1 /* SPDX-License-Identifier: LGPL-2.1+ */
7 //#include <sys/statvfs.h>
10 #include "alloc-util.h"
11 //#include "bus-util.h"
12 #include "cgroup-util.h"
13 //#include "dev-setup.h"
14 //#include "efivars.h"
20 //#include "missing.h"
22 #include "mount-setup.h"
23 #include "mount-util.h"
24 #include "path-util.h"
26 //#include "smack-util.h"
28 #include "user-util.h"
32 /// Additional includes needed by elogind
33 #include "string-util.h"
35 typedef enum MountMode {
38 MNT_IN_CONTAINER = 1 << 1,
39 MNT_CHECK_WRITABLE = 1 << 2,
42 typedef struct MountPoint {
48 bool (*condition_fn)(void);
52 /* The first three entries we might need before SELinux is up. The
53 * fourth (securityfs) is needed by IMA to load a custom policy. The
54 * other ones we can delay until SELinux and IMA are loaded. When
55 * SMACK is enabled we need smackfs, too, so it's a fifth one. */
57 #define N_EARLY_MOUNT 5
59 #define N_EARLY_MOUNT 4
62 static const MountPoint mount_table[] = {
63 #if 0 /// UNNEEDED by elogind
64 { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
65 NULL, MNT_FATAL|MNT_IN_CONTAINER },
66 { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
67 NULL, MNT_FATAL|MNT_IN_CONTAINER },
68 { "devtmpfs", "/dev", "devtmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME,
69 NULL, MNT_FATAL|MNT_IN_CONTAINER },
70 { "securityfs", "/sys/kernel/security", "securityfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
73 { "smackfs", "/sys/fs/smackfs", "smackfs", "smackfsdef=*", MS_NOSUID|MS_NOEXEC|MS_NODEV,
74 mac_smack_use, MNT_FATAL },
75 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
76 mac_smack_use, MNT_FATAL },
78 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
79 NULL, MNT_FATAL|MNT_IN_CONTAINER },
80 { "devpts", "/dev/pts", "devpts", "mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC,
81 NULL, MNT_IN_CONTAINER },
83 { "tmpfs", "/run", "tmpfs", "mode=755,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
84 mac_smack_use, MNT_FATAL },
86 { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
87 NULL, MNT_FATAL|MNT_IN_CONTAINER },
89 { "cgroup2", "/sys/fs/cgroup", "cgroup2", "nsdelegate", MS_NOSUID|MS_NOEXEC|MS_NODEV,
90 cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
91 { "cgroup2", "/sys/fs/cgroup", "cgroup2", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
92 cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
93 { "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
94 cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
95 { "cgroup2", "/sys/fs/cgroup/unified", "cgroup2", "nsdelegate", MS_NOSUID|MS_NOEXEC|MS_NODEV,
96 cg_is_hybrid_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
97 { "cgroup2", "/sys/fs/cgroup/unified", "cgroup2", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
98 cg_is_hybrid_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
99 #if 0 /// UNNEEDED by elogind
100 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
101 cg_is_legacy_wanted, MNT_IN_CONTAINER },
102 { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV,
103 cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
104 { "pstore", "/sys/fs/pstore", "pstore", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
107 { "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
108 is_efi_boot, MNT_NONE },
110 { "bpf", "/sys/fs/bpf", "bpf", "mode=700", MS_NOSUID|MS_NOEXEC|MS_NODEV,
113 { "cgroup", "/sys/fs/cgroup/elogind", "cgroup", "none,name=elogind,release_agent="SYSTEMD_CGROUP_AGENT_PATH",xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
114 cg_is_legacy_wanted, MNT_IN_CONTAINER },
115 { "cgroup", "/sys/fs/cgroup/elogind", "cgroup", "none,name=elogind,release_agent="SYSTEMD_CGROUP_AGENT_PATH, MS_NOSUID|MS_NOEXEC|MS_NODEV,
116 cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
120 #if 0 /// UNNEEDED by elogind
121 /* These are API file systems that might be mounted by other software,
122 * we just list them here so that we know that we should ignore them */
124 static const char ignore_paths[] =
125 /* SELinux file systems */
127 /* Container bind mounts */
132 bool mount_point_is_api(const char *path) {
135 /* Checks if this mount point is considered "API", and hence
136 * should be ignored */
138 for (i = 0; i < ELEMENTSOF(mount_table); i ++)
139 if (path_equal(path, mount_table[i].where))
142 return path_startswith(path, "/sys/fs/cgroup/");
145 bool mount_point_ignore(const char *path) {
148 NULSTR_FOREACH(i, ignore_paths)
149 if (path_equal(path, i))
156 static int mount_one(const MountPoint *p, bool relabel) {
161 priority = (p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG;
163 if (p->condition_fn && !p->condition_fn())
166 /* Relabel first, just in case */
168 (void) label_fix(p->where, LABEL_IGNORE_ENOENT|LABEL_IGNORE_EROFS);
170 r = path_is_mount_point(p->where, NULL, AT_SYMLINK_FOLLOW);
171 if (r < 0 && r != -ENOENT) {
172 log_full_errno(priority, r, "Failed to determine whether %s is a mount point: %m", p->where);
173 return (p->mode & MNT_FATAL) ? r : 0;
178 /* Skip securityfs in a container */
179 if (!(p->mode & MNT_IN_CONTAINER) && detect_container() > 0)
182 /* The access mode here doesn't really matter too much, since
183 * the mounted file system will take precedence anyway. */
185 (void) mkdir_p_label(p->where, 0755);
187 (void) mkdir_p(p->where, 0755);
189 log_debug("Mounting %s to %s of type %s with options %s.",
200 log_full_errno(priority, errno, "Failed to mount %s at %s: %m", p->type, p->where);
201 return (p->mode & MNT_FATAL) ? -errno : 0;
204 /* Relabel again, since we now mounted something fresh here */
206 (void) label_fix(p->where, 0);
208 if (p->mode & MNT_CHECK_WRITABLE) {
209 if (access(p->where, W_OK) < 0) {
212 (void) umount(p->where);
213 (void) rmdir(p->where);
215 log_full_errno(priority, r, "Mount point %s not writable after mounting: %m", p->where);
216 return (p->mode & MNT_FATAL) ? r : 0;
223 static int mount_points_setup(unsigned n, bool loaded_policy) {
227 for (i = 0; i < n; i ++) {
230 j = mount_one(mount_table + i, loaded_policy);
231 if (j != 0 && r >= 0)
238 #if 0 /// UNNEEDED by elogind
239 int mount_setup_early(void) {
240 assert_cc(N_EARLY_MOUNT <= ELEMENTSOF(mount_table));
242 /* Do a minimal mount of /proc and friends to enable the most
243 * basic stuff, such as SELinux */
244 return mount_points_setup(N_EARLY_MOUNT, false);
247 int mount_cgroup_controllers(char ***join_controllers) {
248 _cleanup_set_free_free_ Set *controllers = NULL;
249 bool has_argument = !!join_controllers;
252 if (!cg_is_legacy_wanted())
255 /* Mount all available cgroup controllers that are built into the kernel. */
259 * mount "cpu" + "cpuacct" together, and "net_cls" + "net_prio".
261 * We'd like to add "cpuset" to the mix, but "cpuset" doesn't really
262 * work for groups with no initialized attributes.
264 join_controllers = (char**[]) {
265 STRV_MAKE("cpu", "cpuacct"),
266 STRV_MAKE("net_cls", "net_prio"),
270 r = cg_kernel_controllers(&controllers);
272 return log_error_errno(r, "Failed to enumerate cgroup controllers: %m");
275 _cleanup_free_ char *options = NULL, *controller = NULL, *where = NULL;
279 .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV,
280 .mode = MNT_IN_CONTAINER,
284 controller = set_steal_first(controllers);
288 for (k = join_controllers; *k; k++)
289 if (strv_find(*k, controller))
295 for (i = *k, j = *k; *i; i++) {
297 if (!streq(*i, controller)) {
298 _cleanup_free_ char *t;
300 t = set_remove(controllers, *i);
313 options = strv_join(*k, ",");
317 options = TAKE_PTR(controller);
319 where = strappend("/sys/fs/cgroup/", options);
326 r = mount_one(&p, true);
333 for (i = *k; *i; i++) {
334 _cleanup_free_ char *t = NULL;
336 t = strappend("/sys/fs/cgroup/", *i);
340 r = symlink(options, t);
342 #ifdef SMACK_RUN_LABEL
343 _cleanup_free_ char *src;
344 src = strappend("/sys/fs/cgroup/", options);
347 r = mac_smack_copy(t, src);
348 if (r < 0 && r != -EOPNOTSUPP)
349 return log_error_errno(r, "Failed to copy smack label from %s to %s: %m", src, t);
351 } else if (errno != EEXIST)
352 return log_error_errno(errno, "Failed to create symlink %s: %m", t);
357 /* Now that we mounted everything, let's make the tmpfs the
358 * cgroup file systems are mounted into read-only. */
359 (void) mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
364 #if HAVE_SELINUX || ENABLE_SMACK
367 const struct stat *sb,
369 struct FTW *ftwbuf) {
371 /* No need to label /dev twice in a row... */
372 if (_unlikely_(ftwbuf->level == 0))
375 (void) label_fix(fpath, 0);
377 /* /run/initramfs is static data and big, no need to
378 * dynamically relabel its contents at boot... */
379 if (_unlikely_(ftwbuf->level == 1 &&
381 streq(fpath, "/run/initramfs")))
382 return FTW_SKIP_SUBTREE;
387 static int relabel_cgroup_filesystems(void) {
391 r = cg_all_unified();
393 /* Temporarily remount the root cgroup filesystem to give it a proper label. Do this
394 only when the filesystem has been already populated by a previous instance of systemd
395 running from initrd. Otherwise don't remount anything and leave the filesystem read-write
396 for the cgroup filesystems to be mounted inside. */
397 if (statfs("/sys/fs/cgroup", &st) < 0)
398 return log_error_errno(errno, "Failed to determine mount flags for /sys/fs/cgroup: %m");
400 if (st.f_flags & ST_RDONLY)
401 (void) mount(NULL, "/sys/fs/cgroup", NULL, MS_REMOUNT, NULL);
403 (void) label_fix("/sys/fs/cgroup", 0);
404 (void) nftw("/sys/fs/cgroup", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
406 if (st.f_flags & ST_RDONLY)
407 (void) mount(NULL, "/sys/fs/cgroup", NULL, MS_REMOUNT|MS_RDONLY, NULL);
410 return log_error_errno(r, "Failed to determine whether we are in all unified mode: %m");
417 int mount_setup(bool loaded_policy) {
420 r = mount_points_setup(ELEMENTSOF(mount_table), loaded_policy);
424 #if 0 /// elogind does not control /, /dev, /run and /run/systemd/* are setup elsewhere.
425 #if HAVE_SELINUX || ENABLE_SMACK
426 /* Nodes in devtmpfs and /run need to be manually updated for
427 * the appropriate labels, after mounting. The other virtual
428 * API file systems like /sys and /proc do not need that, they
429 * use the same label for all their files. */
431 usec_t before_relabel, after_relabel;
432 char timespan[FORMAT_TIMESPAN_MAX];
434 before_relabel = now(CLOCK_MONOTONIC);
436 (void) nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
437 (void) nftw("/dev/shm", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
438 (void) nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
440 r = relabel_cgroup_filesystems();
444 after_relabel = now(CLOCK_MONOTONIC);
446 log_info("Relabelled /dev, /run and /sys/fs/cgroup in %s.",
447 format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0));
451 /* Create a few default symlinks, which are normally created
452 * by udevd, but some scripts might need them before we start
454 dev_setup(NULL, UID_INVALID, GID_INVALID);
456 /* Mark the root directory as shared in regards to mount propagation. The kernel defaults to "private", but we
457 * think it makes more sense to have a default of "shared" so that nspawn and the container tools work out of
458 * the box. If specific setups need other settings they can reset the propagation mode to private if
459 * needed. Note that we set this only when we are invoked directly by the kernel. If we are invoked by a
460 * container manager we assume the container manager knows what it is doing (for example, because it set up
461 * some directories with different propagation modes). */
462 if (detect_container() <= 0)
463 if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
464 log_warning_errno(errno, "Failed to set up the root directory for shared mount propagation: %m");
467 /* Create a few directories we always want around, Note that sd_booted() checks for /run/systemd/system, so
468 * this mkdir really needs to stay for good, otherwise software that copied sd-daemon.c into their sources will
469 * misdetect systemd. */
470 (void) mkdir_label("/run/systemd", 0755);
471 #if 0 /// Yeah, but elogind is not used with systemd, so this directory would be toxic.
472 (void) mkdir_label("/run/systemd/system", 0755);
475 /* Set up inaccessible (and empty) file nodes of all types */
476 (void) mkdir_label("/run/systemd/inaccessible", 0000);
477 (void) mknod("/run/systemd/inaccessible/reg", S_IFREG | 0000, 0);
478 (void) mkdir_label("/run/systemd/inaccessible/dir", 0000);
479 (void) mkfifo("/run/systemd/inaccessible/fifo", 0000);
480 (void) mknod("/run/systemd/inaccessible/sock", S_IFSOCK | 0000, 0);
482 /* The following two are likely to fail if we lack the privs for it (for example in an userns environment, if
483 * CAP_SYS_MKNOD is missing, or if a device node policy prohibit major/minor of 0 device nodes to be
484 * created). But that's entirely fine. Consumers of these files should carry fallback to use a different node
485 * then, for example /run/systemd/inaccessible/sock, which is close enough in behaviour and semantics for most
487 (void) mknod("/run/systemd/inaccessible/chr", S_IFCHR | 0000, makedev(0, 0));
488 (void) mknod("/run/systemd/inaccessible/blk", S_IFBLK | 0000, makedev(0, 0));