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
))
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 s
->s_maxbytes
= MAX_NON_LFS
;
221 s
->s_op
= &default_op
;
222 s
->s_time_gran
= 1000000000;
223 s
->cleancache_poolid
= -1;
225 s
->s_shrink
.seeks
= DEFAULT_SEEKS
;
226 s
->s_shrink
.shrink
= prune_super
;
227 s
->s_shrink
.batch
= 1024;
235 free_percpu(s
->s_files
);
237 destroy_sb_writers(s
);
244 * destroy_super - frees a superblock
245 * @s: superblock to free
247 * Frees a superblock.
249 static inline void destroy_super(struct super_block
*s
)
252 free_percpu(s
->s_files
);
254 destroy_sb_writers(s
);
256 WARN_ON(!list_empty(&s
->s_mounts
));
262 /* Superblock refcounting */
265 * Drop a superblock's refcount. The caller must hold sb_lock.
267 static void __put_super(struct super_block
*sb
)
269 if (!--sb
->s_count
) {
270 list_del_init(&sb
->s_list
);
276 * put_super - drop a temporary reference to superblock
277 * @sb: superblock in question
279 * Drops a temporary reference, frees superblock if there's no
282 static void put_super(struct super_block
*sb
)
286 spin_unlock(&sb_lock
);
291 * deactivate_locked_super - drop an active reference to superblock
292 * @s: superblock to deactivate
294 * Drops an active reference to superblock, converting it into a temprory
295 * one if there is no other active references left. In that case we
296 * tell fs driver to shut it down and drop the temporary reference we
299 * Caller holds exclusive lock on superblock; that lock is released.
301 void deactivate_locked_super(struct super_block
*s
)
303 struct file_system_type
*fs
= s
->s_type
;
304 if (atomic_dec_and_test(&s
->s_active
)) {
305 cleancache_invalidate_fs(s
);
308 /* caches are now gone, we can safely kill the shrinker now */
309 unregister_shrinker(&s
->s_shrink
);
313 up_write(&s
->s_umount
);
317 EXPORT_SYMBOL(deactivate_locked_super
);
320 * deactivate_super - drop an active reference to superblock
321 * @s: superblock to deactivate
323 * Variant of deactivate_locked_super(), except that superblock is *not*
324 * locked by caller. If we are going to drop the final active reference,
325 * lock will be acquired prior to that.
327 void deactivate_super(struct super_block
*s
)
329 if (!atomic_add_unless(&s
->s_active
, -1, 1)) {
330 down_write(&s
->s_umount
);
331 deactivate_locked_super(s
);
335 EXPORT_SYMBOL(deactivate_super
);
338 * grab_super - acquire an active reference
339 * @s: reference we are trying to make active
341 * Tries to acquire an active reference. grab_super() is used when we
342 * had just found a superblock in super_blocks or fs_type->fs_supers
343 * and want to turn it into a full-blown active reference. grab_super()
344 * is called with sb_lock held and drops it. Returns 1 in case of
345 * success, 0 if we had failed (superblock contents was already dead or
346 * dying when grab_super() had been called).
348 static int grab_super(struct super_block
*s
) __releases(sb_lock
)
350 if (atomic_inc_not_zero(&s
->s_active
)) {
351 spin_unlock(&sb_lock
);
354 /* it's going away */
356 spin_unlock(&sb_lock
);
357 /* wait for it to die */
358 down_write(&s
->s_umount
);
359 up_write(&s
->s_umount
);
365 * grab_super_passive - acquire a passive reference
366 * @sb: reference we are trying to grab
368 * Tries to acquire a passive reference. This is used in places where we
369 * cannot take an active reference but we need to ensure that the
370 * superblock does not go away while we are working on it. It returns
371 * false if a reference was not gained, and returns true with the s_umount
372 * lock held in read mode if a reference is gained. On successful return,
373 * the caller must drop the s_umount lock and the passive reference when
376 bool grab_super_passive(struct super_block
*sb
)
379 if (hlist_unhashed(&sb
->s_instances
)) {
380 spin_unlock(&sb_lock
);
385 spin_unlock(&sb_lock
);
387 if (down_read_trylock(&sb
->s_umount
)) {
388 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
390 up_read(&sb
->s_umount
);
398 * Superblock locking. We really ought to get rid of these two.
400 void lock_super(struct super_block
* sb
)
402 mutex_lock(&sb
->s_lock
);
405 void unlock_super(struct super_block
* sb
)
407 mutex_unlock(&sb
->s_lock
);
410 EXPORT_SYMBOL(lock_super
);
411 EXPORT_SYMBOL(unlock_super
);
414 * generic_shutdown_super - common helper for ->kill_sb()
415 * @sb: superblock to kill
417 * generic_shutdown_super() does all fs-independent work on superblock
418 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
419 * that need destruction out of superblock, call generic_shutdown_super()
420 * and release aforementioned objects. Note: dentries and inodes _are_
421 * taken care of and do not need specific handling.
423 * Upon calling this function, the filesystem may no longer alter or
424 * rearrange the set of dentries belonging to this super_block, nor may it
425 * change the attachments of dentries to inodes.
427 void generic_shutdown_super(struct super_block
*sb
)
429 const struct super_operations
*sop
= sb
->s_op
;
432 shrink_dcache_for_umount(sb
);
434 sb
->s_flags
&= ~MS_ACTIVE
;
436 fsnotify_unmount_inodes(&sb
->s_inodes
);
443 if (!list_empty(&sb
->s_inodes
)) {
444 printk("VFS: Busy inodes after unmount of %s. "
445 "Self-destruct in 5 seconds. Have a nice day...\n",
450 /* should be initialized for __put_super_and_need_restart() */
451 hlist_del_init(&sb
->s_instances
);
452 spin_unlock(&sb_lock
);
453 up_write(&sb
->s_umount
);
456 EXPORT_SYMBOL(generic_shutdown_super
);
459 * sget - find or create a superblock
460 * @type: filesystem type superblock should belong to
461 * @test: comparison callback
462 * @set: setup callback
463 * @flags: mount flags
464 * @data: argument to each of them
466 struct super_block
*sget(struct file_system_type
*type
,
467 int (*test
)(struct super_block
*,void *),
468 int (*set
)(struct super_block
*,void *),
472 struct super_block
*s
= NULL
;
473 struct hlist_node
*node
;
474 struct super_block
*old
;
480 hlist_for_each_entry(old
, node
, &type
->fs_supers
, s_instances
) {
481 if (!test(old
, data
))
483 if (!grab_super(old
))
486 up_write(&s
->s_umount
);
490 down_write(&old
->s_umount
);
491 if (unlikely(!(old
->s_flags
& MS_BORN
))) {
492 deactivate_locked_super(old
);
499 spin_unlock(&sb_lock
);
500 s
= alloc_super(type
, flags
);
502 return ERR_PTR(-ENOMEM
);
508 spin_unlock(&sb_lock
);
509 up_write(&s
->s_umount
);
514 strlcpy(s
->s_id
, type
->name
, sizeof(s
->s_id
));
515 list_add_tail(&s
->s_list
, &super_blocks
);
516 hlist_add_head(&s
->s_instances
, &type
->fs_supers
);
517 spin_unlock(&sb_lock
);
518 get_filesystem(type
);
519 register_shrinker(&s
->s_shrink
);
525 void drop_super(struct super_block
*sb
)
527 up_read(&sb
->s_umount
);
531 EXPORT_SYMBOL(drop_super
);
534 * iterate_supers - call function for all active superblocks
535 * @f: function to call
536 * @arg: argument to pass to it
538 * Scans the superblock list and calls given function, passing it
539 * locked superblock and given argument.
541 void iterate_supers(void (*f
)(struct super_block
*, void *), void *arg
)
543 struct super_block
*sb
, *p
= NULL
;
546 list_for_each_entry(sb
, &super_blocks
, s_list
) {
547 if (hlist_unhashed(&sb
->s_instances
))
550 spin_unlock(&sb_lock
);
552 down_read(&sb
->s_umount
);
553 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
555 up_read(&sb
->s_umount
);
564 spin_unlock(&sb_lock
);
568 * iterate_supers_type - call function for superblocks of given type
570 * @f: function to call
571 * @arg: argument to pass to it
573 * Scans the superblock list and calls given function, passing it
574 * locked superblock and given argument.
576 void iterate_supers_type(struct file_system_type
*type
,
577 void (*f
)(struct super_block
*, void *), void *arg
)
579 struct super_block
*sb
, *p
= NULL
;
580 struct hlist_node
*node
;
583 hlist_for_each_entry(sb
, node
, &type
->fs_supers
, s_instances
) {
585 spin_unlock(&sb_lock
);
587 down_read(&sb
->s_umount
);
588 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
590 up_read(&sb
->s_umount
);
599 spin_unlock(&sb_lock
);
602 EXPORT_SYMBOL(iterate_supers_type
);
605 * get_super - get the superblock of a device
606 * @bdev: device to get the superblock for
608 * Scans the superblock list and finds the superblock of the file system
609 * mounted on the device given. %NULL is returned if no match is found.
612 struct super_block
*get_super(struct block_device
*bdev
)
614 struct super_block
*sb
;
621 list_for_each_entry(sb
, &super_blocks
, s_list
) {
622 if (hlist_unhashed(&sb
->s_instances
))
624 if (sb
->s_bdev
== bdev
) {
626 spin_unlock(&sb_lock
);
627 down_read(&sb
->s_umount
);
629 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
631 up_read(&sb
->s_umount
);
632 /* nope, got unmounted */
638 spin_unlock(&sb_lock
);
642 EXPORT_SYMBOL(get_super
);
645 * get_super_thawed - get thawed superblock of a device
646 * @bdev: device to get the superblock for
648 * Scans the superblock list and finds the superblock of the file system
649 * mounted on the device. The superblock is returned once it is thawed
650 * (or immediately if it was not frozen). %NULL is returned if no match
653 struct super_block
*get_super_thawed(struct block_device
*bdev
)
656 struct super_block
*s
= get_super(bdev
);
657 if (!s
|| s
->s_writers
.frozen
== SB_UNFROZEN
)
659 up_read(&s
->s_umount
);
660 wait_event(s
->s_writers
.wait_unfrozen
,
661 s
->s_writers
.frozen
== SB_UNFROZEN
);
665 EXPORT_SYMBOL(get_super_thawed
);
668 * get_active_super - get an active reference to the superblock of a device
669 * @bdev: device to get the superblock for
671 * Scans the superblock list and finds the superblock of the file system
672 * mounted on the device given. Returns the superblock with an active
673 * reference or %NULL if none was found.
675 struct super_block
*get_active_super(struct block_device
*bdev
)
677 struct super_block
*sb
;
684 list_for_each_entry(sb
, &super_blocks
, s_list
) {
685 if (hlist_unhashed(&sb
->s_instances
))
687 if (sb
->s_bdev
== bdev
) {
688 if (grab_super(sb
)) /* drops sb_lock */
694 spin_unlock(&sb_lock
);
698 struct super_block
*user_get_super(dev_t dev
)
700 struct super_block
*sb
;
704 list_for_each_entry(sb
, &super_blocks
, s_list
) {
705 if (hlist_unhashed(&sb
->s_instances
))
707 if (sb
->s_dev
== dev
) {
709 spin_unlock(&sb_lock
);
710 down_read(&sb
->s_umount
);
712 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
714 up_read(&sb
->s_umount
);
715 /* nope, got unmounted */
721 spin_unlock(&sb_lock
);
726 * do_remount_sb - asks filesystem to change mount options.
727 * @sb: superblock in question
728 * @flags: numeric part of options
729 * @data: the rest of options
730 * @force: whether or not to force the change
732 * Alters the mount options of a mounted file system.
734 int do_remount_sb(struct super_block
*sb
, int flags
, void *data
, int force
)
739 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
)
743 if (!(flags
& MS_RDONLY
) && bdev_read_only(sb
->s_bdev
))
747 if (flags
& MS_RDONLY
)
749 shrink_dcache_sb(sb
);
752 remount_ro
= (flags
& MS_RDONLY
) && !(sb
->s_flags
& MS_RDONLY
);
754 /* If we are remounting RDONLY and current sb is read/write,
755 make sure there are no rw files opened */
760 retval
= sb_prepare_remount_readonly(sb
);
766 if (sb
->s_op
->remount_fs
) {
767 retval
= sb
->s_op
->remount_fs(sb
, &flags
, data
);
770 goto cancel_readonly
;
771 /* If forced remount, go ahead despite any errors */
772 WARN(1, "forced remount of a %s fs returned %i\n",
773 sb
->s_type
->name
, retval
);
776 sb
->s_flags
= (sb
->s_flags
& ~MS_RMT_MASK
) | (flags
& MS_RMT_MASK
);
777 /* Needs to be ordered wrt mnt_is_readonly() */
779 sb
->s_readonly_remount
= 0;
782 * Some filesystems modify their metadata via some other path than the
783 * bdev buffer cache (eg. use a private mapping, or directories in
784 * pagecache, etc). Also file data modifications go via their own
785 * mappings. So If we try to mount readonly then copy the filesystem
786 * from bdev, we could get stale data, so invalidate it to give a best
787 * effort at coherency.
789 if (remount_ro
&& sb
->s_bdev
)
790 invalidate_bdev(sb
->s_bdev
);
794 sb
->s_readonly_remount
= 0;
798 static void do_emergency_remount(struct work_struct
*work
)
800 struct super_block
*sb
, *p
= NULL
;
803 list_for_each_entry(sb
, &super_blocks
, s_list
) {
804 if (hlist_unhashed(&sb
->s_instances
))
807 spin_unlock(&sb_lock
);
808 down_write(&sb
->s_umount
);
809 if (sb
->s_root
&& sb
->s_bdev
&& (sb
->s_flags
& MS_BORN
) &&
810 !(sb
->s_flags
& MS_RDONLY
)) {
812 * What lock protects sb->s_flags??
814 do_remount_sb(sb
, MS_RDONLY
, NULL
, 1);
816 up_write(&sb
->s_umount
);
824 spin_unlock(&sb_lock
);
826 printk("Emergency Remount complete\n");
829 void emergency_remount(void)
831 struct work_struct
*work
;
833 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
835 INIT_WORK(work
, do_emergency_remount
);
841 * Unnamed block devices are dummy devices used by virtual
842 * filesystems which don't use real block-devices. -- jrs
845 static DEFINE_IDA(unnamed_dev_ida
);
846 static DEFINE_SPINLOCK(unnamed_dev_lock
);/* protects the above */
847 static int unnamed_dev_start
= 0; /* don't bother trying below it */
849 int get_anon_bdev(dev_t
*p
)
855 if (ida_pre_get(&unnamed_dev_ida
, GFP_ATOMIC
) == 0)
857 spin_lock(&unnamed_dev_lock
);
858 error
= ida_get_new_above(&unnamed_dev_ida
, unnamed_dev_start
, &dev
);
860 unnamed_dev_start
= dev
+ 1;
861 spin_unlock(&unnamed_dev_lock
);
862 if (error
== -EAGAIN
)
863 /* We raced and lost with another CPU. */
868 if ((dev
& MAX_IDR_MASK
) == (1 << MINORBITS
)) {
869 spin_lock(&unnamed_dev_lock
);
870 ida_remove(&unnamed_dev_ida
, dev
);
871 if (unnamed_dev_start
> dev
)
872 unnamed_dev_start
= dev
;
873 spin_unlock(&unnamed_dev_lock
);
876 *p
= MKDEV(0, dev
& MINORMASK
);
879 EXPORT_SYMBOL(get_anon_bdev
);
881 void free_anon_bdev(dev_t dev
)
883 int slot
= MINOR(dev
);
884 spin_lock(&unnamed_dev_lock
);
885 ida_remove(&unnamed_dev_ida
, slot
);
886 if (slot
< unnamed_dev_start
)
887 unnamed_dev_start
= slot
;
888 spin_unlock(&unnamed_dev_lock
);
890 EXPORT_SYMBOL(free_anon_bdev
);
892 int set_anon_super(struct super_block
*s
, void *data
)
894 int error
= get_anon_bdev(&s
->s_dev
);
896 s
->s_bdi
= &noop_backing_dev_info
;
900 EXPORT_SYMBOL(set_anon_super
);
902 void kill_anon_super(struct super_block
*sb
)
904 dev_t dev
= sb
->s_dev
;
905 generic_shutdown_super(sb
);
909 EXPORT_SYMBOL(kill_anon_super
);
911 void kill_litter_super(struct super_block
*sb
)
914 d_genocide(sb
->s_root
);
918 EXPORT_SYMBOL(kill_litter_super
);
920 static int ns_test_super(struct super_block
*sb
, void *data
)
922 return sb
->s_fs_info
== data
;
925 static int ns_set_super(struct super_block
*sb
, void *data
)
927 sb
->s_fs_info
= data
;
928 return set_anon_super(sb
, NULL
);
931 struct dentry
*mount_ns(struct file_system_type
*fs_type
, int flags
,
932 void *data
, int (*fill_super
)(struct super_block
*, void *, int))
934 struct super_block
*sb
;
936 sb
= sget(fs_type
, ns_test_super
, ns_set_super
, flags
, data
);
942 err
= fill_super(sb
, data
, flags
& MS_SILENT
? 1 : 0);
944 deactivate_locked_super(sb
);
948 sb
->s_flags
|= MS_ACTIVE
;
951 return dget(sb
->s_root
);
954 EXPORT_SYMBOL(mount_ns
);
957 static int set_bdev_super(struct super_block
*s
, void *data
)
960 s
->s_dev
= s
->s_bdev
->bd_dev
;
963 * We set the bdi here to the queue backing, file systems can
964 * overwrite this in ->fill_super()
966 s
->s_bdi
= &bdev_get_queue(s
->s_bdev
)->backing_dev_info
;
970 static int test_bdev_super(struct super_block
*s
, void *data
)
972 return (void *)s
->s_bdev
== data
;
975 struct dentry
*mount_bdev(struct file_system_type
*fs_type
,
976 int flags
, const char *dev_name
, void *data
,
977 int (*fill_super
)(struct super_block
*, void *, int))
979 struct block_device
*bdev
;
980 struct super_block
*s
;
981 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
984 if (!(flags
& MS_RDONLY
))
987 bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
989 return ERR_CAST(bdev
);
992 * once the super is inserted into the list by sget, s_umount
993 * will protect the lockfs code from trying to start a snapshot
994 * while we are mounting
996 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
997 if (bdev
->bd_fsfreeze_count
> 0) {
998 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1002 s
= sget(fs_type
, test_bdev_super
, set_bdev_super
, flags
| MS_NOSEC
,
1004 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1009 if ((flags
^ s
->s_flags
) & MS_RDONLY
) {
1010 deactivate_locked_super(s
);
1016 * s_umount nests inside bd_mutex during
1017 * __invalidate_device(). blkdev_put() acquires
1018 * bd_mutex and can't be called under s_umount. Drop
1019 * s_umount temporarily. This is safe as we're
1020 * holding an active reference.
1022 up_write(&s
->s_umount
);
1023 blkdev_put(bdev
, mode
);
1024 down_write(&s
->s_umount
);
1026 char b
[BDEVNAME_SIZE
];
1029 strlcpy(s
->s_id
, bdevname(bdev
, b
), sizeof(s
->s_id
));
1030 sb_set_blocksize(s
, block_size(bdev
));
1031 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1033 deactivate_locked_super(s
);
1037 s
->s_flags
|= MS_ACTIVE
;
1041 return dget(s
->s_root
);
1046 blkdev_put(bdev
, mode
);
1048 return ERR_PTR(error
);
1050 EXPORT_SYMBOL(mount_bdev
);
1052 void kill_block_super(struct super_block
*sb
)
1054 struct block_device
*bdev
= sb
->s_bdev
;
1055 fmode_t mode
= sb
->s_mode
;
1057 bdev
->bd_super
= NULL
;
1058 generic_shutdown_super(sb
);
1059 sync_blockdev(bdev
);
1060 WARN_ON_ONCE(!(mode
& FMODE_EXCL
));
1061 blkdev_put(bdev
, mode
| FMODE_EXCL
);
1064 EXPORT_SYMBOL(kill_block_super
);
1067 struct dentry
*mount_nodev(struct file_system_type
*fs_type
,
1068 int flags
, void *data
,
1069 int (*fill_super
)(struct super_block
*, void *, int))
1072 struct super_block
*s
= sget(fs_type
, NULL
, set_anon_super
, flags
, NULL
);
1077 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1079 deactivate_locked_super(s
);
1080 return ERR_PTR(error
);
1082 s
->s_flags
|= MS_ACTIVE
;
1083 return dget(s
->s_root
);
1085 EXPORT_SYMBOL(mount_nodev
);
1087 static int compare_single(struct super_block
*s
, void *p
)
1092 struct dentry
*mount_single(struct file_system_type
*fs_type
,
1093 int flags
, void *data
,
1094 int (*fill_super
)(struct super_block
*, void *, int))
1096 struct super_block
*s
;
1099 s
= sget(fs_type
, compare_single
, set_anon_super
, flags
, NULL
);
1103 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1105 deactivate_locked_super(s
);
1106 return ERR_PTR(error
);
1108 s
->s_flags
|= MS_ACTIVE
;
1110 do_remount_sb(s
, flags
, data
, 0);
1112 return dget(s
->s_root
);
1114 EXPORT_SYMBOL(mount_single
);
1117 mount_fs(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
1119 struct dentry
*root
;
1120 struct super_block
*sb
;
1121 char *secdata
= NULL
;
1122 int error
= -ENOMEM
;
1124 if (data
&& !(type
->fs_flags
& FS_BINARY_MOUNTDATA
)) {
1125 secdata
= alloc_secdata();
1129 error
= security_sb_copy_data(data
, secdata
);
1131 goto out_free_secdata
;
1134 root
= type
->mount(type
, flags
, name
, data
);
1136 error
= PTR_ERR(root
);
1137 goto out_free_secdata
;
1141 WARN_ON(!sb
->s_bdi
);
1142 WARN_ON(sb
->s_bdi
== &default_backing_dev_info
);
1143 sb
->s_flags
|= MS_BORN
;
1145 error
= security_sb_kern_mount(sb
, flags
, secdata
);
1150 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1151 * but s_maxbytes was an unsigned long long for many releases. Throw
1152 * this warning for a little while to try and catch filesystems that
1153 * violate this rule.
1155 WARN((sb
->s_maxbytes
< 0), "%s set sb->s_maxbytes to "
1156 "negative value (%lld)\n", type
->name
, sb
->s_maxbytes
);
1158 up_write(&sb
->s_umount
);
1159 free_secdata(secdata
);
1163 deactivate_locked_super(sb
);
1165 free_secdata(secdata
);
1167 return ERR_PTR(error
);
1171 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1174 void __sb_end_write(struct super_block
*sb
, int level
)
1176 percpu_counter_dec(&sb
->s_writers
.counter
[level
-1]);
1178 * Make sure s_writers are updated before we wake up waiters in
1182 if (waitqueue_active(&sb
->s_writers
.wait
))
1183 wake_up(&sb
->s_writers
.wait
);
1184 rwsem_release(&sb
->s_writers
.lock_map
[level
-1], 1, _RET_IP_
);
1186 EXPORT_SYMBOL(__sb_end_write
);
1188 #ifdef CONFIG_LOCKDEP
1190 * We want lockdep to tell us about possible deadlocks with freezing but
1191 * it's it bit tricky to properly instrument it. Getting a freeze protection
1192 * works as getting a read lock but there are subtle problems. XFS for example
1193 * gets freeze protection on internal level twice in some cases, which is OK
1194 * only because we already hold a freeze protection also on higher level. Due
1195 * to these cases we have to tell lockdep we are doing trylock when we
1196 * already hold a freeze protection for a higher freeze level.
1198 static void acquire_freeze_lock(struct super_block
*sb
, int level
, bool trylock
,
1204 for (i
= 0; i
< level
- 1; i
++)
1205 if (lock_is_held(&sb
->s_writers
.lock_map
[i
])) {
1210 rwsem_acquire_read(&sb
->s_writers
.lock_map
[level
-1], 0, trylock
, ip
);
1215 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1218 int __sb_start_write(struct super_block
*sb
, int level
, bool wait
)
1221 if (unlikely(sb
->s_writers
.frozen
>= level
)) {
1224 wait_event(sb
->s_writers
.wait_unfrozen
,
1225 sb
->s_writers
.frozen
< level
);
1228 #ifdef CONFIG_LOCKDEP
1229 acquire_freeze_lock(sb
, level
, !wait
, _RET_IP_
);
1231 percpu_counter_inc(&sb
->s_writers
.counter
[level
-1]);
1233 * Make sure counter is updated before we check for frozen.
1234 * freeze_super() first sets frozen and then checks the counter.
1237 if (unlikely(sb
->s_writers
.frozen
>= level
)) {
1238 __sb_end_write(sb
, level
);
1243 EXPORT_SYMBOL(__sb_start_write
);
1246 * sb_wait_write - wait until all writers to given file system finish
1247 * @sb: the super for which we wait
1248 * @level: type of writers we wait for (normal vs page fault)
1250 * This function waits until there are no writers of given type to given file
1251 * system. Caller of this function should make sure there can be no new writers
1252 * of type @level before calling this function. Otherwise this function can
1255 static void sb_wait_write(struct super_block
*sb
, int level
)
1260 * We just cycle-through lockdep here so that it does not complain
1261 * about returning with lock to userspace
1263 rwsem_acquire(&sb
->s_writers
.lock_map
[level
-1], 0, 0, _THIS_IP_
);
1264 rwsem_release(&sb
->s_writers
.lock_map
[level
-1], 1, _THIS_IP_
);
1270 * We use a barrier in prepare_to_wait() to separate setting
1271 * of frozen and checking of the counter
1273 prepare_to_wait(&sb
->s_writers
.wait
, &wait
,
1274 TASK_UNINTERRUPTIBLE
);
1276 writers
= percpu_counter_sum(&sb
->s_writers
.counter
[level
-1]);
1280 finish_wait(&sb
->s_writers
.wait
, &wait
);
1285 * freeze_super - lock the filesystem and force it into a consistent state
1286 * @sb: the super to lock
1288 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1289 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1292 * During this function, sb->s_writers.frozen goes through these values:
1294 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1296 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1297 * writes should be blocked, though page faults are still allowed. We wait for
1298 * all writes to complete and then proceed to the next stage.
1300 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1301 * but internal fs threads can still modify the filesystem (although they
1302 * should not dirty new pages or inodes), writeback can run etc. After waiting
1303 * for all running page faults we sync the filesystem which will clean all
1304 * dirty pages and inodes (no new dirty pages or inodes can be created when
1307 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1308 * modification are blocked (e.g. XFS preallocation truncation on inode
1309 * reclaim). This is usually implemented by blocking new transactions for
1310 * filesystems that have them and need this additional guard. After all
1311 * internal writers are finished we call ->freeze_fs() to finish filesystem
1312 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1313 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1315 * sb->s_writers.frozen is protected by sb->s_umount.
1317 int freeze_super(struct super_block
*sb
)
1321 atomic_inc(&sb
->s_active
);
1322 down_write(&sb
->s_umount
);
1323 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
) {
1324 deactivate_locked_super(sb
);
1328 if (!(sb
->s_flags
& MS_BORN
)) {
1329 up_write(&sb
->s_umount
);
1330 return 0; /* sic - it's "nothing to do" */
1333 if (sb
->s_flags
& MS_RDONLY
) {
1334 /* Nothing to do really... */
1335 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1336 up_write(&sb
->s_umount
);
1340 /* From now on, no new normal writers can start */
1341 sb
->s_writers
.frozen
= SB_FREEZE_WRITE
;
1344 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1345 up_write(&sb
->s_umount
);
1347 sb_wait_write(sb
, SB_FREEZE_WRITE
);
1349 /* Now we go and block page faults... */
1350 down_write(&sb
->s_umount
);
1351 sb
->s_writers
.frozen
= SB_FREEZE_PAGEFAULT
;
1354 sb_wait_write(sb
, SB_FREEZE_PAGEFAULT
);
1356 /* All writers are done so after syncing there won't be dirty data */
1357 sync_filesystem(sb
);
1359 /* Now wait for internal filesystem counter */
1360 sb
->s_writers
.frozen
= SB_FREEZE_FS
;
1362 sb_wait_write(sb
, SB_FREEZE_FS
);
1364 if (sb
->s_op
->freeze_fs
) {
1365 ret
= sb
->s_op
->freeze_fs(sb
);
1368 "VFS:Filesystem freeze failed\n");
1369 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1371 wake_up(&sb
->s_writers
.wait_unfrozen
);
1372 deactivate_locked_super(sb
);
1377 * This is just for debugging purposes so that fs can warn if it
1378 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1380 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1381 up_write(&sb
->s_umount
);
1384 EXPORT_SYMBOL(freeze_super
);
1387 * thaw_super -- unlock filesystem
1388 * @sb: the super to thaw
1390 * Unlocks the filesystem and marks it writeable again after freeze_super().
1392 int thaw_super(struct super_block
*sb
)
1396 down_write(&sb
->s_umount
);
1397 if (sb
->s_writers
.frozen
== SB_UNFROZEN
) {
1398 up_write(&sb
->s_umount
);
1402 if (sb
->s_flags
& MS_RDONLY
)
1405 if (sb
->s_op
->unfreeze_fs
) {
1406 error
= sb
->s_op
->unfreeze_fs(sb
);
1409 "VFS:Filesystem thaw failed\n");
1410 up_write(&sb
->s_umount
);
1416 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1418 wake_up(&sb
->s_writers
.wait_unfrozen
);
1419 deactivate_locked_super(sb
);
1423 EXPORT_SYMBOL(thaw_super
);