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"
50 typedef enum MountMode {
53 MNT_IN_CONTAINER = 1 << 1,
56 typedef struct MountPoint {
62 bool (*condition_fn)(void);
66 /* The first three entries we might need before SELinux is up. The
67 * fourth (securityfs) is needed by IMA to load a custom policy. The
68 * other ones we can delay until SELinux and IMA are loaded. */
69 #define N_EARLY_MOUNT 4
71 static const MountPoint mount_table[] = {
72 { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
73 NULL, MNT_FATAL|MNT_IN_CONTAINER },
74 { "sysfs", "/sys", "sysfs", 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 { "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
82 is_efi_boot, MNT_NONE },
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 },
88 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
89 NULL, MNT_FATAL|MNT_IN_CONTAINER },
90 { "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
91 NULL, MNT_IN_CONTAINER },
92 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV,
93 NULL, MNT_IN_CONTAINER },
96 /* These are API file systems that might be mounted by other software,
97 * we just list them here so that we know that we should ignore them */
99 static const char ignore_paths[] =
100 /* SELinux file systems */
103 /* Legacy cgroup mount points */
106 /* Legacy kernel file system */
108 /* Container bind mounts */
113 bool mount_point_is_api(const char *path) {
116 /* Checks if this mount point is considered "API", and hence
117 * should be ignored */
119 for (i = 0; i < ELEMENTSOF(mount_table); i ++)
120 if (path_equal(path, mount_table[i].where))
123 return path_startswith(path, "/sys/fs/cgroup/");
126 bool mount_point_ignore(const char *path) {
129 NULSTR_FOREACH(i, ignore_paths)
130 if (path_equal(path, i))
136 static int mount_one(const MountPoint *p, bool relabel) {
141 if (p->condition_fn && !p->condition_fn())
144 /* Relabel first, just in case */
146 label_fix(p->where, true, true);
148 r = path_is_mount_point(p->where, true);
155 /* Skip securityfs in a container */
156 if (!(p->mode & MNT_IN_CONTAINER) && detect_container(NULL) > 0)
159 /* The access mode here doesn't really matter too much, since
160 * the mounted file system will take precedence anyway. */
161 mkdir_p_label(p->where, 0755);
163 log_debug("Mounting %s to %s of type %s with options %s.",
174 log_full((p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG, "Failed to mount %s: %s", p->where, strerror(errno));
175 return (p->mode & MNT_FATAL) ? -errno : 0;
178 /* Relabel again, since we now mounted something fresh here */
180 label_fix(p->where, false, false);
185 int mount_setup_early(void) {
189 assert_cc(N_EARLY_MOUNT <= ELEMENTSOF(mount_table));
191 /* Do a minimal mount of /proc and friends to enable the most
192 * basic stuff, such as SELinux */
193 for (i = 0; i < N_EARLY_MOUNT; i ++) {
196 j = mount_one(mount_table + i, false);
204 int mount_cgroup_controllers(char ***join_controllers) {
210 /* Mount all available cgroup controllers that are built into the kernel. */
212 f = fopen("/proc/cgroups", "re");
214 log_error("Failed to enumerate cgroup controllers: %m");
218 controllers = set_new(string_hash_func, string_compare_func);
224 /* Ignore the header line */
225 (void) fgets(buf, sizeof(buf), f);
231 if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) {
236 log_error("Failed to parse /proc/cgroups.");
246 r = set_put(controllers, controller);
248 log_error("Failed to add controller to set.");
256 char *controller, *where, *options;
259 controller = set_steal_first(controllers);
263 if (join_controllers)
264 for (k = join_controllers; *k; k++)
265 if (strv_find(*k, controller))
271 for (i = *k, j = *k; *i; i++) {
273 if (!streq(*i, controller)) {
276 t = set_remove(controllers, *i);
289 options = strv_join(*k, ",");
297 options = controller;
301 where = strappend("/sys/fs/cgroup/", options);
313 p.flags = MS_NOSUID|MS_NOEXEC|MS_NODEV;
315 r = mount_one(&p, true);
324 if (r > 0 && k && *k) {
327 for (i = *k; *i; i++) {
330 t = strappend("/sys/fs/cgroup/", *i);
337 r = symlink(options, t);
340 if (r < 0 && errno != EEXIST) {
341 log_error("Failed to create symlink: %m");
355 set_free_free(controllers);
364 const struct stat *sb,
366 struct FTW *ftwbuf) {
368 /* No need to label /dev twice in a row... */
369 if (_unlikely_(ftwbuf->level == 0))
372 label_fix(fpath, false, false);
374 /* /run/initramfs is static data and big, no need to
375 * dynamically relabel its contents at boot... */
376 if (_unlikely_(ftwbuf->level == 1 &&
378 streq(fpath, "/run/initramfs")))
379 return FTW_SKIP_SUBTREE;
384 int mount_setup(bool loaded_policy) {
386 static const char relabel[] =
387 "/run/initramfs/root-fsck\0"
388 "/run/initramfs/shutdown\0";
394 for (i = 0; i < ELEMENTSOF(mount_table); i ++) {
395 r = mount_one(mount_table + i, true);
401 /* Nodes in devtmpfs and /run need to be manually updated for
402 * the appropriate labels, after mounting. The other virtual
403 * API file systems like /sys and /proc do not need that, they
404 * use the same label for all their files. */
406 usec_t before_relabel, after_relabel;
407 char timespan[FORMAT_TIMESPAN_MAX];
409 before_relabel = now(CLOCK_MONOTONIC);
411 nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
412 nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
414 /* Explicitly relabel these */
415 NULSTR_FOREACH(j, relabel)
416 label_fix(j, true, false);
418 after_relabel = now(CLOCK_MONOTONIC);
420 log_info("Relabelled /dev and /run in %s.",
421 format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel));
424 /* Create a few default symlinks, which are normally created
425 * by udevd, but some scripts might need them before we start
429 /* Mark the root directory as shared in regards to mount
430 * propagation. The kernel defaults to "private", but we think
431 * it makes more sense to have a default of "shared" so that
432 * nspawn and the container tools work out of the box. If
433 * specific setups need other settings they can reset the
434 * propagation mode to private if needed. */
435 if (detect_container(NULL) <= 0)
436 if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
437 log_warning("Failed to set up the root directory for shared mount propagation: %m");
439 /* Create a few directories we always want around */
440 mkdir_label("/run/systemd", 0755);
441 mkdir_label("/run/systemd/system", 0755);