]> git.ipfire.org Git - thirdparty/systemd.git/blame - src/shared/mount-util.c
projects / thirdparty / systemd.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge pull request #30284 from YHNdnzj/fstab-wantedby-defaultdeps
[thirdparty/systemd.git] / src / shared / mount-util.c
CommitLineData
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 9#include <linux/loop.h>
3657d3a0 10#if WANT_LINUX_FS_H
35fd3558 11#include <linux/fs.h>
3657d3a0 12#endif
4349cd7c 13
b5efdb8a 14#include "alloc-util.h"
f461a28d 15#include "chase.h"
70599967 16#include "dissect-image.h"
45a68ed3 17#include "exec-util.h"
9e7f941a 18#include "extract-word.h"
4349cd7c
LP		 19#include "fd-util.h"
20#include "fileio.h"
e1873695 21#include "fs-util.h"
e2341b6b 22#include "glyph-util.h"
93cc7779 23#include "hashmap.h"
2e776ed6 24#include "initrd-util.h"
0690160e 25#include "label-util.h"
13dcfe46 26#include "libmount-util.h"
1c092b62 27#include "missing_mount.h"
35fd3558 28#include "missing_syscall.h"
35cd0ba5 29#include "mkdir-label.h"
4349cd7c 30#include "mount-util.h"
049af8ad 31#include "mountpoint-util.h"
2338a175 32#include "namespace-util.h"
4349cd7c
LP		 33#include "parse-util.h"
34#include "path-util.h"
6af52c3a 35#include "process-util.h"
4349cd7c 36#include "set.h"
f63a2c48 37#include "sort-util.h"
28126409 38#include "stat-util.h"
15a5e950 39#include "stdio-util.h"
57c10a56 40#include "string-table.h"
4349cd7c 41#include "string-util.h"
6b7c9f8b 42#include "strv.h"
6af52c3a 43#include "tmpfile-util.h"
70599967 44#include "user-util.h"
4349cd7c 45
84bcb394 46int umount_recursive_full(const char *prefix, int flags, char **keep) {
890c14e3 47 _cleanup_fclose_ FILE *f = NULL;
4349cd7c
LP		 48 int n = 0, r;
49
9d0619de
LP		 50 /* Try to umount everything recursively below a directory. Also, take care of stacked mounts, and
51 * keep unmounting them until they are gone. */
4349cd7c 52
890c14e3
LP		 53 f = fopen("/proc/self/mountinfo", "re"); /* Pin the file, in case we unmount /proc/ as part of the logic here */
54 if (!f)
55 return log_debug_errno(errno, "Failed to open /proc/self/mountinfo: %m");
56
57 for (;;) {
13dcfe46
ZJS		 58 _cleanup_(mnt_free_tablep) struct libmnt_table *table = NULL;
59 _cleanup_(mnt_free_iterp) struct libmnt_iter *iter = NULL;
890c14e3 60 bool again = false;
4349cd7c 61
890c14e3 62 r = libmount_parse("/proc/self/mountinfo", f, &table, &iter);
fdeea3f4 63 if (r < 0)
13dcfe46 64 return log_debug_errno(r, "Failed to parse /proc/self/mountinfo: %m");
35bbbf85 65
4349cd7c 66 for (;;) {
84bcb394 67 bool shall_keep = false;
13dcfe46
ZJS		 68 struct libmnt_fs *fs;
69 const char *path;
4349cd7c 70
13dcfe46
ZJS		 71 r = mnt_table_next_fs(table, iter, &fs);
72 if (r == 1)
73 break;
74 if (r < 0)
75 return log_debug_errno(r, "Failed to get next entry from /proc/self/mountinfo: %m");
4349cd7c 76
13dcfe46
ZJS		 77 path = mnt_fs_get_target(fs);
78 if (!path)
79 continue;
4349cd7c 80
ef742415 81 if (prefix && !path_startswith(path, prefix)) {
62c505e1 82 log_trace("Not unmounting %s, outside of prefix: %s", path, prefix);
4349cd7c 83 continue;
ef742415 84 }
4349cd7c 85
84bcb394
LP		 86 STRV_FOREACH(k, keep)
87 /* Match against anything in the path to the dirs to keep, or below the dirs to keep */
88 if (path_startswith(path, *k) || path_startswith(*k, path)) {
89 shall_keep = true;
90 break;
91 }
92 if (shall_keep) {
93 log_debug("Not unmounting %s, referenced by keep list.", path);
94 continue;
95 }
96
827ea521
LP		 97 if (umount2(path, flags | UMOUNT_NOFOLLOW) < 0) {
98 log_debug_errno(errno, "Failed to umount %s, ignoring: %m", path);
4349cd7c
LP		 99 continue;
100 }
101
30868c1c 102 log_trace("Successfully unmounted %s", path);
6b7c9f8b 103
4349cd7c
LP		 104 again = true;
105 n++;
106
107 break;
108 }
890c14e3
LP		 109
110 if (!again)
111 break;
112
113 rewind(f);
114 }
4349cd7c 115
13dcfe46 116 return n;
4349cd7c
LP		 117}
118
4f5644db
LP		 119#define MS_CONVERTIBLE_FLAGS (MS_RDONLY|MS_NOSUID|MS_NODEV|MS_NOEXEC|MS_NOSYMFOLLOW)
120
121static uint64_t ms_flags_to_mount_attr(unsigned long a) {
122 uint64_t f = 0;
123
124 if (FLAGS_SET(a, MS_RDONLY))
125 f |= MOUNT_ATTR_RDONLY;
126
127 if (FLAGS_SET(a, MS_NOSUID))
128 f |= MOUNT_ATTR_NOSUID;
129
130 if (FLAGS_SET(a, MS_NODEV))
131 f |= MOUNT_ATTR_NODEV;
132
133 if (FLAGS_SET(a, MS_NOEXEC))
134 f |= MOUNT_ATTR_NOEXEC;
135
136 if (FLAGS_SET(a, MS_NOSYMFOLLOW))
137 f |= MOUNT_ATTR_NOSYMFOLLOW;
138
139 return f;
140}
141
142static bool skip_mount_set_attr = false;
143
be3f3752 144/* Use this function only if you do not have direct access to /proc/self/mountinfo but the caller can open it
64e82c19
LP		 145 * for you. This is the case when /proc is masked or not mounted. Otherwise, use bind_remount_recursive. */
146int bind_remount_recursive_with_mountinfo(
147 const char *prefix,
148 unsigned long new_flags,
149 unsigned long flags_mask,
6b000af4 150 char **deny_list,
64e82c19
LP		 151 FILE *proc_self_mountinfo) {
152
0289948e 153 _cleanup_fclose_ FILE *proc_self_mountinfo_opened = NULL;
ba8dced2 154 _cleanup_set_free_ Set *done = NULL;
670e8efd 155 unsigned n_tries = 0;
4349cd7c
LP		 156 int r;
157
8403219f 158 assert(prefix);
ac9de0b3 159
874052c5
LP		 160 if ((flags_mask & ~MS_CONVERTIBLE_FLAGS) == 0 && strv_isempty(deny_list) && !skip_mount_set_attr) {
161 /* Let's take a shortcut for all the flags we know how to convert into mount_setattr() flags */
162
163 if (mount_setattr(AT_FDCWD, prefix, AT_SYMLINK_NOFOLLOW|AT_RECURSIVE,
164 &(struct mount_attr) {
165 .attr_set = ms_flags_to_mount_attr(new_flags & flags_mask),
166 .attr_clr = ms_flags_to_mount_attr(~new_flags & flags_mask),
167 }, MOUNT_ATTR_SIZE_VER0) < 0) {
168
169 log_debug_errno(errno, "mount_setattr() failed, falling back to classic remounting: %m");
170
171 /* We fall through to classic behaviour if not supported (i.e. kernel < 5.12). We
172 * also do this for all other kinds of errors since they are so many different, and
173 * mount_setattr() has no graceful mode where it continues despite seeing errors one
174 * some mounts, but we want that. Moreover mount_setattr() only works on the mount
175 * point inode itself, not a non-mount point inode, and we want to support arbitrary
176 * prefixes here. */
177
178 if (ERRNO_IS_NOT_SUPPORTED(errno)) /* if not supported, then don't bother at all anymore */
179 skip_mount_set_attr = true;
180 } else
181 return 0; /* Nice, this worked! */
182 }
183
0289948e
LP		 184 if (!proc_self_mountinfo) {
185 r = fopen_unlocked("/proc/self/mountinfo", "re", &proc_self_mountinfo_opened);
186 if (r < 0)
187 return r;
188
189 proc_self_mountinfo = proc_self_mountinfo_opened;
190 }
191
ddc155b2
TM		 192 /* Recursively remount a directory (and all its submounts) with desired flags (MS_READONLY,
193 * MS_NOSUID, MS_NOEXEC). If the directory is already mounted, we reuse the mount and simply mark it
194 * MS_BIND|MS_RDONLY (or remove the MS_RDONLY for read-write operation), ditto for other flags. If it
195 * isn't we first make it one. Afterwards we apply (or remove) the flags to all submounts we can
196 * access, too. When mounts are stacked on the same mount point we only care for each individual
197 * "top-level" mount on each point, as we cannot influence/access the underlying mounts anyway. We do
198 * not have any effect on future submounts that might get propagated, they might be writable
4b6ef527
LP		 199 * etc. This includes future submounts that have been triggered via autofs. Also note that we can't
200 * operate atomically here. Mounts established while we process the tree might or might not get
201 * noticed and thus might or might not be covered.
6b7c9f8b 202 *
6b000af4
LP		 203 * If the "deny_list" parameter is specified it may contain a list of subtrees to exclude from the
204 * remount operation. Note that we'll ignore the deny list for the top-level path. */
4349cd7c 205
4349cd7c 206 for (;;) {
13dcfe46
ZJS		 207 _cleanup_(mnt_free_tablep) struct libmnt_table *table = NULL;
208 _cleanup_(mnt_free_iterp) struct libmnt_iter *iter = NULL;
ba8dced2 209 _cleanup_hashmap_free_ Hashmap *todo = NULL;
4349cd7c 210 bool top_autofs = false;
4349cd7c 211
670e8efd
LP		 212 if (n_tries++ >= 32) /* Let's not retry this loop forever */
213 return -EBUSY;
214
ac9de0b3 215 rewind(proc_self_mountinfo);
4349cd7c 216
e2857b3d 217 r = libmount_parse("/proc/self/mountinfo", proc_self_mountinfo, &table, &iter);
13dcfe46
ZJS		 218 if (r < 0)
219 return log_debug_errno(r, "Failed to parse /proc/self/mountinfo: %m");
4349cd7c 220
13dcfe46 221 for (;;) {
ba8dced2
LP		 222 _cleanup_free_ char *d = NULL;
223 const char *path, *type, *opts;
224 unsigned long flags = 0;
13dcfe46 225 struct libmnt_fs *fs;
4349cd7c 226
13dcfe46 227 r = mnt_table_next_fs(table, iter, &fs);
d6bfab11 228 if (r == 1) /* EOF */
13dcfe46 229 break;
4349cd7c 230 if (r < 0)
13dcfe46 231 return log_debug_errno(r, "Failed to get next entry from /proc/self/mountinfo: %m");
4349cd7c 232
13dcfe46 233 path = mnt_fs_get_target(fs);
d6bfab11 234 if (!path)
6b7c9f8b
LP		 235 continue;
236
c6111b85 237 if (!path_startswith(path, prefix))
13dcfe46
ZJS		 238 continue;
239
d6bfab11
LP		 240 type = mnt_fs_get_fstype(fs);
241 if (!type)
242 continue;
243
244 /* Let's ignore autofs mounts. If they aren't triggered yet, we want to avoid
245 * triggering them, as we don't make any guarantees for future submounts anyway. If
246 * they are already triggered, then we will find another entry for this. */
247 if (streq(type, "autofs")) {
248 top_autofs = top_autofs || path_equal(path, prefix);
249 continue;
250 }
251
252 if (set_contains(done, path))
253 continue;
254
6b000af4 255 /* Ignore this mount if it is deny-listed, but only if it isn't the top-level mount
13dcfe46 256 * we shall operate on. */
c6111b85 257 if (!path_equal(path, prefix)) {
6b000af4 258 bool deny_listed = false;
6b7c9f8b 259
6b000af4 260 STRV_FOREACH(i, deny_list) {
c6111b85 261 if (path_equal(*i, prefix))
6b7c9f8b
LP		 262 continue;
263
c6111b85 264 if (!path_startswith(*i, prefix))
6b7c9f8b
LP		 265 continue;
266
13dcfe46 267 if (path_startswith(path, *i)) {
6b000af4 268 deny_listed = true;
61f695f4 269 log_trace("Not remounting %s deny-listed by %s, called for %s", path, *i, prefix);
6b7c9f8b
LP		 270 break;
271 }
272 }
d6bfab11 273
6b000af4 274 if (deny_listed)
6b7c9f8b
LP		 275 continue;
276 }
277
ba8dced2
LP		 278 opts = mnt_fs_get_vfs_options(fs);
279 if (opts) {
280 r = mnt_optstr_get_flags(opts, &flags, mnt_get_builtin_optmap(MNT_LINUX_MAP));
281 if (r < 0)
282 log_debug_errno(r, "Could not get flags for '%s', ignoring: %m", path);
283 }
284
285 d = strdup(path);
286 if (!d)
287 return -ENOMEM;
288
289 r = hashmap_ensure_put(&todo, &path_hash_ops_free, d, ULONG_TO_PTR(flags));
290 if (r == -EEXIST)
e0103063
LB		 291 /* If the same path was recorded, but with different mount flags, update it:
292 * it means a mount point is overmounted, and libmount returns the "bottom" (or
293 * older one) first, but we want to reapply the flags from the "top" (or newer
294 * one). See: https://github.com/systemd/systemd/issues/20032
295 * Note that this shouldn't really fail, as we were just told that the key
296 * exists, and it's an update so we want 'd' to be freed immediately. */
297 r = hashmap_update(todo, d, ULONG_TO_PTR(flags));
d6bfab11
LP		 298 if (r < 0)
299 return r;
ba8dced2
LP		 300 if (r > 0)
301 TAKE_PTR(d);
4349cd7c
LP		 302 }
303
5c5753b9
LP		 304 /* Check if the top-level directory was among what we have seen so far. For that check both
305 * 'done' and 'todo'. Also check 'top_autofs' because if the top-level dir is an autofs we'll
306 * not include it in either set but will set this bool. */
c6111b85 307 if (!set_contains(done, prefix) &&
ba8dced2 308 !(top_autofs || hashmap_contains(todo, prefix))) {
5c5753b9 309
6b7c9f8b 310 /* The prefix directory itself is not yet a mount, make it one. */
c6111b85 311 r = mount_nofollow(prefix, prefix, NULL, MS_BIND|MS_REC, NULL);
511a8cfe
LP		 312 if (r < 0)
313 return r;
4349cd7c 314
5c5753b9
LP		 315 /* Immediately rescan, so that we pick up the new mount's flags */
316 continue;
4349cd7c
LP		 317 }
318
5c5753b9 319 /* If we have no submounts to process anymore, we are done */
ba8dced2 320 if (hashmap_isempty(todo))
5c5753b9
LP		 321 return 0;
322
ba8dced2
LP		 323 for (;;) {
324 unsigned long flags;
325 char *x = NULL;
326
327 /* Take the first mount from our list of mounts to still process */
328 flags = PTR_TO_ULONG(hashmap_steal_first_key_and_value(todo, (void**) &x));
329 if (!x)
330 break;
4349cd7c 331
ba8dced2 332 r = set_ensure_consume(&done, &path_hash_ops_free, x);
4c701096 333 if (IN_SET(r, 0, -EEXIST))
ba8dced2 334 continue; /* Already done */
4349cd7c
LP		 335 if (r < 0)
336 return r;
337
ba8dced2
LP		 338 /* Now, remount this with the new flags set, but exclude MS_RELATIME from it. (It's
339 * the default anyway, thus redundant, and in userns we'll get an error if we try to
340 * explicitly enable it) */
341 r = mount_nofollow(NULL, x, NULL, ((flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags) & ~MS_RELATIME, NULL);
065b4774 342 if (r < 0) {
ba8dced2
LP		 343 int q;
344
345 /* OK, so the remount of this entry failed. We'll ultimately ignore this in
346 * almost all cases (there are simply so many reasons why this can fail,
347 * think autofs, NFS, FUSE, …), but let's generate useful debug messages at
348 * the very least. */
349
350 q = path_is_mount_point(x, NULL, 0);
351 if (IN_SET(q, 0, -ENOENT)) {
352 /* Hmm, whaaaa? The mount point is not actually a mount point? Then
353 * it is either obstructed by a later mount or somebody has been
354 * racing against us and removed it. Either way the mount point
355 * doesn't matter to us, let's ignore it hence. */
356 log_debug_errno(r, "Mount point '%s' to remount is not a mount point anymore, ignoring remount failure: %m", x);
357 continue;
358 }
359 if (q < 0) /* Any other error on this? Just log and continue */
360 log_debug_errno(q, "Failed to determine whether '%s' is a mount point or not, ignoring: %m", x);
361
362 if (((flags ^ new_flags) & flags_mask & ~MS_RELATIME) == 0) { /* ignore MS_RELATIME while comparing */
363 log_debug_errno(r, "Couldn't remount '%s', but the flags already match what we want, hence ignoring: %m", x);
364 continue;
365 }
366
367 /* Make this fatal if this is the top-level mount */
368 if (path_equal(x, prefix))
065b4774
LP		 369 return r;
370
ba8dced2
LP		 371 /* If this is not the top-level mount, then handle this gracefully: log but
372 * otherwise ignore. With NFS, FUSE, autofs there are just too many reasons
373 * this might fail without a chance for us to do anything about it, let's
374 * hence be strict on the top-level mount and lenient on the inner ones. */
375 log_debug_errno(r, "Couldn't remount submount '%s' for unexpected reason, ignoring: %m", x);
ef454fd1
YW		 376 continue;
377 }
98df8089 378
30868c1c 379 log_trace("Remounted %s.", x);
4349cd7c
LP		 380 }
381 }
382}
383
7cce68e1
LP		 384int bind_remount_one_with_mountinfo(
385 const char *path,
386 unsigned long new_flags,
387 unsigned long flags_mask,
388 FILE *proc_self_mountinfo) {
389
390 _cleanup_(mnt_free_tablep) struct libmnt_table *table = NULL;
2c5ff8ea
LP		 391 unsigned long flags = 0;
392 struct libmnt_fs *fs;
393 const char *opts;
7cce68e1
LP		 394 int r;
395
396 assert(path);
397 assert(proc_self_mountinfo);
398
4f5644db
LP		 399 if ((flags_mask & ~MS_CONVERTIBLE_FLAGS) == 0 && !skip_mount_set_attr) {
400 /* Let's take a shortcut for all the flags we know how to convert into mount_setattr() flags */
401
402 if (mount_setattr(AT_FDCWD, path, AT_SYMLINK_NOFOLLOW,
403 &(struct mount_attr) {
404 .attr_set = ms_flags_to_mount_attr(new_flags & flags_mask),
405 .attr_clr = ms_flags_to_mount_attr(~new_flags & flags_mask),
406 }, MOUNT_ATTR_SIZE_VER0) < 0) {
407
408 log_debug_errno(errno, "mount_setattr() didn't work, falling back to classic remounting: %m");
409
410 if (ERRNO_IS_NOT_SUPPORTED(errno)) /* if not supported, then don't bother at all anymore */
411 skip_mount_set_attr = true;
412 } else
413 return 0; /* Nice, this worked! */
414 }
415
7cce68e1
LP		 416 rewind(proc_self_mountinfo);
417
418 table = mnt_new_table();
419 if (!table)
420 return -ENOMEM;
421
422 r = mnt_table_parse_stream(table, proc_self_mountinfo, "/proc/self/mountinfo");
423 if (r < 0)
424 return r;
425
2c5ff8ea 426 fs = mnt_table_find_target(table, path, MNT_ITER_FORWARD);
0338df47
LP		 427 if (!fs) {
428 if (laccess(path, F_OK) < 0) /* Hmm, it's not in the mount table, but does it exist at all? */
429 return -errno;
430
2c5ff8ea 431 return -EINVAL; /* Not a mount point we recognize */
0338df47 432 }
2c5ff8ea
LP		 433
434 opts = mnt_fs_get_vfs_options(fs);
435 if (opts) {
436 r = mnt_optstr_get_flags(opts, &flags, mnt_get_builtin_optmap(MNT_LINUX_MAP));
437 if (r < 0)
438 log_debug_errno(r, "Could not get flags for '%s', ignoring: %m", path);
439 }
7cce68e1 440
2c5ff8ea 441 r = mount_nofollow(NULL, path, NULL, ((flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags) & ~MS_RELATIME, NULL);
b23c6a64
LP		 442 if (r < 0) {
443 if (((flags ^ new_flags) & flags_mask & ~MS_RELATIME) != 0) /* Ignore MS_RELATIME again,
444 * since kernel adds it in
445 * everywhere, because it's the
446 * default. */
447 return r;
448
449 /* Let's handle redundant remounts gracefully */
450 log_debug_errno(r, "Failed to remount '%s' but flags already match what we want, ignoring: %m", path);
451 }
7cce68e1
LP		 452
453 return 0;
454}
455
ea0f3289
LP		 456static int mount_switch_root_pivot(int fd_newroot, const char *path) {
457 assert(fd_newroot >= 0);
458 assert(path);
2e776ed6 459
2e776ed6
CB		 460 /* Let the kernel tuck the new root under the old one. */
461 if (pivot_root(".", ".") < 0)
462 return log_debug_errno(errno, "Failed to pivot root to new rootfs '%s': %m", path);
463
ea0f3289
LP		 464 /* Get rid of the old root and reveal our brand new root. (This will always operate on the top-most
465 * mount on our cwd, regardless what our current directory actually points to.) */
2e776ed6
CB		 466 if (umount2(".", MNT_DETACH) < 0)
467 return log_debug_errno(errno, "Failed to unmount old rootfs: %m");
468
2e776ed6
CB		 469 return 0;
470}
471
ea0f3289
LP		 472static int mount_switch_root_move(int fd_newroot, const char *path) {
473 assert(fd_newroot >= 0);
474 assert(path);
475
ea0f3289
LP		 476 /* Move the new root fs */
477 if (mount(".", "/", NULL, MS_MOVE, NULL) < 0)
57c10a56
CB		 478 return log_debug_errno(errno, "Failed to move new rootfs '%s': %m", path);
479
12cc9601 480 /* Also change root dir */
57c10a56
CB		 481 if (chroot(".") < 0)
482 return log_debug_errno(errno, "Failed to chroot to new rootfs '%s': %m", path);
483
57c10a56
CB		 484 return 0;
485}
486
ea0f3289 487int mount_switch_root_full(const char *path, unsigned long mount_propagation_flag, bool force_ms_move) {
57c10a56 488 _cleanup_close_ int fd_newroot = -EBADF;
bb59b922 489 int r, is_current_root;
57c10a56
CB		 490
491 assert(path);
9d50f850 492 assert(mount_propagation_flag_is_valid(mount_propagation_flag));
57c10a56
CB		 493
494 fd_newroot = open(path, O_PATH|O_DIRECTORY|O_CLOEXEC|O_NOFOLLOW);
495 if (fd_newroot < 0)
496 return log_debug_errno(errno, "Failed to open new rootfs '%s': %m", path);
497
bb59b922
LP		 498 is_current_root = path_is_root_at(fd_newroot, NULL);
499 if (is_current_root < 0)
500 return log_debug_errno(is_current_root, "Failed to determine if target dir is our root already: %m");
501
502 /* Change into the new rootfs. */
503 if (fchdir(fd_newroot) < 0)
504 return log_debug_errno(errno, "Failed to chdir into new rootfs '%s': %m", path);
505
506 /* Make this a NOP if we are supposed to switch to our current root fs. After all, both pivot_root()
507 * and MS_MOVE don't like that. */
508 if (!is_current_root) {
509 if (!force_ms_move) {
510 r = mount_switch_root_pivot(fd_newroot, path);
511 if (r < 0) {
512 log_debug_errno(r, "Failed to pivot into new rootfs '%s', will try to use MS_MOVE instead: %m", path);
513 force_ms_move = true;
514 }
515 }
516 if (force_ms_move) {
517 /* Failed to pivot_root() fallback to MS_MOVE. For example, this may happen if the rootfs is
518 * an initramfs in which case pivot_root() isn't supported. */
519 r = mount_switch_root_move(fd_newroot, path);
520 if (r < 0)
521 return log_debug_errno(r, "Failed to switch to new rootfs '%s' with MS_MOVE: %m", path);
ea0f3289 522 }
57c10a56 523 }
57c10a56 524
9d50f850
YW		 525 /* Finally, let's establish the requested propagation flags. */
526 if (mount_propagation_flag == 0)
527 return 0;
528
529 if (mount(NULL, ".", NULL, mount_propagation_flag | MS_REC, 0) < 0)
57c10a56 530 return log_debug_errno(errno, "Failed to turn new rootfs '%s' into %s mount: %m",
9d50f850 531 mount_propagation_flag_to_string(mount_propagation_flag), path);
57c10a56
CB		 532
533 return 0;
534}
2e776ed6 535
3f2c0bec
LP		 536int repeat_unmount(const char *path, int flags) {
537 bool done = false;
538
539 assert(path);
540
541 /* If there are multiple mounts on a mount point, this
542 * removes them all */
543
544 for (;;) {
545 if (umount2(path, flags) < 0) {
546
547 if (errno == EINVAL)
548 return done;
549
550 return -errno;
551 }
552
553 done = true;
554 }
555}
c4b41707 556
48b747fa
LP		 557int mode_to_inaccessible_node(
558 const char *runtime_dir,
559 mode_t mode,
560 char **ret) {
561
562 /* This function maps a node type to a corresponding inaccessible file node. These nodes are created
563 * during early boot by PID 1. In some cases we lacked the privs to create the character and block
564 * devices (maybe because we run in an userns environment, or miss CAP_SYS_MKNOD, or run with a
565 * devices policy that excludes device nodes with major and minor of 0), but that's fine, in that
566 * case we use an AF_UNIX file node instead, which is not the same, but close enough for most
567 * uses. And most importantly, the kernel allows bind mounts from socket nodes to any non-directory
568 * file nodes, and that's the most important thing that matters.
569 *
570 * Note that the runtime directory argument shall be the top-level runtime directory, i.e. /run/ if
571 * we operate in system context and $XDG_RUNTIME_DIR if we operate in user context. */
572
e5f10caf 573 _cleanup_free_ char *d = NULL;
11e2be3a 574 const char *node;
e5f10caf 575
48b747fa
LP		 576 assert(ret);
577
578 if (!runtime_dir)
579 runtime_dir = "/run";
fe80fcc7 580
11e2be3a
LP		 581 if (S_ISLNK(mode))
582 return -EINVAL;
fe80fcc7 583
11e2be3a 584 node = inode_type_to_string(mode);
e5f10caf
AZ		 585 if (!node)
586 return -EINVAL;
587
11e2be3a 588 d = path_join(runtime_dir, "systemd/inaccessible", node);
48b747fa
LP		 589 if (!d)
590 return -ENOMEM;
591
cbed1dc8
LP		 592 /* On new kernels unprivileged users are permitted to create 0:0 char device nodes (because they also
593 * act as whiteout inode for overlayfs), but no other char or block device nodes. On old kernels no
594 * device node whatsoever may be created by unprivileged processes. Hence, if the caller asks for the
595 * inaccessible block device node let's see if the block device node actually exists, and if not,
596 * fall back to the character device node. From there fall back to the socket device node. This means
597 * in the best case we'll get the right device node type — but if not we'll hopefully at least get a
598 * device node at all. */
599
600 if (S_ISBLK(mode) &&
601 access(d, F_OK) < 0 && errno == ENOENT) {
602 free(d);
603 d = path_join(runtime_dir, "/systemd/inaccessible/chr");
604 if (!d)
605 return -ENOMEM;
606 }
607
608 if (IN_SET(mode & S_IFMT, S_IFBLK, S_IFCHR) &&
609 access(d, F_OK) < 0 && errno == ENOENT) {
48b747fa
LP		 610 free(d);
611 d = path_join(runtime_dir, "/systemd/inaccessible/sock");
612 if (!d)
613 return -ENOMEM;
614 }
e5f10caf 615
48b747fa 616 *ret = TAKE_PTR(d);
e5f10caf 617 return 0;
c4b41707 618}
60e76d48 619
da185cd0 620int mount_flags_to_string(unsigned long flags, char **ret) {
1c092b62 621 static const struct {
da185cd0 622 unsigned long flag;
1c092b62
YW		 623 const char *name;
624 } map[] = {
625 { .flag = MS_RDONLY, .name = "MS_RDONLY", },
626 { .flag = MS_NOSUID, .name = "MS_NOSUID", },
627 { .flag = MS_NODEV, .name = "MS_NODEV", },
628 { .flag = MS_NOEXEC, .name = "MS_NOEXEC", },
629 { .flag = MS_SYNCHRONOUS, .name = "MS_SYNCHRONOUS", },
630 { .flag = MS_REMOUNT, .name = "MS_REMOUNT", },
631 { .flag = MS_MANDLOCK, .name = "MS_MANDLOCK", },
632 { .flag = MS_DIRSYNC, .name = "MS_DIRSYNC", },
633 { .flag = MS_NOSYMFOLLOW, .name = "MS_NOSYMFOLLOW", },
634 { .flag = MS_NOATIME, .name = "MS_NOATIME", },
635 { .flag = MS_NODIRATIME, .name = "MS_NODIRATIME", },
636 { .flag = MS_BIND, .name = "MS_BIND", },
637 { .flag = MS_MOVE, .name = "MS_MOVE", },
638 { .flag = MS_REC, .name = "MS_REC", },
639 { .flag = MS_SILENT, .name = "MS_SILENT", },
640 { .flag = MS_POSIXACL, .name = "MS_POSIXACL", },
641 { .flag = MS_UNBINDABLE, .name = "MS_UNBINDABLE", },
642 { .flag = MS_PRIVATE, .name = "MS_PRIVATE", },
643 { .flag = MS_SLAVE, .name = "MS_SLAVE", },
644 { .flag = MS_SHARED, .name = "MS_SHARED", },
645 { .flag = MS_RELATIME, .name = "MS_RELATIME", },
646 { .flag = MS_KERNMOUNT, .name = "MS_KERNMOUNT", },
647 { .flag = MS_I_VERSION, .name = "MS_I_VERSION", },
648 { .flag = MS_STRICTATIME, .name = "MS_STRICTATIME", },
649 { .flag = MS_LAZYTIME, .name = "MS_LAZYTIME", },
650 };
651 _cleanup_free_ char *str = NULL;
652
4bee2333
YW		 653 assert(ret);
654
1c092b62
YW		 655 for (size_t i = 0; i < ELEMENTSOF(map); i++)
656 if (flags & map[i].flag) {
657 if (!strextend_with_separator(&str, "|", map[i].name))
658 return -ENOMEM;
659 flags &= ~map[i].flag;
660 }
661
662 if (!str || flags != 0)
663 if (strextendf_with_separator(&str, "|", "%lx", flags) < 0)
664 return -ENOMEM;
665
666 *ret = TAKE_PTR(str);
667 return 0;
60e76d48
ZJS		 668}
669
511a8cfe 670int mount_verbose_full(
60e76d48
ZJS		 671 int error_log_level,
672 const char *what,
673 const char *where,
674 const char *type,
675 unsigned long flags,
511a8cfe
LP		 676 const char *options,
677 bool follow_symlink) {
60e76d48 678
6ef8df2b
YW		 679 _cleanup_free_ char *fl = NULL, *o = NULL;
680 unsigned long f;
681 int r;
682
683 r = mount_option_mangle(options, flags, &f, &o);
684 if (r < 0)
685 return log_full_errno(error_log_level, r,
686 "Failed to mangle mount options %s: %m",
687 strempty(options));
60e76d48 688
1c092b62 689 (void) mount_flags_to_string(f, &fl);
60e76d48 690
f56f9c6b
LP		 691 if (FLAGS_SET(f, MS_REMOUNT|MS_BIND))
692 log_debug("Changing mount flags %s (%s \"%s\")...",
6ef8df2b 693 where, strnull(fl), strempty(o));
f56f9c6b
LP		 694 else if (f & MS_REMOUNT)
695 log_debug("Remounting superblock %s (%s \"%s\")...",
6ef8df2b 696 where, strnull(fl), strempty(o));
f56f9c6b
LP		 697 else if (f & (MS_SHARED|MS_PRIVATE|MS_SLAVE|MS_UNBINDABLE))
698 log_debug("Changing mount propagation %s (%s \"%s\")",
699 where, strnull(fl), strempty(o));
700 else if (f & MS_BIND)
60e76d48 701 log_debug("Bind-mounting %s on %s (%s \"%s\")...",
6ef8df2b
YW		 702 what, where, strnull(fl), strempty(o));
703 else if (f & MS_MOVE)
e2341b6b
DT		 704 log_debug("Moving mount %s %s %s (%s \"%s\")...",
705 what, special_glyph(SPECIAL_GLYPH_ARROW_RIGHT), where, strnull(fl), strempty(o));
60e76d48 706 else
3b493d94
LP		 707 log_debug("Mounting %s (%s) on %s (%s \"%s\")...",
708 strna(what), strna(type), where, strnull(fl), strempty(o));
511a8cfe
LP		 709
710 if (follow_symlink)
7c248223 711 r = RET_NERRNO(mount(what, where, type, f, o));
511a8cfe
LP		 712 else
713 r = mount_nofollow(what, where, type, f, o);
714 if (r < 0)
715 return log_full_errno(error_log_level, r,
3ccf6126
LP		 716 "Failed to mount %s (type %s) on %s (%s \"%s\"): %m",
717 strna(what), strna(type), where, strnull(fl), strempty(o));
60e76d48
ZJS		 718 return 0;
719}
720
30f5d104
LP		 721int umount_verbose(
722 int error_log_level,
723 const char *what,
724 int flags) {
725
726 assert(what);
727
60e76d48 728 log_debug("Umounting %s...", what);
30f5d104
LP		 729
730 if (umount2(what, flags) < 0)
731 return log_full_errno(error_log_level, errno,
732 "Failed to unmount %s: %m", what);
733
60e76d48
ZJS		 734 return 0;
735}
83555251 736
7c83d42e
LB		 737int mount_exchange_graceful(int fsmount_fd, const char *dest, bool mount_beneath) {
738 int r;
739
740 assert(fsmount_fd >= 0);
741 assert(dest);
742
743 /* First, try to mount beneath an existing mount point, and if that works, umount the old mount,
744 * which is now at the top. This will ensure we can atomically replace a mount. Note that this works
745 * also in the case where there are submounts down the tree. Mount propagation is allowed but
746 * restricted to layouts that don't end up propagation the new mount on top of the mount stack. If
747 * this is not supported (minimum kernel v6.5), or if there is no mount on the mountpoint, we get
748 * -EINVAL and then we fallback to normal mounting. */
749
750 r = RET_NERRNO(move_mount(
751 fsmount_fd,
752 /* from_path= */ "",
753 /* to_fd= */ -EBADF,
754 dest,
755 MOVE_MOUNT_F_EMPTY_PATH | (mount_beneath ? MOVE_MOUNT_BENEATH : 0)));
756 if (mount_beneath) {
757 if (r == -EINVAL) { /* Fallback if mount_beneath is not supported */
758 log_debug_errno(r,
759 "Failed to mount beneath '%s', falling back to overmount",
760 dest);
761 return RET_NERRNO(move_mount(
762 fsmount_fd,
763 /* from_path= */ "",
764 /* to_fd= */ -EBADF,
765 dest,
766 MOVE_MOUNT_F_EMPTY_PATH));
767 }
768
769 if (r >= 0) /* If it is, now remove the old mount */
770 return umount_verbose(LOG_DEBUG, dest, UMOUNT_NOFOLLOW|MNT_DETACH);
771 }
772
773 return r;
774}
775
9e7f941a
YW		 776int mount_option_mangle(
777 const char *options,
778 unsigned long mount_flags,
779 unsigned long *ret_mount_flags,
780 char **ret_remaining_options) {
781
782 const struct libmnt_optmap *map;
783 _cleanup_free_ char *ret = NULL;
9e7f941a
YW		 784 int r;
785
3ca4ec20 786 /* This extracts mount flags from the mount options, and stores
9e7f941a
YW		 787 * non-mount-flag options to '*ret_remaining_options'.
788 * E.g.,
9f563f27 789 * "rw,nosuid,nodev,relatime,size=1630748k,mode=0700,uid=1000,gid=1000"
9e7f941a 790 * is split to MS_NOSUID|MS_NODEV|MS_RELATIME and
9f563f27 791 * "size=1630748k,mode=0700,uid=1000,gid=1000".
3ca4ec20 792 * See more examples in test-mount-util.c.
9e7f941a 793 *
3ca4ec20 794 * If 'options' does not contain any non-mount-flag options,
5238e957 795 * then '*ret_remaining_options' is set to NULL instead of empty string.
3ca4ec20
ZJS		 796 * The validity of options stored in '*ret_remaining_options' is not checked.
797 * If 'options' is NULL, this just copies 'mount_flags' to *ret_mount_flags. */
9e7f941a
YW		 798
799 assert(ret_mount_flags);
800 assert(ret_remaining_options);
801
802 map = mnt_get_builtin_optmap(MNT_LINUX_MAP);
803 if (!map)
804 return -EINVAL;
805
25086b4c 806 for (const char *p = options;;) {
9e7f941a
YW		 807 _cleanup_free_ char *word = NULL;
808 const struct libmnt_optmap *ent;
809
9b23679e 810 r = extract_first_word(&p, &word, ",", EXTRACT_KEEP_QUOTE);
9e7f941a
YW		 811 if (r < 0)
812 return r;
813 if (r == 0)
814 break;
815
816 for (ent = map; ent->name; ent++) {
817 /* All entries in MNT_LINUX_MAP do not take any argument.
818 * Thus, ent->name does not contain "=" or "[=]". */
819 if (!streq(word, ent->name))
820 continue;
821
822 if (!(ent->mask & MNT_INVERT))
823 mount_flags |= ent->id;
824 else if (mount_flags & ent->id)
825 mount_flags ^= ent->id;
826
827 break;
828 }
829
830 /* If 'word' is not a mount flag, then store it in '*ret_remaining_options'. */
ac6086fd
LB		 831 if (!ent->name &&
832 !startswith_no_case(word, "x-") &&
833 !strextend_with_separator(&ret, ",", word))
9e7f941a
YW		 834 return -ENOMEM;
835 }
836
837 *ret_mount_flags = mount_flags;
ae2a15bc 838 *ret_remaining_options = TAKE_PTR(ret);
9e7f941a
YW		 839
840 return 0;
841}
6af52c3a 842
a7e0199e
LB		 843static int mount_in_namespace_legacy(
844 const char *chased_src_path,
845 int chased_src_fd,
846 struct stat *chased_src_st,
6af52c3a
LB		 847 const char *propagate_path,
848 const char *incoming_path,
6af52c3a 849 const char *dest,
a7e0199e
LB		 850 int pidns_fd,
851 int mntns_fd,
852 int root_fd,
6af52c3a 853 bool read_only,
70599967
LB		 854 bool make_file_or_directory,
855 const MountOptions *options,
84be0c71 856 const ImagePolicy *image_policy,
70599967 857 bool is_image) {
6af52c3a 858
71136404 859 _cleanup_close_pair_ int errno_pipe_fd[2] = EBADF_PAIR;
ddb6eeaf 860 char mount_slave[] = "/tmp/propagate.XXXXXX", *mount_tmp, *mount_outside, *p;
6af52c3a
LB		 861 bool mount_slave_created = false, mount_slave_mounted = false,
862 mount_tmp_created = false, mount_tmp_mounted = false,
863 mount_outside_created = false, mount_outside_mounted = false;
6af52c3a
LB		 864 pid_t child;
865 int r;
866
a7e0199e
LB		 867 assert(chased_src_path);
868 assert(chased_src_fd >= 0);
869 assert(chased_src_st);
6af52c3a
LB		 870 assert(propagate_path);
871 assert(incoming_path);
6af52c3a 872 assert(dest);
a7e0199e
LB		 873 assert(pidns_fd >= 0);
874 assert(mntns_fd >= 0);
875 assert(root_fd >= 0);
70599967 876 assert(!options || is_image);
6af52c3a 877
6af52c3a
LB		 878 p = strjoina(propagate_path, "/");
879 r = laccess(p, F_OK);
880 if (r < 0)
881 return log_debug_errno(r == -ENOENT ? SYNTHETIC_ERRNO(EOPNOTSUPP) : r, "Target does not allow propagation of mount points");
882
6af52c3a
LB		 883 /* Our goal is to install a new bind mount into the container,
884 possibly read-only. This is irritatingly complex
885 unfortunately, currently.
886
887 First, we start by creating a private playground in /tmp,
888 that we can mount MS_SLAVE. (Which is necessary, since
889 MS_MOVE cannot be applied to mounts with MS_SHARED parent
890 mounts.) */
891
892 if (!mkdtemp(mount_slave))
893 return log_debug_errno(errno, "Failed to create playground %s: %m", mount_slave);
894
895 mount_slave_created = true;
896
897 r = mount_nofollow_verbose(LOG_DEBUG, mount_slave, mount_slave, NULL, MS_BIND, NULL);
898 if (r < 0)
899 goto finish;
900
901 mount_slave_mounted = true;
902
903 r = mount_nofollow_verbose(LOG_DEBUG, NULL, mount_slave, NULL, MS_SLAVE, NULL);
904 if (r < 0)
905 goto finish;
906
907 /* Second, we mount the source file or directory to a directory inside of our MS_SLAVE playground. */
908 mount_tmp = strjoina(mount_slave, "/mount");
70599967
LB		 909 if (is_image)
910 r = mkdir_p(mount_tmp, 0700);
911 else
a7e0199e 912 r = make_mount_point_inode_from_stat(chased_src_st, mount_tmp, 0700);
6af52c3a
LB		 913 if (r < 0) {
914 log_debug_errno(r, "Failed to create temporary mount point %s: %m", mount_tmp);
915 goto finish;
916 }
917
918 mount_tmp_created = true;
919
70599967 920 if (is_image)
3e107272
LB		 921 r = verity_dissect_and_mount(
922 chased_src_fd,
923 chased_src_path,
924 mount_tmp,
925 options,
926 image_policy,
927 /* required_host_os_release_id= */ NULL,
928 /* required_host_os_release_version_id= */ NULL,
929 /* required_host_os_release_sysext_level= */ NULL,
5897469a 930 /* required_host_os_release_confext_level= */ NULL,
3e107272
LB		 931 /* required_sysext_scope= */ NULL,
932 /* ret_image= */ NULL);
70599967 933 else
ddb6eeaf 934 r = mount_follow_verbose(LOG_DEBUG, FORMAT_PROC_FD_PATH(chased_src_fd), mount_tmp, NULL, MS_BIND, NULL);
6af52c3a
LB		 935 if (r < 0)
936 goto finish;
937
938 mount_tmp_mounted = true;
939
940 /* Third, we remount the new bind mount read-only if requested. */
941 if (read_only) {
942 r = mount_nofollow_verbose(LOG_DEBUG, NULL, mount_tmp, NULL, MS_BIND|MS_REMOUNT|MS_RDONLY, NULL);
943 if (r < 0)
944 goto finish;
945 }
946
947 /* Fourth, we move the new bind mount into the propagation directory. This way it will appear there read-only
948 * right-away. */
949
950 mount_outside = strjoina(propagate_path, "/XXXXXX");
a7e0199e 951 if (is_image || S_ISDIR(chased_src_st->st_mode))
6af52c3a
LB		 952 r = mkdtemp(mount_outside) ? 0 : -errno;
953 else {
954 r = mkostemp_safe(mount_outside);
955 safe_close(r);
956 }
957 if (r < 0) {
958 log_debug_errno(r, "Cannot create propagation file or directory %s: %m", mount_outside);
959 goto finish;
960 }
961
962 mount_outside_created = true;
963
964 r = mount_nofollow_verbose(LOG_DEBUG, mount_tmp, mount_outside, NULL, MS_MOVE, NULL);
965 if (r < 0)
966 goto finish;
967
968 mount_outside_mounted = true;
969 mount_tmp_mounted = false;
970
a7e0199e 971 if (is_image || S_ISDIR(chased_src_st->st_mode))
6af52c3a
LB		 972 (void) rmdir(mount_tmp);
973 else
974 (void) unlink(mount_tmp);
975 mount_tmp_created = false;
976
977 (void) umount_verbose(LOG_DEBUG, mount_slave, UMOUNT_NOFOLLOW);
978 mount_slave_mounted = false;
979
980 (void) rmdir(mount_slave);
981 mount_slave_created = false;
982
983 if (pipe2(errno_pipe_fd, O_CLOEXEC|O_NONBLOCK) < 0) {
984 log_debug_errno(errno, "Failed to create pipe: %m");
985 goto finish;
986 }
987
e9ccae31 988 r = namespace_fork("(sd-bindmnt)", "(sd-bindmnt-inner)", NULL, 0, FORK_RESET_SIGNALS|FORK_DEATHSIG_SIGTERM,
98f654fd 989 pidns_fd, mntns_fd, -1, -1, root_fd, &child);
6af52c3a
LB		 990 if (r < 0)
991 goto finish;
992 if (r == 0) {
03469b77 993 _cleanup_free_ char *mount_outside_fn = NULL, *mount_inside = NULL;
6af52c3a
LB		 994
995 errno_pipe_fd[0] = safe_close(errno_pipe_fd[0]);
996
6af52c3a 997 if (make_file_or_directory) {
70599967
LB		 998 if (!is_image) {
999 (void) mkdir_parents(dest, 0755);
a7e0199e 1000 (void) make_mount_point_inode_from_stat(chased_src_st, dest, 0700);
70599967
LB		 1001 } else
1002 (void) mkdir_p(dest, 0755);
6af52c3a
LB		 1003 }
1004
1005 /* Fifth, move the mount to the right place inside */
03469b77
LP		 1006 r = path_extract_filename(mount_outside, &mount_outside_fn);
1007 if (r < 0) {
1008 log_debug_errno(r, "Failed to extract filename from propagation file or directory '%s': %m", mount_outside);
1009 goto child_fail;
1010 }
1011
1012 mount_inside = path_join(incoming_path, mount_outside_fn);
1013 if (!mount_inside) {
1014 r = log_oom_debug();
1015 goto child_fail;
1016 }
1017
1018 r = mount_nofollow_verbose(LOG_DEBUG, mount_inside, dest, NULL, MS_MOVE, NULL);
6af52c3a
LB		 1019 if (r < 0)
1020 goto child_fail;
1021
1022 _exit(EXIT_SUCCESS);
1023
1024 child_fail:
1025 (void) write(errno_pipe_fd[1], &r, sizeof(r));
1026 errno_pipe_fd[1] = safe_close(errno_pipe_fd[1]);
1027
1028 _exit(EXIT_FAILURE);
1029 }
1030
1031 errno_pipe_fd[1] = safe_close(errno_pipe_fd[1]);
1032
1033 r = wait_for_terminate_and_check("(sd-bindmnt)", child, 0);
1034 if (r < 0) {
1035 log_debug_errno(r, "Failed to wait for child: %m");
1036 goto finish;
1037 }
1038 if (r != EXIT_SUCCESS) {
1039 if (read(errno_pipe_fd[0], &r, sizeof(r)) == sizeof(r))
1040 log_debug_errno(r, "Failed to mount: %m");
1041 else
1042 log_debug("Child failed.");
1043 goto finish;
1044 }
1045
1046finish:
1047 if (mount_outside_mounted)
1048 (void) umount_verbose(LOG_DEBUG, mount_outside, UMOUNT_NOFOLLOW);
1049 if (mount_outside_created) {
a7e0199e 1050 if (is_image || S_ISDIR(chased_src_st->st_mode))
6af52c3a
LB		 1051 (void) rmdir(mount_outside);
1052 else
1053 (void) unlink(mount_outside);
1054 }
1055
1056 if (mount_tmp_mounted)
1057 (void) umount_verbose(LOG_DEBUG, mount_tmp, UMOUNT_NOFOLLOW);
1058 if (mount_tmp_created) {
a7e0199e 1059 if (is_image || S_ISDIR(chased_src_st->st_mode))
6af52c3a
LB		 1060 (void) rmdir(mount_tmp);
1061 else
1062 (void) unlink(mount_tmp);
1063 }
1064
1065 if (mount_slave_mounted)
1066 (void) umount_verbose(LOG_DEBUG, mount_slave, UMOUNT_NOFOLLOW);
1067 if (mount_slave_created)
1068 (void) rmdir(mount_slave);
1069
1070 return r;
1071}
70599967 1072
a7e0199e 1073static int mount_in_namespace(
83d5bbaf 1074 const PidRef *target,
a7e0199e
LB		 1075 const char *propagate_path,
1076 const char *incoming_path,
1077 const char *src,
1078 const char *dest,
1079 bool read_only,
1080 bool make_file_or_directory,
1081 const MountOptions *options,
1082 const ImagePolicy *image_policy,
1083 bool is_image) {
1084
ddba5a1a 1085 _cleanup_(dissected_image_unrefp) DissectedImage *img = NULL;
71136404 1086 _cleanup_close_pair_ int errno_pipe_fd[2] = EBADF_PAIR;
a7e0199e
LB		 1087 _cleanup_close_ int mntns_fd = -EBADF, root_fd = -EBADF, pidns_fd = -EBADF, chased_src_fd = -EBADF,
1088 new_mount_fd = -EBADF;
1089 _cleanup_free_ char *chased_src_path = NULL;
1090 struct stat st;
1091 pid_t child;
1092 int r;
1093
a7e0199e
LB		 1094 assert(propagate_path);
1095 assert(incoming_path);
1096 assert(src);
1097 assert(dest);
1098 assert(!options || is_image);
1099
5f48198a
LB		 1100 if (!pidref_is_set(target))
1101 return -ESRCH;
1102
1103 r = namespace_open(target->pid, &pidns_fd, &mntns_fd, NULL, NULL, &root_fd);
a7e0199e
LB		 1104 if (r < 0)
1105 return log_debug_errno(r, "Failed to retrieve FDs of the target process' namespace: %m");
1106
5f48198a 1107 r = in_same_namespace(target->pid, 0, NAMESPACE_MOUNT);
a7e0199e
LB		 1108 if (r < 0)
1109 return log_debug_errno(r, "Failed to determine if mount namespaces are equal: %m");
1110 /* We can't add new mounts at runtime if the process wasn't started in a namespace */
1111 if (r > 0)
1112 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to activate bind mount in target, not running in a mount namespace");
1113
0f095d0b
MY		 1114 r = pidref_verify(target);
1115 if (r < 0)
1116 return log_debug_errno(r, "Failed to verify target process '" PID_FMT "': %m", target->pid);
5f48198a 1117
a7e0199e
LB		 1118 r = chase(src, NULL, 0, &chased_src_path, &chased_src_fd);
1119 if (r < 0)
1120 return log_debug_errno(r, "Failed to resolve source path of %s: %m", src);
1121 log_debug("Chased source path of %s to %s", src, chased_src_path);
1122
1123 if (fstat(chased_src_fd, &st) < 0)
1124 return log_debug_errno(errno, "Failed to stat() resolved source path %s: %m", src);
1125 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… */
1126 return log_debug_errno(SYNTHETIC_ERRNO(EOPNOTSUPP), "Source directory %s can't be a symbolic link", src);
1127
ddba5a1a 1128 if (!mount_new_api_supported()) /* Fallback if we can't use the new mount API */
a7e0199e
LB		 1129 return mount_in_namespace_legacy(
1130 chased_src_path,
1131 chased_src_fd,
1132 &st,
1133 propagate_path,
1134 incoming_path,
1135 dest,
1136 pidns_fd,
1137 mntns_fd,
1138 root_fd,
1139 read_only,
1140 make_file_or_directory,
1141 options,
1142 image_policy,
1143 is_image);
1144
ddba5a1a
LB		 1145 if (is_image) {
1146 r = verity_dissect_and_mount(
1147 chased_src_fd,
1148 chased_src_path,
1149 /* dest= */ NULL,
1150 options,
1151 image_policy,
1152 /* required_host_os_release_id= */ NULL,
1153 /* required_host_os_release_version_id= */ NULL,
1154 /* required_host_os_release_sysext_level= */ NULL,
5897469a 1155 /* required_host_os_release_confext_level= */ NULL,
ddba5a1a
LB		 1156 /* required_sysext_scope= */ NULL,
1157 &img);
1158 if (r < 0)
1159 return log_debug_errno(
1160 r,
1161 "Failed to dissect and mount image %s: %m",
1162 chased_src_path);
1163 } else {
1164 new_mount_fd = open_tree(
1165 chased_src_fd,
1166 "",
1167 OPEN_TREE_CLONE|OPEN_TREE_CLOEXEC|AT_SYMLINK_NOFOLLOW|AT_EMPTY_PATH);
1168 if (new_mount_fd < 0)
1169 return log_debug_errno(
1170 errno,
1171 "Failed to open mount point \"%s\": %m",
1172 chased_src_path);
1173
1174 if (read_only && mount_setattr(new_mount_fd, "", AT_EMPTY_PATH,
1175 &(struct mount_attr) {
1176 .attr_set = MOUNT_ATTR_RDONLY,
1177 }, MOUNT_ATTR_SIZE_VER0) < 0)
1178 return log_debug_errno(
1179 errno,
1180 "Failed to set mount flags for \"%s\": %m",
1181 chased_src_path);
1182 }
a7e0199e
LB		 1183
1184 if (pipe2(errno_pipe_fd, O_CLOEXEC|O_NONBLOCK) < 0)
1185 return log_debug_errno(errno, "Failed to create pipe: %m");
1186
1187 r = namespace_fork("(sd-bindmnt)",
1188 "(sd-bindmnt-inner)",
1189 /* except_fds= */ NULL,
1190 /* n_except_fds= */ 0,
e9ccae31 1191 FORK_RESET_SIGNALS|FORK_DEATHSIG_SIGTERM,
a7e0199e
LB		 1192 pidns_fd,
1193 mntns_fd,
1194 /* netns_fd= */ -1,
1195 /* userns_fd= */ -1,
1196 root_fd,
1197 &child);
1198 if (r < 0)
1199 return log_debug_errno(r, "Failed to fork off: %m");
1200 if (r == 0) {
1201 errno_pipe_fd[0] = safe_close(errno_pipe_fd[0]);
1202
ddba5a1a 1203 if (make_file_or_directory)
a7e0199e 1204 (void) mkdir_parents(dest, 0755);
a7e0199e 1205
ddba5a1a 1206 if (img) {
7c83d42e 1207 DissectImageFlags f = DISSECT_IMAGE_TRY_ATOMIC_MOUNT_EXCHANGE;
ddba5a1a
LB		 1208
1209 if (make_file_or_directory)
1210 f |= DISSECT_IMAGE_MKDIR;
1211
1212 if (read_only)
1213 f |= DISSECT_IMAGE_READ_ONLY;
1214
1215 r = dissected_image_mount(
1216 img,
1217 dest,
1218 /* uid_shift= */ UID_INVALID,
1219 /* uid_range= */ UID_INVALID,
1220 /* userns_fd= */ -EBADF,
1221 f);
1222 } else {
1223 if (make_file_or_directory)
1224 (void) make_mount_point_inode_from_stat(&st, dest, 0700);
1225
7c83d42e 1226 r = mount_exchange_graceful(new_mount_fd, dest, /* mount_beneath= */ true);
ddba5a1a
LB		 1227 }
1228 if (r < 0) {
1229 (void) write(errno_pipe_fd[1], &r, sizeof(r));
a7e0199e
LB		 1230 errno_pipe_fd[1] = safe_close(errno_pipe_fd[1]);
1231
1232 _exit(EXIT_FAILURE);
1233 }
1234
1235 _exit(EXIT_SUCCESS);
1236 }
1237
1238 errno_pipe_fd[1] = safe_close(errno_pipe_fd[1]);
1239
1240 r = wait_for_terminate_and_check("(sd-bindmnt)", child, 0);
1241 if (r < 0)
1242 return log_debug_errno(r, "Failed to wait for child: %m");
1243 if (r != EXIT_SUCCESS) {
1244 if (read(errno_pipe_fd[0], &r, sizeof(r)) == sizeof(r))
1245 return log_debug_errno(r, "Failed to mount: %m");
1246
1247 return log_debug_errno(SYNTHETIC_ERRNO(EPROTO), "Child failed.");
1248 }
1249
1250 return 0;
1251}
1252
70599967 1253int bind_mount_in_namespace(
5f48198a 1254 PidRef * target,
70599967
LB		 1255 const char *propagate_path,
1256 const char *incoming_path,
1257 const char *src,
1258 const char *dest,
1259 bool read_only,
1260 bool make_file_or_directory) {
1261
84be0c71 1262 return mount_in_namespace(target, propagate_path, incoming_path, src, dest, read_only, make_file_or_directory, /* options= */ NULL, /* image_policy= */ NULL, /* is_image= */ false);
70599967
LB		 1263}
1264
1265int mount_image_in_namespace(
5f48198a 1266 PidRef * target,
70599967
LB		 1267 const char *propagate_path,
1268 const char *incoming_path,
1269 const char *src,
1270 const char *dest,
1271 bool read_only,
1272 bool make_file_or_directory,
84be0c71
LP		 1273 const MountOptions *options,
1274 const ImagePolicy *image_policy) {
70599967 1275
84be0c71 1276 return mount_in_namespace(target, propagate_path, incoming_path, src, dest, read_only, make_file_or_directory, options, image_policy, /* is_image=*/ true);
70599967 1277}
14a25e1f
LP		 1278
1279int make_mount_point(const char *path) {
1280 int r;
1281
1282 assert(path);
1283
1284 /* If 'path' is already a mount point, does nothing and returns 0. If it is not it makes it one, and returns 1. */
1285
1286 r = path_is_mount_point(path, NULL, 0);
1287 if (r < 0)
1288 return log_debug_errno(r, "Failed to determine whether '%s' is a mount point: %m", path);
1289 if (r > 0)
1290 return 0;
1291
1292 r = mount_nofollow_verbose(LOG_DEBUG, path, path, NULL, MS_BIND|MS_REC, NULL);
1293 if (r < 0)
1294 return r;
1295
1296 return 1;
1297}
35fd3558 1298
f9ad896e
LP		 1299int fd_make_mount_point(int fd) {
1300 int r;
1301
1302 assert(fd >= 0);
1303
1304 r = fd_is_mount_point(fd, NULL, 0);
1305 if (r < 0)
1306 return log_debug_errno(r, "Failed to determine whether file descriptor is a mount point: %m");
1307 if (r > 0)
1308 return 0;
1309
1310 r = mount_follow_verbose(LOG_DEBUG, FORMAT_PROC_FD_PATH(fd), FORMAT_PROC_FD_PATH(fd), NULL, MS_BIND|MS_REC, NULL);
1311 if (r < 0)
1312 return r;
1313
1314 return 1;
1315}
1316
17b798d9 1317int make_userns(uid_t uid_shift, uid_t uid_range, uid_t owner, RemountIdmapping idmapping) {
254d1313 1318 _cleanup_close_ int userns_fd = -EBADF;
50ae2966 1319 _cleanup_free_ char *line = NULL;
35fd3558
LP		 1320
1321 /* Allocates a userns file descriptor with the mapping we need. For this we'll fork off a child
1322 * process whose only purpose is to give us a new user namespace. It's killed when we got it. */
1323
17b798d9
LP		 1324 if (!userns_shift_range_valid(uid_shift, uid_range))
1325 return -EINVAL;
1326
2b2777ed
QD		 1327 if (IN_SET(idmapping, REMOUNT_IDMAPPING_NONE, REMOUNT_IDMAPPING_HOST_ROOT)) {
1328 if (asprintf(&line, UID_FMT " " UID_FMT " " UID_FMT "\n", 0u, uid_shift, uid_range) < 0)
50ae2966 1329 return log_oom_debug();
35fd3558 1330
2b2777ed
QD		 1331 /* If requested we'll include an entry in the mapping so that the host root user can make
1332 * changes to the uidmapped mount like it normally would. Specifically, we'll map the user
1333 * with UID_MAPPED_ROOT on the backing fs to UID 0. This is useful, since nspawn code wants
1334 * to create various missing inodes in the OS tree before booting into it, and this becomes
1335 * very easy and straightforward to do if it can just do it under its own regular UID. Note
1336 * that in that case the container's runtime uidmap (i.e. the one the container payload
1337 * processes run in) will leave this UID unmapped, i.e. if we accidentally leave files owned
1338 * by host root in the already uidmapped tree around they'll show up as owned by 'nobody',
1339 * which is safe. (Of course, we shouldn't leave such inodes around, but always chown() them
1340 * to the container's own UID range, but it's good to have a safety net, in case we
1341 * forget it.) */
1342 if (idmapping == REMOUNT_IDMAPPING_HOST_ROOT)
1343 if (strextendf(&line,
1344 UID_FMT " " UID_FMT " " UID_FMT "\n",
1345 UID_MAPPED_ROOT, 0u, 1u) < 0)
1346 return log_oom_debug();
1347 }
1348
1349 if (idmapping == REMOUNT_IDMAPPING_HOST_OWNER) {
1350 /* Remap the owner of the bind mounted directory to the root user within the container. This
1351 * way every file written by root within the container to the bind-mounted directory will
1352 * be owned by the original user. All other user will remain unmapped. */
1353 if (asprintf(&line, UID_FMT " " UID_FMT " " UID_FMT "\n", owner, uid_shift, 1u) < 0)
1354 return log_oom_debug();
1355 }
1356
35fd3558 1357 /* We always assign the same UID and GID ranges */
979b0ff2
LP		 1358 userns_fd = userns_acquire(line, line);
1359 if (userns_fd < 0)
1360 return log_debug_errno(userns_fd, "Failed to acquire new userns: %m");
35fd3558
LP		 1361
1362 return TAKE_FD(userns_fd);
1363}
1364
17b798d9 1365int remount_idmap_fd(
dba4fa89 1366 char **paths,
17b798d9 1367 int userns_fd) {
35fd3558 1368
35fd3558
LP		 1369 int r;
1370
17b798d9 1371 assert(userns_fd >= 0);
35fd3558 1372
dba4fa89
LP		 1373 /* This remounts all specified paths with the specified userns as idmap. It will do so in in the
1374 * order specified in the strv: the expectation is that the top-level directories are at the
1375 * beginning, and nested directories in the right, so that the tree can be built correctly from left
1376 * to right. */
1377
1378 size_t n = strv_length(paths);
1379 if (n == 0) /* Nothing to do? */
1380 return 0;
1381
1382 int *mount_fds = NULL;
1383 size_t n_mounts_fds = 0;
1384
dba4fa89
LP		 1385 mount_fds = new(int, n);
1386 if (!mount_fds)
1387 return log_oom_debug();
1388
6d5202cc
LB		 1389 CLEANUP_ARRAY(mount_fds, n_mounts_fds, close_many_and_free);
1390
dba4fa89
LP		 1391 for (size_t i = 0; i < n; i++) {
1392 int mntfd;
1393
1394 /* Clone the mount point */
1395 mntfd = mount_fds[n_mounts_fds] = open_tree(-EBADF, paths[i], OPEN_TREE_CLONE | OPEN_TREE_CLOEXEC);
1396 if (mount_fds[n_mounts_fds] < 0)
1397 return log_debug_errno(errno, "Failed to open tree of mounted filesystem '%s': %m", paths[i]);
35fd3558 1398
dba4fa89
LP		 1399 n_mounts_fds++;
1400
1401 /* Set the user namespace mapping attribute on the cloned mount point */
1402 if (mount_setattr(mntfd, "", AT_EMPTY_PATH,
1403 &(struct mount_attr) {
1404 .attr_set = MOUNT_ATTR_IDMAP,
1405 .userns_fd = userns_fd,
1406 }, sizeof(struct mount_attr)) < 0)
1407 return log_debug_errno(errno, "Failed to change bind mount attributes for clone of '%s': %m", paths[i]);
1408 }
1409
1410 for (size_t i = n; i > 0; i--) { /* Unmount the paths right-to-left */
1411 /* Remove the old mount points now that we have a idmapped mounts as replacement for all of them */
1412 r = umount_verbose(LOG_DEBUG, paths[i-1], UMOUNT_NOFOLLOW);
1413 if (r < 0)
1414 return r;
1415 }
1416
1417 for (size_t i = 0; i < n; i++) { /* Mount the replacement mounts left-to-right */
1418 /* And place the cloned version in its place */
1419 log_debug("Mounting idmapped fs to '%s'", paths[i]);
1420 if (move_mount(mount_fds[i], "", -EBADF, paths[i], MOVE_MOUNT_F_EMPTY_PATH) < 0)
1421 return log_debug_errno(errno, "Failed to attach UID mapped mount to '%s': %m", paths[i]);
1422 }
35fd3558
LP		 1423
1424 return 0;
1425}
9c653536 1426
dba4fa89 1427int remount_idmap(char **p, uid_t uid_shift, uid_t uid_range, uid_t owner, RemountIdmapping idmapping) {
17b798d9
LP		 1428 _cleanup_close_ int userns_fd = -EBADF;
1429
1430 userns_fd = make_userns(uid_shift, uid_range, owner, idmapping);
1431 if (userns_fd < 0)
1432 return userns_fd;
1433
1434 return remount_idmap_fd(p, userns_fd);
1435}
1436
f63a2c48
YW		 1437typedef struct SubMount {
1438 char *path;
1439 int mount_fd;
1440} SubMount;
1441
1442static void sub_mount_clear(SubMount *s) {
1443 assert(s);
1444
1445 s->path = mfree(s->path);
1446 s->mount_fd = safe_close(s->mount_fd);
1447}
1448
1449static void sub_mount_array_free(SubMount *s, size_t n) {
1450 assert(s || n == 0);
1451
1452 for (size_t i = 0; i < n; i++)
1453 sub_mount_clear(s + i);
1454
1455 free(s);
1456}
1457
1458static int sub_mount_compare(const SubMount *a, const SubMount *b) {
1459 assert(a);
1460 assert(b);
1461 assert(a->path);
1462 assert(b->path);
1463
1464 return path_compare(a->path, b->path);
1465}
1466
1467static void sub_mount_drop(SubMount *s, size_t n) {
1468 assert(s || n == 0);
1469
1470 for (size_t m = 0, i = 1; i < n; i++) {
1471 if (path_startswith(s[i].path, s[m].path))
1472 sub_mount_clear(s + i);
1473 else
1474 m = i;
1475 }
1476}
1477
1b618bf1
LP		 1478static int get_sub_mounts(
1479 const char *prefix,
1b618bf1
LP		 1480 SubMount **ret_mounts,
1481 size_t *ret_n_mounts) {
f63a2c48
YW		 1482 _cleanup_(mnt_free_tablep) struct libmnt_table *table = NULL;
1483 _cleanup_(mnt_free_iterp) struct libmnt_iter *iter = NULL;
1484 SubMount *mounts = NULL;
1485 size_t n = 0;
1486 int r;
1487
1488 CLEANUP_ARRAY(mounts, n, sub_mount_array_free);
1489
1490 assert(prefix);
1491 assert(ret_mounts);
1492 assert(ret_n_mounts);
1493
1494 r = libmount_parse("/proc/self/mountinfo", NULL, &table, &iter);
1495 if (r < 0)
1496 return log_debug_errno(r, "Failed to parse /proc/self/mountinfo: %m");
1497
1498 for (;;) {
1499 _cleanup_close_ int mount_fd = -EBADF;
1500 _cleanup_free_ char *p = NULL;
1501 struct libmnt_fs *fs;
1502 const char *path;
1503 int id1, id2;
1504
1505 r = mnt_table_next_fs(table, iter, &fs);
1506 if (r == 1)
1507 break; /* EOF */
1508 if (r < 0)
1509 return log_debug_errno(r, "Failed to get next entry from /proc/self/mountinfo: %m");
1510
1511 path = mnt_fs_get_target(fs);
1512 if (!path)
1513 continue;
1514
1515 if (isempty(path_startswith(path, prefix)))
1516 continue;
1517
1518 id1 = mnt_fs_get_id(fs);
1519 r = path_get_mnt_id(path, &id2);
1520 if (r < 0) {
1521 log_debug_errno(r, "Failed to get mount ID of '%s', ignoring: %m", path);
1522 continue;
1523 }
1524 if (id1 != id2) {
1525 /* The path may be hidden by another over-mount or already remounted. */
1526 log_debug("The mount IDs of '%s' obtained by libmount and path_get_mnt_id() are different (%i vs %i), ignoring.",
1527 path, id1, id2);
1528 continue;
1529 }
1530
2b60ce54 1531 mount_fd = open(path, O_CLOEXEC|O_PATH);
f63a2c48
YW		 1532 if (mount_fd < 0) {
1533 if (errno == ENOENT) /* The path may be hidden by another over-mount or already unmounted. */
1534 continue;
1535
1b618bf1 1536 return log_debug_errno(errno, "Failed to open subtree of mounted filesystem '%s': %m", path);
f63a2c48
YW		 1537 }
1538
1539 p = strdup(path);
1540 if (!p)
1541 return log_oom_debug();
1542
1543 if (!GREEDY_REALLOC(mounts, n + 1))
1544 return log_oom_debug();
1545
1546 mounts[n++] = (SubMount) {
1547 .path = TAKE_PTR(p),
1548 .mount_fd = TAKE_FD(mount_fd),
1549 };
1550 }
1551
1552 typesafe_qsort(mounts, n, sub_mount_compare);
1553 sub_mount_drop(mounts, n);
1554
1555 *ret_mounts = TAKE_PTR(mounts);
1556 *ret_n_mounts = n;
1557 return 0;
1558}
1559
1b618bf1
LP		 1560int bind_mount_submounts(
1561 const char *source,
1562 const char *target) {
1563
1564 SubMount *mounts = NULL;
1565 size_t n = 0;
1566 int ret = 0, r;
1567
1568 /* Bind mounts all child mounts of 'source' to 'target'. Useful when setting up a new procfs instance
1569 * with new mount options to copy the original submounts over. */
1570
1571 assert(source);
1572 assert(target);
1573
1574 CLEANUP_ARRAY(mounts, n, sub_mount_array_free);
1575
2b60ce54 1576 r = get_sub_mounts(source, &mounts, &n);
1b618bf1
LP		 1577 if (r < 0)
1578 return r;
1579
1580 FOREACH_ARRAY(m, mounts, n) {
1581 _cleanup_free_ char *t = NULL;
1582 const char *suffix;
1583
1584 if (isempty(m->path))
1585 continue;
1586
1587 assert_se(suffix = path_startswith(m->path, source));
1588
1589 t = path_join(target, suffix);
1590 if (!t)
1591 return -ENOMEM;
1592
1593 r = path_is_mount_point(t, NULL, 0);
1594 if (r < 0) {
1595 log_debug_errno(r, "Failed to detect if '%s' already is a mount point, ignoring: %m", t);
1596 continue;
1597 }
1598 if (r > 0) {
1599 log_debug("Not bind mounting '%s' from '%s' to '%s', since there's already a mountpoint.", suffix, source, target);
1600 continue;
1601 }
1602
1603 r = mount_follow_verbose(LOG_DEBUG, FORMAT_PROC_FD_PATH(m->mount_fd), t, NULL, MS_BIND|MS_REC, NULL);
1604 if (r < 0 && ret == 0)
1605 ret = r;
1606 }
1607
1608 return ret;
1609}
1610
9c653536
ZJS		 1611int make_mount_point_inode_from_stat(const struct stat *st, const char *dest, mode_t mode) {
1612 assert(st);
1613 assert(dest);
1614
1615 if (S_ISDIR(st->st_mode))
1616 return mkdir_label(dest, mode);
1617 else
b6ca2b28 1618 return RET_NERRNO(mknod(dest, S_IFREG|(mode & ~0111), 0));
9c653536
ZJS		 1619}
1620
1621int make_mount_point_inode_from_path(const char *source, const char *dest, mode_t mode) {
1622 struct stat st;
1623
1624 assert(source);
1625 assert(dest);
1626
1627 if (stat(source, &st) < 0)
1628 return -errno;
1629
1630 return make_mount_point_inode_from_stat(&st, dest, mode);
1631}
506c1bb5
DDM		 1632
1633int trigger_automount_at(int dir_fd, const char *path) {
1634 _cleanup_free_ char *nested = NULL;
1635
1636 assert(dir_fd >= 0 || dir_fd == AT_FDCWD);
1637
1638 nested = path_join(path, "a");
1639 if (!nested)
1640 return -ENOMEM;
1641
1642 (void) faccessat(dir_fd, nested, F_OK, 0);
1643
1644 return 0;
1645}
1155f44f
LP		 1646
1647unsigned long credentials_fs_mount_flags(bool ro) {
1648 /* A tight set of mount flags for credentials mounts */
1649 return MS_NODEV|MS_NOEXEC|MS_NOSUID|ms_nosymfollow_supported()|(ro ? MS_RDONLY : 0);
1650}
1651
1652int mount_credentials_fs(const char *path, size_t size, bool ro) {
1653 _cleanup_free_ char *opts = NULL;
1654 int r, noswap_supported;
1655
1656 /* Mounts a file system we can place credentials in, i.e. with tight access modes right from the
1657 * beginning, and ideally swapping turned off. In order of preference:
1658 *
1659 * 1. tmpfs if it supports "noswap"
1660 * 2. ramfs
1661 * 3. tmpfs if it doesn't support "noswap"
1662 */
1663
1664 noswap_supported = mount_option_supported("tmpfs", "noswap", NULL); /* Check explicitly to avoid kmsg noise */
1665 if (noswap_supported > 0) {
1666 _cleanup_free_ char *noswap_opts = NULL;
1667
1668 if (asprintf(&noswap_opts, "mode=0700,nr_inodes=1024,size=%zu,noswap", size) < 0)
1669 return -ENOMEM;
1670
1671 /* Best case: tmpfs with noswap (needs kernel >= 6.3) */
1672
1673 r = mount_nofollow_verbose(
1674 LOG_DEBUG,
1675 "tmpfs",
1676 path,
1677 "tmpfs",
1678 credentials_fs_mount_flags(ro),
1679 noswap_opts);
1680 if (r >= 0)
1681 return r;
1682 }
1683
1684 r = mount_nofollow_verbose(
1685 LOG_DEBUG,
1686 "ramfs",
1687 path,
1688 "ramfs",
1689 credentials_fs_mount_flags(ro),
1690 "mode=0700");
1691 if (r >= 0)
1692 return r;
1693
1694 if (asprintf(&opts, "mode=0700,nr_inodes=1024,size=%zu", size) < 0)
1695 return -ENOMEM;
1696
1697 return mount_nofollow_verbose(
1698 LOG_DEBUG,
1699 "tmpfs",
1700 path,
1701 "tmpfs",
1702 credentials_fs_mount_flags(ro),
1703 opts);
1704}
44cc82bf
LP		 1705
1706int make_fsmount(
1707 int error_log_level,
1708 const char *what,
1709 const char *type,
1710 unsigned long flags,
1711 const char *options,
1712 int userns_fd) {
1713
1714 _cleanup_close_ int fs_fd = -EBADF, mnt_fd = -EBADF;
1715 _cleanup_free_ char *o = NULL;
1716 unsigned long f;
1717 int r;
1718
1719 assert(type);
1720 assert(what);
1721
1722 r = mount_option_mangle(options, flags, &f, &o);
1723 if (r < 0)
1724 return log_full_errno(
1725 error_log_level, r, "Failed to mangle mount options %s: %m",
1726 strempty(options));
1727
1728 if (DEBUG_LOGGING) {
1729 _cleanup_free_ char *fl = NULL;
1730 (void) mount_flags_to_string(f, &fl);
1731
1732 log_debug("Creating mount fd for %s (%s) (%s \"%s\")...",
1733 strna(what), strna(type), strnull(fl), strempty(o));
1734 }
1735
1736 fs_fd = fsopen(type, FSOPEN_CLOEXEC);
1737 if (fs_fd < 0)
1738 return log_full_errno(error_log_level, errno, "Failed to open superblock for \"%s\": %m", type);
1739
1740 if (fsconfig(fs_fd, FSCONFIG_SET_STRING, "source", what, 0) < 0)
1741 return log_full_errno(error_log_level, errno, "Failed to set mount source for \"%s\" to \"%s\": %m", type, what);
1742
1743 if (FLAGS_SET(f, MS_RDONLY))
1744 if (fsconfig(fs_fd, FSCONFIG_SET_FLAG, "ro", NULL, 0) < 0)
1745 return log_full_errno(error_log_level, errno, "Failed to set read only mount flag for \"%s\": %m", type);
1746
1747 for (const char *p = o;;) {
1748 _cleanup_free_ char *word = NULL;
1749 char *eq;
1750
1751 r = extract_first_word(&p, &word, ",", EXTRACT_KEEP_QUOTE);
1752 if (r < 0)
1753 return log_full_errno(error_log_level, r, "Failed to parse mount option string \"%s\": %m", o);
1754 if (r == 0)
1755 break;
1756
1757 eq = strchr(word, '=');
1758 if (eq) {
1759 *eq = 0;
1760 eq++;
1761
1762 if (fsconfig(fs_fd, FSCONFIG_SET_STRING, word, eq, 0) < 0)
1763 return log_full_errno(error_log_level, errno, "Failed to set mount option \"%s=%s\" for \"%s\": %m", word, eq, type);
1764 } else {
1765 if (fsconfig(fs_fd, FSCONFIG_SET_FLAG, word, NULL, 0) < 0)
1766 return log_full_errno(error_log_level, errno, "Failed to set mount flag \"%s\" for \"%s\": %m", word, type);
1767 }
1768 }
1769
1770 if (fsconfig(fs_fd, FSCONFIG_CMD_CREATE, NULL, NULL, 0) < 0)
1771 return log_full_errno(error_log_level, errno, "Failed to realize fs fd for \"%s\" (\"%s\"): %m", what, type);
1772
1773 mnt_fd = fsmount(fs_fd, FSMOUNT_CLOEXEC, 0);
1774 if (mnt_fd < 0)
1775 return log_full_errno(error_log_level, errno, "Failed to create mount fd for \"%s\" (\"%s\"): %m", what, type);
1776
1777 if (mount_setattr(mnt_fd, "", AT_EMPTY_PATH|AT_RECURSIVE,
1778 &(struct mount_attr) {
1779 .attr_set = ms_flags_to_mount_attr(f) | (userns_fd >= 0 ? MOUNT_ATTR_IDMAP : 0),
1780 .userns_fd = userns_fd,
1781 }, MOUNT_ATTR_SIZE_VER0) < 0)
1782 return log_full_errno(error_log_level,
1783 errno,
1784 "Failed to set mount flags for \"%s\" (\"%s\"): %m",
1785 what,
1786 type);
1787
1788 return TAKE_FD(mnt_fd);
1789}
Unnamed repository; edit this file 'description' to name the repository.