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"
47 #include "cgroup-util.h"
49 typedef enum MountMode {
52 MNT_IN_CONTAINER = 1 << 1,
55 typedef struct MountPoint {
61 bool (*condition_fn)(void);
65 /* The first three entries we might need before SELinux is up. The
66 * fourth (securityfs) is needed by IMA to load a custom policy. The
67 * other ones we can delay until SELinux and IMA are loaded. When
68 * SMACK is enabled we need smackfs, too, so it's a fifth one. */
70 #define N_EARLY_MOUNT 5
72 #define N_EARLY_MOUNT 4
75 static const MountPoint mount_table[] = {
76 { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
77 NULL, MNT_FATAL|MNT_IN_CONTAINER },
78 { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
79 NULL, MNT_FATAL|MNT_IN_CONTAINER },
80 { "devtmpfs", "/dev", "devtmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME,
81 NULL, MNT_FATAL|MNT_IN_CONTAINER },
82 { "securityfs", "/sys/kernel/security", "securityfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
85 { "smackfs", "/sys/fs/smackfs", "smackfs", "smackfsdef=*", MS_NOSUID|MS_NOEXEC|MS_NODEV,
86 mac_smack_use, MNT_FATAL },
87 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
88 mac_smack_use, MNT_FATAL },
90 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
91 NULL, MNT_FATAL|MNT_IN_CONTAINER },
92 { "devpts", "/dev/pts", "devpts", "mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC,
93 NULL, MNT_IN_CONTAINER },
95 { "tmpfs", "/run", "tmpfs", "mode=755,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
96 mac_smack_use, MNT_FATAL },
98 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
99 NULL, MNT_FATAL|MNT_IN_CONTAINER },
100 { "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
101 NULL, MNT_FATAL|MNT_IN_CONTAINER },
102 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
103 NULL, MNT_IN_CONTAINER },
104 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV,
105 NULL, MNT_FATAL|MNT_IN_CONTAINER },
106 { "pstore", "/sys/fs/pstore", "pstore", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
109 { "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
110 is_efi_boot, MNT_NONE },
113 { "kdbusfs", "/sys/fs/kdbus", "kdbusfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
114 NULL, MNT_IN_CONTAINER },
118 /* These are API file systems that might be mounted by other software,
119 * we just list them here so that we know that we should ignore them */
121 static const char ignore_paths[] =
122 /* SELinux file systems */
125 /* Legacy cgroup mount points */
128 /* Legacy kernel file system */
130 /* Container bind mounts */
135 bool mount_point_is_api(const char *path) {
138 /* Checks if this mount point is considered "API", and hence
139 * should be ignored */
141 for (i = 0; i < ELEMENTSOF(mount_table); i ++)
142 if (path_equal(path, mount_table[i].where))
145 return path_startswith(path, "/sys/fs/cgroup/");
148 bool mount_point_ignore(const char *path) {
151 NULSTR_FOREACH(i, ignore_paths)
152 if (path_equal(path, i))
158 static int mount_one(const MountPoint *p, bool relabel) {
163 if (p->condition_fn && !p->condition_fn())
166 /* Relabel first, just in case */
168 label_fix(p->where, true, true);
170 r = path_is_mount_point(p->where, true);
177 /* Skip securityfs in a container */
178 if (!(p->mode & MNT_IN_CONTAINER) && detect_container(NULL) > 0)
181 /* The access mode here doesn't really matter too much, since
182 * the mounted file system will take precedence anyway. */
184 mkdir_p_label(p->where, 0755);
186 mkdir_p(p->where, 0755);
188 log_debug("Mounting %s to %s of type %s with options %s.",
199 log_full((p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG, "Failed to mount %s at %s: %m", p->type, p->where);
200 return (p->mode & MNT_FATAL) ? -errno : 0;
203 /* Relabel again, since we now mounted something fresh here */
205 label_fix(p->where, false, false);
210 int mount_setup_early(void) {
214 assert_cc(N_EARLY_MOUNT <= ELEMENTSOF(mount_table));
216 /* Do a minimal mount of /proc and friends to enable the most
217 * basic stuff, such as SELinux */
218 for (i = 0; i < N_EARLY_MOUNT; i ++) {
221 j = mount_one(mount_table + i, false);
229 int mount_cgroup_controllers(char ***join_controllers) {
230 _cleanup_set_free_free_ Set *controllers = NULL;
233 /* Mount all available cgroup controllers that are built into the kernel. */
235 controllers = set_new(&string_hash_ops);
239 r = cg_kernel_controllers(controllers);
241 return log_error_errno(r, "Failed to enumerate cgroup controllers: %m");
244 _cleanup_free_ char *options = NULL, *controller = NULL, *where = NULL;
248 .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV,
249 .mode = MNT_IN_CONTAINER,
253 controller = set_steal_first(controllers);
257 if (join_controllers)
258 for (k = join_controllers; *k; k++)
259 if (strv_find(*k, controller))
265 for (i = *k, j = *k; *i; i++) {
267 if (!streq(*i, controller)) {
268 _cleanup_free_ char *t;
270 t = set_remove(controllers, *i);
282 options = strv_join(*k, ",");
286 options = controller;
290 where = strappend("/sys/fs/cgroup/", options);
297 r = mount_one(&p, true);
301 if (r > 0 && k && *k) {
304 for (i = *k; *i; i++) {
305 _cleanup_free_ char *t = NULL;
307 t = strappend("/sys/fs/cgroup/", *i);
311 r = symlink(options, t);
312 if (r < 0 && errno != EEXIST)
313 return log_error_errno(errno, "Failed to create symlink %s: %m", t);
318 /* Now that we mounted everything, let's make the tmpfs the
319 * cgroup file systems are mounted into read-only. */
320 (void) mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
325 #if defined(HAVE_SELINUX) || defined(HAVE_SMACK)
328 const struct stat *sb,
330 struct FTW *ftwbuf) {
332 /* No need to label /dev twice in a row... */
333 if (_unlikely_(ftwbuf->level == 0))
336 label_fix(fpath, false, false);
338 /* /run/initramfs is static data and big, no need to
339 * dynamically relabel its contents at boot... */
340 if (_unlikely_(ftwbuf->level == 1 &&
342 streq(fpath, "/run/initramfs")))
343 return FTW_SKIP_SUBTREE;
349 int mount_setup(bool loaded_policy) {
353 for (i = 0; i < ELEMENTSOF(mount_table); i ++) {
356 j = mount_one(mount_table + i, loaded_policy);
364 #if defined(HAVE_SELINUX) || defined(HAVE_SMACK)
365 /* Nodes in devtmpfs and /run need to be manually updated for
366 * the appropriate labels, after mounting. The other virtual
367 * API file systems like /sys and /proc do not need that, they
368 * use the same label for all their files. */
370 usec_t before_relabel, after_relabel;
371 char timespan[FORMAT_TIMESPAN_MAX];
373 before_relabel = now(CLOCK_MONOTONIC);
375 nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
376 nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
378 after_relabel = now(CLOCK_MONOTONIC);
380 log_info("Relabelled /dev and /run in %s.",
381 format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0));
385 /* Create a few default symlinks, which are normally created
386 * by udevd, but some scripts might need them before we start
390 /* Mark the root directory as shared in regards to mount
391 * propagation. The kernel defaults to "private", but we think
392 * it makes more sense to have a default of "shared" so that
393 * nspawn and the container tools work out of the box. If
394 * specific setups need other settings they can reset the
395 * propagation mode to private if needed. */
396 if (detect_container(NULL) <= 0)
397 if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
398 log_warning_errno(errno, "Failed to set up the root directory for shared mount propagation: %m");
400 /* Create a few directories we always want around, Note that
401 * sd_booted() checks for /run/systemd/system, so this mkdir
402 * really needs to stay for good, otherwise software that
403 * copied sd-daemon.c into their sources will misdetect
405 mkdir_label("/run/systemd", 0755);
406 mkdir_label("/run/systemd/system", 0755);
407 mkdir_label("/run/systemd/inaccessible", 0000);