]>
Commit | Line | Data |
---|---|---|
db9ecf05 | 1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
4349cd7c | 2 | |
11c3a366 TA |
3 | #include <errno.h> |
4 | #include <stdlib.h> | |
4349cd7c | 5 | #include <sys/mount.h> |
11c3a366 | 6 | #include <sys/stat.h> |
4349cd7c | 7 | #include <sys/statvfs.h> |
11c3a366 | 8 | #include <unistd.h> |
35fd3558 LP |
9 | #include <linux/loop.h> |
10 | #include <linux/fs.h> | |
4349cd7c | 11 | |
b5efdb8a | 12 | #include "alloc-util.h" |
f4351959 | 13 | #include "chase-symlinks.h" |
70599967 | 14 | #include "dissect-image.h" |
45a68ed3 | 15 | #include "exec-util.h" |
9e7f941a | 16 | #include "extract-word.h" |
4349cd7c LP |
17 | #include "fd-util.h" |
18 | #include "fileio.h" | |
e1873695 | 19 | #include "fs-util.h" |
e2341b6b | 20 | #include "glyph-util.h" |
93cc7779 | 21 | #include "hashmap.h" |
9c653536 | 22 | #include "label.h" |
13dcfe46 | 23 | #include "libmount-util.h" |
1c092b62 | 24 | #include "missing_mount.h" |
35fd3558 | 25 | #include "missing_syscall.h" |
35cd0ba5 | 26 | #include "mkdir-label.h" |
4349cd7c | 27 | #include "mount-util.h" |
049af8ad | 28 | #include "mountpoint-util.h" |
2338a175 | 29 | #include "namespace-util.h" |
4349cd7c LP |
30 | #include "parse-util.h" |
31 | #include "path-util.h" | |
6af52c3a | 32 | #include "process-util.h" |
4349cd7c | 33 | #include "set.h" |
28126409 | 34 | #include "stat-util.h" |
15a5e950 | 35 | #include "stdio-util.h" |
4349cd7c | 36 | #include "string-util.h" |
6b7c9f8b | 37 | #include "strv.h" |
6af52c3a | 38 | #include "tmpfile-util.h" |
70599967 | 39 | #include "user-util.h" |
4349cd7c | 40 | |
28126409 LP |
41 | int mount_fd(const char *source, |
42 | int target_fd, | |
43 | const char *filesystemtype, | |
44 | unsigned long mountflags, | |
45 | const void *data) { | |
46 | ||
ddb6eeaf | 47 | if (mount(source, FORMAT_PROC_FD_PATH(target_fd), filesystemtype, mountflags, data) < 0) { |
28126409 LP |
48 | if (errno != ENOENT) |
49 | return -errno; | |
50 | ||
51 | /* ENOENT can mean two things: either that the source is missing, or that /proc/ isn't | |
52 | * mounted. Check for the latter to generate better error messages. */ | |
53 | if (proc_mounted() == 0) | |
54 | return -ENOSYS; | |
55 | ||
56 | return -ENOENT; | |
57 | } | |
58 | ||
59 | return 0; | |
60 | } | |
61 | ||
62 | int mount_nofollow( | |
63 | const char *source, | |
64 | const char *target, | |
65 | const char *filesystemtype, | |
66 | unsigned long mountflags, | |
67 | const void *data) { | |
68 | ||
69 | _cleanup_close_ int fd = -1; | |
70 | ||
71 | /* In almost all cases we want to manipulate the mount table without following symlinks, hence | |
72 | * mount_nofollow() is usually the way to go. The only exceptions are environments where /proc/ is | |
73 | * not available yet, since we need /proc/self/fd/ for this logic to work. i.e. during the early | |
74 | * initialization of namespacing/container stuff where /proc is not yet mounted (and maybe even the | |
75 | * fs to mount) we can only use traditional mount() directly. | |
76 | * | |
77 | * Note that this disables following only for the final component of the target, i.e symlinks within | |
78 | * the path of the target are honoured, as are symlinks in the source path everywhere. */ | |
79 | ||
80 | fd = open(target, O_PATH|O_CLOEXEC|O_NOFOLLOW); | |
81 | if (fd < 0) | |
82 | return -errno; | |
83 | ||
84 | return mount_fd(source, fd, filesystemtype, mountflags, data); | |
85 | } | |
86 | ||
4349cd7c | 87 | int umount_recursive(const char *prefix, int flags) { |
4349cd7c | 88 | int n = 0, r; |
f8b1904f | 89 | bool again; |
4349cd7c | 90 | |
9d0619de LP |
91 | /* Try to umount everything recursively below a directory. Also, take care of stacked mounts, and |
92 | * keep unmounting them until they are gone. */ | |
4349cd7c LP |
93 | |
94 | do { | |
13dcfe46 ZJS |
95 | _cleanup_(mnt_free_tablep) struct libmnt_table *table = NULL; |
96 | _cleanup_(mnt_free_iterp) struct libmnt_iter *iter = NULL; | |
4349cd7c LP |
97 | |
98 | again = false; | |
4349cd7c | 99 | |
2f2d81d9 | 100 | r = libmount_parse("/proc/self/mountinfo", NULL, &table, &iter); |
fdeea3f4 | 101 | if (r < 0) |
13dcfe46 | 102 | return log_debug_errno(r, "Failed to parse /proc/self/mountinfo: %m"); |
35bbbf85 | 103 | |
4349cd7c | 104 | for (;;) { |
13dcfe46 ZJS |
105 | struct libmnt_fs *fs; |
106 | const char *path; | |
4349cd7c | 107 | |
13dcfe46 ZJS |
108 | r = mnt_table_next_fs(table, iter, &fs); |
109 | if (r == 1) | |
110 | break; | |
111 | if (r < 0) | |
112 | return log_debug_errno(r, "Failed to get next entry from /proc/self/mountinfo: %m"); | |
4349cd7c | 113 | |
13dcfe46 ZJS |
114 | path = mnt_fs_get_target(fs); |
115 | if (!path) | |
116 | continue; | |
4349cd7c | 117 | |
13dcfe46 | 118 | if (!path_startswith(path, prefix)) |
4349cd7c LP |
119 | continue; |
120 | ||
827ea521 LP |
121 | if (umount2(path, flags | UMOUNT_NOFOLLOW) < 0) { |
122 | log_debug_errno(errno, "Failed to umount %s, ignoring: %m", path); | |
4349cd7c LP |
123 | continue; |
124 | } | |
125 | ||
13dcfe46 | 126 | log_debug("Successfully unmounted %s", path); |
6b7c9f8b | 127 | |
4349cd7c LP |
128 | again = true; |
129 | n++; | |
130 | ||
131 | break; | |
132 | } | |
4349cd7c LP |
133 | } while (again); |
134 | ||
13dcfe46 | 135 | return n; |
4349cd7c LP |
136 | } |
137 | ||
4f5644db LP |
138 | #define MS_CONVERTIBLE_FLAGS (MS_RDONLY|MS_NOSUID|MS_NODEV|MS_NOEXEC|MS_NOSYMFOLLOW) |
139 | ||
140 | static uint64_t ms_flags_to_mount_attr(unsigned long a) { | |
141 | uint64_t f = 0; | |
142 | ||
143 | if (FLAGS_SET(a, MS_RDONLY)) | |
144 | f |= MOUNT_ATTR_RDONLY; | |
145 | ||
146 | if (FLAGS_SET(a, MS_NOSUID)) | |
147 | f |= MOUNT_ATTR_NOSUID; | |
148 | ||
149 | if (FLAGS_SET(a, MS_NODEV)) | |
150 | f |= MOUNT_ATTR_NODEV; | |
151 | ||
152 | if (FLAGS_SET(a, MS_NOEXEC)) | |
153 | f |= MOUNT_ATTR_NOEXEC; | |
154 | ||
155 | if (FLAGS_SET(a, MS_NOSYMFOLLOW)) | |
156 | f |= MOUNT_ATTR_NOSYMFOLLOW; | |
157 | ||
158 | return f; | |
159 | } | |
160 | ||
161 | static bool skip_mount_set_attr = false; | |
162 | ||
be3f3752 | 163 | /* Use this function only if you do not have direct access to /proc/self/mountinfo but the caller can open it |
64e82c19 LP |
164 | * for you. This is the case when /proc is masked or not mounted. Otherwise, use bind_remount_recursive. */ |
165 | int bind_remount_recursive_with_mountinfo( | |
166 | const char *prefix, | |
167 | unsigned long new_flags, | |
168 | unsigned long flags_mask, | |
6b000af4 | 169 | char **deny_list, |
64e82c19 LP |
170 | FILE *proc_self_mountinfo) { |
171 | ||
0289948e | 172 | _cleanup_fclose_ FILE *proc_self_mountinfo_opened = NULL; |
ba8dced2 | 173 | _cleanup_set_free_ Set *done = NULL; |
670e8efd | 174 | unsigned n_tries = 0; |
4349cd7c LP |
175 | int r; |
176 | ||
8403219f | 177 | assert(prefix); |
ac9de0b3 | 178 | |
874052c5 LP |
179 | if ((flags_mask & ~MS_CONVERTIBLE_FLAGS) == 0 && strv_isempty(deny_list) && !skip_mount_set_attr) { |
180 | /* Let's take a shortcut for all the flags we know how to convert into mount_setattr() flags */ | |
181 | ||
182 | if (mount_setattr(AT_FDCWD, prefix, AT_SYMLINK_NOFOLLOW|AT_RECURSIVE, | |
183 | &(struct mount_attr) { | |
184 | .attr_set = ms_flags_to_mount_attr(new_flags & flags_mask), | |
185 | .attr_clr = ms_flags_to_mount_attr(~new_flags & flags_mask), | |
186 | }, MOUNT_ATTR_SIZE_VER0) < 0) { | |
187 | ||
188 | log_debug_errno(errno, "mount_setattr() failed, falling back to classic remounting: %m"); | |
189 | ||
190 | /* We fall through to classic behaviour if not supported (i.e. kernel < 5.12). We | |
191 | * also do this for all other kinds of errors since they are so many different, and | |
192 | * mount_setattr() has no graceful mode where it continues despite seeing errors one | |
193 | * some mounts, but we want that. Moreover mount_setattr() only works on the mount | |
194 | * point inode itself, not a non-mount point inode, and we want to support arbitrary | |
195 | * prefixes here. */ | |
196 | ||
197 | if (ERRNO_IS_NOT_SUPPORTED(errno)) /* if not supported, then don't bother at all anymore */ | |
198 | skip_mount_set_attr = true; | |
199 | } else | |
200 | return 0; /* Nice, this worked! */ | |
201 | } | |
202 | ||
0289948e LP |
203 | if (!proc_self_mountinfo) { |
204 | r = fopen_unlocked("/proc/self/mountinfo", "re", &proc_self_mountinfo_opened); | |
205 | if (r < 0) | |
206 | return r; | |
207 | ||
208 | proc_self_mountinfo = proc_self_mountinfo_opened; | |
209 | } | |
210 | ||
ddc155b2 TM |
211 | /* Recursively remount a directory (and all its submounts) with desired flags (MS_READONLY, |
212 | * MS_NOSUID, MS_NOEXEC). If the directory is already mounted, we reuse the mount and simply mark it | |
213 | * MS_BIND|MS_RDONLY (or remove the MS_RDONLY for read-write operation), ditto for other flags. If it | |
214 | * isn't we first make it one. Afterwards we apply (or remove) the flags to all submounts we can | |
215 | * access, too. When mounts are stacked on the same mount point we only care for each individual | |
216 | * "top-level" mount on each point, as we cannot influence/access the underlying mounts anyway. We do | |
217 | * not have any effect on future submounts that might get propagated, they might be writable | |
4b6ef527 LP |
218 | * etc. This includes future submounts that have been triggered via autofs. Also note that we can't |
219 | * operate atomically here. Mounts established while we process the tree might or might not get | |
220 | * noticed and thus might or might not be covered. | |
6b7c9f8b | 221 | * |
6b000af4 LP |
222 | * If the "deny_list" parameter is specified it may contain a list of subtrees to exclude from the |
223 | * remount operation. Note that we'll ignore the deny list for the top-level path. */ | |
4349cd7c | 224 | |
4349cd7c | 225 | for (;;) { |
13dcfe46 ZJS |
226 | _cleanup_(mnt_free_tablep) struct libmnt_table *table = NULL; |
227 | _cleanup_(mnt_free_iterp) struct libmnt_iter *iter = NULL; | |
ba8dced2 | 228 | _cleanup_hashmap_free_ Hashmap *todo = NULL; |
4349cd7c | 229 | bool top_autofs = false; |
4349cd7c | 230 | |
670e8efd LP |
231 | if (n_tries++ >= 32) /* Let's not retry this loop forever */ |
232 | return -EBUSY; | |
233 | ||
ac9de0b3 | 234 | rewind(proc_self_mountinfo); |
4349cd7c | 235 | |
e2857b3d | 236 | r = libmount_parse("/proc/self/mountinfo", proc_self_mountinfo, &table, &iter); |
13dcfe46 ZJS |
237 | if (r < 0) |
238 | return log_debug_errno(r, "Failed to parse /proc/self/mountinfo: %m"); | |
4349cd7c | 239 | |
13dcfe46 | 240 | for (;;) { |
ba8dced2 LP |
241 | _cleanup_free_ char *d = NULL; |
242 | const char *path, *type, *opts; | |
243 | unsigned long flags = 0; | |
13dcfe46 | 244 | struct libmnt_fs *fs; |
4349cd7c | 245 | |
13dcfe46 | 246 | r = mnt_table_next_fs(table, iter, &fs); |
d6bfab11 | 247 | if (r == 1) /* EOF */ |
13dcfe46 | 248 | break; |
4349cd7c | 249 | if (r < 0) |
13dcfe46 | 250 | return log_debug_errno(r, "Failed to get next entry from /proc/self/mountinfo: %m"); |
4349cd7c | 251 | |
13dcfe46 | 252 | path = mnt_fs_get_target(fs); |
d6bfab11 | 253 | if (!path) |
6b7c9f8b LP |
254 | continue; |
255 | ||
c6111b85 | 256 | if (!path_startswith(path, prefix)) |
13dcfe46 ZJS |
257 | continue; |
258 | ||
d6bfab11 LP |
259 | type = mnt_fs_get_fstype(fs); |
260 | if (!type) | |
261 | continue; | |
262 | ||
263 | /* Let's ignore autofs mounts. If they aren't triggered yet, we want to avoid | |
264 | * triggering them, as we don't make any guarantees for future submounts anyway. If | |
265 | * they are already triggered, then we will find another entry for this. */ | |
266 | if (streq(type, "autofs")) { | |
267 | top_autofs = top_autofs || path_equal(path, prefix); | |
268 | continue; | |
269 | } | |
270 | ||
271 | if (set_contains(done, path)) | |
272 | continue; | |
273 | ||
6b000af4 | 274 | /* Ignore this mount if it is deny-listed, but only if it isn't the top-level mount |
13dcfe46 | 275 | * we shall operate on. */ |
c6111b85 | 276 | if (!path_equal(path, prefix)) { |
6b000af4 | 277 | bool deny_listed = false; |
6b7c9f8b | 278 | |
6b000af4 | 279 | STRV_FOREACH(i, deny_list) { |
c6111b85 | 280 | if (path_equal(*i, prefix)) |
6b7c9f8b LP |
281 | continue; |
282 | ||
c6111b85 | 283 | if (!path_startswith(*i, prefix)) |
6b7c9f8b LP |
284 | continue; |
285 | ||
13dcfe46 | 286 | if (path_startswith(path, *i)) { |
6b000af4 | 287 | deny_listed = true; |
d6bfab11 | 288 | log_debug("Not remounting %s deny-listed by %s, called for %s", path, *i, prefix); |
6b7c9f8b LP |
289 | break; |
290 | } | |
291 | } | |
d6bfab11 | 292 | |
6b000af4 | 293 | if (deny_listed) |
6b7c9f8b LP |
294 | continue; |
295 | } | |
296 | ||
ba8dced2 LP |
297 | opts = mnt_fs_get_vfs_options(fs); |
298 | if (opts) { | |
299 | r = mnt_optstr_get_flags(opts, &flags, mnt_get_builtin_optmap(MNT_LINUX_MAP)); | |
300 | if (r < 0) | |
301 | log_debug_errno(r, "Could not get flags for '%s', ignoring: %m", path); | |
302 | } | |
303 | ||
304 | d = strdup(path); | |
305 | if (!d) | |
306 | return -ENOMEM; | |
307 | ||
308 | r = hashmap_ensure_put(&todo, &path_hash_ops_free, d, ULONG_TO_PTR(flags)); | |
309 | if (r == -EEXIST) | |
e0103063 LB |
310 | /* If the same path was recorded, but with different mount flags, update it: |
311 | * it means a mount point is overmounted, and libmount returns the "bottom" (or | |
312 | * older one) first, but we want to reapply the flags from the "top" (or newer | |
313 | * one). See: https://github.com/systemd/systemd/issues/20032 | |
314 | * Note that this shouldn't really fail, as we were just told that the key | |
315 | * exists, and it's an update so we want 'd' to be freed immediately. */ | |
316 | r = hashmap_update(todo, d, ULONG_TO_PTR(flags)); | |
d6bfab11 LP |
317 | if (r < 0) |
318 | return r; | |
ba8dced2 LP |
319 | if (r > 0) |
320 | TAKE_PTR(d); | |
4349cd7c LP |
321 | } |
322 | ||
5c5753b9 LP |
323 | /* Check if the top-level directory was among what we have seen so far. For that check both |
324 | * 'done' and 'todo'. Also check 'top_autofs' because if the top-level dir is an autofs we'll | |
325 | * not include it in either set but will set this bool. */ | |
c6111b85 | 326 | if (!set_contains(done, prefix) && |
ba8dced2 | 327 | !(top_autofs || hashmap_contains(todo, prefix))) { |
5c5753b9 | 328 | |
6b7c9f8b | 329 | /* The prefix directory itself is not yet a mount, make it one. */ |
c6111b85 | 330 | r = mount_nofollow(prefix, prefix, NULL, MS_BIND|MS_REC, NULL); |
511a8cfe LP |
331 | if (r < 0) |
332 | return r; | |
4349cd7c | 333 | |
5c5753b9 LP |
334 | /* Immediately rescan, so that we pick up the new mount's flags */ |
335 | continue; | |
4349cd7c LP |
336 | } |
337 | ||
5c5753b9 | 338 | /* If we have no submounts to process anymore, we are done */ |
ba8dced2 | 339 | if (hashmap_isempty(todo)) |
5c5753b9 LP |
340 | return 0; |
341 | ||
ba8dced2 LP |
342 | for (;;) { |
343 | unsigned long flags; | |
344 | char *x = NULL; | |
345 | ||
346 | /* Take the first mount from our list of mounts to still process */ | |
347 | flags = PTR_TO_ULONG(hashmap_steal_first_key_and_value(todo, (void**) &x)); | |
348 | if (!x) | |
349 | break; | |
4349cd7c | 350 | |
ba8dced2 | 351 | r = set_ensure_consume(&done, &path_hash_ops_free, x); |
4c701096 | 352 | if (IN_SET(r, 0, -EEXIST)) |
ba8dced2 | 353 | continue; /* Already done */ |
4349cd7c LP |
354 | if (r < 0) |
355 | return r; | |
356 | ||
ba8dced2 LP |
357 | /* Now, remount this with the new flags set, but exclude MS_RELATIME from it. (It's |
358 | * the default anyway, thus redundant, and in userns we'll get an error if we try to | |
359 | * explicitly enable it) */ | |
360 | r = mount_nofollow(NULL, x, NULL, ((flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags) & ~MS_RELATIME, NULL); | |
065b4774 | 361 | if (r < 0) { |
ba8dced2 LP |
362 | int q; |
363 | ||
364 | /* OK, so the remount of this entry failed. We'll ultimately ignore this in | |
365 | * almost all cases (there are simply so many reasons why this can fail, | |
366 | * think autofs, NFS, FUSE, …), but let's generate useful debug messages at | |
367 | * the very least. */ | |
368 | ||
369 | q = path_is_mount_point(x, NULL, 0); | |
370 | if (IN_SET(q, 0, -ENOENT)) { | |
371 | /* Hmm, whaaaa? The mount point is not actually a mount point? Then | |
372 | * it is either obstructed by a later mount or somebody has been | |
373 | * racing against us and removed it. Either way the mount point | |
374 | * doesn't matter to us, let's ignore it hence. */ | |
375 | log_debug_errno(r, "Mount point '%s' to remount is not a mount point anymore, ignoring remount failure: %m", x); | |
376 | continue; | |
377 | } | |
378 | if (q < 0) /* Any other error on this? Just log and continue */ | |
379 | log_debug_errno(q, "Failed to determine whether '%s' is a mount point or not, ignoring: %m", x); | |
380 | ||
381 | if (((flags ^ new_flags) & flags_mask & ~MS_RELATIME) == 0) { /* ignore MS_RELATIME while comparing */ | |
382 | log_debug_errno(r, "Couldn't remount '%s', but the flags already match what we want, hence ignoring: %m", x); | |
383 | continue; | |
384 | } | |
385 | ||
386 | /* Make this fatal if this is the top-level mount */ | |
387 | if (path_equal(x, prefix)) | |
065b4774 LP |
388 | return r; |
389 | ||
ba8dced2 LP |
390 | /* If this is not the top-level mount, then handle this gracefully: log but |
391 | * otherwise ignore. With NFS, FUSE, autofs there are just too many reasons | |
392 | * this might fail without a chance for us to do anything about it, let's | |
393 | * hence be strict on the top-level mount and lenient on the inner ones. */ | |
394 | log_debug_errno(r, "Couldn't remount submount '%s' for unexpected reason, ignoring: %m", x); | |
ef454fd1 YW |
395 | continue; |
396 | } | |
98df8089 | 397 | |
ba8dced2 | 398 | log_debug("Remounted %s.", x); |
4349cd7c LP |
399 | } |
400 | } | |
401 | } | |
402 | ||
7cce68e1 LP |
403 | int bind_remount_one_with_mountinfo( |
404 | const char *path, | |
405 | unsigned long new_flags, | |
406 | unsigned long flags_mask, | |
407 | FILE *proc_self_mountinfo) { | |
408 | ||
409 | _cleanup_(mnt_free_tablep) struct libmnt_table *table = NULL; | |
2c5ff8ea LP |
410 | unsigned long flags = 0; |
411 | struct libmnt_fs *fs; | |
412 | const char *opts; | |
7cce68e1 LP |
413 | int r; |
414 | ||
415 | assert(path); | |
416 | assert(proc_self_mountinfo); | |
417 | ||
4f5644db LP |
418 | if ((flags_mask & ~MS_CONVERTIBLE_FLAGS) == 0 && !skip_mount_set_attr) { |
419 | /* Let's take a shortcut for all the flags we know how to convert into mount_setattr() flags */ | |
420 | ||
421 | if (mount_setattr(AT_FDCWD, path, AT_SYMLINK_NOFOLLOW, | |
422 | &(struct mount_attr) { | |
423 | .attr_set = ms_flags_to_mount_attr(new_flags & flags_mask), | |
424 | .attr_clr = ms_flags_to_mount_attr(~new_flags & flags_mask), | |
425 | }, MOUNT_ATTR_SIZE_VER0) < 0) { | |
426 | ||
427 | log_debug_errno(errno, "mount_setattr() didn't work, falling back to classic remounting: %m"); | |
428 | ||
429 | if (ERRNO_IS_NOT_SUPPORTED(errno)) /* if not supported, then don't bother at all anymore */ | |
430 | skip_mount_set_attr = true; | |
431 | } else | |
432 | return 0; /* Nice, this worked! */ | |
433 | } | |
434 | ||
7cce68e1 LP |
435 | rewind(proc_self_mountinfo); |
436 | ||
437 | table = mnt_new_table(); | |
438 | if (!table) | |
439 | return -ENOMEM; | |
440 | ||
441 | r = mnt_table_parse_stream(table, proc_self_mountinfo, "/proc/self/mountinfo"); | |
442 | if (r < 0) | |
443 | return r; | |
444 | ||
2c5ff8ea | 445 | fs = mnt_table_find_target(table, path, MNT_ITER_FORWARD); |
0338df47 LP |
446 | if (!fs) { |
447 | if (laccess(path, F_OK) < 0) /* Hmm, it's not in the mount table, but does it exist at all? */ | |
448 | return -errno; | |
449 | ||
2c5ff8ea | 450 | return -EINVAL; /* Not a mount point we recognize */ |
0338df47 | 451 | } |
2c5ff8ea LP |
452 | |
453 | opts = mnt_fs_get_vfs_options(fs); | |
454 | if (opts) { | |
455 | r = mnt_optstr_get_flags(opts, &flags, mnt_get_builtin_optmap(MNT_LINUX_MAP)); | |
456 | if (r < 0) | |
457 | log_debug_errno(r, "Could not get flags for '%s', ignoring: %m", path); | |
458 | } | |
7cce68e1 | 459 | |
2c5ff8ea | 460 | r = mount_nofollow(NULL, path, NULL, ((flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags) & ~MS_RELATIME, NULL); |
b23c6a64 LP |
461 | if (r < 0) { |
462 | if (((flags ^ new_flags) & flags_mask & ~MS_RELATIME) != 0) /* Ignore MS_RELATIME again, | |
463 | * since kernel adds it in | |
464 | * everywhere, because it's the | |
465 | * default. */ | |
466 | return r; | |
467 | ||
468 | /* Let's handle redundant remounts gracefully */ | |
469 | log_debug_errno(r, "Failed to remount '%s' but flags already match what we want, ignoring: %m", path); | |
470 | } | |
7cce68e1 LP |
471 | |
472 | return 0; | |
473 | } | |
474 | ||
4349cd7c LP |
475 | int mount_move_root(const char *path) { |
476 | assert(path); | |
477 | ||
478 | if (chdir(path) < 0) | |
479 | return -errno; | |
480 | ||
481 | if (mount(path, "/", NULL, MS_MOVE, NULL) < 0) | |
482 | return -errno; | |
483 | ||
484 | if (chroot(".") < 0) | |
485 | return -errno; | |
486 | ||
7c248223 | 487 | return RET_NERRNO(chdir("/")); |
4349cd7c | 488 | } |
4e036b7a | 489 | |
3f2c0bec LP |
490 | int repeat_unmount(const char *path, int flags) { |
491 | bool done = false; | |
492 | ||
493 | assert(path); | |
494 | ||
495 | /* If there are multiple mounts on a mount point, this | |
496 | * removes them all */ | |
497 | ||
498 | for (;;) { | |
499 | if (umount2(path, flags) < 0) { | |
500 | ||
501 | if (errno == EINVAL) | |
502 | return done; | |
503 | ||
504 | return -errno; | |
505 | } | |
506 | ||
507 | done = true; | |
508 | } | |
509 | } | |
c4b41707 | 510 | |
48b747fa LP |
511 | int mode_to_inaccessible_node( |
512 | const char *runtime_dir, | |
513 | mode_t mode, | |
514 | char **ret) { | |
515 | ||
516 | /* This function maps a node type to a corresponding inaccessible file node. These nodes are created | |
517 | * during early boot by PID 1. In some cases we lacked the privs to create the character and block | |
518 | * devices (maybe because we run in an userns environment, or miss CAP_SYS_MKNOD, or run with a | |
519 | * devices policy that excludes device nodes with major and minor of 0), but that's fine, in that | |
520 | * case we use an AF_UNIX file node instead, which is not the same, but close enough for most | |
521 | * uses. And most importantly, the kernel allows bind mounts from socket nodes to any non-directory | |
522 | * file nodes, and that's the most important thing that matters. | |
523 | * | |
524 | * Note that the runtime directory argument shall be the top-level runtime directory, i.e. /run/ if | |
525 | * we operate in system context and $XDG_RUNTIME_DIR if we operate in user context. */ | |
526 | ||
e5f10caf AZ |
527 | _cleanup_free_ char *d = NULL; |
528 | const char *node = NULL; | |
e5f10caf | 529 | |
48b747fa LP |
530 | assert(ret); |
531 | ||
532 | if (!runtime_dir) | |
533 | runtime_dir = "/run"; | |
fe80fcc7 | 534 | |
79893116 | 535 | switch (mode & S_IFMT) { |
c4b41707 | 536 | case S_IFREG: |
48b747fa | 537 | node = "/systemd/inaccessible/reg"; |
e5f10caf | 538 | break; |
fe80fcc7 | 539 | |
c4b41707 | 540 | case S_IFDIR: |
48b747fa | 541 | node = "/systemd/inaccessible/dir"; |
e5f10caf | 542 | break; |
fe80fcc7 | 543 | |
c4b41707 | 544 | case S_IFCHR: |
48b747fa | 545 | node = "/systemd/inaccessible/chr"; |
e5f10caf | 546 | break; |
fe80fcc7 | 547 | |
c4b41707 | 548 | case S_IFBLK: |
48b747fa | 549 | node = "/systemd/inaccessible/blk"; |
e5f10caf | 550 | break; |
fe80fcc7 | 551 | |
c4b41707 | 552 | case S_IFIFO: |
48b747fa | 553 | node = "/systemd/inaccessible/fifo"; |
e5f10caf | 554 | break; |
fe80fcc7 | 555 | |
c4b41707 | 556 | case S_IFSOCK: |
48b747fa | 557 | node = "/systemd/inaccessible/sock"; |
e5f10caf | 558 | break; |
c4b41707 | 559 | } |
e5f10caf AZ |
560 | if (!node) |
561 | return -EINVAL; | |
562 | ||
48b747fa LP |
563 | d = path_join(runtime_dir, node); |
564 | if (!d) | |
565 | return -ENOMEM; | |
566 | ||
cbed1dc8 LP |
567 | /* On new kernels unprivileged users are permitted to create 0:0 char device nodes (because they also |
568 | * act as whiteout inode for overlayfs), but no other char or block device nodes. On old kernels no | |
569 | * device node whatsoever may be created by unprivileged processes. Hence, if the caller asks for the | |
570 | * inaccessible block device node let's see if the block device node actually exists, and if not, | |
571 | * fall back to the character device node. From there fall back to the socket device node. This means | |
572 | * in the best case we'll get the right device node type — but if not we'll hopefully at least get a | |
573 | * device node at all. */ | |
574 | ||
575 | if (S_ISBLK(mode) && | |
576 | access(d, F_OK) < 0 && errno == ENOENT) { | |
577 | free(d); | |
578 | d = path_join(runtime_dir, "/systemd/inaccessible/chr"); | |
579 | if (!d) | |
580 | return -ENOMEM; | |
581 | } | |
582 | ||
583 | if (IN_SET(mode & S_IFMT, S_IFBLK, S_IFCHR) && | |
584 | access(d, F_OK) < 0 && errno == ENOENT) { | |
48b747fa LP |
585 | free(d); |
586 | d = path_join(runtime_dir, "/systemd/inaccessible/sock"); | |
587 | if (!d) | |
588 | return -ENOMEM; | |
589 | } | |
e5f10caf | 590 | |
48b747fa | 591 | *ret = TAKE_PTR(d); |
e5f10caf | 592 | return 0; |
c4b41707 | 593 | } |
60e76d48 | 594 | |
da185cd0 | 595 | int mount_flags_to_string(unsigned long flags, char **ret) { |
1c092b62 | 596 | static const struct { |
da185cd0 | 597 | unsigned long flag; |
1c092b62 YW |
598 | const char *name; |
599 | } map[] = { | |
600 | { .flag = MS_RDONLY, .name = "MS_RDONLY", }, | |
601 | { .flag = MS_NOSUID, .name = "MS_NOSUID", }, | |
602 | { .flag = MS_NODEV, .name = "MS_NODEV", }, | |
603 | { .flag = MS_NOEXEC, .name = "MS_NOEXEC", }, | |
604 | { .flag = MS_SYNCHRONOUS, .name = "MS_SYNCHRONOUS", }, | |
605 | { .flag = MS_REMOUNT, .name = "MS_REMOUNT", }, | |
606 | { .flag = MS_MANDLOCK, .name = "MS_MANDLOCK", }, | |
607 | { .flag = MS_DIRSYNC, .name = "MS_DIRSYNC", }, | |
608 | { .flag = MS_NOSYMFOLLOW, .name = "MS_NOSYMFOLLOW", }, | |
609 | { .flag = MS_NOATIME, .name = "MS_NOATIME", }, | |
610 | { .flag = MS_NODIRATIME, .name = "MS_NODIRATIME", }, | |
611 | { .flag = MS_BIND, .name = "MS_BIND", }, | |
612 | { .flag = MS_MOVE, .name = "MS_MOVE", }, | |
613 | { .flag = MS_REC, .name = "MS_REC", }, | |
614 | { .flag = MS_SILENT, .name = "MS_SILENT", }, | |
615 | { .flag = MS_POSIXACL, .name = "MS_POSIXACL", }, | |
616 | { .flag = MS_UNBINDABLE, .name = "MS_UNBINDABLE", }, | |
617 | { .flag = MS_PRIVATE, .name = "MS_PRIVATE", }, | |
618 | { .flag = MS_SLAVE, .name = "MS_SLAVE", }, | |
619 | { .flag = MS_SHARED, .name = "MS_SHARED", }, | |
620 | { .flag = MS_RELATIME, .name = "MS_RELATIME", }, | |
621 | { .flag = MS_KERNMOUNT, .name = "MS_KERNMOUNT", }, | |
622 | { .flag = MS_I_VERSION, .name = "MS_I_VERSION", }, | |
623 | { .flag = MS_STRICTATIME, .name = "MS_STRICTATIME", }, | |
624 | { .flag = MS_LAZYTIME, .name = "MS_LAZYTIME", }, | |
625 | }; | |
626 | _cleanup_free_ char *str = NULL; | |
627 | ||
4bee2333 YW |
628 | assert(ret); |
629 | ||
1c092b62 YW |
630 | for (size_t i = 0; i < ELEMENTSOF(map); i++) |
631 | if (flags & map[i].flag) { | |
632 | if (!strextend_with_separator(&str, "|", map[i].name)) | |
633 | return -ENOMEM; | |
634 | flags &= ~map[i].flag; | |
635 | } | |
636 | ||
637 | if (!str || flags != 0) | |
638 | if (strextendf_with_separator(&str, "|", "%lx", flags) < 0) | |
639 | return -ENOMEM; | |
640 | ||
641 | *ret = TAKE_PTR(str); | |
642 | return 0; | |
60e76d48 ZJS |
643 | } |
644 | ||
511a8cfe | 645 | int mount_verbose_full( |
60e76d48 ZJS |
646 | int error_log_level, |
647 | const char *what, | |
648 | const char *where, | |
649 | const char *type, | |
650 | unsigned long flags, | |
511a8cfe LP |
651 | const char *options, |
652 | bool follow_symlink) { | |
60e76d48 | 653 | |
6ef8df2b YW |
654 | _cleanup_free_ char *fl = NULL, *o = NULL; |
655 | unsigned long f; | |
656 | int r; | |
657 | ||
658 | r = mount_option_mangle(options, flags, &f, &o); | |
659 | if (r < 0) | |
660 | return log_full_errno(error_log_level, r, | |
661 | "Failed to mangle mount options %s: %m", | |
662 | strempty(options)); | |
60e76d48 | 663 | |
1c092b62 | 664 | (void) mount_flags_to_string(f, &fl); |
60e76d48 | 665 | |
6ef8df2b | 666 | if ((f & MS_REMOUNT) && !what && !type) |
60e76d48 | 667 | log_debug("Remounting %s (%s \"%s\")...", |
6ef8df2b | 668 | where, strnull(fl), strempty(o)); |
60e76d48 ZJS |
669 | else if (!what && !type) |
670 | log_debug("Mounting %s (%s \"%s\")...", | |
6ef8df2b YW |
671 | where, strnull(fl), strempty(o)); |
672 | else if ((f & MS_BIND) && !type) | |
60e76d48 | 673 | log_debug("Bind-mounting %s on %s (%s \"%s\")...", |
6ef8df2b YW |
674 | what, where, strnull(fl), strempty(o)); |
675 | else if (f & MS_MOVE) | |
e2341b6b DT |
676 | log_debug("Moving mount %s %s %s (%s \"%s\")...", |
677 | what, special_glyph(SPECIAL_GLYPH_ARROW_RIGHT), where, strnull(fl), strempty(o)); | |
60e76d48 | 678 | else |
3b493d94 LP |
679 | log_debug("Mounting %s (%s) on %s (%s \"%s\")...", |
680 | strna(what), strna(type), where, strnull(fl), strempty(o)); | |
511a8cfe LP |
681 | |
682 | if (follow_symlink) | |
7c248223 | 683 | r = RET_NERRNO(mount(what, where, type, f, o)); |
511a8cfe LP |
684 | else |
685 | r = mount_nofollow(what, where, type, f, o); | |
686 | if (r < 0) | |
687 | return log_full_errno(error_log_level, r, | |
3ccf6126 LP |
688 | "Failed to mount %s (type %s) on %s (%s \"%s\"): %m", |
689 | strna(what), strna(type), where, strnull(fl), strempty(o)); | |
60e76d48 ZJS |
690 | return 0; |
691 | } | |
692 | ||
30f5d104 LP |
693 | int umount_verbose( |
694 | int error_log_level, | |
695 | const char *what, | |
696 | int flags) { | |
697 | ||
698 | assert(what); | |
699 | ||
60e76d48 | 700 | log_debug("Umounting %s...", what); |
30f5d104 LP |
701 | |
702 | if (umount2(what, flags) < 0) | |
703 | return log_full_errno(error_log_level, errno, | |
704 | "Failed to unmount %s: %m", what); | |
705 | ||
60e76d48 ZJS |
706 | return 0; |
707 | } | |
83555251 | 708 | |
9e7f941a YW |
709 | int mount_option_mangle( |
710 | const char *options, | |
711 | unsigned long mount_flags, | |
712 | unsigned long *ret_mount_flags, | |
713 | char **ret_remaining_options) { | |
714 | ||
715 | const struct libmnt_optmap *map; | |
716 | _cleanup_free_ char *ret = NULL; | |
9e7f941a YW |
717 | int r; |
718 | ||
719 | /* This extracts mount flags from the mount options, and store | |
720 | * non-mount-flag options to '*ret_remaining_options'. | |
721 | * E.g., | |
722 | * "rw,nosuid,nodev,relatime,size=1630748k,mode=700,uid=1000,gid=1000" | |
723 | * is split to MS_NOSUID|MS_NODEV|MS_RELATIME and | |
724 | * "size=1630748k,mode=700,uid=1000,gid=1000". | |
725 | * See more examples in test-mount-utils.c. | |
726 | * | |
727 | * Note that if 'options' does not contain any non-mount-flag options, | |
5238e957 | 728 | * then '*ret_remaining_options' is set to NULL instead of empty string. |
9e7f941a YW |
729 | * Note that this does not check validity of options stored in |
730 | * '*ret_remaining_options'. | |
731 | * Note that if 'options' is NULL, then this just copies 'mount_flags' | |
732 | * to '*ret_mount_flags'. */ | |
733 | ||
734 | assert(ret_mount_flags); | |
735 | assert(ret_remaining_options); | |
736 | ||
737 | map = mnt_get_builtin_optmap(MNT_LINUX_MAP); | |
738 | if (!map) | |
739 | return -EINVAL; | |
740 | ||
25086b4c | 741 | for (const char *p = options;;) { |
9e7f941a YW |
742 | _cleanup_free_ char *word = NULL; |
743 | const struct libmnt_optmap *ent; | |
744 | ||
9b23679e | 745 | r = extract_first_word(&p, &word, ",", EXTRACT_KEEP_QUOTE); |
9e7f941a YW |
746 | if (r < 0) |
747 | return r; | |
748 | if (r == 0) | |
749 | break; | |
750 | ||
751 | for (ent = map; ent->name; ent++) { | |
752 | /* All entries in MNT_LINUX_MAP do not take any argument. | |
753 | * Thus, ent->name does not contain "=" or "[=]". */ | |
754 | if (!streq(word, ent->name)) | |
755 | continue; | |
756 | ||
757 | if (!(ent->mask & MNT_INVERT)) | |
758 | mount_flags |= ent->id; | |
759 | else if (mount_flags & ent->id) | |
760 | mount_flags ^= ent->id; | |
761 | ||
762 | break; | |
763 | } | |
764 | ||
765 | /* If 'word' is not a mount flag, then store it in '*ret_remaining_options'. */ | |
c2bc710b | 766 | if (!ent->name && !strextend_with_separator(&ret, ",", word)) |
9e7f941a YW |
767 | return -ENOMEM; |
768 | } | |
769 | ||
770 | *ret_mount_flags = mount_flags; | |
ae2a15bc | 771 | *ret_remaining_options = TAKE_PTR(ret); |
9e7f941a YW |
772 | |
773 | return 0; | |
774 | } | |
6af52c3a | 775 | |
70599967 | 776 | static int mount_in_namespace( |
6af52c3a LB |
777 | pid_t target, |
778 | const char *propagate_path, | |
779 | const char *incoming_path, | |
780 | const char *src, | |
781 | const char *dest, | |
782 | bool read_only, | |
70599967 LB |
783 | bool make_file_or_directory, |
784 | const MountOptions *options, | |
785 | bool is_image) { | |
6af52c3a LB |
786 | |
787 | _cleanup_close_pair_ int errno_pipe_fd[2] = { -1, -1 }; | |
f7c18d3d | 788 | _cleanup_close_ int self_mntns_fd = -1, mntns_fd = -1, root_fd = -1, pidns_fd = -1, chased_src_fd = -1; |
ddb6eeaf | 789 | char mount_slave[] = "/tmp/propagate.XXXXXX", *mount_tmp, *mount_outside, *p; |
6af52c3a LB |
790 | bool mount_slave_created = false, mount_slave_mounted = false, |
791 | mount_tmp_created = false, mount_tmp_mounted = false, | |
792 | mount_outside_created = false, mount_outside_mounted = false; | |
cedf5b1a | 793 | _cleanup_free_ char *chased_src_path = NULL; |
2338a175 | 794 | struct stat st, self_mntns_st; |
6af52c3a LB |
795 | pid_t child; |
796 | int r; | |
797 | ||
798 | assert(target > 0); | |
799 | assert(propagate_path); | |
800 | assert(incoming_path); | |
801 | assert(src); | |
802 | assert(dest); | |
70599967 | 803 | assert(!options || is_image); |
6af52c3a | 804 | |
98f654fd | 805 | r = namespace_open(target, &pidns_fd, &mntns_fd, NULL, NULL, &root_fd); |
2338a175 LB |
806 | if (r < 0) |
807 | return log_debug_errno(r, "Failed to retrieve FDs of the target process' namespace: %m"); | |
808 | ||
809 | if (fstat(mntns_fd, &st) < 0) | |
810 | return log_debug_errno(errno, "Failed to fstat mount namespace FD of target process: %m"); | |
811 | ||
812 | r = namespace_open(0, NULL, &self_mntns_fd, NULL, NULL, NULL); | |
813 | if (r < 0) | |
814 | return log_debug_errno(r, "Failed to retrieve FDs of systemd's namespace: %m"); | |
815 | ||
816 | if (fstat(self_mntns_fd, &self_mntns_st) < 0) | |
817 | return log_debug_errno(errno, "Failed to fstat mount namespace FD of systemd: %m"); | |
818 | ||
819 | /* We can't add new mounts at runtime if the process wasn't started in a namespace */ | |
a9dac7a6 | 820 | if (stat_inode_same(&st, &self_mntns_st)) |
2338a175 LB |
821 | return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to activate bind mount in target, not running in a mount namespace"); |
822 | ||
ddb6eeaf LP |
823 | /* One day, when bind mounting /proc/self/fd/n works across namespace boundaries we should rework |
824 | * this logic to make use of it... */ | |
6af52c3a LB |
825 | |
826 | p = strjoina(propagate_path, "/"); | |
827 | r = laccess(p, F_OK); | |
828 | if (r < 0) | |
829 | return log_debug_errno(r == -ENOENT ? SYNTHETIC_ERRNO(EOPNOTSUPP) : r, "Target does not allow propagation of mount points"); | |
830 | ||
cedf5b1a | 831 | r = chase_symlinks(src, NULL, 0, &chased_src_path, &chased_src_fd); |
6af52c3a LB |
832 | if (r < 0) |
833 | return log_debug_errno(r, "Failed to resolve source path of %s: %m", src); | |
cedf5b1a | 834 | log_debug("Chased source path of %s to %s", src, chased_src_path); |
6af52c3a | 835 | |
f7c18d3d LB |
836 | if (fstat(chased_src_fd, &st) < 0) |
837 | return log_debug_errno(errno, "Failed to stat() resolved source path %s: %m", src); | |
6af52c3a | 838 | if (S_ISLNK(st.st_mode)) /* This shouldn't really happen, given that we just chased the symlinks above, but let's better be safe… */ |
f7c18d3d | 839 | return log_debug_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "Source directory %s can't be a symbolic link", src); |
6af52c3a LB |
840 | |
841 | /* Our goal is to install a new bind mount into the container, | |
842 | possibly read-only. This is irritatingly complex | |
843 | unfortunately, currently. | |
844 | ||
845 | First, we start by creating a private playground in /tmp, | |
846 | that we can mount MS_SLAVE. (Which is necessary, since | |
847 | MS_MOVE cannot be applied to mounts with MS_SHARED parent | |
848 | mounts.) */ | |
849 | ||
850 | if (!mkdtemp(mount_slave)) | |
851 | return log_debug_errno(errno, "Failed to create playground %s: %m", mount_slave); | |
852 | ||
853 | mount_slave_created = true; | |
854 | ||
855 | r = mount_nofollow_verbose(LOG_DEBUG, mount_slave, mount_slave, NULL, MS_BIND, NULL); | |
856 | if (r < 0) | |
857 | goto finish; | |
858 | ||
859 | mount_slave_mounted = true; | |
860 | ||
861 | r = mount_nofollow_verbose(LOG_DEBUG, NULL, mount_slave, NULL, MS_SLAVE, NULL); | |
862 | if (r < 0) | |
863 | goto finish; | |
864 | ||
865 | /* Second, we mount the source file or directory to a directory inside of our MS_SLAVE playground. */ | |
866 | mount_tmp = strjoina(mount_slave, "/mount"); | |
70599967 LB |
867 | if (is_image) |
868 | r = mkdir_p(mount_tmp, 0700); | |
869 | else | |
870 | r = make_mount_point_inode_from_stat(&st, mount_tmp, 0700); | |
6af52c3a LB |
871 | if (r < 0) { |
872 | log_debug_errno(r, "Failed to create temporary mount point %s: %m", mount_tmp); | |
873 | goto finish; | |
874 | } | |
875 | ||
876 | mount_tmp_created = true; | |
877 | ||
70599967 | 878 | if (is_image) |
cedf5b1a | 879 | r = verity_dissect_and_mount(chased_src_fd, chased_src_path, mount_tmp, options, NULL, NULL, NULL, NULL); |
70599967 | 880 | else |
ddb6eeaf | 881 | r = mount_follow_verbose(LOG_DEBUG, FORMAT_PROC_FD_PATH(chased_src_fd), mount_tmp, NULL, MS_BIND, NULL); |
6af52c3a LB |
882 | if (r < 0) |
883 | goto finish; | |
884 | ||
885 | mount_tmp_mounted = true; | |
886 | ||
887 | /* Third, we remount the new bind mount read-only if requested. */ | |
888 | if (read_only) { | |
889 | r = mount_nofollow_verbose(LOG_DEBUG, NULL, mount_tmp, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL); | |
890 | if (r < 0) | |
891 | goto finish; | |
892 | } | |
893 | ||
894 | /* Fourth, we move the new bind mount into the propagation directory. This way it will appear there read-only | |
895 | * right-away. */ | |
896 | ||
897 | mount_outside = strjoina(propagate_path, "/XXXXXX"); | |
70599967 | 898 | if (is_image || S_ISDIR(st.st_mode)) |
6af52c3a LB |
899 | r = mkdtemp(mount_outside) ? 0 : -errno; |
900 | else { | |
901 | r = mkostemp_safe(mount_outside); | |
902 | safe_close(r); | |
903 | } | |
904 | if (r < 0) { | |
905 | log_debug_errno(r, "Cannot create propagation file or directory %s: %m", mount_outside); | |
906 | goto finish; | |
907 | } | |
908 | ||
909 | mount_outside_created = true; | |
910 | ||
911 | r = mount_nofollow_verbose(LOG_DEBUG, mount_tmp, mount_outside, NULL, MS_MOVE, NULL); | |
912 | if (r < 0) | |
913 | goto finish; | |
914 | ||
915 | mount_outside_mounted = true; | |
916 | mount_tmp_mounted = false; | |
917 | ||
70599967 | 918 | if (is_image || S_ISDIR(st.st_mode)) |
6af52c3a LB |
919 | (void) rmdir(mount_tmp); |
920 | else | |
921 | (void) unlink(mount_tmp); | |
922 | mount_tmp_created = false; | |
923 | ||
924 | (void) umount_verbose(LOG_DEBUG, mount_slave, UMOUNT_NOFOLLOW); | |
925 | mount_slave_mounted = false; | |
926 | ||
927 | (void) rmdir(mount_slave); | |
928 | mount_slave_created = false; | |
929 | ||
930 | if (pipe2(errno_pipe_fd, O_CLOEXEC|O_NONBLOCK) < 0) { | |
931 | log_debug_errno(errno, "Failed to create pipe: %m"); | |
932 | goto finish; | |
933 | } | |
934 | ||
2338a175 | 935 | r = namespace_fork("(sd-bindmnt)", "(sd-bindmnt-inner)", NULL, 0, FORK_RESET_SIGNALS|FORK_DEATHSIG, |
98f654fd | 936 | pidns_fd, mntns_fd, -1, -1, root_fd, &child); |
6af52c3a LB |
937 | if (r < 0) |
938 | goto finish; | |
939 | if (r == 0) { | |
2338a175 | 940 | const char *mount_inside; |
6af52c3a LB |
941 | |
942 | errno_pipe_fd[0] = safe_close(errno_pipe_fd[0]); | |
943 | ||
6af52c3a | 944 | if (make_file_or_directory) { |
70599967 LB |
945 | if (!is_image) { |
946 | (void) mkdir_parents(dest, 0755); | |
947 | (void) make_mount_point_inode_from_stat(&st, dest, 0700); | |
948 | } else | |
949 | (void) mkdir_p(dest, 0755); | |
6af52c3a LB |
950 | } |
951 | ||
952 | /* Fifth, move the mount to the right place inside */ | |
953 | mount_inside = strjoina(incoming_path, basename(mount_outside)); | |
954 | r = mount_nofollow_verbose(LOG_ERR, mount_inside, dest, NULL, MS_MOVE, NULL); | |
955 | if (r < 0) | |
956 | goto child_fail; | |
957 | ||
958 | _exit(EXIT_SUCCESS); | |
959 | ||
960 | child_fail: | |
961 | (void) write(errno_pipe_fd[1], &r, sizeof(r)); | |
962 | errno_pipe_fd[1] = safe_close(errno_pipe_fd[1]); | |
963 | ||
964 | _exit(EXIT_FAILURE); | |
965 | } | |
966 | ||
967 | errno_pipe_fd[1] = safe_close(errno_pipe_fd[1]); | |
968 | ||
969 | r = wait_for_terminate_and_check("(sd-bindmnt)", child, 0); | |
970 | if (r < 0) { | |
971 | log_debug_errno(r, "Failed to wait for child: %m"); | |
972 | goto finish; | |
973 | } | |
974 | if (r != EXIT_SUCCESS) { | |
975 | if (read(errno_pipe_fd[0], &r, sizeof(r)) == sizeof(r)) | |
976 | log_debug_errno(r, "Failed to mount: %m"); | |
977 | else | |
978 | log_debug("Child failed."); | |
979 | goto finish; | |
980 | } | |
981 | ||
982 | finish: | |
983 | if (mount_outside_mounted) | |
984 | (void) umount_verbose(LOG_DEBUG, mount_outside, UMOUNT_NOFOLLOW); | |
985 | if (mount_outside_created) { | |
70599967 | 986 | if (is_image || S_ISDIR(st.st_mode)) |
6af52c3a LB |
987 | (void) rmdir(mount_outside); |
988 | else | |
989 | (void) unlink(mount_outside); | |
990 | } | |
991 | ||
992 | if (mount_tmp_mounted) | |
993 | (void) umount_verbose(LOG_DEBUG, mount_tmp, UMOUNT_NOFOLLOW); | |
994 | if (mount_tmp_created) { | |
70599967 | 995 | if (is_image || S_ISDIR(st.st_mode)) |
6af52c3a LB |
996 | (void) rmdir(mount_tmp); |
997 | else | |
998 | (void) unlink(mount_tmp); | |
999 | } | |
1000 | ||
1001 | if (mount_slave_mounted) | |
1002 | (void) umount_verbose(LOG_DEBUG, mount_slave, UMOUNT_NOFOLLOW); | |
1003 | if (mount_slave_created) | |
1004 | (void) rmdir(mount_slave); | |
1005 | ||
1006 | return r; | |
1007 | } | |
70599967 LB |
1008 | |
1009 | int bind_mount_in_namespace( | |
1010 | pid_t target, | |
1011 | const char *propagate_path, | |
1012 | const char *incoming_path, | |
1013 | const char *src, | |
1014 | const char *dest, | |
1015 | bool read_only, | |
1016 | bool make_file_or_directory) { | |
1017 | ||
1018 | return mount_in_namespace(target, propagate_path, incoming_path, src, dest, read_only, make_file_or_directory, NULL, false); | |
1019 | } | |
1020 | ||
1021 | int mount_image_in_namespace( | |
1022 | pid_t target, | |
1023 | const char *propagate_path, | |
1024 | const char *incoming_path, | |
1025 | const char *src, | |
1026 | const char *dest, | |
1027 | bool read_only, | |
1028 | bool make_file_or_directory, | |
1029 | const MountOptions *options) { | |
1030 | ||
1031 | return mount_in_namespace(target, propagate_path, incoming_path, src, dest, read_only, make_file_or_directory, options, true); | |
1032 | } | |
14a25e1f LP |
1033 | |
1034 | int make_mount_point(const char *path) { | |
1035 | int r; | |
1036 | ||
1037 | assert(path); | |
1038 | ||
1039 | /* If 'path' is already a mount point, does nothing and returns 0. If it is not it makes it one, and returns 1. */ | |
1040 | ||
1041 | r = path_is_mount_point(path, NULL, 0); | |
1042 | if (r < 0) | |
1043 | return log_debug_errno(r, "Failed to determine whether '%s' is a mount point: %m", path); | |
1044 | if (r > 0) | |
1045 | return 0; | |
1046 | ||
1047 | r = mount_nofollow_verbose(LOG_DEBUG, path, path, NULL, MS_BIND|MS_REC, NULL); | |
1048 | if (r < 0) | |
1049 | return r; | |
1050 | ||
1051 | return 1; | |
1052 | } | |
35fd3558 | 1053 | |
50ae2966 | 1054 | static int make_userns(uid_t uid_shift, uid_t uid_range, RemountIdmapFlags flags) { |
35fd3558 | 1055 | _cleanup_close_ int userns_fd = -1; |
50ae2966 | 1056 | _cleanup_free_ char *line = NULL; |
35fd3558 LP |
1057 | |
1058 | /* Allocates a userns file descriptor with the mapping we need. For this we'll fork off a child | |
1059 | * process whose only purpose is to give us a new user namespace. It's killed when we got it. */ | |
1060 | ||
50ae2966 LP |
1061 | if (asprintf(&line, UID_FMT " " UID_FMT " " UID_FMT "\n", 0, uid_shift, uid_range) < 0) |
1062 | return log_oom_debug(); | |
1063 | ||
1064 | /* If requested we'll include an entry in the mapping so that the host root user can make changes to | |
1065 | * the uidmapped mount like it normally would. Specifically, we'll map the user with UID_HOST_ROOT on | |
1066 | * the backing fs to UID 0. This is useful, since nspawn code wants to create various missing inodes | |
1067 | * in the OS tree before booting into it, and this becomes very easy and straightforward to do if it | |
1068 | * can just do it under its own regular UID. Note that in that case the container's runtime uidmap | |
1069 | * (i.e. the one the container payload processes run in) will leave this UID unmapped, i.e. if we | |
1070 | * accidentally leave files owned by host root in the already uidmapped tree around they'll show up | |
1071 | * as owned by 'nobody', which is safe. (Of course, we shouldn't leave such inodes around, but always | |
1072 | * chown() them to the container's own UID range, but it's good to have a safety net, in case we | |
1073 | * forget it.) */ | |
1074 | if (flags & REMOUNT_IDMAP_HOST_ROOT) | |
1075 | if (strextendf(&line, | |
1076 | UID_FMT " " UID_FMT " " UID_FMT "\n", | |
1077 | UID_MAPPED_ROOT, 0, 1) < 0) | |
1078 | return log_oom_debug(); | |
35fd3558 | 1079 | |
35fd3558 | 1080 | /* We always assign the same UID and GID ranges */ |
979b0ff2 LP |
1081 | userns_fd = userns_acquire(line, line); |
1082 | if (userns_fd < 0) | |
1083 | return log_debug_errno(userns_fd, "Failed to acquire new userns: %m"); | |
35fd3558 LP |
1084 | |
1085 | return TAKE_FD(userns_fd); | |
1086 | } | |
1087 | ||
1088 | int remount_idmap( | |
1089 | const char *p, | |
1090 | uid_t uid_shift, | |
50ae2966 LP |
1091 | uid_t uid_range, |
1092 | RemountIdmapFlags flags) { | |
35fd3558 LP |
1093 | |
1094 | _cleanup_close_ int mount_fd = -1, userns_fd = -1; | |
1095 | int r; | |
1096 | ||
1097 | assert(p); | |
1098 | ||
1099 | if (!userns_shift_range_valid(uid_shift, uid_range)) | |
1100 | return -EINVAL; | |
1101 | ||
1102 | /* Clone the mount point */ | |
1103 | mount_fd = open_tree(-1, p, OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC); | |
1104 | if (mount_fd < 0) | |
1105 | return log_debug_errno(errno, "Failed to open tree of mounted filesystem '%s': %m", p); | |
1106 | ||
1107 | /* Create a user namespace mapping */ | |
50ae2966 | 1108 | userns_fd = make_userns(uid_shift, uid_range, flags); |
35fd3558 LP |
1109 | if (userns_fd < 0) |
1110 | return userns_fd; | |
1111 | ||
1112 | /* Set the user namespace mapping attribute on the cloned mount point */ | |
1113 | if (mount_setattr(mount_fd, "", AT_EMPTY_PATH | AT_RECURSIVE, | |
1114 | &(struct mount_attr) { | |
1115 | .attr_set = MOUNT_ATTR_IDMAP, | |
1116 | .userns_fd = userns_fd, | |
1117 | }, sizeof(struct mount_attr)) < 0) | |
1118 | return log_debug_errno(errno, "Failed to change bind mount attributes for '%s': %m", p); | |
1119 | ||
1120 | /* Remove the old mount point */ | |
1121 | r = umount_verbose(LOG_DEBUG, p, UMOUNT_NOFOLLOW); | |
1122 | if (r < 0) | |
1123 | return r; | |
1124 | ||
1125 | /* And place the cloned version in its place */ | |
1126 | if (move_mount(mount_fd, "", -1, p, MOVE_MOUNT_F_EMPTY_PATH) < 0) | |
1127 | return log_debug_errno(errno, "Failed to attach UID mapped mount to '%s': %m", p); | |
1128 | ||
1129 | return 0; | |
1130 | } | |
9c653536 ZJS |
1131 | |
1132 | int make_mount_point_inode_from_stat(const struct stat *st, const char *dest, mode_t mode) { | |
1133 | assert(st); | |
1134 | assert(dest); | |
1135 | ||
1136 | if (S_ISDIR(st->st_mode)) | |
1137 | return mkdir_label(dest, mode); | |
1138 | else | |
1139 | return mknod(dest, S_IFREG|(mode & ~0111), 0); | |
1140 | } | |
1141 | ||
1142 | int make_mount_point_inode_from_path(const char *source, const char *dest, mode_t mode) { | |
1143 | struct stat st; | |
1144 | ||
1145 | assert(source); | |
1146 | assert(dest); | |
1147 | ||
1148 | if (stat(source, &st) < 0) | |
1149 | return -errno; | |
1150 | ||
1151 | return make_mount_point_inode_from_stat(&st, dest, mode); | |
1152 | } |