1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
7 * super.c contains code to handle: - mount structures
9 * - filesystem drivers list
11 * - umount system call
14 * GK 2/5/95 - Changed to support mounting the root fs via NFS
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
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>
42 static int thaw_super_locked(struct super_block *sb);
44 static LIST_HEAD(super_blocks);
45 static DEFINE_SPINLOCK(sb_lock);
47 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
54 * One thing we have to be careful of with a per-sb shrinker is that we don't
55 * drop the last active reference to the superblock from within the shrinker.
56 * If that happens we could trigger unregistering the shrinker from within the
57 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
58 * take a passive reference to the superblock to avoid this from occurring.
60 static unsigned long super_cache_scan(struct shrinker *shrink,
61 struct shrink_control *sc)
63 struct super_block *sb;
70 sb = container_of(shrink, struct super_block, s_shrink);
73 * Deadlock avoidance. We may hold various FS locks, and we don't want
74 * to recurse into the FS that called us in clear_inode() and friends..
76 if (!(sc->gfp_mask & __GFP_FS))
79 if (!trylock_super(sb))
82 if (sb->s_op->nr_cached_objects)
83 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
85 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
86 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
87 total_objects = dentries + inodes + fs_objects + 1;
91 /* proportion the scan between the caches */
92 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
93 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
94 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
97 * prune the dcache first as the icache is pinned by it, then
98 * prune the icache, followed by the filesystem specific caches
100 * Ensure that we always scan at least one object - memcg kmem
101 * accounting uses this to fully empty the caches.
103 sc->nr_to_scan = dentries + 1;
104 freed = prune_dcache_sb(sb, sc);
105 sc->nr_to_scan = inodes + 1;
106 freed += prune_icache_sb(sb, sc);
109 sc->nr_to_scan = fs_objects + 1;
110 freed += sb->s_op->free_cached_objects(sb, sc);
113 up_read(&sb->s_umount);
117 static unsigned long super_cache_count(struct shrinker *shrink,
118 struct shrink_control *sc)
120 struct super_block *sb;
121 long total_objects = 0;
123 sb = container_of(shrink, struct super_block, s_shrink);
126 * We don't call trylock_super() here as it is a scalability bottleneck,
127 * so we're exposed to partial setup state. The shrinker rwsem does not
128 * protect filesystem operations backing list_lru_shrink_count() or
129 * s_op->nr_cached_objects(). Counts can change between
130 * super_cache_count and super_cache_scan, so we really don't need locks
133 * However, if we are currently mounting the superblock, the underlying
134 * filesystem might be in a state of partial construction and hence it
135 * is dangerous to access it. trylock_super() uses a SB_BORN check to
136 * avoid this situation, so do the same here. The memory barrier is
137 * matched with the one in mount_fs() as we don't hold locks here.
139 if (!(sb->s_flags & SB_BORN))
143 if (sb->s_op && sb->s_op->nr_cached_objects)
144 total_objects = sb->s_op->nr_cached_objects(sb, sc);
146 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
147 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
152 total_objects = vfs_pressure_ratio(total_objects);
153 return total_objects;
156 static void destroy_super_work(struct work_struct *work)
158 struct super_block *s = container_of(work, struct super_block,
162 for (i = 0; i < SB_FREEZE_LEVELS; i++)
163 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
167 static void destroy_super_rcu(struct rcu_head *head)
169 struct super_block *s = container_of(head, struct super_block, rcu);
170 INIT_WORK(&s->destroy_work, destroy_super_work);
171 schedule_work(&s->destroy_work);
174 /* Free a superblock that has never been seen by anyone */
175 static void destroy_unused_super(struct super_block *s)
179 up_write(&s->s_umount);
180 list_lru_destroy(&s->s_dentry_lru);
181 list_lru_destroy(&s->s_inode_lru);
183 put_user_ns(s->s_user_ns);
185 free_prealloced_shrinker(&s->s_shrink);
186 /* no delays needed */
187 destroy_super_work(&s->destroy_work);
191 * alloc_super - create new superblock
192 * @type: filesystem type superblock should belong to
193 * @flags: the mount flags
194 * @user_ns: User namespace for the super_block
196 * Allocates and initializes a new &struct super_block. alloc_super()
197 * returns a pointer new superblock or %NULL if allocation had failed.
199 static struct super_block *alloc_super(struct file_system_type *type, int flags,
200 struct user_namespace *user_ns)
202 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
203 static const struct super_operations default_op;
209 INIT_LIST_HEAD(&s->s_mounts);
210 s->s_user_ns = get_user_ns(user_ns);
211 init_rwsem(&s->s_umount);
212 lockdep_set_class(&s->s_umount, &type->s_umount_key);
214 * sget() can have s_umount recursion.
216 * When it cannot find a suitable sb, it allocates a new
217 * one (this one), and tries again to find a suitable old
220 * In case that succeeds, it will acquire the s_umount
221 * lock of the old one. Since these are clearly distrinct
222 * locks, and this object isn't exposed yet, there's no
225 * Annotate this by putting this lock in a different
228 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
230 if (security_sb_alloc(s))
233 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
234 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
236 &type->s_writers_key[i]))
239 s->s_bdi = &noop_backing_dev_info;
241 if (s->s_user_ns != &init_user_ns)
242 s->s_iflags |= SB_I_NODEV;
243 INIT_HLIST_NODE(&s->s_instances);
244 INIT_HLIST_BL_HEAD(&s->s_roots);
245 mutex_init(&s->s_sync_lock);
246 INIT_LIST_HEAD(&s->s_inodes);
247 spin_lock_init(&s->s_inode_list_lock);
248 INIT_LIST_HEAD(&s->s_inodes_wb);
249 spin_lock_init(&s->s_inode_wblist_lock);
252 atomic_set(&s->s_active, 1);
253 mutex_init(&s->s_vfs_rename_mutex);
254 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
255 init_rwsem(&s->s_dquot.dqio_sem);
256 s->s_maxbytes = MAX_NON_LFS;
257 s->s_op = &default_op;
258 s->s_time_gran = 1000000000;
259 s->s_time_min = TIME64_MIN;
260 s->s_time_max = TIME64_MAX;
262 s->s_shrink.seeks = DEFAULT_SEEKS;
263 s->s_shrink.scan_objects = super_cache_scan;
264 s->s_shrink.count_objects = super_cache_count;
265 s->s_shrink.batch = 1024;
266 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
267 if (prealloc_shrinker(&s->s_shrink, "sb-%s", type->name))
269 if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink))
271 if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink))
276 destroy_unused_super(s);
280 /* Superblock refcounting */
283 * Drop a superblock's refcount. The caller must hold sb_lock.
285 static void __put_super(struct super_block *s)
288 list_del_init(&s->s_list);
289 WARN_ON(s->s_dentry_lru.node);
290 WARN_ON(s->s_inode_lru.node);
291 WARN_ON(!list_empty(&s->s_mounts));
293 put_user_ns(s->s_user_ns);
295 call_rcu(&s->rcu, destroy_super_rcu);
300 * put_super - drop a temporary reference to superblock
301 * @sb: superblock in question
303 * Drops a temporary reference, frees superblock if there's no
306 void put_super(struct super_block *sb)
310 spin_unlock(&sb_lock);
315 * deactivate_locked_super - drop an active reference to superblock
316 * @s: superblock to deactivate
318 * Drops an active reference to superblock, converting it into a temporary
319 * one if there is no other active references left. In that case we
320 * tell fs driver to shut it down and drop the temporary reference we
323 * Caller holds exclusive lock on superblock; that lock is released.
325 void deactivate_locked_super(struct super_block *s)
327 struct file_system_type *fs = s->s_type;
328 if (atomic_dec_and_test(&s->s_active)) {
329 unregister_shrinker(&s->s_shrink);
333 * Since list_lru_destroy() may sleep, we cannot call it from
334 * put_super(), where we hold the sb_lock. Therefore we destroy
335 * the lru lists right now.
337 list_lru_destroy(&s->s_dentry_lru);
338 list_lru_destroy(&s->s_inode_lru);
343 up_write(&s->s_umount);
347 EXPORT_SYMBOL(deactivate_locked_super);
350 * deactivate_super - drop an active reference to superblock
351 * @s: superblock to deactivate
353 * Variant of deactivate_locked_super(), except that superblock is *not*
354 * locked by caller. If we are going to drop the final active reference,
355 * lock will be acquired prior to that.
357 void deactivate_super(struct super_block *s)
359 if (!atomic_add_unless(&s->s_active, -1, 1)) {
360 down_write(&s->s_umount);
361 deactivate_locked_super(s);
365 EXPORT_SYMBOL(deactivate_super);
368 * grab_super - acquire an active reference
369 * @s: reference we are trying to make active
371 * Tries to acquire an active reference. grab_super() is used when we
372 * had just found a superblock in super_blocks or fs_type->fs_supers
373 * and want to turn it into a full-blown active reference. grab_super()
374 * is called with sb_lock held and drops it. Returns 1 in case of
375 * success, 0 if we had failed (superblock contents was already dead or
376 * dying when grab_super() had been called). Note that this is only
377 * called for superblocks not in rundown mode (== ones still on ->fs_supers
378 * of their type), so increment of ->s_count is OK here.
380 static int grab_super(struct super_block *s) __releases(sb_lock)
383 spin_unlock(&sb_lock);
384 down_write(&s->s_umount);
385 if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
389 up_write(&s->s_umount);
395 * trylock_super - try to grab ->s_umount shared
396 * @sb: reference we are trying to grab
398 * Try to prevent fs shutdown. This is used in places where we
399 * cannot take an active reference but we need to ensure that the
400 * filesystem is not shut down while we are working on it. It returns
401 * false if we cannot acquire s_umount or if we lose the race and
402 * filesystem already got into shutdown, and returns true with the s_umount
403 * lock held in read mode in case of success. On successful return,
404 * the caller must drop the s_umount lock when done.
406 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
407 * The reason why it's safe is that we are OK with doing trylock instead
408 * of down_read(). There's a couple of places that are OK with that, but
409 * it's very much not a general-purpose interface.
411 bool trylock_super(struct super_block *sb)
413 if (down_read_trylock(&sb->s_umount)) {
414 if (!hlist_unhashed(&sb->s_instances) &&
415 sb->s_root && (sb->s_flags & SB_BORN))
417 up_read(&sb->s_umount);
424 * retire_super - prevents superblock from being reused
425 * @sb: superblock to retire
427 * The function marks superblock to be ignored in superblock test, which
428 * prevents it from being reused for any new mounts. If the superblock has
429 * a private bdi, it also unregisters it, but doesn't reduce the refcount
430 * of the superblock to prevent potential races. The refcount is reduced
431 * by generic_shutdown_super(). The function can not be called
432 * concurrently with generic_shutdown_super(). It is safe to call the
433 * function multiple times, subsequent calls have no effect.
435 * The marker will affect the re-use only for block-device-based
436 * superblocks. Other superblocks will still get marked if this function
437 * is used, but that will not affect their reusability.
439 void retire_super(struct super_block *sb)
441 WARN_ON(!sb->s_bdev);
442 down_write(&sb->s_umount);
443 if (sb->s_iflags & SB_I_PERSB_BDI) {
444 bdi_unregister(sb->s_bdi);
445 sb->s_iflags &= ~SB_I_PERSB_BDI;
447 sb->s_iflags |= SB_I_RETIRED;
448 up_write(&sb->s_umount);
450 EXPORT_SYMBOL(retire_super);
453 * generic_shutdown_super - common helper for ->kill_sb()
454 * @sb: superblock to kill
456 * generic_shutdown_super() does all fs-independent work on superblock
457 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
458 * that need destruction out of superblock, call generic_shutdown_super()
459 * and release aforementioned objects. Note: dentries and inodes _are_
460 * taken care of and do not need specific handling.
462 * Upon calling this function, the filesystem may no longer alter or
463 * rearrange the set of dentries belonging to this super_block, nor may it
464 * change the attachments of dentries to inodes.
466 void generic_shutdown_super(struct super_block *sb)
468 const struct super_operations *sop = sb->s_op;
471 shrink_dcache_for_umount(sb);
473 sb->s_flags &= ~SB_ACTIVE;
475 cgroup_writeback_umount();
477 /* Evict all inodes with zero refcount. */
481 * Clean up and evict any inodes that still have references due
482 * to fsnotify or the security policy.
484 fsnotify_sb_delete(sb);
485 security_sb_delete(sb);
488 * Now that all potentially-encrypted inodes have been evicted,
489 * the fscrypt keyring can be destroyed.
491 fscrypt_destroy_keyring(sb);
493 if (sb->s_dio_done_wq) {
494 destroy_workqueue(sb->s_dio_done_wq);
495 sb->s_dio_done_wq = NULL;
501 if (CHECK_DATA_CORRUPTION(!list_empty(&sb->s_inodes),
502 "VFS: Busy inodes after unmount of %s (%s)",
503 sb->s_id, sb->s_type->name)) {
505 * Adding a proper bailout path here would be hard, but
506 * we can at least make it more likely that a later
507 * iput_final() or such crashes cleanly.
511 spin_lock(&sb->s_inode_list_lock);
512 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
513 inode->i_op = VFS_PTR_POISON;
514 inode->i_sb = VFS_PTR_POISON;
515 inode->i_mapping = VFS_PTR_POISON;
517 spin_unlock(&sb->s_inode_list_lock);
521 /* should be initialized for __put_super_and_need_restart() */
522 hlist_del_init(&sb->s_instances);
523 spin_unlock(&sb_lock);
524 up_write(&sb->s_umount);
525 if (sb->s_bdi != &noop_backing_dev_info) {
526 if (sb->s_iflags & SB_I_PERSB_BDI)
527 bdi_unregister(sb->s_bdi);
529 sb->s_bdi = &noop_backing_dev_info;
533 EXPORT_SYMBOL(generic_shutdown_super);
535 bool mount_capable(struct fs_context *fc)
537 if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT))
538 return capable(CAP_SYS_ADMIN);
540 return ns_capable(fc->user_ns, CAP_SYS_ADMIN);
544 * sget_fc - Find or create a superblock
545 * @fc: Filesystem context.
546 * @test: Comparison callback
547 * @set: Setup callback
549 * Find or create a superblock using the parameters stored in the filesystem
550 * context and the two callback functions.
552 * If an extant superblock is matched, then that will be returned with an
553 * elevated reference count that the caller must transfer or discard.
555 * If no match is made, a new superblock will be allocated and basic
556 * initialisation will be performed (s_type, s_fs_info and s_id will be set and
557 * the set() callback will be invoked), the superblock will be published and it
558 * will be returned in a partially constructed state with SB_BORN and SB_ACTIVE
561 struct super_block *sget_fc(struct fs_context *fc,
562 int (*test)(struct super_block *, struct fs_context *),
563 int (*set)(struct super_block *, struct fs_context *))
565 struct super_block *s = NULL;
566 struct super_block *old;
567 struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns;
573 hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) {
575 goto share_extant_sb;
579 spin_unlock(&sb_lock);
580 s = alloc_super(fc->fs_type, fc->sb_flags, user_ns);
582 return ERR_PTR(-ENOMEM);
586 s->s_fs_info = fc->s_fs_info;
590 spin_unlock(&sb_lock);
591 destroy_unused_super(s);
594 fc->s_fs_info = NULL;
595 s->s_type = fc->fs_type;
596 s->s_iflags |= fc->s_iflags;
597 strscpy(s->s_id, s->s_type->name, sizeof(s->s_id));
598 list_add_tail(&s->s_list, &super_blocks);
599 hlist_add_head(&s->s_instances, &s->s_type->fs_supers);
600 spin_unlock(&sb_lock);
601 get_filesystem(s->s_type);
602 register_shrinker_prepared(&s->s_shrink);
606 if (user_ns != old->s_user_ns) {
607 spin_unlock(&sb_lock);
608 destroy_unused_super(s);
609 return ERR_PTR(-EBUSY);
611 if (!grab_super(old))
613 destroy_unused_super(s);
616 EXPORT_SYMBOL(sget_fc);
619 * sget - find or create a superblock
620 * @type: filesystem type superblock should belong to
621 * @test: comparison callback
622 * @set: setup callback
623 * @flags: mount flags
624 * @data: argument to each of them
626 struct super_block *sget(struct file_system_type *type,
627 int (*test)(struct super_block *,void *),
628 int (*set)(struct super_block *,void *),
632 struct user_namespace *user_ns = current_user_ns();
633 struct super_block *s = NULL;
634 struct super_block *old;
637 /* We don't yet pass the user namespace of the parent
638 * mount through to here so always use &init_user_ns
639 * until that changes.
641 if (flags & SB_SUBMOUNT)
642 user_ns = &init_user_ns;
647 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
648 if (!test(old, data))
650 if (user_ns != old->s_user_ns) {
651 spin_unlock(&sb_lock);
652 destroy_unused_super(s);
653 return ERR_PTR(-EBUSY);
655 if (!grab_super(old))
657 destroy_unused_super(s);
662 spin_unlock(&sb_lock);
663 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
665 return ERR_PTR(-ENOMEM);
671 spin_unlock(&sb_lock);
672 destroy_unused_super(s);
676 strscpy(s->s_id, type->name, sizeof(s->s_id));
677 list_add_tail(&s->s_list, &super_blocks);
678 hlist_add_head(&s->s_instances, &type->fs_supers);
679 spin_unlock(&sb_lock);
680 get_filesystem(type);
681 register_shrinker_prepared(&s->s_shrink);
686 void drop_super(struct super_block *sb)
688 up_read(&sb->s_umount);
692 EXPORT_SYMBOL(drop_super);
694 void drop_super_exclusive(struct super_block *sb)
696 up_write(&sb->s_umount);
699 EXPORT_SYMBOL(drop_super_exclusive);
701 static void __iterate_supers(void (*f)(struct super_block *))
703 struct super_block *sb, *p = NULL;
706 list_for_each_entry(sb, &super_blocks, s_list) {
707 if (hlist_unhashed(&sb->s_instances))
710 spin_unlock(&sb_lock);
721 spin_unlock(&sb_lock);
724 * iterate_supers - call function for all active superblocks
725 * @f: function to call
726 * @arg: argument to pass to it
728 * Scans the superblock list and calls given function, passing it
729 * locked superblock and given argument.
731 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
733 struct super_block *sb, *p = NULL;
736 list_for_each_entry(sb, &super_blocks, s_list) {
737 if (hlist_unhashed(&sb->s_instances))
740 spin_unlock(&sb_lock);
742 down_read(&sb->s_umount);
743 if (sb->s_root && (sb->s_flags & SB_BORN))
745 up_read(&sb->s_umount);
754 spin_unlock(&sb_lock);
758 * iterate_supers_type - call function for superblocks of given type
760 * @f: function to call
761 * @arg: argument to pass to it
763 * Scans the superblock list and calls given function, passing it
764 * locked superblock and given argument.
766 void iterate_supers_type(struct file_system_type *type,
767 void (*f)(struct super_block *, void *), void *arg)
769 struct super_block *sb, *p = NULL;
772 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
774 spin_unlock(&sb_lock);
776 down_read(&sb->s_umount);
777 if (sb->s_root && (sb->s_flags & SB_BORN))
779 up_read(&sb->s_umount);
788 spin_unlock(&sb_lock);
791 EXPORT_SYMBOL(iterate_supers_type);
794 * get_super - get the superblock of a device
795 * @bdev: device to get the superblock for
797 * Scans the superblock list and finds the superblock of the file system
798 * mounted on the device given. %NULL is returned if no match is found.
800 struct super_block *get_super(struct block_device *bdev)
802 struct super_block *sb;
809 list_for_each_entry(sb, &super_blocks, s_list) {
810 if (hlist_unhashed(&sb->s_instances))
812 if (sb->s_bdev == bdev) {
814 spin_unlock(&sb_lock);
815 down_read(&sb->s_umount);
817 if (sb->s_root && (sb->s_flags & SB_BORN))
819 up_read(&sb->s_umount);
820 /* nope, got unmounted */
826 spin_unlock(&sb_lock);
831 * get_active_super - get an active reference to the superblock of a device
832 * @bdev: device to get the superblock for
834 * Scans the superblock list and finds the superblock of the file system
835 * mounted on the device given. Returns the superblock with an active
836 * reference or %NULL if none was found.
838 struct super_block *get_active_super(struct block_device *bdev)
840 struct super_block *sb;
847 list_for_each_entry(sb, &super_blocks, s_list) {
848 if (hlist_unhashed(&sb->s_instances))
850 if (sb->s_bdev == bdev) {
853 up_write(&sb->s_umount);
857 spin_unlock(&sb_lock);
861 struct super_block *user_get_super(dev_t dev, bool excl)
863 struct super_block *sb;
867 list_for_each_entry(sb, &super_blocks, s_list) {
868 if (hlist_unhashed(&sb->s_instances))
870 if (sb->s_dev == dev) {
872 spin_unlock(&sb_lock);
874 down_write(&sb->s_umount);
876 down_read(&sb->s_umount);
878 if (sb->s_root && (sb->s_flags & SB_BORN))
881 up_write(&sb->s_umount);
883 up_read(&sb->s_umount);
884 /* nope, got unmounted */
890 spin_unlock(&sb_lock);
895 * reconfigure_super - asks filesystem to change superblock parameters
896 * @fc: The superblock and configuration
898 * Alters the configuration parameters of a live superblock.
900 int reconfigure_super(struct fs_context *fc)
902 struct super_block *sb = fc->root->d_sb;
904 bool remount_ro = false;
905 bool remount_rw = false;
906 bool force = fc->sb_flags & SB_FORCE;
908 if (fc->sb_flags_mask & ~MS_RMT_MASK)
910 if (sb->s_writers.frozen != SB_UNFROZEN)
913 retval = security_sb_remount(sb, fc->security);
917 if (fc->sb_flags_mask & SB_RDONLY) {
919 if (!(fc->sb_flags & SB_RDONLY) && sb->s_bdev &&
920 bdev_read_only(sb->s_bdev))
923 remount_rw = !(fc->sb_flags & SB_RDONLY) && sb_rdonly(sb);
924 remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb);
928 if (!hlist_empty(&sb->s_pins)) {
929 up_write(&sb->s_umount);
930 group_pin_kill(&sb->s_pins);
931 down_write(&sb->s_umount);
934 if (sb->s_writers.frozen != SB_UNFROZEN)
936 remount_ro = !sb_rdonly(sb);
939 shrink_dcache_sb(sb);
941 /* If we are reconfiguring to RDONLY and current sb is read/write,
942 * make sure there are no files open for writing.
946 sb_start_ro_state_change(sb);
948 retval = sb_prepare_remount_readonly(sb);
952 } else if (remount_rw) {
954 * Protect filesystem's reconfigure code from writes from
955 * userspace until reconfigure finishes.
957 sb_start_ro_state_change(sb);
960 if (fc->ops->reconfigure) {
961 retval = fc->ops->reconfigure(fc);
964 goto cancel_readonly;
965 /* If forced remount, go ahead despite any errors */
966 WARN(1, "forced remount of a %s fs returned %i\n",
967 sb->s_type->name, retval);
971 WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) |
972 (fc->sb_flags & fc->sb_flags_mask)));
973 sb_end_ro_state_change(sb);
976 * Some filesystems modify their metadata via some other path than the
977 * bdev buffer cache (eg. use a private mapping, or directories in
978 * pagecache, etc). Also file data modifications go via their own
979 * mappings. So If we try to mount readonly then copy the filesystem
980 * from bdev, we could get stale data, so invalidate it to give a best
981 * effort at coherency.
983 if (remount_ro && sb->s_bdev)
984 invalidate_bdev(sb->s_bdev);
988 sb_end_ro_state_change(sb);
992 static void do_emergency_remount_callback(struct super_block *sb)
994 down_write(&sb->s_umount);
995 if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
997 struct fs_context *fc;
999 fc = fs_context_for_reconfigure(sb->s_root,
1000 SB_RDONLY | SB_FORCE, SB_RDONLY);
1002 if (parse_monolithic_mount_data(fc, NULL) == 0)
1003 (void)reconfigure_super(fc);
1007 up_write(&sb->s_umount);
1010 static void do_emergency_remount(struct work_struct *work)
1012 __iterate_supers(do_emergency_remount_callback);
1014 printk("Emergency Remount complete\n");
1017 void emergency_remount(void)
1019 struct work_struct *work;
1021 work = kmalloc(sizeof(*work), GFP_ATOMIC);
1023 INIT_WORK(work, do_emergency_remount);
1024 schedule_work(work);
1028 static void do_thaw_all_callback(struct super_block *sb)
1030 down_write(&sb->s_umount);
1031 if (sb->s_root && sb->s_flags & SB_BORN) {
1032 emergency_thaw_bdev(sb);
1033 thaw_super_locked(sb);
1035 up_write(&sb->s_umount);
1039 static void do_thaw_all(struct work_struct *work)
1041 __iterate_supers(do_thaw_all_callback);
1043 printk(KERN_WARNING "Emergency Thaw complete\n");
1047 * emergency_thaw_all -- forcibly thaw every frozen filesystem
1049 * Used for emergency unfreeze of all filesystems via SysRq
1051 void emergency_thaw_all(void)
1053 struct work_struct *work;
1055 work = kmalloc(sizeof(*work), GFP_ATOMIC);
1057 INIT_WORK(work, do_thaw_all);
1058 schedule_work(work);
1062 static DEFINE_IDA(unnamed_dev_ida);
1065 * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1066 * @p: Pointer to a dev_t.
1068 * Filesystems which don't use real block devices can call this function
1069 * to allocate a virtual block device.
1071 * Context: Any context. Frequently called while holding sb_lock.
1072 * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1073 * or -ENOMEM if memory allocation failed.
1075 int get_anon_bdev(dev_t *p)
1080 * Many userspace utilities consider an FSID of 0 invalid.
1081 * Always return at least 1 from get_anon_bdev.
1083 dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1093 EXPORT_SYMBOL(get_anon_bdev);
1095 void free_anon_bdev(dev_t dev)
1097 ida_free(&unnamed_dev_ida, MINOR(dev));
1099 EXPORT_SYMBOL(free_anon_bdev);
1101 int set_anon_super(struct super_block *s, void *data)
1103 return get_anon_bdev(&s->s_dev);
1105 EXPORT_SYMBOL(set_anon_super);
1107 void kill_anon_super(struct super_block *sb)
1109 dev_t dev = sb->s_dev;
1110 generic_shutdown_super(sb);
1111 free_anon_bdev(dev);
1113 EXPORT_SYMBOL(kill_anon_super);
1115 void kill_litter_super(struct super_block *sb)
1118 d_genocide(sb->s_root);
1119 kill_anon_super(sb);
1121 EXPORT_SYMBOL(kill_litter_super);
1123 int set_anon_super_fc(struct super_block *sb, struct fs_context *fc)
1125 return set_anon_super(sb, NULL);
1127 EXPORT_SYMBOL(set_anon_super_fc);
1129 static int test_keyed_super(struct super_block *sb, struct fs_context *fc)
1131 return sb->s_fs_info == fc->s_fs_info;
1134 static int test_single_super(struct super_block *s, struct fs_context *fc)
1139 static int vfs_get_super(struct fs_context *fc, bool reconf,
1140 int (*test)(struct super_block *, struct fs_context *),
1141 int (*fill_super)(struct super_block *sb,
1142 struct fs_context *fc))
1144 struct super_block *sb;
1147 sb = sget_fc(fc, test, set_anon_super_fc);
1152 err = fill_super(sb, fc);
1156 sb->s_flags |= SB_ACTIVE;
1157 fc->root = dget(sb->s_root);
1159 fc->root = dget(sb->s_root);
1161 err = reconfigure_super(fc);
1173 deactivate_locked_super(sb);
1177 int get_tree_nodev(struct fs_context *fc,
1178 int (*fill_super)(struct super_block *sb,
1179 struct fs_context *fc))
1181 return vfs_get_super(fc, false, NULL, fill_super);
1183 EXPORT_SYMBOL(get_tree_nodev);
1185 int get_tree_single(struct fs_context *fc,
1186 int (*fill_super)(struct super_block *sb,
1187 struct fs_context *fc))
1189 return vfs_get_super(fc, false, test_single_super, fill_super);
1191 EXPORT_SYMBOL(get_tree_single);
1193 int get_tree_single_reconf(struct fs_context *fc,
1194 int (*fill_super)(struct super_block *sb,
1195 struct fs_context *fc))
1197 return vfs_get_super(fc, true, test_single_super, fill_super);
1199 EXPORT_SYMBOL(get_tree_single_reconf);
1201 int get_tree_keyed(struct fs_context *fc,
1202 int (*fill_super)(struct super_block *sb,
1203 struct fs_context *fc),
1206 fc->s_fs_info = key;
1207 return vfs_get_super(fc, false, test_keyed_super, fill_super);
1209 EXPORT_SYMBOL(get_tree_keyed);
1213 * Lock a super block that the callers holds a reference to.
1215 * The caller needs to ensure that the super_block isn't being freed while
1216 * calling this function, e.g. by holding a lock over the call to this function
1217 * and the place that clears the pointer to the superblock used by this function
1218 * before freeing the superblock.
1220 static bool lock_active_super(struct super_block *sb)
1222 down_read(&sb->s_umount);
1224 (sb->s_flags & (SB_ACTIVE | SB_BORN)) != (SB_ACTIVE | SB_BORN)) {
1225 up_read(&sb->s_umount);
1231 static void fs_mark_dead(struct block_device *bdev)
1233 struct super_block *sb = bdev->bd_holder;
1235 /* bd_holder_lock ensures that the sb isn't freed */
1236 lockdep_assert_held(&bdev->bd_holder_lock);
1238 if (!lock_active_super(sb))
1241 if (sb->s_op->shutdown)
1242 sb->s_op->shutdown(sb);
1244 up_read(&sb->s_umount);
1247 static const struct blk_holder_ops fs_holder_ops = {
1248 .mark_dead = fs_mark_dead,
1251 static int set_bdev_super(struct super_block *s, void *data)
1253 s->s_dev = *(dev_t *)data;
1257 static int set_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1259 return set_bdev_super(s, fc->sget_key);
1262 static int test_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1264 return !(s->s_iflags & SB_I_RETIRED) &&
1265 s->s_dev == *(dev_t *)fc->sget_key;
1268 int setup_bdev_super(struct super_block *sb, int sb_flags,
1269 struct fs_context *fc)
1271 blk_mode_t mode = sb_open_mode(sb_flags);
1272 struct block_device *bdev;
1274 bdev = blkdev_get_by_dev(sb->s_dev, mode, sb, &fs_holder_ops);
1277 errorf(fc, "%s: Can't open blockdev", fc->source);
1278 return PTR_ERR(bdev);
1282 * This really should be in blkdev_get_by_dev, but right now can't due
1283 * to legacy issues that require us to allow opening a block device node
1284 * writable from userspace even for a read-only block device.
1286 if ((mode & BLK_OPEN_WRITE) && bdev_read_only(bdev)) {
1287 blkdev_put(bdev, sb);
1292 * Until SB_BORN flag is set, there can be no active superblock
1293 * references and thus no filesystem freezing. get_active_super() will
1294 * just loop waiting for SB_BORN so even freeze_bdev() cannot proceed.
1296 * It is enough to check bdev was not frozen before we set s_bdev.
1298 mutex_lock(&bdev->bd_fsfreeze_mutex);
1299 if (bdev->bd_fsfreeze_count > 0) {
1300 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1302 warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
1303 blkdev_put(bdev, sb);
1306 spin_lock(&sb_lock);
1308 sb->s_bdi = bdi_get(bdev->bd_disk->bdi);
1309 if (bdev_stable_writes(bdev))
1310 sb->s_iflags |= SB_I_STABLE_WRITES;
1311 spin_unlock(&sb_lock);
1312 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1314 snprintf(sb->s_id, sizeof(sb->s_id), "%pg", bdev);
1315 shrinker_debugfs_rename(&sb->s_shrink, "sb-%s:%s", sb->s_type->name,
1317 sb_set_blocksize(sb, block_size(bdev));
1320 EXPORT_SYMBOL_GPL(setup_bdev_super);
1323 * get_tree_bdev - Get a superblock based on a single block device
1324 * @fc: The filesystem context holding the parameters
1325 * @fill_super: Helper to initialise a new superblock
1327 int get_tree_bdev(struct fs_context *fc,
1328 int (*fill_super)(struct super_block *,
1329 struct fs_context *))
1331 struct super_block *s;
1336 return invalf(fc, "No source specified");
1338 error = lookup_bdev(fc->source, &dev);
1340 errorf(fc, "%s: Can't lookup blockdev", fc->source);
1344 fc->sb_flags |= SB_NOSEC;
1345 fc->sget_key = &dev;
1346 s = sget_fc(fc, test_bdev_super_fc, set_bdev_super_fc);
1351 /* Don't summarily change the RO/RW state. */
1352 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1353 warnf(fc, "%pg: Can't mount, would change RO state", s->s_bdev);
1354 deactivate_locked_super(s);
1359 * We drop s_umount here because we need to open the bdev and
1360 * bdev->open_mutex ranks above s_umount (blkdev_put() ->
1361 * __invalidate_device()). It is safe because we have active sb
1362 * reference and SB_BORN is not set yet.
1364 up_write(&s->s_umount);
1365 error = setup_bdev_super(s, fc->sb_flags, fc);
1366 down_write(&s->s_umount);
1368 error = fill_super(s, fc);
1370 deactivate_locked_super(s);
1373 s->s_flags |= SB_ACTIVE;
1377 fc->root = dget(s->s_root);
1380 EXPORT_SYMBOL(get_tree_bdev);
1382 static int test_bdev_super(struct super_block *s, void *data)
1384 return !(s->s_iflags & SB_I_RETIRED) && s->s_dev == *(dev_t *)data;
1387 struct dentry *mount_bdev(struct file_system_type *fs_type,
1388 int flags, const char *dev_name, void *data,
1389 int (*fill_super)(struct super_block *, void *, int))
1391 struct super_block *s;
1395 error = lookup_bdev(dev_name, &dev);
1397 return ERR_PTR(error);
1400 s = sget(fs_type, test_bdev_super, set_bdev_super, flags, &dev);
1405 if ((flags ^ s->s_flags) & SB_RDONLY) {
1406 deactivate_locked_super(s);
1407 return ERR_PTR(-EBUSY);
1411 * We drop s_umount here because we need to open the bdev and
1412 * bdev->open_mutex ranks above s_umount (blkdev_put() ->
1413 * __invalidate_device()). It is safe because we have active sb
1414 * reference and SB_BORN is not set yet.
1416 up_write(&s->s_umount);
1417 error = setup_bdev_super(s, flags, NULL);
1418 down_write(&s->s_umount);
1420 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1422 deactivate_locked_super(s);
1423 return ERR_PTR(error);
1426 s->s_flags |= SB_ACTIVE;
1429 return dget(s->s_root);
1431 EXPORT_SYMBOL(mount_bdev);
1433 void kill_block_super(struct super_block *sb)
1435 struct block_device *bdev = sb->s_bdev;
1437 generic_shutdown_super(sb);
1439 sync_blockdev(bdev);
1440 blkdev_put(bdev, sb);
1444 EXPORT_SYMBOL(kill_block_super);
1447 struct dentry *mount_nodev(struct file_system_type *fs_type,
1448 int flags, void *data,
1449 int (*fill_super)(struct super_block *, void *, int))
1452 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1457 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1459 deactivate_locked_super(s);
1460 return ERR_PTR(error);
1462 s->s_flags |= SB_ACTIVE;
1463 return dget(s->s_root);
1465 EXPORT_SYMBOL(mount_nodev);
1467 int reconfigure_single(struct super_block *s,
1468 int flags, void *data)
1470 struct fs_context *fc;
1473 /* The caller really need to be passing fc down into mount_single(),
1474 * then a chunk of this can be removed. [Bollocks -- AV]
1475 * Better yet, reconfiguration shouldn't happen, but rather the second
1476 * mount should be rejected if the parameters are not compatible.
1478 fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK);
1482 ret = parse_monolithic_mount_data(fc, data);
1486 ret = reconfigure_super(fc);
1492 static int compare_single(struct super_block *s, void *p)
1497 struct dentry *mount_single(struct file_system_type *fs_type,
1498 int flags, void *data,
1499 int (*fill_super)(struct super_block *, void *, int))
1501 struct super_block *s;
1504 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1508 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1510 s->s_flags |= SB_ACTIVE;
1512 error = reconfigure_single(s, flags, data);
1514 if (unlikely(error)) {
1515 deactivate_locked_super(s);
1516 return ERR_PTR(error);
1518 return dget(s->s_root);
1520 EXPORT_SYMBOL(mount_single);
1523 * vfs_get_tree - Get the mountable root
1524 * @fc: The superblock configuration context.
1526 * The filesystem is invoked to get or create a superblock which can then later
1527 * be used for mounting. The filesystem places a pointer to the root to be
1528 * used for mounting in @fc->root.
1530 int vfs_get_tree(struct fs_context *fc)
1532 struct super_block *sb;
1538 /* Get the mountable root in fc->root, with a ref on the root and a ref
1539 * on the superblock.
1541 error = fc->ops->get_tree(fc);
1546 pr_err("Filesystem %s get_tree() didn't set fc->root\n",
1548 /* We don't know what the locking state of the superblock is -
1549 * if there is a superblock.
1554 sb = fc->root->d_sb;
1555 WARN_ON(!sb->s_bdi);
1558 * Write barrier is for super_cache_count(). We place it before setting
1559 * SB_BORN as the data dependency between the two functions is the
1560 * superblock structure contents that we just set up, not the SB_BORN
1564 sb->s_flags |= SB_BORN;
1566 error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
1567 if (unlikely(error)) {
1573 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1574 * but s_maxbytes was an unsigned long long for many releases. Throw
1575 * this warning for a little while to try and catch filesystems that
1576 * violate this rule.
1578 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1579 "negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes);
1583 EXPORT_SYMBOL(vfs_get_tree);
1586 * Setup private BDI for given superblock. It gets automatically cleaned up
1587 * in generic_shutdown_super().
1589 int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1591 struct backing_dev_info *bdi;
1595 bdi = bdi_alloc(NUMA_NO_NODE);
1599 va_start(args, fmt);
1600 err = bdi_register_va(bdi, fmt, args);
1606 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1608 sb->s_iflags |= SB_I_PERSB_BDI;
1612 EXPORT_SYMBOL(super_setup_bdi_name);
1615 * Setup private BDI for given superblock. I gets automatically cleaned up
1616 * in generic_shutdown_super().
1618 int super_setup_bdi(struct super_block *sb)
1620 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1622 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1623 atomic_long_inc_return(&bdi_seq));
1625 EXPORT_SYMBOL(super_setup_bdi);
1628 * sb_wait_write - wait until all writers to given file system finish
1629 * @sb: the super for which we wait
1630 * @level: type of writers we wait for (normal vs page fault)
1632 * This function waits until there are no writers of given type to given file
1635 static void sb_wait_write(struct super_block *sb, int level)
1637 percpu_down_write(sb->s_writers.rw_sem + level-1);
1641 * We are going to return to userspace and forget about these locks, the
1642 * ownership goes to the caller of thaw_super() which does unlock().
1644 static void lockdep_sb_freeze_release(struct super_block *sb)
1648 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1649 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1653 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1655 static void lockdep_sb_freeze_acquire(struct super_block *sb)
1659 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1660 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1663 static void sb_freeze_unlock(struct super_block *sb, int level)
1665 for (level--; level >= 0; level--)
1666 percpu_up_write(sb->s_writers.rw_sem + level);
1670 * freeze_super - lock the filesystem and force it into a consistent state
1671 * @sb: the super to lock
1673 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1674 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1677 * During this function, sb->s_writers.frozen goes through these values:
1679 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1681 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1682 * writes should be blocked, though page faults are still allowed. We wait for
1683 * all writes to complete and then proceed to the next stage.
1685 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1686 * but internal fs threads can still modify the filesystem (although they
1687 * should not dirty new pages or inodes), writeback can run etc. After waiting
1688 * for all running page faults we sync the filesystem which will clean all
1689 * dirty pages and inodes (no new dirty pages or inodes can be created when
1692 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1693 * modification are blocked (e.g. XFS preallocation truncation on inode
1694 * reclaim). This is usually implemented by blocking new transactions for
1695 * filesystems that have them and need this additional guard. After all
1696 * internal writers are finished we call ->freeze_fs() to finish filesystem
1697 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1698 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1700 * sb->s_writers.frozen is protected by sb->s_umount.
1702 int freeze_super(struct super_block *sb)
1706 atomic_inc(&sb->s_active);
1707 down_write(&sb->s_umount);
1708 if (sb->s_writers.frozen != SB_UNFROZEN) {
1709 deactivate_locked_super(sb);
1713 if (!(sb->s_flags & SB_BORN)) {
1714 up_write(&sb->s_umount);
1715 return 0; /* sic - it's "nothing to do" */
1718 if (sb_rdonly(sb)) {
1719 /* Nothing to do really... */
1720 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1721 up_write(&sb->s_umount);
1725 sb->s_writers.frozen = SB_FREEZE_WRITE;
1726 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1727 up_write(&sb->s_umount);
1728 sb_wait_write(sb, SB_FREEZE_WRITE);
1729 down_write(&sb->s_umount);
1731 /* Now we go and block page faults... */
1732 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1733 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1735 /* All writers are done so after syncing there won't be dirty data */
1736 ret = sync_filesystem(sb);
1738 sb->s_writers.frozen = SB_UNFROZEN;
1739 sb_freeze_unlock(sb, SB_FREEZE_PAGEFAULT);
1740 deactivate_locked_super(sb);
1744 /* Now wait for internal filesystem counter */
1745 sb->s_writers.frozen = SB_FREEZE_FS;
1746 sb_wait_write(sb, SB_FREEZE_FS);
1748 if (sb->s_op->freeze_fs) {
1749 ret = sb->s_op->freeze_fs(sb);
1752 "VFS:Filesystem freeze failed\n");
1753 sb->s_writers.frozen = SB_UNFROZEN;
1754 sb_freeze_unlock(sb, SB_FREEZE_FS);
1755 deactivate_locked_super(sb);
1760 * For debugging purposes so that fs can warn if it sees write activity
1761 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1763 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1764 lockdep_sb_freeze_release(sb);
1765 up_write(&sb->s_umount);
1768 EXPORT_SYMBOL(freeze_super);
1770 static int thaw_super_locked(struct super_block *sb)
1774 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1775 up_write(&sb->s_umount);
1779 if (sb_rdonly(sb)) {
1780 sb->s_writers.frozen = SB_UNFROZEN;
1784 lockdep_sb_freeze_acquire(sb);
1786 if (sb->s_op->unfreeze_fs) {
1787 error = sb->s_op->unfreeze_fs(sb);
1790 "VFS:Filesystem thaw failed\n");
1791 lockdep_sb_freeze_release(sb);
1792 up_write(&sb->s_umount);
1797 sb->s_writers.frozen = SB_UNFROZEN;
1798 sb_freeze_unlock(sb, SB_FREEZE_FS);
1800 deactivate_locked_super(sb);
1805 * thaw_super -- unlock filesystem
1806 * @sb: the super to thaw
1808 * Unlocks the filesystem and marks it writeable again after freeze_super().
1810 int thaw_super(struct super_block *sb)
1812 down_write(&sb->s_umount);
1813 return thaw_super_locked(sb);
1815 EXPORT_SYMBOL(thaw_super);
1818 * Create workqueue for deferred direct IO completions. We allocate the
1819 * workqueue when it's first needed. This avoids creating workqueue for
1820 * filesystems that don't need it and also allows us to create the workqueue
1821 * late enough so the we can include s_id in the name of the workqueue.
1823 int sb_init_dio_done_wq(struct super_block *sb)
1825 struct workqueue_struct *old;
1826 struct workqueue_struct *wq = alloc_workqueue("dio/%s",
1832 * This has to be atomic as more DIOs can race to create the workqueue
1834 old = cmpxchg(&sb->s_dio_done_wq, NULL, wq);
1835 /* Someone created workqueue before us? Free ours... */
1837 destroy_workqueue(wq);