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/blkdev.h>
26 #include <linux/mount.h>
27 #include <linux/security.h>
28 #include <linux/writeback.h> /* for the emergency remount stuff */
29 #include <linux/idr.h>
30 #include <linux/mutex.h>
31 #include <linux/backing-dev.h>
32 #include <linux/rculist_bl.h>
33 #include <linux/cleancache.h>
34 #include <linux/fsnotify.h>
35 #include <linux/lockdep.h>
39 static LIST_HEAD(super_blocks
);
40 static DEFINE_SPINLOCK(sb_lock
);
42 static char *sb_writers_name
[SB_FREEZE_LEVELS
] = {
49 * One thing we have to be careful of with a per-sb shrinker is that we don't
50 * drop the last active reference to the superblock from within the shrinker.
51 * If that happens we could trigger unregistering the shrinker from within the
52 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
53 * take a passive reference to the superblock to avoid this from occurring.
55 static unsigned long super_cache_scan(struct shrinker
*shrink
,
56 struct shrink_control
*sc
)
58 struct super_block
*sb
;
65 sb
= container_of(shrink
, struct super_block
, s_shrink
);
68 * Deadlock avoidance. We may hold various FS locks, and we don't want
69 * to recurse into the FS that called us in clear_inode() and friends..
71 if (!(sc
->gfp_mask
& __GFP_FS
))
74 if (!grab_super_passive(sb
))
77 if (sb
->s_op
->nr_cached_objects
)
78 fs_objects
= sb
->s_op
->nr_cached_objects(sb
, sc
->nid
);
80 inodes
= list_lru_count_node(&sb
->s_inode_lru
, sc
->nid
);
81 dentries
= list_lru_count_node(&sb
->s_dentry_lru
, sc
->nid
);
82 total_objects
= dentries
+ inodes
+ fs_objects
+ 1;
86 /* proportion the scan between the caches */
87 dentries
= mult_frac(sc
->nr_to_scan
, dentries
, total_objects
);
88 inodes
= mult_frac(sc
->nr_to_scan
, inodes
, total_objects
);
91 * prune the dcache first as the icache is pinned by it, then
92 * prune the icache, followed by the filesystem specific caches
94 freed
= prune_dcache_sb(sb
, dentries
, sc
->nid
);
95 freed
+= prune_icache_sb(sb
, inodes
, sc
->nid
);
98 fs_objects
= mult_frac(sc
->nr_to_scan
, fs_objects
,
100 freed
+= sb
->s_op
->free_cached_objects(sb
, fs_objects
,
108 static unsigned long super_cache_count(struct shrinker
*shrink
,
109 struct shrink_control
*sc
)
111 struct super_block
*sb
;
112 long total_objects
= 0;
114 sb
= container_of(shrink
, struct super_block
, s_shrink
);
117 * Don't call grab_super_passive as it is a potential
118 * scalability bottleneck. The counts could get updated
119 * between super_cache_count and super_cache_scan anyway.
120 * Call to super_cache_count with shrinker_rwsem held
121 * ensures the safety of call to list_lru_count_node() and
122 * s_op->nr_cached_objects().
124 if (sb
->s_op
&& sb
->s_op
->nr_cached_objects
)
125 total_objects
= sb
->s_op
->nr_cached_objects(sb
,
128 total_objects
+= list_lru_count_node(&sb
->s_dentry_lru
,
130 total_objects
+= list_lru_count_node(&sb
->s_inode_lru
,
133 total_objects
= vfs_pressure_ratio(total_objects
);
134 return total_objects
;
138 * destroy_super - frees a superblock
139 * @s: superblock to free
141 * Frees a superblock.
143 static void destroy_super(struct super_block
*s
)
146 list_lru_destroy(&s
->s_dentry_lru
);
147 list_lru_destroy(&s
->s_inode_lru
);
148 for (i
= 0; i
< SB_FREEZE_LEVELS
; i
++)
149 percpu_counter_destroy(&s
->s_writers
.counter
[i
]);
151 WARN_ON(!list_empty(&s
->s_mounts
));
158 * alloc_super - create new superblock
159 * @type: filesystem type superblock should belong to
160 * @flags: the mount flags
162 * Allocates and initializes a new &struct super_block. alloc_super()
163 * returns a pointer new superblock or %NULL if allocation had failed.
165 static struct super_block
*alloc_super(struct file_system_type
*type
, int flags
)
167 struct super_block
*s
= kzalloc(sizeof(struct super_block
), GFP_USER
);
168 static const struct super_operations default_op
;
174 INIT_LIST_HEAD(&s
->s_mounts
);
176 if (security_sb_alloc(s
))
179 for (i
= 0; i
< SB_FREEZE_LEVELS
; i
++) {
180 if (percpu_counter_init(&s
->s_writers
.counter
[i
], 0,
183 lockdep_init_map(&s
->s_writers
.lock_map
[i
], sb_writers_name
[i
],
184 &type
->s_writers_key
[i
], 0);
186 init_waitqueue_head(&s
->s_writers
.wait
);
187 init_waitqueue_head(&s
->s_writers
.wait_unfrozen
);
188 s
->s_bdi
= &noop_backing_dev_info
;
190 INIT_HLIST_NODE(&s
->s_instances
);
191 INIT_HLIST_BL_HEAD(&s
->s_anon
);
192 INIT_LIST_HEAD(&s
->s_inodes
);
194 if (list_lru_init(&s
->s_dentry_lru
))
196 if (list_lru_init(&s
->s_inode_lru
))
199 init_rwsem(&s
->s_umount
);
200 lockdep_set_class(&s
->s_umount
, &type
->s_umount_key
);
202 * sget() can have s_umount recursion.
204 * When it cannot find a suitable sb, it allocates a new
205 * one (this one), and tries again to find a suitable old
208 * In case that succeeds, it will acquire the s_umount
209 * lock of the old one. Since these are clearly distrinct
210 * locks, and this object isn't exposed yet, there's no
213 * Annotate this by putting this lock in a different
216 down_write_nested(&s
->s_umount
, SINGLE_DEPTH_NESTING
);
218 atomic_set(&s
->s_active
, 1);
219 mutex_init(&s
->s_vfs_rename_mutex
);
220 lockdep_set_class(&s
->s_vfs_rename_mutex
, &type
->s_vfs_rename_key
);
221 mutex_init(&s
->s_dquot
.dqio_mutex
);
222 mutex_init(&s
->s_dquot
.dqonoff_mutex
);
223 s
->s_maxbytes
= MAX_NON_LFS
;
224 s
->s_op
= &default_op
;
225 s
->s_time_gran
= 1000000000;
226 s
->cleancache_poolid
= -1;
228 s
->s_shrink
.seeks
= DEFAULT_SEEKS
;
229 s
->s_shrink
.scan_objects
= super_cache_scan
;
230 s
->s_shrink
.count_objects
= super_cache_count
;
231 s
->s_shrink
.batch
= 1024;
232 s
->s_shrink
.flags
= SHRINKER_NUMA_AWARE
;
240 /* Superblock refcounting */
243 * Drop a superblock's refcount. The caller must hold sb_lock.
245 static void __put_super(struct super_block
*sb
)
247 if (!--sb
->s_count
) {
248 list_del_init(&sb
->s_list
);
254 * put_super - drop a temporary reference to superblock
255 * @sb: superblock in question
257 * Drops a temporary reference, frees superblock if there's no
260 static void put_super(struct super_block
*sb
)
264 spin_unlock(&sb_lock
);
269 * deactivate_locked_super - drop an active reference to superblock
270 * @s: superblock to deactivate
272 * Drops an active reference to superblock, converting it into a temprory
273 * one if there is no other active references left. In that case we
274 * tell fs driver to shut it down and drop the temporary reference we
277 * Caller holds exclusive lock on superblock; that lock is released.
279 void deactivate_locked_super(struct super_block
*s
)
281 struct file_system_type
*fs
= s
->s_type
;
282 if (atomic_dec_and_test(&s
->s_active
)) {
283 cleancache_invalidate_fs(s
);
284 unregister_shrinker(&s
->s_shrink
);
290 up_write(&s
->s_umount
);
294 EXPORT_SYMBOL(deactivate_locked_super
);
297 * deactivate_super - drop an active reference to superblock
298 * @s: superblock to deactivate
300 * Variant of deactivate_locked_super(), except that superblock is *not*
301 * locked by caller. If we are going to drop the final active reference,
302 * lock will be acquired prior to that.
304 void deactivate_super(struct super_block
*s
)
306 if (!atomic_add_unless(&s
->s_active
, -1, 1)) {
307 down_write(&s
->s_umount
);
308 deactivate_locked_super(s
);
312 EXPORT_SYMBOL(deactivate_super
);
315 * grab_super - acquire an active reference
316 * @s: reference we are trying to make active
318 * Tries to acquire an active reference. grab_super() is used when we
319 * had just found a superblock in super_blocks or fs_type->fs_supers
320 * and want to turn it into a full-blown active reference. grab_super()
321 * is called with sb_lock held and drops it. Returns 1 in case of
322 * success, 0 if we had failed (superblock contents was already dead or
323 * dying when grab_super() had been called). Note that this is only
324 * called for superblocks not in rundown mode (== ones still on ->fs_supers
325 * of their type), so increment of ->s_count is OK here.
327 static int grab_super(struct super_block
*s
) __releases(sb_lock
)
330 spin_unlock(&sb_lock
);
331 down_write(&s
->s_umount
);
332 if ((s
->s_flags
& MS_BORN
) && atomic_inc_not_zero(&s
->s_active
)) {
336 up_write(&s
->s_umount
);
342 * grab_super_passive - acquire a passive reference
343 * @sb: reference we are trying to grab
345 * Tries to acquire a passive reference. This is used in places where we
346 * cannot take an active reference but we need to ensure that the
347 * superblock does not go away while we are working on it. It returns
348 * false if a reference was not gained, and returns true with the s_umount
349 * lock held in read mode if a reference is gained. On successful return,
350 * the caller must drop the s_umount lock and the passive reference when
353 bool grab_super_passive(struct super_block
*sb
)
356 if (hlist_unhashed(&sb
->s_instances
)) {
357 spin_unlock(&sb_lock
);
362 spin_unlock(&sb_lock
);
364 if (down_read_trylock(&sb
->s_umount
)) {
365 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
367 up_read(&sb
->s_umount
);
375 * generic_shutdown_super - common helper for ->kill_sb()
376 * @sb: superblock to kill
378 * generic_shutdown_super() does all fs-independent work on superblock
379 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
380 * that need destruction out of superblock, call generic_shutdown_super()
381 * and release aforementioned objects. Note: dentries and inodes _are_
382 * taken care of and do not need specific handling.
384 * Upon calling this function, the filesystem may no longer alter or
385 * rearrange the set of dentries belonging to this super_block, nor may it
386 * change the attachments of dentries to inodes.
388 void generic_shutdown_super(struct super_block
*sb
)
390 const struct super_operations
*sop
= sb
->s_op
;
393 shrink_dcache_for_umount(sb
);
395 sb
->s_flags
&= ~MS_ACTIVE
;
397 fsnotify_unmount_inodes(&sb
->s_inodes
);
401 if (sb
->s_dio_done_wq
) {
402 destroy_workqueue(sb
->s_dio_done_wq
);
403 sb
->s_dio_done_wq
= NULL
;
409 if (!list_empty(&sb
->s_inodes
)) {
410 printk("VFS: Busy inodes after unmount of %s. "
411 "Self-destruct in 5 seconds. Have a nice day...\n",
416 /* should be initialized for __put_super_and_need_restart() */
417 hlist_del_init(&sb
->s_instances
);
418 spin_unlock(&sb_lock
);
419 up_write(&sb
->s_umount
);
422 EXPORT_SYMBOL(generic_shutdown_super
);
425 * sget - find or create a superblock
426 * @type: filesystem type superblock should belong to
427 * @test: comparison callback
428 * @set: setup callback
429 * @flags: mount flags
430 * @data: argument to each of them
432 struct super_block
*sget(struct file_system_type
*type
,
433 int (*test
)(struct super_block
*,void *),
434 int (*set
)(struct super_block
*,void *),
438 struct super_block
*s
= NULL
;
439 struct super_block
*old
;
445 hlist_for_each_entry(old
, &type
->fs_supers
, s_instances
) {
446 if (!test(old
, data
))
448 if (!grab_super(old
))
451 up_write(&s
->s_umount
);
459 spin_unlock(&sb_lock
);
460 s
= alloc_super(type
, flags
);
462 return ERR_PTR(-ENOMEM
);
468 spin_unlock(&sb_lock
);
469 up_write(&s
->s_umount
);
474 strlcpy(s
->s_id
, type
->name
, sizeof(s
->s_id
));
475 list_add_tail(&s
->s_list
, &super_blocks
);
476 hlist_add_head(&s
->s_instances
, &type
->fs_supers
);
477 spin_unlock(&sb_lock
);
478 get_filesystem(type
);
479 register_shrinker(&s
->s_shrink
);
485 void drop_super(struct super_block
*sb
)
487 up_read(&sb
->s_umount
);
491 EXPORT_SYMBOL(drop_super
);
494 * iterate_supers - call function for all active superblocks
495 * @f: function to call
496 * @arg: argument to pass to it
498 * Scans the superblock list and calls given function, passing it
499 * locked superblock and given argument.
501 void iterate_supers(void (*f
)(struct super_block
*, void *), void *arg
)
503 struct super_block
*sb
, *p
= NULL
;
506 list_for_each_entry(sb
, &super_blocks
, s_list
) {
507 if (hlist_unhashed(&sb
->s_instances
))
510 spin_unlock(&sb_lock
);
512 down_read(&sb
->s_umount
);
513 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
515 up_read(&sb
->s_umount
);
524 spin_unlock(&sb_lock
);
528 * iterate_supers_type - call function for superblocks of given type
530 * @f: function to call
531 * @arg: argument to pass to it
533 * Scans the superblock list and calls given function, passing it
534 * locked superblock and given argument.
536 void iterate_supers_type(struct file_system_type
*type
,
537 void (*f
)(struct super_block
*, void *), void *arg
)
539 struct super_block
*sb
, *p
= NULL
;
542 hlist_for_each_entry(sb
, &type
->fs_supers
, s_instances
) {
544 spin_unlock(&sb_lock
);
546 down_read(&sb
->s_umount
);
547 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
549 up_read(&sb
->s_umount
);
558 spin_unlock(&sb_lock
);
561 EXPORT_SYMBOL(iterate_supers_type
);
564 * get_super - get the superblock of a device
565 * @bdev: device to get the superblock for
567 * Scans the superblock list and finds the superblock of the file system
568 * mounted on the device given. %NULL is returned if no match is found.
571 struct super_block
*get_super(struct block_device
*bdev
)
573 struct super_block
*sb
;
580 list_for_each_entry(sb
, &super_blocks
, s_list
) {
581 if (hlist_unhashed(&sb
->s_instances
))
583 if (sb
->s_bdev
== bdev
) {
585 spin_unlock(&sb_lock
);
586 down_read(&sb
->s_umount
);
588 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
590 up_read(&sb
->s_umount
);
591 /* nope, got unmounted */
597 spin_unlock(&sb_lock
);
601 EXPORT_SYMBOL(get_super
);
604 * get_super_thawed - get thawed superblock of a device
605 * @bdev: device to get the superblock for
607 * Scans the superblock list and finds the superblock of the file system
608 * mounted on the device. The superblock is returned once it is thawed
609 * (or immediately if it was not frozen). %NULL is returned if no match
612 struct super_block
*get_super_thawed(struct block_device
*bdev
)
615 struct super_block
*s
= get_super(bdev
);
616 if (!s
|| s
->s_writers
.frozen
== SB_UNFROZEN
)
618 up_read(&s
->s_umount
);
619 wait_event(s
->s_writers
.wait_unfrozen
,
620 s
->s_writers
.frozen
== SB_UNFROZEN
);
624 EXPORT_SYMBOL(get_super_thawed
);
627 * get_active_super - get an active reference to the superblock of a device
628 * @bdev: device to get the superblock for
630 * Scans the superblock list and finds the superblock of the file system
631 * mounted on the device given. Returns the superblock with an active
632 * reference or %NULL if none was found.
634 struct super_block
*get_active_super(struct block_device
*bdev
)
636 struct super_block
*sb
;
643 list_for_each_entry(sb
, &super_blocks
, s_list
) {
644 if (hlist_unhashed(&sb
->s_instances
))
646 if (sb
->s_bdev
== bdev
) {
649 up_write(&sb
->s_umount
);
653 spin_unlock(&sb_lock
);
657 struct super_block
*user_get_super(dev_t dev
)
659 struct super_block
*sb
;
663 list_for_each_entry(sb
, &super_blocks
, s_list
) {
664 if (hlist_unhashed(&sb
->s_instances
))
666 if (sb
->s_dev
== dev
) {
668 spin_unlock(&sb_lock
);
669 down_read(&sb
->s_umount
);
671 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
673 up_read(&sb
->s_umount
);
674 /* nope, got unmounted */
680 spin_unlock(&sb_lock
);
685 * do_remount_sb - asks filesystem to change mount options.
686 * @sb: superblock in question
687 * @flags: numeric part of options
688 * @data: the rest of options
689 * @force: whether or not to force the change
691 * Alters the mount options of a mounted file system.
693 int do_remount_sb(struct super_block
*sb
, int flags
, void *data
, int force
)
698 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
)
702 if (!(flags
& MS_RDONLY
) && bdev_read_only(sb
->s_bdev
))
706 remount_ro
= (flags
& MS_RDONLY
) && !(sb
->s_flags
& MS_RDONLY
);
709 if (sb
->s_pins
.first
) {
710 up_write(&sb
->s_umount
);
712 down_write(&sb
->s_umount
);
715 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
)
717 remount_ro
= (flags
& MS_RDONLY
) && !(sb
->s_flags
& MS_RDONLY
);
720 shrink_dcache_sb(sb
);
722 /* If we are remounting RDONLY and current sb is read/write,
723 make sure there are no rw files opened */
726 sb
->s_readonly_remount
= 1;
729 retval
= sb_prepare_remount_readonly(sb
);
735 if (sb
->s_op
->remount_fs
) {
736 retval
= sb
->s_op
->remount_fs(sb
, &flags
, data
);
739 goto cancel_readonly
;
740 /* If forced remount, go ahead despite any errors */
741 WARN(1, "forced remount of a %s fs returned %i\n",
742 sb
->s_type
->name
, retval
);
745 sb
->s_flags
= (sb
->s_flags
& ~MS_RMT_MASK
) | (flags
& MS_RMT_MASK
);
746 /* Needs to be ordered wrt mnt_is_readonly() */
748 sb
->s_readonly_remount
= 0;
751 * Some filesystems modify their metadata via some other path than the
752 * bdev buffer cache (eg. use a private mapping, or directories in
753 * pagecache, etc). Also file data modifications go via their own
754 * mappings. So If we try to mount readonly then copy the filesystem
755 * from bdev, we could get stale data, so invalidate it to give a best
756 * effort at coherency.
758 if (remount_ro
&& sb
->s_bdev
)
759 invalidate_bdev(sb
->s_bdev
);
763 sb
->s_readonly_remount
= 0;
767 static void do_emergency_remount(struct work_struct
*work
)
769 struct super_block
*sb
, *p
= NULL
;
772 list_for_each_entry(sb
, &super_blocks
, s_list
) {
773 if (hlist_unhashed(&sb
->s_instances
))
776 spin_unlock(&sb_lock
);
777 down_write(&sb
->s_umount
);
778 if (sb
->s_root
&& sb
->s_bdev
&& (sb
->s_flags
& MS_BORN
) &&
779 !(sb
->s_flags
& MS_RDONLY
)) {
781 * What lock protects sb->s_flags??
783 do_remount_sb(sb
, MS_RDONLY
, NULL
, 1);
785 up_write(&sb
->s_umount
);
793 spin_unlock(&sb_lock
);
795 printk("Emergency Remount complete\n");
798 void emergency_remount(void)
800 struct work_struct
*work
;
802 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
804 INIT_WORK(work
, do_emergency_remount
);
810 * Unnamed block devices are dummy devices used by virtual
811 * filesystems which don't use real block-devices. -- jrs
814 static DEFINE_IDA(unnamed_dev_ida
);
815 static DEFINE_SPINLOCK(unnamed_dev_lock
);/* protects the above */
816 /* Many userspace utilities consider an FSID of 0 invalid.
817 * Always return at least 1 from get_anon_bdev.
819 static int unnamed_dev_start
= 1;
821 int get_anon_bdev(dev_t
*p
)
827 if (ida_pre_get(&unnamed_dev_ida
, GFP_ATOMIC
) == 0)
829 spin_lock(&unnamed_dev_lock
);
830 error
= ida_get_new_above(&unnamed_dev_ida
, unnamed_dev_start
, &dev
);
832 unnamed_dev_start
= dev
+ 1;
833 spin_unlock(&unnamed_dev_lock
);
834 if (error
== -EAGAIN
)
835 /* We raced and lost with another CPU. */
840 if (dev
== (1 << MINORBITS
)) {
841 spin_lock(&unnamed_dev_lock
);
842 ida_remove(&unnamed_dev_ida
, dev
);
843 if (unnamed_dev_start
> dev
)
844 unnamed_dev_start
= dev
;
845 spin_unlock(&unnamed_dev_lock
);
848 *p
= MKDEV(0, dev
& MINORMASK
);
851 EXPORT_SYMBOL(get_anon_bdev
);
853 void free_anon_bdev(dev_t dev
)
855 int slot
= MINOR(dev
);
856 spin_lock(&unnamed_dev_lock
);
857 ida_remove(&unnamed_dev_ida
, slot
);
858 if (slot
< unnamed_dev_start
)
859 unnamed_dev_start
= slot
;
860 spin_unlock(&unnamed_dev_lock
);
862 EXPORT_SYMBOL(free_anon_bdev
);
864 int set_anon_super(struct super_block
*s
, void *data
)
866 return get_anon_bdev(&s
->s_dev
);
869 EXPORT_SYMBOL(set_anon_super
);
871 void kill_anon_super(struct super_block
*sb
)
873 dev_t dev
= sb
->s_dev
;
874 generic_shutdown_super(sb
);
878 EXPORT_SYMBOL(kill_anon_super
);
880 void kill_litter_super(struct super_block
*sb
)
883 d_genocide(sb
->s_root
);
887 EXPORT_SYMBOL(kill_litter_super
);
889 static int ns_test_super(struct super_block
*sb
, void *data
)
891 return sb
->s_fs_info
== data
;
894 static int ns_set_super(struct super_block
*sb
, void *data
)
896 sb
->s_fs_info
= data
;
897 return set_anon_super(sb
, NULL
);
900 struct dentry
*mount_ns(struct file_system_type
*fs_type
, int flags
,
901 void *data
, int (*fill_super
)(struct super_block
*, void *, int))
903 struct super_block
*sb
;
905 sb
= sget(fs_type
, ns_test_super
, ns_set_super
, flags
, data
);
911 err
= fill_super(sb
, data
, flags
& MS_SILENT
? 1 : 0);
913 deactivate_locked_super(sb
);
917 sb
->s_flags
|= MS_ACTIVE
;
920 return dget(sb
->s_root
);
923 EXPORT_SYMBOL(mount_ns
);
926 static int set_bdev_super(struct super_block
*s
, void *data
)
929 s
->s_dev
= s
->s_bdev
->bd_dev
;
932 * We set the bdi here to the queue backing, file systems can
933 * overwrite this in ->fill_super()
935 s
->s_bdi
= &bdev_get_queue(s
->s_bdev
)->backing_dev_info
;
939 static int test_bdev_super(struct super_block
*s
, void *data
)
941 return (void *)s
->s_bdev
== data
;
944 struct dentry
*mount_bdev(struct file_system_type
*fs_type
,
945 int flags
, const char *dev_name
, void *data
,
946 int (*fill_super
)(struct super_block
*, void *, int))
948 struct block_device
*bdev
;
949 struct super_block
*s
;
950 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
953 if (!(flags
& MS_RDONLY
))
956 bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
958 return ERR_CAST(bdev
);
961 * once the super is inserted into the list by sget, s_umount
962 * will protect the lockfs code from trying to start a snapshot
963 * while we are mounting
965 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
966 if (bdev
->bd_fsfreeze_count
> 0) {
967 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
971 s
= sget(fs_type
, test_bdev_super
, set_bdev_super
, flags
| MS_NOSEC
,
973 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
978 if ((flags
^ s
->s_flags
) & MS_RDONLY
) {
979 deactivate_locked_super(s
);
985 * s_umount nests inside bd_mutex during
986 * __invalidate_device(). blkdev_put() acquires
987 * bd_mutex and can't be called under s_umount. Drop
988 * s_umount temporarily. This is safe as we're
989 * holding an active reference.
991 up_write(&s
->s_umount
);
992 blkdev_put(bdev
, mode
);
993 down_write(&s
->s_umount
);
995 char b
[BDEVNAME_SIZE
];
998 strlcpy(s
->s_id
, bdevname(bdev
, b
), sizeof(s
->s_id
));
999 sb_set_blocksize(s
, block_size(bdev
));
1000 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1002 deactivate_locked_super(s
);
1006 s
->s_flags
|= MS_ACTIVE
;
1010 return dget(s
->s_root
);
1015 blkdev_put(bdev
, mode
);
1017 return ERR_PTR(error
);
1019 EXPORT_SYMBOL(mount_bdev
);
1021 void kill_block_super(struct super_block
*sb
)
1023 struct block_device
*bdev
= sb
->s_bdev
;
1024 fmode_t mode
= sb
->s_mode
;
1026 bdev
->bd_super
= NULL
;
1027 generic_shutdown_super(sb
);
1028 sync_blockdev(bdev
);
1029 WARN_ON_ONCE(!(mode
& FMODE_EXCL
));
1030 blkdev_put(bdev
, mode
| FMODE_EXCL
);
1033 EXPORT_SYMBOL(kill_block_super
);
1036 struct dentry
*mount_nodev(struct file_system_type
*fs_type
,
1037 int flags
, void *data
,
1038 int (*fill_super
)(struct super_block
*, void *, int))
1041 struct super_block
*s
= sget(fs_type
, NULL
, set_anon_super
, flags
, NULL
);
1046 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1048 deactivate_locked_super(s
);
1049 return ERR_PTR(error
);
1051 s
->s_flags
|= MS_ACTIVE
;
1052 return dget(s
->s_root
);
1054 EXPORT_SYMBOL(mount_nodev
);
1056 static int compare_single(struct super_block
*s
, void *p
)
1061 struct dentry
*mount_single(struct file_system_type
*fs_type
,
1062 int flags
, void *data
,
1063 int (*fill_super
)(struct super_block
*, void *, int))
1065 struct super_block
*s
;
1068 s
= sget(fs_type
, compare_single
, set_anon_super
, flags
, NULL
);
1072 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1074 deactivate_locked_super(s
);
1075 return ERR_PTR(error
);
1077 s
->s_flags
|= MS_ACTIVE
;
1079 do_remount_sb(s
, flags
, data
, 0);
1081 return dget(s
->s_root
);
1083 EXPORT_SYMBOL(mount_single
);
1086 mount_fs(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
1088 struct dentry
*root
;
1089 struct super_block
*sb
;
1090 char *secdata
= NULL
;
1091 int error
= -ENOMEM
;
1093 if (data
&& !(type
->fs_flags
& FS_BINARY_MOUNTDATA
)) {
1094 secdata
= alloc_secdata();
1098 error
= security_sb_copy_data(data
, secdata
);
1100 goto out_free_secdata
;
1103 root
= type
->mount(type
, flags
, name
, data
);
1105 error
= PTR_ERR(root
);
1106 goto out_free_secdata
;
1110 WARN_ON(!sb
->s_bdi
);
1111 sb
->s_flags
|= MS_BORN
;
1113 error
= security_sb_kern_mount(sb
, flags
, secdata
);
1118 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1119 * but s_maxbytes was an unsigned long long for many releases. Throw
1120 * this warning for a little while to try and catch filesystems that
1121 * violate this rule.
1123 WARN((sb
->s_maxbytes
< 0), "%s set sb->s_maxbytes to "
1124 "negative value (%lld)\n", type
->name
, sb
->s_maxbytes
);
1126 up_write(&sb
->s_umount
);
1127 free_secdata(secdata
);
1131 deactivate_locked_super(sb
);
1133 free_secdata(secdata
);
1135 return ERR_PTR(error
);
1139 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1142 void __sb_end_write(struct super_block
*sb
, int level
)
1144 percpu_counter_dec(&sb
->s_writers
.counter
[level
-1]);
1146 * Make sure s_writers are updated before we wake up waiters in
1150 if (waitqueue_active(&sb
->s_writers
.wait
))
1151 wake_up(&sb
->s_writers
.wait
);
1152 rwsem_release(&sb
->s_writers
.lock_map
[level
-1], 1, _RET_IP_
);
1154 EXPORT_SYMBOL(__sb_end_write
);
1156 #ifdef CONFIG_LOCKDEP
1158 * We want lockdep to tell us about possible deadlocks with freezing but
1159 * it's it bit tricky to properly instrument it. Getting a freeze protection
1160 * works as getting a read lock but there are subtle problems. XFS for example
1161 * gets freeze protection on internal level twice in some cases, which is OK
1162 * only because we already hold a freeze protection also on higher level. Due
1163 * to these cases we have to tell lockdep we are doing trylock when we
1164 * already hold a freeze protection for a higher freeze level.
1166 static void acquire_freeze_lock(struct super_block
*sb
, int level
, bool trylock
,
1172 for (i
= 0; i
< level
- 1; i
++)
1173 if (lock_is_held(&sb
->s_writers
.lock_map
[i
])) {
1178 rwsem_acquire_read(&sb
->s_writers
.lock_map
[level
-1], 0, trylock
, ip
);
1183 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1186 int __sb_start_write(struct super_block
*sb
, int level
, bool wait
)
1189 if (unlikely(sb
->s_writers
.frozen
>= level
)) {
1192 wait_event(sb
->s_writers
.wait_unfrozen
,
1193 sb
->s_writers
.frozen
< level
);
1196 #ifdef CONFIG_LOCKDEP
1197 acquire_freeze_lock(sb
, level
, !wait
, _RET_IP_
);
1199 percpu_counter_inc(&sb
->s_writers
.counter
[level
-1]);
1201 * Make sure counter is updated before we check for frozen.
1202 * freeze_super() first sets frozen and then checks the counter.
1205 if (unlikely(sb
->s_writers
.frozen
>= level
)) {
1206 __sb_end_write(sb
, level
);
1211 EXPORT_SYMBOL(__sb_start_write
);
1214 * sb_wait_write - wait until all writers to given file system finish
1215 * @sb: the super for which we wait
1216 * @level: type of writers we wait for (normal vs page fault)
1218 * This function waits until there are no writers of given type to given file
1219 * system. Caller of this function should make sure there can be no new writers
1220 * of type @level before calling this function. Otherwise this function can
1223 static void sb_wait_write(struct super_block
*sb
, int level
)
1228 * We just cycle-through lockdep here so that it does not complain
1229 * about returning with lock to userspace
1231 rwsem_acquire(&sb
->s_writers
.lock_map
[level
-1], 0, 0, _THIS_IP_
);
1232 rwsem_release(&sb
->s_writers
.lock_map
[level
-1], 1, _THIS_IP_
);
1238 * We use a barrier in prepare_to_wait() to separate setting
1239 * of frozen and checking of the counter
1241 prepare_to_wait(&sb
->s_writers
.wait
, &wait
,
1242 TASK_UNINTERRUPTIBLE
);
1244 writers
= percpu_counter_sum(&sb
->s_writers
.counter
[level
-1]);
1248 finish_wait(&sb
->s_writers
.wait
, &wait
);
1253 * freeze_super - lock the filesystem and force it into a consistent state
1254 * @sb: the super to lock
1256 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1257 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1260 * During this function, sb->s_writers.frozen goes through these values:
1262 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1264 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1265 * writes should be blocked, though page faults are still allowed. We wait for
1266 * all writes to complete and then proceed to the next stage.
1268 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1269 * but internal fs threads can still modify the filesystem (although they
1270 * should not dirty new pages or inodes), writeback can run etc. After waiting
1271 * for all running page faults we sync the filesystem which will clean all
1272 * dirty pages and inodes (no new dirty pages or inodes can be created when
1275 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1276 * modification are blocked (e.g. XFS preallocation truncation on inode
1277 * reclaim). This is usually implemented by blocking new transactions for
1278 * filesystems that have them and need this additional guard. After all
1279 * internal writers are finished we call ->freeze_fs() to finish filesystem
1280 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1281 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1283 * sb->s_writers.frozen is protected by sb->s_umount.
1285 int freeze_super(struct super_block
*sb
)
1289 atomic_inc(&sb
->s_active
);
1290 down_write(&sb
->s_umount
);
1291 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
) {
1292 deactivate_locked_super(sb
);
1296 if (!(sb
->s_flags
& MS_BORN
)) {
1297 up_write(&sb
->s_umount
);
1298 return 0; /* sic - it's "nothing to do" */
1301 if (sb
->s_flags
& MS_RDONLY
) {
1302 /* Nothing to do really... */
1303 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1304 up_write(&sb
->s_umount
);
1308 /* From now on, no new normal writers can start */
1309 sb
->s_writers
.frozen
= SB_FREEZE_WRITE
;
1312 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1313 up_write(&sb
->s_umount
);
1315 sb_wait_write(sb
, SB_FREEZE_WRITE
);
1317 /* Now we go and block page faults... */
1318 down_write(&sb
->s_umount
);
1319 sb
->s_writers
.frozen
= SB_FREEZE_PAGEFAULT
;
1322 sb_wait_write(sb
, SB_FREEZE_PAGEFAULT
);
1324 /* All writers are done so after syncing there won't be dirty data */
1325 sync_filesystem(sb
);
1327 /* Now wait for internal filesystem counter */
1328 sb
->s_writers
.frozen
= SB_FREEZE_FS
;
1330 sb_wait_write(sb
, SB_FREEZE_FS
);
1332 if (sb
->s_op
->freeze_fs
) {
1333 ret
= sb
->s_op
->freeze_fs(sb
);
1336 "VFS:Filesystem freeze failed\n");
1337 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1339 wake_up(&sb
->s_writers
.wait_unfrozen
);
1340 deactivate_locked_super(sb
);
1345 * This is just for debugging purposes so that fs can warn if it
1346 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1348 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1349 up_write(&sb
->s_umount
);
1352 EXPORT_SYMBOL(freeze_super
);
1355 * thaw_super -- unlock filesystem
1356 * @sb: the super to thaw
1358 * Unlocks the filesystem and marks it writeable again after freeze_super().
1360 int thaw_super(struct super_block
*sb
)
1364 down_write(&sb
->s_umount
);
1365 if (sb
->s_writers
.frozen
== SB_UNFROZEN
) {
1366 up_write(&sb
->s_umount
);
1370 if (sb
->s_flags
& MS_RDONLY
)
1373 if (sb
->s_op
->unfreeze_fs
) {
1374 error
= sb
->s_op
->unfreeze_fs(sb
);
1377 "VFS:Filesystem thaw failed\n");
1378 up_write(&sb
->s_umount
);
1384 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1386 wake_up(&sb
->s_writers
.wait_unfrozen
);
1387 deactivate_locked_super(sb
);
1391 EXPORT_SYMBOL(thaw_super
);