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 "mount-setup.h"
29 #include "dev-setup.h"
38 #include "path-util.h"
42 #include "smack-util.h"
43 #include "cgroup-util.h"
45 typedef enum MountMode
{
48 MNT_IN_CONTAINER
= 1 << 1,
51 typedef struct MountPoint
{
57 bool (*condition_fn
)(void);
61 /* The first three entries we might need before SELinux is up. The
62 * fourth (securityfs) is needed by IMA to load a custom policy. The
63 * other ones we can delay until SELinux and IMA are loaded. When
64 * SMACK is enabled we need smackfs, too, so it's a fifth one. */
66 #define N_EARLY_MOUNT 5
68 #define N_EARLY_MOUNT 4
71 static const MountPoint mount_table
[] = {
72 { "sysfs", "/sys", "sysfs", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
73 NULL
, MNT_FATAL
|MNT_IN_CONTAINER
},
74 { "proc", "/proc", "proc", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
75 NULL
, MNT_FATAL
|MNT_IN_CONTAINER
},
76 { "devtmpfs", "/dev", "devtmpfs", "mode=755", MS_NOSUID
|MS_STRICTATIME
,
77 NULL
, MNT_FATAL
|MNT_IN_CONTAINER
},
78 { "securityfs", "/sys/kernel/security", "securityfs", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
81 { "smackfs", "/sys/fs/smackfs", "smackfs", "smackfsdef=*", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
82 mac_smack_use
, MNT_FATAL
},
83 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777,smackfsroot=*", MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
,
84 mac_smack_use
, MNT_FATAL
},
86 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
,
87 NULL
, MNT_FATAL
|MNT_IN_CONTAINER
},
88 { "devpts", "/dev/pts", "devpts", "mode=620,gid=" STRINGIFY(TTY_GID
), MS_NOSUID
|MS_NOEXEC
,
89 NULL
, MNT_IN_CONTAINER
},
91 { "tmpfs", "/run", "tmpfs", "mode=755,smackfsroot=*", MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
,
92 mac_smack_use
, MNT_FATAL
},
94 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
,
95 NULL
, MNT_FATAL
|MNT_IN_CONTAINER
},
96 { "cgroup", "/sys/fs/cgroup", "cgroup", "__DEVEL__sane_behavior", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
97 cg_is_unified_wanted
, MNT_FATAL
|MNT_IN_CONTAINER
},
98 { "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
,
99 cg_is_legacy_wanted
, MNT_FATAL
|MNT_IN_CONTAINER
},
100 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd,xattr", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
101 cg_is_legacy_wanted
, MNT_IN_CONTAINER
},
102 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
103 cg_is_legacy_wanted
, MNT_FATAL
|MNT_IN_CONTAINER
},
104 { "pstore", "/sys/fs/pstore", "pstore", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
107 { "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
108 is_efi_boot
, MNT_NONE
},
110 { "kdbusfs", "/sys/fs/kdbus", "kdbusfs", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
111 is_kdbus_wanted
, MNT_IN_CONTAINER
},
114 /* These are API file systems that might be mounted by other software,
115 * we just list them here so that we know that we should ignore them */
117 static const char ignore_paths
[] =
118 /* SELinux file systems */
120 /* Container bind mounts */
125 bool mount_point_is_api(const char *path
) {
128 /* Checks if this mount point is considered "API", and hence
129 * should be ignored */
131 for (i
= 0; i
< ELEMENTSOF(mount_table
); i
++)
132 if (path_equal(path
, mount_table
[i
].where
))
135 return path_startswith(path
, "/sys/fs/cgroup/");
138 bool mount_point_ignore(const char *path
) {
141 NULSTR_FOREACH(i
, ignore_paths
)
142 if (path_equal(path
, i
))
148 static int mount_one(const MountPoint
*p
, bool relabel
) {
153 if (p
->condition_fn
&& !p
->condition_fn())
156 /* Relabel first, just in case */
158 label_fix(p
->where
, true, true);
160 r
= path_is_mount_point(p
->where
, AT_SYMLINK_FOLLOW
);
161 if (r
< 0 && r
!= -ENOENT
)
166 /* Skip securityfs in a container */
167 if (!(p
->mode
& MNT_IN_CONTAINER
) && detect_container() > 0)
170 /* The access mode here doesn't really matter too much, since
171 * the mounted file system will take precedence anyway. */
173 mkdir_p_label(p
->where
, 0755);
175 mkdir_p(p
->where
, 0755);
177 log_debug("Mounting %s to %s of type %s with options %s.",
188 log_full((p
->mode
& MNT_FATAL
) ? LOG_ERR
: LOG_DEBUG
, "Failed to mount %s at %s: %m", p
->type
, p
->where
);
189 return (p
->mode
& MNT_FATAL
) ? -errno
: 0;
192 /* Relabel again, since we now mounted something fresh here */
194 label_fix(p
->where
, false, false);
199 int mount_setup_early(void) {
203 assert_cc(N_EARLY_MOUNT
<= ELEMENTSOF(mount_table
));
205 /* Do a minimal mount of /proc and friends to enable the most
206 * basic stuff, such as SELinux */
207 for (i
= 0; i
< N_EARLY_MOUNT
; i
++) {
210 j
= mount_one(mount_table
+ i
, false);
211 if (j
!= 0 && r
>= 0)
218 int mount_cgroup_controllers(char ***join_controllers
) {
219 _cleanup_set_free_free_ Set
*controllers
= NULL
;
222 if (!cg_is_legacy_wanted())
225 /* Mount all available cgroup controllers that are built into the kernel. */
227 controllers
= set_new(&string_hash_ops
);
231 r
= cg_kernel_controllers(controllers
);
233 return log_error_errno(r
, "Failed to enumerate cgroup controllers: %m");
236 _cleanup_free_
char *options
= NULL
, *controller
= NULL
, *where
= NULL
;
240 .flags
= MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
241 .mode
= MNT_IN_CONTAINER
,
245 controller
= set_steal_first(controllers
);
249 if (join_controllers
)
250 for (k
= join_controllers
; *k
; k
++)
251 if (strv_find(*k
, controller
))
257 for (i
= *k
, j
= *k
; *i
; i
++) {
259 if (!streq(*i
, controller
)) {
260 _cleanup_free_
char *t
;
262 t
= set_remove(controllers
, *i
);
274 options
= strv_join(*k
, ",");
278 options
= controller
;
282 where
= strappend("/sys/fs/cgroup/", options
);
289 r
= mount_one(&p
, true);
293 if (r
> 0 && k
&& *k
) {
296 for (i
= *k
; *i
; i
++) {
297 _cleanup_free_
char *t
= NULL
;
299 t
= strappend("/sys/fs/cgroup/", *i
);
303 r
= symlink(options
, t
);
304 if (r
< 0 && errno
!= EEXIST
)
305 return log_error_errno(errno
, "Failed to create symlink %s: %m", t
);
306 #ifdef SMACK_RUN_LABEL
307 r
= mac_smack_copy(t
, options
);
308 if (r
< 0 && r
!= -EOPNOTSUPP
)
309 return log_error_errno(r
, "Failed to copy smack label from %s to %s: %m", options
, t
);
315 /* Now that we mounted everything, let's make the tmpfs the
316 * cgroup file systems are mounted into read-only. */
317 (void) mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_STRICTATIME
|MS_RDONLY
, "mode=755");
322 #if defined(HAVE_SELINUX) || defined(HAVE_SMACK)
325 const struct stat
*sb
,
327 struct FTW
*ftwbuf
) {
329 /* No need to label /dev twice in a row... */
330 if (_unlikely_(ftwbuf
->level
== 0))
333 label_fix(fpath
, false, false);
335 /* /run/initramfs is static data and big, no need to
336 * dynamically relabel its contents at boot... */
337 if (_unlikely_(ftwbuf
->level
== 1 &&
339 streq(fpath
, "/run/initramfs")))
340 return FTW_SKIP_SUBTREE
;
346 int mount_setup(bool loaded_policy
) {
350 for (i
= 0; i
< ELEMENTSOF(mount_table
); i
++) {
353 j
= mount_one(mount_table
+ i
, loaded_policy
);
354 if (j
!= 0 && r
>= 0)
361 #if defined(HAVE_SELINUX) || defined(HAVE_SMACK)
362 /* Nodes in devtmpfs and /run need to be manually updated for
363 * the appropriate labels, after mounting. The other virtual
364 * API file systems like /sys and /proc do not need that, they
365 * use the same label for all their files. */
367 usec_t before_relabel
, after_relabel
;
368 char timespan
[FORMAT_TIMESPAN_MAX
];
370 before_relabel
= now(CLOCK_MONOTONIC
);
372 nftw("/dev", nftw_cb
, 64, FTW_MOUNT
|FTW_PHYS
|FTW_ACTIONRETVAL
);
373 nftw("/run", nftw_cb
, 64, FTW_MOUNT
|FTW_PHYS
|FTW_ACTIONRETVAL
);
375 after_relabel
= now(CLOCK_MONOTONIC
);
377 log_info("Relabelled /dev and /run in %s.",
378 format_timespan(timespan
, sizeof(timespan
), after_relabel
- before_relabel
, 0));
382 /* Create a few default symlinks, which are normally created
383 * by udevd, but some scripts might need them before we start
385 dev_setup(NULL
, UID_INVALID
, GID_INVALID
);
387 /* Mark the root directory as shared in regards to mount
388 * propagation. The kernel defaults to "private", but we think
389 * it makes more sense to have a default of "shared" so that
390 * nspawn and the container tools work out of the box. If
391 * specific setups need other settings they can reset the
392 * propagation mode to private if needed. */
393 if (detect_container() <= 0)
394 if (mount(NULL
, "/", NULL
, MS_REC
|MS_SHARED
, NULL
) < 0)
395 log_warning_errno(errno
, "Failed to set up the root directory for shared mount propagation: %m");
397 /* Create a few directories we always want around, Note that
398 * sd_booted() checks for /run/systemd/system, so this mkdir
399 * really needs to stay for good, otherwise software that
400 * copied sd-daemon.c into their sources will misdetect
402 mkdir_label("/run/systemd", 0755);
403 mkdir_label("/run/systemd/system", 0755);
404 mkdir_label("/run/systemd/inaccessible", 0000);