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. 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 { "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
100 NULL, MNT_FATAL|MNT_IN_CONTAINER },
101 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
102 NULL, MNT_IN_CONTAINER },
103 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV,
104 NULL, MNT_FATAL|MNT_IN_CONTAINER },
105 { "pstore", "/sys/fs/pstore", "pstore", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
108 { "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
109 is_efi_boot, MNT_NONE },
112 { "kdbusfs", "/sys/fs/kdbus", "kdbusfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
113 NULL, 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 */
124 /* Legacy cgroup mount points */
127 /* Legacy kernel file system */
129 /* Container bind mounts */
134 bool mount_point_is_api(const char *path) {
137 /* Checks if this mount point is considered "API", and hence
138 * should be ignored */
140 for (i = 0; i < ELEMENTSOF(mount_table); i ++)
141 if (path_equal(path, mount_table[i].where))
144 return path_startswith(path, "/sys/fs/cgroup/");
147 bool mount_point_ignore(const char *path) {
150 NULSTR_FOREACH(i, ignore_paths)
151 if (path_equal(path, i))
157 static int mount_one(const MountPoint *p, bool relabel) {
162 if (p->condition_fn && !p->condition_fn())
165 /* Relabel first, just in case */
167 label_fix(p->where, true, true);
169 r = path_is_mount_point(p->where, true);
176 /* Skip securityfs in a container */
177 if (!(p->mode & MNT_IN_CONTAINER) && detect_container(NULL) > 0)
180 /* The access mode here doesn't really matter too much, since
181 * the mounted file system will take precedence anyway. */
183 mkdir_p_label(p->where, 0755);
185 mkdir_p(p->where, 0755);
187 log_debug("Mounting %s to %s of type %s with options %s.",
198 log_full((p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG, "Failed to mount %s at %s: %m", p->type, p->where);
199 return (p->mode & MNT_FATAL) ? -errno : 0;
202 /* Relabel again, since we now mounted something fresh here */
204 label_fix(p->where, false, false);
209 int mount_setup_early(void) {
213 assert_cc(N_EARLY_MOUNT <= ELEMENTSOF(mount_table));
215 /* Do a minimal mount of /proc and friends to enable the most
216 * basic stuff, such as SELinux */
217 for (i = 0; i < N_EARLY_MOUNT; i ++) {
220 j = mount_one(mount_table + i, false);
228 int mount_cgroup_controllers(char ***join_controllers) {
229 _cleanup_set_free_free_ Set *controllers = NULL;
230 _cleanup_fclose_ FILE *f;
234 /* Mount all available cgroup controllers that are built into the kernel. */
236 f = fopen("/proc/cgroups", "re");
238 log_error("Failed to enumerate cgroup controllers: %m");
242 controllers = set_new(&string_hash_ops);
246 /* Ignore the header line */
247 (void) fgets(buf, sizeof(buf), f);
253 if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) {
258 log_error("Failed to parse /proc/cgroups.");
267 r = set_consume(controllers, controller);
269 log_error("Failed to add controller to set.");
275 _cleanup_free_ char *options = NULL, *controller = NULL, *where = NULL;
279 .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV,
280 .mode = MNT_IN_CONTAINER,
284 controller = set_steal_first(controllers);
288 if (join_controllers)
289 for (k = join_controllers; *k; k++)
290 if (strv_find(*k, controller))
296 for (i = *k, j = *k; *i; i++) {
298 if (!streq(*i, controller)) {
299 _cleanup_free_ char *t;
301 t = set_remove(controllers, *i);
313 options = strv_join(*k, ",");
317 options = controller;
321 where = strappend("/sys/fs/cgroup/", options);
328 r = mount_one(&p, true);
332 if (r > 0 && k && *k) {
335 for (i = *k; *i; i++) {
336 _cleanup_free_ char *t = NULL;
338 t = strappend("/sys/fs/cgroup/", *i);
342 r = symlink(options, t);
343 if (r < 0 && errno != EEXIST) {
344 log_error("Failed to create symlink %s: %m", t);
351 /* Now that we mounted everything, let's make the tmpfs the
352 * cgroup file systems are mounted into read-only. */
353 mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
358 #if defined(HAVE_SELINUX) || defined(HAVE_SMACK)
361 const struct stat *sb,
363 struct FTW *ftwbuf) {
365 /* No need to label /dev twice in a row... */
366 if (_unlikely_(ftwbuf->level == 0))
369 label_fix(fpath, false, false);
371 /* /run/initramfs is static data and big, no need to
372 * dynamically relabel its contents at boot... */
373 if (_unlikely_(ftwbuf->level == 1 &&
375 streq(fpath, "/run/initramfs")))
376 return FTW_SKIP_SUBTREE;
382 int mount_setup(bool loaded_policy) {
386 for (i = 0; i < ELEMENTSOF(mount_table); i ++) {
389 j = mount_one(mount_table + i, loaded_policy);
397 #if defined(HAVE_SELINUX) || defined(HAVE_SMACK)
398 /* Nodes in devtmpfs and /run need to be manually updated for
399 * the appropriate labels, after mounting. The other virtual
400 * API file systems like /sys and /proc do not need that, they
401 * use the same label for all their files. */
403 usec_t before_relabel, after_relabel;
404 char timespan[FORMAT_TIMESPAN_MAX];
406 before_relabel = now(CLOCK_MONOTONIC);
408 nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
409 nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
411 after_relabel = now(CLOCK_MONOTONIC);
413 log_info("Relabelled /dev and /run in %s.",
414 format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0));
418 /* Create a few default symlinks, which are normally created
419 * by udevd, but some scripts might need them before we start
423 /* Mark the root directory as shared in regards to mount
424 * propagation. The kernel defaults to "private", but we think
425 * it makes more sense to have a default of "shared" so that
426 * nspawn and the container tools work out of the box. If
427 * specific setups need other settings they can reset the
428 * propagation mode to private if needed. */
429 if (detect_container(NULL) <= 0)
430 if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
431 log_warning("Failed to set up the root directory for shared mount propagation: %m");
433 /* Create a few directories we always want around, Note that
434 * sd_booted() checks for /run/systemd/system, so this mkdir
435 * really needs to stay for good, otherwise software that
436 * copied sd-daemon.c into their sources will misdetect
438 mkdir_label("/run/systemd", 0755);
439 mkdir_label("/run/systemd/system", 0755);
440 mkdir_label("/run/systemd/inaccessible", 0000);