4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * super.c contains code to handle: - mount structures
8 * - filesystem drivers list
10 * - umount system call
13 * GK 2/5/95 - Changed to support mounting the root fs via NFS
15 * Added kerneld support: Jacques Gelinas and Bjorn Ekwall
16 * Added change_root: Werner Almesberger & Hans Lermen, Feb '96
17 * Added options to /proc/mounts:
18 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
19 * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
20 * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
23 #include <linux/export.h>
24 #include <linux/slab.h>
25 #include <linux/acct.h>
26 #include <linux/blkdev.h>
27 #include <linux/mount.h>
28 #include <linux/security.h>
29 #include <linux/writeback.h> /* for the emergency remount stuff */
30 #include <linux/idr.h>
31 #include <linux/mutex.h>
32 #include <linux/backing-dev.h>
33 #include <linux/rculist_bl.h>
34 #include <linux/cleancache.h>
35 #include <linux/fsnotify.h>
36 #include <linux/lockdep.h>
40 LIST_HEAD(super_blocks);
41 DEFINE_SPINLOCK(sb_lock);
43 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
50 * One thing we have to be careful of with a per-sb shrinker is that we don't
51 * drop the last active reference to the superblock from within the shrinker.
52 * If that happens we could trigger unregistering the shrinker from within the
53 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
54 * take a passive reference to the superblock to avoid this from occurring.
56 static int prune_super(struct shrinker *shrink, struct shrink_control *sc)
58 struct super_block *sb;
62 sb = container_of(shrink, struct super_block, s_shrink);
65 * Deadlock avoidance. We may hold various FS locks, and we don't want
66 * to recurse into the FS that called us in clear_inode() and friends..
68 if (sc->nr_to_scan && !(sc->gfp_mask & __GFP_FS))
71 if (!grab_super_passive(sb))
74 if (sb->s_op && sb->s_op->nr_cached_objects)
75 fs_objects = sb->s_op->nr_cached_objects(sb);
77 total_objects = sb->s_nr_dentry_unused +
78 sb->s_nr_inodes_unused + fs_objects + 1;
84 /* proportion the scan between the caches */
85 dentries = (sc->nr_to_scan * sb->s_nr_dentry_unused) /
87 inodes = (sc->nr_to_scan * sb->s_nr_inodes_unused) /
90 fs_objects = (sc->nr_to_scan * fs_objects) /
93 * prune the dcache first as the icache is pinned by it, then
94 * prune the icache, followed by the filesystem specific caches
96 prune_dcache_sb(sb, dentries);
97 prune_icache_sb(sb, inodes);
99 if (fs_objects && sb->s_op->free_cached_objects) {
100 sb->s_op->free_cached_objects(sb, fs_objects);
101 fs_objects = sb->s_op->nr_cached_objects(sb);
103 total_objects = sb->s_nr_dentry_unused +
104 sb->s_nr_inodes_unused + fs_objects;
107 total_objects = (total_objects / 100) * sysctl_vfs_cache_pressure;
109 return total_objects;
112 static int init_sb_writers(struct super_block *s, struct file_system_type *type)
117 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
118 err = percpu_counter_init(&s->s_writers.counter[i], 0);
121 lockdep_init_map(&s->s_writers.lock_map[i], sb_writers_name[i],
122 &type->s_writers_key[i], 0);
124 init_waitqueue_head(&s->s_writers.wait);
125 init_waitqueue_head(&s->s_writers.wait_unfrozen);
129 percpu_counter_destroy(&s->s_writers.counter[i]);
133 static void destroy_sb_writers(struct super_block *s)
137 for (i = 0; i < SB_FREEZE_LEVELS; i++)
138 percpu_counter_destroy(&s->s_writers.counter[i]);
142 * alloc_super - create new superblock
143 * @type: filesystem type superblock should belong to
144 * @flags: the mount flags
146 * Allocates and initializes a new &struct super_block. alloc_super()
147 * returns a pointer new superblock or %NULL if allocation had failed.
149 static struct super_block *alloc_super(struct file_system_type *type, int flags)
151 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
152 static const struct super_operations default_op;
155 if (security_sb_alloc(s)) {
157 * We cannot call security_sb_free() without
158 * security_sb_alloc() succeeding. So bail out manually
165 s->s_files = alloc_percpu(struct list_head);
171 for_each_possible_cpu(i)
172 INIT_LIST_HEAD(per_cpu_ptr(s->s_files, i));
175 INIT_LIST_HEAD(&s->s_files);
177 if (init_sb_writers(s, type))
180 s->s_bdi = &default_backing_dev_info;
181 INIT_HLIST_NODE(&s->s_instances);
182 INIT_HLIST_BL_HEAD(&s->s_anon);
183 INIT_LIST_HEAD(&s->s_inodes);
184 INIT_LIST_HEAD(&s->s_dentry_lru);
185 INIT_LIST_HEAD(&s->s_inode_lru);
186 spin_lock_init(&s->s_inode_lru_lock);
187 INIT_LIST_HEAD(&s->s_mounts);
188 init_rwsem(&s->s_umount);
189 mutex_init(&s->s_lock);
190 lockdep_set_class(&s->s_umount, &type->s_umount_key);
192 * The locking rules for s_lock are up to the
193 * filesystem. For example ext3fs has different
194 * lock ordering than usbfs:
196 lockdep_set_class(&s->s_lock, &type->s_lock_key);
198 * sget() can have s_umount recursion.
200 * When it cannot find a suitable sb, it allocates a new
201 * one (this one), and tries again to find a suitable old
204 * In case that succeeds, it will acquire the s_umount
205 * lock of the old one. Since these are clearly distrinct
206 * locks, and this object isn't exposed yet, there's no
209 * Annotate this by putting this lock in a different
212 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
214 atomic_set(&s->s_active, 1);
215 mutex_init(&s->s_vfs_rename_mutex);
216 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
217 mutex_init(&s->s_dquot.dqio_mutex);
218 mutex_init(&s->s_dquot.dqonoff_mutex);
219 init_rwsem(&s->s_dquot.dqptr_sem);
220 s->s_maxbytes = MAX_NON_LFS;
221 s->s_op = &default_op;
222 s->s_time_gran = 1000000000;
223 s->cleancache_poolid = -1;
225 s->s_shrink.seeks = DEFAULT_SEEKS;
226 s->s_shrink.shrink = prune_super;
227 s->s_shrink.batch = 1024;
235 free_percpu(s->s_files);
237 destroy_sb_writers(s);
244 * destroy_super - frees a superblock
245 * @s: superblock to free
247 * Frees a superblock.
249 static inline void destroy_super(struct super_block *s)
252 free_percpu(s->s_files);
254 destroy_sb_writers(s);
256 WARN_ON(!list_empty(&s->s_mounts));
262 /* Superblock refcounting */
265 * Drop a superblock's refcount. The caller must hold sb_lock.
267 static void __put_super(struct super_block *sb)
269 if (!--sb->s_count) {
270 list_del_init(&sb->s_list);
276 * put_super - drop a temporary reference to superblock
277 * @sb: superblock in question
279 * Drops a temporary reference, frees superblock if there's no
282 static void put_super(struct super_block *sb)
286 spin_unlock(&sb_lock);
291 * deactivate_locked_super - drop an active reference to superblock
292 * @s: superblock to deactivate
294 * Drops an active reference to superblock, converting it into a temprory
295 * one if there is no other active references left. In that case we
296 * tell fs driver to shut it down and drop the temporary reference we
299 * Caller holds exclusive lock on superblock; that lock is released.
301 void deactivate_locked_super(struct super_block *s)
303 struct file_system_type *fs = s->s_type;
304 if (atomic_dec_and_test(&s->s_active)) {
305 cleancache_invalidate_fs(s);
308 /* caches are now gone, we can safely kill the shrinker now */
309 unregister_shrinker(&s->s_shrink);
312 * We need to call rcu_barrier so all the delayed rcu free
313 * inodes are flushed before we release the fs module.
319 up_write(&s->s_umount);
323 EXPORT_SYMBOL(deactivate_locked_super);
326 * deactivate_super - drop an active reference to superblock
327 * @s: superblock to deactivate
329 * Variant of deactivate_locked_super(), except that superblock is *not*
330 * locked by caller. If we are going to drop the final active reference,
331 * lock will be acquired prior to that.
333 void deactivate_super(struct super_block *s)
335 if (!atomic_add_unless(&s->s_active, -1, 1)) {
336 down_write(&s->s_umount);
337 deactivate_locked_super(s);
341 EXPORT_SYMBOL(deactivate_super);
344 * grab_super - acquire an active reference
345 * @s: reference we are trying to make active
347 * Tries to acquire an active reference. grab_super() is used when we
348 * had just found a superblock in super_blocks or fs_type->fs_supers
349 * and want to turn it into a full-blown active reference. grab_super()
350 * is called with sb_lock held and drops it. Returns 1 in case of
351 * success, 0 if we had failed (superblock contents was already dead or
352 * dying when grab_super() had been called).
354 static int grab_super(struct super_block *s) __releases(sb_lock)
356 if (atomic_inc_not_zero(&s->s_active)) {
357 spin_unlock(&sb_lock);
360 /* it's going away */
362 spin_unlock(&sb_lock);
363 /* wait for it to die */
364 down_write(&s->s_umount);
365 up_write(&s->s_umount);
371 * grab_super_passive - acquire a passive reference
372 * @sb: reference we are trying to grab
374 * Tries to acquire a passive reference. This is used in places where we
375 * cannot take an active reference but we need to ensure that the
376 * superblock does not go away while we are working on it. It returns
377 * false if a reference was not gained, and returns true with the s_umount
378 * lock held in read mode if a reference is gained. On successful return,
379 * the caller must drop the s_umount lock and the passive reference when
382 bool grab_super_passive(struct super_block *sb)
385 if (hlist_unhashed(&sb->s_instances)) {
386 spin_unlock(&sb_lock);
391 spin_unlock(&sb_lock);
393 if (down_read_trylock(&sb->s_umount)) {
394 if (sb->s_root && (sb->s_flags & MS_BORN))
396 up_read(&sb->s_umount);
404 * Superblock locking. We really ought to get rid of these two.
406 void lock_super(struct super_block * sb)
408 mutex_lock(&sb->s_lock);
411 void unlock_super(struct super_block * sb)
413 mutex_unlock(&sb->s_lock);
416 EXPORT_SYMBOL(lock_super);
417 EXPORT_SYMBOL(unlock_super);
420 * generic_shutdown_super - common helper for ->kill_sb()
421 * @sb: superblock to kill
423 * generic_shutdown_super() does all fs-independent work on superblock
424 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
425 * that need destruction out of superblock, call generic_shutdown_super()
426 * and release aforementioned objects. Note: dentries and inodes _are_
427 * taken care of and do not need specific handling.
429 * Upon calling this function, the filesystem may no longer alter or
430 * rearrange the set of dentries belonging to this super_block, nor may it
431 * change the attachments of dentries to inodes.
433 void generic_shutdown_super(struct super_block *sb)
435 const struct super_operations *sop = sb->s_op;
438 shrink_dcache_for_umount(sb);
440 sb->s_flags &= ~MS_ACTIVE;
442 fsnotify_unmount_inodes(&sb->s_inodes);
449 if (!list_empty(&sb->s_inodes)) {
450 printk("VFS: Busy inodes after unmount of %s. "
451 "Self-destruct in 5 seconds. Have a nice day...\n",
456 /* should be initialized for __put_super_and_need_restart() */
457 hlist_del_init(&sb->s_instances);
458 spin_unlock(&sb_lock);
459 up_write(&sb->s_umount);
462 EXPORT_SYMBOL(generic_shutdown_super);
465 * sget - find or create a superblock
466 * @type: filesystem type superblock should belong to
467 * @test: comparison callback
468 * @set: setup callback
469 * @flags: mount flags
470 * @data: argument to each of them
472 struct super_block *sget(struct file_system_type *type,
473 int (*test)(struct super_block *,void *),
474 int (*set)(struct super_block *,void *),
478 struct super_block *s = NULL;
479 struct hlist_node *node;
480 struct super_block *old;
486 hlist_for_each_entry(old, node, &type->fs_supers, s_instances) {
487 if (!test(old, data))
489 if (!grab_super(old))
492 up_write(&s->s_umount);
496 down_write(&old->s_umount);
497 if (unlikely(!(old->s_flags & MS_BORN))) {
498 deactivate_locked_super(old);
505 spin_unlock(&sb_lock);
506 s = alloc_super(type, flags);
508 return ERR_PTR(-ENOMEM);
514 spin_unlock(&sb_lock);
515 up_write(&s->s_umount);
520 strlcpy(s->s_id, type->name, sizeof(s->s_id));
521 list_add_tail(&s->s_list, &super_blocks);
522 hlist_add_head(&s->s_instances, &type->fs_supers);
523 spin_unlock(&sb_lock);
524 get_filesystem(type);
525 register_shrinker(&s->s_shrink);
531 void drop_super(struct super_block *sb)
533 up_read(&sb->s_umount);
537 EXPORT_SYMBOL(drop_super);
540 * iterate_supers - call function for all active superblocks
541 * @f: function to call
542 * @arg: argument to pass to it
544 * Scans the superblock list and calls given function, passing it
545 * locked superblock and given argument.
547 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
549 struct super_block *sb, *p = NULL;
552 list_for_each_entry(sb, &super_blocks, s_list) {
553 if (hlist_unhashed(&sb->s_instances))
556 spin_unlock(&sb_lock);
558 down_read(&sb->s_umount);
559 if (sb->s_root && (sb->s_flags & MS_BORN))
561 up_read(&sb->s_umount);
570 spin_unlock(&sb_lock);
574 * iterate_supers_type - call function for superblocks of given type
576 * @f: function to call
577 * @arg: argument to pass to it
579 * Scans the superblock list and calls given function, passing it
580 * locked superblock and given argument.
582 void iterate_supers_type(struct file_system_type *type,
583 void (*f)(struct super_block *, void *), void *arg)
585 struct super_block *sb, *p = NULL;
586 struct hlist_node *node;
589 hlist_for_each_entry(sb, node, &type->fs_supers, s_instances) {
591 spin_unlock(&sb_lock);
593 down_read(&sb->s_umount);
594 if (sb->s_root && (sb->s_flags & MS_BORN))
596 up_read(&sb->s_umount);
605 spin_unlock(&sb_lock);
608 EXPORT_SYMBOL(iterate_supers_type);
611 * get_super - get the superblock of a device
612 * @bdev: device to get the superblock for
614 * Scans the superblock list and finds the superblock of the file system
615 * mounted on the device given. %NULL is returned if no match is found.
618 struct super_block *get_super(struct block_device *bdev)
620 struct super_block *sb;
627 list_for_each_entry(sb, &super_blocks, s_list) {
628 if (hlist_unhashed(&sb->s_instances))
630 if (sb->s_bdev == bdev) {
632 spin_unlock(&sb_lock);
633 down_read(&sb->s_umount);
635 if (sb->s_root && (sb->s_flags & MS_BORN))
637 up_read(&sb->s_umount);
638 /* nope, got unmounted */
644 spin_unlock(&sb_lock);
648 EXPORT_SYMBOL(get_super);
651 * get_super_thawed - get thawed superblock of a device
652 * @bdev: device to get the superblock for
654 * Scans the superblock list and finds the superblock of the file system
655 * mounted on the device. The superblock is returned once it is thawed
656 * (or immediately if it was not frozen). %NULL is returned if no match
659 struct super_block *get_super_thawed(struct block_device *bdev)
662 struct super_block *s = get_super(bdev);
663 if (!s || s->s_writers.frozen == SB_UNFROZEN)
665 up_read(&s->s_umount);
666 wait_event(s->s_writers.wait_unfrozen,
667 s->s_writers.frozen == SB_UNFROZEN);
671 EXPORT_SYMBOL(get_super_thawed);
674 * get_active_super - get an active reference to the superblock of a device
675 * @bdev: device to get the superblock for
677 * Scans the superblock list and finds the superblock of the file system
678 * mounted on the device given. Returns the superblock with an active
679 * reference or %NULL if none was found.
681 struct super_block *get_active_super(struct block_device *bdev)
683 struct super_block *sb;
690 list_for_each_entry(sb, &super_blocks, s_list) {
691 if (hlist_unhashed(&sb->s_instances))
693 if (sb->s_bdev == bdev) {
694 if (grab_super(sb)) /* drops sb_lock */
700 spin_unlock(&sb_lock);
704 struct super_block *user_get_super(dev_t dev)
706 struct super_block *sb;
710 list_for_each_entry(sb, &super_blocks, s_list) {
711 if (hlist_unhashed(&sb->s_instances))
713 if (sb->s_dev == dev) {
715 spin_unlock(&sb_lock);
716 down_read(&sb->s_umount);
718 if (sb->s_root && (sb->s_flags & MS_BORN))
720 up_read(&sb->s_umount);
721 /* nope, got unmounted */
727 spin_unlock(&sb_lock);
732 * do_remount_sb - asks filesystem to change mount options.
733 * @sb: superblock in question
734 * @flags: numeric part of options
735 * @data: the rest of options
736 * @force: whether or not to force the change
738 * Alters the mount options of a mounted file system.
740 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
745 if (sb->s_writers.frozen != SB_UNFROZEN)
749 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
753 if (flags & MS_RDONLY)
755 shrink_dcache_sb(sb);
758 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
760 /* If we are remounting RDONLY and current sb is read/write,
761 make sure there are no rw files opened */
766 retval = sb_prepare_remount_readonly(sb);
772 if (sb->s_op->remount_fs) {
773 retval = sb->s_op->remount_fs(sb, &flags, data);
776 goto cancel_readonly;
777 /* If forced remount, go ahead despite any errors */
778 WARN(1, "forced remount of a %s fs returned %i\n",
779 sb->s_type->name, retval);
782 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
783 /* Needs to be ordered wrt mnt_is_readonly() */
785 sb->s_readonly_remount = 0;
788 * Some filesystems modify their metadata via some other path than the
789 * bdev buffer cache (eg. use a private mapping, or directories in
790 * pagecache, etc). Also file data modifications go via their own
791 * mappings. So If we try to mount readonly then copy the filesystem
792 * from bdev, we could get stale data, so invalidate it to give a best
793 * effort at coherency.
795 if (remount_ro && sb->s_bdev)
796 invalidate_bdev(sb->s_bdev);
800 sb->s_readonly_remount = 0;
804 static void do_emergency_remount(struct work_struct *work)
806 struct super_block *sb, *p = NULL;
809 list_for_each_entry(sb, &super_blocks, s_list) {
810 if (hlist_unhashed(&sb->s_instances))
813 spin_unlock(&sb_lock);
814 down_write(&sb->s_umount);
815 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
816 !(sb->s_flags & MS_RDONLY)) {
818 * What lock protects sb->s_flags??
820 do_remount_sb(sb, MS_RDONLY, NULL, 1);
822 up_write(&sb->s_umount);
830 spin_unlock(&sb_lock);
832 printk("Emergency Remount complete\n");
835 void emergency_remount(void)
837 struct work_struct *work;
839 work = kmalloc(sizeof(*work), GFP_ATOMIC);
841 INIT_WORK(work, do_emergency_remount);
847 * Unnamed block devices are dummy devices used by virtual
848 * filesystems which don't use real block-devices. -- jrs
851 static DEFINE_IDA(unnamed_dev_ida);
852 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
853 static int unnamed_dev_start = 0; /* don't bother trying below it */
855 int get_anon_bdev(dev_t *p)
861 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
863 spin_lock(&unnamed_dev_lock);
864 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
866 unnamed_dev_start = dev + 1;
867 spin_unlock(&unnamed_dev_lock);
868 if (error == -EAGAIN)
869 /* We raced and lost with another CPU. */
874 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
875 spin_lock(&unnamed_dev_lock);
876 ida_remove(&unnamed_dev_ida, dev);
877 if (unnamed_dev_start > dev)
878 unnamed_dev_start = dev;
879 spin_unlock(&unnamed_dev_lock);
882 *p = MKDEV(0, dev & MINORMASK);
885 EXPORT_SYMBOL(get_anon_bdev);
887 void free_anon_bdev(dev_t dev)
889 int slot = MINOR(dev);
890 spin_lock(&unnamed_dev_lock);
891 ida_remove(&unnamed_dev_ida, slot);
892 if (slot < unnamed_dev_start)
893 unnamed_dev_start = slot;
894 spin_unlock(&unnamed_dev_lock);
896 EXPORT_SYMBOL(free_anon_bdev);
898 int set_anon_super(struct super_block *s, void *data)
900 int error = get_anon_bdev(&s->s_dev);
902 s->s_bdi = &noop_backing_dev_info;
906 EXPORT_SYMBOL(set_anon_super);
908 void kill_anon_super(struct super_block *sb)
910 dev_t dev = sb->s_dev;
911 generic_shutdown_super(sb);
915 EXPORT_SYMBOL(kill_anon_super);
917 void kill_litter_super(struct super_block *sb)
920 d_genocide(sb->s_root);
924 EXPORT_SYMBOL(kill_litter_super);
926 static int ns_test_super(struct super_block *sb, void *data)
928 return sb->s_fs_info == data;
931 static int ns_set_super(struct super_block *sb, void *data)
933 sb->s_fs_info = data;
934 return set_anon_super(sb, NULL);
937 struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
938 void *data, int (*fill_super)(struct super_block *, void *, int))
940 struct super_block *sb;
942 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
948 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
950 deactivate_locked_super(sb);
954 sb->s_flags |= MS_ACTIVE;
957 return dget(sb->s_root);
960 EXPORT_SYMBOL(mount_ns);
963 static int set_bdev_super(struct super_block *s, void *data)
966 s->s_dev = s->s_bdev->bd_dev;
969 * We set the bdi here to the queue backing, file systems can
970 * overwrite this in ->fill_super()
972 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
976 static int test_bdev_super(struct super_block *s, void *data)
978 return (void *)s->s_bdev == data;
981 struct dentry *mount_bdev(struct file_system_type *fs_type,
982 int flags, const char *dev_name, void *data,
983 int (*fill_super)(struct super_block *, void *, int))
985 struct block_device *bdev;
986 struct super_block *s;
987 fmode_t mode = FMODE_READ | FMODE_EXCL;
990 if (!(flags & MS_RDONLY))
993 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
995 return ERR_CAST(bdev);
998 * once the super is inserted into the list by sget, s_umount
999 * will protect the lockfs code from trying to start a snapshot
1000 * while we are mounting
1002 mutex_lock(&bdev->bd_fsfreeze_mutex);
1003 if (bdev->bd_fsfreeze_count > 0) {
1004 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1008 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
1010 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1015 if ((flags ^ s->s_flags) & MS_RDONLY) {
1016 deactivate_locked_super(s);
1022 * s_umount nests inside bd_mutex during
1023 * __invalidate_device(). blkdev_put() acquires
1024 * bd_mutex and can't be called under s_umount. Drop
1025 * s_umount temporarily. This is safe as we're
1026 * holding an active reference.
1028 up_write(&s->s_umount);
1029 blkdev_put(bdev, mode);
1030 down_write(&s->s_umount);
1032 char b[BDEVNAME_SIZE];
1035 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1036 sb_set_blocksize(s, block_size(bdev));
1037 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1039 deactivate_locked_super(s);
1043 s->s_flags |= MS_ACTIVE;
1047 return dget(s->s_root);
1052 blkdev_put(bdev, mode);
1054 return ERR_PTR(error);
1056 EXPORT_SYMBOL(mount_bdev);
1058 void kill_block_super(struct super_block *sb)
1060 struct block_device *bdev = sb->s_bdev;
1061 fmode_t mode = sb->s_mode;
1063 bdev->bd_super = NULL;
1064 generic_shutdown_super(sb);
1065 sync_blockdev(bdev);
1066 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1067 blkdev_put(bdev, mode | FMODE_EXCL);
1070 EXPORT_SYMBOL(kill_block_super);
1073 struct dentry *mount_nodev(struct file_system_type *fs_type,
1074 int flags, void *data,
1075 int (*fill_super)(struct super_block *, void *, int))
1078 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1083 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1085 deactivate_locked_super(s);
1086 return ERR_PTR(error);
1088 s->s_flags |= MS_ACTIVE;
1089 return dget(s->s_root);
1091 EXPORT_SYMBOL(mount_nodev);
1093 static int compare_single(struct super_block *s, void *p)
1098 struct dentry *mount_single(struct file_system_type *fs_type,
1099 int flags, void *data,
1100 int (*fill_super)(struct super_block *, void *, int))
1102 struct super_block *s;
1105 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1109 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1111 deactivate_locked_super(s);
1112 return ERR_PTR(error);
1114 s->s_flags |= MS_ACTIVE;
1116 do_remount_sb(s, flags, data, 0);
1118 return dget(s->s_root);
1120 EXPORT_SYMBOL(mount_single);
1123 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1125 struct dentry *root;
1126 struct super_block *sb;
1127 char *secdata = NULL;
1128 int error = -ENOMEM;
1130 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1131 secdata = alloc_secdata();
1135 error = security_sb_copy_data(data, secdata);
1137 goto out_free_secdata;
1140 root = type->mount(type, flags, name, data);
1142 error = PTR_ERR(root);
1143 goto out_free_secdata;
1147 WARN_ON(!sb->s_bdi);
1148 WARN_ON(sb->s_bdi == &default_backing_dev_info);
1149 sb->s_flags |= MS_BORN;
1151 error = security_sb_kern_mount(sb, flags, secdata);
1156 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1157 * but s_maxbytes was an unsigned long long for many releases. Throw
1158 * this warning for a little while to try and catch filesystems that
1159 * violate this rule.
1161 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1162 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1164 up_write(&sb->s_umount);
1165 free_secdata(secdata);
1169 deactivate_locked_super(sb);
1171 free_secdata(secdata);
1173 return ERR_PTR(error);
1177 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1180 void __sb_end_write(struct super_block *sb, int level)
1182 percpu_counter_dec(&sb->s_writers.counter[level-1]);
1184 * Make sure s_writers are updated before we wake up waiters in
1188 if (waitqueue_active(&sb->s_writers.wait))
1189 wake_up(&sb->s_writers.wait);
1190 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _RET_IP_);
1192 EXPORT_SYMBOL(__sb_end_write);
1194 #ifdef CONFIG_LOCKDEP
1196 * We want lockdep to tell us about possible deadlocks with freezing but
1197 * it's it bit tricky to properly instrument it. Getting a freeze protection
1198 * works as getting a read lock but there are subtle problems. XFS for example
1199 * gets freeze protection on internal level twice in some cases, which is OK
1200 * only because we already hold a freeze protection also on higher level. Due
1201 * to these cases we have to tell lockdep we are doing trylock when we
1202 * already hold a freeze protection for a higher freeze level.
1204 static void acquire_freeze_lock(struct super_block *sb, int level, bool trylock,
1210 for (i = 0; i < level - 1; i++)
1211 if (lock_is_held(&sb->s_writers.lock_map[i])) {
1216 rwsem_acquire_read(&sb->s_writers.lock_map[level-1], 0, trylock, ip);
1221 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1224 int __sb_start_write(struct super_block *sb, int level, bool wait)
1227 if (unlikely(sb->s_writers.frozen >= level)) {
1230 wait_event(sb->s_writers.wait_unfrozen,
1231 sb->s_writers.frozen < level);
1234 #ifdef CONFIG_LOCKDEP
1235 acquire_freeze_lock(sb, level, !wait, _RET_IP_);
1237 percpu_counter_inc(&sb->s_writers.counter[level-1]);
1239 * Make sure counter is updated before we check for frozen.
1240 * freeze_super() first sets frozen and then checks the counter.
1243 if (unlikely(sb->s_writers.frozen >= level)) {
1244 __sb_end_write(sb, level);
1249 EXPORT_SYMBOL(__sb_start_write);
1252 * sb_wait_write - wait until all writers to given file system finish
1253 * @sb: the super for which we wait
1254 * @level: type of writers we wait for (normal vs page fault)
1256 * This function waits until there are no writers of given type to given file
1257 * system. Caller of this function should make sure there can be no new writers
1258 * of type @level before calling this function. Otherwise this function can
1261 static void sb_wait_write(struct super_block *sb, int level)
1266 * We just cycle-through lockdep here so that it does not complain
1267 * about returning with lock to userspace
1269 rwsem_acquire(&sb->s_writers.lock_map[level-1], 0, 0, _THIS_IP_);
1270 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _THIS_IP_);
1276 * We use a barrier in prepare_to_wait() to separate setting
1277 * of frozen and checking of the counter
1279 prepare_to_wait(&sb->s_writers.wait, &wait,
1280 TASK_UNINTERRUPTIBLE);
1282 writers = percpu_counter_sum(&sb->s_writers.counter[level-1]);
1286 finish_wait(&sb->s_writers.wait, &wait);
1291 * freeze_super - lock the filesystem and force it into a consistent state
1292 * @sb: the super to lock
1294 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1295 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1298 * During this function, sb->s_writers.frozen goes through these values:
1300 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1302 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1303 * writes should be blocked, though page faults are still allowed. We wait for
1304 * all writes to complete and then proceed to the next stage.
1306 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1307 * but internal fs threads can still modify the filesystem (although they
1308 * should not dirty new pages or inodes), writeback can run etc. After waiting
1309 * for all running page faults we sync the filesystem which will clean all
1310 * dirty pages and inodes (no new dirty pages or inodes can be created when
1313 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1314 * modification are blocked (e.g. XFS preallocation truncation on inode
1315 * reclaim). This is usually implemented by blocking new transactions for
1316 * filesystems that have them and need this additional guard. After all
1317 * internal writers are finished we call ->freeze_fs() to finish filesystem
1318 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1319 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1321 * sb->s_writers.frozen is protected by sb->s_umount.
1323 int freeze_super(struct super_block *sb)
1327 atomic_inc(&sb->s_active);
1328 down_write(&sb->s_umount);
1329 if (sb->s_writers.frozen != SB_UNFROZEN) {
1330 deactivate_locked_super(sb);
1334 if (!(sb->s_flags & MS_BORN)) {
1335 up_write(&sb->s_umount);
1336 return 0; /* sic - it's "nothing to do" */
1339 if (sb->s_flags & MS_RDONLY) {
1340 /* Nothing to do really... */
1341 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1342 up_write(&sb->s_umount);
1346 /* From now on, no new normal writers can start */
1347 sb->s_writers.frozen = SB_FREEZE_WRITE;
1350 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1351 up_write(&sb->s_umount);
1353 sb_wait_write(sb, SB_FREEZE_WRITE);
1355 /* Now we go and block page faults... */
1356 down_write(&sb->s_umount);
1357 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1360 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1362 /* All writers are done so after syncing there won't be dirty data */
1363 sync_filesystem(sb);
1365 /* Now wait for internal filesystem counter */
1366 sb->s_writers.frozen = SB_FREEZE_FS;
1368 sb_wait_write(sb, SB_FREEZE_FS);
1370 if (sb->s_op->freeze_fs) {
1371 ret = sb->s_op->freeze_fs(sb);
1374 "VFS:Filesystem freeze failed\n");
1375 sb->s_writers.frozen = SB_UNFROZEN;
1377 wake_up(&sb->s_writers.wait_unfrozen);
1378 deactivate_locked_super(sb);
1383 * This is just for debugging purposes so that fs can warn if it
1384 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1386 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1387 up_write(&sb->s_umount);
1390 EXPORT_SYMBOL(freeze_super);
1393 * thaw_super -- unlock filesystem
1394 * @sb: the super to thaw
1396 * Unlocks the filesystem and marks it writeable again after freeze_super().
1398 int thaw_super(struct super_block *sb)
1402 down_write(&sb->s_umount);
1403 if (sb->s_writers.frozen == SB_UNFROZEN) {
1404 up_write(&sb->s_umount);
1408 if (sb->s_flags & MS_RDONLY)
1411 if (sb->s_op->unfreeze_fs) {
1412 error = sb->s_op->unfreeze_fs(sb);
1415 "VFS:Filesystem thaw failed\n");
1416 up_write(&sb->s_umount);
1422 sb->s_writers.frozen = SB_UNFROZEN;
1424 wake_up(&sb->s_writers.wait_unfrozen);
1425 deactivate_locked_super(sb);
1429 EXPORT_SYMBOL(thaw_super);