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 { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
71 NULL, MNT_FATAL|MNT_IN_CONTAINER },
72 { "proc", "/proc", "proc", 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 },
96 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
97 NULL, MNT_IN_CONTAINER },
98 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV,
99 NULL, MNT_FATAL|MNT_IN_CONTAINER },
100 { "pstore", "/sys/fs/pstore", "pstore", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
103 { "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
104 is_efi_boot, MNT_NONE },
108 /* These are API file systems that might be mounted by other software,
109 * we just list them here so that we know that we should ignore them */
111 static const char ignore_paths[] =
112 /* SELinux file systems */
115 /* Legacy cgroup mount points */
118 /* Legacy kernel file system */
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, true);
167 /* Skip securityfs in a container */
168 if (!(p->mode & MNT_IN_CONTAINER) && detect_container(NULL) > 0)
171 /* The access mode here doesn't really matter too much, since
172 * the mounted file system will take precedence anyway. */
174 mkdir_p_label(p->where, 0755);
176 mkdir_p(p->where, 0755);
178 log_debug("Mounting %s to %s of type %s with options %s.",
189 log_full((p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG, "Failed to mount %s at %s: %m", p->type, p->where);
190 return (p->mode & MNT_FATAL) ? -errno : 0;
193 /* Relabel again, since we now mounted something fresh here */
195 label_fix(p->where, false, false);
200 int mount_setup_early(void) {
204 assert_cc(N_EARLY_MOUNT <= ELEMENTSOF(mount_table));
206 /* Do a minimal mount of /proc and friends to enable the most
207 * basic stuff, such as SELinux */
208 for (i = 0; i < N_EARLY_MOUNT; i ++) {
211 j = mount_one(mount_table + i, false);
219 int mount_cgroup_controllers(char ***join_controllers) {
220 _cleanup_set_free_free_ Set *controllers = NULL;
221 _cleanup_fclose_ FILE *f;
225 /* Mount all available cgroup controllers that are built into the kernel. */
227 f = fopen("/proc/cgroups", "re");
229 log_error("Failed to enumerate cgroup controllers: %m");
233 controllers = set_new(string_hash_func, string_compare_func);
237 /* Ignore the header line */
238 (void) fgets(buf, sizeof(buf), f);
244 if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) {
249 log_error("Failed to parse /proc/cgroups.");
258 r = set_consume(controllers, controller);
260 log_error("Failed to add controller to set.");
266 _cleanup_free_ char *options = NULL, *controller = NULL, *where = NULL;
270 .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV,
271 .mode = MNT_IN_CONTAINER,
275 controller = set_steal_first(controllers);
279 if (join_controllers)
280 for (k = join_controllers; *k; k++)
281 if (strv_find(*k, controller))
287 for (i = *k, j = *k; *i; i++) {
289 if (!streq(*i, controller)) {
290 _cleanup_free_ char *t;
292 t = set_remove(controllers, *i);
304 options = strv_join(*k, ",");
308 options = controller;
312 where = strappend("/sys/fs/cgroup/", options);
319 r = mount_one(&p, true);
323 if (r > 0 && k && *k) {
326 for (i = *k; *i; i++) {
327 _cleanup_free_ char *t = NULL;
329 t = strappend("/sys/fs/cgroup/", *i);
333 r = symlink(options, t);
334 if (r < 0 && errno != EEXIST) {
335 log_error("Failed to create symlink %s: %m", t);
342 /* Now that we mounted everything, let's make the tmpfs the
343 * cgroup file systems are mounted into read-only. */
344 mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
351 const struct stat *sb,
353 struct FTW *ftwbuf) {
355 /* No need to label /dev twice in a row... */
356 if (_unlikely_(ftwbuf->level == 0))
359 label_fix(fpath, false, false);
361 /* /run/initramfs is static data and big, no need to
362 * dynamically relabel its contents at boot... */
363 if (_unlikely_(ftwbuf->level == 1 &&
365 streq(fpath, "/run/initramfs")))
366 return FTW_SKIP_SUBTREE;
371 int mount_setup(bool loaded_policy) {
375 for (i = 0; i < ELEMENTSOF(mount_table); i ++) {
376 r = mount_one(mount_table + i, true);
382 /* Nodes in devtmpfs and /run need to be manually updated for
383 * the appropriate labels, after mounting. The other virtual
384 * API file systems like /sys and /proc do not need that, they
385 * use the same label for all their files. */
387 usec_t before_relabel, after_relabel;
388 char timespan[FORMAT_TIMESPAN_MAX];
390 before_relabel = now(CLOCK_MONOTONIC);
392 nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
393 nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
395 after_relabel = now(CLOCK_MONOTONIC);
397 log_info("Relabelled /dev and /run in %s.",
398 format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0));
401 /* Create a few default symlinks, which are normally created
402 * by udevd, but some scripts might need them before we start
406 /* Mark the root directory as shared in regards to mount
407 * propagation. The kernel defaults to "private", but we think
408 * it makes more sense to have a default of "shared" so that
409 * nspawn and the container tools work out of the box. If
410 * specific setups need other settings they can reset the
411 * propagation mode to private if needed. */
412 if (detect_container(NULL) <= 0)
413 if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
414 log_warning("Failed to set up the root directory for shared mount propagation: %m");
416 /* Create a few directories we always want around, Note that
417 * sd_booted() checks for /run/systemd/system, so this mkdir
418 * really needs to stay for good, otherwise software that
419 * copied sd-daemon.c into their sources will misdetect
421 mkdir_label("/run/systemd", 0755);
422 mkdir_label("/run/systemd/system", 0755);
423 mkdir_label("/run/systemd/inaccessible", 0000);