity and reusability */
254 static const struct {
255 const char *level_env
;
256 const char *level_env_print
;
257 } image_class_info
[_IMAGE_CLASS_MAX
] = {
259 .level_env
= "SYSEXT_LEVEL",
260 .level_env_print
= " SYSEXT_LEVEL=",
263 .level_env
= "CONFEXT_LEVEL",
264 .level_env_print
= " CONFEXT_LEVEL=",
268 DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(mount_mode
, MountMode
);
270 static const char *mount_entry_path(const MountEntry
*p
) {
273 /* Returns the path of this bind mount. If the malloc()-allocated ->path_buffer field is set we return that,
274 * otherwise the stack/static ->path field is returned. */
276 return p
->path_malloc
?: p
->path_const
;
279 static const char *mount_entry_unprefixed_path(const MountEntry
*p
) {
282 /* Returns the unprefixed path (ie: before prefix_where_needed() ran), if any */
284 return p
->unprefixed_path_malloc
?: p
->unprefixed_path_const
?: mount_entry_path(p
);
287 static void mount_entry_consume_prefix(MountEntry
*p
, char *new_path
) {
289 assert(p
->path_malloc
|| p
->path_const
);
292 /* Saves current path in unprefixed_ variable, and takes over new_path */
294 free_and_replace(p
->unprefixed_path_malloc
, p
->path_malloc
);
295 /* If we didn't have a path on the heap, then it's a static one */
296 if (!p
->unprefixed_path_malloc
)
297 p
->unprefixed_path_const
= p
->path_const
;
298 p
->path_malloc
= new_path
;
299 p
->has_prefix
= true;
302 static bool mount_entry_read_only(const MountEntry
*p
) {
305 return p
->read_only
|| IN_SET(p
->mode
, MOUNT_READ_ONLY
, MOUNT_INACCESSIBLE
, MOUNT_PRIVATE_TMP_READ_ONLY
);
308 static bool mount_entry_noexec(const MountEntry
*p
) {
311 return p
->noexec
|| IN_SET(p
->mode
, MOUNT_NOEXEC
, MOUNT_INACCESSIBLE
, MOUNT_PRIVATE_SYSFS
, MOUNT_BIND_SYSFS
, MOUNT_PROCFS
);
314 static bool mount_entry_exec(const MountEntry
*p
) {
317 return p
->exec
|| p
->mode
== MOUNT_EXEC
;
320 static const char *mount_entry_source(const MountEntry
*p
) {
323 return p
->source_malloc
?: p
->source_const
;
326 static const char *mount_entry_options(const MountEntry
*p
) {
329 return p
->options_malloc
?: p
->options_const
;
332 static void mount_entry_done(MountEntry
*p
) {
335 p
->path_malloc
= mfree(p
->path_malloc
);
336 p
->unprefixed_path_malloc
= mfree(p
->unprefixed_path_malloc
);
337 p
->source_malloc
= mfree(p
->source_malloc
);
338 p
->options_malloc
= mfree(p
->options_malloc
);
341 static void mount_list_done(MountList
*ml
) {
344 FOREACH_ARRAY(m
, ml
->mounts
, ml
->n_mounts
)
347 ml
->mounts
= mfree(ml
->mounts
);
351 static MountEntry
*mount_list_extend(MountList
*ml
) {
354 if (!GREEDY_REALLOC0(ml
->mounts
, ml
->n_mounts
+1))
357 return ml
->mounts
+ ml
->n_mounts
++;
360 static int append_access_mounts(MountList
*ml
, char **strv
, MountMode mode
, bool forcibly_require_prefix
) {
363 /* Adds a list of user-supplied READ_WRITE/READ_WRITE_IMPLICIT/READ_ONLY/INACCESSIBLE entries */
365 STRV_FOREACH(i
, strv
) {
366 bool ignore
= false, needs_prefix
= false;
369 /* Look for any prefixes */
370 if (startswith(e
, "-")) {
374 if (startswith(e
, "+")) {
379 if (!path_is_absolute(e
))
380 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Path is not absolute: %s", e
);
382 MountEntry
*me
= mount_list_extend(ml
);
384 return log_oom_debug();
390 .has_prefix
= !needs_prefix
&& !forcibly_require_prefix
,
397 static int append_empty_dir_mounts(MountList
*ml
, char **strv
) {
400 /* Adds tmpfs mounts to provide readable but empty directories. This is primarily used to implement the
401 * "/private/" boundary directories for DynamicUser=1. */
403 STRV_FOREACH(i
, strv
) {
404 MountEntry
*me
= mount_list_extend(ml
);
406 return log_oom_debug();
410 .mode
= MOUNT_EMPTY_DIR
,
413 .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
,
414 .flags
= MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
,
421 static int append_bind_mounts(MountList
*ml
, const BindMount
*binds
, size_t n
) {
423 assert(binds
|| n
== 0);
425 FOREACH_ARRAY(b
, binds
, n
) {
426 MountEntry
*me
= mount_list_extend(ml
);
428 return log_oom_debug();
431 .path_const
= b
->destination
,
432 .mode
= b
->recursive
? MOUNT_BIND_RECURSIVE
: MOUNT_BIND
,
433 .read_only
= b
->read_only
,
435 .source_const
= b
->source
,
436 .ignore
= b
->ignore_enoent
,
443 static int append_mount_images(MountList
*ml
, const MountImage
*mount_images
, size_t n
) {
445 assert(mount_images
|| n
== 0);
447 FOREACH_ARRAY(m
, mount_images
, n
) {
448 MountEntry
*me
= mount_list_extend(ml
);
450 return log_oom_debug();
453 .path_const
= m
->destination
,
455 .source_const
= m
->source
,
456 .image_options_const
= m
->mount_options
,
457 .ignore
= m
->ignore_enoent
,
464 static int append_extensions(
467 const char *extension_dir
,
469 const MountImage
*mount_images
,
471 char **extension_directories
) {
473 _cleanup_strv_free_
char **overlays
= NULL
;
478 if (n
== 0 && strv_isempty(extension_directories
))
481 assert(extension_dir
);
483 /* Prepare a list of overlays, that will have as each element a string suitable for being
484 * passed as a lowerdir= parameter, so start with the hierarchy on the root.
485 * The overlays vector will have the same number of elements and will correspond to the
486 * hierarchies vector, so they can be iterated upon together. */
487 STRV_FOREACH(hierarchy
, hierarchies
) {
488 _cleanup_free_
char *prefixed_hierarchy
= NULL
;
490 prefixed_hierarchy
= path_join(root
, *hierarchy
);
491 if (!prefixed_hierarchy
)
494 r
= strv_consume(&overlays
, TAKE_PTR(prefixed_hierarchy
));
499 /* First, prepare a mount for each image, but these won't be visible to the unit, instead
500 * they will be mounted in our propagate directory, and used as a source for the overlay. */
501 for (size_t i
= 0; i
< n
; i
++) {
502 _cleanup_free_
char *mount_point
= NULL
;
503 const MountImage
*m
= mount_images
+ i
;
505 if (asprintf(&mount_point
, "%s/%zu", extension_dir
, i
) < 0)
508 for (size_t j
= 0; hierarchies
&& hierarchies
[j
]; ++j
) {
509 _cleanup_free_
char *prefixed_hierarchy
= NULL
, *escaped
= NULL
, *lowerdir
= NULL
;
511 prefixed_hierarchy
= path_join(mount_point
, hierarchies
[j
]);
512 if (!prefixed_hierarchy
)
515 escaped
= shell_escape(prefixed_hierarchy
, ",:");
519 /* Note that lowerdir= parameters are in 'reverse' order, so the
520 * top-most directory in the overlay comes first in the list. */
521 lowerdir
= strjoin(escaped
, ":", overlays
[j
]);
525 free_and_replace(overlays
[j
], lowerdir
);
528 MountEntry
*me
= mount_list_extend(ml
);
530 return log_oom_debug();
533 .path_malloc
= TAKE_PTR(mount_point
),
534 .image_options_const
= m
->mount_options
,
535 .ignore
= m
->ignore_enoent
,
536 .source_const
= m
->source
,
537 .mode
= MOUNT_EXTENSION_IMAGE
,
542 /* Secondly, extend the lowerdir= parameters with each ExtensionDirectory.
543 * Bind mount them in the same location as the ExtensionImages, so that we
544 * can check that they are valid trees (extension-release.d). */
545 STRV_FOREACH(extension_directory
, extension_directories
) {
546 _cleanup_free_
char *mount_point
= NULL
, *source
= NULL
;
547 const char *e
= *extension_directory
;
548 bool ignore_enoent
= false;
550 /* Pick up the counter where the ExtensionImages left it. */
551 if (asprintf(&mount_point
, "%s/%zu", extension_dir
, n
++) < 0)
554 /* Look for any prefixes */
555 if (startswith(e
, "-")) {
557 ignore_enoent
= true;
559 /* Ignore this for now */
560 if (startswith(e
, "+"))
567 for (size_t j
= 0; hierarchies
&& hierarchies
[j
]; ++j
) {
568 _cleanup_free_
char *prefixed_hierarchy
= NULL
, *escaped
= NULL
, *lowerdir
= NULL
;
570 prefixed_hierarchy
= path_join(mount_point
, hierarchies
[j
]);
571 if (!prefixed_hierarchy
)
574 escaped
= shell_escape(prefixed_hierarchy
, ",:");
578 /* Note that lowerdir= parameters are in 'reverse' order, so the
579 * top-most directory in the overlay comes first in the list. */
580 lowerdir
= strjoin(escaped
, ":", overlays
[j
]);
584 free_and_replace(overlays
[j
], lowerdir
);
587 MountEntry
*me
= mount_list_extend(ml
);
589 return log_oom_debug();
592 .path_malloc
= TAKE_PTR(mount_point
),
593 .source_malloc
= TAKE_PTR(source
),
594 .mode
= MOUNT_EXTENSION_DIRECTORY
,
595 .ignore
= ignore_enoent
,
601 /* Then, for each hierarchy, prepare an overlay with the list of lowerdir= strings
603 for (size_t i
= 0; hierarchies
&& hierarchies
[i
]; ++i
) {
604 _cleanup_free_
char *prefixed_hierarchy
= NULL
;
606 prefixed_hierarchy
= path_join(root
, hierarchies
[i
]);
607 if (!prefixed_hierarchy
)
610 MountEntry
*me
= mount_list_extend(ml
);
612 return log_oom_debug();
615 .path_malloc
= TAKE_PTR(prefixed_hierarchy
),
616 .options_malloc
= TAKE_PTR(overlays
[i
]),
617 .mode
= MOUNT_OVERLAY
,
619 .ignore
= true, /* If the source image doesn't set the ignore bit it will fail earlier. */
626 static int append_tmpfs_mounts(MountList
*ml
, const TemporaryFileSystem
*tmpfs
, size_t n
) {
628 assert(tmpfs
|| n
== 0);
630 FOREACH_ARRAY(t
, tmpfs
, n
) {
631 _cleanup_free_
char *o
= NULL
, *str
= NULL
;
636 if (!path_is_absolute(t
->path
))
637 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
), "Path is not absolute: %s", t
->path
);
639 str
= strjoin("mode=0755" NESTED_TMPFS_LIMITS
",", t
->options
);
643 r
= mount_option_mangle(str
, MS_NODEV
|MS_STRICTATIME
, &flags
, &o
);
645 return log_debug_errno(r
, "Failed to parse mount option '%s': %m", str
);
647 ro
= flags
& MS_RDONLY
;
651 MountEntry
*me
= mount_list_extend(ml
);
653 return log_oom_debug();
656 .path_const
= t
->path
,
659 .options_malloc
= TAKE_PTR(o
),
667 static int append_static_mounts(MountList
*ml
, const MountEntry
*mounts
, size_t n
, bool ignore_protect
) {
669 assert(mounts
|| n
== 0);
671 /* Adds a list of static pre-defined entries */
673 FOREACH_ARRAY(m
, mounts
, n
) {
674 MountEntry
*me
= mount_list_extend(ml
);
676 return log_oom_debug();
679 .path_const
= mount_entry_path(m
),
681 .ignore
= m
->ignore
|| ignore_protect
,
688 static int append_protect_home(MountList
*ml
, ProtectHome protect_home
, bool ignore_protect
) {
691 switch (protect_home
) {
693 case PROTECT_HOME_NO
:
696 case PROTECT_HOME_READ_ONLY
:
697 return append_static_mounts(ml
, protect_home_read_only_table
, ELEMENTSOF(protect_home_read_only_table
), ignore_protect
);
699 case PROTECT_HOME_TMPFS
:
700 return append_static_mounts(ml
, protect_home_tmpfs_table
, ELEMENTSOF(protect_home_tmpfs_table
), ignore_protect
);
702 case PROTECT_HOME_YES
:
703 return append_static_mounts(ml
, protect_home_yes_table
, ELEMENTSOF(protect_home_yes_table
), ignore_protect
);
706 assert_not_reached();
710 static int append_protect_system(MountList
*ml
, ProtectSystem protect_system
, bool ignore_protect
) {
713 switch (protect_system
) {
715 case PROTECT_SYSTEM_NO
:
718 case PROTECT_SYSTEM_STRICT
:
719 return append_static_mounts(ml
, protect_system_strict_table
, ELEMENTSOF(protect_system_strict_table
), ignore_protect
);
721 case PROTECT_SYSTEM_YES
:
722 return append_static_mounts(ml
, protect_system_yes_table
, ELEMENTSOF(protect_system_yes_table
), ignore_protect
);
724 case PROTECT_SYSTEM_FULL
:
725 return append_static_mounts(ml
, protect_system_full_table
, ELEMENTSOF(protect_system_full_table
), ignore_protect
);
728 assert_not_reached();
732 static int mount_path_compare(const MountEntry
*a
, const MountEntry
*b
) {
735 /* ExtensionImages/Directories will be used by other mounts as a base, so sort them first
736 * regardless of the prefix - they are set up in the propagate directory anyway */
737 d
= -CMP(a
->mode
== MOUNT_EXTENSION_IMAGE
, b
->mode
== MOUNT_EXTENSION_IMAGE
);
740 d
= -CMP(a
->mode
== MOUNT_EXTENSION_DIRECTORY
, b
->mode
== MOUNT_EXTENSION_DIRECTORY
);
744 /* If the paths are not equal, then order prefixes first */
745 d
= path_compare(mount_entry_path(a
), mount_entry_path(b
));
749 /* If the paths are equal, check the mode */
750 return CMP((int) a
->mode
, (int) b
->mode
);
753 static int prefix_where_needed(MountList
*ml
, const char *root_directory
) {
754 /* Prefixes all paths in the bind mount table with the root directory if the entry needs that. */
758 FOREACH_ARRAY(me
, ml
->mounts
, ml
->n_mounts
) {
764 s
= path_join(root_directory
, mount_entry_path(me
));
768 mount_entry_consume_prefix(me
, s
);
774 static void drop_duplicates(MountList
*ml
) {
775 MountEntry
*f
, *t
, *previous
;
779 /* Drops duplicate entries. Expects that the array is properly ordered already. */
781 for (f
= ml
->mounts
, t
= ml
->mounts
, previous
= NULL
; f
< ml
->mounts
+ ml
->n_mounts
; f
++) {
783 /* The first one wins (which is the one with the more restrictive mode), see mount_path_compare()
784 * above. Note that we only drop duplicates that haven't been mounted yet. */
786 path_equal(mount_entry_path(f
), mount_entry_path(previous
)) &&
787 f
->state
== MOUNT_PENDING
&& previous
->state
== MOUNT_PENDING
) {
788 log_debug("%s (%s) is duplicate.", mount_entry_path(f
), mount_mode_to_string(f
->mode
));
789 /* Propagate the flags to the remaining entry */
790 previous
->read_only
= previous
->read_only
|| mount_entry_read_only(f
);
791 previous
->noexec
= previous
->noexec
|| mount_entry_noexec(f
);
792 previous
->exec
= previous
->exec
|| mount_entry_exec(f
);
802 ml
->n_mounts
= t
- ml
->mounts
;
805 static void drop_inaccessible(MountList
*ml
) {
807 const char *clear
= NULL
;
811 /* Drops all entries obstructed by another entry further up the tree. Expects that the array is properly
812 * ordered already. */
814 for (f
= ml
->mounts
, t
= ml
->mounts
; f
< ml
->mounts
+ ml
->n_mounts
; f
++) {
816 /* If we found a path set for INACCESSIBLE earlier, and this entry has it as prefix we should drop
817 * it, as inaccessible paths really should drop the entire subtree. */
818 if (clear
&& path_startswith(mount_entry_path(f
), clear
)) {
819 log_debug("%s is masked by %s.", mount_entry_path(f
), clear
);
824 clear
= f
->mode
== MOUNT_INACCESSIBLE
? mount_entry_path(f
) : NULL
;
830 ml
->n_mounts
= t
- ml
->mounts
;
833 static void drop_nop(MountList
*ml
) {
838 /* Drops all entries which have an immediate parent that has the same type, as they are redundant. Assumes the
839 * list is ordered by prefixes. */
841 for (f
= ml
->mounts
, t
= ml
->mounts
; f
< ml
->mounts
+ ml
->n_mounts
; f
++) {
843 /* Only suppress such subtrees for READ_ONLY, READ_WRITE and READ_WRITE_IMPLICIT entries */
844 if (IN_SET(f
->mode
, MOUNT_READ_ONLY
, MOUNT_READ_WRITE
, MOUNT_READ_WRITE_IMPLICIT
)) {
845 MountEntry
*found
= NULL
;
847 /* Now let's find the first parent of the entry we are looking at. */
848 for (MountEntry
*p
= PTR_SUB1(t
, ml
->mounts
); p
; p
= PTR_SUB1(p
, ml
->mounts
))
849 if (path_startswith(mount_entry_path(f
), mount_entry_path(p
))) {
854 /* We found it, let's see if it's the same mode, if so, we can drop this entry */
855 if (found
&& found
->mode
== f
->mode
) {
856 log_debug("%s (%s) is made redundant by %s (%s)",
857 mount_entry_path(f
), mount_mode_to_string(f
->mode
),
858 mount_entry_path(found
), mount_mode_to_string(found
->mode
));
868 ml
->n_mounts
= t
- ml
->mounts
;
871 static void drop_outside_root(MountList
*ml
, const char *root_directory
) {
880 /* Drops all mounts that are outside of the root directory. */
882 for (f
= ml
->mounts
, t
= ml
->mounts
; f
< ml
->mounts
+ ml
->n_mounts
; f
++) {
884 /* ExtensionImages/Directories bases are opened in /run/systemd/unit-extensions on the host */
885 if (!IN_SET(f
->mode
, MOUNT_EXTENSION_IMAGE
, MOUNT_EXTENSION_DIRECTORY
) && !path_startswith(mount_entry_path(f
), root_directory
)) {
886 log_debug("%s is outside of root directory.", mount_entry_path(f
));
895 ml
->n_mounts
= t
- ml
->mounts
;
898 static int clone_device_node(
900 const char *temporary_mount
,
901 bool *make_devnode
) {
903 _cleanup_free_
char *sl
= NULL
;
904 const char *dn
, *bn
, *t
;
908 if (stat(d
, &st
) < 0) {
909 if (errno
== ENOENT
) {
910 log_debug_errno(errno
, "Device node '%s' to clone does not exist, ignoring.", d
);
914 return log_debug_errno(errno
, "Failed to stat() device node '%s' to clone, ignoring: %m", d
);
917 if (!S_ISBLK(st
.st_mode
) &&
918 !S_ISCHR(st
.st_mode
))
919 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
920 "Device node '%s' to clone is not a device node, ignoring.",
923 dn
= strjoina(temporary_mount
, d
);
925 /* First, try to create device node properly */
927 mac_selinux_create_file_prepare(d
, st
.st_mode
);
928 r
= mknod(dn
, st
.st_mode
, st
.st_rdev
);
929 mac_selinux_create_file_clear();
933 return log_debug_errno(errno
, "mknod failed for %s: %m", d
);
935 /* This didn't work, let's not try this again for the next iterations. */
936 *make_devnode
= false;
939 /* We're about to fall back to bind-mounting the device node. So create a dummy bind-mount target.
940 * Do not prepare device-node SELinux label (see issue 13762) */
941 r
= mknod(dn
, S_IFREG
, 0);
942 if (r
< 0 && errno
!= EEXIST
)
943 return log_debug_errno(errno
, "mknod() fallback failed for '%s': %m", d
);
945 /* Fallback to bind-mounting: The assumption here is that all used device nodes carry standard
946 * properties. Specifically, the devices nodes we bind-mount should either be owned by root:root or
947 * root:tty (e.g. /dev/tty, /dev/ptmx) and should not carry ACLs. */
948 r
= mount_nofollow_verbose(LOG_DEBUG
, d
, dn
, NULL
, MS_BIND
, NULL
);
953 bn
= path_startswith(d
, "/dev/");
957 /* Create symlinks like /dev/char/1:9 → ../urandom */
958 if (asprintf(&sl
, "%s/dev/%s/" DEVNUM_FORMAT_STR
,
960 S_ISCHR(st
.st_mode
) ? "char" : "block",
961 DEVNUM_FORMAT_VAL(st
.st_rdev
)) < 0)
962 return log_oom_debug();
964 (void) mkdir_parents(sl
, 0755);
966 t
= strjoina("../", bn
);
967 if (symlink(t
, sl
) < 0)
968 log_debug_errno(errno
, "Failed to symlink '%s' to '%s', ignoring: %m", t
, sl
);
973 static char *settle_runtime_dir(RuntimeScope scope
) {
976 if (scope
!= RUNTIME_SCOPE_USER
)
977 return strdup("/run/");
979 if (asprintf(&runtime_dir
, "/run/user/" UID_FMT
, geteuid()) < 0)
985 static int create_temporary_mount_point(RuntimeScope scope
, char **ret
) {
986 _cleanup_free_
char *runtime_dir
= NULL
, *temporary_mount
= NULL
;
990 runtime_dir
= settle_runtime_dir(scope
);
992 return log_oom_debug();
994 temporary_mount
= path_join(runtime_dir
, "systemd/namespace-XXXXXX");
995 if (!temporary_mount
)
996 return log_oom_debug();
998 if (!mkdtemp(temporary_mount
))
999 return log_debug_errno(errno
, "Failed to create temporary directory '%s': %m", temporary_mount
);
1001 *ret
= TAKE_PTR(temporary_mount
);
1005 static int mount_private_dev(MountEntry
*m
, RuntimeScope scope
) {
1006 static const char devnodes
[] =
1014 _cleanup_free_
char *temporary_mount
= NULL
;
1015 const char *dev
= NULL
, *devpts
= NULL
, *devshm
= NULL
, *devhugepages
= NULL
, *devmqueue
= NULL
, *devlog
= NULL
, *devptmx
= NULL
;
1016 bool can_mknod
= true;
1021 r
= create_temporary_mount_point(scope
, &temporary_mount
);
1025 dev
= strjoina(temporary_mount
, "/dev");
1026 (void) mkdir(dev
, 0755);
1027 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", dev
, "tmpfs", DEV_MOUNT_OPTIONS
, "mode=0755" TMPFS_LIMITS_PRIVATE_DEV
);
1031 r
= label_fix_full(AT_FDCWD
, dev
, "/dev", 0);
1033 log_debug_errno(r
, "Failed to fix label of '%s' as /dev: %m", dev
);
1037 devpts
= strjoina(temporary_mount
, "/dev/pts");
1038 (void) mkdir(devpts
, 0755);
1039 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev/pts", devpts
, NULL
, MS_BIND
, NULL
);
1043 /* /dev/ptmx can either be a device node or a symlink to /dev/pts/ptmx.
1044 * When /dev/ptmx a device node, /dev/pts/ptmx has 000 permissions making it inaccessible.
1045 * Thus, in that case make a clone.
1046 * In nspawn and other containers it will be a symlink, in that case make it a symlink. */
1047 r
= is_symlink("/dev/ptmx");
1049 log_debug_errno(r
, "Failed to detect whether /dev/ptmx is a symlink or not: %m");
1052 devptmx
= strjoina(temporary_mount
, "/dev/ptmx");
1053 if (symlink("pts/ptmx", devptmx
) < 0) {
1054 r
= log_debug_errno(errno
, "Failed to create a symlink '%s' to pts/ptmx: %m", devptmx
);
1058 r
= clone_device_node("/dev/ptmx", temporary_mount
, &can_mknod
);
1063 devshm
= strjoina(temporary_mount
, "/dev/shm");
1064 (void) mkdir(devshm
, 0755);
1065 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev/shm", devshm
, NULL
, MS_BIND
, NULL
);
1069 devmqueue
= strjoina(temporary_mount
, "/dev/mqueue");
1070 (void) mkdir(devmqueue
, 0755);
1071 (void) mount_nofollow_verbose(LOG_DEBUG
, "/dev/mqueue", devmqueue
, NULL
, MS_BIND
, NULL
);
1073 devhugepages
= strjoina(temporary_mount
, "/dev/hugepages");
1074 (void) mkdir(devhugepages
, 0755);
1075 (void) mount_nofollow_verbose(LOG_DEBUG
, "/dev/hugepages", devhugepages
, NULL
, MS_BIND
, NULL
);
1077 devlog
= strjoina(temporary_mount
, "/dev/log");
1078 if (symlink("/run/systemd/journal/dev-log", devlog
) < 0)
1079 log_debug_errno(errno
, "Failed to create a symlink '%s' to /run/systemd/journal/dev-log, ignoring: %m", devlog
);
1081 NULSTR_FOREACH(d
, devnodes
) {
1082 r
= clone_device_node(d
, temporary_mount
, &can_mknod
);
1083 /* ENXIO means the *source* is not a device file, skip creation in that case */
1084 if (r
< 0 && r
!= -ENXIO
)
1088 r
= dev_setup(temporary_mount
, UID_INVALID
, GID_INVALID
);
1090 log_debug_errno(r
, "Failed to set up basic device tree at '%s', ignoring: %m", temporary_mount
);
1092 /* Make the bind mount read-only. */
1093 r
= mount_nofollow_verbose(LOG_DEBUG
, NULL
, dev
, NULL
, MS_REMOUNT
|MS_BIND
|MS_RDONLY
, NULL
);
1097 /* Create the /dev directory if missing. It is more likely to be missing when the service is started
1098 * with RootDirectory. This is consistent with mount units creating the mount points when missing. */
1099 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1101 /* Unmount everything in old /dev */
1102 r
= umount_recursive(mount_entry_path(m
), 0);
1104 log_debug_errno(r
, "Failed to unmount directories below '%s', ignoring: %m", mount_entry_path(m
));
1106 r
= mount_nofollow_verbose(LOG_DEBUG
, dev
, mount_entry_path(m
), NULL
, MS_MOVE
, NULL
);
1111 (void) rmdir(temporary_mount
);
1117 (void) umount_verbose(LOG_DEBUG
, devpts
, UMOUNT_NOFOLLOW
);
1120 (void) umount_verbose(LOG_DEBUG
, devshm
, UMOUNT_NOFOLLOW
);
1123 (void) umount_verbose(LOG_DEBUG
, devhugepages
, UMOUNT_NOFOLLOW
);
1126 (void) umount_verbose(LOG_DEBUG
, devmqueue
, UMOUNT_NOFOLLOW
);
1128 (void) umount_verbose(LOG_DEBUG
, dev
, UMOUNT_NOFOLLOW
);
1130 (void) rmdir(temporary_mount
);
1135 static int mount_bind_dev(const MountEntry
*m
) {
1140 /* Implements the little brother of mount_private_dev(): simply bind mounts the host's /dev into the
1141 * service's /dev. This is only used when RootDirectory= is set. */
1143 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1145 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
1147 return log_debug_errno(r
, "Unable to determine whether /dev is already mounted: %m");
1148 if (r
> 0) /* make this a NOP if /dev is already a mount point */
1151 r
= mount_nofollow_verbose(LOG_DEBUG
, "/dev", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
);
1158 static int mount_bind_sysfs(const MountEntry
*m
) {
1163 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1165 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
1167 return log_debug_errno(r
, "Unable to determine whether /sys is already mounted: %m");
1168 if (r
> 0) /* make this a NOP if /sys is already a mount point */
1171 /* Bind mount the host's version so that we get all child mounts of it, too. */
1172 r
= mount_nofollow_verbose(LOG_DEBUG
, "/sys", mount_entry_path(m
), NULL
, MS_BIND
|MS_REC
, NULL
);
1179 static int mount_private_apivfs(
1181 const char *entry_path
,
1182 const char *bind_source
,
1184 RuntimeScope scope
) {
1186 _cleanup_(rmdir_and_freep
) char *temporary_mount
= NULL
;
1191 assert(bind_source
);
1193 (void) mkdir_p_label(entry_path
, 0755);
1195 /* First, check if we have enough privileges to mount a new instance. Note, a new sysfs instance
1196 * cannot be mounted on an already existing mount. Let's use a temporary place. */
1197 r
= create_temporary_mount_point(scope
, &temporary_mount
);
1201 r
= mount_nofollow_verbose(LOG_DEBUG
, fstype
, temporary_mount
, fstype
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, opts
);
1202 if (r
== -EINVAL
&& opts
)
1203 /* If this failed with EINVAL then this likely means the textual hidepid= stuff for procfs is
1204 * not supported by the kernel, and thus the per-instance hidepid= neither, which means we
1205 * really don't want to use it, since it would affect our host's /proc mount. Hence let's
1206 * gracefully fallback to a classic, unrestricted version. */
1207 r
= mount_nofollow_verbose(LOG_DEBUG
, fstype
, temporary_mount
, fstype
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
, /* opts = */ NULL
);
1208 if (ERRNO_IS_NEG_PRIVILEGE(r
)) {
1209 /* When we do not have enough privileges to mount a new instance, fall back to use an
1210 * existing mount. */
1212 r
= path_is_mount_point(entry_path
, /* root = */ NULL
, /* flags = */ 0);
1214 return log_debug_errno(r
, "Unable to determine whether '%s' is already mounted: %m", entry_path
);
1216 return 0; /* Use the current mount as is. */
1218 /* We lack permissions to mount a new instance, and it is not already mounted. But we can
1219 * access the host's, so as a final fallback bind-mount it to the destination, as most likely
1220 * we are inside a user manager in an unprivileged user namespace. */
1221 r
= mount_nofollow_verbose(LOG_DEBUG
, bind_source
, entry_path
, /* fstype = */ NULL
, MS_BIND
|MS_REC
, /* opts = */ NULL
);
1230 /* OK. We have a new mount instance. Let's clear an existing mount and its submounts. */
1231 r
= umount_recursive(entry_path
, /* flags = */ 0);
1233 log_debug_errno(r
, "Failed to unmount directories below '%s', ignoring: %m", entry_path
);
1235 /* Then, move the new mount instance. */
1236 r
= mount_nofollow_verbose(LOG_DEBUG
, temporary_mount
, entry_path
, /* fstype = */ NULL
, MS_MOVE
, /* opts = */ NULL
);
1240 /* We mounted a new instance now. Let's bind mount the children over now. This matters for nspawn
1241 * where a bunch of files are overmounted, in particular the boot id. */
1242 (void) bind_mount_submounts(bind_source
, entry_path
);
1246 static int mount_private_sysfs(const MountEntry
*m
, const NamespaceParameters
*p
) {
1249 return mount_private_apivfs("sysfs", mount_entry_path(m
), "/sys", /* opts = */ NULL
, p
->runtime_scope
);
1252 static int mount_procfs(const MountEntry
*m
, const NamespaceParameters
*p
) {
1253 _cleanup_free_
char *opts
= NULL
;
1258 if (p
->protect_proc
!= PROTECT_PROC_DEFAULT
||
1259 p
->proc_subset
!= PROC_SUBSET_ALL
) {
1261 /* Starting with kernel 5.8 procfs' hidepid= logic is truly per-instance (previously it
1262 * pretended to be per-instance but actually was per-namespace), hence let's make use of it
1263 * if requested. To make sure this logic succeeds only on kernels where hidepid= is
1264 * per-instance, we'll exclusively use the textual value for hidepid=, since support was
1265 * added in the same commit: if it's supported it is thus also per-instance. */
1267 const char *hpv
= p
->protect_proc
== PROTECT_PROC_DEFAULT
?
1269 protect_proc_to_string(p
->protect_proc
);
1271 /* hidepid= support was added in 5.8, so we can use fsconfig()/fsopen() (which were added in
1272 * 5.2) to check if hidepid= is supported. This avoids a noisy dmesg log by the kernel when
1273 * trying to use hidepid= on systems where it isn't supported. The same applies for subset=.
1274 * fsopen()/fsconfig() was also backported on some distros which allows us to detect
1275 * hidepid=/subset= support in even more scenarios. */
1277 if (mount_option_supported("proc", "hidepid", hpv
) != 0) {
1278 opts
= strjoin("hidepid=", hpv
);
1283 if (p
->proc_subset
== PROC_SUBSET_PID
&&
1284 mount_option_supported("proc", "subset", "pid") != 0)
1285 if (!strextend_with_separator(&opts
, ",", "subset=pid"))
1289 /* Mount a new instance, so that we get the one that matches our user namespace, if we are running in
1290 * one. i.e we don't reuse existing mounts here under any condition, we want a new instance owned by
1291 * our user namespace and with our hidepid= settings applied. Hence, let's get rid of everything
1292 * mounted on /proc/ first. */
1293 return mount_private_apivfs("proc", mount_entry_path(m
), "/proc", opts
, p
->runtime_scope
);
1296 static int mount_tmpfs(const MountEntry
*m
) {
1297 const char *entry_path
, *inner_path
;
1302 entry_path
= mount_entry_path(m
);
1303 inner_path
= mount_entry_unprefixed_path(m
);
1305 /* First, get rid of everything that is below if there is anything. Then, overmount with our new
1308 (void) mkdir_p_label(entry_path
, 0755);
1309 (void) umount_recursive(entry_path
, 0);
1311 r
= mount_nofollow_verbose(LOG_DEBUG
, "tmpfs", entry_path
, "tmpfs", m
->flags
, mount_entry_options(m
));
1315 r
= label_fix_full(AT_FDCWD
, entry_path
, inner_path
, 0);
1317 return log_debug_errno(r
, "Failed to fix label of '%s' as '%s': %m", entry_path
, inner_path
);
1322 static int mount_run(const MountEntry
*m
) {
1327 r
= path_is_mount_point(mount_entry_path(m
), NULL
, 0);
1328 if (r
< 0 && r
!= -ENOENT
)
1329 return log_debug_errno(r
, "Unable to determine whether /run is already mounted: %m");
1330 if (r
> 0) /* make this a NOP if /run is already a mount point */
1333 return mount_tmpfs(m
);
1336 static int mount_mqueuefs(const MountEntry
*m
) {
1338 const char *entry_path
;
1342 entry_path
= mount_entry_path(m
);
1344 (void) mkdir_p_label(entry_path
, 0755);
1345 (void) umount_recursive(entry_path
, 0);
1347 r
= mount_nofollow_verbose(LOG_DEBUG
, "mqueue", entry_path
, "mqueue", m
->flags
, mount_entry_options(m
));
1354 static int mount_image(
1355 const MountEntry
*m
,
1356 const char *root_directory
,
1357 const ImagePolicy
*image_policy
) {
1359 _cleanup_free_
char *host_os_release_id
= NULL
, *host_os_release_version_id
= NULL
,
1360 *host_os_release_sysext_level
= NULL
, *host_os_release_confext_level
= NULL
,
1361 *extension_name
= NULL
;
1366 r
= path_extract_filename(mount_entry_source(m
), &extension_name
);
1368 return log_debug_errno(r
, "Failed to extract extension name from %s: %m", mount_entry_source(m
));
1370 if (m
->mode
== MOUNT_EXTENSION_IMAGE
) {
1371 r
= parse_os_release(
1372 empty_to_root(root_directory
),
1373 "ID", &host_os_release_id
,
1374 "VERSION_ID", &host_os_release_version_id
,
1375 image_class_info
[IMAGE_SYSEXT
].level_env
, &host_os_release_sysext_level
,
1376 image_class_info
[IMAGE_CONFEXT
].level_env
, &host_os_release_confext_level
,
1379 return log_debug_errno(r
, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory
));
1380 if (isempty(host_os_release_id
))
1381 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
));
1384 r
= verity_dissect_and_mount(
1386 mount_entry_source(m
),
1387 mount_entry_path(m
),
1388 m
->image_options_const
,
1391 host_os_release_version_id
,
1392 host_os_release_sysext_level
,
1393 host_os_release_confext_level
,
1394 /* required_sysext_scope= */ NULL
,
1395 /* ret_image= */ NULL
);
1396 if (r
== -ENOENT
&& m
->ignore
)
1398 if (r
== -ESTALE
&& host_os_release_id
)
1399 return log_error_errno(r
, // FIXME: this should not be logged ad LOG_ERR, as it will result in duplicate logging.
1400 "Failed to mount image %s, extension-release metadata does not match the lower layer's: ID=%s%s%s%s%s%s%s",
1401 mount_entry_source(m
),
1403 host_os_release_version_id
? " VERSION_ID=" : "",
1404 strempty(host_os_release_version_id
),
1405 host_os_release_sysext_level
? image_class_info
[IMAGE_SYSEXT
].level_env_print
: "",
1406 strempty(host_os_release_sysext_level
),
1407 host_os_release_confext_level
? image_class_info
[IMAGE_CONFEXT
].level_env_print
: "",
1408 strempty(host_os_release_confext_level
));
1410 return log_debug_errno(r
, "Failed to mount image %s on %s: %m", mount_entry_source(m
), mount_entry_path(m
));
1415 static int mount_overlay(const MountEntry
*m
) {
1416 const char *options
;
1421 options
= strjoina("lowerdir=", mount_entry_options(m
));
1423 (void) mkdir_p_label(mount_entry_path(m
), 0755);
1425 r
= mount_nofollow_verbose(LOG_DEBUG
, "overlay", mount_entry_path(m
), "overlay", MS_RDONLY
, options
);
1426 if (r
== -ENOENT
&& m
->ignore
)
1434 static int follow_symlink(
1435 const char *root_directory
,
1438 _cleanup_free_
char *target
= NULL
;
1441 /* Let's chase symlinks, but only one step at a time. That's because depending where the symlink points we
1442 * might need to change the order in which we mount stuff. Hence: let's normalize piecemeal, and do one step at
1443 * a time by specifying CHASE_STEP. This function returns 0 if we resolved one step, and > 0 if we reached the
1444 * end and already have a fully normalized name. */
1446 r
= chase(mount_entry_path(m
), root_directory
, CHASE_STEP
|CHASE_NONEXISTENT
, &target
, NULL
);
1448 return log_debug_errno(r
, "Failed to chase symlinks '%s': %m", mount_entry_path(m
));
1449 if (r
> 0) /* Reached the end, nothing more to resolve */
1452 if (m
->n_followed
>= CHASE_MAX
) /* put a boundary on things */
1453 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
1454 "Symlink loop on '%s'.",
1455 mount_entry_path(m
));
1457 log_debug("Followed mount entry path symlink %s %s %s.",
1458 mount_entry_path(m
), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT
), target
);
1460 mount_entry_consume_prefix(m
, TAKE_PTR(target
));
1467 static int apply_one_mount(
1468 const char *root_directory
,
1470 const NamespaceParameters
*p
) {
1472 _cleanup_free_
char *inaccessible
= NULL
;
1473 bool rbind
= true, make
= false;
1477 /* Return 1 when the mount should be post-processed (remounted r/o, etc.), 0 otherwise. In most
1478 * cases post-processing is the right thing, the typical exception is when the mount is gracefully
1484 log_debug("Applying namespace mount on %s", mount_entry_path(m
));
1488 case MOUNT_INACCESSIBLE
: {
1489 _cleanup_free_
char *runtime_dir
= NULL
;
1492 /* First, get rid of everything that is below if there
1493 * is anything... Then, overmount it with an
1494 * inaccessible path. */
1495 (void) umount_recursive(mount_entry_path(m
), 0);
1497 if (lstat(mount_entry_path(m
), &target
) < 0) {
1498 if (errno
== ENOENT
&& m
->ignore
)
1501 return log_debug_errno(errno
, "Failed to lstat() %s to determine what to mount over it: %m",
1502 mount_entry_path(m
));
1505 /* We don't pass the literal runtime scope through here but one based purely on our UID. This
1506 * means that the root user's --user services will use the host's inaccessible inodes rather
1507 * then root's private ones. This is preferable since it means device nodes that are
1508 * overmounted to make them inaccessible will be overmounted with a device node, rather than
1509 * an AF_UNIX socket inode. */
1510 runtime_dir
= settle_runtime_dir(geteuid() == 0 ? RUNTIME_SCOPE_SYSTEM
: RUNTIME_SCOPE_USER
);
1512 return log_oom_debug();
1514 r
= mode_to_inaccessible_node(runtime_dir
, target
.st_mode
, &inaccessible
);
1516 return log_debug_errno(SYNTHETIC_ERRNO(ELOOP
),
1517 "File type not supported for inaccessible mounts. Note that symlinks are not allowed");
1518 what
= inaccessible
;
1522 case MOUNT_READ_ONLY
:
1523 case MOUNT_READ_WRITE
:
1524 case MOUNT_READ_WRITE_IMPLICIT
:
1527 r
= path_is_mount_point(mount_entry_path(m
), root_directory
, 0);
1528 if (r
== -ENOENT
&& m
->ignore
)
1531 return log_debug_errno(r
, "Failed to determine whether %s is already a mount point: %m",
1532 mount_entry_path(m
));
1533 if (r
> 0) /* Nothing to do here, it is already a mount. We just later toggle the MS_RDONLY
1534 * and MS_NOEXEC bits for the mount point if needed. */
1536 /* This isn't a mount point yet, let's make it one. */
1537 what
= mount_entry_path(m
);
1540 case MOUNT_EXTENSION_DIRECTORY
: {
1541 _cleanup_free_
char *host_os_release_id
= NULL
, *host_os_release_version_id
= NULL
,
1542 *host_os_release_level
= NULL
, *extension_name
= NULL
;
1543 _cleanup_strv_free_
char **extension_release
= NULL
;
1544 ImageClass
class = IMAGE_SYSEXT
;
1546 r
= path_extract_filename(mount_entry_source(m
), &extension_name
);
1548 return log_debug_errno(r
, "Failed to extract extension name from %s: %m", mount_entry_source(m
));
1550 r
= load_extension_release_pairs(mount_entry_source(m
), IMAGE_SYSEXT
, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1552 r
= load_extension_release_pairs(mount_entry_source(m
), IMAGE_CONFEXT
, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1554 class = IMAGE_CONFEXT
;
1557 return log_debug_errno(r
, "Failed to acquire 'extension-release' data of extension tree %s: %m", mount_entry_source(m
));
1559 r
= parse_os_release(
1560 empty_to_root(root_directory
),
1561 "ID", &host_os_release_id
,
1562 "VERSION_ID", &host_os_release_version_id
,
1563 image_class_info
[class].level_env
, &host_os_release_level
,
1566 return log_debug_errno(r
, "Failed to acquire 'os-release' data of OS tree '%s': %m", empty_to_root(root_directory
));
1567 if (isempty(host_os_release_id
))
1568 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
));
1570 r
= load_extension_release_pairs(mount_entry_source(m
), class, extension_name
, /* relax_extension_release_check= */ false, &extension_release
);
1571 if (r
== -ENOENT
&& m
->ignore
)
1574 return log_debug_errno(r
, "Failed to parse directory %s extension-release metadata: %m", extension_name
);
1576 r
= extension_release_validate(
1579 host_os_release_version_id
,
1580 host_os_release_level
,
1581 /* host_extension_scope */ NULL
, /* Leave empty, we need to accept both system and portable */
1585 return log_debug_errno(SYNTHETIC_ERRNO(ESTALE
), "Directory %s extension-release metadata does not match the root's", extension_name
);
1587 return log_debug_errno(r
, "Failed to compare directory %s extension-release metadata with the root's os-release: %m", extension_name
);
1596 case MOUNT_BIND_RECURSIVE
: {
1597 _cleanup_free_
char *chased
= NULL
;
1599 /* Since mount() will always follow symlinks we chase the symlinks on our own first. Note
1600 * that bind mount source paths are always relative to the host root, hence we pass NULL as
1601 * root directory to chase() here. */
1603 r
= chase(mount_entry_source(m
), NULL
, CHASE_TRAIL_SLASH
, &chased
, NULL
);
1604 if (r
== -ENOENT
&& m
->ignore
) {
1605 log_debug_errno(r
, "Path %s does not exist, ignoring.", mount_entry_source(m
));
1609 return log_debug_errno(r
, "Failed to follow symlinks on %s: %m", mount_entry_source(m
));
1611 log_debug("Followed source symlinks %s %s %s.",
1612 mount_entry_source(m
), special_glyph(SPECIAL_GLYPH_ARROW_RIGHT
), chased
);
1614 free_and_replace(m
->source_malloc
, chased
);
1616 what
= mount_entry_source(m
);
1621 case MOUNT_EMPTY_DIR
:
1623 return mount_tmpfs(m
);
1625 case MOUNT_PRIVATE_TMP
:
1626 case MOUNT_PRIVATE_TMP_READ_ONLY
:
1627 what
= mount_entry_source(m
);
1631 case MOUNT_PRIVATE_DEV
:
1632 return mount_private_dev(m
, p
->runtime_scope
);
1634 case MOUNT_BIND_DEV
:
1635 return mount_bind_dev(m
);
1637 case MOUNT_PRIVATE_SYSFS
:
1638 return mount_private_sysfs(m
, p
);
1640 case MOUNT_BIND_SYSFS
:
1641 return mount_bind_sysfs(m
);
1644 return mount_procfs(m
, p
);
1647 return mount_run(m
);
1649 case MOUNT_MQUEUEFS
:
1650 return mount_mqueuefs(m
);
1653 return mount_image(m
, NULL
, p
->mount_image_policy
);
1655 case MOUNT_EXTENSION_IMAGE
:
1656 return mount_image(m
, root_directory
, p
->extension_image_policy
);
1659 return mount_overlay(m
);
1662 assert_not_reached();
1667 r
= mount_nofollow_verbose(LOG_DEBUG
, what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
);
1669 bool try_again
= false;
1671 if (r
== -ENOENT
&& make
) {
1674 /* Hmm, either the source or the destination are missing. Let's see if we can create
1675 the destination, then try again. */
1677 (void) mkdir_parents(mount_entry_path(m
), 0755);
1679 q
= make_mount_point_inode_from_path(what
, mount_entry_path(m
), 0755);
1681 if (q
!= -EEXIST
) // FIXME: this shouldn't be logged at LOG_WARNING, but be bubbled up, and logged there to avoid duplicate logging
1682 log_warning_errno(q
, "Failed to create destination mount point node '%s', ignoring: %m",
1683 mount_entry_path(m
));
1689 r
= mount_nofollow_verbose(LOG_DEBUG
, what
, mount_entry_path(m
), NULL
, MS_BIND
|(rbind
? MS_REC
: 0), NULL
);
1691 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
1694 log_debug("Successfully mounted %s to %s", what
, mount_entry_path(m
));
1698 static int make_read_only(const MountEntry
*m
, char **deny_list
, FILE *proc_self_mountinfo
) {
1699 unsigned long new_flags
= 0, flags_mask
= 0;
1704 assert(proc_self_mountinfo
);
1706 if (m
->state
!= MOUNT_APPLIED
)
1709 if (mount_entry_read_only(m
) || m
->mode
== MOUNT_PRIVATE_DEV
) {
1710 new_flags
|= MS_RDONLY
;
1711 flags_mask
|= MS_RDONLY
;
1715 new_flags
|= MS_NOSUID
;
1716 flags_mask
|= MS_NOSUID
;
1719 if (flags_mask
== 0) /* No Change? */
1722 /* We generally apply these changes recursively, except for /dev, and the cases we know there's
1723 * nothing further down. Set /dev readonly, but not submounts like /dev/shm. Also, we only set the
1724 * per-mount read-only flag. We can't set it on the superblock, if we are inside a user namespace
1725 * and running Linux <= 4.17. */
1727 mount_entry_read_only(m
) &&
1728 !IN_SET(m
->mode
, MOUNT_EMPTY_DIR
, MOUNT_TMPFS
);
1730 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, deny_list
, proc_self_mountinfo
);
1732 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, proc_self_mountinfo
);
1734 /* Note that we only turn on the MS_RDONLY flag here, we never turn it off. Something that was marked
1735 * read-only already stays this way. This improves compatibility with container managers, where we
1736 * won't attempt to undo read-only mounts already applied. */
1738 if (r
== -ENOENT
&& m
->ignore
)
1741 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1742 submounts
? " and its submounts" : "");
1746 static int make_noexec(const MountEntry
*m
, char **deny_list
, FILE *proc_self_mountinfo
) {
1747 unsigned long new_flags
= 0, flags_mask
= 0;
1752 assert(proc_self_mountinfo
);
1754 if (m
->state
!= MOUNT_APPLIED
)
1757 if (mount_entry_noexec(m
)) {
1758 new_flags
|= MS_NOEXEC
;
1759 flags_mask
|= MS_NOEXEC
;
1760 } else if (mount_entry_exec(m
)) {
1761 new_flags
&= ~MS_NOEXEC
;
1762 flags_mask
|= MS_NOEXEC
;
1765 if (flags_mask
== 0) /* No Change? */
1768 submounts
= !IN_SET(m
->mode
, MOUNT_EMPTY_DIR
, MOUNT_TMPFS
);
1771 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, deny_list
, proc_self_mountinfo
);
1773 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), new_flags
, flags_mask
, proc_self_mountinfo
);
1775 if (r
== -ENOENT
&& m
->ignore
)
1778 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1779 submounts
? " and its submounts" : "");
1783 static int make_nosuid(const MountEntry
*m
, FILE *proc_self_mountinfo
) {
1788 assert(proc_self_mountinfo
);
1790 if (m
->state
!= MOUNT_APPLIED
)
1793 submounts
= !IN_SET(m
->mode
, MOUNT_EMPTY_DIR
, MOUNT_TMPFS
);
1795 r
= bind_remount_recursive_with_mountinfo(mount_entry_path(m
), MS_NOSUID
, MS_NOSUID
, NULL
, proc_self_mountinfo
);
1797 r
= bind_remount_one_with_mountinfo(mount_entry_path(m
), MS_NOSUID
, MS_NOSUID
, proc_self_mountinfo
);
1798 if (r
== -ENOENT
&& m
->ignore
)
1801 return log_debug_errno(r
, "Failed to re-mount '%s'%s: %m", mount_entry_path(m
),
1802 submounts
? " and its submounts" : "");
1806 static bool namespace_parameters_mount_apivfs(const NamespaceParameters
*p
) {
1810 * ProtectControlGroups= and ProtectKernelTunables= imply MountAPIVFS=,
1811 * since to protect the API VFS mounts, they need to be around in the
1815 return p
->mount_apivfs
||
1816 p
->protect_control_groups
||
1817 p
->protect_kernel_tunables
||
1818 p
->protect_proc
!= PROTECT_PROC_DEFAULT
||
1819 p
->proc_subset
!= PROC_SUBSET_ALL
;
1822 /* Walk all mount entries and dropping any unused mounts. This affects all
1824 * - that are implicitly protected by a path that has been rendered inaccessible
1825 * - whose immediate parent requests the same protection mode as the mount itself
1826 * - that are outside of the relevant root directory
1827 * - which are duplicates
1829 static void drop_unused_mounts(MountList
*ml
, const char *root_directory
) {
1831 assert(root_directory
);
1833 assert(ml
->mounts
|| ml
->n_mounts
== 0);
1835 typesafe_qsort(ml
->mounts
, ml
->n_mounts
, mount_path_compare
);
1837 drop_duplicates(ml
);
1838 drop_outside_root(ml
, root_directory
);
1839 drop_inaccessible(ml
);
1843 static int create_symlinks_from_tuples(const char *root
, char **strv_symlinks
) {
1846 STRV_FOREACH_PAIR(src
, dst
, strv_symlinks
) {
1847 _cleanup_free_
char *src_abs
= NULL
, *dst_abs
= NULL
;
1849 src_abs
= path_join(root
, *src
);
1850 dst_abs
= path_join(root
, *dst
);
1851 if (!src_abs
|| !dst_abs
)
1854 r
= mkdir_parents_label(dst_abs
, 0755);
1856 return log_debug_errno(
1858 "Failed to create parent directory for symlink '%s': %m",
1861 r
= symlink_idempotent(src_abs
, dst_abs
, true);
1863 return log_debug_errno(
1865 "Failed to create symlink from '%s' to '%s': %m",
1873 static void mount_entry_path_debug_string(const char *root
, MountEntry
*m
, char **error_path
) {
1876 /* Create a string suitable for debugging logs, stripping for example the local working directory.
1877 * For example, with a BindPaths=/var/bar that does not exist on the host:
1880 * foo.service: Failed to set up mount namespacing: /run/systemd/unit-root/var/bar: No such file or directory
1882 * foo.service: Failed to set up mount namespacing: /var/bar: No such file or directory
1884 * Note that this is an error path, so no OOM check is done on purpose. */
1889 if (!mount_entry_path(m
)) {
1895 const char *e
= startswith(mount_entry_path(m
), root
);
1897 *error_path
= strdup(e
);
1902 *error_path
= strdup(mount_entry_path(m
));
1906 static int apply_mounts(
1909 const NamespaceParameters
*p
,
1910 char **error_path
) {
1912 _cleanup_fclose_
FILE *proc_self_mountinfo
= NULL
;
1913 _cleanup_free_
char **deny_list
= NULL
;
1920 if (ml
->n_mounts
== 0) /* Shortcut: nothing to do */
1923 /* Open /proc/self/mountinfo now as it may become unavailable if we mount anything on top of
1924 * /proc. For example, this is the case with the option: 'InaccessiblePaths=/proc'. */
1925 proc_self_mountinfo
= fopen("/proc/self/mountinfo", "re");
1926 if (!proc_self_mountinfo
) {
1930 *error_path
= strdup("/proc/self/mountinfo");
1932 return log_debug_errno(r
, "Failed to open /proc/self/mountinfo: %m");
1935 /* First round, establish all mounts we need */
1939 FOREACH_ARRAY(m
, ml
->mounts
, ml
->n_mounts
) {
1941 if (m
->state
!= MOUNT_PENDING
)
1944 /* ExtensionImages/Directories are first opened in the propagate directory, not in the root_directory */
1945 r
= follow_symlink(!IN_SET(m
->mode
, MOUNT_EXTENSION_IMAGE
, MOUNT_EXTENSION_DIRECTORY
) ? root
: NULL
, m
);
1947 mount_entry_path_debug_string(root
, m
, error_path
);
1951 /* We hit a symlinked mount point. The entry got rewritten and might
1952 * point to a very different place now. Let's normalize the changed
1953 * list, and start from the beginning. After all to mount the entry
1954 * at the new location we might need some other mounts first */
1959 /* Returns 1 if the mount should be post-processed, 0 otherwise */
1960 r
= apply_one_mount(root
, m
, p
);
1962 mount_entry_path_debug_string(root
, m
, error_path
);
1965 m
->state
= r
== 0 ? MOUNT_SKIPPED
: MOUNT_APPLIED
;
1971 drop_unused_mounts(ml
, root
);
1974 /* Now that all filesystems have been set up, but before the
1975 * read-only switches are flipped, create the exec dirs and other symlinks.
1976 * Note that when /var/lib is not empty/tmpfs, these symlinks will already
1977 * exist, which means this will be a no-op. */
1978 r
= create_symlinks_from_tuples(root
, p
->symlinks
);
1980 return log_debug_errno(r
, "Failed to set up symlinks inside mount namespace: %m");
1982 /* Create a deny list we can pass to bind_mount_recursive() */
1983 deny_list
= new(char*, ml
->n_mounts
+1);
1986 for (size_t j
= 0; j
< ml
->n_mounts
; j
++)
1987 deny_list
[j
] = (char*) mount_entry_path(ml
->mounts
+j
);
1988 deny_list
[ml
->n_mounts
] = NULL
;
1990 /* Second round, flip the ro bits if necessary. */
1991 FOREACH_ARRAY(m
, ml
->mounts
, ml
->n_mounts
) {
1992 r
= make_read_only(m
, deny_list
, proc_self_mountinfo
);
1994 mount_entry_path_debug_string(root
, m
, error_path
);
1999 /* Third round, flip the noexec bits with a simplified deny list. */
2000 for (size_t j
= 0; j
< ml
->n_mounts
; j
++)
2001 if (IN_SET((ml
->mounts
+j
)->mode
, MOUNT_EXEC
, MOUNT_NOEXEC
))
2002 deny_list
[j
] = (char*) mount_entry_path(ml
->mounts
+j
);
2003 deny_list
[ml
->n_mounts
] = NULL
;
2005 FOREACH_ARRAY(m
, ml
->mounts
, ml
->n_mounts
) {
2006 r
= make_noexec(m
, deny_list
, proc_self_mountinfo
);
2008 mount_entry_path_debug_string(root
, m
, error_path
);
2013 /* Fourth round, flip the nosuid bits without a deny list. */
2014 if (p
->mount_nosuid
)
2015 FOREACH_ARRAY(m
, ml
->mounts
, ml
->n_mounts
) {
2016 r
= make_nosuid(m
, proc_self_mountinfo
);
2018 mount_entry_path_debug_string(root
, m
, error_path
);
2026 static bool root_read_only(
2027 char **read_only_paths
,
2028 ProtectSystem protect_system
) {
2030 /* Determine whether the root directory is going to be read-only given the configured settings. */
2032 if (protect_system
== PROTECT_SYSTEM_STRICT
)
2035 if (prefixed_path_strv_contains(read_only_paths
, "/"))
2041 static bool home_read_only(
2042 char** read_only_paths
,
2043 char** inaccessible_paths
,
2044 char** empty_directories
,
2045 const BindMount
*bind_mounts
,
2046 size_t n_bind_mounts
,
2047 const TemporaryFileSystem
*temporary_filesystems
,
2048 size_t n_temporary_filesystems
,
2049 ProtectHome protect_home
) {
2051 /* Determine whether the /home directory is going to be read-only given the configured settings. Yes,
2052 * this is a bit sloppy, since we don't bother checking for cases where / is affected by multiple
2055 if (protect_home
!= PROTECT_HOME_NO
)
2058 if (prefixed_path_strv_contains(read_only_paths
, "/home") ||
2059 prefixed_path_strv_contains(inaccessible_paths
, "/home") ||
2060 prefixed_path_strv_contains(empty_directories
, "/home"))
2063 for (size_t i
= 0; i
< n_temporary_filesystems
; i
++)
2064 if (path_equal(temporary_filesystems
[i
].path
, "/home"))
2067 /* If /home is overmounted with some dir from the host it's not writable. */
2068 for (size_t i
= 0; i
< n_bind_mounts
; i
++)
2069 if (path_equal(bind_mounts
[i
].destination
, "/home"))
2075 int setup_namespace(const NamespaceParameters
*p
, char **error_path
) {
2077 _cleanup_(loop_device_unrefp
) LoopDevice
*loop_device
= NULL
;
2078 _cleanup_(dissected_image_unrefp
) DissectedImage
*dissected_image
= NULL
;
2079 _cleanup_strv_free_
char **hierarchies
= NULL
;
2080 _cleanup_(mount_list_done
) MountList ml
= {};
2081 bool require_prefix
= false;
2083 DissectImageFlags dissect_image_flags
=
2084 DISSECT_IMAGE_GENERIC_ROOT
|
2085 DISSECT_IMAGE_REQUIRE_ROOT
|
2086 DISSECT_IMAGE_DISCARD_ON_LOOP
|
2087 DISSECT_IMAGE_RELAX_VAR_CHECK
|
2088 DISSECT_IMAGE_FSCK
|
2089 DISSECT_IMAGE_USR_NO_ROOT
|
2090 DISSECT_IMAGE_GROWFS
|
2091 DISSECT_IMAGE_ADD_PARTITION_DEVICES
|
2092 DISSECT_IMAGE_PIN_PARTITION_DEVICES
;
2097 /* Make sure that all mknod(), mkdir() calls we do are unaffected by the umask, and the access modes
2098 * we configure take effect */
2099 BLOCK_WITH_UMASK(0000);
2101 bool setup_propagate
= !isempty(p
->propagate_dir
) && !isempty(p
->incoming_dir
);
2102 unsigned long mount_propagation_flag
= p
->mount_propagation_flag
!= 0 ? p
->mount_propagation_flag
: MS_SHARED
;
2104 if (p
->root_image
) {
2105 /* Make the whole image read-only if we can determine that we only access it in a read-only fashion. */
2106 if (root_read_only(p
->read_only_paths
,
2107 p
->protect_system
) &&
2108 home_read_only(p
->read_only_paths
, p
->inaccessible_paths
, p
->empty_directories
,
2109 p
->bind_mounts
, p
->n_bind_mounts
, p
->temporary_filesystems
, p
->n_temporary_filesystems
,
2111 strv_isempty(p
->read_write_paths
))
2112 dissect_image_flags
|= DISSECT_IMAGE_READ_ONLY
;
2114 SET_FLAG(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
, p
->verity
&& p
->verity
->data_path
);
2116 r
= loop_device_make_by_path(
2118 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_DEVICE_READ_ONLY
) ? O_RDONLY
: -1 /* < 0 means writable if possible, read-only as fallback */,
2119 /* sector_size= */ UINT32_MAX
,
2120 FLAGS_SET(dissect_image_flags
, DISSECT_IMAGE_NO_PARTITION_TABLE
) ? 0 : LO_FLAGS_PARTSCAN
,
2124 return log_debug_errno(r
, "Failed to create loop device for root image: %m");
2126 r
= dissect_loop_device(
2129 p
->root_image_options
,
2130 p
->root_image_policy
,
2131 dissect_image_flags
,
2134 return log_debug_errno(r
, "Failed to dissect image: %m");
2136 r
= dissected_image_load_verity_sig_partition(
2143 r
= dissected_image_decrypt(
2147 dissect_image_flags
);
2149 return log_debug_errno(r
, "Failed to decrypt dissected image: %m");
2152 if (p
->root_directory
)
2153 root
= p
->root_directory
;
2155 /* /run/systemd should have been created by PID 1 early on already, but in some cases, like
2156 * when running tests (test-execute), it might not have been created yet so let's make sure
2157 * we create it if it doesn't already exist. */
2158 (void) mkdir_p_label("/run/systemd", 0755);
2160 /* Always create the mount namespace in a temporary directory, instead of operating directly
2161 * in the root. The temporary directory prevents any mounts from being potentially obscured
2162 * my other mounts we already applied. We use the same mount point for all images, which is
2163 * safe, since they all live in their own namespaces after all, and hence won't see each
2164 * other. (Note: this directory is also created by PID 1 early on, we create it here for
2165 * similar reasons as /run/systemd/ first.) */
2166 root
= "/run/systemd/mount-rootfs";
2167 (void) mkdir_label(root
, 0555);
2169 require_prefix
= true;
2172 if (p
->n_extension_images
> 0 || !strv_isempty(p
->extension_directories
)) {
2173 /* Hierarchy population needs to be done for sysext and confext extension images */
2174 r
= parse_env_extension_hierarchies(&hierarchies
, "SYSTEMD_SYSEXT_AND_CONFEXT_HIERARCHIES");
2179 r
= append_access_mounts(&ml
, p
->read_write_paths
, MOUNT_READ_WRITE
, require_prefix
);
2183 r
= append_access_mounts(&ml
, p
->read_only_paths
, MOUNT_READ_ONLY
, require_prefix
);
2187 r
= append_access_mounts(&ml
, p
->inaccessible_paths
, MOUNT_INACCESSIBLE
, require_prefix
);
2191 r
= append_access_mounts(&ml
, p
->exec_paths
, MOUNT_EXEC
, require_prefix
);
2195 r
= append_access_mounts(&ml
, p
->no_exec_paths
, MOUNT_NOEXEC
, require_prefix
);
2199 r
= append_empty_dir_mounts(&ml
, p
->empty_directories
);
2203 r
= append_bind_mounts(&ml
, p
->bind_mounts
, p
->n_bind_mounts
);
2207 r
= append_tmpfs_mounts(&ml
, p
->temporary_filesystems
, p
->n_temporary_filesystems
);
2212 bool ro
= streq(p
->tmp_dir
, RUN_SYSTEMD_EMPTY
);
2214 MountEntry
*me
= mount_list_extend(&ml
);
2216 return log_oom_debug();
2218 *me
= (MountEntry
) {
2219 .path_const
= "/tmp",
2220 .mode
= ro
? MOUNT_PRIVATE_TMP_READ_ONLY
: MOUNT_PRIVATE_TMP
,
2221 .source_const
= p
->tmp_dir
,
2225 if (p
->var_tmp_dir
) {
2226 bool ro
= streq(p
->var_tmp_dir
, RUN_SYSTEMD_EMPTY
);
2228 MountEntry
*me
= mount_list_extend(&ml
);
2230 return log_oom_debug();
2232 *me
= (MountEntry
) {
2233 .path_const
= "/var/tmp",
2234 .mode
= ro
? MOUNT_PRIVATE_TMP_READ_ONLY
: MOUNT_PRIVATE_TMP
,
2235 .source_const
= p
->var_tmp_dir
,
2239 r
= append_mount_images(&ml
, p
->mount_images
, p
->n_mount_images
);
2243 r
= append_extensions(&ml
, root
, p
->extension_dir
, hierarchies
, p
->extension_images
, p
->n_extension_images
, p
->extension_directories
);
2247 if (p
->private_dev
) {
2248 MountEntry
*me
= mount_list_extend(&ml
);
2250 return log_oom_debug();
2252 *me
= (MountEntry
) {
2253 .path_const
= "/dev",
2254 .mode
= MOUNT_PRIVATE_DEV
,
2255 .flags
= DEV_MOUNT_OPTIONS
,
2259 /* In case /proc is successfully mounted with pid tree subset only (ProcSubset=pid), the protective
2260 mounts to non-pid /proc paths would fail. But the pid only option may have failed gracefully, so
2261 let's try the mounts but it's not fatal if they don't succeed. */
2262 bool ignore_protect_proc
= p
->ignore_protect_paths
|| p
->proc_subset
== PROC_SUBSET_PID
;
2263 if (p
->protect_kernel_tunables
) {
2264 r
= append_static_mounts(&ml
,
2265 protect_kernel_tunables_proc_table
,
2266 ELEMENTSOF(protect_kernel_tunables_proc_table
),
2267 ignore_protect_proc
);
2271 r
= append_static_mounts(&ml
,
2272 protect_kernel_tunables_sys_table
,
2273 ELEMENTSOF(protect_kernel_tunables_sys_table
),
2274 p
->ignore_protect_paths
);
2279 if (p
->protect_kernel_modules
) {
2280 r
= append_static_mounts(&ml
,
2281 protect_kernel_modules_table
,
2282 ELEMENTSOF(protect_kernel_modules_table
),
2283 p
->ignore_protect_paths
);
2288 if (p
->protect_kernel_logs
) {
2289 r
= append_static_mounts(&ml
,
2290 protect_kernel_logs_proc_table
,
2291 ELEMENTSOF(protect_kernel_logs_proc_table
),
2292 ignore_protect_proc
);
2296 r
= append_static_mounts(&ml
,
2297 protect_kernel_logs_dev_table
,
2298 ELEMENTSOF(protect_kernel_logs_dev_table
),
2299 p
->ignore_protect_paths
);
2304 if (p
->protect_control_groups
) {
2305 MountEntry
*me
= mount_list_extend(&ml
);
2307 return log_oom_debug();
2309 *me
= (MountEntry
) {
2310 .path_const
= "/sys/fs/cgroup",
2311 .mode
= MOUNT_READ_ONLY
,
2315 r
= append_protect_home(&ml
, p
->protect_home
, p
->ignore_protect_paths
);
2319 r
= append_protect_system(&ml
, p
->protect_system
, false);
2323 if (namespace_parameters_mount_apivfs(p
)) {
2324 r
= append_static_mounts(&ml
,
2326 ELEMENTSOF(apivfs_table
),
2327 p
->ignore_protect_paths
);
2332 /* Note, if proc is mounted with subset=pid then neither of the two paths will exist, i.e. they are
2333 * implicitly protected by the mount option. */
2334 if (p
->protect_hostname
) {
2335 r
= append_static_mounts(
2337 protect_hostname_table
,
2338 ELEMENTSOF(protect_hostname_table
),
2339 ignore_protect_proc
);
2344 if (p
->private_network
) {
2345 MountEntry
*me
= mount_list_extend(&ml
);
2347 return log_oom_debug();
2349 *me
= (MountEntry
) {
2350 .path_const
= "/sys",
2351 .mode
= MOUNT_PRIVATE_SYSFS
,
2355 if (p
->private_ipc
) {
2356 MountEntry
*me
= mount_list_extend(&ml
);
2358 return log_oom_debug();
2360 *me
= (MountEntry
) {
2361 .path_const
= "/dev/mqueue",
2362 .mode
= MOUNT_MQUEUEFS
,
2363 .flags
= MS_NOSUID
| MS_NODEV
| MS_NOEXEC
| MS_RELATIME
,
2367 if (p
->creds_path
) {
2368 /* If our service has a credentials store configured, then bind that one in, but hide
2369 * everything else. */
2371 MountEntry
*me
= mount_list_extend(&ml
);
2373 return log_oom_debug();
2375 *me
= (MountEntry
) {
2376 .path_const
= "/run/credentials",
2377 .mode
= MOUNT_TMPFS
,
2379 .options_const
= "mode=0755" TMPFS_LIMITS_EMPTY_OR_ALMOST
,
2380 .flags
= MS_NODEV
|MS_STRICTATIME
|MS_NOSUID
|MS_NOEXEC
,
2383 me
= mount_list_extend(&ml
);
2385 return log_oom_debug();
2387 *me
= (MountEntry
) {
2388 .path_const
= p
->creds_path
,
2391 .source_const
= p
->creds_path
,
2395 /* If our service has no credentials store configured, then make the whole credentials tree
2396 * inaccessible wholesale. */
2398 MountEntry
*me
= mount_list_extend(&ml
);
2400 return log_oom_debug();
2402 *me
= (MountEntry
) {
2403 .path_const
= "/run/credentials",
2404 .mode
= MOUNT_INACCESSIBLE
,
2409 if (p
->log_namespace
) {
2410 _cleanup_free_
char *q
= NULL
;
2412 q
= strjoin("/run/systemd/journal.", p
->log_namespace
);
2414 return log_oom_debug();
2416 MountEntry
*me
= mount_list_extend(&ml
);
2418 return log_oom_debug();
2420 *me
= (MountEntry
) {
2421 .path_const
= "/run/systemd/journal",
2422 .mode
= MOUNT_BIND_RECURSIVE
,
2424 .source_malloc
= TAKE_PTR(q
),
2428 /* Will be used to add bind mounts at runtime */
2429 if (setup_propagate
) {
2430 MountEntry
*me
= mount_list_extend(&ml
);
2432 return log_oom_debug();
2434 *me
= (MountEntry
) {
2435 .source_const
= p
->propagate_dir
,
2436 .path_const
= p
->incoming_dir
,
2442 if (p
->notify_socket
) {
2443 MountEntry
*me
= mount_list_extend(&ml
);
2445 return log_oom_debug();
2447 *me
= (MountEntry
) {
2448 .path_const
= p
->notify_socket
,
2449 .source_const
= p
->notify_socket
,
2455 if (p
->host_os_release_stage
) {
2456 MountEntry
*me
= mount_list_extend(&ml
);
2458 return log_oom_debug();
2460 *me
= (MountEntry
) {
2461 .path_const
= "/run/host/.os-release-stage/",
2462 .source_const
= p
->host_os_release_stage
,
2465 .ignore
= true, /* Live copy, don't hard-fail if it goes missing */
2469 /* Prepend the root directory where that's necessary */
2470 r
= prefix_where_needed(&ml
, root
);
2474 drop_unused_mounts(&ml
, root
);
2476 /* All above is just preparation, figuring out what to do. Let's now actually start doing something. */
2478 if (unshare(CLONE_NEWNS
) < 0) {
2479 r
= log_debug_errno(errno
, "Failed to unshare the mount namespace: %m");
2481 if (ERRNO_IS_PRIVILEGE(r
) ||
2482 ERRNO_IS_NOT_SUPPORTED(r
))
2483 /* If the kernel doesn't support namespaces, or when there's a MAC or seccomp filter
2484 * in place that doesn't allow us to create namespaces (or a missing cap), then
2485 * propagate a recognizable error back, which the caller can use to detect this case
2486 * (and only this) and optionally continue without namespacing applied. */
2492 /* Create the source directory to allow runtime propagation of mounts */
2493 if (setup_propagate
)
2494 (void) mkdir_p(p
->propagate_dir
, 0600);
2496 if (p
->n_extension_images
> 0 || !strv_isempty(p
->extension_directories
))
2497 /* ExtensionImages/Directories mountpoint directories will be created while parsing the
2498 * mounts to create, so have the parent ready */
2499 (void) mkdir_p(p
->extension_dir
, 0600);
2501 /* Remount / as SLAVE so that nothing now mounted in the namespace
2502 * shows up in the parent */
2503 if (mount(NULL
, "/", NULL
, MS_SLAVE
|MS_REC
, NULL
) < 0)
2504 return log_debug_errno(errno
, "Failed to remount '/' as SLAVE: %m");
2506 if (p
->root_image
) {
2507 /* A root image is specified, mount it to the right place */
2508 r
= dissected_image_mount(
2511 /* uid_shift= */ UID_INVALID
,
2512 /* uid_range= */ UID_INVALID
,
2513 /* userns_fd= */ -EBADF
,
2514 dissect_image_flags
);
2516 return log_debug_errno(r
, "Failed to mount root image: %m");
2518 /* Now release the block device lock, so that udevd is free to call BLKRRPART on the device
2520 r
= loop_device_flock(loop_device
, LOCK_UN
);
2522 return log_debug_errno(r
, "Failed to release lock on loopback block device: %m");
2524 r
= dissected_image_relinquish(dissected_image
);
2526 return log_debug_errno(r
, "Failed to relinquish dissected image: %m");
2528 } else if (p
->root_directory
) {
2530 /* A root directory is specified. Turn its directory into bind mount, if it isn't one yet. */
2531 r
= path_is_mount_point(root
, NULL
, AT_SYMLINK_FOLLOW
);
2533 return log_debug_errno(r
, "Failed to detect that %s is a mount point or not: %m", root
);
2535 r
= mount_nofollow_verbose(LOG_DEBUG
, root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2541 /* Let's mount the main root directory to the root directory to use */
2542 r
= mount_nofollow_verbose(LOG_DEBUG
, "/", root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2547 /* Try to set up the new root directory before mounting anything else there. */
2548 if (p
->root_image
|| p
->root_directory
)
2549 (void) base_filesystem_create(root
, UID_INVALID
, GID_INVALID
);
2551 /* Now make the magic happen */
2552 r
= apply_mounts(&ml
, root
, p
, error_path
);
2556 /* MS_MOVE does not work on MS_SHARED so the remount MS_SHARED will be done later */
2557 r
= mount_switch_root(root
, /* mount_propagation_flag = */ 0);
2558 if (r
== -EINVAL
&& p
->root_directory
) {
2559 /* If we are using root_directory and we don't have privileges (ie: user manager in a user
2560 * namespace) and the root_directory is already a mount point in the parent namespace,
2561 * MS_MOVE will fail as we don't have permission to change it (with EINVAL rather than
2562 * EPERM). Attempt to bind-mount it over itself (like we do above if it's not already a
2563 * mount point) and try again. */
2564 r
= mount_nofollow_verbose(LOG_DEBUG
, root
, root
, NULL
, MS_BIND
|MS_REC
, NULL
);
2567 r
= mount_switch_root(root
, /* mount_propagation_flag = */ 0);
2570 return log_debug_errno(r
, "Failed to mount root with MS_MOVE: %m");
2572 /* Remount / as the desired mode. Note that this will not reestablish propagation from our side to
2573 * the host, since what's disconnected is disconnected. */
2574 if (mount(NULL
, "/", NULL
, mount_propagation_flag
| MS_REC
, NULL
) < 0)
2575 return log_debug_errno(errno
, "Failed to remount '/' with desired mount flags: %m");
2577 /* bind_mount_in_namespace() will MS_MOVE into that directory, and that's only supported for
2578 * non-shared mounts. This needs to happen after remounting / or it will fail. */
2579 if (setup_propagate
&& mount(NULL
, p
->incoming_dir
, NULL
, MS_SLAVE
, NULL
) < 0)
2580 return log_debug_errno(errno
, "Failed to remount %s with MS_SLAVE: %m", p
->incoming_dir
);
2585 void bind_mount_free_many(BindMount
*b
, size_t n
) {
2586 assert(b
|| n
== 0);
2588 for (size_t i
= 0; i
< n
; i
++) {
2590 free(b
[i
].destination
);
2596 int bind_mount_add(BindMount
**b
, size_t *n
, const BindMount
*item
) {
2597 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
2604 s
= strdup(item
->source
);
2608 d
= strdup(item
->destination
);
2612 c
= reallocarray(*b
, *n
+ 1, sizeof(BindMount
));
2618 c
[(*n
) ++] = (BindMount
) {
2619 .source
= TAKE_PTR(s
),
2620 .destination
= TAKE_PTR(d
),
2621 .read_only
= item
->read_only
,
2622 .nosuid
= item
->nosuid
,
2623 .recursive
= item
->recursive
,
2624 .ignore_enoent
= item
->ignore_enoent
,
2630 MountImage
* mount_image_free_many(MountImage
*m
, size_t *n
) {
2632 assert(m
|| *n
== 0);
2634 for (size_t i
= 0; i
< *n
; i
++) {
2636 free(m
[i
].destination
);
2637 mount_options_free_all(m
[i
].mount_options
);
2645 int mount_image_add(MountImage
**m
, size_t *n
, const MountImage
*item
) {
2646 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
2647 _cleanup_(mount_options_free_allp
) MountOptions
*options
= NULL
;
2654 s
= strdup(item
->source
);
2658 if (item
->destination
) {
2659 d
= strdup(item
->destination
);
2664 LIST_FOREACH(mount_options
, i
, item
->mount_options
) {
2665 _cleanup_(mount_options_free_allp
) MountOptions
*o
= NULL
;
2667 o
= new(MountOptions
, 1);
2671 *o
= (MountOptions
) {
2672 .partition_designator
= i
->partition_designator
,
2673 .options
= strdup(i
->options
),
2678 LIST_APPEND(mount_options
, options
, TAKE_PTR(o
));
2681 c
= reallocarray(*m
, *n
+ 1, sizeof(MountImage
));
2687 c
[(*n
) ++] = (MountImage
) {
2688 .source
= TAKE_PTR(s
),
2689 .destination
= TAKE_PTR(d
),
2690 .mount_options
= TAKE_PTR(options
),
2691 .ignore_enoent
= item
->ignore_enoent
,
2698 void temporary_filesystem_free_many(TemporaryFileSystem
*t
, size_t n
) {
2699 assert(t
|| n
== 0);
2701 for (size_t i
= 0; i
< n
; i
++) {
2709 int temporary_filesystem_add(
2710 TemporaryFileSystem
**t
,
2713 const char *options
) {
2715 _cleanup_free_
char *p
= NULL
, *o
= NULL
;
2716 TemporaryFileSystem
*c
;
2726 if (!isempty(options
)) {
2727 o
= strdup(options
);
2732 c
= reallocarray(*t
, *n
+ 1, sizeof(TemporaryFileSystem
));
2738 c
[(*n
) ++] = (TemporaryFileSystem
) {
2739 .path
= TAKE_PTR(p
),
2740 .options
= TAKE_PTR(o
),
2746 static int make_tmp_prefix(const char *prefix
) {
2747 _cleanup_free_
char *t
= NULL
;
2748 _cleanup_close_
int fd
= -EBADF
;
2751 /* Don't do anything unless we know the dir is actually missing */
2752 r
= access(prefix
, F_OK
);
2755 if (errno
!= ENOENT
)
2759 r
= mkdir_parents(prefix
, 0755);
2763 r
= tempfn_random(prefix
, NULL
, &t
);
2767 /* umask will corrupt this access mode, but that doesn't matter, we need to call chmod() anyway for
2768 * the suid bit, below. */
2769 fd
= open_mkdir_at(AT_FDCWD
, t
, O_EXCL
|O_CLOEXEC
, 0777);
2773 r
= RET_NERRNO(fchmod(fd
, 01777));
2779 r
= RET_NERRNO(rename(t
, prefix
));
2782 return r
== -EEXIST
? 0 : r
; /* it's fine if someone else created the dir by now */
2789 static int setup_one_tmp_dir(const char *id
, const char *prefix
, char **path
, char **tmp_path
) {
2790 _cleanup_free_
char *x
= NULL
;
2791 _cleanup_free_
char *y
= NULL
;
2800 /* We include the boot id in the directory so that after a
2801 * reboot we can easily identify obsolete directories. */
2803 r
= sd_id128_get_boot(&boot_id
);
2807 x
= strjoin(prefix
, "/systemd-private-", SD_ID128_TO_STRING(boot_id
), "-", id
, "-XXXXXX");
2811 r
= make_tmp_prefix(prefix
);
2817 if (errno
== EROFS
|| ERRNO_IS_DISK_SPACE(errno
))
2824 y
= strjoin(x
, "/tmp");
2829 if (mkdir(y
, 0777 | S_ISVTX
) < 0)
2832 r
= label_fix_full(AT_FDCWD
, y
, prefix
, 0);
2837 *tmp_path
= TAKE_PTR(y
);
2839 /* Trouble: we failed to create the directory. Instead of failing, let's simulate /tmp being
2840 * read-only. This way the service will get the EROFS result as if it was writing to the real
2843 r
= mkdir_p(RUN_SYSTEMD_EMPTY
, 0500);
2847 r
= free_and_strdup(&x
, RUN_SYSTEMD_EMPTY
);
2852 *path
= TAKE_PTR(x
);
2856 int setup_tmp_dirs(const char *id
, char **tmp_dir
, char **var_tmp_dir
) {
2857 _cleanup_(namespace_cleanup_tmpdirp
) char *a
= NULL
;
2858 _cleanup_(rmdir_and_freep
) char *a_tmp
= NULL
;
2864 assert(var_tmp_dir
);
2866 r
= setup_one_tmp_dir(id
, "/tmp", &a
, &a_tmp
);
2870 r
= setup_one_tmp_dir(id
, "/var/tmp", &b
, NULL
);
2874 a_tmp
= mfree(a_tmp
); /* avoid rmdir */
2875 *tmp_dir
= TAKE_PTR(a
);
2876 *var_tmp_dir
= TAKE_PTR(b
);
2881 int setup_shareable_ns(int ns_storage_socket
[static 2], unsigned long nsflag
) {
2882 _cleanup_close_
int ns
= -EBADF
;
2884 const char *ns_name
, *ns_path
;
2886 assert(ns_storage_socket
);
2887 assert(ns_storage_socket
[0] >= 0);
2888 assert(ns_storage_socket
[1] >= 0);
2890 ns_name
= namespace_single_flag_to_string(nsflag
);
2893 /* We use the passed socketpair as a storage buffer for our
2894 * namespace reference fd. Whatever process runs this first
2895 * shall create a new namespace, all others should just join
2896 * it. To serialize that we use a file lock on the socket
2899 * It's a bit crazy, but hey, works great! */
2901 r
= posix_lock(ns_storage_socket
[0], LOCK_EX
);
2905 CLEANUP_POSIX_UNLOCK(ns_storage_socket
[0]);
2907 ns
= receive_one_fd(ns_storage_socket
[0], MSG_PEEK
|MSG_DONTWAIT
);
2909 /* Yay, found something, so let's join the namespace */
2910 r
= RET_NERRNO(setns(ns
, nsflag
));
2920 /* Nothing stored yet, so let's create a new namespace. */
2922 if (unshare(nsflag
) < 0)
2925 (void) loopback_setup();
2927 ns_path
= strjoina("/proc/self/ns/", ns_name
);
2928 ns
= open(ns_path
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
);
2932 r
= send_one_fd(ns_storage_socket
[1], ns
, MSG_DONTWAIT
);
2939 int open_shareable_ns_path(int ns_storage_socket
[static 2], const char *path
, unsigned long nsflag
) {
2940 _cleanup_close_
int ns
= -EBADF
;
2943 assert(ns_storage_socket
);
2944 assert(ns_storage_socket
[0] >= 0);
2945 assert(ns_storage_socket
[1] >= 0);
2948 /* If the storage socket doesn't contain a ns fd yet, open one via the file system and store it in
2949 * it. This is supposed to be called ahead of time, i.e. before setup_shareable_ns() which will
2950 * allocate a new anonymous ns if needed. */
2952 r
= posix_lock(ns_storage_socket
[0], LOCK_EX
);
2956 CLEANUP_POSIX_UNLOCK(ns_storage_socket
[0]);
2958 ns
= receive_one_fd(ns_storage_socket
[0], MSG_PEEK
|MSG_DONTWAIT
);
2964 /* Nothing stored yet. Open the file from the file system. */
2966 ns
= open(path
, O_RDONLY
|O_NOCTTY
|O_CLOEXEC
);
2970 r
= fd_is_ns(ns
, nsflag
);
2973 if (r
< 0 && r
!= -EUCLEAN
) /* EUCLEAN: we don't know */
2976 r
= send_one_fd(ns_storage_socket
[1], ns
, MSG_DONTWAIT
);
2983 bool ns_type_supported(NamespaceType type
) {
2984 const char *t
, *ns_proc
;
2986 t
= namespace_type_to_string(type
);
2987 if (!t
) /* Don't know how to translate this? Then it's not supported */
2990 ns_proc
= strjoina("/proc/self/ns/", t
);
2991 return access(ns_proc
, F_OK
) == 0;
2994 static const char *const protect_home_table
[_PROTECT_HOME_MAX
] = {
2995 [PROTECT_HOME_NO
] = "no",
2996 [PROTECT_HOME_YES
] = "yes",
2997 [PROTECT_HOME_READ_ONLY
] = "read-only",
2998 [PROTECT_HOME_TMPFS
] = "tmpfs",
3001 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_home
, ProtectHome
, PROTECT_HOME_YES
);
3003 static const char *const protect_system_table
[_PROTECT_SYSTEM_MAX
] = {
3004 [PROTECT_SYSTEM_NO
] = "no",
3005 [PROTECT_SYSTEM_YES
] = "yes",
3006 [PROTECT_SYSTEM_FULL
] = "full",
3007 [PROTECT_SYSTEM_STRICT
] = "strict",
3010 DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(protect_system
, ProtectSystem
, PROTECT_SYSTEM_YES
);
3012 static const char* const namespace_type_table
[] = {
3013 [NAMESPACE_MOUNT
] = "mnt",
3014 [NAMESPACE_CGROUP
] = "cgroup",
3015 [NAMESPACE_UTS
] = "uts",
3016 [NAMESPACE_IPC
] = "ipc",
3017 [NAMESPACE_USER
] = "user",
3018 [NAMESPACE_PID
] = "pid",
3019 [NAMESPACE_NET
] = "net",
3020 [NAMESPACE_TIME
] = "time",
3023 DEFINE_STRING_TABLE_LOOKUP(namespace_type
, NamespaceType
);
3025 static const char* const protect_proc_table
[_PROTECT_PROC_MAX
] = {
3026 [PROTECT_PROC_DEFAULT
] = "default",
3027 [PROTECT_PROC_NOACCESS
] = "noaccess",
3028 [PROTECT_PROC_INVISIBLE
] = "invisible",
3029 [PROTECT_PROC_PTRACEABLE
] = "ptraceable",
3032 DEFINE_STRING_TABLE_LOOKUP(protect_proc
, ProtectProc
);
3034 static const char* const proc_subset_table
[_PROC_SUBSET_MAX
] = {
3035 [PROC_SUBSET_ALL
] = "all",
3036 [PROC_SUBSET_PID
] = "pid",
3039 DEFINE_STRING_TABLE_LOOKUP(proc_subset
, ProcSubset
);