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 init_waitqueue_head(&s
->s_writers
.wait_unfrozen
);
240 s
->s_bdi
= &noop_backing_dev_info
;
242 if (s
->s_user_ns
!= &init_user_ns
)
243 s
->s_iflags
|= SB_I_NODEV
;
244 INIT_HLIST_NODE(&s
->s_instances
);
245 INIT_HLIST_BL_HEAD(&s
->s_roots
);
246 mutex_init(&s
->s_sync_lock
);
247 INIT_LIST_HEAD(&s
->s_inodes
);
248 spin_lock_init(&s
->s_inode_list_lock
);
249 INIT_LIST_HEAD(&s
->s_inodes_wb
);
250 spin_lock_init(&s
->s_inode_wblist_lock
);
253 atomic_set(&s
->s_active
, 1);
254 mutex_init(&s
->s_vfs_rename_mutex
);
255 lockdep_set_class(&s
->s_vfs_rename_mutex
, &type
->s_vfs_rename_key
);
256 init_rwsem(&s
->s_dquot
.dqio_sem
);
257 s
->s_maxbytes
= MAX_NON_LFS
;
258 s
->s_op
= &default_op
;
259 s
->s_time_gran
= 1000000000;
260 s
->s_time_min
= TIME64_MIN
;
261 s
->s_time_max
= TIME64_MAX
;
263 s
->s_shrink
.seeks
= DEFAULT_SEEKS
;
264 s
->s_shrink
.scan_objects
= super_cache_scan
;
265 s
->s_shrink
.count_objects
= super_cache_count
;
266 s
->s_shrink
.batch
= 1024;
267 s
->s_shrink
.flags
= SHRINKER_NUMA_AWARE
| SHRINKER_MEMCG_AWARE
;
268 if (prealloc_shrinker(&s
->s_shrink
, "sb-%s", type
->name
))
270 if (list_lru_init_memcg(&s
->s_dentry_lru
, &s
->s_shrink
))
272 if (list_lru_init_memcg(&s
->s_inode_lru
, &s
->s_shrink
))
277 destroy_unused_super(s
);
281 /* Superblock refcounting */
284 * Drop a superblock's refcount. The caller must hold sb_lock.
286 static void __put_super(struct super_block
*s
)
289 list_del_init(&s
->s_list
);
290 WARN_ON(s
->s_dentry_lru
.node
);
291 WARN_ON(s
->s_inode_lru
.node
);
292 WARN_ON(!list_empty(&s
->s_mounts
));
294 fscrypt_destroy_keyring(s
);
295 put_user_ns(s
->s_user_ns
);
297 call_rcu(&s
->rcu
, destroy_super_rcu
);
302 * put_super - drop a temporary reference to superblock
303 * @sb: superblock in question
305 * Drops a temporary reference, frees superblock if there's no
308 void put_super(struct super_block
*sb
)
312 spin_unlock(&sb_lock
);
317 * deactivate_locked_super - drop an active reference to superblock
318 * @s: superblock to deactivate
320 * Drops an active reference to superblock, converting it into a temporary
321 * one if there is no other active references left. In that case we
322 * tell fs driver to shut it down and drop the temporary reference we
325 * Caller holds exclusive lock on superblock; that lock is released.
327 void deactivate_locked_super(struct super_block
*s
)
329 struct file_system_type
*fs
= s
->s_type
;
330 if (atomic_dec_and_test(&s
->s_active
)) {
331 unregister_shrinker(&s
->s_shrink
);
335 * Since list_lru_destroy() may sleep, we cannot call it from
336 * put_super(), where we hold the sb_lock. Therefore we destroy
337 * the lru lists right now.
339 list_lru_destroy(&s
->s_dentry_lru
);
340 list_lru_destroy(&s
->s_inode_lru
);
345 up_write(&s
->s_umount
);
349 EXPORT_SYMBOL(deactivate_locked_super
);
352 * deactivate_super - drop an active reference to superblock
353 * @s: superblock to deactivate
355 * Variant of deactivate_locked_super(), except that superblock is *not*
356 * locked by caller. If we are going to drop the final active reference,
357 * lock will be acquired prior to that.
359 void deactivate_super(struct super_block
*s
)
361 if (!atomic_add_unless(&s
->s_active
, -1, 1)) {
362 down_write(&s
->s_umount
);
363 deactivate_locked_super(s
);
367 EXPORT_SYMBOL(deactivate_super
);
370 * grab_super - acquire an active reference
371 * @s: reference we are trying to make active
373 * Tries to acquire an active reference. grab_super() is used when we
374 * had just found a superblock in super_blocks or fs_type->fs_supers
375 * and want to turn it into a full-blown active reference. grab_super()
376 * is called with sb_lock held and drops it. Returns 1 in case of
377 * success, 0 if we had failed (superblock contents was already dead or
378 * dying when grab_super() had been called). Note that this is only
379 * called for superblocks not in rundown mode (== ones still on ->fs_supers
380 * of their type), so increment of ->s_count is OK here.
382 static int grab_super(struct super_block
*s
) __releases(sb_lock
)
385 spin_unlock(&sb_lock
);
386 down_write(&s
->s_umount
);
387 if ((s
->s_flags
& SB_BORN
) && atomic_inc_not_zero(&s
->s_active
)) {
391 up_write(&s
->s_umount
);
397 * trylock_super - try to grab ->s_umount shared
398 * @sb: reference we are trying to grab
400 * Try to prevent fs shutdown. This is used in places where we
401 * cannot take an active reference but we need to ensure that the
402 * filesystem is not shut down while we are working on it. It returns
403 * false if we cannot acquire s_umount or if we lose the race and
404 * filesystem already got into shutdown, and returns true with the s_umount
405 * lock held in read mode in case of success. On successful return,
406 * the caller must drop the s_umount lock when done.
408 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
409 * The reason why it's safe is that we are OK with doing trylock instead
410 * of down_read(). There's a couple of places that are OK with that, but
411 * it's very much not a general-purpose interface.
413 bool trylock_super(struct super_block
*sb
)
415 if (down_read_trylock(&sb
->s_umount
)) {
416 if (!hlist_unhashed(&sb
->s_instances
) &&
417 sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
419 up_read(&sb
->s_umount
);
426 * retire_super - prevents superblock from being reused
427 * @sb: superblock to retire
429 * The function marks superblock to be ignored in superblock test, which
430 * prevents it from being reused for any new mounts. If the superblock has
431 * a private bdi, it also unregisters it, but doesn't reduce the refcount
432 * of the superblock to prevent potential races. The refcount is reduced
433 * by generic_shutdown_super(). The function can not be called
434 * concurrently with generic_shutdown_super(). It is safe to call the
435 * function multiple times, subsequent calls have no effect.
437 * The marker will affect the re-use only for block-device-based
438 * superblocks. Other superblocks will still get marked if this function
439 * is used, but that will not affect their reusability.
441 void retire_super(struct super_block
*sb
)
443 WARN_ON(!sb
->s_bdev
);
444 down_write(&sb
->s_umount
);
445 if (sb
->s_iflags
& SB_I_PERSB_BDI
) {
446 bdi_unregister(sb
->s_bdi
);
447 sb
->s_iflags
&= ~SB_I_PERSB_BDI
;
449 sb
->s_iflags
|= SB_I_RETIRED
;
450 up_write(&sb
->s_umount
);
452 EXPORT_SYMBOL(retire_super
);
455 * generic_shutdown_super - common helper for ->kill_sb()
456 * @sb: superblock to kill
458 * generic_shutdown_super() does all fs-independent work on superblock
459 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
460 * that need destruction out of superblock, call generic_shutdown_super()
461 * and release aforementioned objects. Note: dentries and inodes _are_
462 * taken care of and do not need specific handling.
464 * Upon calling this function, the filesystem may no longer alter or
465 * rearrange the set of dentries belonging to this super_block, nor may it
466 * change the attachments of dentries to inodes.
468 void generic_shutdown_super(struct super_block
*sb
)
470 const struct super_operations
*sop
= sb
->s_op
;
473 shrink_dcache_for_umount(sb
);
475 sb
->s_flags
&= ~SB_ACTIVE
;
477 cgroup_writeback_umount();
479 /* evict all inodes with zero refcount */
481 /* only nonzero refcount inodes can have marks */
482 fsnotify_sb_delete(sb
);
483 fscrypt_destroy_keyring(sb
);
484 security_sb_delete(sb
);
486 if (sb
->s_dio_done_wq
) {
487 destroy_workqueue(sb
->s_dio_done_wq
);
488 sb
->s_dio_done_wq
= NULL
;
494 if (!list_empty(&sb
->s_inodes
)) {
495 printk("VFS: Busy inodes after unmount of %s. "
496 "Self-destruct in 5 seconds. Have a nice day...\n",
501 /* should be initialized for __put_super_and_need_restart() */
502 hlist_del_init(&sb
->s_instances
);
503 spin_unlock(&sb_lock
);
504 up_write(&sb
->s_umount
);
505 if (sb
->s_bdi
!= &noop_backing_dev_info
) {
506 if (sb
->s_iflags
& SB_I_PERSB_BDI
)
507 bdi_unregister(sb
->s_bdi
);
509 sb
->s_bdi
= &noop_backing_dev_info
;
513 EXPORT_SYMBOL(generic_shutdown_super
);
515 bool mount_capable(struct fs_context
*fc
)
517 if (!(fc
->fs_type
->fs_flags
& FS_USERNS_MOUNT
))
518 return capable(CAP_SYS_ADMIN
);
520 return ns_capable(fc
->user_ns
, CAP_SYS_ADMIN
);
524 * sget_fc - Find or create a superblock
525 * @fc: Filesystem context.
526 * @test: Comparison callback
527 * @set: Setup callback
529 * Find or create a superblock using the parameters stored in the filesystem
530 * context and the two callback functions.
532 * If an extant superblock is matched, then that will be returned with an
533 * elevated reference count that the caller must transfer or discard.
535 * If no match is made, a new superblock will be allocated and basic
536 * initialisation will be performed (s_type, s_fs_info and s_id will be set and
537 * the set() callback will be invoked), the superblock will be published and it
538 * will be returned in a partially constructed state with SB_BORN and SB_ACTIVE
541 struct super_block
*sget_fc(struct fs_context
*fc
,
542 int (*test
)(struct super_block
*, struct fs_context
*),
543 int (*set
)(struct super_block
*, struct fs_context
*))
545 struct super_block
*s
= NULL
;
546 struct super_block
*old
;
547 struct user_namespace
*user_ns
= fc
->global
? &init_user_ns
: fc
->user_ns
;
553 hlist_for_each_entry(old
, &fc
->fs_type
->fs_supers
, s_instances
) {
555 goto share_extant_sb
;
559 spin_unlock(&sb_lock
);
560 s
= alloc_super(fc
->fs_type
, fc
->sb_flags
, user_ns
);
562 return ERR_PTR(-ENOMEM
);
566 s
->s_fs_info
= fc
->s_fs_info
;
570 spin_unlock(&sb_lock
);
571 destroy_unused_super(s
);
574 fc
->s_fs_info
= NULL
;
575 s
->s_type
= fc
->fs_type
;
576 s
->s_iflags
|= fc
->s_iflags
;
577 strlcpy(s
->s_id
, s
->s_type
->name
, sizeof(s
->s_id
));
578 list_add_tail(&s
->s_list
, &super_blocks
);
579 hlist_add_head(&s
->s_instances
, &s
->s_type
->fs_supers
);
580 spin_unlock(&sb_lock
);
581 get_filesystem(s
->s_type
);
582 register_shrinker_prepared(&s
->s_shrink
);
586 if (user_ns
!= old
->s_user_ns
) {
587 spin_unlock(&sb_lock
);
588 destroy_unused_super(s
);
589 return ERR_PTR(-EBUSY
);
591 if (!grab_super(old
))
593 destroy_unused_super(s
);
596 EXPORT_SYMBOL(sget_fc
);
599 * sget - find or create a superblock
600 * @type: filesystem type superblock should belong to
601 * @test: comparison callback
602 * @set: setup callback
603 * @flags: mount flags
604 * @data: argument to each of them
606 struct super_block
*sget(struct file_system_type
*type
,
607 int (*test
)(struct super_block
*,void *),
608 int (*set
)(struct super_block
*,void *),
612 struct user_namespace
*user_ns
= current_user_ns();
613 struct super_block
*s
= NULL
;
614 struct super_block
*old
;
617 /* We don't yet pass the user namespace of the parent
618 * mount through to here so always use &init_user_ns
619 * until that changes.
621 if (flags
& SB_SUBMOUNT
)
622 user_ns
= &init_user_ns
;
627 hlist_for_each_entry(old
, &type
->fs_supers
, s_instances
) {
628 if (!test(old
, data
))
630 if (user_ns
!= old
->s_user_ns
) {
631 spin_unlock(&sb_lock
);
632 destroy_unused_super(s
);
633 return ERR_PTR(-EBUSY
);
635 if (!grab_super(old
))
637 destroy_unused_super(s
);
642 spin_unlock(&sb_lock
);
643 s
= alloc_super(type
, (flags
& ~SB_SUBMOUNT
), user_ns
);
645 return ERR_PTR(-ENOMEM
);
651 spin_unlock(&sb_lock
);
652 destroy_unused_super(s
);
656 strlcpy(s
->s_id
, type
->name
, sizeof(s
->s_id
));
657 list_add_tail(&s
->s_list
, &super_blocks
);
658 hlist_add_head(&s
->s_instances
, &type
->fs_supers
);
659 spin_unlock(&sb_lock
);
660 get_filesystem(type
);
661 register_shrinker_prepared(&s
->s_shrink
);
666 void drop_super(struct super_block
*sb
)
668 up_read(&sb
->s_umount
);
672 EXPORT_SYMBOL(drop_super
);
674 void drop_super_exclusive(struct super_block
*sb
)
676 up_write(&sb
->s_umount
);
679 EXPORT_SYMBOL(drop_super_exclusive
);
681 static void __iterate_supers(void (*f
)(struct super_block
*))
683 struct super_block
*sb
, *p
= NULL
;
686 list_for_each_entry(sb
, &super_blocks
, s_list
) {
687 if (hlist_unhashed(&sb
->s_instances
))
690 spin_unlock(&sb_lock
);
701 spin_unlock(&sb_lock
);
704 * iterate_supers - call function for all active superblocks
705 * @f: function to call
706 * @arg: argument to pass to it
708 * Scans the superblock list and calls given function, passing it
709 * locked superblock and given argument.
711 void iterate_supers(void (*f
)(struct super_block
*, void *), void *arg
)
713 struct super_block
*sb
, *p
= NULL
;
716 list_for_each_entry(sb
, &super_blocks
, s_list
) {
717 if (hlist_unhashed(&sb
->s_instances
))
720 spin_unlock(&sb_lock
);
722 down_read(&sb
->s_umount
);
723 if (sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
725 up_read(&sb
->s_umount
);
734 spin_unlock(&sb_lock
);
738 * iterate_supers_type - call function for superblocks of given type
740 * @f: function to call
741 * @arg: argument to pass to it
743 * Scans the superblock list and calls given function, passing it
744 * locked superblock and given argument.
746 void iterate_supers_type(struct file_system_type
*type
,
747 void (*f
)(struct super_block
*, void *), void *arg
)
749 struct super_block
*sb
, *p
= NULL
;
752 hlist_for_each_entry(sb
, &type
->fs_supers
, s_instances
) {
754 spin_unlock(&sb_lock
);
756 down_read(&sb
->s_umount
);
757 if (sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
759 up_read(&sb
->s_umount
);
768 spin_unlock(&sb_lock
);
771 EXPORT_SYMBOL(iterate_supers_type
);
774 * get_super - get the superblock of a device
775 * @bdev: device to get the superblock for
777 * Scans the superblock list and finds the superblock of the file system
778 * mounted on the device given. %NULL is returned if no match is found.
780 struct super_block
*get_super(struct block_device
*bdev
)
782 struct super_block
*sb
;
789 list_for_each_entry(sb
, &super_blocks
, s_list
) {
790 if (hlist_unhashed(&sb
->s_instances
))
792 if (sb
->s_bdev
== bdev
) {
794 spin_unlock(&sb_lock
);
795 down_read(&sb
->s_umount
);
797 if (sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
799 up_read(&sb
->s_umount
);
800 /* nope, got unmounted */
806 spin_unlock(&sb_lock
);
811 * get_active_super - get an active reference to the superblock of a device
812 * @bdev: device to get the superblock for
814 * Scans the superblock list and finds the superblock of the file system
815 * mounted on the device given. Returns the superblock with an active
816 * reference or %NULL if none was found.
818 struct super_block
*get_active_super(struct block_device
*bdev
)
820 struct super_block
*sb
;
827 list_for_each_entry(sb
, &super_blocks
, s_list
) {
828 if (hlist_unhashed(&sb
->s_instances
))
830 if (sb
->s_bdev
== bdev
) {
833 up_write(&sb
->s_umount
);
837 spin_unlock(&sb_lock
);
841 struct super_block
*user_get_super(dev_t dev
, bool excl
)
843 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_dev
== dev
) {
852 spin_unlock(&sb_lock
);
854 down_write(&sb
->s_umount
);
856 down_read(&sb
->s_umount
);
858 if (sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
861 up_write(&sb
->s_umount
);
863 up_read(&sb
->s_umount
);
864 /* nope, got unmounted */
870 spin_unlock(&sb_lock
);
875 * reconfigure_super - asks filesystem to change superblock parameters
876 * @fc: The superblock and configuration
878 * Alters the configuration parameters of a live superblock.
880 int reconfigure_super(struct fs_context
*fc
)
882 struct super_block
*sb
= fc
->root
->d_sb
;
884 bool remount_ro
= false;
885 bool force
= fc
->sb_flags
& SB_FORCE
;
887 if (fc
->sb_flags_mask
& ~MS_RMT_MASK
)
889 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
)
892 retval
= security_sb_remount(sb
, fc
->security
);
896 if (fc
->sb_flags_mask
& SB_RDONLY
) {
898 if (!(fc
->sb_flags
& SB_RDONLY
) && sb
->s_bdev
&&
899 bdev_read_only(sb
->s_bdev
))
903 remount_ro
= (fc
->sb_flags
& SB_RDONLY
) && !sb_rdonly(sb
);
907 if (!hlist_empty(&sb
->s_pins
)) {
908 up_write(&sb
->s_umount
);
909 group_pin_kill(&sb
->s_pins
);
910 down_write(&sb
->s_umount
);
913 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
)
915 remount_ro
= !sb_rdonly(sb
);
918 shrink_dcache_sb(sb
);
920 /* If we are reconfiguring to RDONLY and current sb is read/write,
921 * make sure there are no files open for writing.
925 sb
->s_readonly_remount
= 1;
928 retval
= sb_prepare_remount_readonly(sb
);
934 if (fc
->ops
->reconfigure
) {
935 retval
= fc
->ops
->reconfigure(fc
);
938 goto cancel_readonly
;
939 /* If forced remount, go ahead despite any errors */
940 WARN(1, "forced remount of a %s fs returned %i\n",
941 sb
->s_type
->name
, retval
);
945 WRITE_ONCE(sb
->s_flags
, ((sb
->s_flags
& ~fc
->sb_flags_mask
) |
946 (fc
->sb_flags
& fc
->sb_flags_mask
)));
947 /* Needs to be ordered wrt mnt_is_readonly() */
949 sb
->s_readonly_remount
= 0;
952 * Some filesystems modify their metadata via some other path than the
953 * bdev buffer cache (eg. use a private mapping, or directories in
954 * pagecache, etc). Also file data modifications go via their own
955 * mappings. So If we try to mount readonly then copy the filesystem
956 * from bdev, we could get stale data, so invalidate it to give a best
957 * effort at coherency.
959 if (remount_ro
&& sb
->s_bdev
)
960 invalidate_bdev(sb
->s_bdev
);
964 sb
->s_readonly_remount
= 0;
968 static void do_emergency_remount_callback(struct super_block
*sb
)
970 down_write(&sb
->s_umount
);
971 if (sb
->s_root
&& sb
->s_bdev
&& (sb
->s_flags
& SB_BORN
) &&
973 struct fs_context
*fc
;
975 fc
= fs_context_for_reconfigure(sb
->s_root
,
976 SB_RDONLY
| SB_FORCE
, SB_RDONLY
);
978 if (parse_monolithic_mount_data(fc
, NULL
) == 0)
979 (void)reconfigure_super(fc
);
983 up_write(&sb
->s_umount
);
986 static void do_emergency_remount(struct work_struct
*work
)
988 __iterate_supers(do_emergency_remount_callback
);
990 printk("Emergency Remount complete\n");
993 void emergency_remount(void)
995 struct work_struct
*work
;
997 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
999 INIT_WORK(work
, do_emergency_remount
);
1000 schedule_work(work
);
1004 static void do_thaw_all_callback(struct super_block
*sb
)
1006 down_write(&sb
->s_umount
);
1007 if (sb
->s_root
&& sb
->s_flags
& SB_BORN
) {
1008 emergency_thaw_bdev(sb
);
1009 thaw_super_locked(sb
);
1011 up_write(&sb
->s_umount
);
1015 static void do_thaw_all(struct work_struct
*work
)
1017 __iterate_supers(do_thaw_all_callback
);
1019 printk(KERN_WARNING
"Emergency Thaw complete\n");
1023 * emergency_thaw_all -- forcibly thaw every frozen filesystem
1025 * Used for emergency unfreeze of all filesystems via SysRq
1027 void emergency_thaw_all(void)
1029 struct work_struct
*work
;
1031 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
1033 INIT_WORK(work
, do_thaw_all
);
1034 schedule_work(work
);
1038 static DEFINE_IDA(unnamed_dev_ida
);
1041 * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1042 * @p: Pointer to a dev_t.
1044 * Filesystems which don't use real block devices can call this function
1045 * to allocate a virtual block device.
1047 * Context: Any context. Frequently called while holding sb_lock.
1048 * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1049 * or -ENOMEM if memory allocation failed.
1051 int get_anon_bdev(dev_t
*p
)
1056 * Many userspace utilities consider an FSID of 0 invalid.
1057 * Always return at least 1 from get_anon_bdev.
1059 dev
= ida_alloc_range(&unnamed_dev_ida
, 1, (1 << MINORBITS
) - 1,
1069 EXPORT_SYMBOL(get_anon_bdev
);
1071 void free_anon_bdev(dev_t dev
)
1073 ida_free(&unnamed_dev_ida
, MINOR(dev
));
1075 EXPORT_SYMBOL(free_anon_bdev
);
1077 int set_anon_super(struct super_block
*s
, void *data
)
1079 return get_anon_bdev(&s
->s_dev
);
1081 EXPORT_SYMBOL(set_anon_super
);
1083 void kill_anon_super(struct super_block
*sb
)
1085 dev_t dev
= sb
->s_dev
;
1086 generic_shutdown_super(sb
);
1087 free_anon_bdev(dev
);
1089 EXPORT_SYMBOL(kill_anon_super
);
1091 void kill_litter_super(struct super_block
*sb
)
1094 d_genocide(sb
->s_root
);
1095 kill_anon_super(sb
);
1097 EXPORT_SYMBOL(kill_litter_super
);
1099 int set_anon_super_fc(struct super_block
*sb
, struct fs_context
*fc
)
1101 return set_anon_super(sb
, NULL
);
1103 EXPORT_SYMBOL(set_anon_super_fc
);
1105 static int test_keyed_super(struct super_block
*sb
, struct fs_context
*fc
)
1107 return sb
->s_fs_info
== fc
->s_fs_info
;
1110 static int test_single_super(struct super_block
*s
, struct fs_context
*fc
)
1115 static int vfs_get_super(struct fs_context
*fc
, bool reconf
,
1116 int (*test
)(struct super_block
*, struct fs_context
*),
1117 int (*fill_super
)(struct super_block
*sb
,
1118 struct fs_context
*fc
))
1120 struct super_block
*sb
;
1123 sb
= sget_fc(fc
, test
, set_anon_super_fc
);
1128 err
= fill_super(sb
, fc
);
1132 sb
->s_flags
|= SB_ACTIVE
;
1133 fc
->root
= dget(sb
->s_root
);
1135 fc
->root
= dget(sb
->s_root
);
1137 err
= reconfigure_super(fc
);
1149 deactivate_locked_super(sb
);
1153 int get_tree_nodev(struct fs_context
*fc
,
1154 int (*fill_super
)(struct super_block
*sb
,
1155 struct fs_context
*fc
))
1157 return vfs_get_super(fc
, false, NULL
, fill_super
);
1159 EXPORT_SYMBOL(get_tree_nodev
);
1161 int get_tree_single(struct fs_context
*fc
,
1162 int (*fill_super
)(struct super_block
*sb
,
1163 struct fs_context
*fc
))
1165 return vfs_get_super(fc
, false, test_single_super
, fill_super
);
1167 EXPORT_SYMBOL(get_tree_single
);
1169 int get_tree_single_reconf(struct fs_context
*fc
,
1170 int (*fill_super
)(struct super_block
*sb
,
1171 struct fs_context
*fc
))
1173 return vfs_get_super(fc
, true, test_single_super
, fill_super
);
1175 EXPORT_SYMBOL(get_tree_single_reconf
);
1177 int get_tree_keyed(struct fs_context
*fc
,
1178 int (*fill_super
)(struct super_block
*sb
,
1179 struct fs_context
*fc
),
1182 fc
->s_fs_info
= key
;
1183 return vfs_get_super(fc
, false, test_keyed_super
, fill_super
);
1185 EXPORT_SYMBOL(get_tree_keyed
);
1189 static int set_bdev_super(struct super_block
*s
, void *data
)
1192 s
->s_dev
= s
->s_bdev
->bd_dev
;
1193 s
->s_bdi
= bdi_get(s
->s_bdev
->bd_disk
->bdi
);
1195 if (bdev_stable_writes(s
->s_bdev
))
1196 s
->s_iflags
|= SB_I_STABLE_WRITES
;
1200 static int set_bdev_super_fc(struct super_block
*s
, struct fs_context
*fc
)
1202 return set_bdev_super(s
, fc
->sget_key
);
1205 static int test_bdev_super_fc(struct super_block
*s
, struct fs_context
*fc
)
1207 return !(s
->s_iflags
& SB_I_RETIRED
) && s
->s_bdev
== fc
->sget_key
;
1211 * get_tree_bdev - Get a superblock based on a single block device
1212 * @fc: The filesystem context holding the parameters
1213 * @fill_super: Helper to initialise a new superblock
1215 int get_tree_bdev(struct fs_context
*fc
,
1216 int (*fill_super
)(struct super_block
*,
1217 struct fs_context
*))
1219 struct block_device
*bdev
;
1220 struct super_block
*s
;
1221 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1224 if (!(fc
->sb_flags
& SB_RDONLY
))
1225 mode
|= FMODE_WRITE
;
1228 return invalf(fc
, "No source specified");
1230 bdev
= blkdev_get_by_path(fc
->source
, mode
, fc
->fs_type
);
1232 errorf(fc
, "%s: Can't open blockdev", fc
->source
);
1233 return PTR_ERR(bdev
);
1236 /* Once the superblock is inserted into the list by sget_fc(), s_umount
1237 * will protect the lockfs code from trying to start a snapshot while
1240 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
1241 if (bdev
->bd_fsfreeze_count
> 0) {
1242 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1243 warnf(fc
, "%pg: Can't mount, blockdev is frozen", bdev
);
1244 blkdev_put(bdev
, mode
);
1248 fc
->sb_flags
|= SB_NOSEC
;
1249 fc
->sget_key
= bdev
;
1250 s
= sget_fc(fc
, test_bdev_super_fc
, set_bdev_super_fc
);
1251 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1253 blkdev_put(bdev
, mode
);
1258 /* Don't summarily change the RO/RW state. */
1259 if ((fc
->sb_flags
^ s
->s_flags
) & SB_RDONLY
) {
1260 warnf(fc
, "%pg: Can't mount, would change RO state", bdev
);
1261 deactivate_locked_super(s
);
1262 blkdev_put(bdev
, mode
);
1267 * s_umount nests inside open_mutex during
1268 * __invalidate_device(). blkdev_put() acquires
1269 * open_mutex and can't be called under s_umount. Drop
1270 * s_umount temporarily. This is safe as we're
1271 * holding an active reference.
1273 up_write(&s
->s_umount
);
1274 blkdev_put(bdev
, mode
);
1275 down_write(&s
->s_umount
);
1278 snprintf(s
->s_id
, sizeof(s
->s_id
), "%pg", bdev
);
1279 shrinker_debugfs_rename(&s
->s_shrink
, "sb-%s:%s",
1280 fc
->fs_type
->name
, s
->s_id
);
1281 sb_set_blocksize(s
, block_size(bdev
));
1282 error
= fill_super(s
, fc
);
1284 deactivate_locked_super(s
);
1288 s
->s_flags
|= SB_ACTIVE
;
1293 fc
->root
= dget(s
->s_root
);
1296 EXPORT_SYMBOL(get_tree_bdev
);
1298 static int test_bdev_super(struct super_block
*s
, void *data
)
1300 return !(s
->s_iflags
& SB_I_RETIRED
) && (void *)s
->s_bdev
== data
;
1303 struct dentry
*mount_bdev(struct file_system_type
*fs_type
,
1304 int flags
, const char *dev_name
, void *data
,
1305 int (*fill_super
)(struct super_block
*, void *, int))
1307 struct block_device
*bdev
;
1308 struct super_block
*s
;
1309 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1312 if (!(flags
& SB_RDONLY
))
1313 mode
|= FMODE_WRITE
;
1315 bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
1317 return ERR_CAST(bdev
);
1320 * once the super is inserted into the list by sget, s_umount
1321 * will protect the lockfs code from trying to start a snapshot
1322 * while we are mounting
1324 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
1325 if (bdev
->bd_fsfreeze_count
> 0) {
1326 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1330 s
= sget(fs_type
, test_bdev_super
, set_bdev_super
, flags
| SB_NOSEC
,
1332 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1337 if ((flags
^ s
->s_flags
) & SB_RDONLY
) {
1338 deactivate_locked_super(s
);
1344 * s_umount nests inside open_mutex during
1345 * __invalidate_device(). blkdev_put() acquires
1346 * open_mutex and can't be called under s_umount. Drop
1347 * s_umount temporarily. This is safe as we're
1348 * holding an active reference.
1350 up_write(&s
->s_umount
);
1351 blkdev_put(bdev
, mode
);
1352 down_write(&s
->s_umount
);
1355 snprintf(s
->s_id
, sizeof(s
->s_id
), "%pg", bdev
);
1356 shrinker_debugfs_rename(&s
->s_shrink
, "sb-%s:%s",
1357 fs_type
->name
, s
->s_id
);
1358 sb_set_blocksize(s
, block_size(bdev
));
1359 error
= fill_super(s
, data
, flags
& SB_SILENT
? 1 : 0);
1361 deactivate_locked_super(s
);
1365 s
->s_flags
|= SB_ACTIVE
;
1369 return dget(s
->s_root
);
1374 blkdev_put(bdev
, mode
);
1376 return ERR_PTR(error
);
1378 EXPORT_SYMBOL(mount_bdev
);
1380 void kill_block_super(struct super_block
*sb
)
1382 struct block_device
*bdev
= sb
->s_bdev
;
1383 fmode_t mode
= sb
->s_mode
;
1385 bdev
->bd_super
= NULL
;
1386 generic_shutdown_super(sb
);
1387 sync_blockdev(bdev
);
1388 WARN_ON_ONCE(!(mode
& FMODE_EXCL
));
1389 blkdev_put(bdev
, mode
| FMODE_EXCL
);
1392 EXPORT_SYMBOL(kill_block_super
);
1395 struct dentry
*mount_nodev(struct file_system_type
*fs_type
,
1396 int flags
, void *data
,
1397 int (*fill_super
)(struct super_block
*, void *, int))
1400 struct super_block
*s
= sget(fs_type
, NULL
, set_anon_super
, flags
, NULL
);
1405 error
= fill_super(s
, data
, flags
& SB_SILENT
? 1 : 0);
1407 deactivate_locked_super(s
);
1408 return ERR_PTR(error
);
1410 s
->s_flags
|= SB_ACTIVE
;
1411 return dget(s
->s_root
);
1413 EXPORT_SYMBOL(mount_nodev
);
1415 int reconfigure_single(struct super_block
*s
,
1416 int flags
, void *data
)
1418 struct fs_context
*fc
;
1421 /* The caller really need to be passing fc down into mount_single(),
1422 * then a chunk of this can be removed. [Bollocks -- AV]
1423 * Better yet, reconfiguration shouldn't happen, but rather the second
1424 * mount should be rejected if the parameters are not compatible.
1426 fc
= fs_context_for_reconfigure(s
->s_root
, flags
, MS_RMT_MASK
);
1430 ret
= parse_monolithic_mount_data(fc
, data
);
1434 ret
= reconfigure_super(fc
);
1440 static int compare_single(struct super_block
*s
, void *p
)
1445 struct dentry
*mount_single(struct file_system_type
*fs_type
,
1446 int flags
, void *data
,
1447 int (*fill_super
)(struct super_block
*, void *, int))
1449 struct super_block
*s
;
1452 s
= sget(fs_type
, compare_single
, set_anon_super
, flags
, NULL
);
1456 error
= fill_super(s
, data
, flags
& SB_SILENT
? 1 : 0);
1458 s
->s_flags
|= SB_ACTIVE
;
1460 error
= reconfigure_single(s
, flags
, data
);
1462 if (unlikely(error
)) {
1463 deactivate_locked_super(s
);
1464 return ERR_PTR(error
);
1466 return dget(s
->s_root
);
1468 EXPORT_SYMBOL(mount_single
);
1471 * vfs_get_tree - Get the mountable root
1472 * @fc: The superblock configuration context.
1474 * The filesystem is invoked to get or create a superblock which can then later
1475 * be used for mounting. The filesystem places a pointer to the root to be
1476 * used for mounting in @fc->root.
1478 int vfs_get_tree(struct fs_context
*fc
)
1480 struct super_block
*sb
;
1486 /* Get the mountable root in fc->root, with a ref on the root and a ref
1487 * on the superblock.
1489 error
= fc
->ops
->get_tree(fc
);
1494 pr_err("Filesystem %s get_tree() didn't set fc->root\n",
1496 /* We don't know what the locking state of the superblock is -
1497 * if there is a superblock.
1502 sb
= fc
->root
->d_sb
;
1503 WARN_ON(!sb
->s_bdi
);
1506 * Write barrier is for super_cache_count(). We place it before setting
1507 * SB_BORN as the data dependency between the two functions is the
1508 * superblock structure contents that we just set up, not the SB_BORN
1512 sb
->s_flags
|= SB_BORN
;
1514 error
= security_sb_set_mnt_opts(sb
, fc
->security
, 0, NULL
);
1515 if (unlikely(error
)) {
1521 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1522 * but s_maxbytes was an unsigned long long for many releases. Throw
1523 * this warning for a little while to try and catch filesystems that
1524 * violate this rule.
1526 WARN((sb
->s_maxbytes
< 0), "%s set sb->s_maxbytes to "
1527 "negative value (%lld)\n", fc
->fs_type
->name
, sb
->s_maxbytes
);
1531 EXPORT_SYMBOL(vfs_get_tree
);
1534 * Setup private BDI for given superblock. It gets automatically cleaned up
1535 * in generic_shutdown_super().
1537 int super_setup_bdi_name(struct super_block
*sb
, char *fmt
, ...)
1539 struct backing_dev_info
*bdi
;
1543 bdi
= bdi_alloc(NUMA_NO_NODE
);
1547 va_start(args
, fmt
);
1548 err
= bdi_register_va(bdi
, fmt
, args
);
1554 WARN_ON(sb
->s_bdi
!= &noop_backing_dev_info
);
1556 sb
->s_iflags
|= SB_I_PERSB_BDI
;
1560 EXPORT_SYMBOL(super_setup_bdi_name
);
1563 * Setup private BDI for given superblock. I gets automatically cleaned up
1564 * in generic_shutdown_super().
1566 int super_setup_bdi(struct super_block
*sb
)
1568 static atomic_long_t bdi_seq
= ATOMIC_LONG_INIT(0);
1570 return super_setup_bdi_name(sb
, "%.28s-%ld", sb
->s_type
->name
,
1571 atomic_long_inc_return(&bdi_seq
));
1573 EXPORT_SYMBOL(super_setup_bdi
);
1576 * sb_wait_write - wait until all writers to given file system finish
1577 * @sb: the super for which we wait
1578 * @level: type of writers we wait for (normal vs page fault)
1580 * This function waits until there are no writers of given type to given file
1583 static void sb_wait_write(struct super_block
*sb
, int level
)
1585 percpu_down_write(sb
->s_writers
.rw_sem
+ level
-1);
1589 * We are going to return to userspace and forget about these locks, the
1590 * ownership goes to the caller of thaw_super() which does unlock().
1592 static void lockdep_sb_freeze_release(struct super_block
*sb
)
1596 for (level
= SB_FREEZE_LEVELS
- 1; level
>= 0; level
--)
1597 percpu_rwsem_release(sb
->s_writers
.rw_sem
+ level
, 0, _THIS_IP_
);
1601 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1603 static void lockdep_sb_freeze_acquire(struct super_block
*sb
)
1607 for (level
= 0; level
< SB_FREEZE_LEVELS
; ++level
)
1608 percpu_rwsem_acquire(sb
->s_writers
.rw_sem
+ level
, 0, _THIS_IP_
);
1611 static void sb_freeze_unlock(struct super_block
*sb
, int level
)
1613 for (level
--; level
>= 0; level
--)
1614 percpu_up_write(sb
->s_writers
.rw_sem
+ level
);
1618 * freeze_super - lock the filesystem and force it into a consistent state
1619 * @sb: the super to lock
1621 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1622 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1625 * During this function, sb->s_writers.frozen goes through these values:
1627 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1629 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1630 * writes should be blocked, though page faults are still allowed. We wait for
1631 * all writes to complete and then proceed to the next stage.
1633 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1634 * but internal fs threads can still modify the filesystem (although they
1635 * should not dirty new pages or inodes), writeback can run etc. After waiting
1636 * for all running page faults we sync the filesystem which will clean all
1637 * dirty pages and inodes (no new dirty pages or inodes can be created when
1640 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1641 * modification are blocked (e.g. XFS preallocation truncation on inode
1642 * reclaim). This is usually implemented by blocking new transactions for
1643 * filesystems that have them and need this additional guard. After all
1644 * internal writers are finished we call ->freeze_fs() to finish filesystem
1645 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1646 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1648 * sb->s_writers.frozen is protected by sb->s_umount.
1650 int freeze_super(struct super_block
*sb
)
1654 atomic_inc(&sb
->s_active
);
1655 down_write(&sb
->s_umount
);
1656 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
) {
1657 deactivate_locked_super(sb
);
1661 if (!(sb
->s_flags
& SB_BORN
)) {
1662 up_write(&sb
->s_umount
);
1663 return 0; /* sic - it's "nothing to do" */
1666 if (sb_rdonly(sb
)) {
1667 /* Nothing to do really... */
1668 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1669 up_write(&sb
->s_umount
);
1673 sb
->s_writers
.frozen
= SB_FREEZE_WRITE
;
1674 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1675 up_write(&sb
->s_umount
);
1676 sb_wait_write(sb
, SB_FREEZE_WRITE
);
1677 down_write(&sb
->s_umount
);
1679 /* Now we go and block page faults... */
1680 sb
->s_writers
.frozen
= SB_FREEZE_PAGEFAULT
;
1681 sb_wait_write(sb
, SB_FREEZE_PAGEFAULT
);
1683 /* All writers are done so after syncing there won't be dirty data */
1684 ret
= sync_filesystem(sb
);
1686 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1687 sb_freeze_unlock(sb
, SB_FREEZE_PAGEFAULT
);
1688 wake_up(&sb
->s_writers
.wait_unfrozen
);
1689 deactivate_locked_super(sb
);
1693 /* Now wait for internal filesystem counter */
1694 sb
->s_writers
.frozen
= SB_FREEZE_FS
;
1695 sb_wait_write(sb
, SB_FREEZE_FS
);
1697 if (sb
->s_op
->freeze_fs
) {
1698 ret
= sb
->s_op
->freeze_fs(sb
);
1701 "VFS:Filesystem freeze failed\n");
1702 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1703 sb_freeze_unlock(sb
, SB_FREEZE_FS
);
1704 wake_up(&sb
->s_writers
.wait_unfrozen
);
1705 deactivate_locked_super(sb
);
1710 * For debugging purposes so that fs can warn if it sees write activity
1711 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1713 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1714 lockdep_sb_freeze_release(sb
);
1715 up_write(&sb
->s_umount
);
1718 EXPORT_SYMBOL(freeze_super
);
1720 static int thaw_super_locked(struct super_block
*sb
)
1724 if (sb
->s_writers
.frozen
!= SB_FREEZE_COMPLETE
) {
1725 up_write(&sb
->s_umount
);
1729 if (sb_rdonly(sb
)) {
1730 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1734 lockdep_sb_freeze_acquire(sb
);
1736 if (sb
->s_op
->unfreeze_fs
) {
1737 error
= sb
->s_op
->unfreeze_fs(sb
);
1740 "VFS:Filesystem thaw failed\n");
1741 lockdep_sb_freeze_release(sb
);
1742 up_write(&sb
->s_umount
);
1747 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1748 sb_freeze_unlock(sb
, SB_FREEZE_FS
);
1750 wake_up(&sb
->s_writers
.wait_unfrozen
);
1751 deactivate_locked_super(sb
);
1756 * thaw_super -- unlock filesystem
1757 * @sb: the super to thaw
1759 * Unlocks the filesystem and marks it writeable again after freeze_super().
1761 int thaw_super(struct super_block
*sb
)
1763 down_write(&sb
->s_umount
);
1764 return thaw_super_locked(sb
);
1766 EXPORT_SYMBOL(thaw_super
);