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
, sc
->nid
);
81 inodes
= list_lru_count_node(&sb
->s_inode_lru
, sc
->nid
);
82 dentries
= list_lru_count_node(&sb
->s_dentry_lru
, sc
->nid
);
83 total_objects
= dentries
+ inodes
+ fs_objects
+ 1;
85 /* proportion the scan between the caches */
86 dentries
= mult_frac(sc
->nr_to_scan
, dentries
, total_objects
);
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
, sc
->nid
);
94 freed
+= prune_icache_sb(sb
, inodes
, sc
->nid
);
97 fs_objects
= mult_frac(sc
->nr_to_scan
, fs_objects
,
99 freed
+= sb
->s_op
->free_cached_objects(sb
, fs_objects
,
107 static unsigned long super_cache_count(struct shrinker
*shrink
,
108 struct shrink_control
*sc
)
110 struct super_block
*sb
;
111 long total_objects
= 0;
113 sb
= container_of(shrink
, struct super_block
, s_shrink
);
115 if (!grab_super_passive(sb
))
118 if (sb
->s_op
&& sb
->s_op
->nr_cached_objects
)
119 total_objects
= sb
->s_op
->nr_cached_objects(sb
,
122 total_objects
+= list_lru_count_node(&sb
->s_dentry_lru
,
124 total_objects
+= list_lru_count_node(&sb
->s_inode_lru
,
127 total_objects
= vfs_pressure_ratio(total_objects
);
129 return total_objects
;
133 * destroy_super - frees a superblock
134 * @s: superblock to free
136 * Frees a superblock.
138 static void destroy_super(struct super_block
*s
)
141 list_lru_destroy(&s
->s_dentry_lru
);
142 list_lru_destroy(&s
->s_inode_lru
);
144 free_percpu(s
->s_files
);
146 for (i
= 0; i
< SB_FREEZE_LEVELS
; i
++)
147 percpu_counter_destroy(&s
->s_writers
.counter
[i
]);
149 WARN_ON(!list_empty(&s
->s_mounts
));
156 * alloc_super - create new superblock
157 * @type: filesystem type superblock should belong to
158 * @flags: the mount flags
160 * Allocates and initializes a new &struct super_block. alloc_super()
161 * returns a pointer new superblock or %NULL if allocation had failed.
163 static struct super_block
*alloc_super(struct file_system_type
*type
, int flags
)
165 struct super_block
*s
= kzalloc(sizeof(struct super_block
), GFP_USER
);
166 static const struct super_operations default_op
;
172 if (security_sb_alloc(s
))
176 s
->s_files
= alloc_percpu(struct list_head
);
179 for_each_possible_cpu(i
)
180 INIT_LIST_HEAD(per_cpu_ptr(s
->s_files
, i
));
182 INIT_LIST_HEAD(&s
->s_files
);
184 for (i
= 0; i
< SB_FREEZE_LEVELS
; i
++) {
185 if (percpu_counter_init(&s
->s_writers
.counter
[i
], 0) < 0)
187 lockdep_init_map(&s
->s_writers
.lock_map
[i
], sb_writers_name
[i
],
188 &type
->s_writers_key
[i
], 0);
190 init_waitqueue_head(&s
->s_writers
.wait
);
191 init_waitqueue_head(&s
->s_writers
.wait_unfrozen
);
193 s
->s_bdi
= &default_backing_dev_info
;
194 INIT_HLIST_NODE(&s
->s_instances
);
195 INIT_HLIST_BL_HEAD(&s
->s_anon
);
196 INIT_LIST_HEAD(&s
->s_inodes
);
198 if (list_lru_init(&s
->s_dentry_lru
))
200 if (list_lru_init(&s
->s_inode_lru
))
203 INIT_LIST_HEAD(&s
->s_mounts
);
204 init_rwsem(&s
->s_umount
);
205 lockdep_set_class(&s
->s_umount
, &type
->s_umount_key
);
207 * sget() can have s_umount recursion.
209 * When it cannot find a suitable sb, it allocates a new
210 * one (this one), and tries again to find a suitable old
213 * In case that succeeds, it will acquire the s_umount
214 * lock of the old one. Since these are clearly distrinct
215 * locks, and this object isn't exposed yet, there's no
218 * Annotate this by putting this lock in a different
221 down_write_nested(&s
->s_umount
, SINGLE_DEPTH_NESTING
);
223 atomic_set(&s
->s_active
, 1);
224 mutex_init(&s
->s_vfs_rename_mutex
);
225 lockdep_set_class(&s
->s_vfs_rename_mutex
, &type
->s_vfs_rename_key
);
226 mutex_init(&s
->s_dquot
.dqio_mutex
);
227 mutex_init(&s
->s_dquot
.dqonoff_mutex
);
228 init_rwsem(&s
->s_dquot
.dqptr_sem
);
229 s
->s_maxbytes
= MAX_NON_LFS
;
230 s
->s_op
= &default_op
;
231 s
->s_time_gran
= 1000000000;
232 s
->cleancache_poolid
= -1;
234 s
->s_shrink
.seeks
= DEFAULT_SEEKS
;
235 s
->s_shrink
.scan_objects
= super_cache_scan
;
236 s
->s_shrink
.count_objects
= super_cache_count
;
237 s
->s_shrink
.batch
= 1024;
238 s
->s_shrink
.flags
= SHRINKER_NUMA_AWARE
;
246 /* Superblock refcounting */
249 * Drop a superblock's refcount. The caller must hold sb_lock.
251 static void __put_super(struct super_block
*sb
)
253 if (!--sb
->s_count
) {
254 list_del_init(&sb
->s_list
);
260 * put_super - drop a temporary reference to superblock
261 * @sb: superblock in question
263 * Drops a temporary reference, frees superblock if there's no
266 static void put_super(struct super_block
*sb
)
270 spin_unlock(&sb_lock
);
275 * deactivate_locked_super - drop an active reference to superblock
276 * @s: superblock to deactivate
278 * Drops an active reference to superblock, converting it into a temprory
279 * one if there is no other active references left. In that case we
280 * tell fs driver to shut it down and drop the temporary reference we
283 * Caller holds exclusive lock on superblock; that lock is released.
285 void deactivate_locked_super(struct super_block
*s
)
287 struct file_system_type
*fs
= s
->s_type
;
288 if (atomic_dec_and_test(&s
->s_active
)) {
289 cleancache_invalidate_fs(s
);
292 /* caches are now gone, we can safely kill the shrinker now */
293 unregister_shrinker(&s
->s_shrink
);
298 up_write(&s
->s_umount
);
302 EXPORT_SYMBOL(deactivate_locked_super
);
305 * deactivate_super - drop an active reference to superblock
306 * @s: superblock to deactivate
308 * Variant of deactivate_locked_super(), except that superblock is *not*
309 * locked by caller. If we are going to drop the final active reference,
310 * lock will be acquired prior to that.
312 void deactivate_super(struct super_block
*s
)
314 if (!atomic_add_unless(&s
->s_active
, -1, 1)) {
315 down_write(&s
->s_umount
);
316 deactivate_locked_super(s
);
320 EXPORT_SYMBOL(deactivate_super
);
323 * grab_super - acquire an active reference
324 * @s: reference we are trying to make active
326 * Tries to acquire an active reference. grab_super() is used when we
327 * had just found a superblock in super_blocks or fs_type->fs_supers
328 * and want to turn it into a full-blown active reference. grab_super()
329 * is called with sb_lock held and drops it. Returns 1 in case of
330 * success, 0 if we had failed (superblock contents was already dead or
331 * dying when grab_super() had been called). Note that this is only
332 * called for superblocks not in rundown mode (== ones still on ->fs_supers
333 * of their type), so increment of ->s_count is OK here.
335 static int grab_super(struct super_block
*s
) __releases(sb_lock
)
338 spin_unlock(&sb_lock
);
339 down_write(&s
->s_umount
);
340 if ((s
->s_flags
& MS_BORN
) && atomic_inc_not_zero(&s
->s_active
)) {
344 up_write(&s
->s_umount
);
350 * grab_super_passive - acquire a passive reference
351 * @sb: reference we are trying to grab
353 * Tries to acquire a passive reference. This is used in places where we
354 * cannot take an active reference but we need to ensure that the
355 * superblock does not go away while we are working on it. It returns
356 * false if a reference was not gained, and returns true with the s_umount
357 * lock held in read mode if a reference is gained. On successful return,
358 * the caller must drop the s_umount lock and the passive reference when
361 bool grab_super_passive(struct super_block
*sb
)
364 if (hlist_unhashed(&sb
->s_instances
)) {
365 spin_unlock(&sb_lock
);
370 spin_unlock(&sb_lock
);
372 if (down_read_trylock(&sb
->s_umount
)) {
373 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
375 up_read(&sb
->s_umount
);
383 * generic_shutdown_super - common helper for ->kill_sb()
384 * @sb: superblock to kill
386 * generic_shutdown_super() does all fs-independent work on superblock
387 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
388 * that need destruction out of superblock, call generic_shutdown_super()
389 * and release aforementioned objects. Note: dentries and inodes _are_
390 * taken care of and do not need specific handling.
392 * Upon calling this function, the filesystem may no longer alter or
393 * rearrange the set of dentries belonging to this super_block, nor may it
394 * change the attachments of dentries to inodes.
396 void generic_shutdown_super(struct super_block
*sb
)
398 const struct super_operations
*sop
= sb
->s_op
;
401 shrink_dcache_for_umount(sb
);
403 sb
->s_flags
&= ~MS_ACTIVE
;
405 fsnotify_unmount_inodes(&sb
->s_inodes
);
409 if (sb
->s_dio_done_wq
) {
410 destroy_workqueue(sb
->s_dio_done_wq
);
411 sb
->s_dio_done_wq
= NULL
;
417 if (!list_empty(&sb
->s_inodes
)) {
418 printk("VFS: Busy inodes after unmount of %s. "
419 "Self-destruct in 5 seconds. Have a nice day...\n",
424 /* should be initialized for __put_super_and_need_restart() */
425 hlist_del_init(&sb
->s_instances
);
426 spin_unlock(&sb_lock
);
427 up_write(&sb
->s_umount
);
430 EXPORT_SYMBOL(generic_shutdown_super
);
433 * sget - find or create a superblock
434 * @type: filesystem type superblock should belong to
435 * @test: comparison callback
436 * @set: setup callback
437 * @flags: mount flags
438 * @data: argument to each of them
440 struct super_block
*sget(struct file_system_type
*type
,
441 int (*test
)(struct super_block
*,void *),
442 int (*set
)(struct super_block
*,void *),
446 struct super_block
*s
= NULL
;
447 struct super_block
*old
;
453 hlist_for_each_entry(old
, &type
->fs_supers
, s_instances
) {
454 if (!test(old
, data
))
456 if (!grab_super(old
))
459 up_write(&s
->s_umount
);
467 spin_unlock(&sb_lock
);
468 s
= alloc_super(type
, flags
);
470 return ERR_PTR(-ENOMEM
);
476 spin_unlock(&sb_lock
);
477 up_write(&s
->s_umount
);
482 strlcpy(s
->s_id
, type
->name
, sizeof(s
->s_id
));
483 list_add_tail(&s
->s_list
, &super_blocks
);
484 hlist_add_head(&s
->s_instances
, &type
->fs_supers
);
485 spin_unlock(&sb_lock
);
486 get_filesystem(type
);
487 register_shrinker(&s
->s_shrink
);
493 void drop_super(struct super_block
*sb
)
495 up_read(&sb
->s_umount
);
499 EXPORT_SYMBOL(drop_super
);
502 * iterate_supers - call function for all active superblocks
503 * @f: function to call
504 * @arg: argument to pass to it
506 * Scans the superblock list and calls given function, passing it
507 * locked superblock and given argument.
509 void iterate_supers(void (*f
)(struct super_block
*, void *), void *arg
)
511 struct super_block
*sb
, *p
= NULL
;
514 list_for_each_entry(sb
, &super_blocks
, s_list
) {
515 if (hlist_unhashed(&sb
->s_instances
))
518 spin_unlock(&sb_lock
);
520 down_read(&sb
->s_umount
);
521 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
523 up_read(&sb
->s_umount
);
532 spin_unlock(&sb_lock
);
536 * iterate_supers_type - call function for superblocks of given type
538 * @f: function to call
539 * @arg: argument to pass to it
541 * Scans the superblock list and calls given function, passing it
542 * locked superblock and given argument.
544 void iterate_supers_type(struct file_system_type
*type
,
545 void (*f
)(struct super_block
*, void *), void *arg
)
547 struct super_block
*sb
, *p
= NULL
;
550 hlist_for_each_entry(sb
, &type
->fs_supers
, s_instances
) {
552 spin_unlock(&sb_lock
);
554 down_read(&sb
->s_umount
);
555 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
557 up_read(&sb
->s_umount
);
566 spin_unlock(&sb_lock
);
569 EXPORT_SYMBOL(iterate_supers_type
);
572 * get_super - get the superblock of a device
573 * @bdev: device to get the superblock for
575 * Scans the superblock list and finds the superblock of the file system
576 * mounted on the device given. %NULL is returned if no match is found.
579 struct super_block
*get_super(struct block_device
*bdev
)
581 struct super_block
*sb
;
588 list_for_each_entry(sb
, &super_blocks
, s_list
) {
589 if (hlist_unhashed(&sb
->s_instances
))
591 if (sb
->s_bdev
== bdev
) {
593 spin_unlock(&sb_lock
);
594 down_read(&sb
->s_umount
);
596 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
598 up_read(&sb
->s_umount
);
599 /* nope, got unmounted */
605 spin_unlock(&sb_lock
);
609 EXPORT_SYMBOL(get_super
);
612 * get_super_thawed - get thawed superblock of a device
613 * @bdev: device to get the superblock for
615 * Scans the superblock list and finds the superblock of the file system
616 * mounted on the device. The superblock is returned once it is thawed
617 * (or immediately if it was not frozen). %NULL is returned if no match
620 struct super_block
*get_super_thawed(struct block_device
*bdev
)
623 struct super_block
*s
= get_super(bdev
);
624 if (!s
|| s
->s_writers
.frozen
== SB_UNFROZEN
)
626 up_read(&s
->s_umount
);
627 wait_event(s
->s_writers
.wait_unfrozen
,
628 s
->s_writers
.frozen
== SB_UNFROZEN
);
632 EXPORT_SYMBOL(get_super_thawed
);
635 * get_active_super - get an active reference to the superblock of a device
636 * @bdev: device to get the superblock for
638 * Scans the superblock list and finds the superblock of the file system
639 * mounted on the device given. Returns the superblock with an active
640 * reference or %NULL if none was found.
642 struct super_block
*get_active_super(struct block_device
*bdev
)
644 struct super_block
*sb
;
651 list_for_each_entry(sb
, &super_blocks
, s_list
) {
652 if (hlist_unhashed(&sb
->s_instances
))
654 if (sb
->s_bdev
== bdev
) {
657 up_write(&sb
->s_umount
);
661 spin_unlock(&sb_lock
);
665 struct super_block
*user_get_super(dev_t dev
)
667 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_dev
== dev
) {
676 spin_unlock(&sb_lock
);
677 down_read(&sb
->s_umount
);
679 if (sb
->s_root
&& (sb
->s_flags
& MS_BORN
))
681 up_read(&sb
->s_umount
);
682 /* nope, got unmounted */
688 spin_unlock(&sb_lock
);
693 * do_remount_sb - asks filesystem to change mount options.
694 * @sb: superblock in question
695 * @flags: numeric part of options
696 * @data: the rest of options
697 * @force: whether or not to force the change
699 * Alters the mount options of a mounted file system.
701 int do_remount_sb(struct super_block
*sb
, int flags
, void *data
, int force
)
706 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
)
710 if (!(flags
& MS_RDONLY
) && bdev_read_only(sb
->s_bdev
))
714 if (flags
& MS_RDONLY
)
716 shrink_dcache_sb(sb
);
719 remount_ro
= (flags
& MS_RDONLY
) && !(sb
->s_flags
& MS_RDONLY
);
721 /* If we are remounting RDONLY and current sb is read/write,
722 make sure there are no rw files opened */
727 retval
= sb_prepare_remount_readonly(sb
);
733 if (sb
->s_op
->remount_fs
) {
734 retval
= sb
->s_op
->remount_fs(sb
, &flags
, data
);
737 goto cancel_readonly
;
738 /* If forced remount, go ahead despite any errors */
739 WARN(1, "forced remount of a %s fs returned %i\n",
740 sb
->s_type
->name
, retval
);
743 sb
->s_flags
= (sb
->s_flags
& ~MS_RMT_MASK
) | (flags
& MS_RMT_MASK
);
744 /* Needs to be ordered wrt mnt_is_readonly() */
746 sb
->s_readonly_remount
= 0;
749 * Some filesystems modify their metadata via some other path than the
750 * bdev buffer cache (eg. use a private mapping, or directories in
751 * pagecache, etc). Also file data modifications go via their own
752 * mappings. So If we try to mount readonly then copy the filesystem
753 * from bdev, we could get stale data, so invalidate it to give a best
754 * effort at coherency.
756 if (remount_ro
&& sb
->s_bdev
)
757 invalidate_bdev(sb
->s_bdev
);
761 sb
->s_readonly_remount
= 0;
765 static void do_emergency_remount(struct work_struct
*work
)
767 struct super_block
*sb
, *p
= NULL
;
770 list_for_each_entry(sb
, &super_blocks
, s_list
) {
771 if (hlist_unhashed(&sb
->s_instances
))
774 spin_unlock(&sb_lock
);
775 down_write(&sb
->s_umount
);
776 if (sb
->s_root
&& sb
->s_bdev
&& (sb
->s_flags
& MS_BORN
) &&
777 !(sb
->s_flags
& MS_RDONLY
)) {
779 * What lock protects sb->s_flags??
781 do_remount_sb(sb
, MS_RDONLY
, NULL
, 1);
783 up_write(&sb
->s_umount
);
791 spin_unlock(&sb_lock
);
793 printk("Emergency Remount complete\n");
796 void emergency_remount(void)
798 struct work_struct
*work
;
800 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
802 INIT_WORK(work
, do_emergency_remount
);
808 * Unnamed block devices are dummy devices used by virtual
809 * filesystems which don't use real block-devices. -- jrs
812 static DEFINE_IDA(unnamed_dev_ida
);
813 static DEFINE_SPINLOCK(unnamed_dev_lock
);/* protects the above */
814 static int unnamed_dev_start
= 0; /* don't bother trying below it */
816 int get_anon_bdev(dev_t
*p
)
822 if (ida_pre_get(&unnamed_dev_ida
, GFP_ATOMIC
) == 0)
824 spin_lock(&unnamed_dev_lock
);
825 error
= ida_get_new_above(&unnamed_dev_ida
, unnamed_dev_start
, &dev
);
827 unnamed_dev_start
= dev
+ 1;
828 spin_unlock(&unnamed_dev_lock
);
829 if (error
== -EAGAIN
)
830 /* We raced and lost with another CPU. */
835 if (dev
== (1 << MINORBITS
)) {
836 spin_lock(&unnamed_dev_lock
);
837 ida_remove(&unnamed_dev_ida
, dev
);
838 if (unnamed_dev_start
> dev
)
839 unnamed_dev_start
= dev
;
840 spin_unlock(&unnamed_dev_lock
);
843 *p
= MKDEV(0, dev
& MINORMASK
);
846 EXPORT_SYMBOL(get_anon_bdev
);
848 void free_anon_bdev(dev_t dev
)
850 int slot
= MINOR(dev
);
851 spin_lock(&unnamed_dev_lock
);
852 ida_remove(&unnamed_dev_ida
, slot
);
853 if (slot
< unnamed_dev_start
)
854 unnamed_dev_start
= slot
;
855 spin_unlock(&unnamed_dev_lock
);
857 EXPORT_SYMBOL(free_anon_bdev
);
859 int set_anon_super(struct super_block
*s
, void *data
)
861 int error
= get_anon_bdev(&s
->s_dev
);
863 s
->s_bdi
= &noop_backing_dev_info
;
867 EXPORT_SYMBOL(set_anon_super
);
869 void kill_anon_super(struct super_block
*sb
)
871 dev_t dev
= sb
->s_dev
;
872 generic_shutdown_super(sb
);
876 EXPORT_SYMBOL(kill_anon_super
);
878 void kill_litter_super(struct super_block
*sb
)
881 d_genocide(sb
->s_root
);
885 EXPORT_SYMBOL(kill_litter_super
);
887 static int ns_test_super(struct super_block
*sb
, void *data
)
889 return sb
->s_fs_info
== data
;
892 static int ns_set_super(struct super_block
*sb
, void *data
)
894 sb
->s_fs_info
= data
;
895 return set_anon_super(sb
, NULL
);
898 struct dentry
*mount_ns(struct file_system_type
*fs_type
, int flags
,
899 void *data
, int (*fill_super
)(struct super_block
*, void *, int))
901 struct super_block
*sb
;
903 sb
= sget(fs_type
, ns_test_super
, ns_set_super
, flags
, data
);
909 err
= fill_super(sb
, data
, flags
& MS_SILENT
? 1 : 0);
911 deactivate_locked_super(sb
);
915 sb
->s_flags
|= MS_ACTIVE
;
918 return dget(sb
->s_root
);
921 EXPORT_SYMBOL(mount_ns
);
924 static int set_bdev_super(struct super_block
*s
, void *data
)
927 s
->s_dev
= s
->s_bdev
->bd_dev
;
930 * We set the bdi here to the queue backing, file systems can
931 * overwrite this in ->fill_super()
933 s
->s_bdi
= &bdev_get_queue(s
->s_bdev
)->backing_dev_info
;
937 static int test_bdev_super(struct super_block
*s
, void *data
)
939 return (void *)s
->s_bdev
== data
;
942 struct dentry
*mount_bdev(struct file_system_type
*fs_type
,
943 int flags
, const char *dev_name
, void *data
,
944 int (*fill_super
)(struct super_block
*, void *, int))
946 struct block_device
*bdev
;
947 struct super_block
*s
;
948 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
951 if (!(flags
& MS_RDONLY
))
954 bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
956 return ERR_CAST(bdev
);
959 * once the super is inserted into the list by sget, s_umount
960 * will protect the lockfs code from trying to start a snapshot
961 * while we are mounting
963 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
964 if (bdev
->bd_fsfreeze_count
> 0) {
965 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
969 s
= sget(fs_type
, test_bdev_super
, set_bdev_super
, flags
| MS_NOSEC
,
971 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
976 if ((flags
^ s
->s_flags
) & MS_RDONLY
) {
977 deactivate_locked_super(s
);
983 * s_umount nests inside bd_mutex during
984 * __invalidate_device(). blkdev_put() acquires
985 * bd_mutex and can't be called under s_umount. Drop
986 * s_umount temporarily. This is safe as we're
987 * holding an active reference.
989 up_write(&s
->s_umount
);
990 blkdev_put(bdev
, mode
);
991 down_write(&s
->s_umount
);
993 char b
[BDEVNAME_SIZE
];
996 strlcpy(s
->s_id
, bdevname(bdev
, b
), sizeof(s
->s_id
));
997 sb_set_blocksize(s
, block_size(bdev
));
998 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1000 deactivate_locked_super(s
);
1004 s
->s_flags
|= MS_ACTIVE
;
1008 return dget(s
->s_root
);
1013 blkdev_put(bdev
, mode
);
1015 return ERR_PTR(error
);
1017 EXPORT_SYMBOL(mount_bdev
);
1019 void kill_block_super(struct super_block
*sb
)
1021 struct block_device
*bdev
= sb
->s_bdev
;
1022 fmode_t mode
= sb
->s_mode
;
1024 bdev
->bd_super
= NULL
;
1025 generic_shutdown_super(sb
);
1026 sync_blockdev(bdev
);
1027 WARN_ON_ONCE(!(mode
& FMODE_EXCL
));
1028 blkdev_put(bdev
, mode
| FMODE_EXCL
);
1031 EXPORT_SYMBOL(kill_block_super
);
1034 struct dentry
*mount_nodev(struct file_system_type
*fs_type
,
1035 int flags
, void *data
,
1036 int (*fill_super
)(struct super_block
*, void *, int))
1039 struct super_block
*s
= sget(fs_type
, NULL
, set_anon_super
, flags
, NULL
);
1044 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1046 deactivate_locked_super(s
);
1047 return ERR_PTR(error
);
1049 s
->s_flags
|= MS_ACTIVE
;
1050 return dget(s
->s_root
);
1052 EXPORT_SYMBOL(mount_nodev
);
1054 static int compare_single(struct super_block
*s
, void *p
)
1059 struct dentry
*mount_single(struct file_system_type
*fs_type
,
1060 int flags
, void *data
,
1061 int (*fill_super
)(struct super_block
*, void *, int))
1063 struct super_block
*s
;
1066 s
= sget(fs_type
, compare_single
, set_anon_super
, flags
, NULL
);
1070 error
= fill_super(s
, data
, flags
& MS_SILENT
? 1 : 0);
1072 deactivate_locked_super(s
);
1073 return ERR_PTR(error
);
1075 s
->s_flags
|= MS_ACTIVE
;
1077 do_remount_sb(s
, flags
, data
, 0);
1079 return dget(s
->s_root
);
1081 EXPORT_SYMBOL(mount_single
);
1084 mount_fs(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
1086 struct dentry
*root
;
1087 struct super_block
*sb
;
1088 char *secdata
= NULL
;
1089 int error
= -ENOMEM
;
1091 if (data
&& !(type
->fs_flags
& FS_BINARY_MOUNTDATA
)) {
1092 secdata
= alloc_secdata();
1096 error
= security_sb_copy_data(data
, secdata
);
1098 goto out_free_secdata
;
1101 root
= type
->mount(type
, flags
, name
, data
);
1103 error
= PTR_ERR(root
);
1104 goto out_free_secdata
;
1108 WARN_ON(!sb
->s_bdi
);
1109 WARN_ON(sb
->s_bdi
== &default_backing_dev_info
);
1110 sb
->s_flags
|= MS_BORN
;
1112 error
= security_sb_kern_mount(sb
, flags
, secdata
);
1117 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1118 * but s_maxbytes was an unsigned long long for many releases. Throw
1119 * this warning for a little while to try and catch filesystems that
1120 * violate this rule.
1122 WARN((sb
->s_maxbytes
< 0), "%s set sb->s_maxbytes to "
1123 "negative value (%lld)\n", type
->name
, sb
->s_maxbytes
);
1125 up_write(&sb
->s_umount
);
1126 free_secdata(secdata
);
1130 deactivate_locked_super(sb
);
1132 free_secdata(secdata
);
1134 return ERR_PTR(error
);
1138 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1141 void __sb_end_write(struct super_block
*sb
, int level
)
1143 percpu_counter_dec(&sb
->s_writers
.counter
[level
-1]);
1145 * Make sure s_writers are updated before we wake up waiters in
1149 if (waitqueue_active(&sb
->s_writers
.wait
))
1150 wake_up(&sb
->s_writers
.wait
);
1151 rwsem_release(&sb
->s_writers
.lock_map
[level
-1], 1, _RET_IP_
);
1153 EXPORT_SYMBOL(__sb_end_write
);
1155 #ifdef CONFIG_LOCKDEP
1157 * We want lockdep to tell us about possible deadlocks with freezing but
1158 * it's it bit tricky to properly instrument it. Getting a freeze protection
1159 * works as getting a read lock but there are subtle problems. XFS for example
1160 * gets freeze protection on internal level twice in some cases, which is OK
1161 * only because we already hold a freeze protection also on higher level. Due
1162 * to these cases we have to tell lockdep we are doing trylock when we
1163 * already hold a freeze protection for a higher freeze level.
1165 static void acquire_freeze_lock(struct super_block
*sb
, int level
, bool trylock
,
1171 for (i
= 0; i
< level
- 1; i
++)
1172 if (lock_is_held(&sb
->s_writers
.lock_map
[i
])) {
1177 rwsem_acquire_read(&sb
->s_writers
.lock_map
[level
-1], 0, trylock
, ip
);
1182 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1185 int __sb_start_write(struct super_block
*sb
, int level
, bool wait
)
1188 if (unlikely(sb
->s_writers
.frozen
>= level
)) {
1191 wait_event(sb
->s_writers
.wait_unfrozen
,
1192 sb
->s_writers
.frozen
< level
);
1195 #ifdef CONFIG_LOCKDEP
1196 acquire_freeze_lock(sb
, level
, !wait
, _RET_IP_
);
1198 percpu_counter_inc(&sb
->s_writers
.counter
[level
-1]);
1200 * Make sure counter is updated before we check for frozen.
1201 * freeze_super() first sets frozen and then checks the counter.
1204 if (unlikely(sb
->s_writers
.frozen
>= level
)) {
1205 __sb_end_write(sb
, level
);
1210 EXPORT_SYMBOL(__sb_start_write
);
1213 * sb_wait_write - wait until all writers to given file system finish
1214 * @sb: the super for which we wait
1215 * @level: type of writers we wait for (normal vs page fault)
1217 * This function waits until there are no writers of given type to given file
1218 * system. Caller of this function should make sure there can be no new writers
1219 * of type @level before calling this function. Otherwise this function can
1222 static void sb_wait_write(struct super_block
*sb
, int level
)
1227 * We just cycle-through lockdep here so that it does not complain
1228 * about returning with lock to userspace
1230 rwsem_acquire(&sb
->s_writers
.lock_map
[level
-1], 0, 0, _THIS_IP_
);
1231 rwsem_release(&sb
->s_writers
.lock_map
[level
-1], 1, _THIS_IP_
);
1237 * We use a barrier in prepare_to_wait() to separate setting
1238 * of frozen and checking of the counter
1240 prepare_to_wait(&sb
->s_writers
.wait
, &wait
,
1241 TASK_UNINTERRUPTIBLE
);
1243 writers
= percpu_counter_sum(&sb
->s_writers
.counter
[level
-1]);
1247 finish_wait(&sb
->s_writers
.wait
, &wait
);
1252 * freeze_super - lock the filesystem and force it into a consistent state
1253 * @sb: the super to lock
1255 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1256 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1259 * During this function, sb->s_writers.frozen goes through these values:
1261 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1263 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1264 * writes should be blocked, though page faults are still allowed. We wait for
1265 * all writes to complete and then proceed to the next stage.
1267 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1268 * but internal fs threads can still modify the filesystem (although they
1269 * should not dirty new pages or inodes), writeback can run etc. After waiting
1270 * for all running page faults we sync the filesystem which will clean all
1271 * dirty pages and inodes (no new dirty pages or inodes can be created when
1274 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1275 * modification are blocked (e.g. XFS preallocation truncation on inode
1276 * reclaim). This is usually implemented by blocking new transactions for
1277 * filesystems that have them and need this additional guard. After all
1278 * internal writers are finished we call ->freeze_fs() to finish filesystem
1279 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1280 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1282 * sb->s_writers.frozen is protected by sb->s_umount.
1284 int freeze_super(struct super_block
*sb
)
1288 atomic_inc(&sb
->s_active
);
1289 down_write(&sb
->s_umount
);
1290 if (sb
->s_writers
.frozen
!= SB_UNFROZEN
) {
1291 deactivate_locked_super(sb
);
1295 if (!(sb
->s_flags
& MS_BORN
)) {
1296 up_write(&sb
->s_umount
);
1297 return 0; /* sic - it's "nothing to do" */
1300 if (sb
->s_flags
& MS_RDONLY
) {
1301 /* Nothing to do really... */
1302 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1303 up_write(&sb
->s_umount
);
1307 /* From now on, no new normal writers can start */
1308 sb
->s_writers
.frozen
= SB_FREEZE_WRITE
;
1311 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1312 up_write(&sb
->s_umount
);
1314 sb_wait_write(sb
, SB_FREEZE_WRITE
);
1316 /* Now we go and block page faults... */
1317 down_write(&sb
->s_umount
);
1318 sb
->s_writers
.frozen
= SB_FREEZE_PAGEFAULT
;
1321 sb_wait_write(sb
, SB_FREEZE_PAGEFAULT
);
1323 /* All writers are done so after syncing there won't be dirty data */
1324 sync_filesystem(sb
);
1326 /* Now wait for internal filesystem counter */
1327 sb
->s_writers
.frozen
= SB_FREEZE_FS
;
1329 sb_wait_write(sb
, SB_FREEZE_FS
);
1331 if (sb
->s_op
->freeze_fs
) {
1332 ret
= sb
->s_op
->freeze_fs(sb
);
1335 "VFS:Filesystem freeze failed\n");
1336 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1338 wake_up(&sb
->s_writers
.wait_unfrozen
);
1339 deactivate_locked_super(sb
);
1344 * This is just for debugging purposes so that fs can warn if it
1345 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1347 sb
->s_writers
.frozen
= SB_FREEZE_COMPLETE
;
1348 up_write(&sb
->s_umount
);
1351 EXPORT_SYMBOL(freeze_super
);
1354 * thaw_super -- unlock filesystem
1355 * @sb: the super to thaw
1357 * Unlocks the filesystem and marks it writeable again after freeze_super().
1359 int thaw_super(struct super_block
*sb
)
1363 down_write(&sb
->s_umount
);
1364 if (sb
->s_writers
.frozen
== SB_UNFROZEN
) {
1365 up_write(&sb
->s_umount
);
1369 if (sb
->s_flags
& MS_RDONLY
)
1372 if (sb
->s_op
->unfreeze_fs
) {
1373 error
= sb
->s_op
->unfreeze_fs(sb
);
1376 "VFS:Filesystem thaw failed\n");
1377 up_write(&sb
->s_umount
);
1383 sb
->s_writers
.frozen
= SB_UNFROZEN
;
1385 wake_up(&sb
->s_writers
.wait_unfrozen
);
1386 deactivate_locked_super(sb
);
1390 EXPORT_SYMBOL(thaw_super
);