1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 Copyright 2010 Lennart Poettering
10 //#include <sys/statvfs.h>
13 #include "alloc-util.h"
14 //#include "bus-util.h"
15 #include "cgroup-util.h"
16 //#include "dev-setup.h"
17 //#include "efivars.h"
23 //#include "missing.h"
25 #include "mount-setup.h"
26 #include "mount-util.h"
27 #include "path-util.h"
29 //#include "smack-util.h"
31 #include "user-util.h"
35 /// Additional includes needed by elogind
36 #include "string-util.h"
38 typedef enum MountMode {
41 MNT_IN_CONTAINER = 1 << 1,
42 MNT_CHECK_WRITABLE = 1 << 2,
45 typedef struct MountPoint {
51 bool (*condition_fn)(void);
55 /* The first three entries we might need before SELinux is up. The
56 * fourth (securityfs) is needed by IMA to load a custom policy. The
57 * other ones we can delay until SELinux and IMA are loaded. When
58 * SMACK is enabled we need smackfs, too, so it's a fifth one. */
60 #define N_EARLY_MOUNT 5
62 #define N_EARLY_MOUNT 4
65 static const MountPoint mount_table[] = {
66 #if 0 /// UNNEEDED by elogind
67 { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
68 NULL, MNT_FATAL|MNT_IN_CONTAINER },
69 { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
70 NULL, MNT_FATAL|MNT_IN_CONTAINER },
71 { "devtmpfs", "/dev", "devtmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME,
72 NULL, MNT_FATAL|MNT_IN_CONTAINER },
73 { "securityfs", "/sys/kernel/security", "securityfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
76 { "smackfs", "/sys/fs/smackfs", "smackfs", "smackfsdef=*", MS_NOSUID|MS_NOEXEC|MS_NODEV,
77 mac_smack_use, MNT_FATAL },
78 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
79 mac_smack_use, MNT_FATAL },
81 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
82 NULL, MNT_FATAL|MNT_IN_CONTAINER },
83 { "devpts", "/dev/pts", "devpts", "mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC,
84 NULL, MNT_IN_CONTAINER },
86 { "tmpfs", "/run", "tmpfs", "mode=755,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
87 mac_smack_use, MNT_FATAL },
89 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
90 NULL, MNT_FATAL|MNT_IN_CONTAINER },
92 { "cgroup2", "/sys/fs/cgroup", "cgroup2", "nsdelegate", MS_NOSUID|MS_NOEXEC|MS_NODEV,
93 cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
94 { "cgroup2", "/sys/fs/cgroup", "cgroup2", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
95 cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
96 { "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
97 cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
98 { "cgroup2", "/sys/fs/cgroup/unified", "cgroup2", "nsdelegate", MS_NOSUID|MS_NOEXEC|MS_NODEV,
99 cg_is_hybrid_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
100 { "cgroup2", "/sys/fs/cgroup/unified", "cgroup2", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
101 cg_is_hybrid_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
102 #if 0 /// UNNEEDED by elogind
103 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
104 cg_is_legacy_wanted, MNT_IN_CONTAINER },
105 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV,
106 cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
107 { "pstore", "/sys/fs/pstore", "pstore", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
110 { "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
111 is_efi_boot, MNT_NONE },
113 { "bpf", "/sys/fs/bpf", "bpf", "mode=700", MS_NOSUID|MS_NOEXEC|MS_NODEV,
116 { "cgroup", "/sys/fs/cgroup/elogind", "cgroup", "none,name=elogind,release_agent="SYSTEMD_CGROUP_AGENT_PATH",xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
117 cg_is_legacy_wanted, MNT_IN_CONTAINER },
118 { "cgroup", "/sys/fs/cgroup/elogind", "cgroup", "none,name=elogind,release_agent="SYSTEMD_CGROUP_AGENT_PATH, MS_NOSUID|MS_NOEXEC|MS_NODEV,
119 cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
123 #if 0 /// UNNEEDED by elogind
124 /* These are API file systems that might be mounted by other software,
125 * we just list them here so that we know that we should ignore them */
127 static const char ignore_paths[] =
128 /* SELinux file systems */
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))
159 static int mount_one(const MountPoint *p, bool relabel) {
164 priority = (p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG;
166 if (p->condition_fn && !p->condition_fn())
169 /* Relabel first, just in case */
171 (void) label_fix(p->where, LABEL_IGNORE_ENOENT|LABEL_IGNORE_EROFS);
173 r = path_is_mount_point(p->where, NULL, AT_SYMLINK_FOLLOW);
174 if (r < 0 && r != -ENOENT) {
175 log_full_errno(priority, r, "Failed to determine whether %s is a mount point: %m", p->where);
176 return (p->mode & MNT_FATAL) ? r : 0;
181 /* Skip securityfs in a container */
182 if (!(p->mode & MNT_IN_CONTAINER) && detect_container() > 0)
185 /* The access mode here doesn't really matter too much, since
186 * the mounted file system will take precedence anyway. */
188 (void) mkdir_p_label(p->where, 0755);
190 (void) mkdir_p(p->where, 0755);
192 log_debug("Mounting %s to %s of type %s with options %s.",
203 log_full_errno(priority, errno, "Failed to mount %s at %s: %m", p->type, p->where);
204 return (p->mode & MNT_FATAL) ? -errno : 0;
207 /* Relabel again, since we now mounted something fresh here */
209 (void) label_fix(p->where, 0);
211 if (p->mode & MNT_CHECK_WRITABLE) {
212 if (access(p->where, W_OK) < 0) {
215 (void) umount(p->where);
216 (void) rmdir(p->where);
218 log_full_errno(priority, r, "Mount point %s not writable after mounting: %m", p->where);
219 return (p->mode & MNT_FATAL) ? r : 0;
226 static int mount_points_setup(unsigned n, bool loaded_policy) {
230 for (i = 0; i < n; i ++) {
233 j = mount_one(mount_table + i, loaded_policy);
234 if (j != 0 && r >= 0)
241 #if 0 /// UNNEEDED by elogind
242 int mount_setup_early(void) {
243 assert_cc(N_EARLY_MOUNT <= ELEMENTSOF(mount_table));
245 /* Do a minimal mount of /proc and friends to enable the most
246 * basic stuff, such as SELinux */
247 return mount_points_setup(N_EARLY_MOUNT, false);
250 int mount_cgroup_controllers(char ***join_controllers) {
251 _cleanup_set_free_free_ Set *controllers = NULL;
252 bool has_argument = !!join_controllers;
255 if (!cg_is_legacy_wanted())
258 /* Mount all available cgroup controllers that are built into the kernel. */
262 * mount "cpu" + "cpuacct" together, and "net_cls" + "net_prio".
264 * We'd like to add "cpuset" to the mix, but "cpuset" doesn't really
265 * work for groups with no initialized attributes.
267 join_controllers = (char**[]) {
268 STRV_MAKE("cpu", "cpuacct"),
269 STRV_MAKE("net_cls", "net_prio"),
273 r = cg_kernel_controllers(&controllers);
275 return log_error_errno(r, "Failed to enumerate cgroup controllers: %m");
278 _cleanup_free_ char *options = NULL, *controller = NULL, *where = NULL;
282 .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV,
283 .mode = MNT_IN_CONTAINER,
287 controller = set_steal_first(controllers);
291 for (k = join_controllers; *k; k++)
292 if (strv_find(*k, controller))
298 for (i = *k, j = *k; *i; i++) {
300 if (!streq(*i, controller)) {
301 _cleanup_free_ char *t;
303 t = set_remove(controllers, *i);
316 options = strv_join(*k, ",");
320 options = TAKE_PTR(controller);
322 where = strappend("/sys/fs/cgroup/", options);
329 r = mount_one(&p, true);
336 for (i = *k; *i; i++) {
337 _cleanup_free_ char *t = NULL;
339 t = strappend("/sys/fs/cgroup/", *i);
343 r = symlink(options, t);
345 #ifdef SMACK_RUN_LABEL
346 _cleanup_free_ char *src;
347 src = strappend("/sys/fs/cgroup/", options);
350 r = mac_smack_copy(t, src);
351 if (r < 0 && r != -EOPNOTSUPP)
352 return log_error_errno(r, "Failed to copy smack label from %s to %s: %m", src, t);
354 } else if (errno != EEXIST)
355 return log_error_errno(errno, "Failed to create symlink %s: %m", t);
360 /* Now that we mounted everything, let's make the tmpfs the
361 * cgroup file systems are mounted into read-only. */
362 (void) mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
367 #if HAVE_SELINUX || ENABLE_SMACK
370 const struct stat *sb,
372 struct FTW *ftwbuf) {
374 /* No need to label /dev twice in a row... */
375 if (_unlikely_(ftwbuf->level == 0))
378 (void) label_fix(fpath, 0);
380 /* /run/initramfs is static data and big, no need to
381 * dynamically relabel its contents at boot... */
382 if (_unlikely_(ftwbuf->level == 1 &&
384 streq(fpath, "/run/initramfs")))
385 return FTW_SKIP_SUBTREE;
390 static int relabel_cgroup_filesystems(void) {
394 r = cg_all_unified();
396 /* Temporarily remount the root cgroup filesystem to give it a proper label. Do this
397 only when the filesystem has been already populated by a previous instance of systemd
398 running from initrd. Otherwise don't remount anything and leave the filesystem read-write
399 for the cgroup filesystems to be mounted inside. */
400 if (statfs("/sys/fs/cgroup", &st) < 0)
401 return log_error_errno(errno, "Failed to determine mount flags for /sys/fs/cgroup: %m");
403 if (st.f_flags & ST_RDONLY)
404 (void) mount(NULL, "/sys/fs/cgroup", NULL, MS_REMOUNT, NULL);
406 (void) label_fix("/sys/fs/cgroup", 0);
407 (void) nftw("/sys/fs/cgroup", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
409 if (st.f_flags & ST_RDONLY)
410 (void) mount(NULL, "/sys/fs/cgroup", NULL, MS_REMOUNT|MS_RDONLY, NULL);
413 return log_error_errno(r, "Failed to determine whether we are in all unified mode: %m");
420 int mount_setup(bool loaded_policy) {
423 r = mount_points_setup(ELEMENTSOF(mount_table), loaded_policy);
427 #if 0 /// elogind does not control /, /dev, /run and /run/systemd/* are setup elsewhere.
428 #if HAVE_SELINUX || ENABLE_SMACK
429 /* Nodes in devtmpfs and /run need to be manually updated for
430 * the appropriate labels, after mounting. The other virtual
431 * API file systems like /sys and /proc do not need that, they
432 * use the same label for all their files. */
434 usec_t before_relabel, after_relabel;
435 char timespan[FORMAT_TIMESPAN_MAX];
437 before_relabel = now(CLOCK_MONOTONIC);
439 (void) nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
440 (void) nftw("/dev/shm", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
441 (void) nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
443 r = relabel_cgroup_filesystems();
447 after_relabel = now(CLOCK_MONOTONIC);
449 log_info("Relabelled /dev, /run and /sys/fs/cgroup in %s.",
450 format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0));
454 /* Create a few default symlinks, which are normally created
455 * by udevd, but some scripts might need them before we start
457 dev_setup(NULL, UID_INVALID, GID_INVALID);
459 /* Mark the root directory as shared in regards to mount propagation. The kernel defaults to "private", but we
460 * think it makes more sense to have a default of "shared" so that nspawn and the container tools work out of
461 * the box. If specific setups need other settings they can reset the propagation mode to private if
462 * needed. Note that we set this only when we are invoked directly by the kernel. If we are invoked by a
463 * container manager we assume the container manager knows what it is doing (for example, because it set up
464 * some directories with different propagation modes). */
465 if (detect_container() <= 0)
466 if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
467 log_warning_errno(errno, "Failed to set up the root directory for shared mount propagation: %m");
470 /* Create a few directories we always want around, Note that sd_booted() checks for /run/systemd/system, so
471 * this mkdir really needs to stay for good, otherwise software that copied sd-daemon.c into their sources will
472 * misdetect systemd. */
473 (void) mkdir_label("/run/systemd", 0755);
474 #if 0 /// Yeah, but elogind is not used with systemd, so this directory would be toxic.
475 (void) mkdir_label("/run/systemd/system", 0755);
478 /* Set up inaccessible (and empty) file nodes of all types */
479 (void) mkdir_label("/run/systemd/inaccessible", 0000);
480 (void) mknod("/run/systemd/inaccessible/reg", S_IFREG | 0000, 0);
481 (void) mkdir_label("/run/systemd/inaccessible/dir", 0000);
482 (void) mkfifo("/run/systemd/inaccessible/fifo", 0000);
483 (void) mknod("/run/systemd/inaccessible/sock", S_IFSOCK | 0000, 0);
485 /* The following two are likely to fail if we lack the privs for it (for example in an userns environment, if
486 * CAP_SYS_MKNOD is missing, or if a device node policy prohibit major/minor of 0 device nodes to be
487 * created). But that's entirely fine. Consumers of these files should carry fallback to use a different node
488 * then, for example /run/systemd/inaccessible/sock, which is close enough in behaviour and semantics for most
490 (void) mknod("/run/systemd/inaccessible/chr", S_IFCHR | 0000, makedev(0, 0));
491 (void) mknod("/run/systemd/inaccessible/blk", S_IFBLK | 0000, makedev(0, 0));