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 put_user_ns(s
->s_user_ns
);
296 call_rcu(&s
->rcu
, destroy_super_rcu
);
301 * put_super - drop a temporary reference to superblock
302 * @sb: superblock in question
304 * Drops a temporary reference, frees superblock if there's no
307 void put_super(struct super_block
*sb
)
311 spin_unlock(&sb_lock
);
316 * deactivate_locked_super - drop an active reference to superblock
317 * @s: superblock to deactivate
319 * Drops an active reference to superblock, converting it into a temporary
320 * one if there is no other active references left. In that case we
321 * tell fs driver to shut it down and drop the temporary reference we
324 * Caller holds exclusive lock on superblock; that lock is released.
326 void deactivate_locked_super(struct super_block
*s
)
328 struct file_system_type
*fs
= s
->s_type
;
329 if (atomic_dec_and_test(&s
->s_active
)) {
330 unregister_shrinker(&s
->s_shrink
);
334 * Since list_lru_destroy() may sleep, we cannot call it from
335 * put_super(), where we hold the sb_lock. Therefore we destroy
336 * the lru lists right now.
338 list_lru_destroy(&s
->s_dentry_lru
);
339 list_lru_destroy(&s
->s_inode_lru
);
344 up_write(&s
->s_umount
);
348 EXPORT_SYMBOL(deactivate_locked_super
);
351 * deactivate_super - drop an active reference to superblock
352 * @s: superblock to deactivate
354 * Variant of deactivate_locked_super(), except that superblock is *not*
355 * locked by caller. If we are going to drop the final active reference,
356 * lock will be acquired prior to that.
358 void deactivate_super(struct super_block
*s
)
360 if (!atomic_add_unless(&s
->s_active
, -1, 1)) {
361 down_write(&s
->s_umount
);
362 deactivate_locked_super(s
);
366 EXPORT_SYMBOL(deactivate_super
);
369 * grab_super - acquire an active reference
370 * @s: reference we are trying to make active
372 * Tries to acquire an active reference. grab_super() is used when we
373 * had just found a superblock in super_blocks or fs_type->fs_supers
374 * and want to turn it into a full-blown active reference. grab_super()
375 * is called with sb_lock held and drops it. Returns 1 in case of
376 * success, 0 if we had failed (superblock contents was already dead or
377 * dying when grab_super() had been called). Note that this is only
378 * called for superblocks not in rundown mode (== ones still on ->fs_supers
379 * of their type), so increment of ->s_count is OK here.
381 static int grab_super(struct super_block
*s
) __releases(sb_lock
)
384 spin_unlock(&sb_lock
);
385 down_write(&s
->s_umount
);
386 if ((s
->s_flags
& SB_BORN
) && atomic_inc_not_zero(&s
->s_active
)) {
390 up_write(&s
->s_umount
);
396 * trylock_super - try to grab ->s_umount shared
397 * @sb: reference we are trying to grab
399 * Try to prevent fs shutdown. This is used in places where we
400 * cannot take an active reference but we need to ensure that the
401 * filesystem is not shut down while we are working on it. It returns
402 * false if we cannot acquire s_umount or if we lose the race and
403 * filesystem already got into shutdown, and returns true with the s_umount
404 * lock held in read mode in case of success. On successful return,
405 * the caller must drop the s_umount lock when done.
407 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
408 * The reason why it's safe is that we are OK with doing trylock instead
409 * of down_read(). There's a couple of places that are OK with that, but
410 * it's very much not a general-purpose interface.
412 bool trylock_super(struct super_block
*sb
)
414 if (down_read_trylock(&sb
->s_umount
)) {
415 if (!hlist_unhashed(&sb
->s_instances
) &&
416 sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
418 up_read(&sb
->s_umount
);
425 * retire_super - prevents superblock from being reused
426 * @sb: superblock to retire
428 * The function marks superblock to be ignored in superblock test, which
429 * prevents it from being reused for any new mounts. If the superblock has
430 * a private bdi, it also unregisters it, but doesn't reduce the refcount
431 * of the superblock to prevent potential races. The refcount is reduced
432 * by generic_shutdown_super(). The function can not be called
433 * concurrently with generic_shutdown_super(). It is safe to call the
434 * function multiple times, subsequent calls have no effect.
436 * The marker will affect the re-use only for block-device-based
437 * superblocks. Other superblocks will still get marked if this function
438 * is used, but that will not affect their reusability.
440 void retire_super(struct super_block
*sb
)
442 WARN_ON(!sb
->s_bdev
);
443 down_write(&sb
->s_umount
);
444 if (sb
->s_iflags
& SB_I_PERSB_BDI
) {
445 bdi_unregister(sb
->s_bdi
);
446 sb
->s_iflags
&= ~SB_I_PERSB_BDI
;
448 sb
->s_iflags
|= SB_I_RETIRED
;
449 up_write(&sb
->s_umount
);
451 EXPORT_SYMBOL(retire_super
);
454 * generic_shutdown_super - common helper for ->kill_sb()
455 * @sb: superblock to kill
457 * generic_shutdown_super() does all fs-independent work on superblock
458 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
459 * that need destruction out of superblock, call generic_shutdown_super()
460 * and release aforementioned objects. Note: dentries and inodes _are_
461 * taken care of and do not need specific handling.
463 * Upon calling this function, the filesystem may no longer alter or
464 * rearrange the set of dentries belonging to this super_block, nor may it
465 * change the attachments of dentries to inodes.
467 void generic_shutdown_super(struct super_block
*sb
)
469 const struct super_operations
*sop
= sb
->s_op
;
472 shrink_dcache_for_umount(sb
);
474 sb
->s_flags
&= ~SB_ACTIVE
;
476 cgroup_writeback_umount();
478 /* evict all inodes with zero refcount */
480 /* only nonzero refcount inodes can have marks */
481 fsnotify_sb_delete(sb
);
482 fscrypt_destroy_keyring(sb
);
483 security_sb_delete(sb
);
485 if (sb
->s_dio_done_wq
) {
486 destroy_workqueue(sb
->s_dio_done_wq
);
487 sb
->s_dio_done_wq
= NULL
;
493 if (CHECK_DATA_CORRUPTION(!list_empty(&sb
->s_inodes
),
494 "VFS: Busy inodes after unmount of %s (%s)",
495 sb
->s_id
, sb
->s_type
->name
)) {
497 * Adding a proper bailout path here would be hard, but
498 * we can at least make it more likely that a later
499 * iput_final() or such crashes cleanly.
503 spin_lock(&sb
->s_inode_list_lock
);
504 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
505 inode
->i_op
= VFS_PTR_POISON
;
506 inode
->i_sb
= VFS_PTR_POISON
;
507 inode
->i_mapping
= VFS_PTR_POISON
;
509 spin_unlock(&sb
->s_inode_list_lock
);
513 /* should be initialized for __put_super_and_need_restart() */
514 hlist_del_init(&sb
->s_instances
);
515 spin_unlock(&sb_lock
);
516 up_write(&sb
->s_umount
);
517 if (sb
->s_bdi
!= &noop_backing_dev_info
) {
518 if (sb
->s_iflags
& SB_I_PERSB_BDI
)
519 bdi_unregister(sb
->s_bdi
);
521 sb
->s_bdi
= &noop_backing_dev_info
;
525 EXPORT_SYMBOL(generic_shutdown_super
);
527 bool mount_capable(struct fs_context
*fc
)
529 if (!(fc
->fs_type
->fs_flags
& FS_USERNS_MOUNT
))
530 return capable(CAP_SYS_ADMIN
);
532 return ns_capable(fc
->user_ns
, CAP_SYS_ADMIN
);
536 * sget_fc - Find or create a superblock
537 * @fc: Filesystem context.
538 * @test: Comparison callback
539 * @set: Setup callback
541 * Find or create a superblock using the parameters stored in the filesystem
542 * context and the two callback functions.
544 * If an extant superblock is matched, then that will be returned with an
545 * elevated reference count that the caller must transfer or discard.
547 * If no match is made, a new superblock will be allocated and basic
548 * initialisation will be performed (s_type, s_fs_info and s_id will be set and
549 * the set() callback will be invoked), the superblock will be published and it
550 * will be returned in a partially constructed state with SB_BORN and SB_ACTIVE
553 struct super_block
*sget_fc(struct fs_context
*fc
,
554 int (*test
)(struct super_block
*, struct fs_context
*),
555 int (*set
)(struct super_block
*, struct fs_context
*))
557 struct super_block
*s
= NULL
;
558 struct super_block
*old
;
559 struct user_namespace
*user_ns
= fc
->global
? &init_user_ns
: fc
->user_ns
;
565 hlist_for_each_entry(old
, &fc
->fs_type
->fs_supers
, s_instances
) {
567 goto share_extant_sb
;
571 spin_unlock(&sb_lock
);
572 s
= alloc_super(fc
->fs_type
, fc
->sb_flags
, user_ns
);
574 return ERR_PTR(-ENOMEM
);
578 s
->s_fs_info
= fc
->s_fs_info
;
582 spin_unlock(&sb_lock
);
583 destroy_unused_super(s
);
586 fc
->s_fs_info
= NULL
;
587 s
->s_type
= fc
->fs_type
;
588 s
->s_iflags
|= fc
->s_iflags
;
589 strlcpy(s
->s_id
, s
->s_type
->name
, sizeof(s
->s_id
));
590 list_add_tail(&s
->s_list
, &super_blocks
);
591 hlist_add_head(&s
->s_instances
, &s
->s_type
->fs_supers
);
592 spin_unlock(&sb_lock
);
593 get_filesystem(s
->s_type
);
594 register_shrinker_prepared(&s
->s_shrink
);
598 if (user_ns
!= old
->s_user_ns
) {
599 spin_unlock(&sb_lock
);
600 destroy_unused_super(s
);
601 return ERR_PTR(-EBUSY
);
603 if (!grab_super(old
))
605 destroy_unused_super(s
);
608 EXPORT_SYMBOL(sget_fc
);
611 * sget - find or create a superblock
612 * @type: filesystem type superblock should belong to
613 * @test: comparison callback
614 * @set: setup callback
615 * @flags: mount flags
616 * @data: argument to each of them
618 struct super_block
*sget(struct file_system_type
*type
,
619 int (*test
)(struct super_block
*,void *),
620 int (*set
)(struct super_block
*,void *),
624 struct user_namespace
*user_ns
= current_user_ns();
625 struct super_block
*s
= NULL
;
626 struct super_block
*old
;
629 /* We don't yet pass the user namespace of the parent
630 * mount through to here so always use &init_user_ns
631 * until that changes.
633 if (flags
& SB_SUBMOUNT
)
634 user_ns
= &init_user_ns
;
639 hlist_for_each_entry(old
, &type
->fs_supers
, s_instances
) {
640 if (!test(old
, data
))
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
);
654 spin_unlock(&sb_lock
);
655 s
= alloc_super(type
, (flags
& ~SB_SUBMOUNT
), user_ns
);
657 return ERR_PTR(-ENOMEM
);
663 spin_unlock(&sb_lock
);
664 destroy_unused_super(s
);
668 strlcpy(s
->s_id
, type
->name
, sizeof(s
->s_id
));
669 list_add_tail(&s
->s_list
, &super_blocks
);
670 hlist_add_head(&s
->s_instances
, &type
->fs_supers
);
671 spin_unlock(&sb_lock
);
672 get_filesystem(type
);
673 register_shrinker_prepared(&s
->s_shrink
);
678 void drop_super(struct super_block
*sb
)
680 up_read(&sb
->s_umount
);
684 EXPORT_SYMBOL(drop_super
);
686 void drop_super_exclusive(struct super_block
*sb
)
688 up_write(&sb
->s_umount
);
691 EXPORT_SYMBOL(drop_super_exclusive
);
693 static void __iterate_supers(void (*f
)(struct super_block
*))
695 struct super_block
*sb
, *p
= NULL
;
698 list_for_each_entry(sb
, &super_blocks
, s_list
) {
699 if (hlist_unhashed(&sb
->s_instances
))
702 spin_unlock(&sb_lock
);
713 spin_unlock(&sb_lock
);
716 * iterate_supers - call function for all active superblocks
717 * @f: function to call
718 * @arg: argument to pass to it
720 * Scans the superblock list and calls given function, passing it
721 * locked superblock and given argument.
723 void iterate_supers(void (*f
)(struct super_block
*, void *), void *arg
)
725 struct super_block
*sb
, *p
= NULL
;
728 list_for_each_entry(sb
, &super_blocks
, s_list
) {
729 if (hlist_unhashed(&sb
->s_instances
))
732 spin_unlock(&sb_lock
);
734 down_read(&sb
->s_umount
);
735 if (sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
737 up_read(&sb
->s_umount
);
746 spin_unlock(&sb_lock
);
750 * iterate_supers_type - call function for superblocks of given type
752 * @f: function to call
753 * @arg: argument to pass to it
755 * Scans the superblock list and calls given function, passing it
756 * locked superblock and given argument.
758 void iterate_supers_type(struct file_system_type
*type
,
759 void (*f
)(struct super_block
*, void *), void *arg
)
761 struct super_block
*sb
, *p
= NULL
;
764 hlist_for_each_entry(sb
, &type
->fs_supers
, s_instances
) {
766 spin_unlock(&sb_lock
);
768 down_read(&sb
->s_umount
);
769 if (sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
771 up_read(&sb
->s_umount
);
780 spin_unlock(&sb_lock
);
783 EXPORT_SYMBOL(iterate_supers_type
);
786 * get_super - get the superblock of a device
787 * @bdev: device to get the superblock for
789 * Scans the superblock list and finds the superblock of the file system
790 * mounted on the device given. %NULL is returned if no match is found.
792 struct super_block
*get_super(struct block_device
*bdev
)
794 struct super_block
*sb
;
801 list_for_each_entry(sb
, &super_blocks
, s_list
) {
802 if (hlist_unhashed(&sb
->s_instances
))
804 if (sb
->s_bdev
== bdev
) {
806 spin_unlock(&sb_lock
);
807 down_read(&sb
->s_umount
);
809 if (sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
811 up_read(&sb
->s_umount
);
812 /* nope, got unmounted */
818 spin_unlock(&sb_lock
);
823 * get_active_super - get an active reference to the superblock of a device
824 * @bdev: device to get the superblock for
826 * Scans the superblock list and finds the superblock of the file system
827 * mounted on the device given. Returns the superblock with an active
828 * reference or %NULL if none was found.
830 struct super_block
*get_active_super(struct block_device
*bdev
)
832 struct super_block
*sb
;
839 list_for_each_entry(sb
, &super_blocks
, s_list
) {
840 if (hlist_unhashed(&sb
->s_instances
))
842 if (sb
->s_bdev
== bdev
) {
845 up_write(&sb
->s_umount
);
849 spin_unlock(&sb_lock
);
853 struct super_block
*user_get_super(dev_t dev
, bool excl
)
855 struct super_block
*sb
;
859 list_for_each_entry(sb
, &super_blocks
, s_list
) {
860 if (hlist_unhashed(&sb
->s_instances
))
862 if (sb
->s_dev
== dev
) {
864 spin_unlock(&sb_lock
);
866 down_write(&sb
->s_umount
);
868 down_read(&sb
->s_umount
);
870 if (sb
->s_root
&& (sb
->s_flags
& SB_BORN
))
873 up_write(&sb
->s_umount
);
875 up_read(&sb
->s_umount
);
876 /* nope, got unmounted */
882 spin_unlock(&sb_lock
);
887 * reconfigure_super - asks filesystem to change superblock parameters
888 * @fc: The superblock and configuration
890 * Alters the configuration parameters of a live superblock.
892 int reconfigure_super(struct fs_context
*fc
)
894 struct super_block
*sb
= fc
->root
->d_sb
;
896 bool remount_ro
= false;
897 bool force
= fc
->sb_flags
& SB_FORCE
;
899 if (fc
->sb_flags_mask
& ~MS_RMT_MASK
)
901 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
)
904 retval
= security_sb_remount(sb
, fc
->security
);
908 if (fc
->sb_flags_mask
& SB_RDONLY
) {
910 if (!(fc
->sb_flags
& SB_RDONLY
) && sb
->s_bdev
&&
911 bdev_read_only(sb
->s_bdev
))
915 remount_ro
= (fc
->sb_flags
& SB_RDONLY
) && !sb_rdonly(sb
);
919 if (!hlist_empty(&sb
->s_pins
)) {
920 up_write(&sb
->s_umount
);
921 group_pin_kill(&sb
->s_pins
);
922 down_write(&sb
->s_umount
);
925 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
)
927 remount_ro
= !sb_rdonly(sb
);
930 shrink_dcache_sb(sb
);
932 /* If we are reconfiguring to RDONLY and current sb is read/write,
933 * make sure there are no files open for writing.
937 sb
->s_readonly_remount
= 1;
940 retval
= sb_prepare_remount_readonly(sb
);
946 if (fc
->ops
->reconfigure
) {
947 retval
= fc
->ops
->reconfigure(fc
);
950 goto cancel_readonly
;
951 /* If forced remount, go ahead despite any errors */
952 WARN(1, "forced remount of a %s fs returned %i\n",
953 sb
->s_type
->name
, retval
);
957 WRITE_ONCE(sb
->s_flags
, ((sb
->s_flags
& ~fc
->sb_flags_mask
) |
958 (fc
->sb_flags
& fc
->sb_flags_mask
)));
959 /* Needs to be ordered wrt mnt_is_readonly() */
961 sb
->s_readonly_remount
= 0;
964 * Some filesystems modify their metadata via some other path than the
965 * bdev buffer cache (eg. use a private mapping, or directories in
966 * pagecache, etc). Also file data modifications go via their own
967 * mappings. So If we try to mount readonly then copy the filesystem
968 * from bdev, we could get stale data, so invalidate it to give a best
969 * effort at coherency.
971 if (remount_ro
&& sb
->s_bdev
)
972 invalidate_bdev(sb
->s_bdev
);
976 sb
->s_readonly_remount
= 0;
980 static void do_emergency_remount_callback(struct super_block
*sb
)
982 down_write(&sb
->s_umount
);
983 if (sb
->s_root
&& sb
->s_bdev
&& (sb
->s_flags
& SB_BORN
) &&
985 struct fs_context
*fc
;
987 fc
= fs_context_for_reconfigure(sb
->s_root
,
988 SB_RDONLY
| SB_FORCE
, SB_RDONLY
);
990 if (parse_monolithic_mount_data(fc
, NULL
) == 0)
991 (void)reconfigure_super(fc
);
995 up_write(&sb
->s_umount
);
998 static void do_emergency_remount(struct work_struct
*work
)
1000 __iterate_supers(do_emergency_remount_callback
);
1002 printk("Emergency Remount complete\n");
1005 void emergency_remount(void)
1007 struct work_struct
*work
;
1009 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
1011 INIT_WORK(work
, do_emergency_remount
);
1012 schedule_work(work
);
1016 static void do_thaw_all_callback(struct super_block
*sb
)
1018 down_write(&sb
->s_umount
);
1019 if (sb
->s_root
&& sb
->s_flags
& SB_BORN
) {
1020 emergency_thaw_bdev(sb
);
1021 thaw_super_locked(sb
);
1023 up_write(&sb
->s_umount
);
1027 static void do_thaw_all(struct work_struct
*work
)
1029 __iterate_supers(do_thaw_all_callback
);
1031 printk(KERN_WARNING
"Emergency Thaw complete\n");
1035 * emergency_thaw_all -- forcibly thaw every frozen filesystem
1037 * Used for emergency unfreeze of all filesystems via SysRq
1039 void emergency_thaw_all(void)
1041 struct work_struct
*work
;
1043 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
1045 INIT_WORK(work
, do_thaw_all
);
1046 schedule_work(work
);
1050 static DEFINE_IDA(unnamed_dev_ida
);
1053 * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1054 * @p: Pointer to a dev_t.
1056 * Filesystems which don't use real block devices can call this function
1057 * to allocate a virtual block device.
1059 * Context: Any context. Frequently called while holding sb_lock.
1060 * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1061 * or -ENOMEM if memory allocation failed.
1063 int get_anon_bdev(dev_t
*p
)
1068 * Many userspace utilities consider an FSID of 0 invalid.
1069 * Always return at least 1 from get_anon_bdev.
1071 dev
= ida_alloc_range(&unnamed_dev_ida
, 1, (1 << MINORBITS
) - 1,
1081 EXPORT_SYMBOL(get_anon_bdev
);
1083 void free_anon_bdev(dev_t dev
)
1085 ida_free(&unnamed_dev_ida
, MINOR(dev
));
1087 EXPORT_SYMBOL(free_anon_bdev
);
1089 int set_anon_super(struct super_block
*s
, void *data
)
1091 return get_anon_bdev(&s
->s_dev
);
1093 EXPORT_SYMBOL(set_anon_super
);
1095 void kill_anon_super(struct super_block
*sb
)
1097 dev_t dev
= sb
->s_dev
;
1098 generic_shutdown_super(sb
);
1099 free_anon_bdev(dev
);
1101 EXPORT_SYMBOL(kill_anon_super
);
1103 void kill_litter_super(struct super_block
*sb
)
1106 d_genocide(sb
->s_root
);
1107 kill_anon_super(sb
);
1109 EXPORT_SYMBOL(kill_litter_super
);
1111 int set_anon_super_fc(struct super_block
*sb
, struct fs_context
*fc
)
1113 return set_anon_super(sb
, NULL
);
1115 EXPORT_SYMBOL(set_anon_super_fc
);
1117 static int test_keyed_super(struct super_block
*sb
, struct fs_context
*fc
)
1119 return sb
->s_fs_info
== fc
->s_fs_info
;
1122 static int test_single_super(struct super_block
*s
, struct fs_context
*fc
)
1127 static int vfs_get_super(struct fs_context
*fc
, bool reconf
,
1128 int (*test
)(struct super_block
*, struct fs_context
*),
1129 int (*fill_super
)(struct super_block
*sb
,
1130 struct fs_context
*fc
))
1132 struct super_block
*sb
;
1135 sb
= sget_fc(fc
, test
, set_anon_super_fc
);
1140 err
= fill_super(sb
, fc
);
1144 sb
->s_flags
|= SB_ACTIVE
;
1145 fc
->root
= dget(sb
->s_root
);
1147 fc
->root
= dget(sb
->s_root
);
1149 err
= reconfigure_super(fc
);
1161 deactivate_locked_super(sb
);
1165 int get_tree_nodev(struct fs_context
*fc
,
1166 int (*fill_super
)(struct super_block
*sb
,
1167 struct fs_context
*fc
))
1169 return vfs_get_super(fc
, false, NULL
, fill_super
);
1171 EXPORT_SYMBOL(get_tree_nodev
);
1173 int get_tree_single(struct fs_context
*fc
,
1174 int (*fill_super
)(struct super_block
*sb
,
1175 struct fs_context
*fc
))
1177 return vfs_get_super(fc
, false, test_single_super
, fill_super
);
1179 EXPORT_SYMBOL(get_tree_single
);
1181 int get_tree_single_reconf(struct fs_context
*fc
,
1182 int (*fill_super
)(struct super_block
*sb
,
1183 struct fs_context
*fc
))
1185 return vfs_get_super(fc
, true, test_single_super
, fill_super
);
1187 EXPORT_SYMBOL(get_tree_single_reconf
);
1189 int get_tree_keyed(struct fs_context
*fc
,
1190 int (*fill_super
)(struct super_block
*sb
,
1191 struct fs_context
*fc
),
1194 fc
->s_fs_info
= key
;
1195 return vfs_get_super(fc
, false, test_keyed_super
, fill_super
);
1197 EXPORT_SYMBOL(get_tree_keyed
);
1201 static int set_bdev_super(struct super_block
*s
, void *data
)
1204 s
->s_dev
= s
->s_bdev
->bd_dev
;
1205 s
->s_bdi
= bdi_get(s
->s_bdev
->bd_disk
->bdi
);
1207 if (bdev_stable_writes(s
->s_bdev
))
1208 s
->s_iflags
|= SB_I_STABLE_WRITES
;
1212 static int set_bdev_super_fc(struct super_block
*s
, struct fs_context
*fc
)
1214 return set_bdev_super(s
, fc
->sget_key
);
1217 static int test_bdev_super_fc(struct super_block
*s
, struct fs_context
*fc
)
1219 return !(s
->s_iflags
& SB_I_RETIRED
) && s
->s_bdev
== fc
->sget_key
;
1223 * get_tree_bdev - Get a superblock based on a single block device
1224 * @fc: The filesystem context holding the parameters
1225 * @fill_super: Helper to initialise a new superblock
1227 int get_tree_bdev(struct fs_context
*fc
,
1228 int (*fill_super
)(struct super_block
*,
1229 struct fs_context
*))
1231 struct block_device
*bdev
;
1232 struct super_block
*s
;
1233 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1236 if (!(fc
->sb_flags
& SB_RDONLY
))
1237 mode
|= FMODE_WRITE
;
1240 return invalf(fc
, "No source specified");
1242 bdev
= blkdev_get_by_path(fc
->source
, mode
, fc
->fs_type
);
1244 errorf(fc
, "%s: Can't open blockdev", fc
->source
);
1245 return PTR_ERR(bdev
);
1248 /* Once the superblock is inserted into the list by sget_fc(), s_umount
1249 * will protect the lockfs code from trying to start a snapshot while
1252 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
1253 if (bdev
->bd_fsfreeze_count
> 0) {
1254 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1255 warnf(fc
, "%pg: Can't mount, blockdev is frozen", bdev
);
1256 blkdev_put(bdev
, mode
);
1260 fc
->sb_flags
|= SB_NOSEC
;
1261 fc
->sget_key
= bdev
;
1262 s
= sget_fc(fc
, test_bdev_super_fc
, set_bdev_super_fc
);
1263 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1265 blkdev_put(bdev
, mode
);
1270 /* Don't summarily change the RO/RW state. */
1271 if ((fc
->sb_flags
^ s
->s_flags
) & SB_RDONLY
) {
1272 warnf(fc
, "%pg: Can't mount, would change RO state", bdev
);
1273 deactivate_locked_super(s
);
1274 blkdev_put(bdev
, mode
);
1279 * s_umount nests inside open_mutex during
1280 * __invalidate_device(). blkdev_put() acquires
1281 * open_mutex and can't be called under s_umount. Drop
1282 * s_umount temporarily. This is safe as we're
1283 * holding an active reference.
1285 up_write(&s
->s_umount
);
1286 blkdev_put(bdev
, mode
);
1287 down_write(&s
->s_umount
);
1290 snprintf(s
->s_id
, sizeof(s
->s_id
), "%pg", bdev
);
1291 shrinker_debugfs_rename(&s
->s_shrink
, "sb-%s:%s",
1292 fc
->fs_type
->name
, s
->s_id
);
1293 sb_set_blocksize(s
, block_size(bdev
));
1294 error
= fill_super(s
, fc
);
1296 deactivate_locked_super(s
);
1300 s
->s_flags
|= SB_ACTIVE
;
1305 fc
->root
= dget(s
->s_root
);
1308 EXPORT_SYMBOL(get_tree_bdev
);
1310 static int test_bdev_super(struct super_block
*s
, void *data
)
1312 return !(s
->s_iflags
& SB_I_RETIRED
) && (void *)s
->s_bdev
== data
;
1315 struct dentry
*mount_bdev(struct file_system_type
*fs_type
,
1316 int flags
, const char *dev_name
, void *data
,
1317 int (*fill_super
)(struct super_block
*, void *, int))
1319 struct block_device
*bdev
;
1320 struct super_block
*s
;
1321 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1324 if (!(flags
& SB_RDONLY
))
1325 mode
|= FMODE_WRITE
;
1327 bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
1329 return ERR_CAST(bdev
);
1332 * once the super is inserted into the list by sget, s_umount
1333 * will protect the lockfs code from trying to start a snapshot
1334 * while we are mounting
1336 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
1337 if (bdev
->bd_fsfreeze_count
> 0) {
1338 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1342 s
= sget(fs_type
, test_bdev_super
, set_bdev_super
, flags
| SB_NOSEC
,
1344 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1349 if ((flags
^ s
->s_flags
) & SB_RDONLY
) {
1350 deactivate_locked_super(s
);
1356 * s_umount nests inside open_mutex during
1357 * __invalidate_device(). blkdev_put() acquires
1358 * open_mutex and can't be called under s_umount. Drop
1359 * s_umount temporarily. This is safe as we're
1360 * holding an active reference.
1362 up_write(&s
->s_umount
);
1363 blkdev_put(bdev
, mode
);
1364 down_write(&s
->s_umount
);
1367 snprintf(s
->s_id
, sizeof(s
->s_id
), "%pg", bdev
);
1368 shrinker_debugfs_rename(&s
->s_shrink
, "sb-%s:%s",
1369 fs_type
->name
, s
->s_id
);
1370 sb_set_blocksize(s
, block_size(bdev
));
1371 error
= fill_super(s
, data
, flags
& SB_SILENT
? 1 : 0);
1373 deactivate_locked_super(s
);
1377 s
->s_flags
|= SB_ACTIVE
;
1381 return dget(s
->s_root
);
1386 blkdev_put(bdev
, mode
);
1388 return ERR_PTR(error
);
1390 EXPORT_SYMBOL(mount_bdev
);
1392 void kill_block_super(struct super_block
*sb
)
1394 struct block_device
*bdev
= sb
->s_bdev
;
1395 fmode_t mode
= sb
->s_mode
;
1397 bdev
->bd_super
= NULL
;
1398 generic_shutdown_super(sb
);
1399 sync_blockdev(bdev
);
1400 WARN_ON_ONCE(!(mode
& FMODE_EXCL
));
1401 blkdev_put(bdev
, mode
| FMODE_EXCL
);
1404 EXPORT_SYMBOL(kill_block_super
);
1407 struct dentry
*mount_nodev(struct file_system_type
*fs_type
,
1408 int flags
, void *data
,
1409 int (*fill_super
)(struct super_block
*, void *, int))
1412 struct super_block
*s
= sget(fs_type
, NULL
, set_anon_super
, flags
, NULL
);
1417 error
= fill_super(s
, data
, flags
& SB_SILENT
? 1 : 0);
1419 deactivate_locked_super(s
);
1420 return ERR_PTR(error
);
1422 s
->s_flags
|= SB_ACTIVE
;
1423 return dget(s
->s_root
);
1425 EXPORT_SYMBOL(mount_nodev
);
1427 int reconfigure_single(struct super_block
*s
,
1428 int flags
, void *data
)
1430 struct fs_context
*fc
;
1433 /* The caller really need to be passing fc down into mount_single(),
1434 * then a chunk of this can be removed. [Bollocks -- AV]
1435 * Better yet, reconfiguration shouldn't happen, but rather the second
1436 * mount should be rejected if the parameters are not compatible.
1438 fc
= fs_context_for_reconfigure(s
->s_root
, flags
, MS_RMT_MASK
);
1442 ret
= parse_monolithic_mount_data(fc
, data
);
1446 ret
= reconfigure_super(fc
);
1452 static int compare_single(struct super_block
*s
, void *p
)
1457 struct dentry
*mount_single(struct file_system_type
*fs_type
,
1458 int flags
, void *data
,
1459 int (*fill_super
)(struct super_block
*, void *, int))
1461 struct super_block
*s
;
1464 s
= sget(fs_type
, compare_single
, set_anon_super
, flags
, NULL
);
1468 error
= fill_super(s
, data
, flags
& SB_SILENT
? 1 : 0);
1470 s
->s_flags
|= SB_ACTIVE
;
1472 error
= reconfigure_single(s
, flags
, data
);
1474 if (unlikely(error
)) {
1475 deactivate_locked_super(s
);
1476 return ERR_PTR(error
);
1478 return dget(s
->s_root
);
1480 EXPORT_SYMBOL(mount_single
);
1483 * vfs_get_tree - Get the mountable root
1484 * @fc: The superblock configuration context.
1486 * The filesystem is invoked to get or create a superblock which can then later
1487 * be used for mounting. The filesystem places a pointer to the root to be
1488 * used for mounting in @fc->root.
1490 int vfs_get_tree(struct fs_context
*fc
)
1492 struct super_block
*sb
;
1498 /* Get the mountable root in fc->root, with a ref on the root and a ref
1499 * on the superblock.
1501 error
= fc
->ops
->get_tree(fc
);
1506 pr_err("Filesystem %s get_tree() didn't set fc->root\n",
1508 /* We don't know what the locking state of the superblock is -
1509 * if there is a superblock.
1514 sb
= fc
->root
->d_sb
;
1515 WARN_ON(!sb
->s_bdi
);
1518 * Write barrier is for super_cache_count(). We place it before setting
1519 * SB_BORN as the data dependency between the two functions is the
1520 * superblock structure contents that we just set up, not the SB_BORN
1524 sb
->s_flags
|= SB_BORN
;
1526 error
= security_sb_set_mnt_opts(sb
, fc
->security
, 0, NULL
);
1527 if (unlikely(error
)) {
1533 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1534 * but s_maxbytes was an unsigned long long for many releases. Throw
1535 * this warning for a little while to try and catch filesystems that
1536 * violate this rule.
1538 WARN((sb
->s_maxbytes
< 0), "%s set sb->s_maxbytes to "
1539 "negative value (%lld)\n", fc
->fs_type
->name
, sb
->s_maxbytes
);
1543 EXPORT_SYMBOL(vfs_get_tree
);
1546 * Setup private BDI for given superblock. It gets automatically cleaned up
1547 * in generic_shutdown_super().
1549 int super_setup_bdi_name(struct super_block
*sb
, char *fmt
, ...)
1551 struct backing_dev_info
*bdi
;
1555 bdi
= bdi_alloc(NUMA_NO_NODE
);
1559 va_start(args
, fmt
);
1560 err
= bdi_register_va(bdi
, fmt
, args
);
1566 WARN_ON(sb
->s_bdi
!= &noop_backing_dev_info
);
1568 sb
->s_iflags
|= SB_I_PERSB_BDI
;
1572 EXPORT_SYMBOL(super_setup_bdi_name
);
1575 * Setup private BDI for given superblock. I gets automatically cleaned up
1576 * in generic_shutdown_super().
1578 int super_setup_bdi(struct super_block
*sb
)
1580 static atomic_long_t bdi_seq
= ATOMIC_LONG_INIT(0);
1582 return super_setup_bdi_name(sb
, "%.28s-%ld", sb
->s_type
->name
,
1583 atomic_long_inc_return(&bdi_seq
));
1585 EXPORT_SYMBOL(super_setup_bdi
);
1588 * sb_wait_write - wait until all writers to given file system finish
1589 * @sb: the super for which we wait
1590 * @level: type of writers we wait for (normal vs page fault)
1592 * This function waits until there are no writers of given type to given file
1595 static void sb_wait_write(struct super_block
*sb
, int level
)
1597 percpu_down_write(sb
->s_writers
.rw_sem
+ level
-1);
1601 * We are going to return to userspace and forget about these locks, the
1602 * ownership goes to the caller of thaw_super() which does unlock().
1604 static void lockdep_sb_freeze_release(struct super_block
*sb
)
1608 for (level
= SB_FREEZE_LEVELS
- 1; level
>= 0; level
--)
1609 percpu_rwsem_release(sb
->s_writers
.rw_sem
+ level
, 0, _THIS_IP_
);
1613 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1615 static void lockdep_sb_freeze_acquire(struct super_block
*sb
)
1619 for (level
= 0; level
< SB_FREEZE_LEVELS
; ++level
)
1620 percpu_rwsem_acquire(sb
->s_writers
.rw_sem
+ level
, 0, _THIS_IP_
);
1623 static void sb_freeze_unlock(struct super_block
*sb
, int level
)
1625 for (level
--; level
>= 0; level
--)
1626 percpu_up_write(sb
->s_writers
.rw_sem
+ level
);
1630 * freeze_super - lock the filesystem and force it into a consistent state
1631 * @sb: the super to lock
1633 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1634 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1637 * During this function, sb->s_writers.frozen goes through these values:
1639 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1641 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1642 * writes should be blocked, though page faults are still allowed. We wait for
1643 * all writes to complete and then proceed to the next stage.
1645 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1646 * but internal fs threads can still modify the filesystem (although they
1647 * should not dirty new pages or inodes), writeback can run etc. After waiting
1648 * for all running page faults we sync the filesystem which will clean all
1649 * dirty pages and inodes (no new dirty pages or inodes can be created when
1652 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1653 * modification are blocked (e.g. XFS preallocation truncation on inode
1654 * reclaim). This is usually implemented by blocking new transactions for
1655 * filesystems that have them and need this additional guard. After all
1656 * internal writers are finished we call ->freeze_fs() to finish filesystem
1657 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1658 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1660 * sb->s_writers.frozen is protected by sb->s_umount.
1662 int freeze_super(struct super_block
*sb
)
1666 atomic_inc(&sb
->s_active
);
1667 down_write(&sb
->s_umount
);
1668 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
) {
1669 deactivate_locked_super(sb
);
1673 if (!(sb
->s_flags
& SB_BORN
)) {
1674 up_write(&sb
->s_umount
);
1675 return 0; /* sic - it's "nothing to do" */
1678 if (sb_rdonly(sb
)) {
1679 /* Nothing to do really... */
1680 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1681 up_write(&sb
->s_umount
);
1685 sb
->s_writers
.frozen
= SB_FREEZE_WRITE
;
1686 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1687 up_write(&sb
->s_umount
);
1688 sb_wait_write(sb
, SB_FREEZE_WRITE
);
1689 down_write(&sb
->s_umount
);
1691 /* Now we go and block page faults... */
1692 sb
->s_writers
.frozen
= SB_FREEZE_PAGEFAULT
;
1693 sb_wait_write(sb
, SB_FREEZE_PAGEFAULT
);
1695 /* All writers are done so after syncing there won't be dirty data */
1696 ret
= sync_filesystem(sb
);
1698 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1699 sb_freeze_unlock(sb
, SB_FREEZE_PAGEFAULT
);
1700 wake_up(&sb
->s_writers
.wait_unfrozen
);
1701 deactivate_locked_super(sb
);
1705 /* Now wait for internal filesystem counter */
1706 sb
->s_writers
.frozen
= SB_FREEZE_FS
;
1707 sb_wait_write(sb
, SB_FREEZE_FS
);
1709 if (sb
->s_op
->freeze_fs
) {
1710 ret
= sb
->s_op
->freeze_fs(sb
);
1713 "VFS:Filesystem freeze failed\n");
1714 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1715 sb_freeze_unlock(sb
, SB_FREEZE_FS
);
1716 wake_up(&sb
->s_writers
.wait_unfrozen
);
1717 deactivate_locked_super(sb
);
1722 * For debugging purposes so that fs can warn if it sees write activity
1723 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1725 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1726 lockdep_sb_freeze_release(sb
);
1727 up_write(&sb
->s_umount
);
1730 EXPORT_SYMBOL(freeze_super
);
1732 static int thaw_super_locked(struct super_block
*sb
)
1736 if (sb
->s_writers
.frozen
!= SB_FREEZE_COMPLETE
) {
1737 up_write(&sb
->s_umount
);
1741 if (sb_rdonly(sb
)) {
1742 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1746 lockdep_sb_freeze_acquire(sb
);
1748 if (sb
->s_op
->unfreeze_fs
) {
1749 error
= sb
->s_op
->unfreeze_fs(sb
);
1752 "VFS:Filesystem thaw failed\n");
1753 lockdep_sb_freeze_release(sb
);
1754 up_write(&sb
->s_umount
);
1759 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1760 sb_freeze_unlock(sb
, SB_FREEZE_FS
);
1762 wake_up(&sb
->s_writers
.wait_unfrozen
);
1763 deactivate_locked_super(sb
);
1768 * thaw_super -- unlock filesystem
1769 * @sb: the super to thaw
1771 * Unlocks the filesystem and marks it writeable again after freeze_super().
1773 int thaw_super(struct super_block
*sb
)
1775 down_write(&sb
->s_umount
);
1776 return thaw_super_locked(sb
);
1778 EXPORT_SYMBOL(thaw_super
);
1781 * Create workqueue for deferred direct IO completions. We allocate the
1782 * workqueue when it's first needed. This avoids creating workqueue for
1783 * filesystems that don't need it and also allows us to create the workqueue
1784 * late enough so the we can include s_id in the name of the workqueue.
1786 int sb_init_dio_done_wq(struct super_block
*sb
)
1788 struct workqueue_struct
*old
;
1789 struct workqueue_struct
*wq
= alloc_workqueue("dio/%s",
1795 * This has to be atomic as more DIOs can race to create the workqueue
1797 old
= cmpxchg(&sb
->s_dio_done_wq
, NULL
, wq
);
1798 /* Someone created workqueue before us? Free ours... */
1800 destroy_workqueue(wq
);