1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2010 Lennart Poettering
11 #include <sys/mount.h>
12 //#include <sys/statvfs.h>
15 #include "alloc-util.h"
16 //#include "bus-util.h"
17 #include "cgroup-util.h"
18 //#include "dev-setup.h"
19 //#include "efivars.h"
25 //#include "missing.h"
27 #include "mount-setup.h"
28 #include "mount-util.h"
29 #include "path-util.h"
31 //#include "smack-util.h"
33 #include "user-util.h"
37 /// Additional includes needed by elogind
38 #include "string-util.h"
40 typedef enum MountMode {
43 MNT_IN_CONTAINER = 1 << 1,
44 MNT_CHECK_WRITABLE = 1 << 2,
47 typedef struct MountPoint {
53 bool (*condition_fn)(void);
57 /* The first three entries we might need before SELinux is up. The
58 * fourth (securityfs) is needed by IMA to load a custom policy. The
59 * other ones we can delay until SELinux and IMA are loaded. When
60 * SMACK is enabled we need smackfs, too, so it's a fifth one. */
62 #define N_EARLY_MOUNT 5
64 #define N_EARLY_MOUNT 4
67 static const MountPoint mount_table[] = {
68 #if 0 /// UNNEEDED by elogind
69 { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
70 NULL, MNT_FATAL|MNT_IN_CONTAINER },
71 { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
72 NULL, MNT_FATAL|MNT_IN_CONTAINER },
73 { "devtmpfs", "/dev", "devtmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME,
74 NULL, MNT_FATAL|MNT_IN_CONTAINER },
75 { "securityfs", "/sys/kernel/security", "securityfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
78 { "smackfs", "/sys/fs/smackfs", "smackfs", "smackfsdef=*", MS_NOSUID|MS_NOEXEC|MS_NODEV,
79 mac_smack_use, MNT_FATAL },
80 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
81 mac_smack_use, MNT_FATAL },
83 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
84 NULL, MNT_FATAL|MNT_IN_CONTAINER },
85 { "devpts", "/dev/pts", "devpts", "mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC,
86 NULL, MNT_IN_CONTAINER },
88 { "tmpfs", "/run", "tmpfs", "mode=755,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
89 mac_smack_use, MNT_FATAL },
91 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
92 NULL, MNT_FATAL|MNT_IN_CONTAINER },
94 { "cgroup2", "/sys/fs/cgroup", "cgroup2", "nsdelegate", MS_NOSUID|MS_NOEXEC|MS_NODEV,
95 cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
96 { "cgroup2", "/sys/fs/cgroup", "cgroup2", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
97 cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
98 { "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
99 cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
100 { "cgroup2", "/sys/fs/cgroup/unified", "cgroup2", "nsdelegate", MS_NOSUID|MS_NOEXEC|MS_NODEV,
101 cg_is_hybrid_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
102 { "cgroup2", "/sys/fs/cgroup/unified", "cgroup2", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
103 cg_is_hybrid_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
104 #if 0 /// 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 },
109 { "pstore", "/sys/fs/pstore", "pstore", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
112 { "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
113 is_efi_boot, MNT_NONE },
115 { "bpf", "/sys/fs/bpf", "bpf", "mode=700", MS_NOSUID|MS_NOEXEC|MS_NODEV,
118 { "cgroup", "/sys/fs/cgroup/elogind", "cgroup", "none,name=elogind,release_agent="SYSTEMD_CGROUP_AGENT_PATH",xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
119 cg_is_legacy_wanted, MNT_IN_CONTAINER },
120 { "cgroup", "/sys/fs/cgroup/elogind", "cgroup", "none,name=elogind,release_agent="SYSTEMD_CGROUP_AGENT_PATH, MS_NOSUID|MS_NOEXEC|MS_NODEV,
121 cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
125 #if 0 /// UNNEEDED by elogind
126 /* These are API file systems that might be mounted by other software,
127 * we just list them here so that we know that we should ignore them */
129 static const char ignore_paths[] =
130 /* SELinux file systems */
132 /* Container bind mounts */
137 bool mount_point_is_api(const char *path) {
140 /* Checks if this mount point is considered "API", and hence
141 * should be ignored */
143 for (i = 0; i < ELEMENTSOF(mount_table); i ++)
144 if (path_equal(path, mount_table[i].where))
147 return path_startswith(path, "/sys/fs/cgroup/");
150 bool mount_point_ignore(const char *path) {
153 NULSTR_FOREACH(i, ignore_paths)
154 if (path_equal(path, i))
161 static int mount_one(const MountPoint *p, bool relabel) {
166 priority = (p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG;
168 if (p->condition_fn && !p->condition_fn())
171 /* Relabel first, just in case */
173 (void) label_fix(p->where, LABEL_IGNORE_ENOENT|LABEL_IGNORE_EROFS);
175 r = path_is_mount_point(p->where, NULL, AT_SYMLINK_FOLLOW);
176 if (r < 0 && r != -ENOENT) {
177 log_full_errno(priority, r, "Failed to determine whether %s is a mount point: %m", p->where);
178 return (p->mode & MNT_FATAL) ? r : 0;
183 /* Skip securityfs in a container */
184 if (!(p->mode & MNT_IN_CONTAINER) && detect_container() > 0)
187 /* The access mode here doesn't really matter too much, since
188 * the mounted file system will take precedence anyway. */
190 (void) mkdir_p_label(p->where, 0755);
192 (void) mkdir_p(p->where, 0755);
194 log_debug("Mounting %s to %s of type %s with options %s.",
205 log_full_errno(priority, errno, "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 (void) label_fix(p->where, 0);
213 if (p->mode & MNT_CHECK_WRITABLE) {
214 if (access(p->where, W_OK) < 0) {
217 (void) umount(p->where);
218 (void) rmdir(p->where);
220 log_full_errno(priority, r, "Mount point %s not writable after mounting: %m", p->where);
221 return (p->mode & MNT_FATAL) ? r : 0;
228 static int mount_points_setup(unsigned n, bool loaded_policy) {
232 for (i = 0; i < n; i ++) {
235 j = mount_one(mount_table + i, loaded_policy);
236 if (j != 0 && r >= 0)
243 #if 0 /// UNNEEDED by elogind
244 int mount_setup_early(void) {
245 assert_cc(N_EARLY_MOUNT <= ELEMENTSOF(mount_table));
247 /* Do a minimal mount of /proc and friends to enable the most
248 * basic stuff, such as SELinux */
249 return mount_points_setup(N_EARLY_MOUNT, false);
252 int mount_cgroup_controllers(char ***join_controllers) {
253 _cleanup_set_free_free_ Set *controllers = NULL;
254 bool has_argument = !!join_controllers;
257 if (!cg_is_legacy_wanted())
260 /* Mount all available cgroup controllers that are built into the kernel. */
264 * mount "cpu" + "cpuacct" together, and "net_cls" + "net_prio".
266 * We'd like to add "cpuset" to the mix, but "cpuset" doesn't really
267 * work for groups with no initialized attributes.
269 join_controllers = (char**[]) {
270 STRV_MAKE("cpu", "cpuacct"),
271 STRV_MAKE("net_cls", "net_prio"),
275 r = cg_kernel_controllers(&controllers);
277 return log_error_errno(r, "Failed to enumerate cgroup controllers: %m");
280 _cleanup_free_ char *options = NULL, *controller = NULL, *where = NULL;
284 .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV,
285 .mode = MNT_IN_CONTAINER,
289 controller = set_steal_first(controllers);
293 for (k = join_controllers; *k; k++)
294 if (strv_find(*k, controller))
300 for (i = *k, j = *k; *i; i++) {
302 if (!streq(*i, controller)) {
303 _cleanup_free_ char *t;
305 t = set_remove(controllers, *i);
318 options = strv_join(*k, ",");
322 options = TAKE_PTR(controller);
324 where = strappend("/sys/fs/cgroup/", options);
331 r = mount_one(&p, true);
338 for (i = *k; *i; i++) {
339 _cleanup_free_ char *t = NULL;
341 t = strappend("/sys/fs/cgroup/", *i);
345 r = symlink(options, t);
347 #ifdef SMACK_RUN_LABEL
348 _cleanup_free_ char *src;
349 src = strappend("/sys/fs/cgroup/", options);
352 r = mac_smack_copy(t, src);
353 if (r < 0 && r != -EOPNOTSUPP)
354 return log_error_errno(r, "Failed to copy smack label from %s to %s: %m", src, t);
356 } else if (errno != EEXIST)
357 return log_error_errno(errno, "Failed to create symlink %s: %m", t);
362 /* Now that we mounted everything, let's make the tmpfs the
363 * cgroup file systems are mounted into read-only. */
364 (void) mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
369 #if HAVE_SELINUX || ENABLE_SMACK
372 const struct stat *sb,
374 struct FTW *ftwbuf) {
376 /* No need to label /dev twice in a row... */
377 if (_unlikely_(ftwbuf->level == 0))
380 (void) label_fix(fpath, 0);
382 /* /run/initramfs is static data and big, no need to
383 * dynamically relabel its contents at boot... */
384 if (_unlikely_(ftwbuf->level == 1 &&
386 streq(fpath, "/run/initramfs")))
387 return FTW_SKIP_SUBTREE;
392 static int relabel_cgroup_filesystems(void) {
396 r = cg_all_unified();
398 /* Temporarily remount the root cgroup filesystem to give it a proper label. Do this
399 only when the filesystem has been already populated by a previous instance of systemd
400 running from initrd. Otherwise don't remount anything and leave the filesystem read-write
401 for the cgroup filesystems to be mounted inside. */
402 if (statfs("/sys/fs/cgroup", &st) < 0)
403 return log_error_errno(errno, "Failed to determine mount flags for /sys/fs/cgroup: %m");
405 if (st.f_flags & ST_RDONLY)
406 (void) mount(NULL, "/sys/fs/cgroup", NULL, MS_REMOUNT, NULL);
408 (void) label_fix("/sys/fs/cgroup", 0);
409 (void) nftw("/sys/fs/cgroup", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
411 if (st.f_flags & ST_RDONLY)
412 (void) mount(NULL, "/sys/fs/cgroup", NULL, MS_REMOUNT|MS_RDONLY, NULL);
415 return log_error_errno(r, "Failed to determine whether we are in all unified mode: %m");
422 int mount_setup(bool loaded_policy) {
425 r = mount_points_setup(ELEMENTSOF(mount_table), loaded_policy);
429 #if 0 /// elogind does not control /, /dev, /run and /run/systemd/* are setup elsewhere.
430 #if HAVE_SELINUX || ENABLE_SMACK
431 /* Nodes in devtmpfs and /run need to be manually updated for
432 * the appropriate labels, after mounting. The other virtual
433 * API file systems like /sys and /proc do not need that, they
434 * use the same label for all their files. */
436 usec_t before_relabel, after_relabel;
437 char timespan[FORMAT_TIMESPAN_MAX];
439 before_relabel = now(CLOCK_MONOTONIC);
441 (void) nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
442 (void) nftw("/dev/shm", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
443 (void) nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
445 r = relabel_cgroup_filesystems();
449 after_relabel = now(CLOCK_MONOTONIC);
451 log_info("Relabelled /dev, /run and /sys/fs/cgroup in %s.",
452 format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0));
456 /* Create a few default symlinks, which are normally created
457 * by udevd, but some scripts might need them before we start
459 dev_setup(NULL, UID_INVALID, GID_INVALID);
461 /* Mark the root directory as shared in regards to mount propagation. The kernel defaults to "private", but we
462 * think it makes more sense to have a default of "shared" so that nspawn and the container tools work out of
463 * the box. If specific setups need other settings they can reset the propagation mode to private if
464 * needed. Note that we set this only when we are invoked directly by the kernel. If we are invoked by a
465 * container manager we assume the container manager knows what it is doing (for example, because it set up
466 * some directories with different propagation modes). */
467 if (detect_container() <= 0)
468 if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
469 log_warning_errno(errno, "Failed to set up the root directory for shared mount propagation: %m");
472 /* Create a few directories we always want around, Note that sd_booted() checks for /run/systemd/system, so
473 * this mkdir really needs to stay for good, otherwise software that copied sd-daemon.c into their sources will
474 * misdetect systemd. */
475 (void) mkdir_label("/run/systemd", 0755);
476 #if 0 /// Yeah, but elogind is not used with systemd, so this directory would be toxic.
477 (void) mkdir_label("/run/systemd/system", 0755);
480 /* Set up inaccessible (and empty) file nodes of all types */
481 (void) mkdir_label("/run/systemd/inaccessible", 0000);
482 (void) mknod("/run/systemd/inaccessible/reg", S_IFREG | 0000, 0);
483 (void) mkdir_label("/run/systemd/inaccessible/dir", 0000);
484 (void) mkfifo("/run/systemd/inaccessible/fifo", 0000);
485 (void) mknod("/run/systemd/inaccessible/sock", S_IFSOCK | 0000, 0);
487 /* The following two are likely to fail if we lack the privs for it (for example in an userns environment, if
488 * CAP_SYS_MKNOD is missing, or if a device node policy prohibit major/minor of 0 device nodes to be
489 * created). But that's entirely fine. Consumers of these files should carry fallback to use a different node
490 * then, for example /run/systemd/inaccessible/sock, which is close enough in behaviour and semantics for most
492 (void) mknod("/run/systemd/inaccessible/chr", S_IFCHR | 0000, makedev(0, 0));
493 (void) mknod("/run/systemd/inaccessible/blk", S_IFBLK | 0000, makedev(0, 0));