mtd: sa11x0: remove definitions and code left for documentation purposes
[linux-2.6.git] / fs / super.c
blob6277ec6cb60a65e71b2c590260c67a057a1748ca
1 /*
2 * linux/fs/super.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * super.c contains code to handle: - mount structures
7 * - super-block tables
8 * - filesystem drivers list
9 * - mount system call
10 * - umount system call
11 * - ustat 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/module.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 "internal.h"
38 LIST_HEAD(super_blocks);
39 DEFINE_SPINLOCK(sb_lock);
42 * One thing we have to be careful of with a per-sb shrinker is that we don't
43 * drop the last active reference to the superblock from within the shrinker.
44 * If that happens we could trigger unregistering the shrinker from within the
45 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
46 * take a passive reference to the superblock to avoid this from occurring.
48 static int prune_super(struct shrinker *shrink, struct shrink_control *sc)
50 struct super_block *sb;
51 int fs_objects = 0;
52 int total_objects;
54 sb = container_of(shrink, struct super_block, s_shrink);
57 * Deadlock avoidance. We may hold various FS locks, and we don't want
58 * to recurse into the FS that called us in clear_inode() and friends..
60 if (sc->nr_to_scan && !(sc->gfp_mask & __GFP_FS))
61 return -1;
63 if (!grab_super_passive(sb))
64 return !sc->nr_to_scan ? 0 : -1;
66 if (sb->s_op && sb->s_op->nr_cached_objects)
67 fs_objects = sb->s_op->nr_cached_objects(sb);
69 total_objects = sb->s_nr_dentry_unused +
70 sb->s_nr_inodes_unused + fs_objects + 1;
72 if (sc->nr_to_scan) {
73 int dentries;
74 int inodes;
76 /* proportion the scan between the caches */
77 dentries = (sc->nr_to_scan * sb->s_nr_dentry_unused) /
78 total_objects;
79 inodes = (sc->nr_to_scan * sb->s_nr_inodes_unused) /
80 total_objects;
81 if (fs_objects)
82 fs_objects = (sc->nr_to_scan * fs_objects) /
83 total_objects;
85 * prune the dcache first as the icache is pinned by it, then
86 * prune the icache, followed by the filesystem specific caches
88 prune_dcache_sb(sb, dentries);
89 prune_icache_sb(sb, inodes);
91 if (fs_objects && sb->s_op->free_cached_objects) {
92 sb->s_op->free_cached_objects(sb, fs_objects);
93 fs_objects = sb->s_op->nr_cached_objects(sb);
95 total_objects = sb->s_nr_dentry_unused +
96 sb->s_nr_inodes_unused + fs_objects;
99 total_objects = (total_objects / 100) * sysctl_vfs_cache_pressure;
100 drop_super(sb);
101 return total_objects;
105 * alloc_super - create new superblock
106 * @type: filesystem type superblock should belong to
108 * Allocates and initializes a new &struct super_block. alloc_super()
109 * returns a pointer new superblock or %NULL if allocation had failed.
111 static struct super_block *alloc_super(struct file_system_type *type)
113 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
114 static const struct super_operations default_op;
116 if (s) {
117 if (security_sb_alloc(s)) {
118 kfree(s);
119 s = NULL;
120 goto out;
122 #ifdef CONFIG_SMP
123 s->s_files = alloc_percpu(struct list_head);
124 if (!s->s_files) {
125 security_sb_free(s);
126 kfree(s);
127 s = NULL;
128 goto out;
129 } else {
130 int i;
132 for_each_possible_cpu(i)
133 INIT_LIST_HEAD(per_cpu_ptr(s->s_files, i));
135 #else
136 INIT_LIST_HEAD(&s->s_files);
137 #endif
138 s->s_bdi = &default_backing_dev_info;
139 INIT_HLIST_NODE(&s->s_instances);
140 INIT_HLIST_BL_HEAD(&s->s_anon);
141 INIT_LIST_HEAD(&s->s_inodes);
142 INIT_LIST_HEAD(&s->s_dentry_lru);
143 INIT_LIST_HEAD(&s->s_inode_lru);
144 spin_lock_init(&s->s_inode_lru_lock);
145 INIT_LIST_HEAD(&s->s_mounts);
146 init_rwsem(&s->s_umount);
147 mutex_init(&s->s_lock);
148 lockdep_set_class(&s->s_umount, &type->s_umount_key);
150 * The locking rules for s_lock are up to the
151 * filesystem. For example ext3fs has different
152 * lock ordering than usbfs:
154 lockdep_set_class(&s->s_lock, &type->s_lock_key);
156 * sget() can have s_umount recursion.
158 * When it cannot find a suitable sb, it allocates a new
159 * one (this one), and tries again to find a suitable old
160 * one.
162 * In case that succeeds, it will acquire the s_umount
163 * lock of the old one. Since these are clearly distrinct
164 * locks, and this object isn't exposed yet, there's no
165 * risk of deadlocks.
167 * Annotate this by putting this lock in a different
168 * subclass.
170 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
171 s->s_count = 1;
172 atomic_set(&s->s_active, 1);
173 mutex_init(&s->s_vfs_rename_mutex);
174 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
175 mutex_init(&s->s_dquot.dqio_mutex);
176 mutex_init(&s->s_dquot.dqonoff_mutex);
177 init_rwsem(&s->s_dquot.dqptr_sem);
178 init_waitqueue_head(&s->s_wait_unfrozen);
179 s->s_maxbytes = MAX_NON_LFS;
180 s->s_op = &default_op;
181 s->s_time_gran = 1000000000;
182 s->cleancache_poolid = -1;
184 s->s_shrink.seeks = DEFAULT_SEEKS;
185 s->s_shrink.shrink = prune_super;
186 s->s_shrink.batch = 1024;
188 out:
189 return s;
193 * destroy_super - frees a superblock
194 * @s: superblock to free
196 * Frees a superblock.
198 static inline void destroy_super(struct super_block *s)
200 #ifdef CONFIG_SMP
201 free_percpu(s->s_files);
202 #endif
203 security_sb_free(s);
204 WARN_ON(!list_empty(&s->s_mounts));
205 kfree(s->s_subtype);
206 kfree(s->s_options);
207 kfree(s);
210 /* Superblock refcounting */
213 * Drop a superblock's refcount. The caller must hold sb_lock.
215 static void __put_super(struct super_block *sb)
217 if (!--sb->s_count) {
218 list_del_init(&sb->s_list);
219 destroy_super(sb);
224 * put_super - drop a temporary reference to superblock
225 * @sb: superblock in question
227 * Drops a temporary reference, frees superblock if there's no
228 * references left.
230 static void put_super(struct super_block *sb)
232 spin_lock(&sb_lock);
233 __put_super(sb);
234 spin_unlock(&sb_lock);
239 * deactivate_locked_super - drop an active reference to superblock
240 * @s: superblock to deactivate
242 * Drops an active reference to superblock, converting it into a temprory
243 * one if there is no other active references left. In that case we
244 * tell fs driver to shut it down and drop the temporary reference we
245 * had just acquired.
247 * Caller holds exclusive lock on superblock; that lock is released.
249 void deactivate_locked_super(struct super_block *s)
251 struct file_system_type *fs = s->s_type;
252 if (atomic_dec_and_test(&s->s_active)) {
253 cleancache_flush_fs(s);
254 fs->kill_sb(s);
256 /* caches are now gone, we can safely kill the shrinker now */
257 unregister_shrinker(&s->s_shrink);
260 * We need to call rcu_barrier so all the delayed rcu free
261 * inodes are flushed before we release the fs module.
263 rcu_barrier();
264 put_filesystem(fs);
265 put_super(s);
266 } else {
267 up_write(&s->s_umount);
271 EXPORT_SYMBOL(deactivate_locked_super);
274 * deactivate_super - drop an active reference to superblock
275 * @s: superblock to deactivate
277 * Variant of deactivate_locked_super(), except that superblock is *not*
278 * locked by caller. If we are going to drop the final active reference,
279 * lock will be acquired prior to that.
281 void deactivate_super(struct super_block *s)
283 if (!atomic_add_unless(&s->s_active, -1, 1)) {
284 down_write(&s->s_umount);
285 deactivate_locked_super(s);
289 EXPORT_SYMBOL(deactivate_super);
292 * grab_super - acquire an active reference
293 * @s: reference we are trying to make active
295 * Tries to acquire an active reference. grab_super() is used when we
296 * had just found a superblock in super_blocks or fs_type->fs_supers
297 * and want to turn it into a full-blown active reference. grab_super()
298 * is called with sb_lock held and drops it. Returns 1 in case of
299 * success, 0 if we had failed (superblock contents was already dead or
300 * dying when grab_super() had been called).
302 static int grab_super(struct super_block *s) __releases(sb_lock)
304 if (atomic_inc_not_zero(&s->s_active)) {
305 spin_unlock(&sb_lock);
306 return 1;
308 /* it's going away */
309 s->s_count++;
310 spin_unlock(&sb_lock);
311 /* wait for it to die */
312 down_write(&s->s_umount);
313 up_write(&s->s_umount);
314 put_super(s);
315 return 0;
319 * grab_super_passive - acquire a passive reference
320 * @s: reference we are trying to grab
322 * Tries to acquire a passive reference. This is used in places where we
323 * cannot take an active reference but we need to ensure that the
324 * superblock does not go away while we are working on it. It returns
325 * false if a reference was not gained, and returns true with the s_umount
326 * lock held in read mode if a reference is gained. On successful return,
327 * the caller must drop the s_umount lock and the passive reference when
328 * done.
330 bool grab_super_passive(struct super_block *sb)
332 spin_lock(&sb_lock);
333 if (hlist_unhashed(&sb->s_instances)) {
334 spin_unlock(&sb_lock);
335 return false;
338 sb->s_count++;
339 spin_unlock(&sb_lock);
341 if (down_read_trylock(&sb->s_umount)) {
342 if (sb->s_root && (sb->s_flags & MS_BORN))
343 return true;
344 up_read(&sb->s_umount);
347 put_super(sb);
348 return false;
352 * Superblock locking. We really ought to get rid of these two.
354 void lock_super(struct super_block * sb)
356 mutex_lock(&sb->s_lock);
359 void unlock_super(struct super_block * sb)
361 mutex_unlock(&sb->s_lock);
364 EXPORT_SYMBOL(lock_super);
365 EXPORT_SYMBOL(unlock_super);
368 * generic_shutdown_super - common helper for ->kill_sb()
369 * @sb: superblock to kill
371 * generic_shutdown_super() does all fs-independent work on superblock
372 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
373 * that need destruction out of superblock, call generic_shutdown_super()
374 * and release aforementioned objects. Note: dentries and inodes _are_
375 * taken care of and do not need specific handling.
377 * Upon calling this function, the filesystem may no longer alter or
378 * rearrange the set of dentries belonging to this super_block, nor may it
379 * change the attachments of dentries to inodes.
381 void generic_shutdown_super(struct super_block *sb)
383 const struct super_operations *sop = sb->s_op;
385 if (sb->s_root) {
386 shrink_dcache_for_umount(sb);
387 sync_filesystem(sb);
388 sb->s_flags &= ~MS_ACTIVE;
390 fsnotify_unmount_inodes(&sb->s_inodes);
392 evict_inodes(sb);
394 if (sop->put_super)
395 sop->put_super(sb);
397 if (!list_empty(&sb->s_inodes)) {
398 printk("VFS: Busy inodes after unmount of %s. "
399 "Self-destruct in 5 seconds. Have a nice day...\n",
400 sb->s_id);
403 spin_lock(&sb_lock);
404 /* should be initialized for __put_super_and_need_restart() */
405 hlist_del_init(&sb->s_instances);
406 spin_unlock(&sb_lock);
407 up_write(&sb->s_umount);
410 EXPORT_SYMBOL(generic_shutdown_super);
413 * sget - find or create a superblock
414 * @type: filesystem type superblock should belong to
415 * @test: comparison callback
416 * @set: setup callback
417 * @data: argument to each of them
419 struct super_block *sget(struct file_system_type *type,
420 int (*test)(struct super_block *,void *),
421 int (*set)(struct super_block *,void *),
422 void *data)
424 struct super_block *s = NULL;
425 struct hlist_node *node;
426 struct super_block *old;
427 int err;
429 retry:
430 spin_lock(&sb_lock);
431 if (test) {
432 hlist_for_each_entry(old, node, &type->fs_supers, s_instances) {
433 if (!test(old, data))
434 continue;
435 if (!grab_super(old))
436 goto retry;
437 if (s) {
438 up_write(&s->s_umount);
439 destroy_super(s);
440 s = NULL;
442 down_write(&old->s_umount);
443 if (unlikely(!(old->s_flags & MS_BORN))) {
444 deactivate_locked_super(old);
445 goto retry;
447 return old;
450 if (!s) {
451 spin_unlock(&sb_lock);
452 s = alloc_super(type);
453 if (!s)
454 return ERR_PTR(-ENOMEM);
455 goto retry;
458 err = set(s, data);
459 if (err) {
460 spin_unlock(&sb_lock);
461 up_write(&s->s_umount);
462 destroy_super(s);
463 return ERR_PTR(err);
465 s->s_type = type;
466 strlcpy(s->s_id, type->name, sizeof(s->s_id));
467 list_add_tail(&s->s_list, &super_blocks);
468 hlist_add_head(&s->s_instances, &type->fs_supers);
469 spin_unlock(&sb_lock);
470 get_filesystem(type);
471 register_shrinker(&s->s_shrink);
472 return s;
475 EXPORT_SYMBOL(sget);
477 void drop_super(struct super_block *sb)
479 up_read(&sb->s_umount);
480 put_super(sb);
483 EXPORT_SYMBOL(drop_super);
486 * sync_supers - helper for periodic superblock writeback
488 * Call the write_super method if present on all dirty superblocks in
489 * the system. This is for the periodic writeback used by most older
490 * filesystems. For data integrity superblock writeback use
491 * sync_filesystems() instead.
493 * Note: check the dirty flag before waiting, so we don't
494 * hold up the sync while mounting a device. (The newly
495 * mounted device won't need syncing.)
497 void sync_supers(void)
499 struct super_block *sb, *p = NULL;
501 spin_lock(&sb_lock);
502 list_for_each_entry(sb, &super_blocks, s_list) {
503 if (hlist_unhashed(&sb->s_instances))
504 continue;
505 if (sb->s_op->write_super && sb->s_dirt) {
506 sb->s_count++;
507 spin_unlock(&sb_lock);
509 down_read(&sb->s_umount);
510 if (sb->s_root && sb->s_dirt && (sb->s_flags & MS_BORN))
511 sb->s_op->write_super(sb);
512 up_read(&sb->s_umount);
514 spin_lock(&sb_lock);
515 if (p)
516 __put_super(p);
517 p = sb;
520 if (p)
521 __put_super(p);
522 spin_unlock(&sb_lock);
526 * iterate_supers - call function for all active superblocks
527 * @f: function to call
528 * @arg: argument to pass to it
530 * Scans the superblock list and calls given function, passing it
531 * locked superblock and given argument.
533 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
535 struct super_block *sb, *p = NULL;
537 spin_lock(&sb_lock);
538 list_for_each_entry(sb, &super_blocks, s_list) {
539 if (hlist_unhashed(&sb->s_instances))
540 continue;
541 sb->s_count++;
542 spin_unlock(&sb_lock);
544 down_read(&sb->s_umount);
545 if (sb->s_root && (sb->s_flags & MS_BORN))
546 f(sb, arg);
547 up_read(&sb->s_umount);
549 spin_lock(&sb_lock);
550 if (p)
551 __put_super(p);
552 p = sb;
554 if (p)
555 __put_super(p);
556 spin_unlock(&sb_lock);
560 * iterate_supers_type - call function for superblocks of given type
561 * @type: fs type
562 * @f: function to call
563 * @arg: argument to pass to it
565 * Scans the superblock list and calls given function, passing it
566 * locked superblock and given argument.
568 void iterate_supers_type(struct file_system_type *type,
569 void (*f)(struct super_block *, void *), void *arg)
571 struct super_block *sb, *p = NULL;
572 struct hlist_node *node;
574 spin_lock(&sb_lock);
575 hlist_for_each_entry(sb, node, &type->fs_supers, s_instances) {
576 sb->s_count++;
577 spin_unlock(&sb_lock);
579 down_read(&sb->s_umount);
580 if (sb->s_root && (sb->s_flags & MS_BORN))
581 f(sb, arg);
582 up_read(&sb->s_umount);
584 spin_lock(&sb_lock);
585 if (p)
586 __put_super(p);
587 p = sb;
589 if (p)
590 __put_super(p);
591 spin_unlock(&sb_lock);
594 EXPORT_SYMBOL(iterate_supers_type);
597 * get_super - get the superblock of a device
598 * @bdev: device to get the superblock for
600 * Scans the superblock list and finds the superblock of the file system
601 * mounted on the device given. %NULL is returned if no match is found.
604 struct super_block *get_super(struct block_device *bdev)
606 struct super_block *sb;
608 if (!bdev)
609 return NULL;
611 spin_lock(&sb_lock);
612 rescan:
613 list_for_each_entry(sb, &super_blocks, s_list) {
614 if (hlist_unhashed(&sb->s_instances))
615 continue;
616 if (sb->s_bdev == bdev) {
617 sb->s_count++;
618 spin_unlock(&sb_lock);
619 down_read(&sb->s_umount);
620 /* still alive? */
621 if (sb->s_root && (sb->s_flags & MS_BORN))
622 return sb;
623 up_read(&sb->s_umount);
624 /* nope, got unmounted */
625 spin_lock(&sb_lock);
626 __put_super(sb);
627 goto rescan;
630 spin_unlock(&sb_lock);
631 return NULL;
634 EXPORT_SYMBOL(get_super);
637 * get_super_thawed - get thawed superblock of a device
638 * @bdev: device to get the superblock for
640 * Scans the superblock list and finds the superblock of the file system
641 * mounted on the device. The superblock is returned once it is thawed
642 * (or immediately if it was not frozen). %NULL is returned if no match
643 * is found.
645 struct super_block *get_super_thawed(struct block_device *bdev)
647 while (1) {
648 struct super_block *s = get_super(bdev);
649 if (!s || s->s_frozen == SB_UNFROZEN)
650 return s;
651 up_read(&s->s_umount);
652 vfs_check_frozen(s, SB_FREEZE_WRITE);
653 put_super(s);
656 EXPORT_SYMBOL(get_super_thawed);
659 * get_active_super - get an active reference to the superblock of a device
660 * @bdev: device to get the superblock for
662 * Scans the superblock list and finds the superblock of the file system
663 * mounted on the device given. Returns the superblock with an active
664 * reference or %NULL if none was found.
666 struct super_block *get_active_super(struct block_device *bdev)
668 struct super_block *sb;
670 if (!bdev)
671 return NULL;
673 restart:
674 spin_lock(&sb_lock);
675 list_for_each_entry(sb, &super_blocks, s_list) {
676 if (hlist_unhashed(&sb->s_instances))
677 continue;
678 if (sb->s_bdev == bdev) {
679 if (grab_super(sb)) /* drops sb_lock */
680 return sb;
681 else
682 goto restart;
685 spin_unlock(&sb_lock);
686 return NULL;
689 struct super_block *user_get_super(dev_t dev)
691 struct super_block *sb;
693 spin_lock(&sb_lock);
694 rescan:
695 list_for_each_entry(sb, &super_blocks, s_list) {
696 if (hlist_unhashed(&sb->s_instances))
697 continue;
698 if (sb->s_dev == dev) {
699 sb->s_count++;
700 spin_unlock(&sb_lock);
701 down_read(&sb->s_umount);
702 /* still alive? */
703 if (sb->s_root && (sb->s_flags & MS_BORN))
704 return sb;
705 up_read(&sb->s_umount);
706 /* nope, got unmounted */
707 spin_lock(&sb_lock);
708 __put_super(sb);
709 goto rescan;
712 spin_unlock(&sb_lock);
713 return NULL;
717 * do_remount_sb - asks filesystem to change mount options.
718 * @sb: superblock in question
719 * @flags: numeric part of options
720 * @data: the rest of options
721 * @force: whether or not to force the change
723 * Alters the mount options of a mounted file system.
725 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
727 int retval;
728 int remount_ro;
730 if (sb->s_frozen != SB_UNFROZEN)
731 return -EBUSY;
733 #ifdef CONFIG_BLOCK
734 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
735 return -EACCES;
736 #endif
738 if (flags & MS_RDONLY)
739 acct_auto_close(sb);
740 shrink_dcache_sb(sb);
741 sync_filesystem(sb);
743 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
745 /* If we are remounting RDONLY and current sb is read/write,
746 make sure there are no rw files opened */
747 if (remount_ro) {
748 if (force) {
749 mark_files_ro(sb);
750 } else {
751 retval = sb_prepare_remount_readonly(sb);
752 if (retval)
753 return retval;
757 if (sb->s_op->remount_fs) {
758 retval = sb->s_op->remount_fs(sb, &flags, data);
759 if (retval) {
760 if (!force)
761 goto cancel_readonly;
762 /* If forced remount, go ahead despite any errors */
763 WARN(1, "forced remount of a %s fs returned %i\n",
764 sb->s_type->name, retval);
767 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
768 /* Needs to be ordered wrt mnt_is_readonly() */
769 smp_wmb();
770 sb->s_readonly_remount = 0;
773 * Some filesystems modify their metadata via some other path than the
774 * bdev buffer cache (eg. use a private mapping, or directories in
775 * pagecache, etc). Also file data modifications go via their own
776 * mappings. So If we try to mount readonly then copy the filesystem
777 * from bdev, we could get stale data, so invalidate it to give a best
778 * effort at coherency.
780 if (remount_ro && sb->s_bdev)
781 invalidate_bdev(sb->s_bdev);
782 return 0;
784 cancel_readonly:
785 sb->s_readonly_remount = 0;
786 return retval;
789 static void do_emergency_remount(struct work_struct *work)
791 struct super_block *sb, *p = NULL;
793 spin_lock(&sb_lock);
794 list_for_each_entry(sb, &super_blocks, s_list) {
795 if (hlist_unhashed(&sb->s_instances))
796 continue;
797 sb->s_count++;
798 spin_unlock(&sb_lock);
799 down_write(&sb->s_umount);
800 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
801 !(sb->s_flags & MS_RDONLY)) {
803 * What lock protects sb->s_flags??
805 do_remount_sb(sb, MS_RDONLY, NULL, 1);
807 up_write(&sb->s_umount);
808 spin_lock(&sb_lock);
809 if (p)
810 __put_super(p);
811 p = sb;
813 if (p)
814 __put_super(p);
815 spin_unlock(&sb_lock);
816 kfree(work);
817 printk("Emergency Remount complete\n");
820 void emergency_remount(void)
822 struct work_struct *work;
824 work = kmalloc(sizeof(*work), GFP_ATOMIC);
825 if (work) {
826 INIT_WORK(work, do_emergency_remount);
827 schedule_work(work);
832 * Unnamed block devices are dummy devices used by virtual
833 * filesystems which don't use real block-devices. -- jrs
836 static DEFINE_IDA(unnamed_dev_ida);
837 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
838 static int unnamed_dev_start = 0; /* don't bother trying below it */
840 int get_anon_bdev(dev_t *p)
842 int dev;
843 int error;
845 retry:
846 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
847 return -ENOMEM;
848 spin_lock(&unnamed_dev_lock);
849 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
850 if (!error)
851 unnamed_dev_start = dev + 1;
852 spin_unlock(&unnamed_dev_lock);
853 if (error == -EAGAIN)
854 /* We raced and lost with another CPU. */
855 goto retry;
856 else if (error)
857 return -EAGAIN;
859 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
860 spin_lock(&unnamed_dev_lock);
861 ida_remove(&unnamed_dev_ida, dev);
862 if (unnamed_dev_start > dev)
863 unnamed_dev_start = dev;
864 spin_unlock(&unnamed_dev_lock);
865 return -EMFILE;
867 *p = MKDEV(0, dev & MINORMASK);
868 return 0;
870 EXPORT_SYMBOL(get_anon_bdev);
872 void free_anon_bdev(dev_t dev)
874 int slot = MINOR(dev);
875 spin_lock(&unnamed_dev_lock);
876 ida_remove(&unnamed_dev_ida, slot);
877 if (slot < unnamed_dev_start)
878 unnamed_dev_start = slot;
879 spin_unlock(&unnamed_dev_lock);
881 EXPORT_SYMBOL(free_anon_bdev);
883 int set_anon_super(struct super_block *s, void *data)
885 int error = get_anon_bdev(&s->s_dev);
886 if (!error)
887 s->s_bdi = &noop_backing_dev_info;
888 return error;
891 EXPORT_SYMBOL(set_anon_super);
893 void kill_anon_super(struct super_block *sb)
895 dev_t dev = sb->s_dev;
896 generic_shutdown_super(sb);
897 free_anon_bdev(dev);
900 EXPORT_SYMBOL(kill_anon_super);
902 void kill_litter_super(struct super_block *sb)
904 if (sb->s_root)
905 d_genocide(sb->s_root);
906 kill_anon_super(sb);
909 EXPORT_SYMBOL(kill_litter_super);
911 static int ns_test_super(struct super_block *sb, void *data)
913 return sb->s_fs_info == data;
916 static int ns_set_super(struct super_block *sb, void *data)
918 sb->s_fs_info = data;
919 return set_anon_super(sb, NULL);
922 struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
923 void *data, int (*fill_super)(struct super_block *, void *, int))
925 struct super_block *sb;
927 sb = sget(fs_type, ns_test_super, ns_set_super, data);
928 if (IS_ERR(sb))
929 return ERR_CAST(sb);
931 if (!sb->s_root) {
932 int err;
933 sb->s_flags = flags;
934 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
935 if (err) {
936 deactivate_locked_super(sb);
937 return ERR_PTR(err);
940 sb->s_flags |= MS_ACTIVE;
943 return dget(sb->s_root);
946 EXPORT_SYMBOL(mount_ns);
948 #ifdef CONFIG_BLOCK
949 static int set_bdev_super(struct super_block *s, void *data)
951 s->s_bdev = data;
952 s->s_dev = s->s_bdev->bd_dev;
955 * We set the bdi here to the queue backing, file systems can
956 * overwrite this in ->fill_super()
958 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
959 return 0;
962 static int test_bdev_super(struct super_block *s, void *data)
964 return (void *)s->s_bdev == data;
967 struct dentry *mount_bdev(struct file_system_type *fs_type,
968 int flags, const char *dev_name, void *data,
969 int (*fill_super)(struct super_block *, void *, int))
971 struct block_device *bdev;
972 struct super_block *s;
973 fmode_t mode = FMODE_READ | FMODE_EXCL;
974 int error = 0;
976 if (!(flags & MS_RDONLY))
977 mode |= FMODE_WRITE;
979 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
980 if (IS_ERR(bdev))
981 return ERR_CAST(bdev);
984 * once the super is inserted into the list by sget, s_umount
985 * will protect the lockfs code from trying to start a snapshot
986 * while we are mounting
988 mutex_lock(&bdev->bd_fsfreeze_mutex);
989 if (bdev->bd_fsfreeze_count > 0) {
990 mutex_unlock(&bdev->bd_fsfreeze_mutex);
991 error = -EBUSY;
992 goto error_bdev;
994 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
995 mutex_unlock(&bdev->bd_fsfreeze_mutex);
996 if (IS_ERR(s))
997 goto error_s;
999 if (s->s_root) {
1000 if ((flags ^ s->s_flags) & MS_RDONLY) {
1001 deactivate_locked_super(s);
1002 error = -EBUSY;
1003 goto error_bdev;
1007 * s_umount nests inside bd_mutex during
1008 * __invalidate_device(). blkdev_put() acquires
1009 * bd_mutex and can't be called under s_umount. Drop
1010 * s_umount temporarily. This is safe as we're
1011 * holding an active reference.
1013 up_write(&s->s_umount);
1014 blkdev_put(bdev, mode);
1015 down_write(&s->s_umount);
1016 } else {
1017 char b[BDEVNAME_SIZE];
1019 s->s_flags = flags | MS_NOSEC;
1020 s->s_mode = mode;
1021 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1022 sb_set_blocksize(s, block_size(bdev));
1023 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1024 if (error) {
1025 deactivate_locked_super(s);
1026 goto error;
1029 s->s_flags |= MS_ACTIVE;
1030 bdev->bd_super = s;
1033 return dget(s->s_root);
1035 error_s:
1036 error = PTR_ERR(s);
1037 error_bdev:
1038 blkdev_put(bdev, mode);
1039 error:
1040 return ERR_PTR(error);
1042 EXPORT_SYMBOL(mount_bdev);
1044 void kill_block_super(struct super_block *sb)
1046 struct block_device *bdev = sb->s_bdev;
1047 fmode_t mode = sb->s_mode;
1049 bdev->bd_super = NULL;
1050 generic_shutdown_super(sb);
1051 sync_blockdev(bdev);
1052 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1053 blkdev_put(bdev, mode | FMODE_EXCL);
1056 EXPORT_SYMBOL(kill_block_super);
1057 #endif
1059 struct dentry *mount_nodev(struct file_system_type *fs_type,
1060 int flags, void *data,
1061 int (*fill_super)(struct super_block *, void *, int))
1063 int error;
1064 struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
1066 if (IS_ERR(s))
1067 return ERR_CAST(s);
1069 s->s_flags = flags;
1071 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1072 if (error) {
1073 deactivate_locked_super(s);
1074 return ERR_PTR(error);
1076 s->s_flags |= MS_ACTIVE;
1077 return dget(s->s_root);
1079 EXPORT_SYMBOL(mount_nodev);
1081 static int compare_single(struct super_block *s, void *p)
1083 return 1;
1086 struct dentry *mount_single(struct file_system_type *fs_type,
1087 int flags, void *data,
1088 int (*fill_super)(struct super_block *, void *, int))
1090 struct super_block *s;
1091 int error;
1093 s = sget(fs_type, compare_single, set_anon_super, NULL);
1094 if (IS_ERR(s))
1095 return ERR_CAST(s);
1096 if (!s->s_root) {
1097 s->s_flags = flags;
1098 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1099 if (error) {
1100 deactivate_locked_super(s);
1101 return ERR_PTR(error);
1103 s->s_flags |= MS_ACTIVE;
1104 } else {
1105 do_remount_sb(s, flags, data, 0);
1107 return dget(s->s_root);
1109 EXPORT_SYMBOL(mount_single);
1111 struct dentry *
1112 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1114 struct dentry *root;
1115 struct super_block *sb;
1116 char *secdata = NULL;
1117 int error = -ENOMEM;
1119 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1120 secdata = alloc_secdata();
1121 if (!secdata)
1122 goto out;
1124 error = security_sb_copy_data(data, secdata);
1125 if (error)
1126 goto out_free_secdata;
1129 root = type->mount(type, flags, name, data);
1130 if (IS_ERR(root)) {
1131 error = PTR_ERR(root);
1132 goto out_free_secdata;
1134 sb = root->d_sb;
1135 BUG_ON(!sb);
1136 WARN_ON(!sb->s_bdi);
1137 WARN_ON(sb->s_bdi == &default_backing_dev_info);
1138 sb->s_flags |= MS_BORN;
1140 error = security_sb_kern_mount(sb, flags, secdata);
1141 if (error)
1142 goto out_sb;
1145 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1146 * but s_maxbytes was an unsigned long long for many releases. Throw
1147 * this warning for a little while to try and catch filesystems that
1148 * violate this rule.
1150 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1151 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1153 up_write(&sb->s_umount);
1154 free_secdata(secdata);
1155 return root;
1156 out_sb:
1157 dput(root);
1158 deactivate_locked_super(sb);
1159 out_free_secdata:
1160 free_secdata(secdata);
1161 out:
1162 return ERR_PTR(error);
1166 * freeze_super - lock the filesystem and force it into a consistent state
1167 * @sb: the super to lock
1169 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1170 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1171 * -EBUSY.
1173 int freeze_super(struct super_block *sb)
1175 int ret;
1177 atomic_inc(&sb->s_active);
1178 down_write(&sb->s_umount);
1179 if (sb->s_frozen) {
1180 deactivate_locked_super(sb);
1181 return -EBUSY;
1184 if (!(sb->s_flags & MS_BORN)) {
1185 up_write(&sb->s_umount);
1186 return 0; /* sic - it's "nothing to do" */
1189 if (sb->s_flags & MS_RDONLY) {
1190 sb->s_frozen = SB_FREEZE_TRANS;
1191 smp_wmb();
1192 up_write(&sb->s_umount);
1193 return 0;
1196 sb->s_frozen = SB_FREEZE_WRITE;
1197 smp_wmb();
1199 sync_filesystem(sb);
1201 sb->s_frozen = SB_FREEZE_TRANS;
1202 smp_wmb();
1204 sync_blockdev(sb->s_bdev);
1205 if (sb->s_op->freeze_fs) {
1206 ret = sb->s_op->freeze_fs(sb);
1207 if (ret) {
1208 printk(KERN_ERR
1209 "VFS:Filesystem freeze failed\n");
1210 sb->s_frozen = SB_UNFROZEN;
1211 smp_wmb();
1212 wake_up(&sb->s_wait_unfrozen);
1213 deactivate_locked_super(sb);
1214 return ret;
1217 up_write(&sb->s_umount);
1218 return 0;
1220 EXPORT_SYMBOL(freeze_super);
1223 * thaw_super -- unlock filesystem
1224 * @sb: the super to thaw
1226 * Unlocks the filesystem and marks it writeable again after freeze_super().
1228 int thaw_super(struct super_block *sb)
1230 int error;
1232 down_write(&sb->s_umount);
1233 if (sb->s_frozen == SB_UNFROZEN) {
1234 up_write(&sb->s_umount);
1235 return -EINVAL;
1238 if (sb->s_flags & MS_RDONLY)
1239 goto out;
1241 if (sb->s_op->unfreeze_fs) {
1242 error = sb->s_op->unfreeze_fs(sb);
1243 if (error) {
1244 printk(KERN_ERR
1245 "VFS:Filesystem thaw failed\n");
1246 sb->s_frozen = SB_FREEZE_TRANS;
1247 up_write(&sb->s_umount);
1248 return error;
1252 out:
1253 sb->s_frozen = SB_UNFROZEN;
1254 smp_wmb();
1255 wake_up(&sb->s_wait_unfrozen);
1256 deactivate_locked_super(sb);
1258 return 0;
1260 EXPORT_SYMBOL(thaw_super);