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 unsigned long super_cache_scan(struct shrinker
*shrink
,
57 struct shrink_control
*sc
)
59 struct super_block
*sb
;
66 sb
= container_of(shrink
, struct super_block
, s_shrink
);
69 * Deadlock avoidance. We may hold various FS locks, and we don't want
70 * to recurse into the FS that called us in clear_inode() and friends..
72 if (!(sc
->gfp_mask
& __GFP_FS
))
75 if (!grab_super_passive(sb
))
78 if (sb
->s_op
->nr_cached_objects
)
79 fs_objects
= sb
->s_op
->nr_cached_objects(sb
);
81 inodes
= list_lru_count(&sb
->s_inode_lru
);
82 total_objects
= sb
->s_nr_dentry_unused
+ inodes
+ fs_objects
+ 1;
84 /* proportion the scan between the caches */
85 dentries
= mult_frac(sc
->nr_to_scan
, sb
->s_nr_dentry_unused
,
87 inodes
= mult_frac(sc
->nr_to_scan
, inodes
, total_objects
);
90 * prune the dcache first as the icache is pinned by it, then
91 * prune the icache, followed by the filesystem specific caches
93 freed
= prune_dcache_sb(sb
, dentries
);
94 freed
+= prune_icache_sb(sb
, inodes
);
97 fs_objects
= mult_frac(sc
->nr_to_scan
, fs_objects
,
99 freed
+= sb
->s_op
->free_cached_objects(sb
, fs_objects
);
106 static unsigned long super_cache_count(struct shrinker
*shrink
,
107 struct shrink_control
*sc
)
109 struct super_block
*sb
;
110 long total_objects
= 0;
112 sb
= container_of(shrink
, struct super_block
, s_shrink
);
114 if (!grab_super_passive(sb
))
117 if (sb
->s_op
&& sb
->s_op
->nr_cached_objects
)
118 total_objects
= sb
->s_op
->nr_cached_objects(sb
);
120 total_objects
+= sb
->s_nr_dentry_unused
;
121 total_objects
+= list_lru_count(&sb
->s_inode_lru
);
123 total_objects
= vfs_pressure_ratio(total_objects
);
125 return total_objects
;
128 static int init_sb_writers(struct super_block
*s
, struct file_system_type
*type
)
133 for (i
= 0; i
< SB_FREEZE_LEVELS
; i
++) {
134 err
= percpu_counter_init(&s
->s_writers
.counter
[i
], 0);
137 lockdep_init_map(&s
->s_writers
.lock_map
[i
], sb_writers_name
[i
],
138 &type
->s_writers_key
[i
], 0);
140 init_waitqueue_head(&s
->s_writers
.wait
);
141 init_waitqueue_head(&s
->s_writers
.wait_unfrozen
);
145 percpu_counter_destroy(&s
->s_writers
.counter
[i
]);
149 static void destroy_sb_writers(struct super_block
*s
)
153 for (i
= 0; i
< SB_FREEZE_LEVELS
; i
++)
154 percpu_counter_destroy(&s
->s_writers
.counter
[i
]);
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
;
171 if (security_sb_alloc(s
))
175 s
->s_files
= alloc_percpu(struct list_head
);
181 for_each_possible_cpu(i
)
182 INIT_LIST_HEAD(per_cpu_ptr(s
->s_files
, i
));
185 INIT_LIST_HEAD(&s
->s_files
);
187 if (init_sb_writers(s
, type
))
190 s
->s_bdi
= &default_backing_dev_info
;
191 INIT_HLIST_NODE(&s
->s_instances
);
192 INIT_HLIST_BL_HEAD(&s
->s_anon
);
193 INIT_LIST_HEAD(&s
->s_inodes
);
194 INIT_LIST_HEAD(&s
->s_dentry_lru
);
195 spin_lock_init(&s
->s_dentry_lru_lock
);
196 list_lru_init(&s
->s_inode_lru
);
197 INIT_LIST_HEAD(&s
->s_mounts
);
198 init_rwsem(&s
->s_umount
);
199 lockdep_set_class(&s
->s_umount
, &type
->s_umount_key
);
201 * sget() can have s_umount recursion.
203 * When it cannot find a suitable sb, it allocates a new
204 * one (this one), and tries again to find a suitable old
207 * In case that succeeds, it will acquire the s_umount
208 * lock of the old one. Since these are clearly distrinct
209 * locks, and this object isn't exposed yet, there's no
212 * Annotate this by putting this lock in a different
215 down_write_nested(&s
->s_umount
, SINGLE_DEPTH_NESTING
);
217 atomic_set(&s
->s_active
, 1);
218 mutex_init(&s
->s_vfs_rename_mutex
);
219 lockdep_set_class(&s
->s_vfs_rename_mutex
, &type
->s_vfs_rename_key
);
220 mutex_init(&s
->s_dquot
.dqio_mutex
);
221 mutex_init(&s
->s_dquot
.dqonoff_mutex
);
222 init_rwsem(&s
->s_dquot
.dqptr_sem
);
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;
239 free_percpu(s
->s_files
);
241 destroy_sb_writers(s
);
249 * destroy_super - frees a superblock
250 * @s: superblock to free
252 * Frees a superblock.
254 static inline void destroy_super(struct super_block
*s
)
257 free_percpu(s
->s_files
);
259 destroy_sb_writers(s
);
261 WARN_ON(!list_empty(&s
->s_mounts
));
267 /* Superblock refcounting */
270 * Drop a superblock's refcount. The caller must hold sb_lock.
272 static void __put_super(struct super_block
*sb
)
274 if (!--sb
->s_count
) {
275 list_del_init(&sb
->s_list
);
281 * put_super - drop a temporary reference to superblock
282 * @sb: superblock in question
284 * Drops a temporary reference, frees superblock if there's no
287 static void put_super(struct super_block
*sb
)
291 spin_unlock(&sb_lock
);
296 * deactivate_locked_super - drop an active reference to superblock
297 * @s: superblock to deactivate
299 * Drops an active reference to superblock, converting it into a temprory
300 * one if there is no other active references left. In that case we
301 * tell fs driver to shut it down and drop the temporary reference we
304 * Caller holds exclusive lock on superblock; that lock is released.
306 void deactivate_locked_super(struct super_block
*s
)
308 struct file_system_type
*fs
= s
->s_type
;
309 if (atomic_dec_and_test(&s
->s_active
)) {
310 cleancache_invalidate_fs(s
);
313 /* caches are now gone, we can safely kill the shrinker now */
314 unregister_shrinker(&s
->s_shrink
);
318 up_write(&s
->s_umount
);
322 EXPORT_SYMBOL(deactivate_locked_super
);
325 * deactivate_super - drop an active reference to superblock
326 * @s: superblock to deactivate
328 * Variant of deactivate_locked_super(), except that superblock is *not*
329 * locked by caller. If we are going to drop the final active reference,
330 * lock will be acquired prior to that.
332 void deactivate_super(struct super_block
*s
)
334 if (!atomic_add_unless(&s
->s_active
, -1, 1)) {
335 down_write(&s
->s_umount
);
336 deactivate_locked_super(s
);
340 EXPORT_SYMBOL(deactivate_super
);
343 * grab_super - acquire an active reference
344 * @s: reference we are trying to make active
346 * Tries to acquire an active reference. grab_super() is used when we
347 * had just found a superblock in super_blocks or fs_type->fs_supers
348 * and want to turn it into a full-blown active reference. grab_super()
349 * is called with sb_lock held and drops it. Returns 1 in case of
350 * success, 0 if we had failed (superblock contents was already dead or
351 * dying when grab_super() had been called). Note that this is only
352 * called for superblocks not in rundown mode (== ones still on ->fs_supers
353 * of their type), so increment of ->s_count is OK here.
355 static int grab_super(struct super_block
*s
) __releases(sb_lock
)
358 spin_unlock(&sb_lock
);
359 down_write(&s
->s_umount
);
360 if ((s
->s_flags
& MS_BORN
) && atomic_inc_not_zero(&s
->s_active
)) {
364 up_write(&s
->s_umount
);
370 * grab_super_passive - acquire a passive reference
371 * @sb: reference we are trying to grab
373 * Tries to acquire a passive reference. This is used in places where we
374 * cannot take an active reference but we need to ensure that the
375 * superblock does not go away while we are working on it. It returns
376 * false if a reference was not gained, and returns true with the s_umount
377 * lock held in read mode if a reference is gained. On successful return,
378 * the caller must drop the s_umount lock and the passive reference when
381 bool grab_super_passive(struct super_block
*sb
)
384 if (hlist_unhashed(&sb
->s_instances
)) {
385 spin_unlock(&sb_lock
);
390 spin_unlock(&sb_lock
);
392 if (down_read_trylock(&sb
->s_umount
)) {
393 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
395 up_read(&sb
->s_umount
);
403 * generic_shutdown_super - common helper for ->kill_sb()
404 * @sb: superblock to kill
406 * generic_shutdown_super() does all fs-independent work on superblock
407 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
408 * that need destruction out of superblock, call generic_shutdown_super()
409 * and release aforementioned objects. Note: dentries and inodes _are_
410 * taken care of and do not need specific handling.
412 * Upon calling this function, the filesystem may no longer alter or
413 * rearrange the set of dentries belonging to this super_block, nor may it
414 * change the attachments of dentries to inodes.
416 void generic_shutdown_super(struct super_block
*sb
)
418 const struct super_operations
*sop
= sb
->s_op
;
421 shrink_dcache_for_umount(sb
);
423 sb
->s_flags
&= ~MS_ACTIVE
;
425 fsnotify_unmount_inodes(&sb
->s_inodes
);
429 if (sb
->s_dio_done_wq
) {
430 destroy_workqueue(sb
->s_dio_done_wq
);
431 sb
->s_dio_done_wq
= NULL
;
437 if (!list_empty(&sb
->s_inodes
)) {
438 printk("VFS: Busy inodes after unmount of %s. "
439 "Self-destruct in 5 seconds. Have a nice day...\n",
444 /* should be initialized for __put_super_and_need_restart() */
445 hlist_del_init(&sb
->s_instances
);
446 spin_unlock(&sb_lock
);
447 up_write(&sb
->s_umount
);
450 EXPORT_SYMBOL(generic_shutdown_super
);
453 * sget - find or create a superblock
454 * @type: filesystem type superblock should belong to
455 * @test: comparison callback
456 * @set: setup callback
457 * @flags: mount flags
458 * @data: argument to each of them
460 struct super_block
*sget(struct file_system_type
*type
,
461 int (*test
)(struct super_block
*,void *),
462 int (*set
)(struct super_block
*,void *),
466 struct super_block
*s
= NULL
;
467 struct super_block
*old
;
473 hlist_for_each_entry(old
, &type
->fs_supers
, s_instances
) {
474 if (!test(old
, data
))
476 if (!grab_super(old
))
479 up_write(&s
->s_umount
);
487 spin_unlock(&sb_lock
);
488 s
= alloc_super(type
, flags
);
490 return ERR_PTR(-ENOMEM
);
496 spin_unlock(&sb_lock
);
497 up_write(&s
->s_umount
);
502 strlcpy(s
->s_id
, type
->name
, sizeof(s
->s_id
));
503 list_add_tail(&s
->s_list
, &super_blocks
);
504 hlist_add_head(&s
->s_instances
, &type
->fs_supers
);
505 spin_unlock(&sb_lock
);
506 get_filesystem(type
);
507 register_shrinker(&s
->s_shrink
);
513 void drop_super(struct super_block
*sb
)
515 up_read(&sb
->s_umount
);
519 EXPORT_SYMBOL(drop_super
);
522 * iterate_supers - call function for all active superblocks
523 * @f: function to call
524 * @arg: argument to pass to it
526 * Scans the superblock list and calls given function, passing it
527 * locked superblock and given argument.
529 void iterate_supers(void (*f
)(struct super_block
*, void *), void *arg
)
531 struct super_block
*sb
, *p
= NULL
;
534 list_for_each_entry(sb
, &super_blocks
, s_list
) {
535 if (hlist_unhashed(&sb
->s_instances
))
538 spin_unlock(&sb_lock
);
540 down_read(&sb
->s_umount
);
541 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
543 up_read(&sb
->s_umount
);
552 spin_unlock(&sb_lock
);
556 * iterate_supers_type - call function for superblocks of given type
558 * @f: function to call
559 * @arg: argument to pass to it
561 * Scans the superblock list and calls given function, passing it
562 * locked superblock and given argument.
564 void iterate_supers_type(struct file_system_type
*type
,
565 void (*f
)(struct super_block
*, void *), void *arg
)
567 struct super_block
*sb
, *p
= NULL
;
570 hlist_for_each_entry(sb
, &type
->fs_supers
, s_instances
) {
572 spin_unlock(&sb_lock
);
574 down_read(&sb
->s_umount
);
575 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
577 up_read(&sb
->s_umount
);
586 spin_unlock(&sb_lock
);
589 EXPORT_SYMBOL(iterate_supers_type
);
592 * get_super - get the superblock of a device
593 * @bdev: device to get the superblock for
595 * Scans the superblock list and finds the superblock of the file system
596 * mounted on the device given. %NULL is returned if no match is found.
599 struct super_block
*get_super(struct block_device
*bdev
)
601 struct super_block
*sb
;
608 list_for_each_entry(sb
, &super_blocks
, s_list
) {
609 if (hlist_unhashed(&sb
->s_instances
))
611 if (sb
->s_bdev
== bdev
) {
613 spin_unlock(&sb_lock
);
614 down_read(&sb
->s_umount
);
616 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
618 up_read(&sb
->s_umount
);
619 /* nope, got unmounted */
625 spin_unlock(&sb_lock
);
629 EXPORT_SYMBOL(get_super
);
632 * get_super_thawed - get thawed superblock of a device
633 * @bdev: device to get the superblock for
635 * Scans the superblock list and finds the superblock of the file system
636 * mounted on the device. The superblock is returned once it is thawed
637 * (or immediately if it was not frozen). %NULL is returned if no match
640 struct super_block
*get_super_thawed(struct block_device
*bdev
)
643 struct super_block
*s
= get_super(bdev
);
644 if (!s
|| s
->s_writers
.frozen
== SB_UNFROZEN
)
646 up_read(&s
->s_umount
);
647 wait_event(s
->s_writers
.wait_unfrozen
,
648 s
->s_writers
.frozen
== SB_UNFROZEN
);
652 EXPORT_SYMBOL(get_super_thawed
);
655 * get_active_super - get an active reference to the superblock of a device
656 * @bdev: device to get the superblock for
658 * Scans the superblock list and finds the superblock of the file system
659 * mounted on the device given. Returns the superblock with an active
660 * reference or %NULL if none was found.
662 struct super_block
*get_active_super(struct block_device
*bdev
)
664 struct super_block
*sb
;
671 list_for_each_entry(sb
, &super_blocks
, s_list
) {
672 if (hlist_unhashed(&sb
->s_instances
))
674 if (sb
->s_bdev
== bdev
) {
677 up_write(&sb
->s_umount
);
681 spin_unlock(&sb_lock
);
685 struct super_block
*user_get_super(dev_t dev
)
687 struct super_block
*sb
;
691 list_for_each_entry(sb
, &super_blocks
, s_list
) {
692 if (hlist_unhashed(&sb
->s_instances
))
694 if (sb
->s_dev
== dev
) {
696 spin_unlock(&sb_lock
);
697 down_read(&sb
->s_umount
);
699 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
701 up_read(&sb
->s_umount
);
702 /* nope, got unmounted */
708 spin_unlock(&sb_lock
);
713 * do_remount_sb - asks filesystem to change mount options.
714 * @sb: superblock in question
715 * @flags: numeric part of options
716 * @data: the rest of options
717 * @force: whether or not to force the change
719 * Alters the mount options of a mounted file system.
721 int do_remount_sb(struct super_block
*sb
, int flags
, void *data
, int force
)
726 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
)
730 if (!(flags
& MS_RDONLY
) && bdev_read_only(sb
->s_bdev
))
734 if (flags
& MS_RDONLY
)
736 shrink_dcache_sb(sb
);
739 remount_ro
= (flags
& MS_RDONLY
) && !(sb
->s_flags
& MS_RDONLY
);
741 /* If we are remounting RDONLY and current sb is read/write,
742 make sure there are no rw files opened */
747 retval
= sb_prepare_remount_readonly(sb
);
753 if (sb
->s_op
->remount_fs
) {
754 retval
= sb
->s_op
->remount_fs(sb
, &flags
, data
);
757 goto cancel_readonly
;
758 /* If forced remount, go ahead despite any errors */
759 WARN(1, "forced remount of a %s fs returned %i\n",
760 sb
->s_type
->name
, retval
);
763 sb
->s_flags
= (sb
->s_flags
& ~MS_RMT_MASK
) | (flags
& MS_RMT_MASK
);
764 /* Needs to be ordered wrt mnt_is_readonly() */
766 sb
->s_readonly_remount
= 0;
769 * Some filesystems modify their metadata via some other path than the
770 * bdev buffer cache (eg. use a private mapping, or directories in
771 * pagecache, etc). Also file data modifications go via their own
772 * mappings. So If we try to mount readonly then copy the filesystem
773 * from bdev, we could get stale data, so invalidate it to give a best
774 * effort at coherency.
776 if (remount_ro
&& sb
->s_bdev
)
777 invalidate_bdev(sb
->s_bdev
);
781 sb
->s_readonly_remount
= 0;
785 static void do_emergency_remount(struct work_struct
*work
)
787 struct super_block
*sb
, *p
= NULL
;
790 list_for_each_entry(sb
, &super_blocks
, s_list
) {
791 if (hlist_unhashed(&sb
->s_instances
))
794 spin_unlock(&sb_lock
);
795 down_write(&sb
->s_umount
);
796 if (sb
->s_root
&& sb
->s_bdev
&& (sb
->s_flags
& MS_BORN
) &&
797 !(sb
->s_flags
& MS_RDONLY
)) {
799 * What lock protects sb->s_flags??
801 do_remount_sb(sb
, MS_RDONLY
, NULL
, 1);
803 up_write(&sb
->s_umount
);
811 spin_unlock(&sb_lock
);
813 printk("Emergency Remount complete\n");
816 void emergency_remount(void)
818 struct work_struct
*work
;
820 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
822 INIT_WORK(work
, do_emergency_remount
);
828 * Unnamed block devices are dummy devices used by virtual
829 * filesystems which don't use real block-devices. -- jrs
832 static DEFINE_IDA(unnamed_dev_ida
);
833 static DEFINE_SPINLOCK(unnamed_dev_lock
);/* protects the above */
834 static int unnamed_dev_start
= 0; /* don't bother trying below it */
836 int get_anon_bdev(dev_t
*p
)
842 if (ida_pre_get(&unnamed_dev_ida
, GFP_ATOMIC
) == 0)
844 spin_lock(&unnamed_dev_lock
);
845 error
= ida_get_new_above(&unnamed_dev_ida
, unnamed_dev_start
, &dev
);
847 unnamed_dev_start
= dev
+ 1;
848 spin_unlock(&unnamed_dev_lock
);
849 if (error
== -EAGAIN
)
850 /* We raced and lost with another CPU. */
855 if (dev
== (1 << MINORBITS
)) {
856 spin_lock(&unnamed_dev_lock
);
857 ida_remove(&unnamed_dev_ida
, dev
);
858 if (unnamed_dev_start
> dev
)
859 unnamed_dev_start
= dev
;
860 spin_unlock(&unnamed_dev_lock
);
863 *p
= MKDEV(0, dev
& MINORMASK
);
866 EXPORT_SYMBOL(get_anon_bdev
);
868 void free_anon_bdev(dev_t dev
)
870 int slot
= MINOR(dev
);
871 spin_lock(&unnamed_dev_lock
);
872 ida_remove(&unnamed_dev_ida
, slot
);
873 if (slot
< unnamed_dev_start
)
874 unnamed_dev_start
= slot
;
875 spin_unlock(&unnamed_dev_lock
);
877 EXPORT_SYMBOL(free_anon_bdev
);
879 int set_anon_super(struct super_block
*s
, void *data
)
881 int error
= get_anon_bdev(&s
->s_dev
);
883 s
->s_bdi
= &noop_backing_dev_info
;
887 EXPORT_SYMBOL(set_anon_super
);
889 void kill_anon_super(struct super_block
*sb
)
891 dev_t dev
= sb
->s_dev
;
892 generic_shutdown_super(sb
);
896 EXPORT_SYMBOL(kill_anon_super
);
898 void kill_litter_super(struct super_block
*sb
)
901 d_genocide(sb
->s_root
);
905 EXPORT_SYMBOL(kill_litter_super
);
907 static int ns_test_super(struct super_block
*sb
, void *data
)
909 return sb
->s_fs_info
== data
;
912 static int ns_set_super(struct super_block
*sb
, void *data
)
914 sb
->s_fs_info
= data
;
915 return set_anon_super(sb
, NULL
);
918 struct dentry
*mount_ns(struct file_system_type
*fs_type
, int flags
,
919 void *data
, int (*fill_super
)(struct super_block
*, void *, int))
921 struct super_block
*sb
;
923 sb
= sget(fs_type
, ns_test_super
, ns_set_super
, flags
, data
);
929 err
= fill_super(sb
, data
, flags
& MS_SILENT
? 1 : 0);
931 deactivate_locked_super(sb
);
935 sb
->s_flags
|= MS_ACTIVE
;
938 return dget(sb
->s_root
);
941 EXPORT_SYMBOL(mount_ns
);
944 static int set_bdev_super(struct super_block
*s
, void *data
)
947 s
->s_dev
= s
->s_bdev
->bd_dev
;
950 * We set the bdi here to the queue backing, file systems can
951 * overwrite this in ->fill_super()
953 s
->s_bdi
= &bdev_get_queue(s
->s_bdev
)->backing_dev_info
;
957 static int test_bdev_super(struct super_block
*s
, void *data
)
959 return (void *)s
->s_bdev
== data
;
962 struct dentry
*mount_bdev(struct file_system_type
*fs_type
,
963 int flags
, const char *dev_name
, void *data
,
964 int (*fill_super
)(struct super_block
*, void *, int))
966 struct block_device
*bdev
;
967 struct super_block
*s
;
968 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
971 if (!(flags
& MS_RDONLY
))
974 bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
976 return ERR_CAST(bdev
);
979 * once the super is inserted into the list by sget, s_umount
980 * will protect the lockfs code from trying to start a snapshot
981 * while we are mounting
983 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
984 if (bdev
->bd_fsfreeze_count
> 0) {
985 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
989 s
= sget(fs_type
, test_bdev_super
, set_bdev_super
, flags
| MS_NOSEC
,
991 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
996 if ((flags
^ s
->s_flags
) & MS_RDONLY
) {
997 deactivate_locked_super(s
);
1003 * s_umount nests inside bd_mutex during
1004 * __invalidate_device(). blkdev_put() acquires
1005 * bd_mutex and can't be called under s_umount. Drop
1006 * s_umount temporarily. This is safe as we're
1007 * holding an active reference.
1009 up_write(&s
->s_umount
);
1010 blkdev_put(bdev
, mode
);
1011 down_write(&s
->s_umount
);
1013 char b
[BDEVNAME_SIZE
];
1016 strlcpy(s
->s_id
, bdevname(bdev
, b
), sizeof(s
->s_id
));
1017 sb_set_blocksize(s
, block_size(bdev
));
1018 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1020 deactivate_locked_super(s
);
1024 s
->s_flags
|= MS_ACTIVE
;
1028 return dget(s
->s_root
);
1033 blkdev_put(bdev
, mode
);
1035 return ERR_PTR(error
);
1037 EXPORT_SYMBOL(mount_bdev
);
1039 void kill_block_super(struct super_block
*sb
)
1041 struct block_device
*bdev
= sb
->s_bdev
;
1042 fmode_t mode
= sb
->s_mode
;
1044 bdev
->bd_super
= NULL
;
1045 generic_shutdown_super(sb
);
1046 sync_blockdev(bdev
);
1047 WARN_ON_ONCE(!(mode
& FMODE_EXCL
));
1048 blkdev_put(bdev
, mode
| FMODE_EXCL
);
1051 EXPORT_SYMBOL(kill_block_super
);
1054 struct dentry
*mount_nodev(struct file_system_type
*fs_type
,
1055 int flags
, void *data
,
1056 int (*fill_super
)(struct super_block
*, void *, int))
1059 struct super_block
*s
= sget(fs_type
, NULL
, set_anon_super
, flags
, NULL
);
1064 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1066 deactivate_locked_super(s
);
1067 return ERR_PTR(error
);
1069 s
->s_flags
|= MS_ACTIVE
;
1070 return dget(s
->s_root
);
1072 EXPORT_SYMBOL(mount_nodev
);
1074 static int compare_single(struct super_block
*s
, void *p
)
1079 struct dentry
*mount_single(struct file_system_type
*fs_type
,
1080 int flags
, void *data
,
1081 int (*fill_super
)(struct super_block
*, void *, int))
1083 struct super_block
*s
;
1086 s
= sget(fs_type
, compare_single
, set_anon_super
, flags
, NULL
);
1090 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1092 deactivate_locked_super(s
);
1093 return ERR_PTR(error
);
1095 s
->s_flags
|= MS_ACTIVE
;
1097 do_remount_sb(s
, flags
, data
, 0);
1099 return dget(s
->s_root
);
1101 EXPORT_SYMBOL(mount_single
);
1104 mount_fs(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
1106 struct dentry
*root
;
1107 struct super_block
*sb
;
1108 char *secdata
= NULL
;
1109 int error
= -ENOMEM
;
1111 if (data
&& !(type
->fs_flags
& FS_BINARY_MOUNTDATA
)) {
1112 secdata
= alloc_secdata();
1116 error
= security_sb_copy_data(data
, secdata
);
1118 goto out_free_secdata
;
1121 root
= type
->mount(type
, flags
, name
, data
);
1123 error
= PTR_ERR(root
);
1124 goto out_free_secdata
;
1128 WARN_ON(!sb
->s_bdi
);
1129 WARN_ON(sb
->s_bdi
== &default_backing_dev_info
);
1130 sb
->s_flags
|= MS_BORN
;
1132 error
= security_sb_kern_mount(sb
, flags
, secdata
);
1137 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1138 * but s_maxbytes was an unsigned long long for many releases. Throw
1139 * this warning for a little while to try and catch filesystems that
1140 * violate this rule.
1142 WARN((sb
->s_maxbytes
< 0), "%s set sb->s_maxbytes to "
1143 "negative value (%lld)\n", type
->name
, sb
->s_maxbytes
);
1145 up_write(&sb
->s_umount
);
1146 free_secdata(secdata
);
1150 deactivate_locked_super(sb
);
1152 free_secdata(secdata
);
1154 return ERR_PTR(error
);
1158 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1161 void __sb_end_write(struct super_block
*sb
, int level
)
1163 percpu_counter_dec(&sb
->s_writers
.counter
[level
-1]);
1165 * Make sure s_writers are updated before we wake up waiters in
1169 if (waitqueue_active(&sb
->s_writers
.wait
))
1170 wake_up(&sb
->s_writers
.wait
);
1171 rwsem_release(&sb
->s_writers
.lock_map
[level
-1], 1, _RET_IP_
);
1173 EXPORT_SYMBOL(__sb_end_write
);
1175 #ifdef CONFIG_LOCKDEP
1177 * We want lockdep to tell us about possible deadlocks with freezing but
1178 * it's it bit tricky to properly instrument it. Getting a freeze protection
1179 * works as getting a read lock but there are subtle problems. XFS for example
1180 * gets freeze protection on internal level twice in some cases, which is OK
1181 * only because we already hold a freeze protection also on higher level. Due
1182 * to these cases we have to tell lockdep we are doing trylock when we
1183 * already hold a freeze protection for a higher freeze level.
1185 static void acquire_freeze_lock(struct super_block
*sb
, int level
, bool trylock
,
1191 for (i
= 0; i
< level
- 1; i
++)
1192 if (lock_is_held(&sb
->s_writers
.lock_map
[i
])) {
1197 rwsem_acquire_read(&sb
->s_writers
.lock_map
[level
-1], 0, trylock
, ip
);
1202 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1205 int __sb_start_write(struct super_block
*sb
, int level
, bool wait
)
1208 if (unlikely(sb
->s_writers
.frozen
>= level
)) {
1211 wait_event(sb
->s_writers
.wait_unfrozen
,
1212 sb
->s_writers
.frozen
< level
);
1215 #ifdef CONFIG_LOCKDEP
1216 acquire_freeze_lock(sb
, level
, !wait
, _RET_IP_
);
1218 percpu_counter_inc(&sb
->s_writers
.counter
[level
-1]);
1220 * Make sure counter is updated before we check for frozen.
1221 * freeze_super() first sets frozen and then checks the counter.
1224 if (unlikely(sb
->s_writers
.frozen
>= level
)) {
1225 __sb_end_write(sb
, level
);
1230 EXPORT_SYMBOL(__sb_start_write
);
1233 * sb_wait_write - wait until all writers to given file system finish
1234 * @sb: the super for which we wait
1235 * @level: type of writers we wait for (normal vs page fault)
1237 * This function waits until there are no writers of given type to given file
1238 * system. Caller of this function should make sure there can be no new writers
1239 * of type @level before calling this function. Otherwise this function can
1242 static void sb_wait_write(struct super_block
*sb
, int level
)
1247 * We just cycle-through lockdep here so that it does not complain
1248 * about returning with lock to userspace
1250 rwsem_acquire(&sb
->s_writers
.lock_map
[level
-1], 0, 0, _THIS_IP_
);
1251 rwsem_release(&sb
->s_writers
.lock_map
[level
-1], 1, _THIS_IP_
);
1257 * We use a barrier in prepare_to_wait() to separate setting
1258 * of frozen and checking of the counter
1260 prepare_to_wait(&sb
->s_writers
.wait
, &wait
,
1261 TASK_UNINTERRUPTIBLE
);
1263 writers
= percpu_counter_sum(&sb
->s_writers
.counter
[level
-1]);
1267 finish_wait(&sb
->s_writers
.wait
, &wait
);
1272 * freeze_super - lock the filesystem and force it into a consistent state
1273 * @sb: the super to lock
1275 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1276 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1279 * During this function, sb->s_writers.frozen goes through these values:
1281 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1283 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1284 * writes should be blocked, though page faults are still allowed. We wait for
1285 * all writes to complete and then proceed to the next stage.
1287 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1288 * but internal fs threads can still modify the filesystem (although they
1289 * should not dirty new pages or inodes), writeback can run etc. After waiting
1290 * for all running page faults we sync the filesystem which will clean all
1291 * dirty pages and inodes (no new dirty pages or inodes can be created when
1294 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1295 * modification are blocked (e.g. XFS preallocation truncation on inode
1296 * reclaim). This is usually implemented by blocking new transactions for
1297 * filesystems that have them and need this additional guard. After all
1298 * internal writers are finished we call ->freeze_fs() to finish filesystem
1299 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1300 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1302 * sb->s_writers.frozen is protected by sb->s_umount.
1304 int freeze_super(struct super_block
*sb
)
1308 atomic_inc(&sb
->s_active
);
1309 down_write(&sb
->s_umount
);
1310 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
) {
1311 deactivate_locked_super(sb
);
1315 if (!(sb
->s_flags
& MS_BORN
)) {
1316 up_write(&sb
->s_umount
);
1317 return 0; /* sic - it's "nothing to do" */
1320 if (sb
->s_flags
& MS_RDONLY
) {
1321 /* Nothing to do really... */
1322 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1323 up_write(&sb
->s_umount
);
1327 /* From now on, no new normal writers can start */
1328 sb
->s_writers
.frozen
= SB_FREEZE_WRITE
;
1331 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1332 up_write(&sb
->s_umount
);
1334 sb_wait_write(sb
, SB_FREEZE_WRITE
);
1336 /* Now we go and block page faults... */
1337 down_write(&sb
->s_umount
);
1338 sb
->s_writers
.frozen
= SB_FREEZE_PAGEFAULT
;
1341 sb_wait_write(sb
, SB_FREEZE_PAGEFAULT
);
1343 /* All writers are done so after syncing there won't be dirty data */
1344 sync_filesystem(sb
);
1346 /* Now wait for internal filesystem counter */
1347 sb
->s_writers
.frozen
= SB_FREEZE_FS
;
1349 sb_wait_write(sb
, SB_FREEZE_FS
);
1351 if (sb
->s_op
->freeze_fs
) {
1352 ret
= sb
->s_op
->freeze_fs(sb
);
1355 "VFS:Filesystem freeze failed\n");
1356 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1358 wake_up(&sb
->s_writers
.wait_unfrozen
);
1359 deactivate_locked_super(sb
);
1364 * This is just for debugging purposes so that fs can warn if it
1365 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1367 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1368 up_write(&sb
->s_umount
);
1371 EXPORT_SYMBOL(freeze_super
);
1374 * thaw_super -- unlock filesystem
1375 * @sb: the super to thaw
1377 * Unlocks the filesystem and marks it writeable again after freeze_super().
1379 int thaw_super(struct super_block
*sb
)
1383 down_write(&sb
->s_umount
);
1384 if (sb
->s_writers
.frozen
== SB_UNFROZEN
) {
1385 up_write(&sb
->s_umount
);
1389 if (sb
->s_flags
& MS_RDONLY
)
1392 if (sb
->s_op
->unfreeze_fs
) {
1393 error
= sb
->s_op
->unfreeze_fs(sb
);
1396 "VFS:Filesystem thaw failed\n");
1397 up_write(&sb
->s_umount
);
1403 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1405 wake_up(&sb
->s_writers
.wait_unfrozen
);
1406 deactivate_locked_super(sb
);
1410 EXPORT_SYMBOL(thaw_super
);