4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * super.c contains code to handle: - mount structures
8 * - filesystem drivers list
10 * - umount system call
13 * GK 2/5/95 - Changed to support mounting the root fs via NFS
15 * Added kerneld support: Jacques Gelinas and Bjorn Ekwall
16 * Added change_root: Werner Almesberger & Hans Lermen, Feb '96
17 * Added options to /proc/mounts:
18 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
19 * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
20 * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
23 #include <linux/export.h>
24 #include <linux/slab.h>
25 #include <linux/acct.h>
26 #include <linux/blkdev.h>
27 #include <linux/mount.h>
28 #include <linux/security.h>
29 #include <linux/writeback.h> /* for the emergency remount stuff */
30 #include <linux/idr.h>
31 #include <linux/mutex.h>
32 #include <linux/backing-dev.h>
33 #include <linux/rculist_bl.h>
34 #include <linux/cleancache.h>
35 #include <linux/fsnotify.h>
36 #include <linux/lockdep.h>
40 LIST_HEAD(super_blocks);
41 DEFINE_SPINLOCK(sb_lock);
43 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
50 * One thing we have to be careful of with a per-sb shrinker is that we don't
51 * drop the last active reference to the superblock from within the shrinker.
52 * If that happens we could trigger unregistering the shrinker from within the
53 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
54 * take a passive reference to the superblock to avoid this from occurring.
56 static unsigned long super_cache_scan(struct shrinker *shrink,
57 struct shrink_control *sc)
59 struct super_block *sb;
66 sb = container_of(shrink, struct super_block, s_shrink);
69 * Deadlock avoidance. We may hold various FS locks, and we don't want
70 * to recurse into the FS that called us in clear_inode() and friends..
72 if (!(sc->gfp_mask & __GFP_FS))
75 if (!grab_super_passive(sb))
78 if (sb->s_op->nr_cached_objects)
79 fs_objects = sb->s_op->nr_cached_objects(sb, sc->nid);
81 inodes = list_lru_count_node(&sb->s_inode_lru, sc->nid);
82 dentries = list_lru_count_node(&sb->s_dentry_lru, sc->nid);
83 total_objects = dentries + inodes + fs_objects + 1;
85 /* proportion the scan between the caches */
86 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
87 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
90 * prune the dcache first as the icache is pinned by it, then
91 * prune the icache, followed by the filesystem specific caches
93 freed = prune_dcache_sb(sb, dentries, sc->nid);
94 freed += prune_icache_sb(sb, inodes, sc->nid);
97 fs_objects = mult_frac(sc->nr_to_scan, fs_objects,
99 freed += sb->s_op->free_cached_objects(sb, fs_objects,
107 static unsigned long super_cache_count(struct shrinker *shrink,
108 struct shrink_control *sc)
110 struct super_block *sb;
111 long total_objects = 0;
113 sb = container_of(shrink, struct super_block, s_shrink);
115 if (!grab_super_passive(sb))
118 if (sb->s_op && sb->s_op->nr_cached_objects)
119 total_objects = sb->s_op->nr_cached_objects(sb,
122 total_objects += list_lru_count_node(&sb->s_dentry_lru,
124 total_objects += list_lru_count_node(&sb->s_inode_lru,
127 total_objects = vfs_pressure_ratio(total_objects);
129 return total_objects;
132 static int init_sb_writers(struct super_block *s, struct file_system_type *type)
137 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
138 err = percpu_counter_init(&s->s_writers.counter[i], 0);
141 lockdep_init_map(&s->s_writers.lock_map[i], sb_writers_name[i],
142 &type->s_writers_key[i], 0);
144 init_waitqueue_head(&s->s_writers.wait);
145 init_waitqueue_head(&s->s_writers.wait_unfrozen);
149 percpu_counter_destroy(&s->s_writers.counter[i]);
153 static void destroy_sb_writers(struct super_block *s)
157 for (i = 0; i < SB_FREEZE_LEVELS; i++)
158 percpu_counter_destroy(&s->s_writers.counter[i]);
162 * alloc_super - create new superblock
163 * @type: filesystem type superblock should belong to
164 * @flags: the mount flags
166 * Allocates and initializes a new &struct super_block. alloc_super()
167 * returns a pointer new superblock or %NULL if allocation had failed.
169 static struct super_block *alloc_super(struct file_system_type *type, int flags)
171 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
172 static const struct super_operations default_op;
175 if (security_sb_alloc(s))
179 s->s_files = alloc_percpu(struct list_head);
185 for_each_possible_cpu(i)
186 INIT_LIST_HEAD(per_cpu_ptr(s->s_files, i));
189 INIT_LIST_HEAD(&s->s_files);
191 if (init_sb_writers(s, type))
194 s->s_bdi = &default_backing_dev_info;
195 INIT_HLIST_NODE(&s->s_instances);
196 INIT_HLIST_BL_HEAD(&s->s_anon);
197 INIT_LIST_HEAD(&s->s_inodes);
199 if (list_lru_init(&s->s_dentry_lru))
201 if (list_lru_init(&s->s_inode_lru))
202 goto err_out_dentry_lru;
204 INIT_LIST_HEAD(&s->s_mounts);
205 init_rwsem(&s->s_umount);
206 lockdep_set_class(&s->s_umount, &type->s_umount_key);
208 * sget() can have s_umount recursion.
210 * When it cannot find a suitable sb, it allocates a new
211 * one (this one), and tries again to find a suitable old
214 * In case that succeeds, it will acquire the s_umount
215 * lock of the old one. Since these are clearly distrinct
216 * locks, and this object isn't exposed yet, there's no
219 * Annotate this by putting this lock in a different
222 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
224 atomic_set(&s->s_active, 1);
225 mutex_init(&s->s_vfs_rename_mutex);
226 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
227 mutex_init(&s->s_dquot.dqio_mutex);
228 mutex_init(&s->s_dquot.dqonoff_mutex);
229 init_rwsem(&s->s_dquot.dqptr_sem);
230 s->s_maxbytes = MAX_NON_LFS;
231 s->s_op = &default_op;
232 s->s_time_gran = 1000000000;
233 s->cleancache_poolid = -1;
235 s->s_shrink.seeks = DEFAULT_SEEKS;
236 s->s_shrink.scan_objects = super_cache_scan;
237 s->s_shrink.count_objects = super_cache_count;
238 s->s_shrink.batch = 1024;
239 s->s_shrink.flags = SHRINKER_NUMA_AWARE;
245 list_lru_destroy(&s->s_dentry_lru);
250 free_percpu(s->s_files);
252 destroy_sb_writers(s);
260 * destroy_super - frees a superblock
261 * @s: superblock to free
263 * Frees a superblock.
265 static inline void destroy_super(struct super_block *s)
267 list_lru_destroy(&s->s_dentry_lru);
268 list_lru_destroy(&s->s_inode_lru);
270 free_percpu(s->s_files);
272 destroy_sb_writers(s);
274 WARN_ON(!list_empty(&s->s_mounts));
280 /* Superblock refcounting */
283 * Drop a superblock's refcount. The caller must hold sb_lock.
285 static void __put_super(struct super_block *sb)
287 if (!--sb->s_count) {
288 list_del_init(&sb->s_list);
294 * put_super - drop a temporary reference to superblock
295 * @sb: superblock in question
297 * Drops a temporary reference, frees superblock if there's no
300 static void put_super(struct super_block *sb)
304 spin_unlock(&sb_lock);
309 * deactivate_locked_super - drop an active reference to superblock
310 * @s: superblock to deactivate
312 * Drops an active reference to superblock, converting it into a temprory
313 * one if there is no other active references left. In that case we
314 * tell fs driver to shut it down and drop the temporary reference we
317 * Caller holds exclusive lock on superblock; that lock is released.
319 void deactivate_locked_super(struct super_block *s)
321 struct file_system_type *fs = s->s_type;
322 if (atomic_dec_and_test(&s->s_active)) {
323 cleancache_invalidate_fs(s);
326 /* caches are now gone, we can safely kill the shrinker now */
327 unregister_shrinker(&s->s_shrink);
332 up_write(&s->s_umount);
336 EXPORT_SYMBOL(deactivate_locked_super);
339 * deactivate_super - drop an active reference to superblock
340 * @s: superblock to deactivate
342 * Variant of deactivate_locked_super(), except that superblock is *not*
343 * locked by caller. If we are going to drop the final active reference,
344 * lock will be acquired prior to that.
346 void deactivate_super(struct super_block *s)
348 if (!atomic_add_unless(&s->s_active, -1, 1)) {
349 down_write(&s->s_umount);
350 deactivate_locked_super(s);
354 EXPORT_SYMBOL(deactivate_super);
357 * grab_super - acquire an active reference
358 * @s: reference we are trying to make active
360 * Tries to acquire an active reference. grab_super() is used when we
361 * had just found a superblock in super_blocks or fs_type->fs_supers
362 * and want to turn it into a full-blown active reference. grab_super()
363 * is called with sb_lock held and drops it. Returns 1 in case of
364 * success, 0 if we had failed (superblock contents was already dead or
365 * dying when grab_super() had been called). Note that this is only
366 * called for superblocks not in rundown mode (== ones still on ->fs_supers
367 * of their type), so increment of ->s_count is OK here.
369 static int grab_super(struct super_block *s) __releases(sb_lock)
372 spin_unlock(&sb_lock);
373 down_write(&s->s_umount);
374 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
378 up_write(&s->s_umount);
384 * grab_super_passive - acquire a passive reference
385 * @sb: reference we are trying to grab
387 * Tries to acquire a passive reference. This is used in places where we
388 * cannot take an active reference but we need to ensure that the
389 * superblock does not go away while we are working on it. It returns
390 * false if a reference was not gained, and returns true with the s_umount
391 * lock held in read mode if a reference is gained. On successful return,
392 * the caller must drop the s_umount lock and the passive reference when
395 bool grab_super_passive(struct super_block *sb)
398 if (hlist_unhashed(&sb->s_instances)) {
399 spin_unlock(&sb_lock);
404 spin_unlock(&sb_lock);
406 if (down_read_trylock(&sb->s_umount)) {
407 if (sb->s_root && (sb->s_flags & MS_BORN))
409 up_read(&sb->s_umount);
417 * generic_shutdown_super - common helper for ->kill_sb()
418 * @sb: superblock to kill
420 * generic_shutdown_super() does all fs-independent work on superblock
421 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
422 * that need destruction out of superblock, call generic_shutdown_super()
423 * and release aforementioned objects. Note: dentries and inodes _are_
424 * taken care of and do not need specific handling.
426 * Upon calling this function, the filesystem may no longer alter or
427 * rearrange the set of dentries belonging to this super_block, nor may it
428 * change the attachments of dentries to inodes.
430 void generic_shutdown_super(struct super_block *sb)
432 const struct super_operations *sop = sb->s_op;
435 shrink_dcache_for_umount(sb);
437 sb->s_flags &= ~MS_ACTIVE;
439 fsnotify_unmount_inodes(&sb->s_inodes);
443 if (sb->s_dio_done_wq) {
444 destroy_workqueue(sb->s_dio_done_wq);
445 sb->s_dio_done_wq = NULL;
451 if (!list_empty(&sb->s_inodes)) {
452 printk("VFS: Busy inodes after unmount of %s. "
453 "Self-destruct in 5 seconds. Have a nice day...\n",
458 /* should be initialized for __put_super_and_need_restart() */
459 hlist_del_init(&sb->s_instances);
460 spin_unlock(&sb_lock);
461 up_write(&sb->s_umount);
464 EXPORT_SYMBOL(generic_shutdown_super);
467 * sget - find or create a superblock
468 * @type: filesystem type superblock should belong to
469 * @test: comparison callback
470 * @set: setup callback
471 * @flags: mount flags
472 * @data: argument to each of them
474 struct super_block *sget(struct file_system_type *type,
475 int (*test)(struct super_block *,void *),
476 int (*set)(struct super_block *,void *),
480 struct super_block *s = NULL;
481 struct super_block *old;
487 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
488 if (!test(old, data))
490 if (!grab_super(old))
493 up_write(&s->s_umount);
501 spin_unlock(&sb_lock);
502 s = alloc_super(type, flags);
504 return ERR_PTR(-ENOMEM);
510 spin_unlock(&sb_lock);
511 up_write(&s->s_umount);
516 strlcpy(s->s_id, type->name, sizeof(s->s_id));
517 list_add_tail(&s->s_list, &super_blocks);
518 hlist_add_head(&s->s_instances, &type->fs_supers);
519 spin_unlock(&sb_lock);
520 get_filesystem(type);
521 register_shrinker(&s->s_shrink);
527 void drop_super(struct super_block *sb)
529 up_read(&sb->s_umount);
533 EXPORT_SYMBOL(drop_super);
536 * iterate_supers - call function for all active superblocks
537 * @f: function to call
538 * @arg: argument to pass to it
540 * Scans the superblock list and calls given function, passing it
541 * locked superblock and given argument.
543 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
545 struct super_block *sb, *p = NULL;
548 list_for_each_entry(sb, &super_blocks, s_list) {
549 if (hlist_unhashed(&sb->s_instances))
552 spin_unlock(&sb_lock);
554 down_read(&sb->s_umount);
555 if (sb->s_root && (sb->s_flags & MS_BORN))
557 up_read(&sb->s_umount);
566 spin_unlock(&sb_lock);
570 * iterate_supers_type - call function for superblocks of given type
572 * @f: function to call
573 * @arg: argument to pass to it
575 * Scans the superblock list and calls given function, passing it
576 * locked superblock and given argument.
578 void iterate_supers_type(struct file_system_type *type,
579 void (*f)(struct super_block *, void *), void *arg)
581 struct super_block *sb, *p = NULL;
584 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
586 spin_unlock(&sb_lock);
588 down_read(&sb->s_umount);
589 if (sb->s_root && (sb->s_flags & MS_BORN))
591 up_read(&sb->s_umount);
600 spin_unlock(&sb_lock);
603 EXPORT_SYMBOL(iterate_supers_type);
606 * get_super - get the superblock of a device
607 * @bdev: device to get the superblock for
609 * Scans the superblock list and finds the superblock of the file system
610 * mounted on the device given. %NULL is returned if no match is found.
613 struct super_block *get_super(struct block_device *bdev)
615 struct super_block *sb;
622 list_for_each_entry(sb, &super_blocks, s_list) {
623 if (hlist_unhashed(&sb->s_instances))
625 if (sb->s_bdev == bdev) {
627 spin_unlock(&sb_lock);
628 down_read(&sb->s_umount);
630 if (sb->s_root && (sb->s_flags & MS_BORN))
632 up_read(&sb->s_umount);
633 /* nope, got unmounted */
639 spin_unlock(&sb_lock);
643 EXPORT_SYMBOL(get_super);
646 * get_super_thawed - get thawed superblock of a device
647 * @bdev: device to get the superblock for
649 * Scans the superblock list and finds the superblock of the file system
650 * mounted on the device. The superblock is returned once it is thawed
651 * (or immediately if it was not frozen). %NULL is returned if no match
654 struct super_block *get_super_thawed(struct block_device *bdev)
657 struct super_block *s = get_super(bdev);
658 if (!s || s->s_writers.frozen == SB_UNFROZEN)
660 up_read(&s->s_umount);
661 wait_event(s->s_writers.wait_unfrozen,
662 s->s_writers.frozen == SB_UNFROZEN);
666 EXPORT_SYMBOL(get_super_thawed);
669 * get_active_super - get an active reference to the superblock of a device
670 * @bdev: device to get the superblock for
672 * Scans the superblock list and finds the superblock of the file system
673 * mounted on the device given. Returns the superblock with an active
674 * reference or %NULL if none was found.
676 struct super_block *get_active_super(struct block_device *bdev)
678 struct super_block *sb;
685 list_for_each_entry(sb, &super_blocks, s_list) {
686 if (hlist_unhashed(&sb->s_instances))
688 if (sb->s_bdev == bdev) {
691 up_write(&sb->s_umount);
695 spin_unlock(&sb_lock);
699 struct super_block *user_get_super(dev_t dev)
701 struct super_block *sb;
705 list_for_each_entry(sb, &super_blocks, s_list) {
706 if (hlist_unhashed(&sb->s_instances))
708 if (sb->s_dev == dev) {
710 spin_unlock(&sb_lock);
711 down_read(&sb->s_umount);
713 if (sb->s_root && (sb->s_flags & MS_BORN))
715 up_read(&sb->s_umount);
716 /* nope, got unmounted */
722 spin_unlock(&sb_lock);
727 * do_remount_sb - asks filesystem to change mount options.
728 * @sb: superblock in question
729 * @flags: numeric part of options
730 * @data: the rest of options
731 * @force: whether or not to force the change
733 * Alters the mount options of a mounted file system.
735 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
740 if (sb->s_writers.frozen != SB_UNFROZEN)
744 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
748 if (flags & MS_RDONLY)
750 shrink_dcache_sb(sb);
753 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
755 /* If we are remounting RDONLY and current sb is read/write,
756 make sure there are no rw files opened */
761 retval = sb_prepare_remount_readonly(sb);
767 if (sb->s_op->remount_fs) {
768 retval = sb->s_op->remount_fs(sb, &flags, data);
771 goto cancel_readonly;
772 /* If forced remount, go ahead despite any errors */
773 WARN(1, "forced remount of a %s fs returned %i\n",
774 sb->s_type->name, retval);
777 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
778 /* Needs to be ordered wrt mnt_is_readonly() */
780 sb->s_readonly_remount = 0;
783 * Some filesystems modify their metadata via some other path than the
784 * bdev buffer cache (eg. use a private mapping, or directories in
785 * pagecache, etc). Also file data modifications go via their own
786 * mappings. So If we try to mount readonly then copy the filesystem
787 * from bdev, we could get stale data, so invalidate it to give a best
788 * effort at coherency.
790 if (remount_ro && sb->s_bdev)
791 invalidate_bdev(sb->s_bdev);
795 sb->s_readonly_remount = 0;
799 static void do_emergency_remount(struct work_struct *work)
801 struct super_block *sb, *p = NULL;
804 list_for_each_entry(sb, &super_blocks, s_list) {
805 if (hlist_unhashed(&sb->s_instances))
808 spin_unlock(&sb_lock);
809 down_write(&sb->s_umount);
810 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
811 !(sb->s_flags & MS_RDONLY)) {
813 * What lock protects sb->s_flags??
815 do_remount_sb(sb, MS_RDONLY, NULL, 1);
817 up_write(&sb->s_umount);
825 spin_unlock(&sb_lock);
827 printk("Emergency Remount complete\n");
830 void emergency_remount(void)
832 struct work_struct *work;
834 work = kmalloc(sizeof(*work), GFP_ATOMIC);
836 INIT_WORK(work, do_emergency_remount);
842 * Unnamed block devices are dummy devices used by virtual
843 * filesystems which don't use real block-devices. -- jrs
846 static DEFINE_IDA(unnamed_dev_ida);
847 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
848 static int unnamed_dev_start = 0; /* don't bother trying below it */
850 int get_anon_bdev(dev_t *p)
856 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
858 spin_lock(&unnamed_dev_lock);
859 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
861 unnamed_dev_start = dev + 1;
862 spin_unlock(&unnamed_dev_lock);
863 if (error == -EAGAIN)
864 /* We raced and lost with another CPU. */
869 if (dev == (1 << MINORBITS)) {
870 spin_lock(&unnamed_dev_lock);
871 ida_remove(&unnamed_dev_ida, dev);
872 if (unnamed_dev_start > dev)
873 unnamed_dev_start = dev;
874 spin_unlock(&unnamed_dev_lock);
877 *p = MKDEV(0, dev & MINORMASK);
880 EXPORT_SYMBOL(get_anon_bdev);
882 void free_anon_bdev(dev_t dev)
884 int slot = MINOR(dev);
885 spin_lock(&unnamed_dev_lock);
886 ida_remove(&unnamed_dev_ida, slot);
887 if (slot < unnamed_dev_start)
888 unnamed_dev_start = slot;
889 spin_unlock(&unnamed_dev_lock);
891 EXPORT_SYMBOL(free_anon_bdev);
893 int set_anon_super(struct super_block *s, void *data)
895 int error = get_anon_bdev(&s->s_dev);
897 s->s_bdi = &noop_backing_dev_info;
901 EXPORT_SYMBOL(set_anon_super);
903 void kill_anon_super(struct super_block *sb)
905 dev_t dev = sb->s_dev;
906 generic_shutdown_super(sb);
910 EXPORT_SYMBOL(kill_anon_super);
912 void kill_litter_super(struct super_block *sb)
915 d_genocide(sb->s_root);
919 EXPORT_SYMBOL(kill_litter_super);
921 static int ns_test_super(struct super_block *sb, void *data)
923 return sb->s_fs_info == data;
926 static int ns_set_super(struct super_block *sb, void *data)
928 sb->s_fs_info = data;
929 return set_anon_super(sb, NULL);
932 struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
933 void *data, int (*fill_super)(struct super_block *, void *, int))
935 struct super_block *sb;
937 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
943 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
945 deactivate_locked_super(sb);
949 sb->s_flags |= MS_ACTIVE;
952 return dget(sb->s_root);
955 EXPORT_SYMBOL(mount_ns);
958 static int set_bdev_super(struct super_block *s, void *data)
961 s->s_dev = s->s_bdev->bd_dev;
964 * We set the bdi here to the queue backing, file systems can
965 * overwrite this in ->fill_super()
967 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
971 static int test_bdev_super(struct super_block *s, void *data)
973 return (void *)s->s_bdev == data;
976 struct dentry *mount_bdev(struct file_system_type *fs_type,
977 int flags, const char *dev_name, void *data,
978 int (*fill_super)(struct super_block *, void *, int))
980 struct block_device *bdev;
981 struct super_block *s;
982 fmode_t mode = FMODE_READ | FMODE_EXCL;
985 if (!(flags & MS_RDONLY))
988 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
990 return ERR_CAST(bdev);
993 * once the super is inserted into the list by sget, s_umount
994 * will protect the lockfs code from trying to start a snapshot
995 * while we are mounting
997 mutex_lock(&bdev->bd_fsfreeze_mutex);
998 if (bdev->bd_fsfreeze_count > 0) {
999 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1003 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
1005 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1010 if ((flags ^ s->s_flags) & MS_RDONLY) {
1011 deactivate_locked_super(s);
1017 * s_umount nests inside bd_mutex during
1018 * __invalidate_device(). blkdev_put() acquires
1019 * bd_mutex and can't be called under s_umount. Drop
1020 * s_umount temporarily. This is safe as we're
1021 * holding an active reference.
1023 up_write(&s->s_umount);
1024 blkdev_put(bdev, mode);
1025 down_write(&s->s_umount);
1027 char b[BDEVNAME_SIZE];
1030 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1031 sb_set_blocksize(s, block_size(bdev));
1032 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1034 deactivate_locked_super(s);
1038 s->s_flags |= MS_ACTIVE;
1042 return dget(s->s_root);
1047 blkdev_put(bdev, mode);
1049 return ERR_PTR(error);
1051 EXPORT_SYMBOL(mount_bdev);
1053 void kill_block_super(struct super_block *sb)
1055 struct block_device *bdev = sb->s_bdev;
1056 fmode_t mode = sb->s_mode;
1058 bdev->bd_super = NULL;
1059 generic_shutdown_super(sb);
1060 sync_blockdev(bdev);
1061 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1062 blkdev_put(bdev, mode | FMODE_EXCL);
1065 EXPORT_SYMBOL(kill_block_super);
1068 struct dentry *mount_nodev(struct file_system_type *fs_type,
1069 int flags, void *data,
1070 int (*fill_super)(struct super_block *, void *, int))
1073 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1078 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1080 deactivate_locked_super(s);
1081 return ERR_PTR(error);
1083 s->s_flags |= MS_ACTIVE;
1084 return dget(s->s_root);
1086 EXPORT_SYMBOL(mount_nodev);
1088 static int compare_single(struct super_block *s, void *p)
1093 struct dentry *mount_single(struct file_system_type *fs_type,
1094 int flags, void *data,
1095 int (*fill_super)(struct super_block *, void *, int))
1097 struct super_block *s;
1100 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1104 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1106 deactivate_locked_super(s);
1107 return ERR_PTR(error);
1109 s->s_flags |= MS_ACTIVE;
1111 do_remount_sb(s, flags, data, 0);
1113 return dget(s->s_root);
1115 EXPORT_SYMBOL(mount_single);
1118 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1120 struct dentry *root;
1121 struct super_block *sb;
1122 char *secdata = NULL;
1123 int error = -ENOMEM;
1125 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1126 secdata = alloc_secdata();
1130 error = security_sb_copy_data(data, secdata);
1132 goto out_free_secdata;
1135 root = type->mount(type, flags, name, data);
1137 error = PTR_ERR(root);
1138 goto out_free_secdata;
1142 WARN_ON(!sb->s_bdi);
1143 WARN_ON(sb->s_bdi == &default_backing_dev_info);
1144 sb->s_flags |= MS_BORN;
1146 error = security_sb_kern_mount(sb, flags, secdata);
1151 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1152 * but s_maxbytes was an unsigned long long for many releases. Throw
1153 * this warning for a little while to try and catch filesystems that
1154 * violate this rule.
1156 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1157 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1159 up_write(&sb->s_umount);
1160 free_secdata(secdata);
1164 deactivate_locked_super(sb);
1166 free_secdata(secdata);
1168 return ERR_PTR(error);
1172 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1175 void __sb_end_write(struct super_block *sb, int level)
1177 percpu_counter_dec(&sb->s_writers.counter[level-1]);
1179 * Make sure s_writers are updated before we wake up waiters in
1183 if (waitqueue_active(&sb->s_writers.wait))
1184 wake_up(&sb->s_writers.wait);
1185 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _RET_IP_);
1187 EXPORT_SYMBOL(__sb_end_write);
1189 #ifdef CONFIG_LOCKDEP
1191 * We want lockdep to tell us about possible deadlocks with freezing but
1192 * it's it bit tricky to properly instrument it. Getting a freeze protection
1193 * works as getting a read lock but there are subtle problems. XFS for example
1194 * gets freeze protection on internal level twice in some cases, which is OK
1195 * only because we already hold a freeze protection also on higher level. Due
1196 * to these cases we have to tell lockdep we are doing trylock when we
1197 * already hold a freeze protection for a higher freeze level.
1199 static void acquire_freeze_lock(struct super_block *sb, int level, bool trylock,
1205 for (i = 0; i < level - 1; i++)
1206 if (lock_is_held(&sb->s_writers.lock_map[i])) {
1211 rwsem_acquire_read(&sb->s_writers.lock_map[level-1], 0, trylock, ip);
1216 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1219 int __sb_start_write(struct super_block *sb, int level, bool wait)
1222 if (unlikely(sb->s_writers.frozen >= level)) {
1225 wait_event(sb->s_writers.wait_unfrozen,
1226 sb->s_writers.frozen < level);
1229 #ifdef CONFIG_LOCKDEP
1230 acquire_freeze_lock(sb, level, !wait, _RET_IP_);
1232 percpu_counter_inc(&sb->s_writers.counter[level-1]);
1234 * Make sure counter is updated before we check for frozen.
1235 * freeze_super() first sets frozen and then checks the counter.
1238 if (unlikely(sb->s_writers.frozen >= level)) {
1239 __sb_end_write(sb, level);
1244 EXPORT_SYMBOL(__sb_start_write);
1247 * sb_wait_write - wait until all writers to given file system finish
1248 * @sb: the super for which we wait
1249 * @level: type of writers we wait for (normal vs page fault)
1251 * This function waits until there are no writers of given type to given file
1252 * system. Caller of this function should make sure there can be no new writers
1253 * of type @level before calling this function. Otherwise this function can
1256 static void sb_wait_write(struct super_block *sb, int level)
1261 * We just cycle-through lockdep here so that it does not complain
1262 * about returning with lock to userspace
1264 rwsem_acquire(&sb->s_writers.lock_map[level-1], 0, 0, _THIS_IP_);
1265 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _THIS_IP_);
1271 * We use a barrier in prepare_to_wait() to separate setting
1272 * of frozen and checking of the counter
1274 prepare_to_wait(&sb->s_writers.wait, &wait,
1275 TASK_UNINTERRUPTIBLE);
1277 writers = percpu_counter_sum(&sb->s_writers.counter[level-1]);
1281 finish_wait(&sb->s_writers.wait, &wait);
1286 * freeze_super - lock the filesystem and force it into a consistent state
1287 * @sb: the super to lock
1289 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1290 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1293 * During this function, sb->s_writers.frozen goes through these values:
1295 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1297 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1298 * writes should be blocked, though page faults are still allowed. We wait for
1299 * all writes to complete and then proceed to the next stage.
1301 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1302 * but internal fs threads can still modify the filesystem (although they
1303 * should not dirty new pages or inodes), writeback can run etc. After waiting
1304 * for all running page faults we sync the filesystem which will clean all
1305 * dirty pages and inodes (no new dirty pages or inodes can be created when
1308 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1309 * modification are blocked (e.g. XFS preallocation truncation on inode
1310 * reclaim). This is usually implemented by blocking new transactions for
1311 * filesystems that have them and need this additional guard. After all
1312 * internal writers are finished we call ->freeze_fs() to finish filesystem
1313 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1314 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1316 * sb->s_writers.frozen is protected by sb->s_umount.
1318 int freeze_super(struct super_block *sb)
1322 atomic_inc(&sb->s_active);
1323 down_write(&sb->s_umount);
1324 if (sb->s_writers.frozen != SB_UNFROZEN) {
1325 deactivate_locked_super(sb);
1329 if (!(sb->s_flags & MS_BORN)) {
1330 up_write(&sb->s_umount);
1331 return 0; /* sic - it's "nothing to do" */
1334 if (sb->s_flags & MS_RDONLY) {
1335 /* Nothing to do really... */
1336 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1337 up_write(&sb->s_umount);
1341 /* From now on, no new normal writers can start */
1342 sb->s_writers.frozen = SB_FREEZE_WRITE;
1345 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1346 up_write(&sb->s_umount);
1348 sb_wait_write(sb, SB_FREEZE_WRITE);
1350 /* Now we go and block page faults... */
1351 down_write(&sb->s_umount);
1352 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1355 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1357 /* All writers are done so after syncing there won't be dirty data */
1358 sync_filesystem(sb);
1360 /* Now wait for internal filesystem counter */
1361 sb->s_writers.frozen = SB_FREEZE_FS;
1363 sb_wait_write(sb, SB_FREEZE_FS);
1365 if (sb->s_op->freeze_fs) {
1366 ret = sb->s_op->freeze_fs(sb);
1369 "VFS:Filesystem freeze failed\n");
1370 sb->s_writers.frozen = SB_UNFROZEN;
1372 wake_up(&sb->s_writers.wait_unfrozen);
1373 deactivate_locked_super(sb);
1378 * This is just for debugging purposes so that fs can warn if it
1379 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1381 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1382 up_write(&sb->s_umount);
1385 EXPORT_SYMBOL(freeze_super);
1388 * thaw_super -- unlock filesystem
1389 * @sb: the super to thaw
1391 * Unlocks the filesystem and marks it writeable again after freeze_super().
1393 int thaw_super(struct super_block *sb)
1397 down_write(&sb->s_umount);
1398 if (sb->s_writers.frozen == SB_UNFROZEN) {
1399 up_write(&sb->s_umount);
1403 if (sb->s_flags & MS_RDONLY)
1406 if (sb->s_op->unfreeze_fs) {
1407 error = sb->s_op->unfreeze_fs(sb);
1410 "VFS:Filesystem thaw failed\n");
1411 up_write(&sb->s_umount);
1417 sb->s_writers.frozen = SB_UNFROZEN;
1419 wake_up(&sb->s_writers.wait_unfrozen);
1420 deactivate_locked_super(sb);
1424 EXPORT_SYMBOL(thaw_super);