Merge back earlier 'pm-sleep' material for v3.18.
[linux/fpc-iii.git] / fs / super.c
blobb9a214d2fe98b8b37a7560ebf1a7d1dc7c5d83e0
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/export.h>
24 #include <linux/slab.h>
25 #include <linux/blkdev.h>
26 #include <linux/mount.h>
27 #include <linux/security.h>
28 #include <linux/writeback.h> /* for the emergency remount stuff */
29 #include <linux/idr.h>
30 #include <linux/mutex.h>
31 #include <linux/backing-dev.h>
32 #include <linux/rculist_bl.h>
33 #include <linux/cleancache.h>
34 #include <linux/fsnotify.h>
35 #include <linux/lockdep.h>
36 #include "internal.h"
39 LIST_HEAD(super_blocks);
40 DEFINE_SPINLOCK(sb_lock);
42 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
43 "sb_writers",
44 "sb_pagefaults",
45 "sb_internal",
49 * One thing we have to be careful of with a per-sb shrinker is that we don't
50 * drop the last active reference to the superblock from within the shrinker.
51 * If that happens we could trigger unregistering the shrinker from within the
52 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
53 * take a passive reference to the superblock to avoid this from occurring.
55 static unsigned long super_cache_scan(struct shrinker *shrink,
56 struct shrink_control *sc)
58 struct super_block *sb;
59 long fs_objects = 0;
60 long total_objects;
61 long freed = 0;
62 long dentries;
63 long inodes;
65 sb = container_of(shrink, struct super_block, s_shrink);
68 * Deadlock avoidance. We may hold various FS locks, and we don't want
69 * to recurse into the FS that called us in clear_inode() and friends..
71 if (!(sc->gfp_mask & __GFP_FS))
72 return SHRINK_STOP;
74 if (!grab_super_passive(sb))
75 return SHRINK_STOP;
77 if (sb->s_op->nr_cached_objects)
78 fs_objects = sb->s_op->nr_cached_objects(sb, sc->nid);
80 inodes = list_lru_count_node(&sb->s_inode_lru, sc->nid);
81 dentries = list_lru_count_node(&sb->s_dentry_lru, sc->nid);
82 total_objects = dentries + inodes + fs_objects + 1;
84 /* proportion the scan between the caches */
85 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
86 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
89 * prune the dcache first as the icache is pinned by it, then
90 * prune the icache, followed by the filesystem specific caches
92 freed = prune_dcache_sb(sb, dentries, sc->nid);
93 freed += prune_icache_sb(sb, inodes, sc->nid);
95 if (fs_objects) {
96 fs_objects = mult_frac(sc->nr_to_scan, fs_objects,
97 total_objects);
98 freed += sb->s_op->free_cached_objects(sb, fs_objects,
99 sc->nid);
102 drop_super(sb);
103 return freed;
106 static unsigned long super_cache_count(struct shrinker *shrink,
107 struct shrink_control *sc)
109 struct super_block *sb;
110 long total_objects = 0;
112 sb = container_of(shrink, struct super_block, s_shrink);
115 * Don't call grab_super_passive as it is a potential
116 * scalability bottleneck. The counts could get updated
117 * between super_cache_count and super_cache_scan anyway.
118 * Call to super_cache_count with shrinker_rwsem held
119 * ensures the safety of call to list_lru_count_node() and
120 * s_op->nr_cached_objects().
122 if (sb->s_op && sb->s_op->nr_cached_objects)
123 total_objects = sb->s_op->nr_cached_objects(sb,
124 sc->nid);
126 total_objects += list_lru_count_node(&sb->s_dentry_lru,
127 sc->nid);
128 total_objects += list_lru_count_node(&sb->s_inode_lru,
129 sc->nid);
131 total_objects = vfs_pressure_ratio(total_objects);
132 return total_objects;
136 * destroy_super - frees a superblock
137 * @s: superblock to free
139 * Frees a superblock.
141 static void destroy_super(struct super_block *s)
143 int i;
144 list_lru_destroy(&s->s_dentry_lru);
145 list_lru_destroy(&s->s_inode_lru);
146 for (i = 0; i < SB_FREEZE_LEVELS; i++)
147 percpu_counter_destroy(&s->s_writers.counter[i]);
148 security_sb_free(s);
149 WARN_ON(!list_empty(&s->s_mounts));
150 kfree(s->s_subtype);
151 kfree(s->s_options);
152 kfree_rcu(s, rcu);
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;
167 int i;
169 if (!s)
170 return NULL;
172 INIT_LIST_HEAD(&s->s_mounts);
174 if (security_sb_alloc(s))
175 goto fail;
177 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
178 if (percpu_counter_init(&s->s_writers.counter[i], 0) < 0)
179 goto fail;
180 lockdep_init_map(&s->s_writers.lock_map[i], sb_writers_name[i],
181 &type->s_writers_key[i], 0);
183 init_waitqueue_head(&s->s_writers.wait);
184 init_waitqueue_head(&s->s_writers.wait_unfrozen);
185 s->s_flags = flags;
186 s->s_bdi = &default_backing_dev_info;
187 INIT_HLIST_NODE(&s->s_instances);
188 INIT_HLIST_BL_HEAD(&s->s_anon);
189 INIT_LIST_HEAD(&s->s_inodes);
191 if (list_lru_init(&s->s_dentry_lru))
192 goto fail;
193 if (list_lru_init(&s->s_inode_lru))
194 goto fail;
196 init_rwsem(&s->s_umount);
197 lockdep_set_class(&s->s_umount, &type->s_umount_key);
199 * sget() can have s_umount recursion.
201 * When it cannot find a suitable sb, it allocates a new
202 * one (this one), and tries again to find a suitable old
203 * one.
205 * In case that succeeds, it will acquire the s_umount
206 * lock of the old one. Since these are clearly distrinct
207 * locks, and this object isn't exposed yet, there's no
208 * risk of deadlocks.
210 * Annotate this by putting this lock in a different
211 * subclass.
213 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
214 s->s_count = 1;
215 atomic_set(&s->s_active, 1);
216 mutex_init(&s->s_vfs_rename_mutex);
217 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
218 mutex_init(&s->s_dquot.dqio_mutex);
219 mutex_init(&s->s_dquot.dqonoff_mutex);
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.scan_objects = super_cache_scan;
227 s->s_shrink.count_objects = super_cache_count;
228 s->s_shrink.batch = 1024;
229 s->s_shrink.flags = SHRINKER_NUMA_AWARE;
230 return s;
232 fail:
233 destroy_super(s);
234 return NULL;
237 /* Superblock refcounting */
240 * Drop a superblock's refcount. The caller must hold sb_lock.
242 static void __put_super(struct super_block *sb)
244 if (!--sb->s_count) {
245 list_del_init(&sb->s_list);
246 destroy_super(sb);
251 * put_super - drop a temporary reference to superblock
252 * @sb: superblock in question
254 * Drops a temporary reference, frees superblock if there's no
255 * references left.
257 static void put_super(struct super_block *sb)
259 spin_lock(&sb_lock);
260 __put_super(sb);
261 spin_unlock(&sb_lock);
266 * deactivate_locked_super - drop an active reference to superblock
267 * @s: superblock to deactivate
269 * Drops an active reference to superblock, converting it into a temprory
270 * one if there is no other active references left. In that case we
271 * tell fs driver to shut it down and drop the temporary reference we
272 * had just acquired.
274 * Caller holds exclusive lock on superblock; that lock is released.
276 void deactivate_locked_super(struct super_block *s)
278 struct file_system_type *fs = s->s_type;
279 if (atomic_dec_and_test(&s->s_active)) {
280 cleancache_invalidate_fs(s);
281 unregister_shrinker(&s->s_shrink);
282 fs->kill_sb(s);
284 put_filesystem(fs);
285 put_super(s);
286 } else {
287 up_write(&s->s_umount);
291 EXPORT_SYMBOL(deactivate_locked_super);
294 * deactivate_super - drop an active reference to superblock
295 * @s: superblock to deactivate
297 * Variant of deactivate_locked_super(), except that superblock is *not*
298 * locked by caller. If we are going to drop the final active reference,
299 * lock will be acquired prior to that.
301 void deactivate_super(struct super_block *s)
303 if (!atomic_add_unless(&s->s_active, -1, 1)) {
304 down_write(&s->s_umount);
305 deactivate_locked_super(s);
309 EXPORT_SYMBOL(deactivate_super);
312 * grab_super - acquire an active reference
313 * @s: reference we are trying to make active
315 * Tries to acquire an active reference. grab_super() is used when we
316 * had just found a superblock in super_blocks or fs_type->fs_supers
317 * and want to turn it into a full-blown active reference. grab_super()
318 * is called with sb_lock held and drops it. Returns 1 in case of
319 * success, 0 if we had failed (superblock contents was already dead or
320 * dying when grab_super() had been called). Note that this is only
321 * called for superblocks not in rundown mode (== ones still on ->fs_supers
322 * of their type), so increment of ->s_count is OK here.
324 static int grab_super(struct super_block *s) __releases(sb_lock)
326 s->s_count++;
327 spin_unlock(&sb_lock);
328 down_write(&s->s_umount);
329 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
330 put_super(s);
331 return 1;
333 up_write(&s->s_umount);
334 put_super(s);
335 return 0;
339 * grab_super_passive - acquire a passive reference
340 * @sb: reference we are trying to grab
342 * Tries to acquire a passive reference. This is used in places where we
343 * cannot take an active reference but we need to ensure that the
344 * superblock does not go away while we are working on it. It returns
345 * false if a reference was not gained, and returns true with the s_umount
346 * lock held in read mode if a reference is gained. On successful return,
347 * the caller must drop the s_umount lock and the passive reference when
348 * done.
350 bool grab_super_passive(struct super_block *sb)
352 spin_lock(&sb_lock);
353 if (hlist_unhashed(&sb->s_instances)) {
354 spin_unlock(&sb_lock);
355 return false;
358 sb->s_count++;
359 spin_unlock(&sb_lock);
361 if (down_read_trylock(&sb->s_umount)) {
362 if (sb->s_root && (sb->s_flags & MS_BORN))
363 return true;
364 up_read(&sb->s_umount);
367 put_super(sb);
368 return false;
372 * generic_shutdown_super - common helper for ->kill_sb()
373 * @sb: superblock to kill
375 * generic_shutdown_super() does all fs-independent work on superblock
376 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
377 * that need destruction out of superblock, call generic_shutdown_super()
378 * and release aforementioned objects. Note: dentries and inodes _are_
379 * taken care of and do not need specific handling.
381 * Upon calling this function, the filesystem may no longer alter or
382 * rearrange the set of dentries belonging to this super_block, nor may it
383 * change the attachments of dentries to inodes.
385 void generic_shutdown_super(struct super_block *sb)
387 const struct super_operations *sop = sb->s_op;
389 if (sb->s_root) {
390 shrink_dcache_for_umount(sb);
391 sync_filesystem(sb);
392 sb->s_flags &= ~MS_ACTIVE;
394 fsnotify_unmount_inodes(&sb->s_inodes);
396 evict_inodes(sb);
398 if (sb->s_dio_done_wq) {
399 destroy_workqueue(sb->s_dio_done_wq);
400 sb->s_dio_done_wq = NULL;
403 if (sop->put_super)
404 sop->put_super(sb);
406 if (!list_empty(&sb->s_inodes)) {
407 printk("VFS: Busy inodes after unmount of %s. "
408 "Self-destruct in 5 seconds. Have a nice day...\n",
409 sb->s_id);
412 spin_lock(&sb_lock);
413 /* should be initialized for __put_super_and_need_restart() */
414 hlist_del_init(&sb->s_instances);
415 spin_unlock(&sb_lock);
416 up_write(&sb->s_umount);
419 EXPORT_SYMBOL(generic_shutdown_super);
422 * sget - find or create a superblock
423 * @type: filesystem type superblock should belong to
424 * @test: comparison callback
425 * @set: setup callback
426 * @flags: mount flags
427 * @data: argument to each of them
429 struct super_block *sget(struct file_system_type *type,
430 int (*test)(struct super_block *,void *),
431 int (*set)(struct super_block *,void *),
432 int flags,
433 void *data)
435 struct super_block *s = NULL;
436 struct super_block *old;
437 int err;
439 retry:
440 spin_lock(&sb_lock);
441 if (test) {
442 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
443 if (!test(old, data))
444 continue;
445 if (!grab_super(old))
446 goto retry;
447 if (s) {
448 up_write(&s->s_umount);
449 destroy_super(s);
450 s = NULL;
452 return old;
455 if (!s) {
456 spin_unlock(&sb_lock);
457 s = alloc_super(type, flags);
458 if (!s)
459 return ERR_PTR(-ENOMEM);
460 goto retry;
463 err = set(s, data);
464 if (err) {
465 spin_unlock(&sb_lock);
466 up_write(&s->s_umount);
467 destroy_super(s);
468 return ERR_PTR(err);
470 s->s_type = type;
471 strlcpy(s->s_id, type->name, sizeof(s->s_id));
472 list_add_tail(&s->s_list, &super_blocks);
473 hlist_add_head(&s->s_instances, &type->fs_supers);
474 spin_unlock(&sb_lock);
475 get_filesystem(type);
476 register_shrinker(&s->s_shrink);
477 return s;
480 EXPORT_SYMBOL(sget);
482 void drop_super(struct super_block *sb)
484 up_read(&sb->s_umount);
485 put_super(sb);
488 EXPORT_SYMBOL(drop_super);
491 * iterate_supers - call function for all active superblocks
492 * @f: function to call
493 * @arg: argument to pass to it
495 * Scans the superblock list and calls given function, passing it
496 * locked superblock and given argument.
498 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
500 struct super_block *sb, *p = NULL;
502 spin_lock(&sb_lock);
503 list_for_each_entry(sb, &super_blocks, s_list) {
504 if (hlist_unhashed(&sb->s_instances))
505 continue;
506 sb->s_count++;
507 spin_unlock(&sb_lock);
509 down_read(&sb->s_umount);
510 if (sb->s_root && (sb->s_flags & MS_BORN))
511 f(sb, arg);
512 up_read(&sb->s_umount);
514 spin_lock(&sb_lock);
515 if (p)
516 __put_super(p);
517 p = sb;
519 if (p)
520 __put_super(p);
521 spin_unlock(&sb_lock);
525 * iterate_supers_type - call function for superblocks of given type
526 * @type: fs type
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_type(struct file_system_type *type,
534 void (*f)(struct super_block *, void *), void *arg)
536 struct super_block *sb, *p = NULL;
538 spin_lock(&sb_lock);
539 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
540 sb->s_count++;
541 spin_unlock(&sb_lock);
543 down_read(&sb->s_umount);
544 if (sb->s_root && (sb->s_flags & MS_BORN))
545 f(sb, arg);
546 up_read(&sb->s_umount);
548 spin_lock(&sb_lock);
549 if (p)
550 __put_super(p);
551 p = sb;
553 if (p)
554 __put_super(p);
555 spin_unlock(&sb_lock);
558 EXPORT_SYMBOL(iterate_supers_type);
561 * get_super - get the superblock of a device
562 * @bdev: device to get the superblock for
564 * Scans the superblock list and finds the superblock of the file system
565 * mounted on the device given. %NULL is returned if no match is found.
568 struct super_block *get_super(struct block_device *bdev)
570 struct super_block *sb;
572 if (!bdev)
573 return NULL;
575 spin_lock(&sb_lock);
576 rescan:
577 list_for_each_entry(sb, &super_blocks, s_list) {
578 if (hlist_unhashed(&sb->s_instances))
579 continue;
580 if (sb->s_bdev == bdev) {
581 sb->s_count++;
582 spin_unlock(&sb_lock);
583 down_read(&sb->s_umount);
584 /* still alive? */
585 if (sb->s_root && (sb->s_flags & MS_BORN))
586 return sb;
587 up_read(&sb->s_umount);
588 /* nope, got unmounted */
589 spin_lock(&sb_lock);
590 __put_super(sb);
591 goto rescan;
594 spin_unlock(&sb_lock);
595 return NULL;
598 EXPORT_SYMBOL(get_super);
601 * get_super_thawed - get thawed superblock of a device
602 * @bdev: device to get the superblock for
604 * Scans the superblock list and finds the superblock of the file system
605 * mounted on the device. The superblock is returned once it is thawed
606 * (or immediately if it was not frozen). %NULL is returned if no match
607 * is found.
609 struct super_block *get_super_thawed(struct block_device *bdev)
611 while (1) {
612 struct super_block *s = get_super(bdev);
613 if (!s || s->s_writers.frozen == SB_UNFROZEN)
614 return s;
615 up_read(&s->s_umount);
616 wait_event(s->s_writers.wait_unfrozen,
617 s->s_writers.frozen == SB_UNFROZEN);
618 put_super(s);
621 EXPORT_SYMBOL(get_super_thawed);
624 * get_active_super - get an active reference to the superblock of a device
625 * @bdev: device to get the superblock for
627 * Scans the superblock list and finds the superblock of the file system
628 * mounted on the device given. Returns the superblock with an active
629 * reference or %NULL if none was found.
631 struct super_block *get_active_super(struct block_device *bdev)
633 struct super_block *sb;
635 if (!bdev)
636 return NULL;
638 restart:
639 spin_lock(&sb_lock);
640 list_for_each_entry(sb, &super_blocks, s_list) {
641 if (hlist_unhashed(&sb->s_instances))
642 continue;
643 if (sb->s_bdev == bdev) {
644 if (!grab_super(sb))
645 goto restart;
646 up_write(&sb->s_umount);
647 return sb;
650 spin_unlock(&sb_lock);
651 return NULL;
654 struct super_block *user_get_super(dev_t dev)
656 struct super_block *sb;
658 spin_lock(&sb_lock);
659 rescan:
660 list_for_each_entry(sb, &super_blocks, s_list) {
661 if (hlist_unhashed(&sb->s_instances))
662 continue;
663 if (sb->s_dev == dev) {
664 sb->s_count++;
665 spin_unlock(&sb_lock);
666 down_read(&sb->s_umount);
667 /* still alive? */
668 if (sb->s_root && (sb->s_flags & MS_BORN))
669 return sb;
670 up_read(&sb->s_umount);
671 /* nope, got unmounted */
672 spin_lock(&sb_lock);
673 __put_super(sb);
674 goto rescan;
677 spin_unlock(&sb_lock);
678 return NULL;
682 * do_remount_sb - asks filesystem to change mount options.
683 * @sb: superblock in question
684 * @flags: numeric part of options
685 * @data: the rest of options
686 * @force: whether or not to force the change
688 * Alters the mount options of a mounted file system.
690 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
692 int retval;
693 int remount_ro;
695 if (sb->s_writers.frozen != SB_UNFROZEN)
696 return -EBUSY;
698 #ifdef CONFIG_BLOCK
699 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
700 return -EACCES;
701 #endif
703 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
705 if (remount_ro) {
706 if (sb->s_pins.first) {
707 up_write(&sb->s_umount);
708 sb_pin_kill(sb);
709 down_write(&sb->s_umount);
710 if (!sb->s_root)
711 return 0;
712 if (sb->s_writers.frozen != SB_UNFROZEN)
713 return -EBUSY;
714 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
717 shrink_dcache_sb(sb);
719 /* If we are remounting RDONLY and current sb is read/write,
720 make sure there are no rw files opened */
721 if (remount_ro) {
722 if (force) {
723 sb->s_readonly_remount = 1;
724 smp_wmb();
725 } else {
726 retval = sb_prepare_remount_readonly(sb);
727 if (retval)
728 return retval;
732 if (sb->s_op->remount_fs) {
733 retval = sb->s_op->remount_fs(sb, &flags, data);
734 if (retval) {
735 if (!force)
736 goto cancel_readonly;
737 /* If forced remount, go ahead despite any errors */
738 WARN(1, "forced remount of a %s fs returned %i\n",
739 sb->s_type->name, retval);
742 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
743 /* Needs to be ordered wrt mnt_is_readonly() */
744 smp_wmb();
745 sb->s_readonly_remount = 0;
748 * Some filesystems modify their metadata via some other path than the
749 * bdev buffer cache (eg. use a private mapping, or directories in
750 * pagecache, etc). Also file data modifications go via their own
751 * mappings. So If we try to mount readonly then copy the filesystem
752 * from bdev, we could get stale data, so invalidate it to give a best
753 * effort at coherency.
755 if (remount_ro && sb->s_bdev)
756 invalidate_bdev(sb->s_bdev);
757 return 0;
759 cancel_readonly:
760 sb->s_readonly_remount = 0;
761 return retval;
764 static void do_emergency_remount(struct work_struct *work)
766 struct super_block *sb, *p = NULL;
768 spin_lock(&sb_lock);
769 list_for_each_entry(sb, &super_blocks, s_list) {
770 if (hlist_unhashed(&sb->s_instances))
771 continue;
772 sb->s_count++;
773 spin_unlock(&sb_lock);
774 down_write(&sb->s_umount);
775 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
776 !(sb->s_flags & MS_RDONLY)) {
778 * What lock protects sb->s_flags??
780 do_remount_sb(sb, MS_RDONLY, NULL, 1);
782 up_write(&sb->s_umount);
783 spin_lock(&sb_lock);
784 if (p)
785 __put_super(p);
786 p = sb;
788 if (p)
789 __put_super(p);
790 spin_unlock(&sb_lock);
791 kfree(work);
792 printk("Emergency Remount complete\n");
795 void emergency_remount(void)
797 struct work_struct *work;
799 work = kmalloc(sizeof(*work), GFP_ATOMIC);
800 if (work) {
801 INIT_WORK(work, do_emergency_remount);
802 schedule_work(work);
807 * Unnamed block devices are dummy devices used by virtual
808 * filesystems which don't use real block-devices. -- jrs
811 static DEFINE_IDA(unnamed_dev_ida);
812 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
813 /* Many userspace utilities consider an FSID of 0 invalid.
814 * Always return at least 1 from get_anon_bdev.
816 static int unnamed_dev_start = 1;
818 int get_anon_bdev(dev_t *p)
820 int dev;
821 int error;
823 retry:
824 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
825 return -ENOMEM;
826 spin_lock(&unnamed_dev_lock);
827 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
828 if (!error)
829 unnamed_dev_start = dev + 1;
830 spin_unlock(&unnamed_dev_lock);
831 if (error == -EAGAIN)
832 /* We raced and lost with another CPU. */
833 goto retry;
834 else if (error)
835 return -EAGAIN;
837 if (dev == (1 << MINORBITS)) {
838 spin_lock(&unnamed_dev_lock);
839 ida_remove(&unnamed_dev_ida, dev);
840 if (unnamed_dev_start > dev)
841 unnamed_dev_start = dev;
842 spin_unlock(&unnamed_dev_lock);
843 return -EMFILE;
845 *p = MKDEV(0, dev & MINORMASK);
846 return 0;
848 EXPORT_SYMBOL(get_anon_bdev);
850 void free_anon_bdev(dev_t dev)
852 int slot = MINOR(dev);
853 spin_lock(&unnamed_dev_lock);
854 ida_remove(&unnamed_dev_ida, slot);
855 if (slot < unnamed_dev_start)
856 unnamed_dev_start = slot;
857 spin_unlock(&unnamed_dev_lock);
859 EXPORT_SYMBOL(free_anon_bdev);
861 int set_anon_super(struct super_block *s, void *data)
863 int error = get_anon_bdev(&s->s_dev);
864 if (!error)
865 s->s_bdi = &noop_backing_dev_info;
866 return error;
869 EXPORT_SYMBOL(set_anon_super);
871 void kill_anon_super(struct super_block *sb)
873 dev_t dev = sb->s_dev;
874 generic_shutdown_super(sb);
875 free_anon_bdev(dev);
878 EXPORT_SYMBOL(kill_anon_super);
880 void kill_litter_super(struct super_block *sb)
882 if (sb->s_root)
883 d_genocide(sb->s_root);
884 kill_anon_super(sb);
887 EXPORT_SYMBOL(kill_litter_super);
889 static int ns_test_super(struct super_block *sb, void *data)
891 return sb->s_fs_info == data;
894 static int ns_set_super(struct super_block *sb, void *data)
896 sb->s_fs_info = data;
897 return set_anon_super(sb, NULL);
900 struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
901 void *data, int (*fill_super)(struct super_block *, void *, int))
903 struct super_block *sb;
905 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
906 if (IS_ERR(sb))
907 return ERR_CAST(sb);
909 if (!sb->s_root) {
910 int err;
911 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
912 if (err) {
913 deactivate_locked_super(sb);
914 return ERR_PTR(err);
917 sb->s_flags |= MS_ACTIVE;
920 return dget(sb->s_root);
923 EXPORT_SYMBOL(mount_ns);
925 #ifdef CONFIG_BLOCK
926 static int set_bdev_super(struct super_block *s, void *data)
928 s->s_bdev = data;
929 s->s_dev = s->s_bdev->bd_dev;
932 * We set the bdi here to the queue backing, file systems can
933 * overwrite this in ->fill_super()
935 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
936 return 0;
939 static int test_bdev_super(struct super_block *s, void *data)
941 return (void *)s->s_bdev == data;
944 struct dentry *mount_bdev(struct file_system_type *fs_type,
945 int flags, const char *dev_name, void *data,
946 int (*fill_super)(struct super_block *, void *, int))
948 struct block_device *bdev;
949 struct super_block *s;
950 fmode_t mode = FMODE_READ | FMODE_EXCL;
951 int error = 0;
953 if (!(flags & MS_RDONLY))
954 mode |= FMODE_WRITE;
956 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
957 if (IS_ERR(bdev))
958 return ERR_CAST(bdev);
961 * once the super is inserted into the list by sget, s_umount
962 * will protect the lockfs code from trying to start a snapshot
963 * while we are mounting
965 mutex_lock(&bdev->bd_fsfreeze_mutex);
966 if (bdev->bd_fsfreeze_count > 0) {
967 mutex_unlock(&bdev->bd_fsfreeze_mutex);
968 error = -EBUSY;
969 goto error_bdev;
971 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
972 bdev);
973 mutex_unlock(&bdev->bd_fsfreeze_mutex);
974 if (IS_ERR(s))
975 goto error_s;
977 if (s->s_root) {
978 if ((flags ^ s->s_flags) & MS_RDONLY) {
979 deactivate_locked_super(s);
980 error = -EBUSY;
981 goto error_bdev;
985 * s_umount nests inside bd_mutex during
986 * __invalidate_device(). blkdev_put() acquires
987 * bd_mutex and can't be called under s_umount. Drop
988 * s_umount temporarily. This is safe as we're
989 * holding an active reference.
991 up_write(&s->s_umount);
992 blkdev_put(bdev, mode);
993 down_write(&s->s_umount);
994 } else {
995 char b[BDEVNAME_SIZE];
997 s->s_mode = mode;
998 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
999 sb_set_blocksize(s, block_size(bdev));
1000 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1001 if (error) {
1002 deactivate_locked_super(s);
1003 goto error;
1006 s->s_flags |= MS_ACTIVE;
1007 bdev->bd_super = s;
1010 return dget(s->s_root);
1012 error_s:
1013 error = PTR_ERR(s);
1014 error_bdev:
1015 blkdev_put(bdev, mode);
1016 error:
1017 return ERR_PTR(error);
1019 EXPORT_SYMBOL(mount_bdev);
1021 void kill_block_super(struct super_block *sb)
1023 struct block_device *bdev = sb->s_bdev;
1024 fmode_t mode = sb->s_mode;
1026 bdev->bd_super = NULL;
1027 generic_shutdown_super(sb);
1028 sync_blockdev(bdev);
1029 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1030 blkdev_put(bdev, mode | FMODE_EXCL);
1033 EXPORT_SYMBOL(kill_block_super);
1034 #endif
1036 struct dentry *mount_nodev(struct file_system_type *fs_type,
1037 int flags, void *data,
1038 int (*fill_super)(struct super_block *, void *, int))
1040 int error;
1041 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1043 if (IS_ERR(s))
1044 return ERR_CAST(s);
1046 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1047 if (error) {
1048 deactivate_locked_super(s);
1049 return ERR_PTR(error);
1051 s->s_flags |= MS_ACTIVE;
1052 return dget(s->s_root);
1054 EXPORT_SYMBOL(mount_nodev);
1056 static int compare_single(struct super_block *s, void *p)
1058 return 1;
1061 struct dentry *mount_single(struct file_system_type *fs_type,
1062 int flags, void *data,
1063 int (*fill_super)(struct super_block *, void *, int))
1065 struct super_block *s;
1066 int error;
1068 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1069 if (IS_ERR(s))
1070 return ERR_CAST(s);
1071 if (!s->s_root) {
1072 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1073 if (error) {
1074 deactivate_locked_super(s);
1075 return ERR_PTR(error);
1077 s->s_flags |= MS_ACTIVE;
1078 } else {
1079 do_remount_sb(s, flags, data, 0);
1081 return dget(s->s_root);
1083 EXPORT_SYMBOL(mount_single);
1085 struct dentry *
1086 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1088 struct dentry *root;
1089 struct super_block *sb;
1090 char *secdata = NULL;
1091 int error = -ENOMEM;
1093 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1094 secdata = alloc_secdata();
1095 if (!secdata)
1096 goto out;
1098 error = security_sb_copy_data(data, secdata);
1099 if (error)
1100 goto out_free_secdata;
1103 root = type->mount(type, flags, name, data);
1104 if (IS_ERR(root)) {
1105 error = PTR_ERR(root);
1106 goto out_free_secdata;
1108 sb = root->d_sb;
1109 BUG_ON(!sb);
1110 WARN_ON(!sb->s_bdi);
1111 WARN_ON(sb->s_bdi == &default_backing_dev_info);
1112 sb->s_flags |= MS_BORN;
1114 error = security_sb_kern_mount(sb, flags, secdata);
1115 if (error)
1116 goto out_sb;
1119 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1120 * but s_maxbytes was an unsigned long long for many releases. Throw
1121 * this warning for a little while to try and catch filesystems that
1122 * violate this rule.
1124 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1125 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1127 up_write(&sb->s_umount);
1128 free_secdata(secdata);
1129 return root;
1130 out_sb:
1131 dput(root);
1132 deactivate_locked_super(sb);
1133 out_free_secdata:
1134 free_secdata(secdata);
1135 out:
1136 return ERR_PTR(error);
1140 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1141 * instead.
1143 void __sb_end_write(struct super_block *sb, int level)
1145 percpu_counter_dec(&sb->s_writers.counter[level-1]);
1147 * Make sure s_writers are updated before we wake up waiters in
1148 * freeze_super().
1150 smp_mb();
1151 if (waitqueue_active(&sb->s_writers.wait))
1152 wake_up(&sb->s_writers.wait);
1153 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _RET_IP_);
1155 EXPORT_SYMBOL(__sb_end_write);
1157 #ifdef CONFIG_LOCKDEP
1159 * We want lockdep to tell us about possible deadlocks with freezing but
1160 * it's it bit tricky to properly instrument it. Getting a freeze protection
1161 * works as getting a read lock but there are subtle problems. XFS for example
1162 * gets freeze protection on internal level twice in some cases, which is OK
1163 * only because we already hold a freeze protection also on higher level. Due
1164 * to these cases we have to tell lockdep we are doing trylock when we
1165 * already hold a freeze protection for a higher freeze level.
1167 static void acquire_freeze_lock(struct super_block *sb, int level, bool trylock,
1168 unsigned long ip)
1170 int i;
1172 if (!trylock) {
1173 for (i = 0; i < level - 1; i++)
1174 if (lock_is_held(&sb->s_writers.lock_map[i])) {
1175 trylock = true;
1176 break;
1179 rwsem_acquire_read(&sb->s_writers.lock_map[level-1], 0, trylock, ip);
1181 #endif
1184 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1185 * instead.
1187 int __sb_start_write(struct super_block *sb, int level, bool wait)
1189 retry:
1190 if (unlikely(sb->s_writers.frozen >= level)) {
1191 if (!wait)
1192 return 0;
1193 wait_event(sb->s_writers.wait_unfrozen,
1194 sb->s_writers.frozen < level);
1197 #ifdef CONFIG_LOCKDEP
1198 acquire_freeze_lock(sb, level, !wait, _RET_IP_);
1199 #endif
1200 percpu_counter_inc(&sb->s_writers.counter[level-1]);
1202 * Make sure counter is updated before we check for frozen.
1203 * freeze_super() first sets frozen and then checks the counter.
1205 smp_mb();
1206 if (unlikely(sb->s_writers.frozen >= level)) {
1207 __sb_end_write(sb, level);
1208 goto retry;
1210 return 1;
1212 EXPORT_SYMBOL(__sb_start_write);
1215 * sb_wait_write - wait until all writers to given file system finish
1216 * @sb: the super for which we wait
1217 * @level: type of writers we wait for (normal vs page fault)
1219 * This function waits until there are no writers of given type to given file
1220 * system. Caller of this function should make sure there can be no new writers
1221 * of type @level before calling this function. Otherwise this function can
1222 * livelock.
1224 static void sb_wait_write(struct super_block *sb, int level)
1226 s64 writers;
1229 * We just cycle-through lockdep here so that it does not complain
1230 * about returning with lock to userspace
1232 rwsem_acquire(&sb->s_writers.lock_map[level-1], 0, 0, _THIS_IP_);
1233 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _THIS_IP_);
1235 do {
1236 DEFINE_WAIT(wait);
1239 * We use a barrier in prepare_to_wait() to separate setting
1240 * of frozen and checking of the counter
1242 prepare_to_wait(&sb->s_writers.wait, &wait,
1243 TASK_UNINTERRUPTIBLE);
1245 writers = percpu_counter_sum(&sb->s_writers.counter[level-1]);
1246 if (writers)
1247 schedule();
1249 finish_wait(&sb->s_writers.wait, &wait);
1250 } while (writers);
1254 * freeze_super - lock the filesystem and force it into a consistent state
1255 * @sb: the super to lock
1257 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1258 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1259 * -EBUSY.
1261 * During this function, sb->s_writers.frozen goes through these values:
1263 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1265 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1266 * writes should be blocked, though page faults are still allowed. We wait for
1267 * all writes to complete and then proceed to the next stage.
1269 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1270 * but internal fs threads can still modify the filesystem (although they
1271 * should not dirty new pages or inodes), writeback can run etc. After waiting
1272 * for all running page faults we sync the filesystem which will clean all
1273 * dirty pages and inodes (no new dirty pages or inodes can be created when
1274 * sync is running).
1276 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1277 * modification are blocked (e.g. XFS preallocation truncation on inode
1278 * reclaim). This is usually implemented by blocking new transactions for
1279 * filesystems that have them and need this additional guard. After all
1280 * internal writers are finished we call ->freeze_fs() to finish filesystem
1281 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1282 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1284 * sb->s_writers.frozen is protected by sb->s_umount.
1286 int freeze_super(struct super_block *sb)
1288 int ret;
1290 atomic_inc(&sb->s_active);
1291 down_write(&sb->s_umount);
1292 if (sb->s_writers.frozen != SB_UNFROZEN) {
1293 deactivate_locked_super(sb);
1294 return -EBUSY;
1297 if (!(sb->s_flags & MS_BORN)) {
1298 up_write(&sb->s_umount);
1299 return 0; /* sic - it's "nothing to do" */
1302 if (sb->s_flags & MS_RDONLY) {
1303 /* Nothing to do really... */
1304 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1305 up_write(&sb->s_umount);
1306 return 0;
1309 /* From now on, no new normal writers can start */
1310 sb->s_writers.frozen = SB_FREEZE_WRITE;
1311 smp_wmb();
1313 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1314 up_write(&sb->s_umount);
1316 sb_wait_write(sb, SB_FREEZE_WRITE);
1318 /* Now we go and block page faults... */
1319 down_write(&sb->s_umount);
1320 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1321 smp_wmb();
1323 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1325 /* All writers are done so after syncing there won't be dirty data */
1326 sync_filesystem(sb);
1328 /* Now wait for internal filesystem counter */
1329 sb->s_writers.frozen = SB_FREEZE_FS;
1330 smp_wmb();
1331 sb_wait_write(sb, SB_FREEZE_FS);
1333 if (sb->s_op->freeze_fs) {
1334 ret = sb->s_op->freeze_fs(sb);
1335 if (ret) {
1336 printk(KERN_ERR
1337 "VFS:Filesystem freeze failed\n");
1338 sb->s_writers.frozen = SB_UNFROZEN;
1339 smp_wmb();
1340 wake_up(&sb->s_writers.wait_unfrozen);
1341 deactivate_locked_super(sb);
1342 return ret;
1346 * This is just for debugging purposes so that fs can warn if it
1347 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1349 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1350 up_write(&sb->s_umount);
1351 return 0;
1353 EXPORT_SYMBOL(freeze_super);
1356 * thaw_super -- unlock filesystem
1357 * @sb: the super to thaw
1359 * Unlocks the filesystem and marks it writeable again after freeze_super().
1361 int thaw_super(struct super_block *sb)
1363 int error;
1365 down_write(&sb->s_umount);
1366 if (sb->s_writers.frozen == SB_UNFROZEN) {
1367 up_write(&sb->s_umount);
1368 return -EINVAL;
1371 if (sb->s_flags & MS_RDONLY)
1372 goto out;
1374 if (sb->s_op->unfreeze_fs) {
1375 error = sb->s_op->unfreeze_fs(sb);
1376 if (error) {
1377 printk(KERN_ERR
1378 "VFS:Filesystem thaw failed\n");
1379 up_write(&sb->s_umount);
1380 return error;
1384 out:
1385 sb->s_writers.frozen = SB_UNFROZEN;
1386 smp_wmb();
1387 wake_up(&sb->s_writers.wait_unfrozen);
1388 deactivate_locked_super(sb);
1390 return 0;
1392 EXPORT_SYMBOL(thaw_super);