1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
6 #include <sys/statvfs.h>
9 #include "alloc-util.h"
11 #include "cgroup-setup.h"
12 #include "cgroup-util.h"
13 #include "conf-files.h"
14 #include "dev-setup.h"
15 #include "dirent-util.h"
16 #include "efi-loader.h"
23 #include "mkdir-label.h"
24 #include "mount-setup.h"
25 #include "mount-util.h"
26 #include "mountpoint-util.h"
27 #include "nulstr-util.h"
28 #include "path-util.h"
29 #include "recurse-dir.h"
31 #include "smack-util.h"
33 #include "user-util.h"
36 typedef enum MountMode
{
39 MNT_IN_CONTAINER
= 1 << 1,
40 MNT_CHECK_WRITABLE
= 1 << 2,
41 MNT_FOLLOW_SYMLINK
= 1 << 3,
44 typedef struct MountPoint
{
50 bool (*condition_fn
)(void);
54 /* The first three entries we might need before SELinux is up. The
55 * fourth (securityfs) is needed by IMA to load a custom policy. The
56 * other ones we can delay until SELinux and IMA are loaded. When
57 * SMACK is enabled we need smackfs, too, so it's a fifth one. */
59 #define N_EARLY_MOUNT 5
61 #define N_EARLY_MOUNT 4
64 static const MountPoint mount_table
[] = {
65 { "proc", "/proc", "proc", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
66 NULL
, MNT_FATAL
|MNT_IN_CONTAINER
|MNT_FOLLOW_SYMLINK
},
67 { "sysfs", "/sys", "sysfs", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
68 NULL
, MNT_FATAL
|MNT_IN_CONTAINER
},
69 { "devtmpfs", "/dev", "devtmpfs", "mode=0755" TMPFS_LIMITS_DEV
, MS_NOSUID
|MS_STRICTATIME
,
70 NULL
, MNT_FATAL
|MNT_IN_CONTAINER
},
71 { "securityfs", "/sys/kernel/security", "securityfs", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
74 { "smackfs", "/sys/fs/smackfs", "smackfs", "smackfsdef=*", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
75 mac_smack_use
, MNT_FATAL
},
76 { "tmpfs", "/dev/shm", "tmpfs", "mode=01777,smackfsroot=*", MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
,
77 mac_smack_use
, MNT_FATAL
},
79 { "tmpfs", "/dev/shm", "tmpfs", "mode=01777", MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
,
80 NULL
, MNT_FATAL
|MNT_IN_CONTAINER
},
81 { "devpts", "/dev/pts", "devpts", "mode=0620,gid=" STRINGIFY(TTY_GID
), MS_NOSUID
|MS_NOEXEC
,
82 NULL
, MNT_IN_CONTAINER
},
84 { "tmpfs", "/run", "tmpfs", "mode=0755,smackfsroot=*" TMPFS_LIMITS_RUN
, MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
,
85 mac_smack_use
, MNT_FATAL
},
87 { "tmpfs", "/run", "tmpfs", "mode=0755" TMPFS_LIMITS_RUN
, MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
,
88 NULL
, MNT_FATAL
|MNT_IN_CONTAINER
},
89 { "cgroup2", "/sys/fs/cgroup", "cgroup2", "nsdelegate,memory_recursiveprot", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
90 cg_is_unified_wanted
, MNT_IN_CONTAINER
|MNT_CHECK_WRITABLE
},
91 { "cgroup2", "/sys/fs/cgroup", "cgroup2", "nsdelegate", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
92 cg_is_unified_wanted
, MNT_IN_CONTAINER
|MNT_CHECK_WRITABLE
},
93 { "cgroup2", "/sys/fs/cgroup", "cgroup2", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
94 cg_is_unified_wanted
, MNT_IN_CONTAINER
|MNT_CHECK_WRITABLE
},
95 { "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=0755" TMPFS_LIMITS_SYS_FS_CGROUP
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
,
96 cg_is_legacy_wanted
, MNT_FATAL
|MNT_IN_CONTAINER
},
97 { "cgroup2", "/sys/fs/cgroup/unified", "cgroup2", "nsdelegate", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
98 cg_is_hybrid_wanted
, MNT_IN_CONTAINER
|MNT_CHECK_WRITABLE
},
99 { "cgroup2", "/sys/fs/cgroup/unified", "cgroup2", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
100 cg_is_hybrid_wanted
, MNT_IN_CONTAINER
|MNT_CHECK_WRITABLE
},
101 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd,xattr", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
102 cg_is_legacy_wanted
, MNT_IN_CONTAINER
},
103 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
104 cg_is_legacy_wanted
, MNT_FATAL
|MNT_IN_CONTAINER
},
106 { "pstore", "/sys/fs/pstore", "pstore", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
110 { "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
111 is_efi_boot
, MNT_NONE
},
113 { "bpf", "/sys/fs/bpf", "bpf", "mode=0700", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
117 bool mount_point_is_api(const char *path
) {
120 /* Checks if this mount point is considered "API", and hence
121 * should be ignored */
123 for (i
= 0; i
< ELEMENTSOF(mount_table
); i
++)
124 if (path_equal(path
, mount_table
[i
].where
))
127 return path_startswith(path
, "/sys/fs/cgroup/");
130 bool mount_point_ignore(const char *path
) {
131 /* These are API file systems that might be mounted by other software, we just list them here so that
132 * we know that we should ignore them. */
134 /* SELinux file systems */
136 /* Container bind mounts */
140 "/proc/sys/kernel/random/boot_id")
141 if (path_equal(path
, i
))
144 if (path_startswith(path
, "/run/host")) /* All mounts passed in from the container manager are
145 * something we better ignore. */
151 static int mount_one(const MountPoint
*p
, bool relabel
) {
156 priority
= (p
->mode
& MNT_FATAL
) ? LOG_ERR
: LOG_DEBUG
;
158 if (p
->condition_fn
&& !p
->condition_fn())
161 /* Relabel first, just in case */
163 (void) label_fix(p
->where
, LABEL_IGNORE_ENOENT
|LABEL_IGNORE_EROFS
);
165 r
= path_is_mount_point(p
->where
, NULL
, AT_SYMLINK_FOLLOW
);
166 if (r
< 0 && r
!= -ENOENT
) {
167 log_full_errno(priority
, r
, "Failed to determine whether %s is a mount point: %m", p
->where
);
168 return (p
->mode
& MNT_FATAL
) ? r
: 0;
173 /* Skip securityfs in a container */
174 if (!(p
->mode
& MNT_IN_CONTAINER
) && detect_container() > 0)
177 /* The access mode here doesn't really matter too much, since
178 * the mounted file system will take precedence anyway. */
180 (void) mkdir_p_label(p
->where
, 0755);
182 (void) mkdir_p(p
->where
, 0755);
184 log_debug("Mounting %s to %s of type %s with options %s.",
190 if (FLAGS_SET(p
->mode
, MNT_FOLLOW_SYMLINK
))
191 r
= RET_NERRNO(mount(p
->what
, p
->where
, p
->type
, p
->flags
, p
->options
));
193 r
= mount_nofollow(p
->what
, p
->where
, p
->type
, p
->flags
, p
->options
);
195 log_full_errno(priority
, r
, "Failed to mount %s at %s: %m", p
->type
, p
->where
);
196 return (p
->mode
& MNT_FATAL
) ? r
: 0;
199 /* Relabel again, since we now mounted something fresh here */
201 (void) label_fix(p
->where
, 0);
203 if (p
->mode
& MNT_CHECK_WRITABLE
) {
204 if (access(p
->where
, W_OK
) < 0) {
207 (void) umount2(p
->where
, UMOUNT_NOFOLLOW
);
208 (void) rmdir(p
->where
);
210 log_full_errno(priority
, r
, "Mount point %s not writable after mounting: %m", p
->where
);
211 return (p
->mode
& MNT_FATAL
) ? r
: 0;
218 static int mount_points_setup(unsigned n
, bool loaded_policy
) {
222 for (i
= 0; i
< n
; i
++) {
225 j
= mount_one(mount_table
+ i
, loaded_policy
);
226 if (j
!= 0 && r
>= 0)
233 int mount_setup_early(void) {
234 assert_cc(N_EARLY_MOUNT
<= ELEMENTSOF(mount_table
));
236 /* Do a minimal mount of /proc and friends to enable the most
237 * basic stuff, such as SELinux */
238 return mount_points_setup(N_EARLY_MOUNT
, false);
241 static const char *join_with(const char *controller
) {
243 static const char* const pairs
[] = {
245 "net_cls", "net_prio",
251 /* This will lookup which controller to mount another controller with. Input is a controller name, and output
252 * is the other controller name. The function works both ways: you can input one and get the other, and input
253 * the other to get the one. */
255 STRV_FOREACH_PAIR(x
, y
, pairs
) {
256 if (streq(controller
, *x
))
258 if (streq(controller
, *y
))
265 static int symlink_controller(const char *target
, const char *alias
) {
272 a
= strjoina("/sys/fs/cgroup/", alias
);
274 r
= symlink_idempotent(target
, a
, false);
276 return log_error_errno(r
, "Failed to create symlink %s: %m", a
);
278 #if HAVE_SMACK_RUN_LABEL
281 p
= strjoina("/sys/fs/cgroup/", target
);
283 r
= mac_smack_copy(a
, p
);
284 if (r
< 0 && r
!= -EOPNOTSUPP
)
285 return log_error_errno(r
, "Failed to copy smack label from %s to %s: %m", p
, a
);
291 int mount_cgroup_controllers(void) {
292 _cleanup_set_free_ Set
*controllers
= NULL
;
295 if (!cg_is_legacy_wanted())
298 /* Mount all available cgroup controllers that are built into the kernel. */
299 r
= cg_kernel_controllers(&controllers
);
301 return log_error_errno(r
, "Failed to enumerate cgroup controllers: %m");
304 _cleanup_free_
char *options
= NULL
, *controller
= NULL
, *where
= NULL
;
305 const char *other_controller
;
309 .flags
= MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
310 .mode
= MNT_IN_CONTAINER
,
313 controller
= set_steal_first(controllers
);
317 /* Check if we shall mount this together with another controller */
318 other_controller
= join_with(controller
);
319 if (other_controller
) {
320 _cleanup_free_
char *c
= NULL
;
322 /* Check if the other controller is actually available in the kernel too */
323 c
= set_remove(controllers
, other_controller
);
326 /* Join the two controllers into one string, and maintain a stable ordering */
327 if (strcmp(controller
, other_controller
) < 0)
328 options
= strjoin(controller
, ",", other_controller
);
330 options
= strjoin(other_controller
, ",", controller
);
336 /* The simple case, where there's only one controller to mount together */
338 options
= TAKE_PTR(controller
);
340 where
= path_join("/sys/fs/cgroup", options
);
347 r
= mount_one(&p
, true);
351 /* Create symlinks from the individual controller names, in case we have a joined mount */
353 (void) symlink_controller(options
, controller
);
354 if (other_controller
)
355 (void) symlink_controller(options
, other_controller
);
358 /* Now that we mounted everything, let's make the tmpfs the cgroup file systems are mounted into read-only. */
359 (void) mount_nofollow("tmpfs", "/sys/fs/cgroup", "tmpfs",
360 MS_REMOUNT
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
|MS_RDONLY
,
361 "mode=0755" TMPFS_LIMITS_SYS_FS_CGROUP
);
366 #if HAVE_SELINUX || ENABLE_SMACK
367 static int relabel_cb(
368 RecurseDirEvent event
,
372 const struct dirent
*de
,
373 const struct statx
*sx
,
378 case RECURSE_DIR_LEAVE
:
379 case RECURSE_DIR_SKIP_MOUNT
:
380 /* If we already saw this dirent when entering it or this is a dirent that on a different
381 * mount, don't relabel it. */
382 return RECURSE_DIR_CONTINUE
;
384 case RECURSE_DIR_ENTER
:
385 /* /run/initramfs is static data and big, no need to dynamically relabel its contents at boot... */
386 if (path_equal(path
, "/run/initramfs"))
387 return RECURSE_DIR_SKIP_ENTRY
;
392 /* Otherwise, label it, even if we had trouble stat()ing it and similar. SELinux can figure this out */
393 (void) label_fix(path
, 0);
394 return RECURSE_DIR_CONTINUE
;
398 static int relabel_tree(const char *path
) {
401 r
= recurse_dir_at(AT_FDCWD
, path
, 0, UINT_MAX
, RECURSE_DIR_ENSURE_TYPE
|RECURSE_DIR_SAME_MOUNT
, relabel_cb
, NULL
);
403 log_debug_errno(r
, "Failed to recursively relabel '%s': %m", path
);
408 static int relabel_cgroup_filesystems(void) {
412 r
= cg_all_unified();
414 /* Temporarily remount the root cgroup filesystem to give it a proper label. Do this
415 only when the filesystem has been already populated by a previous instance of systemd
416 running from initrd. Otherwise don't remount anything and leave the filesystem read-write
417 for the cgroup filesystems to be mounted inside. */
418 if (statfs("/sys/fs/cgroup", &st
) < 0)
419 return log_error_errno(errno
, "Failed to determine mount flags for /sys/fs/cgroup: %m");
421 if (st
.f_flags
& ST_RDONLY
)
422 (void) mount_nofollow(NULL
, "/sys/fs/cgroup", NULL
, MS_REMOUNT
, NULL
);
424 (void) label_fix("/sys/fs/cgroup", 0);
425 (void) relabel_tree("/sys/fs/cgroup");
427 if (st
.f_flags
& ST_RDONLY
)
428 (void) mount_nofollow(NULL
, "/sys/fs/cgroup", NULL
, MS_REMOUNT
|MS_RDONLY
, NULL
);
431 return log_error_errno(r
, "Failed to determine whether we are in all unified mode: %m");
436 static int relabel_extra(void) {
437 _cleanup_strv_free_
char **files
= NULL
;
440 /* Support for relabelling additional files or directories after loading the policy. For this, code in the
441 * initrd simply has to drop in *.relabel files into /run/systemd/relabel-extra.d/. We'll read all such files
442 * expecting one absolute path by line and will relabel each (and everyone below that in case the path refers
443 * to a directory). These drop-in files are supposed to be absolutely minimal, and do not understand comments
444 * and such. After the operation succeeded the files are removed, and the drop-in directory as well, if
448 r
= conf_files_list(&files
, ".relabel", NULL
,
449 CONF_FILES_FILTER_MASKED
| CONF_FILES_REGULAR
,
450 "/run/systemd/relabel-extra.d/");
452 return log_error_errno(r
, "Failed to enumerate /run/systemd/relabel-extra.d/, ignoring: %m");
454 STRV_FOREACH(file
, files
) {
455 _cleanup_fclose_
FILE *f
= NULL
;
457 f
= fopen(*file
, "re");
459 log_warning_errno(errno
, "Failed to open %s, ignoring: %m", *file
);
464 _cleanup_free_
char *line
= NULL
;
466 r
= read_line(f
, LONG_LINE_MAX
, &line
);
468 log_warning_errno(r
, "Failed to read %s, ignoring: %m", *file
);
471 if (r
== 0) /* EOF */
476 if (!path_is_normalized(line
)) {
477 log_warning("Path to relabel is not normalized, ignoring: %s", line
);
481 if (!path_is_absolute(line
)) {
482 log_warning("Path to relabel is not absolute, ignoring: %s", line
);
486 log_debug("Relabelling additional file/directory '%s'.", line
);
487 (void) label_fix(line
, 0);
488 (void) relabel_tree(line
);
492 if (unlink(*file
) < 0)
493 log_warning_errno(errno
, "Failed to remove %s, ignoring: %m", *file
);
496 /* Remove when we complete things. */
497 if (rmdir("/run/systemd/relabel-extra.d") < 0 &&
499 log_warning_errno(errno
, "Failed to remove /run/systemd/relabel-extra.d/ directory: %m");
505 int mount_setup(bool loaded_policy
, bool leave_propagation
) {
508 r
= mount_points_setup(ELEMENTSOF(mount_table
), loaded_policy
);
512 #if HAVE_SELINUX || ENABLE_SMACK
513 /* Nodes in devtmpfs and /run need to be manually updated for
514 * the appropriate labels, after mounting. The other virtual
515 * API file systems like /sys and /proc do not need that, they
516 * use the same label for all their files. */
518 usec_t before_relabel
, after_relabel
;
521 before_relabel
= now(CLOCK_MONOTONIC
);
523 FOREACH_STRING(i
, "/dev", "/dev/shm", "/run")
524 (void) relabel_tree(i
);
526 (void) relabel_cgroup_filesystems();
528 n_extra
= relabel_extra();
530 after_relabel
= now(CLOCK_MONOTONIC
);
532 log_info("Relabelled /dev, /dev/shm, /run, /sys/fs/cgroup%s in %s.",
533 n_extra
> 0 ? ", additional files" : "",
534 FORMAT_TIMESPAN(after_relabel
- before_relabel
, 0));
538 /* Create a few default symlinks, which are normally created
539 * by udevd, but some scripts might need them before we start
541 dev_setup(NULL
, UID_INVALID
, GID_INVALID
);
543 /* Mark the root directory as shared in regards to mount propagation. The kernel defaults to "private", but we
544 * think it makes more sense to have a default of "shared" so that nspawn and the container tools work out of
545 * the box. If specific setups need other settings they can reset the propagation mode to private if
546 * needed. Note that we set this only when we are invoked directly by the kernel. If we are invoked by a
547 * container manager we assume the container manager knows what it is doing (for example, because it set up
548 * some directories with different propagation modes). */
549 if (detect_container() <= 0 && !leave_propagation
)
550 if (mount(NULL
, "/", NULL
, MS_REC
|MS_SHARED
, NULL
) < 0)
551 log_warning_errno(errno
, "Failed to set up the root directory for shared mount propagation: %m");
553 /* Create a few directories we always want around, Note that sd_booted() checks for /run/systemd/system, so
554 * this mkdir really needs to stay for good, otherwise software that copied sd-daemon.c into their sources will
555 * misdetect systemd. */
556 (void) mkdir_label("/run/systemd", 0755);
557 (void) mkdir_label("/run/systemd/system", 0755);
559 /* Make sure we have a mount point to hide in sandboxes */
560 (void) mkdir_label("/run/credentials", 0755);
562 /* Also create /run/systemd/inaccessible nodes, so that we always have something to mount
563 * inaccessible nodes from. If we run in a container the host might have created these for us already
564 * in /run/host/inaccessible/. Use those if we can, since that way we likely get access to block/char
565 * device nodes that are inaccessible, and if userns is used to nodes that are on mounts owned by a
566 * userns outside the container and thus nicely read-only and not remountable. */
567 if (access("/run/host/inaccessible/", F_OK
) < 0) {
569 log_debug_errno(errno
, "Failed to check if /run/host/inaccessible exists, ignoring: %m");
571 (void) make_inaccessible_nodes("/run/systemd", UID_INVALID
, GID_INVALID
);
573 (void) symlink("../host/inaccessible", "/run/systemd/inaccessible");