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/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>
39 LIST_HEAD(super_blocks);
40 DEFINE_SPINLOCK(sb_lock);
42 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
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;
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))
74 if (!grab_super_passive(sb))
77 if (sb->s_op->nr_cached_objects)
78 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
80 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
81 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
82 total_objects = dentries + inodes + fs_objects + 1;
86 /* proportion the scan between the caches */
87 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
88 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
89 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
92 * prune the dcache first as the icache is pinned by it, then
93 * prune the icache, followed by the filesystem specific caches
95 sc->nr_to_scan = dentries;
96 freed = prune_dcache_sb(sb, sc);
97 sc->nr_to_scan = inodes;
98 freed += prune_icache_sb(sb, sc);
101 sc->nr_to_scan = fs_objects;
102 freed += sb->s_op->free_cached_objects(sb, sc);
109 static unsigned long super_cache_count(struct shrinker *shrink,
110 struct shrink_control *sc)
112 struct super_block *sb;
113 long total_objects = 0;
115 sb = container_of(shrink, struct super_block, s_shrink);
118 * Don't call grab_super_passive as it is a potential
119 * scalability bottleneck. The counts could get updated
120 * between super_cache_count and super_cache_scan anyway.
121 * Call to super_cache_count with shrinker_rwsem held
122 * ensures the safety of call to list_lru_shrink_count() and
123 * s_op->nr_cached_objects().
125 if (sb->s_op && sb->s_op->nr_cached_objects)
126 total_objects = sb->s_op->nr_cached_objects(sb, sc);
128 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
129 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
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)
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]);
149 WARN_ON(!list_empty(&s->s_mounts));
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;
172 INIT_LIST_HEAD(&s->s_mounts);
174 if (security_sb_alloc(s))
177 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
178 if (percpu_counter_init(&s->s_writers.counter[i], 0,
181 lockdep_init_map(&s->s_writers.lock_map[i], sb_writers_name[i],
182 &type->s_writers_key[i], 0);
184 init_waitqueue_head(&s->s_writers.wait);
185 init_waitqueue_head(&s->s_writers.wait_unfrozen);
187 s->s_bdi = &default_backing_dev_info;
188 INIT_HLIST_NODE(&s->s_instances);
189 INIT_HLIST_BL_HEAD(&s->s_anon);
190 INIT_LIST_HEAD(&s->s_inodes);
192 if (list_lru_init_memcg(&s->s_dentry_lru))
194 if (list_lru_init_memcg(&s->s_inode_lru))
197 init_rwsem(&s->s_umount);
198 lockdep_set_class(&s->s_umount, &type->s_umount_key);
200 * sget() can have s_umount recursion.
202 * When it cannot find a suitable sb, it allocates a new
203 * one (this one), and tries again to find a suitable old
206 * In case that succeeds, it will acquire the s_umount
207 * lock of the old one. Since these are clearly distrinct
208 * locks, and this object isn't exposed yet, there's no
211 * Annotate this by putting this lock in a different
214 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
216 atomic_set(&s->s_active, 1);
217 mutex_init(&s->s_vfs_rename_mutex);
218 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
219 mutex_init(&s->s_dquot.dqio_mutex);
220 mutex_init(&s->s_dquot.dqonoff_mutex);
221 s->s_maxbytes = MAX_NON_LFS;
222 s->s_op = &default_op;
223 s->s_time_gran = 1000000000;
224 s->cleancache_poolid = -1;
226 s->s_shrink.seeks = DEFAULT_SEEKS;
227 s->s_shrink.scan_objects = super_cache_scan;
228 s->s_shrink.count_objects = super_cache_count;
229 s->s_shrink.batch = 1024;
230 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
238 /* Superblock refcounting */
241 * Drop a superblock's refcount. The caller must hold sb_lock.
243 static void __put_super(struct super_block *sb)
245 if (!--sb->s_count) {
246 list_del_init(&sb->s_list);
252 * put_super - drop a temporary reference to superblock
253 * @sb: superblock in question
255 * Drops a temporary reference, frees superblock if there's no
258 static void put_super(struct super_block *sb)
262 spin_unlock(&sb_lock);
267 * deactivate_locked_super - drop an active reference to superblock
268 * @s: superblock to deactivate
270 * Drops an active reference to superblock, converting it into a temprory
271 * one if there is no other active references left. In that case we
272 * tell fs driver to shut it down and drop the temporary reference we
275 * Caller holds exclusive lock on superblock; that lock is released.
277 void deactivate_locked_super(struct super_block *s)
279 struct file_system_type *fs = s->s_type;
280 if (atomic_dec_and_test(&s->s_active)) {
281 cleancache_invalidate_fs(s);
282 unregister_shrinker(&s->s_shrink);
286 * Since list_lru_destroy() may sleep, we cannot call it from
287 * put_super(), where we hold the sb_lock. Therefore we destroy
288 * the lru lists right now.
290 list_lru_destroy(&s->s_dentry_lru);
291 list_lru_destroy(&s->s_inode_lru);
296 up_write(&s->s_umount);
300 EXPORT_SYMBOL(deactivate_locked_super);
303 * deactivate_super - drop an active reference to superblock
304 * @s: superblock to deactivate
306 * Variant of deactivate_locked_super(), except that superblock is *not*
307 * locked by caller. If we are going to drop the final active reference,
308 * lock will be acquired prior to that.
310 void deactivate_super(struct super_block *s)
312 if (!atomic_add_unless(&s->s_active, -1, 1)) {
313 down_write(&s->s_umount);
314 deactivate_locked_super(s);
318 EXPORT_SYMBOL(deactivate_super);
321 * grab_super - acquire an active reference
322 * @s: reference we are trying to make active
324 * Tries to acquire an active reference. grab_super() is used when we
325 * had just found a superblock in super_blocks or fs_type->fs_supers
326 * and want to turn it into a full-blown active reference. grab_super()
327 * is called with sb_lock held and drops it. Returns 1 in case of
328 * success, 0 if we had failed (superblock contents was already dead or
329 * dying when grab_super() had been called). Note that this is only
330 * called for superblocks not in rundown mode (== ones still on ->fs_supers
331 * of their type), so increment of ->s_count is OK here.
333 static int grab_super(struct super_block *s) __releases(sb_lock)
336 spin_unlock(&sb_lock);
337 down_write(&s->s_umount);
338 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
342 up_write(&s->s_umount);
348 * grab_super_passive - acquire a passive reference
349 * @sb: reference we are trying to grab
351 * Tries to acquire a passive reference. This is used in places where we
352 * cannot take an active reference but we need to ensure that the
353 * superblock does not go away while we are working on it. It returns
354 * false if a reference was not gained, and returns true with the s_umount
355 * lock held in read mode if a reference is gained. On successful return,
356 * the caller must drop the s_umount lock and the passive reference when
359 bool grab_super_passive(struct super_block *sb)
362 if (hlist_unhashed(&sb->s_instances)) {
363 spin_unlock(&sb_lock);
368 spin_unlock(&sb_lock);
370 if (down_read_trylock(&sb->s_umount)) {
371 if (sb->s_root && (sb->s_flags & MS_BORN))
373 up_read(&sb->s_umount);
381 * generic_shutdown_super - common helper for ->kill_sb()
382 * @sb: superblock to kill
384 * generic_shutdown_super() does all fs-independent work on superblock
385 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
386 * that need destruction out of superblock, call generic_shutdown_super()
387 * and release aforementioned objects. Note: dentries and inodes _are_
388 * taken care of and do not need specific handling.
390 * Upon calling this function, the filesystem may no longer alter or
391 * rearrange the set of dentries belonging to this super_block, nor may it
392 * change the attachments of dentries to inodes.
394 void generic_shutdown_super(struct super_block *sb)
396 const struct super_operations *sop = sb->s_op;
399 shrink_dcache_for_umount(sb);
401 sb->s_flags &= ~MS_ACTIVE;
403 fsnotify_unmount_inodes(&sb->s_inodes);
407 if (sb->s_dio_done_wq) {
408 destroy_workqueue(sb->s_dio_done_wq);
409 sb->s_dio_done_wq = NULL;
415 if (!list_empty(&sb->s_inodes)) {
416 printk("VFS: Busy inodes after unmount of %s. "
417 "Self-destruct in 5 seconds. Have a nice day...\n",
422 /* should be initialized for __put_super_and_need_restart() */
423 hlist_del_init(&sb->s_instances);
424 spin_unlock(&sb_lock);
425 up_write(&sb->s_umount);
428 EXPORT_SYMBOL(generic_shutdown_super);
431 * sget - find or create a superblock
432 * @type: filesystem type superblock should belong to
433 * @test: comparison callback
434 * @set: setup callback
435 * @flags: mount flags
436 * @data: argument to each of them
438 struct super_block *sget(struct file_system_type *type,
439 int (*test)(struct super_block *,void *),
440 int (*set)(struct super_block *,void *),
444 struct super_block *s = NULL;
445 struct super_block *old;
451 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
452 if (!test(old, data))
454 if (!grab_super(old))
457 up_write(&s->s_umount);
465 spin_unlock(&sb_lock);
466 s = alloc_super(type, flags);
468 return ERR_PTR(-ENOMEM);
474 spin_unlock(&sb_lock);
475 up_write(&s->s_umount);
480 strlcpy(s->s_id, type->name, sizeof(s->s_id));
481 list_add_tail(&s->s_list, &super_blocks);
482 hlist_add_head(&s->s_instances, &type->fs_supers);
483 spin_unlock(&sb_lock);
484 get_filesystem(type);
485 register_shrinker(&s->s_shrink);
491 void drop_super(struct super_block *sb)
493 up_read(&sb->s_umount);
497 EXPORT_SYMBOL(drop_super);
500 * iterate_supers - call function for all active superblocks
501 * @f: function to call
502 * @arg: argument to pass to it
504 * Scans the superblock list and calls given function, passing it
505 * locked superblock and given argument.
507 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
509 struct super_block *sb, *p = NULL;
512 list_for_each_entry(sb, &super_blocks, s_list) {
513 if (hlist_unhashed(&sb->s_instances))
516 spin_unlock(&sb_lock);
518 down_read(&sb->s_umount);
519 if (sb->s_root && (sb->s_flags & MS_BORN))
521 up_read(&sb->s_umount);
530 spin_unlock(&sb_lock);
534 * iterate_supers_type - call function for superblocks of given type
536 * @f: function to call
537 * @arg: argument to pass to it
539 * Scans the superblock list and calls given function, passing it
540 * locked superblock and given argument.
542 void iterate_supers_type(struct file_system_type *type,
543 void (*f)(struct super_block *, void *), void *arg)
545 struct super_block *sb, *p = NULL;
548 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
550 spin_unlock(&sb_lock);
552 down_read(&sb->s_umount);
553 if (sb->s_root && (sb->s_flags & MS_BORN))
555 up_read(&sb->s_umount);
564 spin_unlock(&sb_lock);
567 EXPORT_SYMBOL(iterate_supers_type);
570 * get_super - get the superblock of a device
571 * @bdev: device to get the superblock for
573 * Scans the superblock list and finds the superblock of the file system
574 * mounted on the device given. %NULL is returned if no match is found.
577 struct super_block *get_super(struct block_device *bdev)
579 struct super_block *sb;
586 list_for_each_entry(sb, &super_blocks, s_list) {
587 if (hlist_unhashed(&sb->s_instances))
589 if (sb->s_bdev == bdev) {
591 spin_unlock(&sb_lock);
592 down_read(&sb->s_umount);
594 if (sb->s_root && (sb->s_flags & MS_BORN))
596 up_read(&sb->s_umount);
597 /* nope, got unmounted */
603 spin_unlock(&sb_lock);
607 EXPORT_SYMBOL(get_super);
610 * get_super_thawed - get thawed superblock of a device
611 * @bdev: device to get the superblock for
613 * Scans the superblock list and finds the superblock of the file system
614 * mounted on the device. The superblock is returned once it is thawed
615 * (or immediately if it was not frozen). %NULL is returned if no match
618 struct super_block *get_super_thawed(struct block_device *bdev)
621 struct super_block *s = get_super(bdev);
622 if (!s || s->s_writers.frozen == SB_UNFROZEN)
624 up_read(&s->s_umount);
625 wait_event(s->s_writers.wait_unfrozen,
626 s->s_writers.frozen == SB_UNFROZEN);
630 EXPORT_SYMBOL(get_super_thawed);
633 * get_active_super - get an active reference to the superblock of a device
634 * @bdev: device to get the superblock for
636 * Scans the superblock list and finds the superblock of the file system
637 * mounted on the device given. Returns the superblock with an active
638 * reference or %NULL if none was found.
640 struct super_block *get_active_super(struct block_device *bdev)
642 struct super_block *sb;
649 list_for_each_entry(sb, &super_blocks, s_list) {
650 if (hlist_unhashed(&sb->s_instances))
652 if (sb->s_bdev == bdev) {
655 up_write(&sb->s_umount);
659 spin_unlock(&sb_lock);
663 struct super_block *user_get_super(dev_t dev)
665 struct super_block *sb;
669 list_for_each_entry(sb, &super_blocks, s_list) {
670 if (hlist_unhashed(&sb->s_instances))
672 if (sb->s_dev == dev) {
674 spin_unlock(&sb_lock);
675 down_read(&sb->s_umount);
677 if (sb->s_root && (sb->s_flags & MS_BORN))
679 up_read(&sb->s_umount);
680 /* nope, got unmounted */
686 spin_unlock(&sb_lock);
691 * do_remount_sb - asks filesystem to change mount options.
692 * @sb: superblock in question
693 * @flags: numeric part of options
694 * @data: the rest of options
695 * @force: whether or not to force the change
697 * Alters the mount options of a mounted file system.
699 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
704 if (sb->s_writers.frozen != SB_UNFROZEN)
708 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
712 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
715 if (sb->s_pins.first) {
716 up_write(&sb->s_umount);
718 down_write(&sb->s_umount);
721 if (sb->s_writers.frozen != SB_UNFROZEN)
723 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
726 shrink_dcache_sb(sb);
728 /* If we are remounting RDONLY and current sb is read/write,
729 make sure there are no rw files opened */
732 sb->s_readonly_remount = 1;
735 retval = sb_prepare_remount_readonly(sb);
741 if (sb->s_op->remount_fs) {
742 retval = sb->s_op->remount_fs(sb, &flags, data);
745 goto cancel_readonly;
746 /* If forced remount, go ahead despite any errors */
747 WARN(1, "forced remount of a %s fs returned %i\n",
748 sb->s_type->name, retval);
751 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
752 /* Needs to be ordered wrt mnt_is_readonly() */
754 sb->s_readonly_remount = 0;
757 * Some filesystems modify their metadata via some other path than the
758 * bdev buffer cache (eg. use a private mapping, or directories in
759 * pagecache, etc). Also file data modifications go via their own
760 * mappings. So If we try to mount readonly then copy the filesystem
761 * from bdev, we could get stale data, so invalidate it to give a best
762 * effort at coherency.
764 if (remount_ro && sb->s_bdev)
765 invalidate_bdev(sb->s_bdev);
769 sb->s_readonly_remount = 0;
773 static void do_emergency_remount(struct work_struct *work)
775 struct super_block *sb, *p = NULL;
778 list_for_each_entry(sb, &super_blocks, s_list) {
779 if (hlist_unhashed(&sb->s_instances))
782 spin_unlock(&sb_lock);
783 down_write(&sb->s_umount);
784 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
785 !(sb->s_flags & MS_RDONLY)) {
787 * What lock protects sb->s_flags??
789 do_remount_sb(sb, MS_RDONLY, NULL, 1);
791 up_write(&sb->s_umount);
799 spin_unlock(&sb_lock);
801 printk("Emergency Remount complete\n");
804 void emergency_remount(void)
806 struct work_struct *work;
808 work = kmalloc(sizeof(*work), GFP_ATOMIC);
810 INIT_WORK(work, do_emergency_remount);
816 * Unnamed block devices are dummy devices used by virtual
817 * filesystems which don't use real block-devices. -- jrs
820 static DEFINE_IDA(unnamed_dev_ida);
821 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
822 /* Many userspace utilities consider an FSID of 0 invalid.
823 * Always return at least 1 from get_anon_bdev.
825 static int unnamed_dev_start = 1;
827 int get_anon_bdev(dev_t *p)
833 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
835 spin_lock(&unnamed_dev_lock);
836 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
838 unnamed_dev_start = dev + 1;
839 spin_unlock(&unnamed_dev_lock);
840 if (error == -EAGAIN)
841 /* We raced and lost with another CPU. */
846 if (dev == (1 << MINORBITS)) {
847 spin_lock(&unnamed_dev_lock);
848 ida_remove(&unnamed_dev_ida, dev);
849 if (unnamed_dev_start > dev)
850 unnamed_dev_start = dev;
851 spin_unlock(&unnamed_dev_lock);
854 *p = MKDEV(0, dev & MINORMASK);
857 EXPORT_SYMBOL(get_anon_bdev);
859 void free_anon_bdev(dev_t dev)
861 int slot = MINOR(dev);
862 spin_lock(&unnamed_dev_lock);
863 ida_remove(&unnamed_dev_ida, slot);
864 if (slot < unnamed_dev_start)
865 unnamed_dev_start = slot;
866 spin_unlock(&unnamed_dev_lock);
868 EXPORT_SYMBOL(free_anon_bdev);
870 int set_anon_super(struct super_block *s, void *data)
872 int error = get_anon_bdev(&s->s_dev);
874 s->s_bdi = &noop_backing_dev_info;
878 EXPORT_SYMBOL(set_anon_super);
880 void kill_anon_super(struct super_block *sb)
882 dev_t dev = sb->s_dev;
883 generic_shutdown_super(sb);
887 EXPORT_SYMBOL(kill_anon_super);
889 void kill_litter_super(struct super_block *sb)
892 d_genocide(sb->s_root);
896 EXPORT_SYMBOL(kill_litter_super);
898 static int ns_test_super(struct super_block *sb, void *data)
900 return sb->s_fs_info == data;
903 static int ns_set_super(struct super_block *sb, void *data)
905 sb->s_fs_info = data;
906 return set_anon_super(sb, NULL);
909 struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
910 void *data, int (*fill_super)(struct super_block *, void *, int))
912 struct super_block *sb;
914 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
920 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
922 deactivate_locked_super(sb);
926 sb->s_flags |= MS_ACTIVE;
929 return dget(sb->s_root);
932 EXPORT_SYMBOL(mount_ns);
935 static int set_bdev_super(struct super_block *s, void *data)
938 s->s_dev = s->s_bdev->bd_dev;
941 * We set the bdi here to the queue backing, file systems can
942 * overwrite this in ->fill_super()
944 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
948 static int test_bdev_super(struct super_block *s, void *data)
950 return (void *)s->s_bdev == data;
953 struct dentry *mount_bdev(struct file_system_type *fs_type,
954 int flags, const char *dev_name, void *data,
955 int (*fill_super)(struct super_block *, void *, int))
957 struct block_device *bdev;
958 struct super_block *s;
959 fmode_t mode = FMODE_READ | FMODE_EXCL;
962 if (!(flags & MS_RDONLY))
965 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
967 return ERR_CAST(bdev);
970 * once the super is inserted into the list by sget, s_umount
971 * will protect the lockfs code from trying to start a snapshot
972 * while we are mounting
974 mutex_lock(&bdev->bd_fsfreeze_mutex);
975 if (bdev->bd_fsfreeze_count > 0) {
976 mutex_unlock(&bdev->bd_fsfreeze_mutex);
980 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
982 mutex_unlock(&bdev->bd_fsfreeze_mutex);
987 if ((flags ^ s->s_flags) & MS_RDONLY) {
988 deactivate_locked_super(s);
994 * s_umount nests inside bd_mutex during
995 * __invalidate_device(). blkdev_put() acquires
996 * bd_mutex and can't be called under s_umount. Drop
997 * s_umount temporarily. This is safe as we're
998 * holding an active reference.
1000 up_write(&s->s_umount);
1001 blkdev_put(bdev, mode);
1002 down_write(&s->s_umount);
1004 char b[BDEVNAME_SIZE];
1007 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1008 sb_set_blocksize(s, block_size(bdev));
1009 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1011 deactivate_locked_super(s);
1015 s->s_flags |= MS_ACTIVE;
1019 return dget(s->s_root);
1024 blkdev_put(bdev, mode);
1026 return ERR_PTR(error);
1028 EXPORT_SYMBOL(mount_bdev);
1030 void kill_block_super(struct super_block *sb)
1032 struct block_device *bdev = sb->s_bdev;
1033 fmode_t mode = sb->s_mode;
1035 bdev->bd_super = NULL;
1036 generic_shutdown_super(sb);
1037 sync_blockdev(bdev);
1038 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1039 blkdev_put(bdev, mode | FMODE_EXCL);
1042 EXPORT_SYMBOL(kill_block_super);
1045 struct dentry *mount_nodev(struct file_system_type *fs_type,
1046 int flags, void *data,
1047 int (*fill_super)(struct super_block *, void *, int))
1050 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1055 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1057 deactivate_locked_super(s);
1058 return ERR_PTR(error);
1060 s->s_flags |= MS_ACTIVE;
1061 return dget(s->s_root);
1063 EXPORT_SYMBOL(mount_nodev);
1065 static int compare_single(struct super_block *s, void *p)
1070 struct dentry *mount_single(struct file_system_type *fs_type,
1071 int flags, void *data,
1072 int (*fill_super)(struct super_block *, void *, int))
1074 struct super_block *s;
1077 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1081 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1083 deactivate_locked_super(s);
1084 return ERR_PTR(error);
1086 s->s_flags |= MS_ACTIVE;
1088 do_remount_sb(s, flags, data, 0);
1090 return dget(s->s_root);
1092 EXPORT_SYMBOL(mount_single);
1095 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1097 struct dentry *root;
1098 struct super_block *sb;
1099 char *secdata = NULL;
1100 int error = -ENOMEM;
1102 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1103 secdata = alloc_secdata();
1107 error = security_sb_copy_data(data, secdata);
1109 goto out_free_secdata;
1112 root = type->mount(type, flags, name, data);
1114 error = PTR_ERR(root);
1115 goto out_free_secdata;
1119 WARN_ON(!sb->s_bdi);
1120 WARN_ON(sb->s_bdi == &default_backing_dev_info);
1121 sb->s_flags |= MS_BORN;
1123 error = security_sb_kern_mount(sb, flags, secdata);
1128 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1129 * but s_maxbytes was an unsigned long long for many releases. Throw
1130 * this warning for a little while to try and catch filesystems that
1131 * violate this rule.
1133 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1134 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1136 up_write(&sb->s_umount);
1137 free_secdata(secdata);
1141 deactivate_locked_super(sb);
1143 free_secdata(secdata);
1145 return ERR_PTR(error);
1149 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1152 void __sb_end_write(struct super_block *sb, int level)
1154 percpu_counter_dec(&sb->s_writers.counter[level-1]);
1156 * Make sure s_writers are updated before we wake up waiters in
1160 if (waitqueue_active(&sb->s_writers.wait))
1161 wake_up(&sb->s_writers.wait);
1162 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _RET_IP_);
1164 EXPORT_SYMBOL(__sb_end_write);
1166 #ifdef CONFIG_LOCKDEP
1168 * We want lockdep to tell us about possible deadlocks with freezing but
1169 * it's it bit tricky to properly instrument it. Getting a freeze protection
1170 * works as getting a read lock but there are subtle problems. XFS for example
1171 * gets freeze protection on internal level twice in some cases, which is OK
1172 * only because we already hold a freeze protection also on higher level. Due
1173 * to these cases we have to tell lockdep we are doing trylock when we
1174 * already hold a freeze protection for a higher freeze level.
1176 static void acquire_freeze_lock(struct super_block *sb, int level, bool trylock,
1182 for (i = 0; i < level - 1; i++)
1183 if (lock_is_held(&sb->s_writers.lock_map[i])) {
1188 rwsem_acquire_read(&sb->s_writers.lock_map[level-1], 0, trylock, ip);
1193 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1196 int __sb_start_write(struct super_block *sb, int level, bool wait)
1199 if (unlikely(sb->s_writers.frozen >= level)) {
1202 wait_event(sb->s_writers.wait_unfrozen,
1203 sb->s_writers.frozen < level);
1206 #ifdef CONFIG_LOCKDEP
1207 acquire_freeze_lock(sb, level, !wait, _RET_IP_);
1209 percpu_counter_inc(&sb->s_writers.counter[level-1]);
1211 * Make sure counter is updated before we check for frozen.
1212 * freeze_super() first sets frozen and then checks the counter.
1215 if (unlikely(sb->s_writers.frozen >= level)) {
1216 __sb_end_write(sb, level);
1221 EXPORT_SYMBOL(__sb_start_write);
1224 * sb_wait_write - wait until all writers to given file system finish
1225 * @sb: the super for which we wait
1226 * @level: type of writers we wait for (normal vs page fault)
1228 * This function waits until there are no writers of given type to given file
1229 * system. Caller of this function should make sure there can be no new writers
1230 * of type @level before calling this function. Otherwise this function can
1233 static void sb_wait_write(struct super_block *sb, int level)
1238 * We just cycle-through lockdep here so that it does not complain
1239 * about returning with lock to userspace
1241 rwsem_acquire(&sb->s_writers.lock_map[level-1], 0, 0, _THIS_IP_);
1242 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _THIS_IP_);
1248 * We use a barrier in prepare_to_wait() to separate setting
1249 * of frozen and checking of the counter
1251 prepare_to_wait(&sb->s_writers.wait, &wait,
1252 TASK_UNINTERRUPTIBLE);
1254 writers = percpu_counter_sum(&sb->s_writers.counter[level-1]);
1258 finish_wait(&sb->s_writers.wait, &wait);
1263 * freeze_super - lock the filesystem and force it into a consistent state
1264 * @sb: the super to lock
1266 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1267 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1270 * During this function, sb->s_writers.frozen goes through these values:
1272 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1274 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1275 * writes should be blocked, though page faults are still allowed. We wait for
1276 * all writes to complete and then proceed to the next stage.
1278 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1279 * but internal fs threads can still modify the filesystem (although they
1280 * should not dirty new pages or inodes), writeback can run etc. After waiting
1281 * for all running page faults we sync the filesystem which will clean all
1282 * dirty pages and inodes (no new dirty pages or inodes can be created when
1285 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1286 * modification are blocked (e.g. XFS preallocation truncation on inode
1287 * reclaim). This is usually implemented by blocking new transactions for
1288 * filesystems that have them and need this additional guard. After all
1289 * internal writers are finished we call ->freeze_fs() to finish filesystem
1290 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1291 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1293 * sb->s_writers.frozen is protected by sb->s_umount.
1295 int freeze_super(struct super_block *sb)
1299 atomic_inc(&sb->s_active);
1300 down_write(&sb->s_umount);
1301 if (sb->s_writers.frozen != SB_UNFROZEN) {
1302 deactivate_locked_super(sb);
1306 if (!(sb->s_flags & MS_BORN)) {
1307 up_write(&sb->s_umount);
1308 return 0; /* sic - it's "nothing to do" */
1311 if (sb->s_flags & MS_RDONLY) {
1312 /* Nothing to do really... */
1313 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1314 up_write(&sb->s_umount);
1318 /* From now on, no new normal writers can start */
1319 sb->s_writers.frozen = SB_FREEZE_WRITE;
1322 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1323 up_write(&sb->s_umount);
1325 sb_wait_write(sb, SB_FREEZE_WRITE);
1327 /* Now we go and block page faults... */
1328 down_write(&sb->s_umount);
1329 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1332 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1334 /* All writers are done so after syncing there won't be dirty data */
1335 sync_filesystem(sb);
1337 /* Now wait for internal filesystem counter */
1338 sb->s_writers.frozen = SB_FREEZE_FS;
1340 sb_wait_write(sb, SB_FREEZE_FS);
1342 if (sb->s_op->freeze_fs) {
1343 ret = sb->s_op->freeze_fs(sb);
1346 "VFS:Filesystem freeze failed\n");
1347 sb->s_writers.frozen = SB_UNFROZEN;
1349 wake_up(&sb->s_writers.wait_unfrozen);
1350 deactivate_locked_super(sb);
1355 * This is just for debugging purposes so that fs can warn if it
1356 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1358 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1359 up_write(&sb->s_umount);
1362 EXPORT_SYMBOL(freeze_super);
1365 * thaw_super -- unlock filesystem
1366 * @sb: the super to thaw
1368 * Unlocks the filesystem and marks it writeable again after freeze_super().
1370 int thaw_super(struct super_block *sb)
1374 down_write(&sb->s_umount);
1375 if (sb->s_writers.frozen == SB_UNFROZEN) {
1376 up_write(&sb->s_umount);
1380 if (sb->s_flags & MS_RDONLY)
1383 if (sb->s_op->unfreeze_fs) {
1384 error = sb->s_op->unfreeze_fs(sb);
1387 "VFS:Filesystem thaw failed\n");
1388 up_write(&sb->s_umount);
1394 sb->s_writers.frozen = SB_UNFROZEN;
1396 wake_up(&sb->s_writers.wait_unfrozen);
1397 deactivate_locked_super(sb);
1401 EXPORT_SYMBOL(thaw_super);