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>
28 #include "mount-setup.h"
29 //#include "dev-setup.h"
30 //#include "bus-util.h"
38 #include "path-util.h"
39 //#include "missing.h"
41 //#include "efivars.h"
42 //#include "smack-util.h"
43 #include "cgroup-util.h"
45 typedef enum MountMode {
48 MNT_IN_CONTAINER = 1 << 1,
51 typedef struct MountPoint {
57 bool (*condition_fn)(void);
61 /* The first three entries we might need before SELinux is up. The
62 * fourth (securityfs) is needed by IMA to load a custom policy. The
63 * other ones we can delay until SELinux and IMA are loaded. When
64 * SMACK is enabled we need smackfs, too, so it's a fifth one. */
66 #define N_EARLY_MOUNT 5
68 #define N_EARLY_MOUNT 4
71 static const MountPoint mount_table[] = {
72 /// UNNEEDED by elogind
74 { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
75 NULL, MNT_FATAL|MNT_IN_CONTAINER },
76 { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
77 NULL, MNT_FATAL|MNT_IN_CONTAINER },
78 { "devtmpfs", "/dev", "devtmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME,
79 NULL, MNT_FATAL|MNT_IN_CONTAINER },
80 { "securityfs", "/sys/kernel/security", "securityfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
83 { "smackfs", "/sys/fs/smackfs", "smackfs", "smackfsdef=*", MS_NOSUID|MS_NOEXEC|MS_NODEV,
84 mac_smack_use, MNT_FATAL },
85 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
86 mac_smack_use, MNT_FATAL },
88 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
89 NULL, MNT_FATAL|MNT_IN_CONTAINER },
90 { "devpts", "/dev/pts", "devpts", "mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC,
91 NULL, MNT_IN_CONTAINER },
93 { "tmpfs", "/run", "tmpfs", "mode=755,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
94 mac_smack_use, MNT_FATAL },
96 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
97 NULL, MNT_FATAL|MNT_IN_CONTAINER },
98 { "cgroup", "/sys/fs/cgroup", "cgroup", "__DEVEL__sane_behavior", MS_NOSUID|MS_NOEXEC|MS_NODEV,
99 cg_is_unified_wanted, MNT_FATAL|MNT_IN_CONTAINER },
101 { "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
102 cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
103 /// UNNEEDED by elogind
105 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
106 cg_is_legacy_wanted, MNT_IN_CONTAINER },
107 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV,
108 cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
110 { "cgroup", "/sys/fs/cgroup/elogind", "cgroup", "none,name=elogind,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
111 cg_is_legacy_wanted, MNT_IN_CONTAINER },
112 { "cgroup", "/sys/fs/cgroup/elogind", "cgroup", "none,name=elogind", MS_NOSUID|MS_NOEXEC|MS_NODEV,
113 cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
115 /// UNNEEDED by elogind
117 { "pstore", "/sys/fs/pstore", "pstore", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
120 { "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
121 is_efi_boot, MNT_NONE },
123 { "kdbusfs", "/sys/fs/kdbus", "kdbusfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
124 is_kdbus_wanted, MNT_IN_CONTAINER },
128 /// UNNEEDED by elogind
130 /* These are API file systems that might be mounted by other software,
131 * we just list them here so that we know that we should ignore them */
133 static const char ignore_paths[] =
134 /* SELinux file systems */
136 /* Container bind mounts */
141 bool mount_point_is_api(const char *path) {
144 /* Checks if this mount point is considered "API", and hence
145 * should be ignored */
147 for (i = 0; i < ELEMENTSOF(mount_table); i ++)
148 if (path_equal(path, mount_table[i].where))
151 return path_startswith(path, "/sys/fs/cgroup/");
154 bool mount_point_ignore(const char *path) {
157 NULSTR_FOREACH(i, ignore_paths)
158 if (path_equal(path, i))
165 static int mount_one(const MountPoint *p, bool relabel) {
170 if (p->condition_fn && !p->condition_fn())
173 /* Relabel first, just in case */
175 label_fix(p->where, true, true);
177 r = path_is_mount_point(p->where, AT_SYMLINK_FOLLOW);
178 if (r < 0 && r != -ENOENT)
183 /* Skip securityfs in a container */
184 if (!(p->mode & MNT_IN_CONTAINER) && detect_container(NULL) > 0)
187 /* The access mode here doesn't really matter too much, since
188 * the mounted file system will take precedence anyway. */
190 mkdir_p_label(p->where, 0755);
192 mkdir_p(p->where, 0755);
194 log_debug("Mounting %s to %s of type %s with options %s.",
205 log_full((p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG, "Failed to mount %s at %s: %m", p->type, p->where);
206 return (p->mode & MNT_FATAL) ? -errno : 0;
209 /* Relabel again, since we now mounted something fresh here */
211 label_fix(p->where, false, false);
216 /// UNNEEDED by elogind
218 int mount_setup_early(void) {
222 assert_cc(N_EARLY_MOUNT <= ELEMENTSOF(mount_table));
224 /* Do a minimal mount of /proc and friends to enable the most
225 * basic stuff, such as SELinux */
226 for (i = 0; i < N_EARLY_MOUNT; i ++) {
229 j = mount_one(mount_table + i, false);
237 int mount_cgroup_controllers(char ***join_controllers) {
238 _cleanup_set_free_free_ Set *controllers = NULL;
241 if (!cg_is_legacy_wanted())
244 /* Mount all available cgroup controllers that are built into the kernel. */
246 controllers = set_new(&string_hash_ops);
250 r = cg_kernel_controllers(controllers);
252 return log_error_errno(r, "Failed to enumerate cgroup controllers: %m");
255 _cleanup_free_ char *options = NULL, *controller = NULL, *where = NULL;
259 .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV,
260 .mode = MNT_IN_CONTAINER,
264 controller = set_steal_first(controllers);
268 if (join_controllers)
269 for (k = join_controllers; *k; k++)
270 if (strv_find(*k, controller))
276 for (i = *k, j = *k; *i; i++) {
278 if (!streq(*i, controller)) {
279 _cleanup_free_ char *t;
281 t = set_remove(controllers, *i);
293 options = strv_join(*k, ",");
297 options = controller;
301 where = strappend("/sys/fs/cgroup/", options);
308 r = mount_one(&p, true);
312 if (r > 0 && k && *k) {
315 for (i = *k; *i; i++) {
316 _cleanup_free_ char *t = NULL;
318 t = strappend("/sys/fs/cgroup/", *i);
322 r = symlink(options, t);
323 if (r < 0 && errno != EEXIST)
324 return log_error_errno(errno, "Failed to create symlink %s: %m", t);
329 /* Now that we mounted everything, let's make the tmpfs the
330 * cgroup file systems are mounted into read-only. */
331 (void) mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
336 #if defined(HAVE_SELINUX) || defined(HAVE_SMACK)
339 const struct stat *sb,
341 struct FTW *ftwbuf) {
343 /* No need to label /dev twice in a row... */
344 if (_unlikely_(ftwbuf->level == 0))
347 label_fix(fpath, false, false);
349 /* /run/initramfs is static data and big, no need to
350 * dynamically relabel its contents at boot... */
351 if (_unlikely_(ftwbuf->level == 1 &&
353 streq(fpath, "/run/initramfs")))
354 return FTW_SKIP_SUBTREE;
361 int mount_setup(bool loaded_policy) {
365 for (i = 0; i < ELEMENTSOF(mount_table); i ++) {
368 j = mount_one(mount_table + i, loaded_policy);
376 /// elogind does not control /, /dev, /run and /run/systemd/* are setup elsewhere.
378 #if defined(HAVE_SELINUX) || defined(HAVE_SMACK)
379 /* Nodes in devtmpfs and /run need to be manually updated for
380 * the appropriate labels, after mounting. The other virtual
381 * API file systems like /sys and /proc do not need that, they
382 * use the same label for all their files. */
384 usec_t before_relabel, after_relabel;
385 char timespan[FORMAT_TIMESPAN_MAX];
387 before_relabel = now(CLOCK_MONOTONIC);
389 nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
390 nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
392 after_relabel = now(CLOCK_MONOTONIC);
394 log_info("Relabelled /dev and /run in %s.",
395 format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0));
399 /* Create a few default symlinks, which are normally created
400 * by udevd, but some scripts might need them before we start
402 dev_setup(NULL, UID_INVALID, GID_INVALID);
404 /* Mark the root directory as shared in regards to mount
405 * propagation. The kernel defaults to "private", but we think
406 * it makes more sense to have a default of "shared" so that
407 * nspawn and the container tools work out of the box. If
408 * specific setups need other settings they can reset the
409 * propagation mode to private if needed. */
410 if (detect_container(NULL) <= 0)
411 if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
412 log_warning_errno(errno, "Failed to set up the root directory for shared mount propagation: %m");
414 /* Create a few directories we always want around, Note that
415 * sd_booted() checks for /run/systemd/system, so this mkdir
416 * really needs to stay for good, otherwise software that
417 * copied sd-daemon.c into their sources will misdetect
419 mkdir_label("/run/systemd", 0755);
420 mkdir_label("/run/systemd/system", 0755);
421 mkdir_label("/run/systemd/inaccessible", 0000);