1 /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
4 This file is part of systemd.
6 Copyright 2010 Lennart Poettering
8 systemd is free software; you can redistribute it and/or modify it
9 under the terms of the GNU Lesser General Public License as published by
10 the Free Software Foundation; either version 2.1 of the License, or
11 (at your option) any later version.
13 systemd is distributed in the hope that it will be useful, but
14 WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 Lesser General Public License for more details.
18 You should have received a copy of the GNU Lesser General Public License
19 along with systemd; If not, see <http://www.gnu.org/licenses/>.
22 #include <sys/mount.h>
28 #include "alloc-util.h"
30 #include "cgroup-util.h"
31 #include "dev-setup.h"
38 #include "mount-setup.h"
39 #include "mount-util.h"
40 #include "path-util.h"
42 #include "smack-util.h"
44 #include "user-util.h"
48 typedef enum MountMode
{
51 MNT_IN_CONTAINER
= 1 << 1,
54 typedef struct MountPoint
{
60 bool (*condition_fn
)(void);
64 /* The first three entries we might need before SELinux is up. The
65 * fourth (securityfs) is needed by IMA to load a custom policy. The
66 * other ones we can delay until SELinux and IMA are loaded. When
67 * SMACK is enabled we need smackfs, too, so it's a fifth one. */
69 #define N_EARLY_MOUNT 5
71 #define N_EARLY_MOUNT 4
74 static const MountPoint mount_table
[] = {
75 { "sysfs", "/sys", "sysfs", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
76 NULL
, MNT_FATAL
|MNT_IN_CONTAINER
},
77 { "proc", "/proc", "proc", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
78 NULL
, MNT_FATAL
|MNT_IN_CONTAINER
},
79 { "devtmpfs", "/dev", "devtmpfs", "mode=755", MS_NOSUID
|MS_STRICTATIME
,
80 NULL
, MNT_FATAL
|MNT_IN_CONTAINER
},
81 { "securityfs", "/sys/kernel/security", "securityfs", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
84 { "smackfs", "/sys/fs/smackfs", "smackfs", "smackfsdef=*", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
85 mac_smack_use
, MNT_FATAL
},
86 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777,smackfsroot=*", MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
,
87 mac_smack_use
, MNT_FATAL
},
89 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
,
90 NULL
, MNT_FATAL
|MNT_IN_CONTAINER
},
91 { "devpts", "/dev/pts", "devpts", "mode=620,gid=" STRINGIFY(TTY_GID
), MS_NOSUID
|MS_NOEXEC
,
92 NULL
, MNT_IN_CONTAINER
},
94 { "tmpfs", "/run", "tmpfs", "mode=755,smackfsroot=*", MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
,
95 mac_smack_use
, MNT_FATAL
},
97 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
,
98 NULL
, MNT_FATAL
|MNT_IN_CONTAINER
},
99 { "cgroup", "/sys/fs/cgroup", "cgroup", "__DEVEL__sane_behavior", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
100 cg_is_unified_wanted
, MNT_FATAL
|MNT_IN_CONTAINER
},
101 { "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
,
102 cg_is_legacy_wanted
, MNT_FATAL
|MNT_IN_CONTAINER
},
103 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd,xattr", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
104 cg_is_legacy_wanted
, MNT_IN_CONTAINER
},
105 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
106 cg_is_legacy_wanted
, MNT_FATAL
|MNT_IN_CONTAINER
},
107 { "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 { "kdbusfs", "/sys/fs/kdbus", "kdbusfs", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
114 is_kdbus_wanted
, MNT_IN_CONTAINER
},
117 /* These are API file systems that might be mounted by other software,
118 * we just list them here so that we know that we should ignore them */
120 static const char ignore_paths
[] =
121 /* SELinux file systems */
123 /* Container bind mounts */
128 bool mount_point_is_api(const char *path
) {
131 /* Checks if this mount point is considered "API", and hence
132 * should be ignored */
134 for (i
= 0; i
< ELEMENTSOF(mount_table
); i
++)
135 if (path_equal(path
, mount_table
[i
].where
))
138 return path_startswith(path
, "/sys/fs/cgroup/");
141 bool mount_point_ignore(const char *path
) {
144 NULSTR_FOREACH(i
, ignore_paths
)
145 if (path_equal(path
, i
))
151 static int mount_one(const MountPoint
*p
, bool relabel
) {
156 if (p
->condition_fn
&& !p
->condition_fn())
159 /* Relabel first, just in case */
161 label_fix(p
->where
, true, true);
163 r
= path_is_mount_point(p
->where
, AT_SYMLINK_FOLLOW
);
164 if (r
< 0 && r
!= -ENOENT
)
169 /* Skip securityfs in a container */
170 if (!(p
->mode
& MNT_IN_CONTAINER
) && detect_container() > 0)
173 /* The access mode here doesn't really matter too much, since
174 * the mounted file system will take precedence anyway. */
176 mkdir_p_label(p
->where
, 0755);
178 mkdir_p(p
->where
, 0755);
180 log_debug("Mounting %s to %s of type %s with options %s.",
191 log_full((p
->mode
& MNT_FATAL
) ? LOG_ERR
: LOG_DEBUG
, "Failed to mount %s at %s: %m", p
->type
, p
->where
);
192 return (p
->mode
& MNT_FATAL
) ? -errno
: 0;
195 /* Relabel again, since we now mounted something fresh here */
197 label_fix(p
->where
, false, false);
202 int mount_setup_early(void) {
206 assert_cc(N_EARLY_MOUNT
<= ELEMENTSOF(mount_table
));
208 /* Do a minimal mount of /proc and friends to enable the most
209 * basic stuff, such as SELinux */
210 for (i
= 0; i
< N_EARLY_MOUNT
; i
++) {
213 j
= mount_one(mount_table
+ i
, false);
214 if (j
!= 0 && r
>= 0)
221 int mount_cgroup_controllers(char ***join_controllers
) {
222 _cleanup_set_free_free_ Set
*controllers
= NULL
;
225 if (!cg_is_legacy_wanted())
228 /* Mount all available cgroup controllers that are built into the kernel. */
230 controllers
= set_new(&string_hash_ops
);
234 r
= cg_kernel_controllers(controllers
);
236 return log_error_errno(r
, "Failed to enumerate cgroup controllers: %m");
239 _cleanup_free_
char *options
= NULL
, *controller
= NULL
, *where
= NULL
;
243 .flags
= MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
244 .mode
= MNT_IN_CONTAINER
,
248 controller
= set_steal_first(controllers
);
252 if (join_controllers
)
253 for (k
= join_controllers
; *k
; k
++)
254 if (strv_find(*k
, controller
))
260 for (i
= *k
, j
= *k
; *i
; i
++) {
262 if (!streq(*i
, controller
)) {
263 _cleanup_free_
char *t
;
265 t
= set_remove(controllers
, *i
);
277 options
= strv_join(*k
, ",");
281 options
= controller
;
285 where
= strappend("/sys/fs/cgroup/", options
);
292 r
= mount_one(&p
, true);
296 if (r
> 0 && k
&& *k
) {
299 for (i
= *k
; *i
; i
++) {
300 _cleanup_free_
char *t
= NULL
;
302 t
= strappend("/sys/fs/cgroup/", *i
);
306 r
= symlink(options
, t
);
307 if (r
< 0 && errno
!= EEXIST
)
308 return log_error_errno(errno
, "Failed to create symlink %s: %m", t
);
309 #ifdef SMACK_RUN_LABEL
310 r
= mac_smack_copy(t
, options
);
311 if (r
< 0 && r
!= -EOPNOTSUPP
)
312 return log_error_errno(r
, "Failed to copy smack label from %s to %s: %m", options
, t
);
318 /* Now that we mounted everything, let's make the tmpfs the
319 * cgroup file systems are mounted into read-only. */
320 (void) mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
|MS_RDONLY
, "mode=755");
325 #if defined(HAVE_SELINUX) || defined(HAVE_SMACK)
328 const struct stat
*sb
,
330 struct FTW
*ftwbuf
) {
332 /* No need to label /dev twice in a row... */
333 if (_unlikely_(ftwbuf
->level
== 0))
336 label_fix(fpath
, false, false);
338 /* /run/initramfs is static data and big, no need to
339 * dynamically relabel its contents at boot... */
340 if (_unlikely_(ftwbuf
->level
== 1 &&
342 streq(fpath
, "/run/initramfs")))
343 return FTW_SKIP_SUBTREE
;
349 int mount_setup(bool loaded_policy
) {
353 for (i
= 0; i
< ELEMENTSOF(mount_table
); i
++) {
356 j
= mount_one(mount_table
+ i
, loaded_policy
);
357 if (j
!= 0 && r
>= 0)
364 #if defined(HAVE_SELINUX) || defined(HAVE_SMACK)
365 /* Nodes in devtmpfs and /run need to be manually updated for
366 * the appropriate labels, after mounting. The other virtual
367 * API file systems like /sys and /proc do not need that, they
368 * use the same label for all their files. */
370 usec_t before_relabel
, after_relabel
;
371 char timespan
[FORMAT_TIMESPAN_MAX
];
373 before_relabel
= now(CLOCK_MONOTONIC
);
375 nftw("/dev", nftw_cb
, 64, FTW_MOUNT
|FTW_PHYS
|FTW_ACTIONRETVAL
);
376 nftw("/run", nftw_cb
, 64, FTW_MOUNT
|FTW_PHYS
|FTW_ACTIONRETVAL
);
378 after_relabel
= now(CLOCK_MONOTONIC
);
380 log_info("Relabelled /dev and /run in %s.",
381 format_timespan(timespan
, sizeof(timespan
), after_relabel
- before_relabel
, 0));
385 /* Create a few default symlinks, which are normally created
386 * by udevd, but some scripts might need them before we start
388 dev_setup(NULL
, UID_INVALID
, GID_INVALID
);
390 /* Mark the root directory as shared in regards to mount
391 * propagation. The kernel defaults to "private", but we think
392 * it makes more sense to have a default of "shared" so that
393 * nspawn and the container tools work out of the box. If
394 * specific setups need other settings they can reset the
395 * propagation mode to private if needed. */
396 if (detect_container() <= 0)
397 if (mount(NULL
, "/", NULL
, MS_REC
|MS_SHARED
, NULL
) < 0)
398 log_warning_errno(errno
, "Failed to set up the root directory for shared mount propagation: %m");
400 /* Create a few directories we always want around, Note that
401 * sd_booted() checks for /run/systemd/system, so this mkdir
402 * really needs to stay for good, otherwise software that
403 * copied sd-daemon.c into their sources will misdetect
405 mkdir_label("/run/systemd", 0755);
406 mkdir_label("/run/systemd/system", 0755);
407 mkdir_label("/run/systemd/inaccessible", 0000);