4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * super.c contains code to handle: - mount structures
8 * - filesystem drivers list
10 * - umount system call
13 * GK 2/5/95 - Changed to support mounting the root fs via NFS
15 * Added kerneld support: Jacques Gelinas and Bjorn Ekwall
16 * Added change_root: Werner Almesberger & Hans Lermen, Feb '96
17 * Added options to /proc/mounts:
18 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
19 * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
20 * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
23 #include <linux/export.h>
24 #include <linux/slab.h>
25 #include <linux/acct.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/cleancache.h>
35 #include <linux/fsnotify.h>
36 #include <linux/lockdep.h>
40 LIST_HEAD(super_blocks
);
41 DEFINE_SPINLOCK(sb_lock
);
43 static char *sb_writers_name
[SB_FREEZE_LEVELS
] = {
50 * One thing we have to be careful of with a per-sb shrinker is that we don't
51 * drop the last active reference to the superblock from within the shrinker.
52 * If that happens we could trigger unregistering the shrinker from within the
53 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
54 * take a passive reference to the superblock to avoid this from occurring.
56 static int prune_super(struct shrinker
*shrink
, struct shrink_control
*sc
)
58 struct super_block
*sb
;
62 sb
= container_of(shrink
, struct super_block
, s_shrink
);
65 * Deadlock avoidance. We may hold various FS locks, and we don't want
66 * to recurse into the FS that called us in clear_inode() and friends..
68 if (sc
->nr_to_scan
&& !(sc
->gfp_mask
& __GFP_FS
))
71 if (!grab_super_passive(sb
))
72 return !sc
->nr_to_scan
? 0 : -1;
74 if (sb
->s_op
&& sb
->s_op
->nr_cached_objects
)
75 fs_objects
= sb
->s_op
->nr_cached_objects(sb
);
77 total_objects
= sb
->s_nr_dentry_unused
+
78 sb
->s_nr_inodes_unused
+ fs_objects
+ 1;
84 /* proportion the scan between the caches */
85 dentries
= (sc
->nr_to_scan
* sb
->s_nr_dentry_unused
) /
87 inodes
= (sc
->nr_to_scan
* sb
->s_nr_inodes_unused
) /
90 fs_objects
= (sc
->nr_to_scan
* fs_objects
) /
93 * prune the dcache first as the icache is pinned by it, then
94 * prune the icache, followed by the filesystem specific caches
96 prune_dcache_sb(sb
, dentries
);
97 prune_icache_sb(sb
, inodes
);
99 if (fs_objects
&& sb
->s_op
->free_cached_objects
) {
100 sb
->s_op
->free_cached_objects(sb
, fs_objects
);
101 fs_objects
= sb
->s_op
->nr_cached_objects(sb
);
103 total_objects
= sb
->s_nr_dentry_unused
+
104 sb
->s_nr_inodes_unused
+ fs_objects
;
107 total_objects
= (total_objects
/ 100) * sysctl_vfs_cache_pressure
;
109 return total_objects
;
112 static int init_sb_writers(struct super_block
*s
, struct file_system_type
*type
)
117 for (i
= 0; i
< SB_FREEZE_LEVELS
; i
++) {
118 err
= percpu_counter_init(&s
->s_writers
.counter
[i
], 0);
121 lockdep_init_map(&s
->s_writers
.lock_map
[i
], sb_writers_name
[i
],
122 &type
->s_writers_key
[i
], 0);
124 init_waitqueue_head(&s
->s_writers
.wait
);
125 init_waitqueue_head(&s
->s_writers
.wait_unfrozen
);
129 percpu_counter_destroy(&s
->s_writers
.counter
[i
]);
133 static void destroy_sb_writers(struct super_block
*s
)
137 for (i
= 0; i
< SB_FREEZE_LEVELS
; i
++)
138 percpu_counter_destroy(&s
->s_writers
.counter
[i
]);
142 * alloc_super - create new superblock
143 * @type: filesystem type superblock should belong to
144 * @flags: the mount flags
146 * Allocates and initializes a new &struct super_block. alloc_super()
147 * returns a pointer new superblock or %NULL if allocation had failed.
149 static struct super_block
*alloc_super(struct file_system_type
*type
, int flags
)
151 struct super_block
*s
= kzalloc(sizeof(struct super_block
), GFP_USER
);
152 static const struct super_operations default_op
;
155 if (security_sb_alloc(s
)) {
157 * We cannot call security_sb_free() without
158 * security_sb_alloc() succeeding. So bail out manually
165 s
->s_files
= alloc_percpu(struct list_head
);
171 for_each_possible_cpu(i
)
172 INIT_LIST_HEAD(per_cpu_ptr(s
->s_files
, i
));
175 INIT_LIST_HEAD(&s
->s_files
);
177 if (init_sb_writers(s
, type
))
180 s
->s_bdi
= &default_backing_dev_info
;
181 INIT_HLIST_NODE(&s
->s_instances
);
182 INIT_HLIST_BL_HEAD(&s
->s_anon
);
183 INIT_LIST_HEAD(&s
->s_inodes
);
184 INIT_LIST_HEAD(&s
->s_dentry_lru
);
185 INIT_LIST_HEAD(&s
->s_inode_lru
);
186 spin_lock_init(&s
->s_inode_lru_lock
);
187 INIT_LIST_HEAD(&s
->s_mounts
);
188 init_rwsem(&s
->s_umount
);
189 mutex_init(&s
->s_lock
);
190 lockdep_set_class(&s
->s_umount
, &type
->s_umount_key
);
192 * The locking rules for s_lock are up to the
193 * filesystem. For example ext3fs has different
194 * lock ordering than usbfs:
196 lockdep_set_class(&s
->s_lock
, &type
->s_lock_key
);
198 * sget() can have s_umount recursion.
200 * When it cannot find a suitable sb, it allocates a new
201 * one (this one), and tries again to find a suitable old
204 * In case that succeeds, it will acquire the s_umount
205 * lock of the old one. Since these are clearly distrinct
206 * locks, and this object isn't exposed yet, there's no
209 * Annotate this by putting this lock in a different
212 down_write_nested(&s
->s_umount
, SINGLE_DEPTH_NESTING
);
214 atomic_set(&s
->s_active
, 1);
215 mutex_init(&s
->s_vfs_rename_mutex
);
216 lockdep_set_class(&s
->s_vfs_rename_mutex
, &type
->s_vfs_rename_key
);
217 mutex_init(&s
->s_dquot
.dqio_mutex
);
218 mutex_init(&s
->s_dquot
.dqonoff_mutex
);
219 init_rwsem(&s
->s_dquot
.dqptr_sem
);
220 init_waitqueue_head(&s
->s_wait_unfrozen
);
221 s
->s_maxbytes
= MAX_NON_LFS
;
222 s
->s_op
= &default_op
;
223 s
->s_time_gran
= 1000000000;
224 s
->cleancache_poolid
= -1;
226 s
->s_shrink
.seeks
= DEFAULT_SEEKS
;
227 s
->s_shrink
.shrink
= prune_super
;
228 s
->s_shrink
.batch
= 1024;
236 free_percpu(s
->s_files
);
238 destroy_sb_writers(s
);
245 * destroy_super - frees a superblock
246 * @s: superblock to free
248 * Frees a superblock.
250 static inline void destroy_super(struct super_block
*s
)
253 free_percpu(s
->s_files
);
255 destroy_sb_writers(s
);
257 WARN_ON(!list_empty(&s
->s_mounts
));
263 /* Superblock refcounting */
266 * Drop a superblock's refcount. The caller must hold sb_lock.
268 static void __put_super(struct super_block
*sb
)
270 if (!--sb
->s_count
) {
271 list_del_init(&sb
->s_list
);
277 * put_super - drop a temporary reference to superblock
278 * @sb: superblock in question
280 * Drops a temporary reference, frees superblock if there's no
283 static void put_super(struct super_block
*sb
)
287 spin_unlock(&sb_lock
);
292 * deactivate_locked_super - drop an active reference to superblock
293 * @s: superblock to deactivate
295 * Drops an active reference to superblock, converting it into a temprory
296 * one if there is no other active references left. In that case we
297 * tell fs driver to shut it down and drop the temporary reference we
300 * Caller holds exclusive lock on superblock; that lock is released.
302 void deactivate_locked_super(struct super_block
*s
)
304 struct file_system_type
*fs
= s
->s_type
;
305 if (atomic_dec_and_test(&s
->s_active
)) {
306 cleancache_invalidate_fs(s
);
309 /* caches are now gone, we can safely kill the shrinker now */
310 unregister_shrinker(&s
->s_shrink
);
313 * We need to call rcu_barrier so all the delayed rcu free
314 * inodes are flushed before we release the fs module.
320 up_write(&s
->s_umount
);
324 EXPORT_SYMBOL(deactivate_locked_super
);
327 * deactivate_super - drop an active reference to superblock
328 * @s: superblock to deactivate
330 * Variant of deactivate_locked_super(), except that superblock is *not*
331 * locked by caller. If we are going to drop the final active reference,
332 * lock will be acquired prior to that.
334 void deactivate_super(struct super_block
*s
)
336 if (!atomic_add_unless(&s
->s_active
, -1, 1)) {
337 down_write(&s
->s_umount
);
338 deactivate_locked_super(s
);
342 EXPORT_SYMBOL(deactivate_super
);
345 * grab_super - acquire an active reference
346 * @s: reference we are trying to make active
348 * Tries to acquire an active reference. grab_super() is used when we
349 * had just found a superblock in super_blocks or fs_type->fs_supers
350 * and want to turn it into a full-blown active reference. grab_super()
351 * is called with sb_lock held and drops it. Returns 1 in case of
352 * success, 0 if we had failed (superblock contents was already dead or
353 * dying when grab_super() had been called).
355 static int grab_super(struct super_block
*s
) __releases(sb_lock
)
357 if (atomic_inc_not_zero(&s
->s_active
)) {
358 spin_unlock(&sb_lock
);
361 /* it's going away */
363 spin_unlock(&sb_lock
);
364 /* wait for it to die */
365 down_write(&s
->s_umount
);
366 up_write(&s
->s_umount
);
372 * grab_super_passive - acquire a passive reference
373 * @s: reference we are trying to grab
375 * Tries to acquire a passive reference. This is used in places where we
376 * cannot take an active reference but we need to ensure that the
377 * superblock does not go away while we are working on it. It returns
378 * false if a reference was not gained, and returns true with the s_umount
379 * lock held in read mode if a reference is gained. On successful return,
380 * the caller must drop the s_umount lock and the passive reference when
383 bool grab_super_passive(struct super_block
*sb
)
386 if (hlist_unhashed(&sb
->s_instances
)) {
387 spin_unlock(&sb_lock
);
392 spin_unlock(&sb_lock
);
394 if (down_read_trylock(&sb
->s_umount
)) {
395 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
397 up_read(&sb
->s_umount
);
405 * Superblock locking. We really ought to get rid of these two.
407 void lock_super(struct super_block
* sb
)
409 mutex_lock(&sb
->s_lock
);
412 void unlock_super(struct super_block
* sb
)
414 mutex_unlock(&sb
->s_lock
);
417 EXPORT_SYMBOL(lock_super
);
418 EXPORT_SYMBOL(unlock_super
);
421 * generic_shutdown_super - common helper for ->kill_sb()
422 * @sb: superblock to kill
424 * generic_shutdown_super() does all fs-independent work on superblock
425 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
426 * that need destruction out of superblock, call generic_shutdown_super()
427 * and release aforementioned objects. Note: dentries and inodes _are_
428 * taken care of and do not need specific handling.
430 * Upon calling this function, the filesystem may no longer alter or
431 * rearrange the set of dentries belonging to this super_block, nor may it
432 * change the attachments of dentries to inodes.
434 void generic_shutdown_super(struct super_block
*sb
)
436 const struct super_operations
*sop
= sb
->s_op
;
439 shrink_dcache_for_umount(sb
);
441 sb
->s_flags
&= ~MS_ACTIVE
;
443 fsnotify_unmount_inodes(&sb
->s_inodes
);
450 if (!list_empty(&sb
->s_inodes
)) {
451 printk("VFS: Busy inodes after unmount of %s. "
452 "Self-destruct in 5 seconds. Have a nice day...\n",
457 /* should be initialized for __put_super_and_need_restart() */
458 hlist_del_init(&sb
->s_instances
);
459 spin_unlock(&sb_lock
);
460 up_write(&sb
->s_umount
);
463 EXPORT_SYMBOL(generic_shutdown_super
);
466 * sget - find or create a superblock
467 * @type: filesystem type superblock should belong to
468 * @test: comparison callback
469 * @set: setup callback
470 * @flags: mount flags
471 * @data: argument to each of them
473 struct super_block
*sget(struct file_system_type
*type
,
474 int (*test
)(struct super_block
*,void *),
475 int (*set
)(struct super_block
*,void *),
479 struct super_block
*s
= NULL
;
480 struct hlist_node
*node
;
481 struct super_block
*old
;
487 hlist_for_each_entry(old
, node
, &type
->fs_supers
, s_instances
) {
488 if (!test(old
, data
))
490 if (!grab_super(old
))
493 up_write(&s
->s_umount
);
497 down_write(&old
->s_umount
);
498 if (unlikely(!(old
->s_flags
& MS_BORN
))) {
499 deactivate_locked_super(old
);
506 spin_unlock(&sb_lock
);
507 s
= alloc_super(type
, flags
);
509 return ERR_PTR(-ENOMEM
);
515 spin_unlock(&sb_lock
);
516 up_write(&s
->s_umount
);
521 strlcpy(s
->s_id
, type
->name
, sizeof(s
->s_id
));
522 list_add_tail(&s
->s_list
, &super_blocks
);
523 hlist_add_head(&s
->s_instances
, &type
->fs_supers
);
524 spin_unlock(&sb_lock
);
525 get_filesystem(type
);
526 register_shrinker(&s
->s_shrink
);
532 void drop_super(struct super_block
*sb
)
534 up_read(&sb
->s_umount
);
538 EXPORT_SYMBOL(drop_super
);
541 * sync_supers - helper for periodic superblock writeback
543 * Call the write_super method if present on all dirty superblocks in
544 * the system. This is for the periodic writeback used by most older
545 * filesystems. For data integrity superblock writeback use
546 * sync_filesystems() instead.
548 * Note: check the dirty flag before waiting, so we don't
549 * hold up the sync while mounting a device. (The newly
550 * mounted device won't need syncing.)
552 void sync_supers(void)
554 struct super_block
*sb
, *p
= NULL
;
557 list_for_each_entry(sb
, &super_blocks
, s_list
) {
558 if (hlist_unhashed(&sb
->s_instances
))
560 if (sb
->s_op
->write_super
&& sb
->s_dirt
) {
562 spin_unlock(&sb_lock
);
564 down_read(&sb
->s_umount
);
565 if (sb
->s_root
&& sb
->s_dirt
&& (sb
->s_flags
& MS_BORN
))
566 sb
->s_op
->write_super(sb
);
567 up_read(&sb
->s_umount
);
577 spin_unlock(&sb_lock
);
581 * iterate_supers - call function for all active superblocks
582 * @f: function to call
583 * @arg: argument to pass to it
585 * Scans the superblock list and calls given function, passing it
586 * locked superblock and given argument.
588 void iterate_supers(void (*f
)(struct super_block
*, void *), void *arg
)
590 struct super_block
*sb
, *p
= NULL
;
593 list_for_each_entry(sb
, &super_blocks
, s_list
) {
594 if (hlist_unhashed(&sb
->s_instances
))
597 spin_unlock(&sb_lock
);
599 down_read(&sb
->s_umount
);
600 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
602 up_read(&sb
->s_umount
);
611 spin_unlock(&sb_lock
);
615 * iterate_supers_type - call function for superblocks of given type
617 * @f: function to call
618 * @arg: argument to pass to it
620 * Scans the superblock list and calls given function, passing it
621 * locked superblock and given argument.
623 void iterate_supers_type(struct file_system_type
*type
,
624 void (*f
)(struct super_block
*, void *), void *arg
)
626 struct super_block
*sb
, *p
= NULL
;
627 struct hlist_node
*node
;
630 hlist_for_each_entry(sb
, node
, &type
->fs_supers
, s_instances
) {
632 spin_unlock(&sb_lock
);
634 down_read(&sb
->s_umount
);
635 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
637 up_read(&sb
->s_umount
);
646 spin_unlock(&sb_lock
);
649 EXPORT_SYMBOL(iterate_supers_type
);
652 * get_super - get the superblock of a device
653 * @bdev: device to get the superblock for
655 * Scans the superblock list and finds the superblock of the file system
656 * mounted on the device given. %NULL is returned if no match is found.
659 struct super_block
*get_super(struct block_device
*bdev
)
661 struct super_block
*sb
;
668 list_for_each_entry(sb
, &super_blocks
, s_list
) {
669 if (hlist_unhashed(&sb
->s_instances
))
671 if (sb
->s_bdev
== bdev
) {
673 spin_unlock(&sb_lock
);
674 down_read(&sb
->s_umount
);
676 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
678 up_read(&sb
->s_umount
);
679 /* nope, got unmounted */
685 spin_unlock(&sb_lock
);
689 EXPORT_SYMBOL(get_super
);
692 * get_super_thawed - get thawed superblock of a device
693 * @bdev: device to get the superblock for
695 * Scans the superblock list and finds the superblock of the file system
696 * mounted on the device. The superblock is returned once it is thawed
697 * (or immediately if it was not frozen). %NULL is returned if no match
700 struct super_block
*get_super_thawed(struct block_device
*bdev
)
703 struct super_block
*s
= get_super(bdev
);
704 if (!s
|| s
->s_writers
.frozen
== SB_UNFROZEN
)
706 up_read(&s
->s_umount
);
707 wait_event(s
->s_writers
.wait_unfrozen
,
708 s
->s_writers
.frozen
== SB_UNFROZEN
);
712 EXPORT_SYMBOL(get_super_thawed
);
715 * get_active_super - get an active reference to the superblock of a device
716 * @bdev: device to get the superblock for
718 * Scans the superblock list and finds the superblock of the file system
719 * mounted on the device given. Returns the superblock with an active
720 * reference or %NULL if none was found.
722 struct super_block
*get_active_super(struct block_device
*bdev
)
724 struct super_block
*sb
;
731 list_for_each_entry(sb
, &super_blocks
, s_list
) {
732 if (hlist_unhashed(&sb
->s_instances
))
734 if (sb
->s_bdev
== bdev
) {
735 if (grab_super(sb
)) /* drops sb_lock */
741 spin_unlock(&sb_lock
);
745 struct super_block
*user_get_super(dev_t dev
)
747 struct super_block
*sb
;
751 list_for_each_entry(sb
, &super_blocks
, s_list
) {
752 if (hlist_unhashed(&sb
->s_instances
))
754 if (sb
->s_dev
== dev
) {
756 spin_unlock(&sb_lock
);
757 down_read(&sb
->s_umount
);
759 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
761 up_read(&sb
->s_umount
);
762 /* nope, got unmounted */
768 spin_unlock(&sb_lock
);
773 * do_remount_sb - asks filesystem to change mount options.
774 * @sb: superblock in question
775 * @flags: numeric part of options
776 * @data: the rest of options
777 * @force: whether or not to force the change
779 * Alters the mount options of a mounted file system.
781 int do_remount_sb(struct super_block
*sb
, int flags
, void *data
, int force
)
786 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
)
790 if (!(flags
& MS_RDONLY
) && bdev_read_only(sb
->s_bdev
))
794 if (flags
& MS_RDONLY
)
796 shrink_dcache_sb(sb
);
799 remount_ro
= (flags
& MS_RDONLY
) && !(sb
->s_flags
& MS_RDONLY
);
801 /* If we are remounting RDONLY and current sb is read/write,
802 make sure there are no rw files opened */
807 retval
= sb_prepare_remount_readonly(sb
);
813 if (sb
->s_op
->remount_fs
) {
814 retval
= sb
->s_op
->remount_fs(sb
, &flags
, data
);
817 goto cancel_readonly
;
818 /* If forced remount, go ahead despite any errors */
819 WARN(1, "forced remount of a %s fs returned %i\n",
820 sb
->s_type
->name
, retval
);
823 sb
->s_flags
= (sb
->s_flags
& ~MS_RMT_MASK
) | (flags
& MS_RMT_MASK
);
824 /* Needs to be ordered wrt mnt_is_readonly() */
826 sb
->s_readonly_remount
= 0;
829 * Some filesystems modify their metadata via some other path than the
830 * bdev buffer cache (eg. use a private mapping, or directories in
831 * pagecache, etc). Also file data modifications go via their own
832 * mappings. So If we try to mount readonly then copy the filesystem
833 * from bdev, we could get stale data, so invalidate it to give a best
834 * effort at coherency.
836 if (remount_ro
&& sb
->s_bdev
)
837 invalidate_bdev(sb
->s_bdev
);
841 sb
->s_readonly_remount
= 0;
845 static void do_emergency_remount(struct work_struct
*work
)
847 struct super_block
*sb
, *p
= NULL
;
850 list_for_each_entry(sb
, &super_blocks
, s_list
) {
851 if (hlist_unhashed(&sb
->s_instances
))
854 spin_unlock(&sb_lock
);
855 down_write(&sb
->s_umount
);
856 if (sb
->s_root
&& sb
->s_bdev
&& (sb
->s_flags
& MS_BORN
) &&
857 !(sb
->s_flags
& MS_RDONLY
)) {
859 * What lock protects sb->s_flags??
861 do_remount_sb(sb
, MS_RDONLY
, NULL
, 1);
863 up_write(&sb
->s_umount
);
871 spin_unlock(&sb_lock
);
873 printk("Emergency Remount complete\n");
876 void emergency_remount(void)
878 struct work_struct
*work
;
880 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
882 INIT_WORK(work
, do_emergency_remount
);
888 * Unnamed block devices are dummy devices used by virtual
889 * filesystems which don't use real block-devices. -- jrs
892 static DEFINE_IDA(unnamed_dev_ida
);
893 static DEFINE_SPINLOCK(unnamed_dev_lock
);/* protects the above */
894 static int unnamed_dev_start
= 0; /* don't bother trying below it */
896 int get_anon_bdev(dev_t
*p
)
902 if (ida_pre_get(&unnamed_dev_ida
, GFP_ATOMIC
) == 0)
904 spin_lock(&unnamed_dev_lock
);
905 error
= ida_get_new_above(&unnamed_dev_ida
, unnamed_dev_start
, &dev
);
907 unnamed_dev_start
= dev
+ 1;
908 spin_unlock(&unnamed_dev_lock
);
909 if (error
== -EAGAIN
)
910 /* We raced and lost with another CPU. */
915 if ((dev
& MAX_ID_MASK
) == (1 << MINORBITS
)) {
916 spin_lock(&unnamed_dev_lock
);
917 ida_remove(&unnamed_dev_ida
, dev
);
918 if (unnamed_dev_start
> dev
)
919 unnamed_dev_start
= dev
;
920 spin_unlock(&unnamed_dev_lock
);
923 *p
= MKDEV(0, dev
& MINORMASK
);
926 EXPORT_SYMBOL(get_anon_bdev
);
928 void free_anon_bdev(dev_t dev
)
930 int slot
= MINOR(dev
);
931 spin_lock(&unnamed_dev_lock
);
932 ida_remove(&unnamed_dev_ida
, slot
);
933 if (slot
< unnamed_dev_start
)
934 unnamed_dev_start
= slot
;
935 spin_unlock(&unnamed_dev_lock
);
937 EXPORT_SYMBOL(free_anon_bdev
);
939 int set_anon_super(struct super_block
*s
, void *data
)
941 int error
= get_anon_bdev(&s
->s_dev
);
943 s
->s_bdi
= &noop_backing_dev_info
;
947 EXPORT_SYMBOL(set_anon_super
);
949 void kill_anon_super(struct super_block
*sb
)
951 dev_t dev
= sb
->s_dev
;
952 generic_shutdown_super(sb
);
956 EXPORT_SYMBOL(kill_anon_super
);
958 void kill_litter_super(struct super_block
*sb
)
961 d_genocide(sb
->s_root
);
965 EXPORT_SYMBOL(kill_litter_super
);
967 static int ns_test_super(struct super_block
*sb
, void *data
)
969 return sb
->s_fs_info
== data
;
972 static int ns_set_super(struct super_block
*sb
, void *data
)
974 sb
->s_fs_info
= data
;
975 return set_anon_super(sb
, NULL
);
978 struct dentry
*mount_ns(struct file_system_type
*fs_type
, int flags
,
979 void *data
, int (*fill_super
)(struct super_block
*, void *, int))
981 struct super_block
*sb
;
983 sb
= sget(fs_type
, ns_test_super
, ns_set_super
, flags
, data
);
989 err
= fill_super(sb
, data
, flags
& MS_SILENT
? 1 : 0);
991 deactivate_locked_super(sb
);
995 sb
->s_flags
|= MS_ACTIVE
;
998 return dget(sb
->s_root
);
1001 EXPORT_SYMBOL(mount_ns
);
1004 static int set_bdev_super(struct super_block
*s
, void *data
)
1007 s
->s_dev
= s
->s_bdev
->bd_dev
;
1010 * We set the bdi here to the queue backing, file systems can
1011 * overwrite this in ->fill_super()
1013 s
->s_bdi
= &bdev_get_queue(s
->s_bdev
)->backing_dev_info
;
1017 static int test_bdev_super(struct super_block
*s
, void *data
)
1019 return (void *)s
->s_bdev
== data
;
1022 struct dentry
*mount_bdev(struct file_system_type
*fs_type
,
1023 int flags
, const char *dev_name
, void *data
,
1024 int (*fill_super
)(struct super_block
*, void *, int))
1026 struct block_device
*bdev
;
1027 struct super_block
*s
;
1028 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1031 if (!(flags
& MS_RDONLY
))
1032 mode
|= FMODE_WRITE
;
1034 bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
1036 return ERR_CAST(bdev
);
1039 * once the super is inserted into the list by sget, s_umount
1040 * will protect the lockfs code from trying to start a snapshot
1041 * while we are mounting
1043 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
1044 if (bdev
->bd_fsfreeze_count
> 0) {
1045 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1049 s
= sget(fs_type
, test_bdev_super
, set_bdev_super
, flags
| MS_NOSEC
,
1051 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1056 if ((flags
^ s
->s_flags
) & MS_RDONLY
) {
1057 deactivate_locked_super(s
);
1063 * s_umount nests inside bd_mutex during
1064 * __invalidate_device(). blkdev_put() acquires
1065 * bd_mutex and can't be called under s_umount. Drop
1066 * s_umount temporarily. This is safe as we're
1067 * holding an active reference.
1069 up_write(&s
->s_umount
);
1070 blkdev_put(bdev
, mode
);
1071 down_write(&s
->s_umount
);
1073 char b
[BDEVNAME_SIZE
];
1076 strlcpy(s
->s_id
, bdevname(bdev
, b
), sizeof(s
->s_id
));
1077 sb_set_blocksize(s
, block_size(bdev
));
1078 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1080 deactivate_locked_super(s
);
1084 s
->s_flags
|= MS_ACTIVE
;
1088 return dget(s
->s_root
);
1093 blkdev_put(bdev
, mode
);
1095 return ERR_PTR(error
);
1097 EXPORT_SYMBOL(mount_bdev
);
1099 void kill_block_super(struct super_block
*sb
)
1101 struct block_device
*bdev
= sb
->s_bdev
;
1102 fmode_t mode
= sb
->s_mode
;
1104 bdev
->bd_super
= NULL
;
1105 generic_shutdown_super(sb
);
1106 sync_blockdev(bdev
);
1107 WARN_ON_ONCE(!(mode
& FMODE_EXCL
));
1108 blkdev_put(bdev
, mode
| FMODE_EXCL
);
1111 EXPORT_SYMBOL(kill_block_super
);
1114 struct dentry
*mount_nodev(struct file_system_type
*fs_type
,
1115 int flags
, void *data
,
1116 int (*fill_super
)(struct super_block
*, void *, int))
1119 struct super_block
*s
= sget(fs_type
, NULL
, set_anon_super
, flags
, NULL
);
1124 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1126 deactivate_locked_super(s
);
1127 return ERR_PTR(error
);
1129 s
->s_flags
|= MS_ACTIVE
;
1130 return dget(s
->s_root
);
1132 EXPORT_SYMBOL(mount_nodev
);
1134 static int compare_single(struct super_block
*s
, void *p
)
1139 struct dentry
*mount_single(struct file_system_type
*fs_type
,
1140 int flags
, void *data
,
1141 int (*fill_super
)(struct super_block
*, void *, int))
1143 struct super_block
*s
;
1146 s
= sget(fs_type
, compare_single
, set_anon_super
, flags
, NULL
);
1150 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1152 deactivate_locked_super(s
);
1153 return ERR_PTR(error
);
1155 s
->s_flags
|= MS_ACTIVE
;
1157 do_remount_sb(s
, flags
, data
, 0);
1159 return dget(s
->s_root
);
1161 EXPORT_SYMBOL(mount_single
);
1164 mount_fs(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
1166 struct dentry
*root
;
1167 struct super_block
*sb
;
1168 char *secdata
= NULL
;
1169 int error
= -ENOMEM
;
1171 if (data
&& !(type
->fs_flags
& FS_BINARY_MOUNTDATA
)) {
1172 secdata
= alloc_secdata();
1176 error
= security_sb_copy_data(data
, secdata
);
1178 goto out_free_secdata
;
1181 root
= type
->mount(type
, flags
, name
, data
);
1183 error
= PTR_ERR(root
);
1184 goto out_free_secdata
;
1188 WARN_ON(!sb
->s_bdi
);
1189 WARN_ON(sb
->s_bdi
== &default_backing_dev_info
);
1190 sb
->s_flags
|= MS_BORN
;
1192 error
= security_sb_kern_mount(sb
, flags
, secdata
);
1197 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1198 * but s_maxbytes was an unsigned long long for many releases. Throw
1199 * this warning for a little while to try and catch filesystems that
1200 * violate this rule.
1202 WARN((sb
->s_maxbytes
< 0), "%s set sb->s_maxbytes to "
1203 "negative value (%lld)\n", type
->name
, sb
->s_maxbytes
);
1205 up_write(&sb
->s_umount
);
1206 free_secdata(secdata
);
1210 deactivate_locked_super(sb
);
1212 free_secdata(secdata
);
1214 return ERR_PTR(error
);
1218 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1221 void __sb_end_write(struct super_block
*sb
, int level
)
1223 percpu_counter_dec(&sb
->s_writers
.counter
[level
-1]);
1225 * Make sure s_writers are updated before we wake up waiters in
1229 if (waitqueue_active(&sb
->s_writers
.wait
))
1230 wake_up(&sb
->s_writers
.wait
);
1231 rwsem_release(&sb
->s_writers
.lock_map
[level
-1], 1, _RET_IP_
);
1233 EXPORT_SYMBOL(__sb_end_write
);
1235 #ifdef CONFIG_LOCKDEP
1237 * We want lockdep to tell us about possible deadlocks with freezing but
1238 * it's it bit tricky to properly instrument it. Getting a freeze protection
1239 * works as getting a read lock but there are subtle problems. XFS for example
1240 * gets freeze protection on internal level twice in some cases, which is OK
1241 * only because we already hold a freeze protection also on higher level. Due
1242 * to these cases we have to tell lockdep we are doing trylock when we
1243 * already hold a freeze protection for a higher freeze level.
1245 static void acquire_freeze_lock(struct super_block
*sb
, int level
, bool trylock
,
1251 for (i
= 0; i
< level
- 1; i
++)
1252 if (lock_is_held(&sb
->s_writers
.lock_map
[i
])) {
1257 rwsem_acquire_read(&sb
->s_writers
.lock_map
[level
-1], 0, trylock
, ip
);
1262 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1265 int __sb_start_write(struct super_block
*sb
, int level
, bool wait
)
1268 if (unlikely(sb
->s_writers
.frozen
>= level
)) {
1271 wait_event(sb
->s_writers
.wait_unfrozen
,
1272 sb
->s_writers
.frozen
< level
);
1275 #ifdef CONFIG_LOCKDEP
1276 acquire_freeze_lock(sb
, level
, !wait
, _RET_IP_
);
1278 percpu_counter_inc(&sb
->s_writers
.counter
[level
-1]);
1280 * Make sure counter is updated before we check for frozen.
1281 * freeze_super() first sets frozen and then checks the counter.
1284 if (unlikely(sb
->s_writers
.frozen
>= level
)) {
1285 __sb_end_write(sb
, level
);
1290 EXPORT_SYMBOL(__sb_start_write
);
1293 * sb_wait_write - wait until all writers to given file system finish
1294 * @sb: the super for which we wait
1295 * @level: type of writers we wait for (normal vs page fault)
1297 * This function waits until there are no writers of given type to given file
1298 * system. Caller of this function should make sure there can be no new writers
1299 * of type @level before calling this function. Otherwise this function can
1302 static void sb_wait_write(struct super_block
*sb
, int level
)
1307 * We just cycle-through lockdep here so that it does not complain
1308 * about returning with lock to userspace
1310 rwsem_acquire(&sb
->s_writers
.lock_map
[level
-1], 0, 0, _THIS_IP_
);
1311 rwsem_release(&sb
->s_writers
.lock_map
[level
-1], 1, _THIS_IP_
);
1317 * We use a barrier in prepare_to_wait() to separate setting
1318 * of frozen and checking of the counter
1320 prepare_to_wait(&sb
->s_writers
.wait
, &wait
,
1321 TASK_UNINTERRUPTIBLE
);
1323 writers
= percpu_counter_sum(&sb
->s_writers
.counter
[level
-1]);
1327 finish_wait(&sb
->s_writers
.wait
, &wait
);
1332 * freeze_super - lock the filesystem and force it into a consistent state
1333 * @sb: the super to lock
1335 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1336 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1339 * During this function, sb->s_writers.frozen goes through these values:
1341 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1343 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1344 * writes should be blocked, though page faults are still allowed. We wait for
1345 * all writes to complete and then proceed to the next stage.
1347 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1348 * but internal fs threads can still modify the filesystem (although they
1349 * should not dirty new pages or inodes), writeback can run etc. After waiting
1350 * for all running page faults we sync the filesystem which will clean all
1351 * dirty pages and inodes (no new dirty pages or inodes can be created when
1354 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1355 * modification are blocked (e.g. XFS preallocation truncation on inode
1356 * reclaim). This is usually implemented by blocking new transactions for
1357 * filesystems that have them and need this additional guard. After all
1358 * internal writers are finished we call ->freeze_fs() to finish filesystem
1359 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1360 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1362 * sb->s_writers.frozen is protected by sb->s_umount.
1364 int freeze_super(struct super_block
*sb
)
1368 atomic_inc(&sb
->s_active
);
1369 down_write(&sb
->s_umount
);
1370 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
) {
1371 deactivate_locked_super(sb
);
1375 if (!(sb
->s_flags
& MS_BORN
)) {
1376 up_write(&sb
->s_umount
);
1377 return 0; /* sic - it's "nothing to do" */
1380 if (sb
->s_flags
& MS_RDONLY
) {
1381 /* Nothing to do really... */
1382 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1383 up_write(&sb
->s_umount
);
1387 /* From now on, no new normal writers can start */
1388 sb
->s_writers
.frozen
= SB_FREEZE_WRITE
;
1391 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1392 up_write(&sb
->s_umount
);
1394 sb_wait_write(sb
, SB_FREEZE_WRITE
);
1396 /* Now we go and block page faults... */
1397 down_write(&sb
->s_umount
);
1398 sb
->s_writers
.frozen
= SB_FREEZE_PAGEFAULT
;
1401 sb_wait_write(sb
, SB_FREEZE_PAGEFAULT
);
1403 /* All writers are done so after syncing there won't be dirty data */
1404 sync_filesystem(sb
);
1406 /* Now wait for internal filesystem counter */
1407 sb
->s_writers
.frozen
= SB_FREEZE_FS
;
1409 sb_wait_write(sb
, SB_FREEZE_FS
);
1411 if (sb
->s_op
->freeze_fs
) {
1412 ret
= sb
->s_op
->freeze_fs(sb
);
1415 "VFS:Filesystem freeze failed\n");
1416 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1418 wake_up(&sb
->s_writers
.wait_unfrozen
);
1419 deactivate_locked_super(sb
);
1424 * This is just for debugging purposes so that fs can warn if it
1425 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1427 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1428 up_write(&sb
->s_umount
);
1431 EXPORT_SYMBOL(freeze_super
);
1434 * thaw_super -- unlock filesystem
1435 * @sb: the super to thaw
1437 * Unlocks the filesystem and marks it writeable again after freeze_super().
1439 int thaw_super(struct super_block
*sb
)
1443 down_write(&sb
->s_umount
);
1444 if (sb
->s_writers
.frozen
== SB_UNFROZEN
) {
1445 up_write(&sb
->s_umount
);
1449 if (sb
->s_flags
& MS_RDONLY
)
1452 if (sb
->s_op
->unfreeze_fs
) {
1453 error
= sb
->s_op
->unfreeze_fs(sb
);
1456 "VFS:Filesystem thaw failed\n");
1457 up_write(&sb
->s_umount
);
1463 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1465 wake_up(&sb
->s_writers
.wait_unfrozen
);
1466 deactivate_locked_super(sb
);
1470 EXPORT_SYMBOL(thaw_super
);