]>
Commit | Line | Data |
---|---|---|
d6c9574f | 1 | /*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/ |
8e274523 LP |
2 | |
3 | /*** | |
4 | This file is part of systemd. | |
5 | ||
6 | Copyright 2010 Lennart Poettering | |
7 | ||
8 | systemd is free software; you can redistribute it and/or modify it | |
5430f7f2 LP |
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 | |
8e274523 LP |
11 | (at your option) any later version. |
12 | ||
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 | |
5430f7f2 | 16 | Lesser General Public License for more details. |
8e274523 | 17 | |
5430f7f2 | 18 | You should have received a copy of the GNU Lesser General Public License |
8e274523 LP |
19 | along with systemd; If not, see <http://www.gnu.org/licenses/>. |
20 | ***/ | |
21 | ||
22 | #include <sys/mount.h> | |
23 | #include <errno.h> | |
8e274523 | 24 | #include <stdlib.h> |
5c0532d1 | 25 | #include <unistd.h> |
1829dc9d | 26 | #include <ftw.h> |
8e274523 LP |
27 | |
28 | #include "mount-setup.h" | |
5ba2dc25 | 29 | #include "dev-setup.h" |
8e274523 | 30 | #include "log.h" |
c9af1080 LP |
31 | #include "macro.h" |
32 | #include "util.h" | |
5275d3c1 | 33 | #include "label.h" |
0c85a4f3 LP |
34 | #include "set.h" |
35 | #include "strv.h" | |
49e942b2 | 36 | #include "mkdir.h" |
9eb977db | 37 | #include "path-util.h" |
48ac500b | 38 | #include "missing.h" |
c481f78b | 39 | #include "virt.h" |
34e5a31e | 40 | #include "efivars.h" |
8552b176 | 41 | #include "smack-util.h" |
b12afc8c | 42 | #include "cgroup-util.h" |
bef2733f | 43 | |
6aa220e0 KS |
44 | typedef enum MountMode { |
45 | MNT_NONE = 0, | |
46 | MNT_FATAL = 1 << 0, | |
47 | MNT_IN_CONTAINER = 1 << 1, | |
48 | } MountMode; | |
49 | ||
ca714c0e LP |
50 | typedef struct MountPoint { |
51 | const char *what; | |
52 | const char *where; | |
53 | const char *type; | |
54 | const char *options; | |
55 | unsigned long flags; | |
6aa220e0 KS |
56 | bool (*condition_fn)(void); |
57 | MountMode mode; | |
ca714c0e LP |
58 | } MountPoint; |
59 | ||
4ef31082 | 60 | /* The first three entries we might need before SELinux is up. The |
160481f6 | 61 | * fourth (securityfs) is needed by IMA to load a custom policy. The |
7c96ab1d LP |
62 | * other ones we can delay until SELinux and IMA are loaded. When |
63 | * SMACK is enabled we need smackfs, too, so it's a fifth one. */ | |
64 | #ifdef HAVE_SMACK | |
ffbd2c4d | 65 | #define N_EARLY_MOUNT 5 |
7c96ab1d LP |
66 | #else |
67 | #define N_EARLY_MOUNT 4 | |
68 | #endif | |
4ef31082 | 69 | |
ca714c0e | 70 | static const MountPoint mount_table[] = { |
68d4c452 LP |
71 | { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, |
72 | NULL, MNT_FATAL|MNT_IN_CONTAINER }, | |
73 | { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, | |
74 | NULL, MNT_FATAL|MNT_IN_CONTAINER }, | |
75 | { "devtmpfs", "/dev", "devtmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME, | |
76 | NULL, MNT_FATAL|MNT_IN_CONTAINER }, | |
77 | { "securityfs", "/sys/kernel/security", "securityfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, | |
78 | NULL, MNT_NONE }, | |
d407c940 | 79 | #ifdef HAVE_SMACK |
68d4c452 LP |
80 | { "smackfs", "/sys/fs/smackfs", "smackfs", "smackfsdef=*", MS_NOSUID|MS_NOEXEC|MS_NODEV, |
81 | mac_smack_use, MNT_FATAL }, | |
82 | { "tmpfs", "/dev/shm", "tmpfs", "mode=1777,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME, | |
83 | mac_smack_use, MNT_FATAL }, | |
d407c940 | 84 | #endif |
68d4c452 LP |
85 | { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME, |
86 | NULL, MNT_FATAL|MNT_IN_CONTAINER }, | |
87 | { "devpts", "/dev/pts", "devpts", "mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC, | |
88 | NULL, MNT_IN_CONTAINER }, | |
d407c940 | 89 | #ifdef HAVE_SMACK |
68d4c452 LP |
90 | { "tmpfs", "/run", "tmpfs", "mode=755,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME, |
91 | mac_smack_use, MNT_FATAL }, | |
d407c940 | 92 | #endif |
68d4c452 LP |
93 | { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME, |
94 | NULL, MNT_FATAL|MNT_IN_CONTAINER }, | |
95 | { "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, | |
96 | NULL, MNT_FATAL|MNT_IN_CONTAINER }, | |
97 | { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV, | |
98 | NULL, MNT_IN_CONTAINER }, | |
99 | { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV, | |
100 | NULL, MNT_FATAL|MNT_IN_CONTAINER }, | |
101 | { "pstore", "/sys/fs/pstore", "pstore", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, | |
102 | NULL, MNT_NONE }, | |
c06bf414 | 103 | #ifdef ENABLE_EFI |
68d4c452 LP |
104 | { "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, |
105 | is_efi_boot, MNT_NONE }, | |
c06bf414 | 106 | #endif |
63cc4c31 DM |
107 | { "kdbusfs", "/sys/fs/kdbus", "kdbusfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, |
108 | NULL, MNT_IN_CONTAINER }, | |
63cc4c31 DM |
109 | }; |
110 | ||
949c6510 | 111 | /* These are API file systems that might be mounted by other software, |
46ff0ed7 | 112 | * we just list them here so that we know that we should ignore them */ |
949c6510 | 113 | |
eaeb18db LP |
114 | static const char ignore_paths[] = |
115 | /* SELinux file systems */ | |
116 | "/sys/fs/selinux\0" | |
eaeb18db LP |
117 | /* Container bind mounts */ |
118 | "/proc/sys\0" | |
119 | "/dev/console\0" | |
c481f78b | 120 | "/proc/kmsg\0"; |
949c6510 | 121 | |
dad08730 LP |
122 | bool mount_point_is_api(const char *path) { |
123 | unsigned i; | |
124 | ||
125 | /* Checks if this mount point is considered "API", and hence | |
126 | * should be ignored */ | |
127 | ||
ca714c0e | 128 | for (i = 0; i < ELEMENTSOF(mount_table); i ++) |
449ddb2d | 129 | if (path_equal(path, mount_table[i].where)) |
dad08730 LP |
130 | return true; |
131 | ||
57f2a956 KS |
132 | return path_startswith(path, "/sys/fs/cgroup/"); |
133 | } | |
134 | ||
135 | bool mount_point_ignore(const char *path) { | |
eaeb18db | 136 | const char *i; |
57f2a956 | 137 | |
eaeb18db LP |
138 | NULSTR_FOREACH(i, ignore_paths) |
139 | if (path_equal(path, i)) | |
949c6510 LP |
140 | return true; |
141 | ||
57f2a956 | 142 | return false; |
dad08730 LP |
143 | } |
144 | ||
4ef31082 | 145 | static int mount_one(const MountPoint *p, bool relabel) { |
8e274523 LP |
146 | int r; |
147 | ||
ca714c0e | 148 | assert(p); |
8e274523 | 149 | |
6aa220e0 KS |
150 | if (p->condition_fn && !p->condition_fn()) |
151 | return 0; | |
152 | ||
51b4af2c | 153 | /* Relabel first, just in case */ |
4ef31082 | 154 | if (relabel) |
c9bc0764 | 155 | label_fix(p->where, true, true); |
51b4af2c | 156 | |
e26d6ce5 | 157 | r = path_is_mount_point(p->where, AT_SYMLINK_FOLLOW); |
64f75d7a | 158 | if (r < 0 && r != -ENOENT) |
8e274523 | 159 | return r; |
8e274523 | 160 | if (r > 0) |
51b4af2c | 161 | return 0; |
8e274523 | 162 | |
c481f78b | 163 | /* Skip securityfs in a container */ |
6aa220e0 | 164 | if (!(p->mode & MNT_IN_CONTAINER) && detect_container(NULL) > 0) |
c481f78b LP |
165 | return 0; |
166 | ||
a04f58d6 LP |
167 | /* The access mode here doesn't really matter too much, since |
168 | * the mounted file system will take precedence anyway. */ | |
c4bfd169 LP |
169 | if (relabel) |
170 | mkdir_p_label(p->where, 0755); | |
171 | else | |
172 | mkdir_p(p->where, 0755); | |
a04f58d6 | 173 | |
8e274523 | 174 | log_debug("Mounting %s to %s of type %s with options %s.", |
ca714c0e LP |
175 | p->what, |
176 | p->where, | |
177 | p->type, | |
178 | strna(p->options)); | |
179 | ||
180 | if (mount(p->what, | |
181 | p->where, | |
182 | p->type, | |
183 | p->flags, | |
184 | p->options) < 0) { | |
99a17ada | 185 | log_full((p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG, "Failed to mount %s at %s: %m", p->type, p->where); |
6aa220e0 | 186 | return (p->mode & MNT_FATAL) ? -errno : 0; |
8e274523 LP |
187 | } |
188 | ||
51b4af2c | 189 | /* Relabel again, since we now mounted something fresh here */ |
4ef31082 | 190 | if (relabel) |
c9bc0764 | 191 | label_fix(p->where, false, false); |
5275d3c1 | 192 | |
0c85a4f3 | 193 | return 1; |
8e274523 LP |
194 | } |
195 | ||
4ef31082 LP |
196 | int mount_setup_early(void) { |
197 | unsigned i; | |
198 | int r = 0; | |
199 | ||
200 | assert_cc(N_EARLY_MOUNT <= ELEMENTSOF(mount_table)); | |
201 | ||
202 | /* Do a minimal mount of /proc and friends to enable the most | |
203 | * basic stuff, such as SELinux */ | |
204 | for (i = 0; i < N_EARLY_MOUNT; i ++) { | |
205 | int j; | |
206 | ||
207 | j = mount_one(mount_table + i, false); | |
208 | if (r == 0) | |
209 | r = j; | |
210 | } | |
211 | ||
212 | return r; | |
213 | } | |
214 | ||
0c85a4f3 | 215 | int mount_cgroup_controllers(char ***join_controllers) { |
a6b26d90 | 216 | _cleanup_set_free_free_ Set *controllers = NULL; |
a641dcd9 | 217 | int r; |
2076ca54 | 218 | |
670802d4 | 219 | /* Mount all available cgroup controllers that are built into the kernel. */ |
2076ca54 | 220 | |
d5099efc | 221 | controllers = set_new(&string_hash_ops); |
a6b26d90 ZJS |
222 | if (!controllers) |
223 | return log_oom(); | |
0c85a4f3 | 224 | |
b12afc8c LP |
225 | r = cg_kernel_controllers(controllers); |
226 | if (r < 0) | |
227 | return log_error_errno(r, "Failed to enumerate cgroup controllers: %m"); | |
0c85a4f3 LP |
228 | |
229 | for (;;) { | |
a641dcd9 | 230 | _cleanup_free_ char *options = NULL, *controller = NULL, *where = NULL; |
a6b26d90 ZJS |
231 | MountPoint p = { |
232 | .what = "cgroup", | |
233 | .type = "cgroup", | |
234 | .flags = MS_NOSUID|MS_NOEXEC|MS_NODEV, | |
235 | .mode = MNT_IN_CONTAINER, | |
236 | }; | |
0c85a4f3 LP |
237 | char ***k = NULL; |
238 | ||
239 | controller = set_steal_first(controllers); | |
240 | if (!controller) | |
241 | break; | |
242 | ||
243 | if (join_controllers) | |
244 | for (k = join_controllers; *k; k++) | |
245 | if (strv_find(*k, controller)) | |
246 | break; | |
247 | ||
248 | if (k && *k) { | |
249 | char **i, **j; | |
250 | ||
251 | for (i = *k, j = *k; *i; i++) { | |
252 | ||
253 | if (!streq(*i, controller)) { | |
a641dcd9 | 254 | _cleanup_free_ char *t; |
0c85a4f3 LP |
255 | |
256 | t = set_remove(controllers, *i); | |
257 | if (!t) { | |
258 | free(*i); | |
259 | continue; | |
260 | } | |
0c85a4f3 LP |
261 | } |
262 | ||
263 | *(j++) = *i; | |
264 | } | |
265 | ||
266 | *j = NULL; | |
267 | ||
268 | options = strv_join(*k, ","); | |
a6b26d90 ZJS |
269 | if (!options) |
270 | return log_oom(); | |
0c85a4f3 LP |
271 | } else { |
272 | options = controller; | |
273 | controller = NULL; | |
274 | } | |
275 | ||
a641dcd9 LP |
276 | where = strappend("/sys/fs/cgroup/", options); |
277 | if (!where) | |
278 | return log_oom(); | |
279 | ||
280 | p.where = where; | |
0c85a4f3 | 281 | p.options = options; |
2076ca54 | 282 | |
4ef31082 | 283 | r = mount_one(&p, true); |
a6b26d90 ZJS |
284 | if (r < 0) |
285 | return r; | |
0c85a4f3 LP |
286 | |
287 | if (r > 0 && k && *k) { | |
288 | char **i; | |
289 | ||
290 | for (i = *k; *i; i++) { | |
a641dcd9 LP |
291 | _cleanup_free_ char *t = NULL; |
292 | ||
293 | t = strappend("/sys/fs/cgroup/", *i); | |
294 | if (!t) | |
295 | return log_oom(); | |
0c85a4f3 LP |
296 | |
297 | r = symlink(options, t); | |
4a62c710 MS |
298 | if (r < 0 && errno != EEXIST) |
299 | return log_error_errno(errno, "Failed to create symlink %s: %m", t); | |
0c85a4f3 LP |
300 | } |
301 | } | |
2076ca54 LP |
302 | } |
303 | ||
679142ce LP |
304 | /* Now that we mounted everything, let's make the tmpfs the |
305 | * cgroup file systems are mounted into read-only. */ | |
b12afc8c | 306 | (void) mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755"); |
679142ce | 307 | |
a6b26d90 | 308 | return 0; |
2076ca54 LP |
309 | } |
310 | ||
0fff82e5 | 311 | #if defined(HAVE_SELINUX) || defined(HAVE_SMACK) |
1829dc9d LP |
312 | static int nftw_cb( |
313 | const char *fpath, | |
314 | const struct stat *sb, | |
315 | int tflag, | |
316 | struct FTW *ftwbuf) { | |
317 | ||
9fe117ea | 318 | /* No need to label /dev twice in a row... */ |
edb49778 LP |
319 | if (_unlikely_(ftwbuf->level == 0)) |
320 | return FTW_CONTINUE; | |
321 | ||
c9bc0764 | 322 | label_fix(fpath, false, false); |
af65c248 | 323 | |
edb49778 | 324 | /* /run/initramfs is static data and big, no need to |
af65c248 | 325 | * dynamically relabel its contents at boot... */ |
edb49778 LP |
326 | if (_unlikely_(ftwbuf->level == 1 && |
327 | tflag == FTW_D && | |
328 | streq(fpath, "/run/initramfs"))) | |
329 | return FTW_SKIP_SUBTREE; | |
9fe117ea | 330 | |
edb49778 | 331 | return FTW_CONTINUE; |
1829dc9d | 332 | }; |
0fff82e5 | 333 | #endif |
1829dc9d | 334 | |
0b3325e7 | 335 | int mount_setup(bool loaded_policy) { |
dad08730 | 336 | unsigned i; |
68d4c452 | 337 | int r = 0; |
8e274523 | 338 | |
4ef31082 | 339 | for (i = 0; i < ELEMENTSOF(mount_table); i ++) { |
68d4c452 | 340 | int j; |
4ef31082 | 341 | |
68d4c452 LP |
342 | j = mount_one(mount_table + i, loaded_policy); |
343 | if (r == 0) | |
344 | r = j; | |
4ef31082 | 345 | } |
8e274523 | 346 | |
68d4c452 LP |
347 | if (r < 0) |
348 | return r; | |
349 | ||
0fff82e5 | 350 | #if defined(HAVE_SELINUX) || defined(HAVE_SMACK) |
f1d19aa4 LP |
351 | /* Nodes in devtmpfs and /run need to be manually updated for |
352 | * the appropriate labels, after mounting. The other virtual | |
353 | * API file systems like /sys and /proc do not need that, they | |
354 | * use the same label for all their files. */ | |
0b3325e7 LP |
355 | if (loaded_policy) { |
356 | usec_t before_relabel, after_relabel; | |
357 | char timespan[FORMAT_TIMESPAN_MAX]; | |
358 | ||
359 | before_relabel = now(CLOCK_MONOTONIC); | |
360 | ||
edb49778 LP |
361 | nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL); |
362 | nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL); | |
0b3325e7 LP |
363 | |
364 | after_relabel = now(CLOCK_MONOTONIC); | |
365 | ||
366 | log_info("Relabelled /dev and /run in %s.", | |
2fa4092c | 367 | format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0)); |
3bbecb2f | 368 | } |
0fff82e5 | 369 | #endif |
1829dc9d | 370 | |
5c0532d1 | 371 | /* Create a few default symlinks, which are normally created |
f1d19aa4 | 372 | * by udevd, but some scripts might need them before we start |
5c0532d1 | 373 | * udevd. */ |
03cfe0d5 | 374 | dev_setup(NULL, UID_INVALID, GID_INVALID); |
5c0532d1 | 375 | |
b3ac5f8c LP |
376 | /* Mark the root directory as shared in regards to mount |
377 | * propagation. The kernel defaults to "private", but we think | |
378 | * it makes more sense to have a default of "shared" so that | |
379 | * nspawn and the container tools work out of the box. If | |
380 | * specific setups need other settings they can reset the | |
381 | * propagation mode to private if needed. */ | |
c481f78b LP |
382 | if (detect_container(NULL) <= 0) |
383 | if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0) | |
56f64d95 | 384 | log_warning_errno(errno, "Failed to set up the root directory for shared mount propagation: %m"); |
b3ac5f8c | 385 | |
66e41181 LP |
386 | /* Create a few directories we always want around, Note that |
387 | * sd_booted() checks for /run/systemd/system, so this mkdir | |
388 | * really needs to stay for good, otherwise software that | |
389 | * copied sd-daemon.c into their sources will misdetect | |
390 | * systemd. */ | |
d2e54fae KS |
391 | mkdir_label("/run/systemd", 0755); |
392 | mkdir_label("/run/systemd/system", 0755); | |
c17ec25e | 393 | mkdir_label("/run/systemd/inaccessible", 0000); |
b925e726 | 394 | |
0c85a4f3 | 395 | return 0; |
8e274523 | 396 | } |