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"
37 #include "path-util.h"
41 #include "smack-util.h"
42 #include "cgroup-util.h"
44 typedef enum MountMode {
47 MNT_IN_CONTAINER = 1 << 1,
50 typedef struct MountPoint {
56 bool (*condition_fn)(void);
60 /* The first three entries we might need before SELinux is up. The
61 * fourth (securityfs) is needed by IMA to load a custom policy. The
62 * other ones we can delay until SELinux and IMA are loaded. When
63 * SMACK is enabled we need smackfs, too, so it's a fifth one. */
65 #define N_EARLY_MOUNT 5
67 #define N_EARLY_MOUNT 4
70 static const MountPoint mount_table[] = {
71 { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
72 NULL, MNT_FATAL|MNT_IN_CONTAINER },
73 { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
74 NULL, MNT_FATAL|MNT_IN_CONTAINER },
75 { "devtmpfs", "/dev", "devtmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME,
76 NULL, MNT_FATAL|MNT_IN_CONTAINER },
77 { "securityfs", "/sys/kernel/security", "securityfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
80 { "smackfs", "/sys/fs/smackfs", "smackfs", "smackfsdef=*", MS_NOSUID|MS_NOEXEC|MS_NODEV,
81 mac_smack_use, MNT_FATAL },
82 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
83 mac_smack_use, MNT_FATAL },
85 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
86 NULL, MNT_FATAL|MNT_IN_CONTAINER },
87 { "devpts", "/dev/pts", "devpts", "mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC,
88 NULL, MNT_IN_CONTAINER },
90 { "tmpfs", "/run", "tmpfs", "mode=755,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
91 mac_smack_use, MNT_FATAL },
93 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
94 NULL, MNT_FATAL|MNT_IN_CONTAINER },
95 { "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
96 NULL, MNT_FATAL|MNT_IN_CONTAINER },
97 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
98 NULL, MNT_IN_CONTAINER },
99 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV,
100 NULL, MNT_FATAL|MNT_IN_CONTAINER },
101 { "pstore", "/sys/fs/pstore", "pstore", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
104 { "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
105 is_efi_boot, MNT_NONE },
108 { "kdbusfs", "/sys/fs/kdbus", "kdbusfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
109 NULL, MNT_IN_CONTAINER },
113 /* These are API file systems that might be mounted by other software,
114 * we just list them here so that we know that we should ignore them */
116 static const char ignore_paths[] =
117 /* SELinux file systems */
119 /* Container bind mounts */
124 bool mount_point_is_api(const char *path) {
127 /* Checks if this mount point is considered "API", and hence
128 * should be ignored */
130 for (i = 0; i < ELEMENTSOF(mount_table); i ++)
131 if (path_equal(path, mount_table[i].where))
134 return path_startswith(path, "/sys/fs/cgroup/");
137 bool mount_point_ignore(const char *path) {
140 NULSTR_FOREACH(i, ignore_paths)
141 if (path_equal(path, i))
147 static int mount_one(const MountPoint *p, bool relabel) {
152 if (p->condition_fn && !p->condition_fn())
155 /* Relabel first, just in case */
157 label_fix(p->where, true, true);
159 r = path_is_mount_point(p->where, true);
166 /* Skip securityfs in a container */
167 if (!(p->mode & MNT_IN_CONTAINER) && detect_container(NULL) > 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);
218 int mount_cgroup_controllers(char ***join_controllers) {
219 _cleanup_set_free_free_ Set *controllers = NULL;
222 /* Mount all available cgroup controllers that are built into the kernel. */
224 controllers = set_new(&string_hash_ops);
228 r = cg_kernel_controllers(controllers);
230 return log_error_errno(r, "Failed to enumerate cgroup controllers: %m");
233 _cleanup_free_ char *options = NULL, *controller = NULL, *where = NULL;
237 .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV,
238 .mode = MNT_IN_CONTAINER,
242 controller = set_steal_first(controllers);
246 if (join_controllers)
247 for (k = join_controllers; *k; k++)
248 if (strv_find(*k, controller))
254 for (i = *k, j = *k; *i; i++) {
256 if (!streq(*i, controller)) {
257 _cleanup_free_ char *t;
259 t = set_remove(controllers, *i);
271 options = strv_join(*k, ",");
275 options = controller;
279 where = strappend("/sys/fs/cgroup/", options);
286 r = mount_one(&p, true);
290 if (r > 0 && k && *k) {
293 for (i = *k; *i; i++) {
294 _cleanup_free_ char *t = NULL;
296 t = strappend("/sys/fs/cgroup/", *i);
300 r = symlink(options, t);
301 if (r < 0 && errno != EEXIST)
302 return log_error_errno(errno, "Failed to create symlink %s: %m", t);
307 /* Now that we mounted everything, let's make the tmpfs the
308 * cgroup file systems are mounted into read-only. */
309 (void) mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
314 #if defined(HAVE_SELINUX) || defined(HAVE_SMACK)
317 const struct stat *sb,
319 struct FTW *ftwbuf) {
321 /* No need to label /dev twice in a row... */
322 if (_unlikely_(ftwbuf->level == 0))
325 label_fix(fpath, false, false);
327 /* /run/initramfs is static data and big, no need to
328 * dynamically relabel its contents at boot... */
329 if (_unlikely_(ftwbuf->level == 1 &&
331 streq(fpath, "/run/initramfs")))
332 return FTW_SKIP_SUBTREE;
338 int mount_setup(bool loaded_policy) {
342 for (i = 0; i < ELEMENTSOF(mount_table); i ++) {
345 j = mount_one(mount_table + i, loaded_policy);
353 #if defined(HAVE_SELINUX) || defined(HAVE_SMACK)
354 /* Nodes in devtmpfs and /run need to be manually updated for
355 * the appropriate labels, after mounting. The other virtual
356 * API file systems like /sys and /proc do not need that, they
357 * use the same label for all their files. */
359 usec_t before_relabel, after_relabel;
360 char timespan[FORMAT_TIMESPAN_MAX];
362 before_relabel = now(CLOCK_MONOTONIC);
364 nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
365 nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
367 after_relabel = now(CLOCK_MONOTONIC);
369 log_info("Relabelled /dev and /run in %s.",
370 format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0));
374 /* Create a few default symlinks, which are normally created
375 * by udevd, but some scripts might need them before we start
379 /* Mark the root directory as shared in regards to mount
380 * propagation. The kernel defaults to "private", but we think
381 * it makes more sense to have a default of "shared" so that
382 * nspawn and the container tools work out of the box. If
383 * specific setups need other settings they can reset the
384 * propagation mode to private if needed. */
385 if (detect_container(NULL) <= 0)
386 if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
387 log_warning_errno(errno, "Failed to set up the root directory for shared mount propagation: %m");
389 /* Create a few directories we always want around, Note that
390 * sd_booted() checks for /run/systemd/system, so this mkdir
391 * really needs to stay for good, otherwise software that
392 * copied sd-daemon.c into their sources will misdetect
394 mkdir_label("/run/systemd", 0755);
395 mkdir_label("/run/systemd/system", 0755);
396 mkdir_label("/run/systemd/inaccessible", 0000);