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] = {
53 static inline void super_lock(struct super_block *sb, bool excl)
56 down_write(&sb->s_umount);
58 down_read(&sb->s_umount);
61 static inline void super_unlock(struct super_block *sb, bool excl)
64 up_write(&sb->s_umount);
66 up_read(&sb->s_umount);
69 static inline void super_lock_excl(struct super_block *sb)
74 static inline void super_lock_shared(struct super_block *sb)
76 super_lock(sb, false);
79 static inline void super_unlock_excl(struct super_block *sb)
81 super_unlock(sb, true);
84 static inline void super_unlock_shared(struct super_block *sb)
86 super_unlock(sb, false);
90 * One thing we have to be careful of with a per-sb shrinker is that we don't
91 * drop the last active reference to the superblock from within the shrinker.
92 * If that happens we could trigger unregistering the shrinker from within the
93 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
94 * take a passive reference to the superblock to avoid this from occurring.
96 static unsigned long super_cache_scan(struct shrinker *shrink,
97 struct shrink_control *sc)
99 struct super_block *sb;
106 sb = container_of(shrink, struct super_block, s_shrink);
109 * Deadlock avoidance. We may hold various FS locks, and we don't want
110 * to recurse into the FS that called us in clear_inode() and friends..
112 if (!(sc->gfp_mask & __GFP_FS))
115 if (!super_trylock_shared(sb))
118 if (sb->s_op->nr_cached_objects)
119 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
121 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
122 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
123 total_objects = dentries + inodes + fs_objects + 1;
127 /* proportion the scan between the caches */
128 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
129 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
130 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
133 * prune the dcache first as the icache is pinned by it, then
134 * prune the icache, followed by the filesystem specific caches
136 * Ensure that we always scan at least one object - memcg kmem
137 * accounting uses this to fully empty the caches.
139 sc->nr_to_scan = dentries + 1;
140 freed = prune_dcache_sb(sb, sc);
141 sc->nr_to_scan = inodes + 1;
142 freed += prune_icache_sb(sb, sc);
145 sc->nr_to_scan = fs_objects + 1;
146 freed += sb->s_op->free_cached_objects(sb, sc);
149 super_unlock_shared(sb);
153 static unsigned long super_cache_count(struct shrinker *shrink,
154 struct shrink_control *sc)
156 struct super_block *sb;
157 long total_objects = 0;
159 sb = container_of(shrink, struct super_block, s_shrink);
162 * We don't call super_trylock_shared() here as it is a scalability
163 * bottleneck, so we're exposed to partial setup state. The shrinker
164 * rwsem does not protect filesystem operations backing
165 * list_lru_shrink_count() or s_op->nr_cached_objects(). Counts can
166 * change between super_cache_count and super_cache_scan, so we really
167 * don't need locks here.
169 * However, if we are currently mounting the superblock, the underlying
170 * filesystem might be in a state of partial construction and hence it
171 * is dangerous to access it. super_trylock_shared() uses a SB_BORN check
172 * to avoid this situation, so do the same here. The memory barrier is
173 * matched with the one in mount_fs() as we don't hold locks here.
175 if (!(sb->s_flags & SB_BORN))
179 if (sb->s_op && sb->s_op->nr_cached_objects)
180 total_objects = sb->s_op->nr_cached_objects(sb, sc);
182 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
183 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
188 total_objects = vfs_pressure_ratio(total_objects);
189 return total_objects;
192 static void destroy_super_work(struct work_struct *work)
194 struct super_block *s = container_of(work, struct super_block,
198 for (i = 0; i < SB_FREEZE_LEVELS; i++)
199 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
203 static void destroy_super_rcu(struct rcu_head *head)
205 struct super_block *s = container_of(head, struct super_block, rcu);
206 INIT_WORK(&s->destroy_work, destroy_super_work);
207 schedule_work(&s->destroy_work);
210 /* Free a superblock that has never been seen by anyone */
211 static void destroy_unused_super(struct super_block *s)
215 super_unlock_excl(s);
216 list_lru_destroy(&s->s_dentry_lru);
217 list_lru_destroy(&s->s_inode_lru);
219 put_user_ns(s->s_user_ns);
221 free_prealloced_shrinker(&s->s_shrink);
222 /* no delays needed */
223 destroy_super_work(&s->destroy_work);
227 * alloc_super - create new superblock
228 * @type: filesystem type superblock should belong to
229 * @flags: the mount flags
230 * @user_ns: User namespace for the super_block
232 * Allocates and initializes a new &struct super_block. alloc_super()
233 * returns a pointer new superblock or %NULL if allocation had failed.
235 static struct super_block *alloc_super(struct file_system_type *type, int flags,
236 struct user_namespace *user_ns)
238 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
239 static const struct super_operations default_op;
245 INIT_LIST_HEAD(&s->s_mounts);
246 s->s_user_ns = get_user_ns(user_ns);
247 init_rwsem(&s->s_umount);
248 lockdep_set_class(&s->s_umount, &type->s_umount_key);
250 * sget() can have s_umount recursion.
252 * When it cannot find a suitable sb, it allocates a new
253 * one (this one), and tries again to find a suitable old
256 * In case that succeeds, it will acquire the s_umount
257 * lock of the old one. Since these are clearly distrinct
258 * locks, and this object isn't exposed yet, there's no
261 * Annotate this by putting this lock in a different
264 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
266 if (security_sb_alloc(s))
269 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
270 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
272 &type->s_writers_key[i]))
275 s->s_bdi = &noop_backing_dev_info;
277 if (s->s_user_ns != &init_user_ns)
278 s->s_iflags |= SB_I_NODEV;
279 INIT_HLIST_NODE(&s->s_instances);
280 INIT_HLIST_BL_HEAD(&s->s_roots);
281 mutex_init(&s->s_sync_lock);
282 INIT_LIST_HEAD(&s->s_inodes);
283 spin_lock_init(&s->s_inode_list_lock);
284 INIT_LIST_HEAD(&s->s_inodes_wb);
285 spin_lock_init(&s->s_inode_wblist_lock);
288 atomic_set(&s->s_active, 1);
289 mutex_init(&s->s_vfs_rename_mutex);
290 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
291 init_rwsem(&s->s_dquot.dqio_sem);
292 s->s_maxbytes = MAX_NON_LFS;
293 s->s_op = &default_op;
294 s->s_time_gran = 1000000000;
295 s->s_time_min = TIME64_MIN;
296 s->s_time_max = TIME64_MAX;
298 s->s_shrink.seeks = DEFAULT_SEEKS;
299 s->s_shrink.scan_objects = super_cache_scan;
300 s->s_shrink.count_objects = super_cache_count;
301 s->s_shrink.batch = 1024;
302 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
303 if (prealloc_shrinker(&s->s_shrink, "sb-%s", type->name))
305 if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink))
307 if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink))
312 destroy_unused_super(s);
316 /* Superblock refcounting */
319 * Drop a superblock's refcount. The caller must hold sb_lock.
321 static void __put_super(struct super_block *s)
324 list_del_init(&s->s_list);
325 WARN_ON(s->s_dentry_lru.node);
326 WARN_ON(s->s_inode_lru.node);
327 WARN_ON(!list_empty(&s->s_mounts));
329 put_user_ns(s->s_user_ns);
331 call_rcu(&s->rcu, destroy_super_rcu);
336 * put_super - drop a temporary reference to superblock
337 * @sb: superblock in question
339 * Drops a temporary reference, frees superblock if there's no
342 void put_super(struct super_block *sb)
346 spin_unlock(&sb_lock);
351 * deactivate_locked_super - drop an active reference to superblock
352 * @s: superblock to deactivate
354 * Drops an active reference to superblock, converting it into a temporary
355 * one if there is no other active references left. In that case we
356 * tell fs driver to shut it down and drop the temporary reference we
359 * Caller holds exclusive lock on superblock; that lock is released.
361 void deactivate_locked_super(struct super_block *s)
363 struct file_system_type *fs = s->s_type;
364 if (atomic_dec_and_test(&s->s_active)) {
365 unregister_shrinker(&s->s_shrink);
369 * Since list_lru_destroy() may sleep, we cannot call it from
370 * put_super(), where we hold the sb_lock. Therefore we destroy
371 * the lru lists right now.
373 list_lru_destroy(&s->s_dentry_lru);
374 list_lru_destroy(&s->s_inode_lru);
379 super_unlock_excl(s);
383 EXPORT_SYMBOL(deactivate_locked_super);
386 * deactivate_super - drop an active reference to superblock
387 * @s: superblock to deactivate
389 * Variant of deactivate_locked_super(), except that superblock is *not*
390 * locked by caller. If we are going to drop the final active reference,
391 * lock will be acquired prior to that.
393 void deactivate_super(struct super_block *s)
395 if (!atomic_add_unless(&s->s_active, -1, 1)) {
397 deactivate_locked_super(s);
401 EXPORT_SYMBOL(deactivate_super);
404 * grab_super - acquire an active reference
405 * @s: reference we are trying to make active
407 * Tries to acquire an active reference. grab_super() is used when we
408 * had just found a superblock in super_blocks or fs_type->fs_supers
409 * and want to turn it into a full-blown active reference. grab_super()
410 * is called with sb_lock held and drops it. Returns 1 in case of
411 * success, 0 if we had failed (superblock contents was already dead or
412 * dying when grab_super() had been called). Note that this is only
413 * called for superblocks not in rundown mode (== ones still on ->fs_supers
414 * of their type), so increment of ->s_count is OK here.
416 static int grab_super(struct super_block *s) __releases(sb_lock)
419 spin_unlock(&sb_lock);
421 if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
425 super_unlock_excl(s);
431 * super_trylock_shared - try to grab ->s_umount shared
432 * @sb: reference we are trying to grab
434 * Try to prevent fs shutdown. This is used in places where we
435 * cannot take an active reference but we need to ensure that the
436 * filesystem is not shut down while we are working on it. It returns
437 * false if we cannot acquire s_umount or if we lose the race and
438 * filesystem already got into shutdown, and returns true with the s_umount
439 * lock held in read mode in case of success. On successful return,
440 * the caller must drop the s_umount lock when done.
442 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
443 * The reason why it's safe is that we are OK with doing trylock instead
444 * of down_read(). There's a couple of places that are OK with that, but
445 * it's very much not a general-purpose interface.
447 bool super_trylock_shared(struct super_block *sb)
449 if (down_read_trylock(&sb->s_umount)) {
450 if (!hlist_unhashed(&sb->s_instances) &&
451 sb->s_root && (sb->s_flags & SB_BORN))
453 super_unlock_shared(sb);
460 * retire_super - prevents superblock from being reused
461 * @sb: superblock to retire
463 * The function marks superblock to be ignored in superblock test, which
464 * prevents it from being reused for any new mounts. If the superblock has
465 * a private bdi, it also unregisters it, but doesn't reduce the refcount
466 * of the superblock to prevent potential races. The refcount is reduced
467 * by generic_shutdown_super(). The function can not be called
468 * concurrently with generic_shutdown_super(). It is safe to call the
469 * function multiple times, subsequent calls have no effect.
471 * The marker will affect the re-use only for block-device-based
472 * superblocks. Other superblocks will still get marked if this function
473 * is used, but that will not affect their reusability.
475 void retire_super(struct super_block *sb)
477 WARN_ON(!sb->s_bdev);
479 if (sb->s_iflags & SB_I_PERSB_BDI) {
480 bdi_unregister(sb->s_bdi);
481 sb->s_iflags &= ~SB_I_PERSB_BDI;
483 sb->s_iflags |= SB_I_RETIRED;
484 super_unlock_excl(sb);
486 EXPORT_SYMBOL(retire_super);
489 * generic_shutdown_super - common helper for ->kill_sb()
490 * @sb: superblock to kill
492 * generic_shutdown_super() does all fs-independent work on superblock
493 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
494 * that need destruction out of superblock, call generic_shutdown_super()
495 * and release aforementioned objects. Note: dentries and inodes _are_
496 * taken care of and do not need specific handling.
498 * Upon calling this function, the filesystem may no longer alter or
499 * rearrange the set of dentries belonging to this super_block, nor may it
500 * change the attachments of dentries to inodes.
502 void generic_shutdown_super(struct super_block *sb)
504 const struct super_operations *sop = sb->s_op;
507 shrink_dcache_for_umount(sb);
509 sb->s_flags &= ~SB_ACTIVE;
511 cgroup_writeback_umount();
513 /* Evict all inodes with zero refcount. */
517 * Clean up and evict any inodes that still have references due
518 * to fsnotify or the security policy.
520 fsnotify_sb_delete(sb);
521 security_sb_delete(sb);
524 * Now that all potentially-encrypted inodes have been evicted,
525 * the fscrypt keyring can be destroyed.
527 fscrypt_destroy_keyring(sb);
529 if (sb->s_dio_done_wq) {
530 destroy_workqueue(sb->s_dio_done_wq);
531 sb->s_dio_done_wq = NULL;
537 if (CHECK_DATA_CORRUPTION(!list_empty(&sb->s_inodes),
538 "VFS: Busy inodes after unmount of %s (%s)",
539 sb->s_id, sb->s_type->name)) {
541 * Adding a proper bailout path here would be hard, but
542 * we can at least make it more likely that a later
543 * iput_final() or such crashes cleanly.
547 spin_lock(&sb->s_inode_list_lock);
548 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
549 inode->i_op = VFS_PTR_POISON;
550 inode->i_sb = VFS_PTR_POISON;
551 inode->i_mapping = VFS_PTR_POISON;
553 spin_unlock(&sb->s_inode_list_lock);
557 /* should be initialized for __put_super_and_need_restart() */
558 hlist_del_init(&sb->s_instances);
559 spin_unlock(&sb_lock);
560 super_unlock_excl(sb);
561 if (sb->s_bdi != &noop_backing_dev_info) {
562 if (sb->s_iflags & SB_I_PERSB_BDI)
563 bdi_unregister(sb->s_bdi);
565 sb->s_bdi = &noop_backing_dev_info;
569 EXPORT_SYMBOL(generic_shutdown_super);
571 bool mount_capable(struct fs_context *fc)
573 if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT))
574 return capable(CAP_SYS_ADMIN);
576 return ns_capable(fc->user_ns, CAP_SYS_ADMIN);
580 * sget_fc - Find or create a superblock
581 * @fc: Filesystem context.
582 * @test: Comparison callback
583 * @set: Setup callback
585 * Find or create a superblock using the parameters stored in the filesystem
586 * context and the two callback functions.
588 * If an extant superblock is matched, then that will be returned with an
589 * elevated reference count that the caller must transfer or discard.
591 * If no match is made, a new superblock will be allocated and basic
592 * initialisation will be performed (s_type, s_fs_info and s_id will be set and
593 * the set() callback will be invoked), the superblock will be published and it
594 * will be returned in a partially constructed state with SB_BORN and SB_ACTIVE
597 struct super_block *sget_fc(struct fs_context *fc,
598 int (*test)(struct super_block *, struct fs_context *),
599 int (*set)(struct super_block *, struct fs_context *))
601 struct super_block *s = NULL;
602 struct super_block *old;
603 struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns;
609 hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) {
611 goto share_extant_sb;
615 spin_unlock(&sb_lock);
616 s = alloc_super(fc->fs_type, fc->sb_flags, user_ns);
618 return ERR_PTR(-ENOMEM);
622 s->s_fs_info = fc->s_fs_info;
626 spin_unlock(&sb_lock);
627 destroy_unused_super(s);
630 fc->s_fs_info = NULL;
631 s->s_type = fc->fs_type;
632 s->s_iflags |= fc->s_iflags;
633 strscpy(s->s_id, s->s_type->name, sizeof(s->s_id));
634 list_add_tail(&s->s_list, &super_blocks);
635 hlist_add_head(&s->s_instances, &s->s_type->fs_supers);
636 spin_unlock(&sb_lock);
637 get_filesystem(s->s_type);
638 register_shrinker_prepared(&s->s_shrink);
642 if (user_ns != old->s_user_ns) {
643 spin_unlock(&sb_lock);
644 destroy_unused_super(s);
645 return ERR_PTR(-EBUSY);
647 if (!grab_super(old))
649 destroy_unused_super(s);
652 EXPORT_SYMBOL(sget_fc);
655 * sget - find or create a superblock
656 * @type: filesystem type superblock should belong to
657 * @test: comparison callback
658 * @set: setup callback
659 * @flags: mount flags
660 * @data: argument to each of them
662 struct super_block *sget(struct file_system_type *type,
663 int (*test)(struct super_block *,void *),
664 int (*set)(struct super_block *,void *),
668 struct user_namespace *user_ns = current_user_ns();
669 struct super_block *s = NULL;
670 struct super_block *old;
673 /* We don't yet pass the user namespace of the parent
674 * mount through to here so always use &init_user_ns
675 * until that changes.
677 if (flags & SB_SUBMOUNT)
678 user_ns = &init_user_ns;
683 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
684 if (!test(old, data))
686 if (user_ns != old->s_user_ns) {
687 spin_unlock(&sb_lock);
688 destroy_unused_super(s);
689 return ERR_PTR(-EBUSY);
691 if (!grab_super(old))
693 destroy_unused_super(s);
698 spin_unlock(&sb_lock);
699 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
701 return ERR_PTR(-ENOMEM);
707 spin_unlock(&sb_lock);
708 destroy_unused_super(s);
712 strscpy(s->s_id, type->name, sizeof(s->s_id));
713 list_add_tail(&s->s_list, &super_blocks);
714 hlist_add_head(&s->s_instances, &type->fs_supers);
715 spin_unlock(&sb_lock);
716 get_filesystem(type);
717 register_shrinker_prepared(&s->s_shrink);
722 void drop_super(struct super_block *sb)
724 super_unlock_shared(sb);
728 EXPORT_SYMBOL(drop_super);
730 void drop_super_exclusive(struct super_block *sb)
732 super_unlock_excl(sb);
735 EXPORT_SYMBOL(drop_super_exclusive);
737 static void __iterate_supers(void (*f)(struct super_block *))
739 struct super_block *sb, *p = NULL;
742 list_for_each_entry(sb, &super_blocks, s_list) {
743 if (hlist_unhashed(&sb->s_instances))
746 spin_unlock(&sb_lock);
757 spin_unlock(&sb_lock);
760 * iterate_supers - call function for all active superblocks
761 * @f: function to call
762 * @arg: argument to pass to it
764 * Scans the superblock list and calls given function, passing it
765 * locked superblock and given argument.
767 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
769 struct super_block *sb, *p = NULL;
772 list_for_each_entry(sb, &super_blocks, s_list) {
773 if (hlist_unhashed(&sb->s_instances))
776 spin_unlock(&sb_lock);
778 super_lock_shared(sb);
779 if (sb->s_root && (sb->s_flags & SB_BORN))
781 super_unlock_shared(sb);
790 spin_unlock(&sb_lock);
794 * iterate_supers_type - call function for superblocks of given type
796 * @f: function to call
797 * @arg: argument to pass to it
799 * Scans the superblock list and calls given function, passing it
800 * locked superblock and given argument.
802 void iterate_supers_type(struct file_system_type *type,
803 void (*f)(struct super_block *, void *), void *arg)
805 struct super_block *sb, *p = NULL;
808 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
810 spin_unlock(&sb_lock);
812 super_lock_shared(sb);
813 if (sb->s_root && (sb->s_flags & SB_BORN))
815 super_unlock_shared(sb);
824 spin_unlock(&sb_lock);
827 EXPORT_SYMBOL(iterate_supers_type);
830 * get_active_super - get an active reference to the superblock of a device
831 * @bdev: device to get the superblock for
833 * Scans the superblock list and finds the superblock of the file system
834 * mounted on the device given. Returns the superblock with an active
835 * reference or %NULL if none was found.
837 struct super_block *get_active_super(struct block_device *bdev)
839 struct super_block *sb;
846 list_for_each_entry(sb, &super_blocks, s_list) {
847 if (hlist_unhashed(&sb->s_instances))
849 if (sb->s_bdev == bdev) {
852 super_unlock_excl(sb);
856 spin_unlock(&sb_lock);
860 struct super_block *user_get_super(dev_t dev, bool excl)
862 struct super_block *sb;
866 list_for_each_entry(sb, &super_blocks, s_list) {
867 if (hlist_unhashed(&sb->s_instances))
869 if (sb->s_dev == dev) {
871 spin_unlock(&sb_lock);
872 super_lock(sb, excl);
874 if (sb->s_root && (sb->s_flags & SB_BORN))
876 super_unlock(sb, excl);
877 /* nope, got unmounted */
883 spin_unlock(&sb_lock);
888 * reconfigure_super - asks filesystem to change superblock parameters
889 * @fc: The superblock and configuration
891 * Alters the configuration parameters of a live superblock.
893 int reconfigure_super(struct fs_context *fc)
895 struct super_block *sb = fc->root->d_sb;
897 bool remount_ro = false;
898 bool remount_rw = false;
899 bool force = fc->sb_flags & SB_FORCE;
901 if (fc->sb_flags_mask & ~MS_RMT_MASK)
903 if (sb->s_writers.frozen != SB_UNFROZEN)
906 retval = security_sb_remount(sb, fc->security);
910 if (fc->sb_flags_mask & SB_RDONLY) {
912 if (!(fc->sb_flags & SB_RDONLY) && sb->s_bdev &&
913 bdev_read_only(sb->s_bdev))
916 remount_rw = !(fc->sb_flags & SB_RDONLY) && sb_rdonly(sb);
917 remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb);
921 if (!hlist_empty(&sb->s_pins)) {
922 super_unlock_excl(sb);
923 group_pin_kill(&sb->s_pins);
927 if (sb->s_writers.frozen != SB_UNFROZEN)
929 remount_ro = !sb_rdonly(sb);
932 shrink_dcache_sb(sb);
934 /* If we are reconfiguring to RDONLY and current sb is read/write,
935 * make sure there are no files open for writing.
939 sb_start_ro_state_change(sb);
941 retval = sb_prepare_remount_readonly(sb);
945 } else if (remount_rw) {
947 * Protect filesystem's reconfigure code from writes from
948 * userspace until reconfigure finishes.
950 sb_start_ro_state_change(sb);
953 if (fc->ops->reconfigure) {
954 retval = fc->ops->reconfigure(fc);
957 goto cancel_readonly;
958 /* If forced remount, go ahead despite any errors */
959 WARN(1, "forced remount of a %s fs returned %i\n",
960 sb->s_type->name, retval);
964 WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) |
965 (fc->sb_flags & fc->sb_flags_mask)));
966 sb_end_ro_state_change(sb);
969 * Some filesystems modify their metadata via some other path than the
970 * bdev buffer cache (eg. use a private mapping, or directories in
971 * pagecache, etc). Also file data modifications go via their own
972 * mappings. So If we try to mount readonly then copy the filesystem
973 * from bdev, we could get stale data, so invalidate it to give a best
974 * effort at coherency.
976 if (remount_ro && sb->s_bdev)
977 invalidate_bdev(sb->s_bdev);
981 sb_end_ro_state_change(sb);
985 static void do_emergency_remount_callback(struct super_block *sb)
988 if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
990 struct fs_context *fc;
992 fc = fs_context_for_reconfigure(sb->s_root,
993 SB_RDONLY | SB_FORCE, SB_RDONLY);
995 if (parse_monolithic_mount_data(fc, NULL) == 0)
996 (void)reconfigure_super(fc);
1000 super_unlock_excl(sb);
1003 static void do_emergency_remount(struct work_struct *work)
1005 __iterate_supers(do_emergency_remount_callback);
1007 printk("Emergency Remount complete\n");
1010 void emergency_remount(void)
1012 struct work_struct *work;
1014 work = kmalloc(sizeof(*work), GFP_ATOMIC);
1016 INIT_WORK(work, do_emergency_remount);
1017 schedule_work(work);
1021 static void do_thaw_all_callback(struct super_block *sb)
1023 super_lock_excl(sb);
1024 if (sb->s_root && sb->s_flags & SB_BORN) {
1025 emergency_thaw_bdev(sb);
1026 thaw_super_locked(sb);
1028 super_unlock_excl(sb);
1032 static void do_thaw_all(struct work_struct *work)
1034 __iterate_supers(do_thaw_all_callback);
1036 printk(KERN_WARNING "Emergency Thaw complete\n");
1040 * emergency_thaw_all -- forcibly thaw every frozen filesystem
1042 * Used for emergency unfreeze of all filesystems via SysRq
1044 void emergency_thaw_all(void)
1046 struct work_struct *work;
1048 work = kmalloc(sizeof(*work), GFP_ATOMIC);
1050 INIT_WORK(work, do_thaw_all);
1051 schedule_work(work);
1055 static DEFINE_IDA(unnamed_dev_ida);
1058 * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1059 * @p: Pointer to a dev_t.
1061 * Filesystems which don't use real block devices can call this function
1062 * to allocate a virtual block device.
1064 * Context: Any context. Frequently called while holding sb_lock.
1065 * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1066 * or -ENOMEM if memory allocation failed.
1068 int get_anon_bdev(dev_t *p)
1073 * Many userspace utilities consider an FSID of 0 invalid.
1074 * Always return at least 1 from get_anon_bdev.
1076 dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1086 EXPORT_SYMBOL(get_anon_bdev);
1088 void free_anon_bdev(dev_t dev)
1090 ida_free(&unnamed_dev_ida, MINOR(dev));
1092 EXPORT_SYMBOL(free_anon_bdev);
1094 int set_anon_super(struct super_block *s, void *data)
1096 return get_anon_bdev(&s->s_dev);
1098 EXPORT_SYMBOL(set_anon_super);
1100 void kill_anon_super(struct super_block *sb)
1102 dev_t dev = sb->s_dev;
1103 generic_shutdown_super(sb);
1104 free_anon_bdev(dev);
1106 EXPORT_SYMBOL(kill_anon_super);
1108 void kill_litter_super(struct super_block *sb)
1111 d_genocide(sb->s_root);
1112 kill_anon_super(sb);
1114 EXPORT_SYMBOL(kill_litter_super);
1116 int set_anon_super_fc(struct super_block *sb, struct fs_context *fc)
1118 return set_anon_super(sb, NULL);
1120 EXPORT_SYMBOL(set_anon_super_fc);
1122 static int test_keyed_super(struct super_block *sb, struct fs_context *fc)
1124 return sb->s_fs_info == fc->s_fs_info;
1127 static int test_single_super(struct super_block *s, struct fs_context *fc)
1132 static int vfs_get_super(struct fs_context *fc, bool reconf,
1133 int (*test)(struct super_block *, struct fs_context *),
1134 int (*fill_super)(struct super_block *sb,
1135 struct fs_context *fc))
1137 struct super_block *sb;
1140 sb = sget_fc(fc, test, set_anon_super_fc);
1145 err = fill_super(sb, fc);
1149 sb->s_flags |= SB_ACTIVE;
1150 fc->root = dget(sb->s_root);
1152 fc->root = dget(sb->s_root);
1154 err = reconfigure_super(fc);
1166 deactivate_locked_super(sb);
1170 int get_tree_nodev(struct fs_context *fc,
1171 int (*fill_super)(struct super_block *sb,
1172 struct fs_context *fc))
1174 return vfs_get_super(fc, false, NULL, fill_super);
1176 EXPORT_SYMBOL(get_tree_nodev);
1178 int get_tree_single(struct fs_context *fc,
1179 int (*fill_super)(struct super_block *sb,
1180 struct fs_context *fc))
1182 return vfs_get_super(fc, false, test_single_super, fill_super);
1184 EXPORT_SYMBOL(get_tree_single);
1186 int get_tree_single_reconf(struct fs_context *fc,
1187 int (*fill_super)(struct super_block *sb,
1188 struct fs_context *fc))
1190 return vfs_get_super(fc, true, test_single_super, fill_super);
1192 EXPORT_SYMBOL(get_tree_single_reconf);
1194 int get_tree_keyed(struct fs_context *fc,
1195 int (*fill_super)(struct super_block *sb,
1196 struct fs_context *fc),
1199 fc->s_fs_info = key;
1200 return vfs_get_super(fc, false, test_keyed_super, fill_super);
1202 EXPORT_SYMBOL(get_tree_keyed);
1206 * Lock a super block that the callers holds a reference to.
1208 * The caller needs to ensure that the super_block isn't being freed while
1209 * calling this function, e.g. by holding a lock over the call to this function
1210 * and the place that clears the pointer to the superblock used by this function
1211 * before freeing the superblock.
1213 static bool super_lock_shared_active(struct super_block *sb)
1215 super_lock_shared(sb);
1217 (sb->s_flags & (SB_ACTIVE | SB_BORN)) != (SB_ACTIVE | SB_BORN)) {
1218 super_unlock_shared(sb);
1224 static void fs_bdev_mark_dead(struct block_device *bdev, bool surprise)
1226 struct super_block *sb = bdev->bd_holder;
1228 /* bd_holder_lock ensures that the sb isn't freed */
1229 lockdep_assert_held(&bdev->bd_holder_lock);
1231 if (!super_lock_shared_active(sb))
1235 sync_filesystem(sb);
1236 shrink_dcache_sb(sb);
1237 invalidate_inodes(sb);
1238 if (sb->s_op->shutdown)
1239 sb->s_op->shutdown(sb);
1241 super_unlock_shared(sb);
1244 static void fs_bdev_sync(struct block_device *bdev)
1246 struct super_block *sb = bdev->bd_holder;
1248 lockdep_assert_held(&bdev->bd_holder_lock);
1250 if (!super_lock_shared_active(sb))
1252 sync_filesystem(sb);
1253 super_unlock_shared(sb);
1256 const struct blk_holder_ops fs_holder_ops = {
1257 .mark_dead = fs_bdev_mark_dead,
1258 .sync = fs_bdev_sync,
1260 EXPORT_SYMBOL_GPL(fs_holder_ops);
1262 static int set_bdev_super(struct super_block *s, void *data)
1264 s->s_dev = *(dev_t *)data;
1268 static int set_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1270 return set_bdev_super(s, fc->sget_key);
1273 static int test_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1275 return !(s->s_iflags & SB_I_RETIRED) &&
1276 s->s_dev == *(dev_t *)fc->sget_key;
1279 int setup_bdev_super(struct super_block *sb, int sb_flags,
1280 struct fs_context *fc)
1282 blk_mode_t mode = sb_open_mode(sb_flags);
1283 struct block_device *bdev;
1285 bdev = blkdev_get_by_dev(sb->s_dev, mode, sb, &fs_holder_ops);
1288 errorf(fc, "%s: Can't open blockdev", fc->source);
1289 return PTR_ERR(bdev);
1293 * This really should be in blkdev_get_by_dev, but right now can't due
1294 * to legacy issues that require us to allow opening a block device node
1295 * writable from userspace even for a read-only block device.
1297 if ((mode & BLK_OPEN_WRITE) && bdev_read_only(bdev)) {
1298 blkdev_put(bdev, sb);
1303 * Until SB_BORN flag is set, there can be no active superblock
1304 * references and thus no filesystem freezing. get_active_super() will
1305 * just loop waiting for SB_BORN so even freeze_bdev() cannot proceed.
1307 * It is enough to check bdev was not frozen before we set s_bdev.
1309 mutex_lock(&bdev->bd_fsfreeze_mutex);
1310 if (bdev->bd_fsfreeze_count > 0) {
1311 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1313 warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
1314 blkdev_put(bdev, sb);
1317 spin_lock(&sb_lock);
1319 sb->s_bdi = bdi_get(bdev->bd_disk->bdi);
1320 if (bdev_stable_writes(bdev))
1321 sb->s_iflags |= SB_I_STABLE_WRITES;
1322 spin_unlock(&sb_lock);
1323 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1325 snprintf(sb->s_id, sizeof(sb->s_id), "%pg", bdev);
1326 shrinker_debugfs_rename(&sb->s_shrink, "sb-%s:%s", sb->s_type->name,
1328 sb_set_blocksize(sb, block_size(bdev));
1331 EXPORT_SYMBOL_GPL(setup_bdev_super);
1334 * get_tree_bdev - Get a superblock based on a single block device
1335 * @fc: The filesystem context holding the parameters
1336 * @fill_super: Helper to initialise a new superblock
1338 int get_tree_bdev(struct fs_context *fc,
1339 int (*fill_super)(struct super_block *,
1340 struct fs_context *))
1342 struct super_block *s;
1347 return invalf(fc, "No source specified");
1349 error = lookup_bdev(fc->source, &dev);
1351 errorf(fc, "%s: Can't lookup blockdev", fc->source);
1355 fc->sb_flags |= SB_NOSEC;
1356 fc->sget_key = &dev;
1357 s = sget_fc(fc, test_bdev_super_fc, set_bdev_super_fc);
1362 /* Don't summarily change the RO/RW state. */
1363 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1364 warnf(fc, "%pg: Can't mount, would change RO state", s->s_bdev);
1365 deactivate_locked_super(s);
1370 * We drop s_umount here because we need to open the bdev and
1371 * bdev->open_mutex ranks above s_umount (blkdev_put() ->
1372 * bdev_mark_dead()). It is safe because we have active sb
1373 * reference and SB_BORN is not set yet.
1375 super_unlock_excl(s);
1376 error = setup_bdev_super(s, fc->sb_flags, fc);
1379 error = fill_super(s, fc);
1381 deactivate_locked_super(s);
1384 s->s_flags |= SB_ACTIVE;
1388 fc->root = dget(s->s_root);
1391 EXPORT_SYMBOL(get_tree_bdev);
1393 static int test_bdev_super(struct super_block *s, void *data)
1395 return !(s->s_iflags & SB_I_RETIRED) && s->s_dev == *(dev_t *)data;
1398 struct dentry *mount_bdev(struct file_system_type *fs_type,
1399 int flags, const char *dev_name, void *data,
1400 int (*fill_super)(struct super_block *, void *, int))
1402 struct super_block *s;
1406 error = lookup_bdev(dev_name, &dev);
1408 return ERR_PTR(error);
1411 s = sget(fs_type, test_bdev_super, set_bdev_super, flags, &dev);
1416 if ((flags ^ s->s_flags) & SB_RDONLY) {
1417 deactivate_locked_super(s);
1418 return ERR_PTR(-EBUSY);
1422 * We drop s_umount here because we need to open the bdev and
1423 * bdev->open_mutex ranks above s_umount (blkdev_put() ->
1424 * bdev_mark_dead()). It is safe because we have active sb
1425 * reference and SB_BORN is not set yet.
1427 super_unlock_excl(s);
1428 error = setup_bdev_super(s, flags, NULL);
1431 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1433 deactivate_locked_super(s);
1434 return ERR_PTR(error);
1437 s->s_flags |= SB_ACTIVE;
1440 return dget(s->s_root);
1442 EXPORT_SYMBOL(mount_bdev);
1444 void kill_block_super(struct super_block *sb)
1446 struct block_device *bdev = sb->s_bdev;
1448 generic_shutdown_super(sb);
1450 sync_blockdev(bdev);
1451 blkdev_put(bdev, sb);
1455 EXPORT_SYMBOL(kill_block_super);
1458 struct dentry *mount_nodev(struct file_system_type *fs_type,
1459 int flags, void *data,
1460 int (*fill_super)(struct super_block *, void *, int))
1463 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1468 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1470 deactivate_locked_super(s);
1471 return ERR_PTR(error);
1473 s->s_flags |= SB_ACTIVE;
1474 return dget(s->s_root);
1476 EXPORT_SYMBOL(mount_nodev);
1478 int reconfigure_single(struct super_block *s,
1479 int flags, void *data)
1481 struct fs_context *fc;
1484 /* The caller really need to be passing fc down into mount_single(),
1485 * then a chunk of this can be removed. [Bollocks -- AV]
1486 * Better yet, reconfiguration shouldn't happen, but rather the second
1487 * mount should be rejected if the parameters are not compatible.
1489 fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK);
1493 ret = parse_monolithic_mount_data(fc, data);
1497 ret = reconfigure_super(fc);
1503 static int compare_single(struct super_block *s, void *p)
1508 struct dentry *mount_single(struct file_system_type *fs_type,
1509 int flags, void *data,
1510 int (*fill_super)(struct super_block *, void *, int))
1512 struct super_block *s;
1515 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1519 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1521 s->s_flags |= SB_ACTIVE;
1523 error = reconfigure_single(s, flags, data);
1525 if (unlikely(error)) {
1526 deactivate_locked_super(s);
1527 return ERR_PTR(error);
1529 return dget(s->s_root);
1531 EXPORT_SYMBOL(mount_single);
1534 * vfs_get_tree - Get the mountable root
1535 * @fc: The superblock configuration context.
1537 * The filesystem is invoked to get or create a superblock which can then later
1538 * be used for mounting. The filesystem places a pointer to the root to be
1539 * used for mounting in @fc->root.
1541 int vfs_get_tree(struct fs_context *fc)
1543 struct super_block *sb;
1549 /* Get the mountable root in fc->root, with a ref on the root and a ref
1550 * on the superblock.
1552 error = fc->ops->get_tree(fc);
1557 pr_err("Filesystem %s get_tree() didn't set fc->root\n",
1559 /* We don't know what the locking state of the superblock is -
1560 * if there is a superblock.
1565 sb = fc->root->d_sb;
1566 WARN_ON(!sb->s_bdi);
1569 * Write barrier is for super_cache_count(). We place it before setting
1570 * SB_BORN as the data dependency between the two functions is the
1571 * superblock structure contents that we just set up, not the SB_BORN
1575 sb->s_flags |= SB_BORN;
1577 error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
1578 if (unlikely(error)) {
1584 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1585 * but s_maxbytes was an unsigned long long for many releases. Throw
1586 * this warning for a little while to try and catch filesystems that
1587 * violate this rule.
1589 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1590 "negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes);
1594 EXPORT_SYMBOL(vfs_get_tree);
1597 * Setup private BDI for given superblock. It gets automatically cleaned up
1598 * in generic_shutdown_super().
1600 int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1602 struct backing_dev_info *bdi;
1606 bdi = bdi_alloc(NUMA_NO_NODE);
1610 va_start(args, fmt);
1611 err = bdi_register_va(bdi, fmt, args);
1617 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1619 sb->s_iflags |= SB_I_PERSB_BDI;
1623 EXPORT_SYMBOL(super_setup_bdi_name);
1626 * Setup private BDI for given superblock. I gets automatically cleaned up
1627 * in generic_shutdown_super().
1629 int super_setup_bdi(struct super_block *sb)
1631 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1633 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1634 atomic_long_inc_return(&bdi_seq));
1636 EXPORT_SYMBOL(super_setup_bdi);
1639 * sb_wait_write - wait until all writers to given file system finish
1640 * @sb: the super for which we wait
1641 * @level: type of writers we wait for (normal vs page fault)
1643 * This function waits until there are no writers of given type to given file
1646 static void sb_wait_write(struct super_block *sb, int level)
1648 percpu_down_write(sb->s_writers.rw_sem + level-1);
1652 * We are going to return to userspace and forget about these locks, the
1653 * ownership goes to the caller of thaw_super() which does unlock().
1655 static void lockdep_sb_freeze_release(struct super_block *sb)
1659 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1660 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1664 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1666 static void lockdep_sb_freeze_acquire(struct super_block *sb)
1670 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1671 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1674 static void sb_freeze_unlock(struct super_block *sb, int level)
1676 for (level--; level >= 0; level--)
1677 percpu_up_write(sb->s_writers.rw_sem + level);
1681 * freeze_super - lock the filesystem and force it into a consistent state
1682 * @sb: the super to lock
1684 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1685 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1688 * During this function, sb->s_writers.frozen goes through these values:
1690 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1692 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1693 * writes should be blocked, though page faults are still allowed. We wait for
1694 * all writes to complete and then proceed to the next stage.
1696 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1697 * but internal fs threads can still modify the filesystem (although they
1698 * should not dirty new pages or inodes), writeback can run etc. After waiting
1699 * for all running page faults we sync the filesystem which will clean all
1700 * dirty pages and inodes (no new dirty pages or inodes can be created when
1703 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1704 * modification are blocked (e.g. XFS preallocation truncation on inode
1705 * reclaim). This is usually implemented by blocking new transactions for
1706 * filesystems that have them and need this additional guard. After all
1707 * internal writers are finished we call ->freeze_fs() to finish filesystem
1708 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1709 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1711 * sb->s_writers.frozen is protected by sb->s_umount.
1713 int freeze_super(struct super_block *sb)
1717 atomic_inc(&sb->s_active);
1718 super_lock_excl(sb);
1719 if (sb->s_writers.frozen != SB_UNFROZEN) {
1720 deactivate_locked_super(sb);
1724 if (!(sb->s_flags & SB_BORN)) {
1725 super_unlock_excl(sb);
1726 return 0; /* sic - it's "nothing to do" */
1729 if (sb_rdonly(sb)) {
1730 /* Nothing to do really... */
1731 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1732 super_unlock_excl(sb);
1736 sb->s_writers.frozen = SB_FREEZE_WRITE;
1737 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1738 super_unlock_excl(sb);
1739 sb_wait_write(sb, SB_FREEZE_WRITE);
1740 super_lock_excl(sb);
1742 /* Now we go and block page faults... */
1743 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1744 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1746 /* All writers are done so after syncing there won't be dirty data */
1747 ret = sync_filesystem(sb);
1749 sb->s_writers.frozen = SB_UNFROZEN;
1750 sb_freeze_unlock(sb, SB_FREEZE_PAGEFAULT);
1751 deactivate_locked_super(sb);
1755 /* Now wait for internal filesystem counter */
1756 sb->s_writers.frozen = SB_FREEZE_FS;
1757 sb_wait_write(sb, SB_FREEZE_FS);
1759 if (sb->s_op->freeze_fs) {
1760 ret = sb->s_op->freeze_fs(sb);
1763 "VFS:Filesystem freeze failed\n");
1764 sb->s_writers.frozen = SB_UNFROZEN;
1765 sb_freeze_unlock(sb, SB_FREEZE_FS);
1766 deactivate_locked_super(sb);
1771 * For debugging purposes so that fs can warn if it sees write activity
1772 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1774 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1775 lockdep_sb_freeze_release(sb);
1776 super_unlock_excl(sb);
1779 EXPORT_SYMBOL(freeze_super);
1781 static int thaw_super_locked(struct super_block *sb)
1785 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1786 super_unlock_excl(sb);
1790 if (sb_rdonly(sb)) {
1791 sb->s_writers.frozen = SB_UNFROZEN;
1795 lockdep_sb_freeze_acquire(sb);
1797 if (sb->s_op->unfreeze_fs) {
1798 error = sb->s_op->unfreeze_fs(sb);
1801 "VFS:Filesystem thaw failed\n");
1802 lockdep_sb_freeze_release(sb);
1803 super_unlock_excl(sb);
1808 sb->s_writers.frozen = SB_UNFROZEN;
1809 sb_freeze_unlock(sb, SB_FREEZE_FS);
1811 deactivate_locked_super(sb);
1816 * thaw_super -- unlock filesystem
1817 * @sb: the super to thaw
1819 * Unlocks the filesystem and marks it writeable again after freeze_super().
1821 int thaw_super(struct super_block *sb)
1823 super_lock_excl(sb);
1824 return thaw_super_locked(sb);
1826 EXPORT_SYMBOL(thaw_super);
1829 * Create workqueue for deferred direct IO completions. We allocate the
1830 * workqueue when it's first needed. This avoids creating workqueue for
1831 * filesystems that don't need it and also allows us to create the workqueue
1832 * late enough so the we can include s_id in the name of the workqueue.
1834 int sb_init_dio_done_wq(struct super_block *sb)
1836 struct workqueue_struct *old;
1837 struct workqueue_struct *wq = alloc_workqueue("dio/%s",
1843 * This has to be atomic as more DIOs can race to create the workqueue
1845 old = cmpxchg(&sb->s_dio_done_wq, NULL, wq);
1846 /* Someone created workqueue before us? Free ours... */
1848 destroy_workqueue(wq);