1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/fs/super.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * super.c contains code to handle: - mount structures
8 * - super-block tables
9 * - filesystem drivers list
10 * - mount system call
11 * - umount system call
12 * - ustat system call
13 *
14 * GK 2/5/95 - Changed to support mounting the root fs via NFS
15 *
16 * Added kerneld support: Jacques Gelinas and Bjorn Ekwall
17 * Added change_root: Werner Almesberger & Hans Lermen, Feb '96
18 * Added options to /proc/mounts:
19 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
20 * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
21 * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
22 */
23
24 #include <linux/export.h>
25 #include <linux/slab.h>
26 #include <linux/blkdev.h>
27 #include <linux/mount.h>
28 #include <linux/security.h>
29 #include <linux/writeback.h> /* for the emergency remount stuff */
30 #include <linux/idr.h>
31 #include <linux/mutex.h>
32 #include <linux/backing-dev.h>
33 #include <linux/rculist_bl.h>
34 #include <linux/fscrypt.h>
35 #include <linux/fsnotify.h>
36 #include <linux/lockdep.h>
37 #include <linux/user_namespace.h>
38 #include <linux/fs_context.h>
39 #include <uapi/linux/mount.h>
40 #include "internal.h"
41
42 static int thaw_super_locked(struct super_block *sb, enum freeze_holder who);
43
44 static LIST_HEAD(super_blocks);
45 static DEFINE_SPINLOCK(sb_lock);
46
47 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
48 "sb_writers",
49 "sb_pagefaults",
50 "sb_internal",
51 };
52
__super_lock(struct super_block * sb,bool excl)53 static inline void __super_lock(struct super_block *sb, bool excl)
54 {
55 if (excl)
56 down_write(&sb->s_umount);
57 else
58 down_read(&sb->s_umount);
59 }
60
super_unlock(struct super_block * sb,bool excl)61 static inline void super_unlock(struct super_block *sb, bool excl)
62 {
63 if (excl)
64 up_write(&sb->s_umount);
65 else
66 up_read(&sb->s_umount);
67 }
68
__super_lock_excl(struct super_block * sb)69 static inline void __super_lock_excl(struct super_block *sb)
70 {
71 __super_lock(sb, true);
72 }
73
super_unlock_excl(struct super_block * sb)74 static inline void super_unlock_excl(struct super_block *sb)
75 {
76 super_unlock(sb, true);
77 }
78
super_unlock_shared(struct super_block * sb)79 static inline void super_unlock_shared(struct super_block *sb)
80 {
81 super_unlock(sb, false);
82 }
83
super_flags(const struct super_block * sb,unsigned int flags)84 static bool super_flags(const struct super_block *sb, unsigned int flags)
85 {
86 /*
87 * Pairs with smp_store_release() in super_wake() and ensures
88 * that we see @flags after we're woken.
89 */
90 return smp_load_acquire(&sb->s_flags) & flags;
91 }
92
93 /**
94 * super_lock - wait for superblock to become ready and lock it
95 * @sb: superblock to wait for
96 * @excl: whether exclusive access is required
97 *
98 * If the superblock has neither passed through vfs_get_tree() or
99 * generic_shutdown_super() yet wait for it to happen. Either superblock
100 * creation will succeed and SB_BORN is set by vfs_get_tree() or we're
101 * woken and we'll see SB_DYING.
102 *
103 * The caller must have acquired a temporary reference on @sb->s_count.
104 *
105 * Return: The function returns true if SB_BORN was set and with
106 * s_umount held. The function returns false if SB_DYING was
107 * set and without s_umount held.
108 */
super_lock(struct super_block * sb,bool excl)109 static __must_check bool super_lock(struct super_block *sb, bool excl)
110 {
111 lockdep_assert_not_held(&sb->s_umount);
112
113 /* wait until the superblock is ready or dying */
114 wait_var_event(&sb->s_flags, super_flags(sb, SB_BORN | SB_DYING));
115
116 /* Don't pointlessly acquire s_umount. */
117 if (super_flags(sb, SB_DYING))
118 return false;
119
120 __super_lock(sb, excl);
121
122 /*
123 * Has gone through generic_shutdown_super() in the meantime.
124 * @sb->s_root is NULL and @sb->s_active is 0. No one needs to
125 * grab a reference to this. Tell them so.
126 */
127 if (sb->s_flags & SB_DYING) {
128 super_unlock(sb, excl);
129 return false;
130 }
131
132 WARN_ON_ONCE(!(sb->s_flags & SB_BORN));
133 return true;
134 }
135
136 /* wait and try to acquire read-side of @sb->s_umount */
super_lock_shared(struct super_block * sb)137 static inline bool super_lock_shared(struct super_block *sb)
138 {
139 return super_lock(sb, false);
140 }
141
142 /* wait and try to acquire write-side of @sb->s_umount */
super_lock_excl(struct super_block * sb)143 static inline bool super_lock_excl(struct super_block *sb)
144 {
145 return super_lock(sb, true);
146 }
147
148 /* wake waiters */
149 #define SUPER_WAKE_FLAGS (SB_BORN | SB_DYING | SB_DEAD)
super_wake(struct super_block * sb,unsigned int flag)150 static void super_wake(struct super_block *sb, unsigned int flag)
151 {
152 WARN_ON_ONCE((flag & ~SUPER_WAKE_FLAGS));
153 WARN_ON_ONCE(hweight32(flag & SUPER_WAKE_FLAGS) > 1);
154
155 /*
156 * Pairs with smp_load_acquire() in super_lock() to make sure
157 * all initializations in the superblock are seen by the user
158 * seeing SB_BORN sent.
159 */
160 smp_store_release(&sb->s_flags, sb->s_flags | flag);
161 /*
162 * Pairs with the barrier in prepare_to_wait_event() to make sure
163 * ___wait_var_event() either sees SB_BORN set or
164 * waitqueue_active() check in wake_up_var() sees the waiter.
165 */
166 smp_mb();
167 wake_up_var(&sb->s_flags);
168 }
169
170 /*
171 * One thing we have to be careful of with a per-sb shrinker is that we don't
172 * drop the last active reference to the superblock from within the shrinker.
173 * If that happens we could trigger unregistering the shrinker from within the
174 * shrinker path and that leads to deadlock on the shrinker_mutex. Hence we
175 * take a passive reference to the superblock to avoid this from occurring.
176 */
super_cache_scan(struct shrinker * shrink,struct shrink_control * sc)177 static unsigned long super_cache_scan(struct shrinker *shrink,
178 struct shrink_control *sc)
179 {
180 struct super_block *sb;
181 long fs_objects = 0;
182 long total_objects;
183 long freed = 0;
184 long dentries;
185 long inodes;
186
187 sb = shrink->private_data;
188
189 /*
190 * Deadlock avoidance. We may hold various FS locks, and we don't want
191 * to recurse into the FS that called us in clear_inode() and friends..
192 */
193 if (!(sc->gfp_mask & __GFP_FS))
194 return SHRINK_STOP;
195
196 if (!super_trylock_shared(sb))
197 return SHRINK_STOP;
198
199 if (sb->s_op->nr_cached_objects)
200 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
201
202 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
203 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
204 total_objects = dentries + inodes + fs_objects + 1;
205 if (!total_objects)
206 total_objects = 1;
207
208 /* proportion the scan between the caches */
209 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
210 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
211 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
212
213 /*
214 * prune the dcache first as the icache is pinned by it, then
215 * prune the icache, followed by the filesystem specific caches
216 *
217 * Ensure that we always scan at least one object - memcg kmem
218 * accounting uses this to fully empty the caches.
219 */
220 sc->nr_to_scan = dentries + 1;
221 freed = prune_dcache_sb(sb, sc);
222 sc->nr_to_scan = inodes + 1;
223 freed += prune_icache_sb(sb, sc);
224
225 if (fs_objects) {
226 sc->nr_to_scan = fs_objects + 1;
227 freed += sb->s_op->free_cached_objects(sb, sc);
228 }
229
230 super_unlock_shared(sb);
231 return freed;
232 }
233
super_cache_count(struct shrinker * shrink,struct shrink_control * sc)234 static unsigned long super_cache_count(struct shrinker *shrink,
235 struct shrink_control *sc)
236 {
237 struct super_block *sb;
238 long total_objects = 0;
239
240 sb = shrink->private_data;
241
242 /*
243 * We don't call super_trylock_shared() here as it is a scalability
244 * bottleneck, so we're exposed to partial setup state. The shrinker
245 * rwsem does not protect filesystem operations backing
246 * list_lru_shrink_count() or s_op->nr_cached_objects(). Counts can
247 * change between super_cache_count and super_cache_scan, so we really
248 * don't need locks here.
249 *
250 * However, if we are currently mounting the superblock, the underlying
251 * filesystem might be in a state of partial construction and hence it
252 * is dangerous to access it. super_trylock_shared() uses a SB_BORN check
253 * to avoid this situation, so do the same here. The memory barrier is
254 * matched with the one in mount_fs() as we don't hold locks here.
255 */
256 if (!(sb->s_flags & SB_BORN))
257 return 0;
258 smp_rmb();
259
260 if (sb->s_op && sb->s_op->nr_cached_objects)
261 total_objects = sb->s_op->nr_cached_objects(sb, sc);
262
263 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
264 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
265
266 if (!total_objects)
267 return SHRINK_EMPTY;
268
269 total_objects = vfs_pressure_ratio(total_objects);
270 return total_objects;
271 }
272
destroy_super_work(struct work_struct * work)273 static void destroy_super_work(struct work_struct *work)
274 {
275 struct super_block *s = container_of(work, struct super_block,
276 destroy_work);
277 fsnotify_sb_free(s);
278 security_sb_free(s);
279 put_user_ns(s->s_user_ns);
280 kfree(s->s_subtype);
281 for (int i = 0; i < SB_FREEZE_LEVELS; i++)
282 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
283 kfree(s);
284 }
285
destroy_super_rcu(struct rcu_head * head)286 static void destroy_super_rcu(struct rcu_head *head)
287 {
288 struct super_block *s = container_of(head, struct super_block, rcu);
289 INIT_WORK(&s->destroy_work, destroy_super_work);
290 schedule_work(&s->destroy_work);
291 }
292
293 /* Free a superblock that has never been seen by anyone */
destroy_unused_super(struct super_block * s)294 static void destroy_unused_super(struct super_block *s)
295 {
296 if (!s)
297 return;
298 super_unlock_excl(s);
299 list_lru_destroy(&s->s_dentry_lru);
300 list_lru_destroy(&s->s_inode_lru);
301 shrinker_free(s->s_shrink);
302 /* no delays needed */
303 destroy_super_work(&s->destroy_work);
304 }
305
306 /**
307 * alloc_super - create new superblock
308 * @type: filesystem type superblock should belong to
309 * @flags: the mount flags
310 * @user_ns: User namespace for the super_block
311 *
312 * Allocates and initializes a new &struct super_block. alloc_super()
313 * returns a pointer new superblock or %NULL if allocation had failed.
314 */
alloc_super(struct file_system_type * type,int flags,struct user_namespace * user_ns)315 static struct super_block *alloc_super(struct file_system_type *type, int flags,
316 struct user_namespace *user_ns)
317 {
318 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_KERNEL);
319 static const struct super_operations default_op;
320 int i;
321
322 if (!s)
323 return NULL;
324
325 INIT_LIST_HEAD(&s->s_mounts);
326 s->s_user_ns = get_user_ns(user_ns);
327 init_rwsem(&s->s_umount);
328 lockdep_set_class(&s->s_umount, &type->s_umount_key);
329 /*
330 * sget() can have s_umount recursion.
331 *
332 * When it cannot find a suitable sb, it allocates a new
333 * one (this one), and tries again to find a suitable old
334 * one.
335 *
336 * In case that succeeds, it will acquire the s_umount
337 * lock of the old one. Since these are clearly distrinct
338 * locks, and this object isn't exposed yet, there's no
339 * risk of deadlocks.
340 *
341 * Annotate this by putting this lock in a different
342 * subclass.
343 */
344 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
345
346 if (security_sb_alloc(s))
347 goto fail;
348
349 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
350 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
351 sb_writers_name[i],
352 &type->s_writers_key[i]))
353 goto fail;
354 }
355 s->s_bdi = &noop_backing_dev_info;
356 s->s_flags = flags;
357 if (s->s_user_ns != &init_user_ns)
358 s->s_iflags |= SB_I_NODEV;
359 INIT_HLIST_NODE(&s->s_instances);
360 INIT_HLIST_BL_HEAD(&s->s_roots);
361 mutex_init(&s->s_sync_lock);
362 INIT_LIST_HEAD(&s->s_inodes);
363 spin_lock_init(&s->s_inode_list_lock);
364 INIT_LIST_HEAD(&s->s_inodes_wb);
365 spin_lock_init(&s->s_inode_wblist_lock);
366
367 s->s_count = 1;
368 atomic_set(&s->s_active, 1);
369 mutex_init(&s->s_vfs_rename_mutex);
370 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
371 init_rwsem(&s->s_dquot.dqio_sem);
372 s->s_maxbytes = MAX_NON_LFS;
373 s->s_op = &default_op;
374 s->s_time_gran = 1000000000;
375 s->s_time_min = TIME64_MIN;
376 s->s_time_max = TIME64_MAX;
377
378 s->s_shrink = shrinker_alloc(SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE,
379 "sb-%s", type->name);
380 if (!s->s_shrink)
381 goto fail;
382
383 s->s_shrink->scan_objects = super_cache_scan;
384 s->s_shrink->count_objects = super_cache_count;
385 s->s_shrink->batch = 1024;
386 s->s_shrink->private_data = s;
387
388 if (list_lru_init_memcg(&s->s_dentry_lru, s->s_shrink))
389 goto fail;
390 if (list_lru_init_memcg(&s->s_inode_lru, s->s_shrink))
391 goto fail;
392 return s;
393
394 fail:
395 destroy_unused_super(s);
396 return NULL;
397 }
398
399 /* Superblock refcounting */
400
401 /*
402 * Drop a superblock's refcount. The caller must hold sb_lock.
403 */
__put_super(struct super_block * s)404 static void __put_super(struct super_block *s)
405 {
406 if (!--s->s_count) {
407 list_del_init(&s->s_list);
408 WARN_ON(s->s_dentry_lru.node);
409 WARN_ON(s->s_inode_lru.node);
410 WARN_ON(!list_empty(&s->s_mounts));
411 call_rcu(&s->rcu, destroy_super_rcu);
412 }
413 }
414
415 /**
416 * put_super - drop a temporary reference to superblock
417 * @sb: superblock in question
418 *
419 * Drops a temporary reference, frees superblock if there's no
420 * references left.
421 */
put_super(struct super_block * sb)422 void put_super(struct super_block *sb)
423 {
424 spin_lock(&sb_lock);
425 __put_super(sb);
426 spin_unlock(&sb_lock);
427 }
428
kill_super_notify(struct super_block * sb)429 static void kill_super_notify(struct super_block *sb)
430 {
431 lockdep_assert_not_held(&sb->s_umount);
432
433 /* already notified earlier */
434 if (sb->s_flags & SB_DEAD)
435 return;
436
437 /*
438 * Remove it from @fs_supers so it isn't found by new
439 * sget{_fc}() walkers anymore. Any concurrent mounter still
440 * managing to grab a temporary reference is guaranteed to
441 * already see SB_DYING and will wait until we notify them about
442 * SB_DEAD.
443 */
444 spin_lock(&sb_lock);
445 hlist_del_init(&sb->s_instances);
446 spin_unlock(&sb_lock);
447
448 /*
449 * Let concurrent mounts know that this thing is really dead.
450 * We don't need @sb->s_umount here as every concurrent caller
451 * will see SB_DYING and either discard the superblock or wait
452 * for SB_DEAD.
453 */
454 super_wake(sb, SB_DEAD);
455 }
456
457 /**
458 * deactivate_locked_super - drop an active reference to superblock
459 * @s: superblock to deactivate
460 *
461 * Drops an active reference to superblock, converting it into a temporary
462 * one if there is no other active references left. In that case we
463 * tell fs driver to shut it down and drop the temporary reference we
464 * had just acquired.
465 *
466 * Caller holds exclusive lock on superblock; that lock is released.
467 */
deactivate_locked_super(struct super_block * s)468 void deactivate_locked_super(struct super_block *s)
469 {
470 struct file_system_type *fs = s->s_type;
471 if (atomic_dec_and_test(&s->s_active)) {
472 shrinker_free(s->s_shrink);
473 fs->kill_sb(s);
474
475 kill_super_notify(s);
476
477 /*
478 * Since list_lru_destroy() may sleep, we cannot call it from
479 * put_super(), where we hold the sb_lock. Therefore we destroy
480 * the lru lists right now.
481 */
482 list_lru_destroy(&s->s_dentry_lru);
483 list_lru_destroy(&s->s_inode_lru);
484
485 put_filesystem(fs);
486 put_super(s);
487 } else {
488 super_unlock_excl(s);
489 }
490 }
491
492 EXPORT_SYMBOL(deactivate_locked_super);
493
494 /**
495 * deactivate_super - drop an active reference to superblock
496 * @s: superblock to deactivate
497 *
498 * Variant of deactivate_locked_super(), except that superblock is *not*
499 * locked by caller. If we are going to drop the final active reference,
500 * lock will be acquired prior to that.
501 */
deactivate_super(struct super_block * s)502 void deactivate_super(struct super_block *s)
503 {
504 if (!atomic_add_unless(&s->s_active, -1, 1)) {
505 __super_lock_excl(s);
506 deactivate_locked_super(s);
507 }
508 }
509
510 EXPORT_SYMBOL(deactivate_super);
511
512 /**
513 * grab_super - acquire an active reference to a superblock
514 * @sb: superblock to acquire
515 *
516 * Acquire a temporary reference on a superblock and try to trade it for
517 * an active reference. This is used in sget{_fc}() to wait for a
518 * superblock to either become SB_BORN or for it to pass through
519 * sb->kill() and be marked as SB_DEAD.
520 *
521 * Return: This returns true if an active reference could be acquired,
522 * false if not.
523 */
grab_super(struct super_block * sb)524 static bool grab_super(struct super_block *sb)
525 {
526 bool locked;
527
528 sb->s_count++;
529 spin_unlock(&sb_lock);
530 locked = super_lock_excl(sb);
531 if (locked) {
532 if (atomic_inc_not_zero(&sb->s_active)) {
533 put_super(sb);
534 return true;
535 }
536 super_unlock_excl(sb);
537 }
538 wait_var_event(&sb->s_flags, super_flags(sb, SB_DEAD));
539 put_super(sb);
540 return false;
541 }
542
543 /*
544 * super_trylock_shared - try to grab ->s_umount shared
545 * @sb: reference we are trying to grab
546 *
547 * Try to prevent fs shutdown. This is used in places where we
548 * cannot take an active reference but we need to ensure that the
549 * filesystem is not shut down while we are working on it. It returns
550 * false if we cannot acquire s_umount or if we lose the race and
551 * filesystem already got into shutdown, and returns true with the s_umount
552 * lock held in read mode in case of success. On successful return,
553 * the caller must drop the s_umount lock when done.
554 *
555 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
556 * The reason why it's safe is that we are OK with doing trylock instead
557 * of down_read(). There's a couple of places that are OK with that, but
558 * it's very much not a general-purpose interface.
559 */
super_trylock_shared(struct super_block * sb)560 bool super_trylock_shared(struct super_block *sb)
561 {
562 if (down_read_trylock(&sb->s_umount)) {
563 if (!(sb->s_flags & SB_DYING) && sb->s_root &&
564 (sb->s_flags & SB_BORN))
565 return true;
566 super_unlock_shared(sb);
567 }
568
569 return false;
570 }
571
572 /**
573 * retire_super - prevents superblock from being reused
574 * @sb: superblock to retire
575 *
576 * The function marks superblock to be ignored in superblock test, which
577 * prevents it from being reused for any new mounts. If the superblock has
578 * a private bdi, it also unregisters it, but doesn't reduce the refcount
579 * of the superblock to prevent potential races. The refcount is reduced
580 * by generic_shutdown_super(). The function can not be called
581 * concurrently with generic_shutdown_super(). It is safe to call the
582 * function multiple times, subsequent calls have no effect.
583 *
584 * The marker will affect the re-use only for block-device-based
585 * superblocks. Other superblocks will still get marked if this function
586 * is used, but that will not affect their reusability.
587 */
retire_super(struct super_block * sb)588 void retire_super(struct super_block *sb)
589 {
590 WARN_ON(!sb->s_bdev);
591 __super_lock_excl(sb);
592 if (sb->s_iflags & SB_I_PERSB_BDI) {
593 bdi_unregister(sb->s_bdi);
594 sb->s_iflags &= ~SB_I_PERSB_BDI;
595 }
596 sb->s_iflags |= SB_I_RETIRED;
597 super_unlock_excl(sb);
598 }
599 EXPORT_SYMBOL(retire_super);
600
601 /**
602 * generic_shutdown_super - common helper for ->kill_sb()
603 * @sb: superblock to kill
604 *
605 * generic_shutdown_super() does all fs-independent work on superblock
606 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
607 * that need destruction out of superblock, call generic_shutdown_super()
608 * and release aforementioned objects. Note: dentries and inodes _are_
609 * taken care of and do not need specific handling.
610 *
611 * Upon calling this function, the filesystem may no longer alter or
612 * rearrange the set of dentries belonging to this super_block, nor may it
613 * change the attachments of dentries to inodes.
614 */
generic_shutdown_super(struct super_block * sb)615 void generic_shutdown_super(struct super_block *sb)
616 {
617 const struct super_operations *sop = sb->s_op;
618
619 if (sb->s_root) {
620 shrink_dcache_for_umount(sb);
621 sync_filesystem(sb);
622 sb->s_flags &= ~SB_ACTIVE;
623
624 cgroup_writeback_umount();
625
626 /* Evict all inodes with zero refcount. */
627 evict_inodes(sb);
628
629 /*
630 * Clean up and evict any inodes that still have references due
631 * to fsnotify or the security policy.
632 */
633 fsnotify_sb_delete(sb);
634 security_sb_delete(sb);
635
636 if (sb->s_dio_done_wq) {
637 destroy_workqueue(sb->s_dio_done_wq);
638 sb->s_dio_done_wq = NULL;
639 }
640
641 if (sop->put_super)
642 sop->put_super(sb);
643
644 /*
645 * Now that all potentially-encrypted inodes have been evicted,
646 * the fscrypt keyring can be destroyed.
647 */
648 fscrypt_destroy_keyring(sb);
649
650 if (CHECK_DATA_CORRUPTION(!list_empty(&sb->s_inodes),
651 "VFS: Busy inodes after unmount of %s (%s)",
652 sb->s_id, sb->s_type->name)) {
653 /*
654 * Adding a proper bailout path here would be hard, but
655 * we can at least make it more likely that a later
656 * iput_final() or such crashes cleanly.
657 */
658 struct inode *inode;
659
660 spin_lock(&sb->s_inode_list_lock);
661 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
662 inode->i_op = VFS_PTR_POISON;
663 inode->i_sb = VFS_PTR_POISON;
664 inode->i_mapping = VFS_PTR_POISON;
665 }
666 spin_unlock(&sb->s_inode_list_lock);
667 }
668 }
669 /*
670 * Broadcast to everyone that grabbed a temporary reference to this
671 * superblock before we removed it from @fs_supers that the superblock
672 * is dying. Every walker of @fs_supers outside of sget{_fc}() will now
673 * discard this superblock and treat it as dead.
674 *
675 * We leave the superblock on @fs_supers so it can be found by
676 * sget{_fc}() until we passed sb->kill_sb().
677 */
678 super_wake(sb, SB_DYING);
679 super_unlock_excl(sb);
680 if (sb->s_bdi != &noop_backing_dev_info) {
681 if (sb->s_iflags & SB_I_PERSB_BDI)
682 bdi_unregister(sb->s_bdi);
683 bdi_put(sb->s_bdi);
684 sb->s_bdi = &noop_backing_dev_info;
685 }
686 }
687
688 EXPORT_SYMBOL(generic_shutdown_super);
689
mount_capable(struct fs_context * fc)690 bool mount_capable(struct fs_context *fc)
691 {
692 if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT))
693 return capable(CAP_SYS_ADMIN);
694 else
695 return ns_capable(fc->user_ns, CAP_SYS_ADMIN);
696 }
697
698 /**
699 * sget_fc - Find or create a superblock
700 * @fc: Filesystem context.
701 * @test: Comparison callback
702 * @set: Setup callback
703 *
704 * Create a new superblock or find an existing one.
705 *
706 * The @test callback is used to find a matching existing superblock.
707 * Whether or not the requested parameters in @fc are taken into account
708 * is specific to the @test callback that is used. They may even be
709 * completely ignored.
710 *
711 * If an extant superblock is matched, it will be returned unless:
712 *
713 * (1) the namespace the filesystem context @fc and the extant
714 * superblock's namespace differ
715 *
716 * (2) the filesystem context @fc has requested that reusing an extant
717 * superblock is not allowed
718 *
719 * In both cases EBUSY will be returned.
720 *
721 * If no match is made, a new superblock will be allocated and basic
722 * initialisation will be performed (s_type, s_fs_info and s_id will be
723 * set and the @set callback will be invoked), the superblock will be
724 * published and it will be returned in a partially constructed state
725 * with SB_BORN and SB_ACTIVE as yet unset.
726 *
727 * Return: On success, an extant or newly created superblock is
728 * returned. On failure an error pointer is returned.
729 */
sget_fc(struct fs_context * fc,int (* test)(struct super_block *,struct fs_context *),int (* set)(struct super_block *,struct fs_context *))730 struct super_block *sget_fc(struct fs_context *fc,
731 int (*test)(struct super_block *, struct fs_context *),
732 int (*set)(struct super_block *, struct fs_context *))
733 {
734 struct super_block *s = NULL;
735 struct super_block *old;
736 struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns;
737 int err;
738
739 retry:
740 spin_lock(&sb_lock);
741 if (test) {
742 hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) {
743 if (test(old, fc))
744 goto share_extant_sb;
745 }
746 }
747 if (!s) {
748 spin_unlock(&sb_lock);
749 s = alloc_super(fc->fs_type, fc->sb_flags, user_ns);
750 if (!s)
751 return ERR_PTR(-ENOMEM);
752 goto retry;
753 }
754
755 s->s_fs_info = fc->s_fs_info;
756 err = set(s, fc);
757 if (err) {
758 s->s_fs_info = NULL;
759 spin_unlock(&sb_lock);
760 destroy_unused_super(s);
761 return ERR_PTR(err);
762 }
763 fc->s_fs_info = NULL;
764 s->s_type = fc->fs_type;
765 s->s_iflags |= fc->s_iflags;
766 strscpy(s->s_id, s->s_type->name, sizeof(s->s_id));
767 /*
768 * Make the superblock visible on @super_blocks and @fs_supers.
769 * It's in a nascent state and users should wait on SB_BORN or
770 * SB_DYING to be set.
771 */
772 list_add_tail(&s->s_list, &super_blocks);
773 hlist_add_head(&s->s_instances, &s->s_type->fs_supers);
774 spin_unlock(&sb_lock);
775 get_filesystem(s->s_type);
776 shrinker_register(s->s_shrink);
777 return s;
778
779 share_extant_sb:
780 if (user_ns != old->s_user_ns || fc->exclusive) {
781 spin_unlock(&sb_lock);
782 destroy_unused_super(s);
783 if (fc->exclusive)
784 warnfc(fc, "reusing existing filesystem not allowed");
785 else
786 warnfc(fc, "reusing existing filesystem in another namespace not allowed");
787 return ERR_PTR(-EBUSY);
788 }
789 if (!grab_super(old))
790 goto retry;
791 destroy_unused_super(s);
792 return old;
793 }
794 EXPORT_SYMBOL(sget_fc);
795
796 /**
797 * sget - find or create a superblock
798 * @type: filesystem type superblock should belong to
799 * @test: comparison callback
800 * @set: setup callback
801 * @flags: mount flags
802 * @data: argument to each of them
803 */
sget(struct file_system_type * type,int (* test)(struct super_block *,void *),int (* set)(struct super_block *,void *),int flags,void * data)804 struct super_block *sget(struct file_system_type *type,
805 int (*test)(struct super_block *,void *),
806 int (*set)(struct super_block *,void *),
807 int flags,
808 void *data)
809 {
810 struct user_namespace *user_ns = current_user_ns();
811 struct super_block *s = NULL;
812 struct super_block *old;
813 int err;
814
815 /* We don't yet pass the user namespace of the parent
816 * mount through to here so always use &init_user_ns
817 * until that changes.
818 */
819 if (flags & SB_SUBMOUNT)
820 user_ns = &init_user_ns;
821
822 retry:
823 spin_lock(&sb_lock);
824 if (test) {
825 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
826 if (!test(old, data))
827 continue;
828 if (user_ns != old->s_user_ns) {
829 spin_unlock(&sb_lock);
830 destroy_unused_super(s);
831 return ERR_PTR(-EBUSY);
832 }
833 if (!grab_super(old))
834 goto retry;
835 destroy_unused_super(s);
836 return old;
837 }
838 }
839 if (!s) {
840 spin_unlock(&sb_lock);
841 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
842 if (!s)
843 return ERR_PTR(-ENOMEM);
844 goto retry;
845 }
846
847 err = set(s, data);
848 if (err) {
849 spin_unlock(&sb_lock);
850 destroy_unused_super(s);
851 return ERR_PTR(err);
852 }
853 s->s_type = type;
854 strscpy(s->s_id, type->name, sizeof(s->s_id));
855 list_add_tail(&s->s_list, &super_blocks);
856 hlist_add_head(&s->s_instances, &type->fs_supers);
857 spin_unlock(&sb_lock);
858 get_filesystem(type);
859 shrinker_register(s->s_shrink);
860 return s;
861 }
862 EXPORT_SYMBOL(sget);
863
drop_super(struct super_block * sb)864 void drop_super(struct super_block *sb)
865 {
866 super_unlock_shared(sb);
867 put_super(sb);
868 }
869
870 EXPORT_SYMBOL(drop_super);
871
drop_super_exclusive(struct super_block * sb)872 void drop_super_exclusive(struct super_block *sb)
873 {
874 super_unlock_excl(sb);
875 put_super(sb);
876 }
877 EXPORT_SYMBOL(drop_super_exclusive);
878
__iterate_supers(void (* f)(struct super_block *))879 static void __iterate_supers(void (*f)(struct super_block *))
880 {
881 struct super_block *sb, *p = NULL;
882
883 spin_lock(&sb_lock);
884 list_for_each_entry(sb, &super_blocks, s_list) {
885 if (super_flags(sb, SB_DYING))
886 continue;
887 sb->s_count++;
888 spin_unlock(&sb_lock);
889
890 f(sb);
891
892 spin_lock(&sb_lock);
893 if (p)
894 __put_super(p);
895 p = sb;
896 }
897 if (p)
898 __put_super(p);
899 spin_unlock(&sb_lock);
900 }
901 /**
902 * iterate_supers - call function for all active superblocks
903 * @f: function to call
904 * @arg: argument to pass to it
905 *
906 * Scans the superblock list and calls given function, passing it
907 * locked superblock and given argument.
908 */
iterate_supers(void (* f)(struct super_block *,void *),void * arg)909 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
910 {
911 struct super_block *sb, *p = NULL;
912
913 spin_lock(&sb_lock);
914 list_for_each_entry(sb, &super_blocks, s_list) {
915 bool locked;
916
917 sb->s_count++;
918 spin_unlock(&sb_lock);
919
920 locked = super_lock_shared(sb);
921 if (locked) {
922 if (sb->s_root)
923 f(sb, arg);
924 super_unlock_shared(sb);
925 }
926
927 spin_lock(&sb_lock);
928 if (p)
929 __put_super(p);
930 p = sb;
931 }
932 if (p)
933 __put_super(p);
934 spin_unlock(&sb_lock);
935 }
936
937 /**
938 * iterate_supers_type - call function for superblocks of given type
939 * @type: fs type
940 * @f: function to call
941 * @arg: argument to pass to it
942 *
943 * Scans the superblock list and calls given function, passing it
944 * locked superblock and given argument.
945 */
iterate_supers_type(struct file_system_type * type,void (* f)(struct super_block *,void *),void * arg)946 void iterate_supers_type(struct file_system_type *type,
947 void (*f)(struct super_block *, void *), void *arg)
948 {
949 struct super_block *sb, *p = NULL;
950
951 spin_lock(&sb_lock);
952 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
953 bool locked;
954
955 sb->s_count++;
956 spin_unlock(&sb_lock);
957
958 locked = super_lock_shared(sb);
959 if (locked) {
960 if (sb->s_root)
961 f(sb, arg);
962 super_unlock_shared(sb);
963 }
964
965 spin_lock(&sb_lock);
966 if (p)
967 __put_super(p);
968 p = sb;
969 }
970 if (p)
971 __put_super(p);
972 spin_unlock(&sb_lock);
973 }
974
975 EXPORT_SYMBOL(iterate_supers_type);
976
user_get_super(dev_t dev,bool excl)977 struct super_block *user_get_super(dev_t dev, bool excl)
978 {
979 struct super_block *sb;
980
981 spin_lock(&sb_lock);
982 list_for_each_entry(sb, &super_blocks, s_list) {
983 if (sb->s_dev == dev) {
984 bool locked;
985
986 sb->s_count++;
987 spin_unlock(&sb_lock);
988 /* still alive? */
989 locked = super_lock(sb, excl);
990 if (locked) {
991 if (sb->s_root)
992 return sb;
993 super_unlock(sb, excl);
994 }
995 /* nope, got unmounted */
996 spin_lock(&sb_lock);
997 __put_super(sb);
998 break;
999 }
1000 }
1001 spin_unlock(&sb_lock);
1002 return NULL;
1003 }
1004
1005 /**
1006 * reconfigure_super - asks filesystem to change superblock parameters
1007 * @fc: The superblock and configuration
1008 *
1009 * Alters the configuration parameters of a live superblock.
1010 */
reconfigure_super(struct fs_context * fc)1011 int reconfigure_super(struct fs_context *fc)
1012 {
1013 struct super_block *sb = fc->root->d_sb;
1014 int retval;
1015 bool remount_ro = false;
1016 bool remount_rw = false;
1017 bool force = fc->sb_flags & SB_FORCE;
1018
1019 if (fc->sb_flags_mask & ~MS_RMT_MASK)
1020 return -EINVAL;
1021 if (sb->s_writers.frozen != SB_UNFROZEN)
1022 return -EBUSY;
1023
1024 retval = security_sb_remount(sb, fc->security);
1025 if (retval)
1026 return retval;
1027
1028 if (fc->sb_flags_mask & SB_RDONLY) {
1029 #ifdef CONFIG_BLOCK
1030 if (!(fc->sb_flags & SB_RDONLY) && sb->s_bdev &&
1031 bdev_read_only(sb->s_bdev))
1032 return -EACCES;
1033 #endif
1034 remount_rw = !(fc->sb_flags & SB_RDONLY) && sb_rdonly(sb);
1035 remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb);
1036 }
1037
1038 if (remount_ro) {
1039 if (!hlist_empty(&sb->s_pins)) {
1040 super_unlock_excl(sb);
1041 group_pin_kill(&sb->s_pins);
1042 __super_lock_excl(sb);
1043 if (!sb->s_root)
1044 return 0;
1045 if (sb->s_writers.frozen != SB_UNFROZEN)
1046 return -EBUSY;
1047 remount_ro = !sb_rdonly(sb);
1048 }
1049 }
1050 shrink_dcache_sb(sb);
1051
1052 /* If we are reconfiguring to RDONLY and current sb is read/write,
1053 * make sure there are no files open for writing.
1054 */
1055 if (remount_ro) {
1056 if (force) {
1057 sb_start_ro_state_change(sb);
1058 } else {
1059 retval = sb_prepare_remount_readonly(sb);
1060 if (retval)
1061 return retval;
1062 }
1063 } else if (remount_rw) {
1064 /*
1065 * Protect filesystem's reconfigure code from writes from
1066 * userspace until reconfigure finishes.
1067 */
1068 sb_start_ro_state_change(sb);
1069 }
1070
1071 if (fc->ops->reconfigure) {
1072 retval = fc->ops->reconfigure(fc);
1073 if (retval) {
1074 if (!force)
1075 goto cancel_readonly;
1076 /* If forced remount, go ahead despite any errors */
1077 WARN(1, "forced remount of a %s fs returned %i\n",
1078 sb->s_type->name, retval);
1079 }
1080 }
1081
1082 WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) |
1083 (fc->sb_flags & fc->sb_flags_mask)));
1084 sb_end_ro_state_change(sb);
1085
1086 /*
1087 * Some filesystems modify their metadata via some other path than the
1088 * bdev buffer cache (eg. use a private mapping, or directories in
1089 * pagecache, etc). Also file data modifications go via their own
1090 * mappings. So If we try to mount readonly then copy the filesystem
1091 * from bdev, we could get stale data, so invalidate it to give a best
1092 * effort at coherency.
1093 */
1094 if (remount_ro && sb->s_bdev)
1095 invalidate_bdev(sb->s_bdev);
1096 return 0;
1097
1098 cancel_readonly:
1099 sb_end_ro_state_change(sb);
1100 return retval;
1101 }
1102
do_emergency_remount_callback(struct super_block * sb)1103 static void do_emergency_remount_callback(struct super_block *sb)
1104 {
1105 bool locked = super_lock_excl(sb);
1106
1107 if (locked && sb->s_root && sb->s_bdev && !sb_rdonly(sb)) {
1108 struct fs_context *fc;
1109
1110 fc = fs_context_for_reconfigure(sb->s_root,
1111 SB_RDONLY | SB_FORCE, SB_RDONLY);
1112 if (!IS_ERR(fc)) {
1113 if (parse_monolithic_mount_data(fc, NULL) == 0)
1114 (void)reconfigure_super(fc);
1115 put_fs_context(fc);
1116 }
1117 }
1118 if (locked)
1119 super_unlock_excl(sb);
1120 }
1121
do_emergency_remount(struct work_struct * work)1122 static void do_emergency_remount(struct work_struct *work)
1123 {
1124 __iterate_supers(do_emergency_remount_callback);
1125 kfree(work);
1126 printk("Emergency Remount complete\n");
1127 }
1128
emergency_remount(void)1129 void emergency_remount(void)
1130 {
1131 struct work_struct *work;
1132
1133 work = kmalloc(sizeof(*work), GFP_ATOMIC);
1134 if (work) {
1135 INIT_WORK(work, do_emergency_remount);
1136 schedule_work(work);
1137 }
1138 }
1139
do_thaw_all_callback(struct super_block * sb)1140 static void do_thaw_all_callback(struct super_block *sb)
1141 {
1142 bool locked = super_lock_excl(sb);
1143
1144 if (locked && sb->s_root) {
1145 if (IS_ENABLED(CONFIG_BLOCK))
1146 while (sb->s_bdev && !bdev_thaw(sb->s_bdev))
1147 pr_warn("Emergency Thaw on %pg\n", sb->s_bdev);
1148 thaw_super_locked(sb, FREEZE_HOLDER_USERSPACE);
1149 return;
1150 }
1151 if (locked)
1152 super_unlock_excl(sb);
1153 }
1154
do_thaw_all(struct work_struct * work)1155 static void do_thaw_all(struct work_struct *work)
1156 {
1157 __iterate_supers(do_thaw_all_callback);
1158 kfree(work);
1159 printk(KERN_WARNING "Emergency Thaw complete\n");
1160 }
1161
1162 /**
1163 * emergency_thaw_all -- forcibly thaw every frozen filesystem
1164 *
1165 * Used for emergency unfreeze of all filesystems via SysRq
1166 */
emergency_thaw_all(void)1167 void emergency_thaw_all(void)
1168 {
1169 struct work_struct *work;
1170
1171 work = kmalloc(sizeof(*work), GFP_ATOMIC);
1172 if (work) {
1173 INIT_WORK(work, do_thaw_all);
1174 schedule_work(work);
1175 }
1176 }
1177
1178 static DEFINE_IDA(unnamed_dev_ida);
1179
1180 /**
1181 * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1182 * @p: Pointer to a dev_t.
1183 *
1184 * Filesystems which don't use real block devices can call this function
1185 * to allocate a virtual block device.
1186 *
1187 * Context: Any context. Frequently called while holding sb_lock.
1188 * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1189 * or -ENOMEM if memory allocation failed.
1190 */
get_anon_bdev(dev_t * p)1191 int get_anon_bdev(dev_t *p)
1192 {
1193 int dev;
1194
1195 /*
1196 * Many userspace utilities consider an FSID of 0 invalid.
1197 * Always return at least 1 from get_anon_bdev.
1198 */
1199 dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1200 GFP_ATOMIC);
1201 if (dev == -ENOSPC)
1202 dev = -EMFILE;
1203 if (dev < 0)
1204 return dev;
1205
1206 *p = MKDEV(0, dev);
1207 return 0;
1208 }
1209 EXPORT_SYMBOL(get_anon_bdev);
1210
free_anon_bdev(dev_t dev)1211 void free_anon_bdev(dev_t dev)
1212 {
1213 ida_free(&unnamed_dev_ida, MINOR(dev));
1214 }
1215 EXPORT_SYMBOL(free_anon_bdev);
1216
set_anon_super(struct super_block * s,void * data)1217 int set_anon_super(struct super_block *s, void *data)
1218 {
1219 return get_anon_bdev(&s->s_dev);
1220 }
1221 EXPORT_SYMBOL(set_anon_super);
1222
kill_anon_super(struct super_block * sb)1223 void kill_anon_super(struct super_block *sb)
1224 {
1225 dev_t dev = sb->s_dev;
1226 generic_shutdown_super(sb);
1227 kill_super_notify(sb);
1228 free_anon_bdev(dev);
1229 }
1230 EXPORT_SYMBOL(kill_anon_super);
1231
kill_litter_super(struct super_block * sb)1232 void kill_litter_super(struct super_block *sb)
1233 {
1234 if (sb->s_root)
1235 d_genocide(sb->s_root);
1236 kill_anon_super(sb);
1237 }
1238 EXPORT_SYMBOL(kill_litter_super);
1239
set_anon_super_fc(struct super_block * sb,struct fs_context * fc)1240 int set_anon_super_fc(struct super_block *sb, struct fs_context *fc)
1241 {
1242 return set_anon_super(sb, NULL);
1243 }
1244 EXPORT_SYMBOL(set_anon_super_fc);
1245
test_keyed_super(struct super_block * sb,struct fs_context * fc)1246 static int test_keyed_super(struct super_block *sb, struct fs_context *fc)
1247 {
1248 return sb->s_fs_info == fc->s_fs_info;
1249 }
1250
test_single_super(struct super_block * s,struct fs_context * fc)1251 static int test_single_super(struct super_block *s, struct fs_context *fc)
1252 {
1253 return 1;
1254 }
1255
vfs_get_super(struct fs_context * fc,int (* test)(struct super_block *,struct fs_context *),int (* fill_super)(struct super_block * sb,struct fs_context * fc))1256 static int vfs_get_super(struct fs_context *fc,
1257 int (*test)(struct super_block *, struct fs_context *),
1258 int (*fill_super)(struct super_block *sb,
1259 struct fs_context *fc))
1260 {
1261 struct super_block *sb;
1262 int err;
1263
1264 sb = sget_fc(fc, test, set_anon_super_fc);
1265 if (IS_ERR(sb))
1266 return PTR_ERR(sb);
1267
1268 if (!sb->s_root) {
1269 err = fill_super(sb, fc);
1270 if (err)
1271 goto error;
1272
1273 sb->s_flags |= SB_ACTIVE;
1274 }
1275
1276 fc->root = dget(sb->s_root);
1277 return 0;
1278
1279 error:
1280 deactivate_locked_super(sb);
1281 return err;
1282 }
1283
get_tree_nodev(struct fs_context * fc,int (* fill_super)(struct super_block * sb,struct fs_context * fc))1284 int get_tree_nodev(struct fs_context *fc,
1285 int (*fill_super)(struct super_block *sb,
1286 struct fs_context *fc))
1287 {
1288 return vfs_get_super(fc, NULL, fill_super);
1289 }
1290 EXPORT_SYMBOL(get_tree_nodev);
1291
get_tree_single(struct fs_context * fc,int (* fill_super)(struct super_block * sb,struct fs_context * fc))1292 int get_tree_single(struct fs_context *fc,
1293 int (*fill_super)(struct super_block *sb,
1294 struct fs_context *fc))
1295 {
1296 return vfs_get_super(fc, test_single_super, fill_super);
1297 }
1298 EXPORT_SYMBOL(get_tree_single);
1299
get_tree_keyed(struct fs_context * fc,int (* fill_super)(struct super_block * sb,struct fs_context * fc),void * key)1300 int get_tree_keyed(struct fs_context *fc,
1301 int (*fill_super)(struct super_block *sb,
1302 struct fs_context *fc),
1303 void *key)
1304 {
1305 fc->s_fs_info = key;
1306 return vfs_get_super(fc, test_keyed_super, fill_super);
1307 }
1308 EXPORT_SYMBOL(get_tree_keyed);
1309
set_bdev_super(struct super_block * s,void * data)1310 static int set_bdev_super(struct super_block *s, void *data)
1311 {
1312 s->s_dev = *(dev_t *)data;
1313 return 0;
1314 }
1315
super_s_dev_set(struct super_block * s,struct fs_context * fc)1316 static int super_s_dev_set(struct super_block *s, struct fs_context *fc)
1317 {
1318 return set_bdev_super(s, fc->sget_key);
1319 }
1320
super_s_dev_test(struct super_block * s,struct fs_context * fc)1321 static int super_s_dev_test(struct super_block *s, struct fs_context *fc)
1322 {
1323 return !(s->s_iflags & SB_I_RETIRED) &&
1324 s->s_dev == *(dev_t *)fc->sget_key;
1325 }
1326
1327 /**
1328 * sget_dev - Find or create a superblock by device number
1329 * @fc: Filesystem context.
1330 * @dev: device number
1331 *
1332 * Find or create a superblock using the provided device number that
1333 * will be stored in fc->sget_key.
1334 *
1335 * If an extant superblock is matched, then that will be returned with
1336 * an elevated reference count that the caller must transfer or discard.
1337 *
1338 * If no match is made, a new superblock will be allocated and basic
1339 * initialisation will be performed (s_type, s_fs_info, s_id, s_dev will
1340 * be set). The superblock will be published and it will be returned in
1341 * a partially constructed state with SB_BORN and SB_ACTIVE as yet
1342 * unset.
1343 *
1344 * Return: an existing or newly created superblock on success, an error
1345 * pointer on failure.
1346 */
sget_dev(struct fs_context * fc,dev_t dev)1347 struct super_block *sget_dev(struct fs_context *fc, dev_t dev)
1348 {
1349 fc->sget_key = &dev;
1350 return sget_fc(fc, super_s_dev_test, super_s_dev_set);
1351 }
1352 EXPORT_SYMBOL(sget_dev);
1353
1354 #ifdef CONFIG_BLOCK
1355 /*
1356 * Lock the superblock that is holder of the bdev. Returns the superblock
1357 * pointer if we successfully locked the superblock and it is alive. Otherwise
1358 * we return NULL and just unlock bdev->bd_holder_lock.
1359 *
1360 * The function must be called with bdev->bd_holder_lock and releases it.
1361 */
bdev_super_lock(struct block_device * bdev,bool excl)1362 static struct super_block *bdev_super_lock(struct block_device *bdev, bool excl)
1363 __releases(&bdev->bd_holder_lock)
1364 {
1365 struct super_block *sb = bdev->bd_holder;
1366 bool locked;
1367
1368 lockdep_assert_held(&bdev->bd_holder_lock);
1369 lockdep_assert_not_held(&sb->s_umount);
1370 lockdep_assert_not_held(&bdev->bd_disk->open_mutex);
1371
1372 /* Make sure sb doesn't go away from under us */
1373 spin_lock(&sb_lock);
1374 sb->s_count++;
1375 spin_unlock(&sb_lock);
1376
1377 mutex_unlock(&bdev->bd_holder_lock);
1378
1379 locked = super_lock(sb, excl);
1380
1381 /*
1382 * If the superblock wasn't already SB_DYING then we hold
1383 * s_umount and can safely drop our temporary reference.
1384 */
1385 put_super(sb);
1386
1387 if (!locked)
1388 return NULL;
1389
1390 if (!sb->s_root || !(sb->s_flags & SB_ACTIVE)) {
1391 super_unlock(sb, excl);
1392 return NULL;
1393 }
1394
1395 return sb;
1396 }
1397
fs_bdev_mark_dead(struct block_device * bdev,bool surprise)1398 static void fs_bdev_mark_dead(struct block_device *bdev, bool surprise)
1399 {
1400 struct super_block *sb;
1401
1402 sb = bdev_super_lock(bdev, false);
1403 if (!sb)
1404 return;
1405
1406 if (!surprise)
1407 sync_filesystem(sb);
1408 shrink_dcache_sb(sb);
1409 invalidate_inodes(sb);
1410 if (sb->s_op->shutdown)
1411 sb->s_op->shutdown(sb);
1412
1413 super_unlock_shared(sb);
1414 }
1415
fs_bdev_sync(struct block_device * bdev)1416 static void fs_bdev_sync(struct block_device *bdev)
1417 {
1418 struct super_block *sb;
1419
1420 sb = bdev_super_lock(bdev, false);
1421 if (!sb)
1422 return;
1423
1424 sync_filesystem(sb);
1425 super_unlock_shared(sb);
1426 }
1427
get_bdev_super(struct block_device * bdev)1428 static struct super_block *get_bdev_super(struct block_device *bdev)
1429 {
1430 bool active = false;
1431 struct super_block *sb;
1432
1433 sb = bdev_super_lock(bdev, true);
1434 if (sb) {
1435 active = atomic_inc_not_zero(&sb->s_active);
1436 super_unlock_excl(sb);
1437 }
1438 if (!active)
1439 return NULL;
1440 return sb;
1441 }
1442
1443 /**
1444 * fs_bdev_freeze - freeze owning filesystem of block device
1445 * @bdev: block device
1446 *
1447 * Freeze the filesystem that owns this block device if it is still
1448 * active.
1449 *
1450 * A filesystem that owns multiple block devices may be frozen from each
1451 * block device and won't be unfrozen until all block devices are
1452 * unfrozen. Each block device can only freeze the filesystem once as we
1453 * nest freezes for block devices in the block layer.
1454 *
1455 * Return: If the freeze was successful zero is returned. If the freeze
1456 * failed a negative error code is returned.
1457 */
fs_bdev_freeze(struct block_device * bdev)1458 static int fs_bdev_freeze(struct block_device *bdev)
1459 {
1460 struct super_block *sb;
1461 int error = 0;
1462
1463 lockdep_assert_held(&bdev->bd_fsfreeze_mutex);
1464
1465 sb = get_bdev_super(bdev);
1466 if (!sb)
1467 return -EINVAL;
1468
1469 if (sb->s_op->freeze_super)
1470 error = sb->s_op->freeze_super(sb,
1471 FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE);
1472 else
1473 error = freeze_super(sb,
1474 FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE);
1475 if (!error)
1476 error = sync_blockdev(bdev);
1477 deactivate_super(sb);
1478 return error;
1479 }
1480
1481 /**
1482 * fs_bdev_thaw - thaw owning filesystem of block device
1483 * @bdev: block device
1484 *
1485 * Thaw the filesystem that owns this block device.
1486 *
1487 * A filesystem that owns multiple block devices may be frozen from each
1488 * block device and won't be unfrozen until all block devices are
1489 * unfrozen. Each block device can only freeze the filesystem once as we
1490 * nest freezes for block devices in the block layer.
1491 *
1492 * Return: If the thaw was successful zero is returned. If the thaw
1493 * failed a negative error code is returned. If this function
1494 * returns zero it doesn't mean that the filesystem is unfrozen
1495 * as it may have been frozen multiple times (kernel may hold a
1496 * freeze or might be frozen from other block devices).
1497 */
fs_bdev_thaw(struct block_device * bdev)1498 static int fs_bdev_thaw(struct block_device *bdev)
1499 {
1500 struct super_block *sb;
1501 int error;
1502
1503 lockdep_assert_held(&bdev->bd_fsfreeze_mutex);
1504
1505 sb = get_bdev_super(bdev);
1506 if (WARN_ON_ONCE(!sb))
1507 return -EINVAL;
1508
1509 if (sb->s_op->thaw_super)
1510 error = sb->s_op->thaw_super(sb,
1511 FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE);
1512 else
1513 error = thaw_super(sb,
1514 FREEZE_MAY_NEST | FREEZE_HOLDER_USERSPACE);
1515 deactivate_super(sb);
1516 return error;
1517 }
1518
1519 const struct blk_holder_ops fs_holder_ops = {
1520 .mark_dead = fs_bdev_mark_dead,
1521 .sync = fs_bdev_sync,
1522 .freeze = fs_bdev_freeze,
1523 .thaw = fs_bdev_thaw,
1524 };
1525 EXPORT_SYMBOL_GPL(fs_holder_ops);
1526
setup_bdev_super(struct super_block * sb,int sb_flags,struct fs_context * fc)1527 int setup_bdev_super(struct super_block *sb, int sb_flags,
1528 struct fs_context *fc)
1529 {
1530 blk_mode_t mode = sb_open_mode(sb_flags);
1531 struct file *bdev_file;
1532 struct block_device *bdev;
1533
1534 bdev_file = bdev_file_open_by_dev(sb->s_dev, mode, sb, &fs_holder_ops);
1535 if (IS_ERR(bdev_file)) {
1536 if (fc)
1537 errorf(fc, "%s: Can't open blockdev", fc->source);
1538 return PTR_ERR(bdev_file);
1539 }
1540 bdev = file_bdev(bdev_file);
1541
1542 /*
1543 * This really should be in blkdev_get_by_dev, but right now can't due
1544 * to legacy issues that require us to allow opening a block device node
1545 * writable from userspace even for a read-only block device.
1546 */
1547 if ((mode & BLK_OPEN_WRITE) && bdev_read_only(bdev)) {
1548 bdev_fput(bdev_file);
1549 return -EACCES;
1550 }
1551
1552 /*
1553 * It is enough to check bdev was not frozen before we set
1554 * s_bdev as freezing will wait until SB_BORN is set.
1555 */
1556 if (atomic_read(&bdev->bd_fsfreeze_count) > 0) {
1557 if (fc)
1558 warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
1559 bdev_fput(bdev_file);
1560 return -EBUSY;
1561 }
1562 spin_lock(&sb_lock);
1563 sb->s_bdev_file = bdev_file;
1564 sb->s_bdev = bdev;
1565 sb->s_bdi = bdi_get(bdev->bd_disk->bdi);
1566 if (bdev_stable_writes(bdev))
1567 sb->s_iflags |= SB_I_STABLE_WRITES;
1568 spin_unlock(&sb_lock);
1569
1570 snprintf(sb->s_id, sizeof(sb->s_id), "%pg", bdev);
1571 shrinker_debugfs_rename(sb->s_shrink, "sb-%s:%s", sb->s_type->name,
1572 sb->s_id);
1573 sb_set_blocksize(sb, block_size(bdev));
1574 return 0;
1575 }
1576 EXPORT_SYMBOL_GPL(setup_bdev_super);
1577
1578 /**
1579 * get_tree_bdev - Get a superblock based on a single block device
1580 * @fc: The filesystem context holding the parameters
1581 * @fill_super: Helper to initialise a new superblock
1582 */
get_tree_bdev(struct fs_context * fc,int (* fill_super)(struct super_block *,struct fs_context *))1583 int get_tree_bdev(struct fs_context *fc,
1584 int (*fill_super)(struct super_block *,
1585 struct fs_context *))
1586 {
1587 struct super_block *s;
1588 int error = 0;
1589 dev_t dev;
1590
1591 if (!fc->source)
1592 return invalf(fc, "No source specified");
1593
1594 error = lookup_bdev(fc->source, &dev);
1595 if (error) {
1596 errorf(fc, "%s: Can't lookup blockdev", fc->source);
1597 return error;
1598 }
1599
1600 fc->sb_flags |= SB_NOSEC;
1601 s = sget_dev(fc, dev);
1602 if (IS_ERR(s))
1603 return PTR_ERR(s);
1604
1605 if (s->s_root) {
1606 /* Don't summarily change the RO/RW state. */
1607 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1608 warnf(fc, "%pg: Can't mount, would change RO state", s->s_bdev);
1609 deactivate_locked_super(s);
1610 return -EBUSY;
1611 }
1612 } else {
1613 error = setup_bdev_super(s, fc->sb_flags, fc);
1614 if (!error)
1615 error = fill_super(s, fc);
1616 if (error) {
1617 deactivate_locked_super(s);
1618 return error;
1619 }
1620 s->s_flags |= SB_ACTIVE;
1621 }
1622
1623 BUG_ON(fc->root);
1624 fc->root = dget(s->s_root);
1625 return 0;
1626 }
1627 EXPORT_SYMBOL(get_tree_bdev);
1628
test_bdev_super(struct super_block * s,void * data)1629 static int test_bdev_super(struct super_block *s, void *data)
1630 {
1631 return !(s->s_iflags & SB_I_RETIRED) && s->s_dev == *(dev_t *)data;
1632 }
1633
mount_bdev(struct file_system_type * fs_type,int flags,const char * dev_name,void * data,int (* fill_super)(struct super_block *,void *,int))1634 struct dentry *mount_bdev(struct file_system_type *fs_type,
1635 int flags, const char *dev_name, void *data,
1636 int (*fill_super)(struct super_block *, void *, int))
1637 {
1638 struct super_block *s;
1639 int error;
1640 dev_t dev;
1641
1642 error = lookup_bdev(dev_name, &dev);
1643 if (error)
1644 return ERR_PTR(error);
1645
1646 flags |= SB_NOSEC;
1647 s = sget(fs_type, test_bdev_super, set_bdev_super, flags, &dev);
1648 if (IS_ERR(s))
1649 return ERR_CAST(s);
1650
1651 if (s->s_root) {
1652 if ((flags ^ s->s_flags) & SB_RDONLY) {
1653 deactivate_locked_super(s);
1654 return ERR_PTR(-EBUSY);
1655 }
1656 } else {
1657 error = setup_bdev_super(s, flags, NULL);
1658 if (!error)
1659 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1660 if (error) {
1661 deactivate_locked_super(s);
1662 return ERR_PTR(error);
1663 }
1664
1665 s->s_flags |= SB_ACTIVE;
1666 }
1667
1668 return dget(s->s_root);
1669 }
1670 EXPORT_SYMBOL(mount_bdev);
1671
kill_block_super(struct super_block * sb)1672 void kill_block_super(struct super_block *sb)
1673 {
1674 struct block_device *bdev = sb->s_bdev;
1675
1676 generic_shutdown_super(sb);
1677 if (bdev) {
1678 sync_blockdev(bdev);
1679 bdev_fput(sb->s_bdev_file);
1680 }
1681 }
1682
1683 EXPORT_SYMBOL(kill_block_super);
1684 #endif
1685
mount_nodev(struct file_system_type * fs_type,int flags,void * data,int (* fill_super)(struct super_block *,void *,int))1686 struct dentry *mount_nodev(struct file_system_type *fs_type,
1687 int flags, void *data,
1688 int (*fill_super)(struct super_block *, void *, int))
1689 {
1690 int error;
1691 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1692
1693 if (IS_ERR(s))
1694 return ERR_CAST(s);
1695
1696 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1697 if (error) {
1698 deactivate_locked_super(s);
1699 return ERR_PTR(error);
1700 }
1701 s->s_flags |= SB_ACTIVE;
1702 return dget(s->s_root);
1703 }
1704 EXPORT_SYMBOL(mount_nodev);
1705
reconfigure_single(struct super_block * s,int flags,void * data)1706 int reconfigure_single(struct super_block *s,
1707 int flags, void *data)
1708 {
1709 struct fs_context *fc;
1710 int ret;
1711
1712 /* The caller really need to be passing fc down into mount_single(),
1713 * then a chunk of this can be removed. [Bollocks -- AV]
1714 * Better yet, reconfiguration shouldn't happen, but rather the second
1715 * mount should be rejected if the parameters are not compatible.
1716 */
1717 fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK);
1718 if (IS_ERR(fc))
1719 return PTR_ERR(fc);
1720
1721 ret = parse_monolithic_mount_data(fc, data);
1722 if (ret < 0)
1723 goto out;
1724
1725 ret = reconfigure_super(fc);
1726 out:
1727 put_fs_context(fc);
1728 return ret;
1729 }
1730
compare_single(struct super_block * s,void * p)1731 static int compare_single(struct super_block *s, void *p)
1732 {
1733 return 1;
1734 }
1735
mount_single(struct file_system_type * fs_type,int flags,void * data,int (* fill_super)(struct super_block *,void *,int))1736 struct dentry *mount_single(struct file_system_type *fs_type,
1737 int flags, void *data,
1738 int (*fill_super)(struct super_block *, void *, int))
1739 {
1740 struct super_block *s;
1741 int error;
1742
1743 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1744 if (IS_ERR(s))
1745 return ERR_CAST(s);
1746 if (!s->s_root) {
1747 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1748 if (!error)
1749 s->s_flags |= SB_ACTIVE;
1750 } else {
1751 error = reconfigure_single(s, flags, data);
1752 }
1753 if (unlikely(error)) {
1754 deactivate_locked_super(s);
1755 return ERR_PTR(error);
1756 }
1757 return dget(s->s_root);
1758 }
1759 EXPORT_SYMBOL(mount_single);
1760
1761 /**
1762 * vfs_get_tree - Get the mountable root
1763 * @fc: The superblock configuration context.
1764 *
1765 * The filesystem is invoked to get or create a superblock which can then later
1766 * be used for mounting. The filesystem places a pointer to the root to be
1767 * used for mounting in @fc->root.
1768 */
vfs_get_tree(struct fs_context * fc)1769 int vfs_get_tree(struct fs_context *fc)
1770 {
1771 struct super_block *sb;
1772 int error;
1773
1774 if (fc->root)
1775 return -EBUSY;
1776
1777 /* Get the mountable root in fc->root, with a ref on the root and a ref
1778 * on the superblock.
1779 */
1780 error = fc->ops->get_tree(fc);
1781 if (error < 0)
1782 return error;
1783
1784 if (!fc->root) {
1785 pr_err("Filesystem %s get_tree() didn't set fc->root\n",
1786 fc->fs_type->name);
1787 /* We don't know what the locking state of the superblock is -
1788 * if there is a superblock.
1789 */
1790 BUG();
1791 }
1792
1793 sb = fc->root->d_sb;
1794 WARN_ON(!sb->s_bdi);
1795
1796 /*
1797 * super_wake() contains a memory barrier which also care of
1798 * ordering for super_cache_count(). We place it before setting
1799 * SB_BORN as the data dependency between the two functions is
1800 * the superblock structure contents that we just set up, not
1801 * the SB_BORN flag.
1802 */
1803 super_wake(sb, SB_BORN);
1804
1805 error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
1806 if (unlikely(error)) {
1807 fc_drop_locked(fc);
1808 return error;
1809 }
1810
1811 /*
1812 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1813 * but s_maxbytes was an unsigned long long for many releases. Throw
1814 * this warning for a little while to try and catch filesystems that
1815 * violate this rule.
1816 */
1817 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1818 "negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes);
1819
1820 return 0;
1821 }
1822 EXPORT_SYMBOL(vfs_get_tree);
1823
1824 /*
1825 * Setup private BDI for given superblock. It gets automatically cleaned up
1826 * in generic_shutdown_super().
1827 */
super_setup_bdi_name(struct super_block * sb,char * fmt,...)1828 int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1829 {
1830 struct backing_dev_info *bdi;
1831 int err;
1832 va_list args;
1833
1834 bdi = bdi_alloc(NUMA_NO_NODE);
1835 if (!bdi)
1836 return -ENOMEM;
1837
1838 va_start(args, fmt);
1839 err = bdi_register_va(bdi, fmt, args);
1840 va_end(args);
1841 if (err) {
1842 bdi_put(bdi);
1843 return err;
1844 }
1845 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1846 sb->s_bdi = bdi;
1847 sb->s_iflags |= SB_I_PERSB_BDI;
1848
1849 return 0;
1850 }
1851 EXPORT_SYMBOL(super_setup_bdi_name);
1852
1853 /*
1854 * Setup private BDI for given superblock. I gets automatically cleaned up
1855 * in generic_shutdown_super().
1856 */
super_setup_bdi(struct super_block * sb)1857 int super_setup_bdi(struct super_block *sb)
1858 {
1859 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1860
1861 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1862 atomic_long_inc_return(&bdi_seq));
1863 }
1864 EXPORT_SYMBOL(super_setup_bdi);
1865
1866 /**
1867 * sb_wait_write - wait until all writers to given file system finish
1868 * @sb: the super for which we wait
1869 * @level: type of writers we wait for (normal vs page fault)
1870 *
1871 * This function waits until there are no writers of given type to given file
1872 * system.
1873 */
sb_wait_write(struct super_block * sb,int level)1874 static void sb_wait_write(struct super_block *sb, int level)
1875 {
1876 percpu_down_write(sb->s_writers.rw_sem + level-1);
1877 }
1878
1879 /*
1880 * We are going to return to userspace and forget about these locks, the
1881 * ownership goes to the caller of thaw_super() which does unlock().
1882 */
lockdep_sb_freeze_release(struct super_block * sb)1883 static void lockdep_sb_freeze_release(struct super_block *sb)
1884 {
1885 int level;
1886
1887 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1888 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1889 }
1890
1891 /*
1892 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1893 */
lockdep_sb_freeze_acquire(struct super_block * sb)1894 static void lockdep_sb_freeze_acquire(struct super_block *sb)
1895 {
1896 int level;
1897
1898 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1899 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1900 }
1901
sb_freeze_unlock(struct super_block * sb,int level)1902 static void sb_freeze_unlock(struct super_block *sb, int level)
1903 {
1904 for (level--; level >= 0; level--)
1905 percpu_up_write(sb->s_writers.rw_sem + level);
1906 }
1907
wait_for_partially_frozen(struct super_block * sb)1908 static int wait_for_partially_frozen(struct super_block *sb)
1909 {
1910 int ret = 0;
1911
1912 do {
1913 unsigned short old = sb->s_writers.frozen;
1914
1915 up_write(&sb->s_umount);
1916 ret = wait_var_event_killable(&sb->s_writers.frozen,
1917 sb->s_writers.frozen != old);
1918 down_write(&sb->s_umount);
1919 } while (ret == 0 &&
1920 sb->s_writers.frozen != SB_UNFROZEN &&
1921 sb->s_writers.frozen != SB_FREEZE_COMPLETE);
1922
1923 return ret;
1924 }
1925
1926 #define FREEZE_HOLDERS (FREEZE_HOLDER_KERNEL | FREEZE_HOLDER_USERSPACE)
1927 #define FREEZE_FLAGS (FREEZE_HOLDERS | FREEZE_MAY_NEST)
1928
freeze_inc(struct super_block * sb,enum freeze_holder who)1929 static inline int freeze_inc(struct super_block *sb, enum freeze_holder who)
1930 {
1931 WARN_ON_ONCE((who & ~FREEZE_FLAGS));
1932 WARN_ON_ONCE(hweight32(who & FREEZE_HOLDERS) > 1);
1933
1934 if (who & FREEZE_HOLDER_KERNEL)
1935 ++sb->s_writers.freeze_kcount;
1936 if (who & FREEZE_HOLDER_USERSPACE)
1937 ++sb->s_writers.freeze_ucount;
1938 return sb->s_writers.freeze_kcount + sb->s_writers.freeze_ucount;
1939 }
1940
freeze_dec(struct super_block * sb,enum freeze_holder who)1941 static inline int freeze_dec(struct super_block *sb, enum freeze_holder who)
1942 {
1943 WARN_ON_ONCE((who & ~FREEZE_FLAGS));
1944 WARN_ON_ONCE(hweight32(who & FREEZE_HOLDERS) > 1);
1945
1946 if ((who & FREEZE_HOLDER_KERNEL) && sb->s_writers.freeze_kcount)
1947 --sb->s_writers.freeze_kcount;
1948 if ((who & FREEZE_HOLDER_USERSPACE) && sb->s_writers.freeze_ucount)
1949 --sb->s_writers.freeze_ucount;
1950 return sb->s_writers.freeze_kcount + sb->s_writers.freeze_ucount;
1951 }
1952
may_freeze(struct super_block * sb,enum freeze_holder who)1953 static inline bool may_freeze(struct super_block *sb, enum freeze_holder who)
1954 {
1955 WARN_ON_ONCE((who & ~FREEZE_FLAGS));
1956 WARN_ON_ONCE(hweight32(who & FREEZE_HOLDERS) > 1);
1957
1958 if (who & FREEZE_HOLDER_KERNEL)
1959 return (who & FREEZE_MAY_NEST) ||
1960 sb->s_writers.freeze_kcount == 0;
1961 if (who & FREEZE_HOLDER_USERSPACE)
1962 return (who & FREEZE_MAY_NEST) ||
1963 sb->s_writers.freeze_ucount == 0;
1964 return false;
1965 }
1966
1967 /**
1968 * freeze_super - lock the filesystem and force it into a consistent state
1969 * @sb: the super to lock
1970 * @who: context that wants to freeze
1971 *
1972 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1973 * freeze_fs. Subsequent calls to this without first thawing the fs may return
1974 * -EBUSY.
1975 *
1976 * @who should be:
1977 * * %FREEZE_HOLDER_USERSPACE if userspace wants to freeze the fs;
1978 * * %FREEZE_HOLDER_KERNEL if the kernel wants to freeze the fs.
1979 * * %FREEZE_MAY_NEST whether nesting freeze and thaw requests is allowed.
1980 *
1981 * The @who argument distinguishes between the kernel and userspace trying to
1982 * freeze the filesystem. Although there cannot be multiple kernel freezes or
1983 * multiple userspace freezes in effect at any given time, the kernel and
1984 * userspace can both hold a filesystem frozen. The filesystem remains frozen
1985 * until there are no kernel or userspace freezes in effect.
1986 *
1987 * A filesystem may hold multiple devices and thus a filesystems may be
1988 * frozen through the block layer via multiple block devices. In this
1989 * case the request is marked as being allowed to nest by passing
1990 * FREEZE_MAY_NEST. The filesystem remains frozen until all block
1991 * devices are unfrozen. If multiple freezes are attempted without
1992 * FREEZE_MAY_NEST -EBUSY will be returned.
1993 *
1994 * During this function, sb->s_writers.frozen goes through these values:
1995 *
1996 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1997 *
1998 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1999 * writes should be blocked, though page faults are still allowed. We wait for
2000 * all writes to complete and then proceed to the next stage.
2001 *
2002 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
2003 * but internal fs threads can still modify the filesystem (although they
2004 * should not dirty new pages or inodes), writeback can run etc. After waiting
2005 * for all running page faults we sync the filesystem which will clean all
2006 * dirty pages and inodes (no new dirty pages or inodes can be created when
2007 * sync is running).
2008 *
2009 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
2010 * modification are blocked (e.g. XFS preallocation truncation on inode
2011 * reclaim). This is usually implemented by blocking new transactions for
2012 * filesystems that have them and need this additional guard. After all
2013 * internal writers are finished we call ->freeze_fs() to finish filesystem
2014 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
2015 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
2016 *
2017 * sb->s_writers.frozen is protected by sb->s_umount.
2018 *
2019 * Return: If the freeze was successful zero is returned. If the freeze
2020 * failed a negative error code is returned.
2021 */
freeze_super(struct super_block * sb,enum freeze_holder who)2022 int freeze_super(struct super_block *sb, enum freeze_holder who)
2023 {
2024 int ret;
2025
2026 if (!super_lock_excl(sb)) {
2027 WARN_ON_ONCE("Dying superblock while freezing!");
2028 return -EINVAL;
2029 }
2030 atomic_inc(&sb->s_active);
2031
2032 retry:
2033 if (sb->s_writers.frozen == SB_FREEZE_COMPLETE) {
2034 if (may_freeze(sb, who))
2035 ret = !!WARN_ON_ONCE(freeze_inc(sb, who) == 1);
2036 else
2037 ret = -EBUSY;
2038 /* All freezers share a single active reference. */
2039 deactivate_locked_super(sb);
2040 return ret;
2041 }
2042
2043 if (sb->s_writers.frozen != SB_UNFROZEN) {
2044 ret = wait_for_partially_frozen(sb);
2045 if (ret) {
2046 deactivate_locked_super(sb);
2047 return ret;
2048 }
2049
2050 goto retry;
2051 }
2052
2053 if (sb_rdonly(sb)) {
2054 /* Nothing to do really... */
2055 WARN_ON_ONCE(freeze_inc(sb, who) > 1);
2056 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
2057 wake_up_var(&sb->s_writers.frozen);
2058 super_unlock_excl(sb);
2059 return 0;
2060 }
2061
2062 sb->s_writers.frozen = SB_FREEZE_WRITE;
2063 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
2064 super_unlock_excl(sb);
2065 sb_wait_write(sb, SB_FREEZE_WRITE);
2066 __super_lock_excl(sb);
2067
2068 /* Now we go and block page faults... */
2069 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
2070 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
2071
2072 /* All writers are done so after syncing there won't be dirty data */
2073 ret = sync_filesystem(sb);
2074 if (ret) {
2075 sb->s_writers.frozen = SB_UNFROZEN;
2076 sb_freeze_unlock(sb, SB_FREEZE_PAGEFAULT);
2077 wake_up_var(&sb->s_writers.frozen);
2078 deactivate_locked_super(sb);
2079 return ret;
2080 }
2081
2082 /* Now wait for internal filesystem counter */
2083 sb->s_writers.frozen = SB_FREEZE_FS;
2084 sb_wait_write(sb, SB_FREEZE_FS);
2085
2086 if (sb->s_op->freeze_fs) {
2087 ret = sb->s_op->freeze_fs(sb);
2088 if (ret) {
2089 printk(KERN_ERR
2090 "VFS:Filesystem freeze failed\n");
2091 sb->s_writers.frozen = SB_UNFROZEN;
2092 sb_freeze_unlock(sb, SB_FREEZE_FS);
2093 wake_up_var(&sb->s_writers.frozen);
2094 deactivate_locked_super(sb);
2095 return ret;
2096 }
2097 }
2098 /*
2099 * For debugging purposes so that fs can warn if it sees write activity
2100 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
2101 */
2102 WARN_ON_ONCE(freeze_inc(sb, who) > 1);
2103 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
2104 wake_up_var(&sb->s_writers.frozen);
2105 lockdep_sb_freeze_release(sb);
2106 super_unlock_excl(sb);
2107 return 0;
2108 }
2109 EXPORT_SYMBOL(freeze_super);
2110
2111 /*
2112 * Undoes the effect of a freeze_super_locked call. If the filesystem is
2113 * frozen both by userspace and the kernel, a thaw call from either source
2114 * removes that state without releasing the other state or unlocking the
2115 * filesystem.
2116 */
thaw_super_locked(struct super_block * sb,enum freeze_holder who)2117 static int thaw_super_locked(struct super_block *sb, enum freeze_holder who)
2118 {
2119 int error = -EINVAL;
2120
2121 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE)
2122 goto out_unlock;
2123
2124 /*
2125 * All freezers share a single active reference.
2126 * So just unlock in case there are any left.
2127 */
2128 if (freeze_dec(sb, who))
2129 goto out_unlock;
2130
2131 if (sb_rdonly(sb)) {
2132 sb->s_writers.frozen = SB_UNFROZEN;
2133 wake_up_var(&sb->s_writers.frozen);
2134 goto out_deactivate;
2135 }
2136
2137 lockdep_sb_freeze_acquire(sb);
2138
2139 if (sb->s_op->unfreeze_fs) {
2140 error = sb->s_op->unfreeze_fs(sb);
2141 if (error) {
2142 pr_err("VFS: Filesystem thaw failed\n");
2143 freeze_inc(sb, who);
2144 lockdep_sb_freeze_release(sb);
2145 goto out_unlock;
2146 }
2147 }
2148
2149 sb->s_writers.frozen = SB_UNFROZEN;
2150 wake_up_var(&sb->s_writers.frozen);
2151 sb_freeze_unlock(sb, SB_FREEZE_FS);
2152 out_deactivate:
2153 deactivate_locked_super(sb);
2154 return 0;
2155
2156 out_unlock:
2157 super_unlock_excl(sb);
2158 return error;
2159 }
2160
2161 /**
2162 * thaw_super -- unlock filesystem
2163 * @sb: the super to thaw
2164 * @who: context that wants to freeze
2165 *
2166 * Unlocks the filesystem and marks it writeable again after freeze_super()
2167 * if there are no remaining freezes on the filesystem.
2168 *
2169 * @who should be:
2170 * * %FREEZE_HOLDER_USERSPACE if userspace wants to thaw the fs;
2171 * * %FREEZE_HOLDER_KERNEL if the kernel wants to thaw the fs.
2172 * * %FREEZE_MAY_NEST whether nesting freeze and thaw requests is allowed
2173 *
2174 * A filesystem may hold multiple devices and thus a filesystems may
2175 * have been frozen through the block layer via multiple block devices.
2176 * The filesystem remains frozen until all block devices are unfrozen.
2177 */
thaw_super(struct super_block * sb,enum freeze_holder who)2178 int thaw_super(struct super_block *sb, enum freeze_holder who)
2179 {
2180 if (!super_lock_excl(sb)) {
2181 WARN_ON_ONCE("Dying superblock while thawing!");
2182 return -EINVAL;
2183 }
2184 return thaw_super_locked(sb, who);
2185 }
2186 EXPORT_SYMBOL(thaw_super);
2187
2188 /*
2189 * Create workqueue for deferred direct IO completions. We allocate the
2190 * workqueue when it's first needed. This avoids creating workqueue for
2191 * filesystems that don't need it and also allows us to create the workqueue
2192 * late enough so the we can include s_id in the name of the workqueue.
2193 */
sb_init_dio_done_wq(struct super_block * sb)2194 int sb_init_dio_done_wq(struct super_block *sb)
2195 {
2196 struct workqueue_struct *old;
2197 struct workqueue_struct *wq = alloc_workqueue("dio/%s",
2198 WQ_MEM_RECLAIM, 0,
2199 sb->s_id);
2200 if (!wq)
2201 return -ENOMEM;
2202 /*
2203 * This has to be atomic as more DIOs can race to create the workqueue
2204 */
2205 old = cmpxchg(&sb->s_dio_done_wq, NULL, wq);
2206 /* Someone created workqueue before us? Free ours... */
2207 if (old)
2208 destroy_workqueue(wq);
2209 return 0;
2210 }
2211 EXPORT_SYMBOL_GPL(sb_init_dio_done_wq);
2212