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>
32 #include "mount-setup.h"
33 #include "dev-setup.h"
41 #include "path-util.h"
45 #include "smack-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. */
67 #define N_EARLY_MOUNT 5
69 static const MountPoint mount_table[] = {
70 { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
71 NULL, MNT_FATAL|MNT_IN_CONTAINER },
72 { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
73 NULL, MNT_FATAL|MNT_IN_CONTAINER },
74 { "devtmpfs", "/dev", "devtmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME,
75 NULL, MNT_FATAL|MNT_IN_CONTAINER },
76 { "securityfs", "/sys/kernel/security", "securityfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
79 { "smackfs", "/sys/fs/smackfs", "smackfs", "smackfsdef=*", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
80 use_smack, MNT_FATAL },
81 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
82 use_smack, MNT_FATAL },
84 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
85 NULL, MNT_FATAL|MNT_IN_CONTAINER },
86 { "devpts", "/dev/pts", "devpts", "mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC,
87 NULL, MNT_IN_CONTAINER },
89 { "tmpfs", "/run", "tmpfs", "mode=755,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
90 use_smack, MNT_FATAL },
92 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
93 NULL, MNT_FATAL|MNT_IN_CONTAINER },
94 { "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
95 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_FATAL|MNT_IN_CONTAINER },
100 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV,
101 NULL, MNT_FATAL|MNT_IN_CONTAINER },
102 { "pstore", "/sys/fs/pstore", "pstore", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
105 { "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
106 is_efi_boot, MNT_NONE },
110 /* These are API file systems that might be mounted by other software,
111 * we just list them here so that we know that we should ignore them */
113 static const char ignore_paths[] =
114 /* SELinux file systems */
117 /* Legacy cgroup mount points */
120 /* Legacy kernel file system */
122 /* Container bind mounts */
127 bool mount_point_is_api(const char *path) {
130 /* Checks if this mount point is considered "API", and hence
131 * should be ignored */
133 for (i = 0; i < ELEMENTSOF(mount_table); i ++)
134 if (path_equal(path, mount_table[i].where))
137 return path_startswith(path, "/sys/fs/cgroup/");
140 bool mount_point_ignore(const char *path) {
143 NULSTR_FOREACH(i, ignore_paths)
144 if (path_equal(path, i))
150 static int mount_one(const MountPoint *p, bool relabel) {
155 if (p->condition_fn && !p->condition_fn())
158 /* Relabel first, just in case */
160 label_fix(p->where, true, true);
162 r = path_is_mount_point(p->where, true);
169 /* Skip securityfs in a container */
170 if (!(p->mode & MNT_IN_CONTAINER) && detect_container(NULL) > 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);
221 int mount_cgroup_controllers(char ***join_controllers) {
222 _cleanup_set_free_free_ Set *controllers = NULL;
223 _cleanup_fclose_ FILE *f;
227 /* Mount all available cgroup controllers that are built into the kernel. */
229 f = fopen("/proc/cgroups", "re");
231 log_error("Failed to enumerate cgroup controllers: %m");
235 controllers = set_new(string_hash_func, string_compare_func);
239 /* Ignore the header line */
240 (void) fgets(buf, sizeof(buf), f);
246 if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) {
251 log_error("Failed to parse /proc/cgroups.");
260 r = set_consume(controllers, controller);
262 log_error("Failed to add controller to set.");
268 _cleanup_free_ char *options = NULL, *controller = NULL, *where = NULL;
272 .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV,
273 .mode = MNT_IN_CONTAINER,
277 controller = set_steal_first(controllers);
281 if (join_controllers)
282 for (k = join_controllers; *k; k++)
283 if (strv_find(*k, controller))
289 for (i = *k, j = *k; *i; i++) {
291 if (!streq(*i, controller)) {
292 _cleanup_free_ char *t;
294 t = set_remove(controllers, *i);
306 options = strv_join(*k, ",");
310 options = controller;
314 where = strappend("/sys/fs/cgroup/", options);
321 r = mount_one(&p, true);
325 if (r > 0 && k && *k) {
328 for (i = *k; *i; i++) {
329 _cleanup_free_ char *t = NULL;
331 t = strappend("/sys/fs/cgroup/", *i);
335 r = symlink(options, t);
336 if (r < 0 && errno != EEXIST) {
337 log_error("Failed to create symlink %s: %m", t);
344 /* Now that we mounted everything, let's make the tmpfs the
345 * cgroup file systems are mounted into read-only. */
346 mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
353 const struct stat *sb,
355 struct FTW *ftwbuf) {
357 /* No need to label /dev twice in a row... */
358 if (_unlikely_(ftwbuf->level == 0))
361 label_fix(fpath, false, false);
363 /* /run/initramfs is static data and big, no need to
364 * dynamically relabel its contents at boot... */
365 if (_unlikely_(ftwbuf->level == 1 &&
367 streq(fpath, "/run/initramfs")))
368 return FTW_SKIP_SUBTREE;
373 int mount_setup(bool loaded_policy) {
377 for (i = 0; i < ELEMENTSOF(mount_table); i ++) {
378 r = mount_one(mount_table + i, true);
384 /* Nodes in devtmpfs and /run need to be manually updated for
385 * the appropriate labels, after mounting. The other virtual
386 * API file systems like /sys and /proc do not need that, they
387 * use the same label for all their files. */
389 usec_t before_relabel, after_relabel;
390 char timespan[FORMAT_TIMESPAN_MAX];
392 before_relabel = now(CLOCK_MONOTONIC);
394 nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
395 nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
397 after_relabel = now(CLOCK_MONOTONIC);
399 log_info("Relabelled /dev and /run in %s.",
400 format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0));
403 /* Create a few default symlinks, which are normally created
404 * by udevd, but some scripts might need them before we start
408 /* Mark the root directory as shared in regards to mount
409 * propagation. The kernel defaults to "private", but we think
410 * it makes more sense to have a default of "shared" so that
411 * nspawn and the container tools work out of the box. If
412 * specific setups need other settings they can reset the
413 * propagation mode to private if needed. */
414 if (detect_container(NULL) <= 0)
415 if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
416 log_warning("Failed to set up the root directory for shared mount propagation: %m");
418 /* Create a few directories we always want around, Note that
419 * sd_booted() checks for /run/systemd/system, so this mkdir
420 * really needs to stay for good, otherwise software that
421 * copied sd-daemon.c into their sources will misdetect
423 mkdir_label("/run/systemd", 0755);
424 mkdir_label("/run/systemd/system", 0755);
425 mkdir_label("/run/systemd/inaccessible", 0000);