4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * super.c contains code to handle: - mount structures
8 * - filesystem drivers list
10 * - umount system call
13 * GK 2/5/95 - Changed to support mounting the root fs via NFS
15 * Added kerneld support: Jacques Gelinas and Bjorn Ekwall
16 * Added change_root: Werner Almesberger & Hans Lermen, Feb '96
17 * Added options to /proc/mounts:
18 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
19 * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
20 * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
23 #include <linux/export.h>
24 #include <linux/slab.h>
25 #include <linux/acct.h>
26 #include <linux/blkdev.h>
27 #include <linux/mount.h>
28 #include <linux/security.h>
29 #include <linux/writeback.h> /* for the emergency remount stuff */
30 #include <linux/idr.h>
31 #include <linux/mutex.h>
32 #include <linux/backing-dev.h>
33 #include <linux/rculist_bl.h>
34 #include <linux/cleancache.h>
35 #include <linux/fsnotify.h>
36 #include <linux/lockdep.h>
40 LIST_HEAD(super_blocks);
41 DEFINE_SPINLOCK(sb_lock);
43 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
50 * One thing we have to be careful of with a per-sb shrinker is that we don't
51 * drop the last active reference to the superblock from within the shrinker.
52 * If that happens we could trigger unregistering the shrinker from within the
53 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
54 * take a passive reference to the superblock to avoid this from occurring.
56 static unsigned long super_cache_scan(struct shrinker *shrink,
57 struct shrink_control *sc)
59 struct super_block *sb;
66 sb = container_of(shrink, struct super_block, s_shrink);
69 * Deadlock avoidance. We may hold various FS locks, and we don't want
70 * to recurse into the FS that called us in clear_inode() and friends..
72 if (!(sc->gfp_mask & __GFP_FS))
75 if (!grab_super_passive(sb))
78 if (sb->s_op && sb->s_op->nr_cached_objects)
79 fs_objects = sb->s_op->nr_cached_objects(sb);
81 inodes = list_lru_count(&sb->s_inode_lru);
82 dentries = list_lru_count(&sb->s_dentry_lru);
83 total_objects = dentries + inodes + fs_objects + 1;
85 /* proportion the scan between the caches */
86 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
87 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
90 * prune the dcache first as the icache is pinned by it, then
91 * prune the icache, followed by the filesystem specific caches
93 freed = prune_dcache_sb(sb, dentries);
94 freed += prune_icache_sb(sb, inodes);
97 fs_objects = mult_frac(sc->nr_to_scan, fs_objects,
99 freed += sb->s_op->free_cached_objects(sb, fs_objects);
106 static unsigned long super_cache_count(struct shrinker *shrink,
107 struct shrink_control *sc)
109 struct super_block *sb;
110 long total_objects = 0;
112 sb = container_of(shrink, struct super_block, s_shrink);
114 if (!grab_super_passive(sb))
117 if (sb->s_op && sb->s_op->nr_cached_objects)
118 total_objects = sb->s_op->nr_cached_objects(sb);
120 total_objects += list_lru_count(&sb->s_dentry_lru);
121 total_objects += list_lru_count(&sb->s_inode_lru);
123 total_objects = vfs_pressure_ratio(total_objects);
125 return total_objects;
128 static int init_sb_writers(struct super_block *s, struct file_system_type *type)
133 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
134 err = percpu_counter_init(&s->s_writers.counter[i], 0);
137 lockdep_init_map(&s->s_writers.lock_map[i], sb_writers_name[i],
138 &type->s_writers_key[i], 0);
140 init_waitqueue_head(&s->s_writers.wait);
141 init_waitqueue_head(&s->s_writers.wait_unfrozen);
145 percpu_counter_destroy(&s->s_writers.counter[i]);
149 static void destroy_sb_writers(struct super_block *s)
153 for (i = 0; i < SB_FREEZE_LEVELS; i++)
154 percpu_counter_destroy(&s->s_writers.counter[i]);
158 * alloc_super - create new superblock
159 * @type: filesystem type superblock should belong to
160 * @flags: the mount flags
162 * Allocates and initializes a new &struct super_block. alloc_super()
163 * returns a pointer new superblock or %NULL if allocation had failed.
165 static struct super_block *alloc_super(struct file_system_type *type, int flags)
167 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
168 static const struct super_operations default_op;
171 if (security_sb_alloc(s)) {
173 * We cannot call security_sb_free() without
174 * security_sb_alloc() succeeding. So bail out manually
181 s->s_files = alloc_percpu(struct list_head);
187 for_each_possible_cpu(i)
188 INIT_LIST_HEAD(per_cpu_ptr(s->s_files, i));
191 INIT_LIST_HEAD(&s->s_files);
193 if (init_sb_writers(s, type))
196 s->s_bdi = &default_backing_dev_info;
197 INIT_HLIST_NODE(&s->s_instances);
198 INIT_HLIST_BL_HEAD(&s->s_anon);
199 INIT_LIST_HEAD(&s->s_inodes);
200 list_lru_init(&s->s_dentry_lru);
201 list_lru_init(&s->s_inode_lru);
202 INIT_LIST_HEAD(&s->s_mounts);
203 init_rwsem(&s->s_umount);
204 lockdep_set_class(&s->s_umount, &type->s_umount_key);
206 * sget() can have s_umount recursion.
208 * When it cannot find a suitable sb, it allocates a new
209 * one (this one), and tries again to find a suitable old
212 * In case that succeeds, it will acquire the s_umount
213 * lock of the old one. Since these are clearly distrinct
214 * locks, and this object isn't exposed yet, there's no
217 * Annotate this by putting this lock in a different
220 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
222 atomic_set(&s->s_active, 1);
223 mutex_init(&s->s_vfs_rename_mutex);
224 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
225 mutex_init(&s->s_dquot.dqio_mutex);
226 mutex_init(&s->s_dquot.dqonoff_mutex);
227 init_rwsem(&s->s_dquot.dqptr_sem);
228 s->s_maxbytes = MAX_NON_LFS;
229 s->s_op = &default_op;
230 s->s_time_gran = 1000000000;
231 s->cleancache_poolid = -1;
233 s->s_shrink.seeks = DEFAULT_SEEKS;
234 s->s_shrink.scan_objects = super_cache_scan;
235 s->s_shrink.count_objects = super_cache_count;
236 s->s_shrink.batch = 1024;
244 free_percpu(s->s_files);
246 destroy_sb_writers(s);
253 * destroy_super - frees a superblock
254 * @s: superblock to free
256 * Frees a superblock.
258 static inline void destroy_super(struct super_block *s)
261 free_percpu(s->s_files);
263 destroy_sb_writers(s);
265 WARN_ON(!list_empty(&s->s_mounts));
271 /* Superblock refcounting */
274 * Drop a superblock's refcount. The caller must hold sb_lock.
276 static void __put_super(struct super_block *sb)
278 if (!--sb->s_count) {
279 list_del_init(&sb->s_list);
285 * put_super - drop a temporary reference to superblock
286 * @sb: superblock in question
288 * Drops a temporary reference, frees superblock if there's no
291 static void put_super(struct super_block *sb)
295 spin_unlock(&sb_lock);
300 * deactivate_locked_super - drop an active reference to superblock
301 * @s: superblock to deactivate
303 * Drops an active reference to superblock, converting it into a temprory
304 * one if there is no other active references left. In that case we
305 * tell fs driver to shut it down and drop the temporary reference we
308 * Caller holds exclusive lock on superblock; that lock is released.
310 void deactivate_locked_super(struct super_block *s)
312 struct file_system_type *fs = s->s_type;
313 if (atomic_dec_and_test(&s->s_active)) {
314 cleancache_invalidate_fs(s);
317 /* caches are now gone, we can safely kill the shrinker now */
318 unregister_shrinker(&s->s_shrink);
322 up_write(&s->s_umount);
326 EXPORT_SYMBOL(deactivate_locked_super);
329 * deactivate_super - drop an active reference to superblock
330 * @s: superblock to deactivate
332 * Variant of deactivate_locked_super(), except that superblock is *not*
333 * locked by caller. If we are going to drop the final active reference,
334 * lock will be acquired prior to that.
336 void deactivate_super(struct super_block *s)
338 if (!atomic_add_unless(&s->s_active, -1, 1)) {
339 down_write(&s->s_umount);
340 deactivate_locked_super(s);
344 EXPORT_SYMBOL(deactivate_super);
347 * grab_super - acquire an active reference
348 * @s: reference we are trying to make active
350 * Tries to acquire an active reference. grab_super() is used when we
351 * had just found a superblock in super_blocks or fs_type->fs_supers
352 * and want to turn it into a full-blown active reference. grab_super()
353 * is called with sb_lock held and drops it. Returns 1 in case of
354 * success, 0 if we had failed (superblock contents was already dead or
355 * dying when grab_super() had been called). Note that this is only
356 * called for superblocks not in rundown mode (== ones still on ->fs_supers
357 * of their type), so increment of ->s_count is OK here.
359 static int grab_super(struct super_block *s) __releases(sb_lock)
362 spin_unlock(&sb_lock);
363 down_write(&s->s_umount);
364 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
368 up_write(&s->s_umount);
374 * grab_super_passive - acquire a passive reference
375 * @sb: reference we are trying to grab
377 * Tries to acquire a passive reference. This is used in places where we
378 * cannot take an active reference but we need to ensure that the
379 * superblock does not go away while we are working on it. It returns
380 * false if a reference was not gained, and returns true with the s_umount
381 * lock held in read mode if a reference is gained. On successful return,
382 * the caller must drop the s_umount lock and the passive reference when
385 bool grab_super_passive(struct super_block *sb)
388 if (hlist_unhashed(&sb->s_instances)) {
389 spin_unlock(&sb_lock);
394 spin_unlock(&sb_lock);
396 if (down_read_trylock(&sb->s_umount)) {
397 if (sb->s_root && (sb->s_flags & MS_BORN))
399 up_read(&sb->s_umount);
407 * generic_shutdown_super - common helper for ->kill_sb()
408 * @sb: superblock to kill
410 * generic_shutdown_super() does all fs-independent work on superblock
411 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
412 * that need destruction out of superblock, call generic_shutdown_super()
413 * and release aforementioned objects. Note: dentries and inodes _are_
414 * taken care of and do not need specific handling.
416 * Upon calling this function, the filesystem may no longer alter or
417 * rearrange the set of dentries belonging to this super_block, nor may it
418 * change the attachments of dentries to inodes.
420 void generic_shutdown_super(struct super_block *sb)
422 const struct super_operations *sop = sb->s_op;
425 shrink_dcache_for_umount(sb);
427 sb->s_flags &= ~MS_ACTIVE;
429 fsnotify_unmount_inodes(&sb->s_inodes);
436 if (!list_empty(&sb->s_inodes)) {
437 printk("VFS: Busy inodes after unmount of %s. "
438 "Self-destruct in 5 seconds. Have a nice day...\n",
443 /* should be initialized for __put_super_and_need_restart() */
444 hlist_del_init(&sb->s_instances);
445 spin_unlock(&sb_lock);
446 up_write(&sb->s_umount);
449 EXPORT_SYMBOL(generic_shutdown_super);
452 * sget - find or create a superblock
453 * @type: filesystem type superblock should belong to
454 * @test: comparison callback
455 * @set: setup callback
456 * @flags: mount flags
457 * @data: argument to each of them
459 struct super_block *sget(struct file_system_type *type,
460 int (*test)(struct super_block *,void *),
461 int (*set)(struct super_block *,void *),
465 struct super_block *s = NULL;
466 struct super_block *old;
472 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
473 if (!test(old, data))
475 if (!grab_super(old))
478 up_write(&s->s_umount);
486 spin_unlock(&sb_lock);
487 s = alloc_super(type, flags);
489 return ERR_PTR(-ENOMEM);
495 spin_unlock(&sb_lock);
496 up_write(&s->s_umount);
501 strlcpy(s->s_id, type->name, sizeof(s->s_id));
502 list_add_tail(&s->s_list, &super_blocks);
503 hlist_add_head(&s->s_instances, &type->fs_supers);
504 spin_unlock(&sb_lock);
505 get_filesystem(type);
506 register_shrinker(&s->s_shrink);
512 void drop_super(struct super_block *sb)
514 up_read(&sb->s_umount);
518 EXPORT_SYMBOL(drop_super);
521 * iterate_supers - call function for all active superblocks
522 * @f: function to call
523 * @arg: argument to pass to it
525 * Scans the superblock list and calls given function, passing it
526 * locked superblock and given argument.
528 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
530 struct super_block *sb, *p = NULL;
533 list_for_each_entry(sb, &super_blocks, s_list) {
534 if (hlist_unhashed(&sb->s_instances))
537 spin_unlock(&sb_lock);
539 down_read(&sb->s_umount);
540 if (sb->s_root && (sb->s_flags & MS_BORN))
542 up_read(&sb->s_umount);
551 spin_unlock(&sb_lock);
555 * iterate_supers_type - call function for superblocks of given type
557 * @f: function to call
558 * @arg: argument to pass to it
560 * Scans the superblock list and calls given function, passing it
561 * locked superblock and given argument.
563 void iterate_supers_type(struct file_system_type *type,
564 void (*f)(struct super_block *, void *), void *arg)
566 struct super_block *sb, *p = NULL;
569 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
571 spin_unlock(&sb_lock);
573 down_read(&sb->s_umount);
574 if (sb->s_root && (sb->s_flags & MS_BORN))
576 up_read(&sb->s_umount);
585 spin_unlock(&sb_lock);
588 EXPORT_SYMBOL(iterate_supers_type);
591 * get_super - get the superblock of a device
592 * @bdev: device to get the superblock for
594 * Scans the superblock list and finds the superblock of the file system
595 * mounted on the device given. %NULL is returned if no match is found.
598 struct super_block *get_super(struct block_device *bdev)
600 struct super_block *sb;
607 list_for_each_entry(sb, &super_blocks, s_list) {
608 if (hlist_unhashed(&sb->s_instances))
610 if (sb->s_bdev == bdev) {
612 spin_unlock(&sb_lock);
613 down_read(&sb->s_umount);
615 if (sb->s_root && (sb->s_flags & MS_BORN))
617 up_read(&sb->s_umount);
618 /* nope, got unmounted */
624 spin_unlock(&sb_lock);
628 EXPORT_SYMBOL(get_super);
631 * get_super_thawed - get thawed superblock of a device
632 * @bdev: device to get the superblock for
634 * Scans the superblock list and finds the superblock of the file system
635 * mounted on the device. The superblock is returned once it is thawed
636 * (or immediately if it was not frozen). %NULL is returned if no match
639 struct super_block *get_super_thawed(struct block_device *bdev)
642 struct super_block *s = get_super(bdev);
643 if (!s || s->s_writers.frozen == SB_UNFROZEN)
645 up_read(&s->s_umount);
646 wait_event(s->s_writers.wait_unfrozen,
647 s->s_writers.frozen == SB_UNFROZEN);
651 EXPORT_SYMBOL(get_super_thawed);
654 * get_active_super - get an active reference to the superblock of a device
655 * @bdev: device to get the superblock for
657 * Scans the superblock list and finds the superblock of the file system
658 * mounted on the device given. Returns the superblock with an active
659 * reference or %NULL if none was found.
661 struct super_block *get_active_super(struct block_device *bdev)
663 struct super_block *sb;
670 list_for_each_entry(sb, &super_blocks, s_list) {
671 if (hlist_unhashed(&sb->s_instances))
673 if (sb->s_bdev == bdev) {
676 up_write(&sb->s_umount);
680 spin_unlock(&sb_lock);
684 struct super_block *user_get_super(dev_t dev)
686 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_dev == dev) {
695 spin_unlock(&sb_lock);
696 down_read(&sb->s_umount);
698 if (sb->s_root && (sb->s_flags & MS_BORN))
700 up_read(&sb->s_umount);
701 /* nope, got unmounted */
707 spin_unlock(&sb_lock);
712 * do_remount_sb - asks filesystem to change mount options.
713 * @sb: superblock in question
714 * @flags: numeric part of options
715 * @data: the rest of options
716 * @force: whether or not to force the change
718 * Alters the mount options of a mounted file system.
720 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
725 if (sb->s_writers.frozen != SB_UNFROZEN)
729 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
733 if (flags & MS_RDONLY)
735 shrink_dcache_sb(sb);
738 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
740 /* If we are remounting RDONLY and current sb is read/write,
741 make sure there are no rw files opened */
746 retval = sb_prepare_remount_readonly(sb);
752 if (sb->s_op->remount_fs) {
753 retval = sb->s_op->remount_fs(sb, &flags, data);
756 goto cancel_readonly;
757 /* If forced remount, go ahead despite any errors */
758 WARN(1, "forced remount of a %s fs returned %i\n",
759 sb->s_type->name, retval);
762 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
763 /* Needs to be ordered wrt mnt_is_readonly() */
765 sb->s_readonly_remount = 0;
768 * Some filesystems modify their metadata via some other path than the
769 * bdev buffer cache (eg. use a private mapping, or directories in
770 * pagecache, etc). Also file data modifications go via their own
771 * mappings. So If we try to mount readonly then copy the filesystem
772 * from bdev, we could get stale data, so invalidate it to give a best
773 * effort at coherency.
775 if (remount_ro && sb->s_bdev)
776 invalidate_bdev(sb->s_bdev);
780 sb->s_readonly_remount = 0;
784 static void do_emergency_remount(struct work_struct *work)
786 struct super_block *sb, *p = NULL;
789 list_for_each_entry(sb, &super_blocks, s_list) {
790 if (hlist_unhashed(&sb->s_instances))
793 spin_unlock(&sb_lock);
794 down_write(&sb->s_umount);
795 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
796 !(sb->s_flags & MS_RDONLY)) {
798 * What lock protects sb->s_flags??
800 do_remount_sb(sb, MS_RDONLY, NULL, 1);
802 up_write(&sb->s_umount);
810 spin_unlock(&sb_lock);
812 printk("Emergency Remount complete\n");
815 void emergency_remount(void)
817 struct work_struct *work;
819 work = kmalloc(sizeof(*work), GFP_ATOMIC);
821 INIT_WORK(work, do_emergency_remount);
827 * Unnamed block devices are dummy devices used by virtual
828 * filesystems which don't use real block-devices. -- jrs
831 static DEFINE_IDA(unnamed_dev_ida);
832 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
833 static int unnamed_dev_start = 0; /* don't bother trying below it */
835 int get_anon_bdev(dev_t *p)
841 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
843 spin_lock(&unnamed_dev_lock);
844 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
846 unnamed_dev_start = dev + 1;
847 spin_unlock(&unnamed_dev_lock);
848 if (error == -EAGAIN)
849 /* We raced and lost with another CPU. */
854 if (dev == (1 << MINORBITS)) {
855 spin_lock(&unnamed_dev_lock);
856 ida_remove(&unnamed_dev_ida, dev);
857 if (unnamed_dev_start > dev)
858 unnamed_dev_start = dev;
859 spin_unlock(&unnamed_dev_lock);
862 *p = MKDEV(0, dev & MINORMASK);
865 EXPORT_SYMBOL(get_anon_bdev);
867 void free_anon_bdev(dev_t dev)
869 int slot = MINOR(dev);
870 spin_lock(&unnamed_dev_lock);
871 ida_remove(&unnamed_dev_ida, slot);
872 if (slot < unnamed_dev_start)
873 unnamed_dev_start = slot;
874 spin_unlock(&unnamed_dev_lock);
876 EXPORT_SYMBOL(free_anon_bdev);
878 int set_anon_super(struct super_block *s, void *data)
880 int error = get_anon_bdev(&s->s_dev);
882 s->s_bdi = &noop_backing_dev_info;
886 EXPORT_SYMBOL(set_anon_super);
888 void kill_anon_super(struct super_block *sb)
890 dev_t dev = sb->s_dev;
891 generic_shutdown_super(sb);
895 EXPORT_SYMBOL(kill_anon_super);
897 void kill_litter_super(struct super_block *sb)
900 d_genocide(sb->s_root);
904 EXPORT_SYMBOL(kill_litter_super);
906 static int ns_test_super(struct super_block *sb, void *data)
908 return sb->s_fs_info == data;
911 static int ns_set_super(struct super_block *sb, void *data)
913 sb->s_fs_info = data;
914 return set_anon_super(sb, NULL);
917 struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
918 void *data, int (*fill_super)(struct super_block *, void *, int))
920 struct super_block *sb;
922 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
928 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
930 deactivate_locked_super(sb);
934 sb->s_flags |= MS_ACTIVE;
937 return dget(sb->s_root);
940 EXPORT_SYMBOL(mount_ns);
943 static int set_bdev_super(struct super_block *s, void *data)
946 s->s_dev = s->s_bdev->bd_dev;
949 * We set the bdi here to the queue backing, file systems can
950 * overwrite this in ->fill_super()
952 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
956 static int test_bdev_super(struct super_block *s, void *data)
958 return (void *)s->s_bdev == data;
961 struct dentry *mount_bdev(struct file_system_type *fs_type,
962 int flags, const char *dev_name, void *data,
963 int (*fill_super)(struct super_block *, void *, int))
965 struct block_device *bdev;
966 struct super_block *s;
967 fmode_t mode = FMODE_READ | FMODE_EXCL;
970 if (!(flags & MS_RDONLY))
973 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
975 return ERR_CAST(bdev);
978 * once the super is inserted into the list by sget, s_umount
979 * will protect the lockfs code from trying to start a snapshot
980 * while we are mounting
982 mutex_lock(&bdev->bd_fsfreeze_mutex);
983 if (bdev->bd_fsfreeze_count > 0) {
984 mutex_unlock(&bdev->bd_fsfreeze_mutex);
988 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
990 mutex_unlock(&bdev->bd_fsfreeze_mutex);
995 if ((flags ^ s->s_flags) & MS_RDONLY) {
996 deactivate_locked_super(s);
1002 * s_umount nests inside bd_mutex during
1003 * __invalidate_device(). blkdev_put() acquires
1004 * bd_mutex and can't be called under s_umount. Drop
1005 * s_umount temporarily. This is safe as we're
1006 * holding an active reference.
1008 up_write(&s->s_umount);
1009 blkdev_put(bdev, mode);
1010 down_write(&s->s_umount);
1012 char b[BDEVNAME_SIZE];
1015 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1016 sb_set_blocksize(s, block_size(bdev));
1017 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1019 deactivate_locked_super(s);
1023 s->s_flags |= MS_ACTIVE;
1027 return dget(s->s_root);
1032 blkdev_put(bdev, mode);
1034 return ERR_PTR(error);
1036 EXPORT_SYMBOL(mount_bdev);
1038 void kill_block_super(struct super_block *sb)
1040 struct block_device *bdev = sb->s_bdev;
1041 fmode_t mode = sb->s_mode;
1043 bdev->bd_super = NULL;
1044 generic_shutdown_super(sb);
1045 sync_blockdev(bdev);
1046 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1047 blkdev_put(bdev, mode | FMODE_EXCL);
1050 EXPORT_SYMBOL(kill_block_super);
1053 struct dentry *mount_nodev(struct file_system_type *fs_type,
1054 int flags, void *data,
1055 int (*fill_super)(struct super_block *, void *, int))
1058 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1063 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1065 deactivate_locked_super(s);
1066 return ERR_PTR(error);
1068 s->s_flags |= MS_ACTIVE;
1069 return dget(s->s_root);
1071 EXPORT_SYMBOL(mount_nodev);
1073 static int compare_single(struct super_block *s, void *p)
1078 struct dentry *mount_single(struct file_system_type *fs_type,
1079 int flags, void *data,
1080 int (*fill_super)(struct super_block *, void *, int))
1082 struct super_block *s;
1085 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1089 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1091 deactivate_locked_super(s);
1092 return ERR_PTR(error);
1094 s->s_flags |= MS_ACTIVE;
1096 do_remount_sb(s, flags, data, 0);
1098 return dget(s->s_root);
1100 EXPORT_SYMBOL(mount_single);
1103 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1105 struct dentry *root;
1106 struct super_block *sb;
1107 char *secdata = NULL;
1108 int error = -ENOMEM;
1110 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1111 secdata = alloc_secdata();
1115 error = security_sb_copy_data(data, secdata);
1117 goto out_free_secdata;
1120 root = type->mount(type, flags, name, data);
1122 error = PTR_ERR(root);
1123 goto out_free_secdata;
1127 WARN_ON(!sb->s_bdi);
1128 WARN_ON(sb->s_bdi == &default_backing_dev_info);
1129 sb->s_flags |= MS_BORN;
1131 error = security_sb_kern_mount(sb, flags, secdata);
1136 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1137 * but s_maxbytes was an unsigned long long for many releases. Throw
1138 * this warning for a little while to try and catch filesystems that
1139 * violate this rule.
1141 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1142 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1144 up_write(&sb->s_umount);
1145 free_secdata(secdata);
1149 deactivate_locked_super(sb);
1151 free_secdata(secdata);
1153 return ERR_PTR(error);
1157 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1160 void __sb_end_write(struct super_block *sb, int level)
1162 percpu_counter_dec(&sb->s_writers.counter[level-1]);
1164 * Make sure s_writers are updated before we wake up waiters in
1168 if (waitqueue_active(&sb->s_writers.wait))
1169 wake_up(&sb->s_writers.wait);
1170 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _RET_IP_);
1172 EXPORT_SYMBOL(__sb_end_write);
1174 #ifdef CONFIG_LOCKDEP
1176 * We want lockdep to tell us about possible deadlocks with freezing but
1177 * it's it bit tricky to properly instrument it. Getting a freeze protection
1178 * works as getting a read lock but there are subtle problems. XFS for example
1179 * gets freeze protection on internal level twice in some cases, which is OK
1180 * only because we already hold a freeze protection also on higher level. Due
1181 * to these cases we have to tell lockdep we are doing trylock when we
1182 * already hold a freeze protection for a higher freeze level.
1184 static void acquire_freeze_lock(struct super_block *sb, int level, bool trylock,
1190 for (i = 0; i < level - 1; i++)
1191 if (lock_is_held(&sb->s_writers.lock_map[i])) {
1196 rwsem_acquire_read(&sb->s_writers.lock_map[level-1], 0, trylock, ip);
1201 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1204 int __sb_start_write(struct super_block *sb, int level, bool wait)
1207 if (unlikely(sb->s_writers.frozen >= level)) {
1210 wait_event(sb->s_writers.wait_unfrozen,
1211 sb->s_writers.frozen < level);
1214 #ifdef CONFIG_LOCKDEP
1215 acquire_freeze_lock(sb, level, !wait, _RET_IP_);
1217 percpu_counter_inc(&sb->s_writers.counter[level-1]);
1219 * Make sure counter is updated before we check for frozen.
1220 * freeze_super() first sets frozen and then checks the counter.
1223 if (unlikely(sb->s_writers.frozen >= level)) {
1224 __sb_end_write(sb, level);
1229 EXPORT_SYMBOL(__sb_start_write);
1232 * sb_wait_write - wait until all writers to given file system finish
1233 * @sb: the super for which we wait
1234 * @level: type of writers we wait for (normal vs page fault)
1236 * This function waits until there are no writers of given type to given file
1237 * system. Caller of this function should make sure there can be no new writers
1238 * of type @level before calling this function. Otherwise this function can
1241 static void sb_wait_write(struct super_block *sb, int level)
1246 * We just cycle-through lockdep here so that it does not complain
1247 * about returning with lock to userspace
1249 rwsem_acquire(&sb->s_writers.lock_map[level-1], 0, 0, _THIS_IP_);
1250 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _THIS_IP_);
1256 * We use a barrier in prepare_to_wait() to separate setting
1257 * of frozen and checking of the counter
1259 prepare_to_wait(&sb->s_writers.wait, &wait,
1260 TASK_UNINTERRUPTIBLE);
1262 writers = percpu_counter_sum(&sb->s_writers.counter[level-1]);
1266 finish_wait(&sb->s_writers.wait, &wait);
1271 * freeze_super - lock the filesystem and force it into a consistent state
1272 * @sb: the super to lock
1274 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1275 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1278 * During this function, sb->s_writers.frozen goes through these values:
1280 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1282 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1283 * writes should be blocked, though page faults are still allowed. We wait for
1284 * all writes to complete and then proceed to the next stage.
1286 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1287 * but internal fs threads can still modify the filesystem (although they
1288 * should not dirty new pages or inodes), writeback can run etc. After waiting
1289 * for all running page faults we sync the filesystem which will clean all
1290 * dirty pages and inodes (no new dirty pages or inodes can be created when
1293 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1294 * modification are blocked (e.g. XFS preallocation truncation on inode
1295 * reclaim). This is usually implemented by blocking new transactions for
1296 * filesystems that have them and need this additional guard. After all
1297 * internal writers are finished we call ->freeze_fs() to finish filesystem
1298 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1299 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1301 * sb->s_writers.frozen is protected by sb->s_umount.
1303 int freeze_super(struct super_block *sb)
1307 atomic_inc(&sb->s_active);
1308 down_write(&sb->s_umount);
1309 if (sb->s_writers.frozen != SB_UNFROZEN) {
1310 deactivate_locked_super(sb);
1314 if (!(sb->s_flags & MS_BORN)) {
1315 up_write(&sb->s_umount);
1316 return 0; /* sic - it's "nothing to do" */
1319 if (sb->s_flags & MS_RDONLY) {
1320 /* Nothing to do really... */
1321 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1322 up_write(&sb->s_umount);
1326 /* From now on, no new normal writers can start */
1327 sb->s_writers.frozen = SB_FREEZE_WRITE;
1330 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1331 up_write(&sb->s_umount);
1333 sb_wait_write(sb, SB_FREEZE_WRITE);
1335 /* Now we go and block page faults... */
1336 down_write(&sb->s_umount);
1337 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1340 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1342 /* All writers are done so after syncing there won't be dirty data */
1343 sync_filesystem(sb);
1345 /* Now wait for internal filesystem counter */
1346 sb->s_writers.frozen = SB_FREEZE_FS;
1348 sb_wait_write(sb, SB_FREEZE_FS);
1350 if (sb->s_op->freeze_fs) {
1351 ret = sb->s_op->freeze_fs(sb);
1354 "VFS:Filesystem freeze failed\n");
1355 sb->s_writers.frozen = SB_UNFROZEN;
1357 wake_up(&sb->s_writers.wait_unfrozen);
1358 deactivate_locked_super(sb);
1363 * This is just for debugging purposes so that fs can warn if it
1364 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1366 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1367 up_write(&sb->s_umount);
1370 EXPORT_SYMBOL(freeze_super);
1373 * thaw_super -- unlock filesystem
1374 * @sb: the super to thaw
1376 * Unlocks the filesystem and marks it writeable again after freeze_super().
1378 int thaw_super(struct super_block *sb)
1382 down_write(&sb->s_umount);
1383 if (sb->s_writers.frozen == SB_UNFROZEN) {
1384 up_write(&sb->s_umount);
1388 if (sb->s_flags & MS_RDONLY)
1391 if (sb->s_op->unfreeze_fs) {
1392 error = sb->s_op->unfreeze_fs(sb);
1395 "VFS:Filesystem thaw failed\n");
1396 up_write(&sb->s_umount);
1402 sb->s_writers.frozen = SB_UNFROZEN;
1404 wake_up(&sb->s_writers.wait_unfrozen);
1405 deactivate_locked_super(sb);
1409 EXPORT_SYMBOL(thaw_super);