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 static LIST_HEAD(super_blocks);
40 static 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 (!trylock_super(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 * Ensure that we always scan at least one object - memcg kmem
96 * accounting uses this to fully empty the caches.
98 sc->nr_to_scan = dentries + 1;
99 freed = prune_dcache_sb(sb, sc);
100 sc->nr_to_scan = inodes + 1;
101 freed += prune_icache_sb(sb, sc);
104 sc->nr_to_scan = fs_objects + 1;
105 freed += sb->s_op->free_cached_objects(sb, sc);
108 up_read(&sb->s_umount);
112 static unsigned long super_cache_count(struct shrinker *shrink,
113 struct shrink_control *sc)
115 struct super_block *sb;
116 long total_objects = 0;
118 sb = container_of(shrink, struct super_block, s_shrink);
121 * We don't call trylock_super() here as it is a scalability bottleneck,
122 * so we're exposed to partial setup state. The shrinker rwsem does not
123 * protect filesystem operations backing list_lru_shrink_count() or
124 * s_op->nr_cached_objects(). Counts can change between
125 * super_cache_count and super_cache_scan, so we really don't need locks
128 * However, if we are currently mounting the superblock, the underlying
129 * filesystem might be in a state of partial construction and hence it
130 * is dangerous to access it. trylock_super() uses a MS_BORN check to
131 * avoid this situation, so do the same here. The memory barrier is
132 * matched with the one in mount_fs() as we don't hold locks here.
134 if (!(sb->s_flags & MS_BORN))
138 if (sb->s_op && sb->s_op->nr_cached_objects)
139 total_objects = sb->s_op->nr_cached_objects(sb, sc);
141 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
142 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
144 total_objects = vfs_pressure_ratio(total_objects);
145 return total_objects;
148 static void destroy_super_work(struct work_struct *work)
150 struct super_block *s = container_of(work, struct super_block,
154 for (i = 0; i < SB_FREEZE_LEVELS; i++)
155 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
159 static void destroy_super_rcu(struct rcu_head *head)
161 struct super_block *s = container_of(head, struct super_block, rcu);
162 INIT_WORK(&s->destroy_work, destroy_super_work);
163 schedule_work(&s->destroy_work);
167 * destroy_super - frees a superblock
168 * @s: superblock to free
170 * Frees a superblock.
172 static void destroy_super(struct super_block *s)
174 list_lru_destroy(&s->s_dentry_lru);
175 list_lru_destroy(&s->s_inode_lru);
177 WARN_ON(!list_empty(&s->s_mounts));
180 call_rcu(&s->rcu, destroy_super_rcu);
184 * alloc_super - create new superblock
185 * @type: filesystem type superblock should belong to
186 * @flags: the mount flags
188 * Allocates and initializes a new &struct super_block. alloc_super()
189 * returns a pointer new superblock or %NULL if allocation had failed.
191 static struct super_block *alloc_super(struct file_system_type *type, int flags)
193 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
194 static const struct super_operations default_op;
200 INIT_LIST_HEAD(&s->s_mounts);
202 if (security_sb_alloc(s))
205 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
206 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
208 &type->s_writers_key[i]))
211 init_waitqueue_head(&s->s_writers.wait_unfrozen);
212 s->s_bdi = &noop_backing_dev_info;
214 INIT_HLIST_NODE(&s->s_instances);
215 INIT_HLIST_BL_HEAD(&s->s_anon);
216 mutex_init(&s->s_sync_lock);
217 INIT_LIST_HEAD(&s->s_inodes);
218 spin_lock_init(&s->s_inode_list_lock);
220 if (list_lru_init_memcg(&s->s_dentry_lru))
222 if (list_lru_init_memcg(&s->s_inode_lru))
225 init_rwsem(&s->s_umount);
226 lockdep_set_class(&s->s_umount, &type->s_umount_key);
228 * sget() can have s_umount recursion.
230 * When it cannot find a suitable sb, it allocates a new
231 * one (this one), and tries again to find a suitable old
234 * In case that succeeds, it will acquire the s_umount
235 * lock of the old one. Since these are clearly distrinct
236 * locks, and this object isn't exposed yet, there's no
239 * Annotate this by putting this lock in a different
242 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
244 atomic_set(&s->s_active, 1);
245 mutex_init(&s->s_vfs_rename_mutex);
246 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
247 mutex_init(&s->s_dquot.dqio_mutex);
248 mutex_init(&s->s_dquot.dqonoff_mutex);
249 s->s_maxbytes = MAX_NON_LFS;
250 s->s_op = &default_op;
251 s->s_time_gran = 1000000000;
252 s->cleancache_poolid = CLEANCACHE_NO_POOL;
254 s->s_shrink.seeks = DEFAULT_SEEKS;
255 s->s_shrink.scan_objects = super_cache_scan;
256 s->s_shrink.count_objects = super_cache_count;
257 s->s_shrink.batch = 1024;
258 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
266 /* Superblock refcounting */
269 * Drop a superblock's refcount. The caller must hold sb_lock.
271 static void __put_super(struct super_block *sb)
273 if (!--sb->s_count) {
274 list_del_init(&sb->s_list);
280 * put_super - drop a temporary reference to superblock
281 * @sb: superblock in question
283 * Drops a temporary reference, frees superblock if there's no
286 static void put_super(struct super_block *sb)
290 spin_unlock(&sb_lock);
295 * deactivate_locked_super - drop an active reference to superblock
296 * @s: superblock to deactivate
298 * Drops an active reference to superblock, converting it into a temprory
299 * one if there is no other active references left. In that case we
300 * tell fs driver to shut it down and drop the temporary reference we
303 * Caller holds exclusive lock on superblock; that lock is released.
305 void deactivate_locked_super(struct super_block *s)
307 struct file_system_type *fs = s->s_type;
308 if (atomic_dec_and_test(&s->s_active)) {
309 cleancache_invalidate_fs(s);
310 unregister_shrinker(&s->s_shrink);
314 * Since list_lru_destroy() may sleep, we cannot call it from
315 * put_super(), where we hold the sb_lock. Therefore we destroy
316 * the lru lists right now.
318 list_lru_destroy(&s->s_dentry_lru);
319 list_lru_destroy(&s->s_inode_lru);
324 up_write(&s->s_umount);
328 EXPORT_SYMBOL(deactivate_locked_super);
331 * deactivate_super - drop an active reference to superblock
332 * @s: superblock to deactivate
334 * Variant of deactivate_locked_super(), except that superblock is *not*
335 * locked by caller. If we are going to drop the final active reference,
336 * lock will be acquired prior to that.
338 void deactivate_super(struct super_block *s)
340 if (!atomic_add_unless(&s->s_active, -1, 1)) {
341 down_write(&s->s_umount);
342 deactivate_locked_super(s);
346 EXPORT_SYMBOL(deactivate_super);
349 * grab_super - acquire an active reference
350 * @s: reference we are trying to make active
352 * Tries to acquire an active reference. grab_super() is used when we
353 * had just found a superblock in super_blocks or fs_type->fs_supers
354 * and want to turn it into a full-blown active reference. grab_super()
355 * is called with sb_lock held and drops it. Returns 1 in case of
356 * success, 0 if we had failed (superblock contents was already dead or
357 * dying when grab_super() had been called). Note that this is only
358 * called for superblocks not in rundown mode (== ones still on ->fs_supers
359 * of their type), so increment of ->s_count is OK here.
361 static int grab_super(struct super_block *s) __releases(sb_lock)
364 spin_unlock(&sb_lock);
365 down_write(&s->s_umount);
366 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
370 up_write(&s->s_umount);
376 * trylock_super - try to grab ->s_umount shared
377 * @sb: reference we are trying to grab
379 * Try to prevent fs shutdown. This is used in places where we
380 * cannot take an active reference but we need to ensure that the
381 * filesystem is not shut down while we are working on it. It returns
382 * false if we cannot acquire s_umount or if we lose the race and
383 * filesystem already got into shutdown, and returns true with the s_umount
384 * lock held in read mode in case of success. On successful return,
385 * the caller must drop the s_umount lock when done.
387 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
388 * The reason why it's safe is that we are OK with doing trylock instead
389 * of down_read(). There's a couple of places that are OK with that, but
390 * it's very much not a general-purpose interface.
392 bool trylock_super(struct super_block *sb)
394 if (down_read_trylock(&sb->s_umount)) {
395 if (!hlist_unhashed(&sb->s_instances) &&
396 sb->s_root && (sb->s_flags & MS_BORN))
398 up_read(&sb->s_umount);
405 * generic_shutdown_super - common helper for ->kill_sb()
406 * @sb: superblock to kill
408 * generic_shutdown_super() does all fs-independent work on superblock
409 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
410 * that need destruction out of superblock, call generic_shutdown_super()
411 * and release aforementioned objects. Note: dentries and inodes _are_
412 * taken care of and do not need specific handling.
414 * Upon calling this function, the filesystem may no longer alter or
415 * rearrange the set of dentries belonging to this super_block, nor may it
416 * change the attachments of dentries to inodes.
418 void generic_shutdown_super(struct super_block *sb)
420 const struct super_operations *sop = sb->s_op;
423 shrink_dcache_for_umount(sb);
425 sb->s_flags &= ~MS_ACTIVE;
427 fsnotify_unmount_inodes(sb);
428 cgroup_writeback_umount();
432 if (sb->s_dio_done_wq) {
433 destroy_workqueue(sb->s_dio_done_wq);
434 sb->s_dio_done_wq = NULL;
440 if (!list_empty(&sb->s_inodes)) {
441 printk("VFS: Busy inodes after unmount of %s. "
442 "Self-destruct in 5 seconds. Have a nice day...\n",
447 /* should be initialized for __put_super_and_need_restart() */
448 hlist_del_init(&sb->s_instances);
449 spin_unlock(&sb_lock);
450 up_write(&sb->s_umount);
453 EXPORT_SYMBOL(generic_shutdown_super);
456 * sget - find or create a superblock
457 * @type: filesystem type superblock should belong to
458 * @test: comparison callback
459 * @set: setup callback
460 * @flags: mount flags
461 * @data: argument to each of them
463 struct super_block *sget(struct file_system_type *type,
464 int (*test)(struct super_block *,void *),
465 int (*set)(struct super_block *,void *),
469 struct super_block *s = NULL;
470 struct super_block *old;
476 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
477 if (!test(old, data))
479 if (!grab_super(old))
482 up_write(&s->s_umount);
490 spin_unlock(&sb_lock);
491 s = alloc_super(type, flags);
493 return ERR_PTR(-ENOMEM);
499 spin_unlock(&sb_lock);
500 up_write(&s->s_umount);
505 strlcpy(s->s_id, type->name, sizeof(s->s_id));
506 list_add_tail(&s->s_list, &super_blocks);
507 hlist_add_head(&s->s_instances, &type->fs_supers);
508 spin_unlock(&sb_lock);
509 get_filesystem(type);
510 err = register_shrinker(&s->s_shrink);
512 deactivate_locked_super(s);
520 void drop_super(struct super_block *sb)
522 up_read(&sb->s_umount);
526 EXPORT_SYMBOL(drop_super);
529 * iterate_supers - call function for all active superblocks
530 * @f: function to call
531 * @arg: argument to pass to it
533 * Scans the superblock list and calls given function, passing it
534 * locked superblock and given argument.
536 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
538 struct super_block *sb, *p = NULL;
541 list_for_each_entry(sb, &super_blocks, s_list) {
542 if (hlist_unhashed(&sb->s_instances))
545 spin_unlock(&sb_lock);
547 down_read(&sb->s_umount);
548 if (sb->s_root && (sb->s_flags & MS_BORN))
550 up_read(&sb->s_umount);
559 spin_unlock(&sb_lock);
563 * iterate_supers_type - call function for superblocks of given type
565 * @f: function to call
566 * @arg: argument to pass to it
568 * Scans the superblock list and calls given function, passing it
569 * locked superblock and given argument.
571 void iterate_supers_type(struct file_system_type *type,
572 void (*f)(struct super_block *, void *), void *arg)
574 struct super_block *sb, *p = NULL;
577 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
579 spin_unlock(&sb_lock);
581 down_read(&sb->s_umount);
582 if (sb->s_root && (sb->s_flags & MS_BORN))
584 up_read(&sb->s_umount);
593 spin_unlock(&sb_lock);
596 EXPORT_SYMBOL(iterate_supers_type);
599 * get_super - get the superblock of a device
600 * @bdev: device to get the superblock for
602 * Scans the superblock list and finds the superblock of the file system
603 * mounted on the device given. %NULL is returned if no match is found.
606 struct super_block *get_super(struct block_device *bdev)
608 struct super_block *sb;
615 list_for_each_entry(sb, &super_blocks, s_list) {
616 if (hlist_unhashed(&sb->s_instances))
618 if (sb->s_bdev == bdev) {
620 spin_unlock(&sb_lock);
621 down_read(&sb->s_umount);
623 if (sb->s_root && (sb->s_flags & MS_BORN))
625 up_read(&sb->s_umount);
626 /* nope, got unmounted */
632 spin_unlock(&sb_lock);
636 EXPORT_SYMBOL(get_super);
639 * get_super_thawed - get thawed superblock of a device
640 * @bdev: device to get the superblock for
642 * Scans the superblock list and finds the superblock of the file system
643 * mounted on the device. The superblock is returned once it is thawed
644 * (or immediately if it was not frozen). %NULL is returned if no match
647 struct super_block *get_super_thawed(struct block_device *bdev)
650 struct super_block *s = get_super(bdev);
651 if (!s || s->s_writers.frozen == SB_UNFROZEN)
653 up_read(&s->s_umount);
654 wait_event(s->s_writers.wait_unfrozen,
655 s->s_writers.frozen == SB_UNFROZEN);
659 EXPORT_SYMBOL(get_super_thawed);
662 * get_active_super - get an active reference to the superblock of a device
663 * @bdev: device to get the superblock for
665 * Scans the superblock list and finds the superblock of the file system
666 * mounted on the device given. Returns the superblock with an active
667 * reference or %NULL if none was found.
669 struct super_block *get_active_super(struct block_device *bdev)
671 struct super_block *sb;
678 list_for_each_entry(sb, &super_blocks, s_list) {
679 if (hlist_unhashed(&sb->s_instances))
681 if (sb->s_bdev == bdev) {
684 up_write(&sb->s_umount);
688 spin_unlock(&sb_lock);
692 struct super_block *user_get_super(dev_t dev)
694 struct super_block *sb;
698 list_for_each_entry(sb, &super_blocks, s_list) {
699 if (hlist_unhashed(&sb->s_instances))
701 if (sb->s_dev == dev) {
703 spin_unlock(&sb_lock);
704 down_read(&sb->s_umount);
706 if (sb->s_root && (sb->s_flags & MS_BORN))
708 up_read(&sb->s_umount);
709 /* nope, got unmounted */
715 spin_unlock(&sb_lock);
720 * do_remount_sb2 - asks filesystem to change mount options.
721 * @mnt: mount we are looking at
722 * @sb: superblock in question
723 * @flags: numeric part of options
724 * @data: the rest of options
725 * @force: whether or not to force the change
727 * Alters the mount options of a mounted file system.
729 int do_remount_sb2(struct vfsmount *mnt, struct super_block *sb, int flags, void *data, int force)
734 if (sb->s_writers.frozen != SB_UNFROZEN)
738 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
742 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
745 if (!hlist_empty(&sb->s_pins)) {
746 up_write(&sb->s_umount);
747 group_pin_kill(&sb->s_pins);
748 down_write(&sb->s_umount);
751 if (sb->s_writers.frozen != SB_UNFROZEN)
753 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
756 shrink_dcache_sb(sb);
758 /* If we are remounting RDONLY and current sb is read/write,
759 make sure there are no rw files opened */
762 sb->s_readonly_remount = 1;
765 retval = sb_prepare_remount_readonly(sb);
771 if (mnt && sb->s_op->remount_fs2) {
772 retval = sb->s_op->remount_fs2(mnt, sb, &flags, data);
775 goto cancel_readonly;
776 /* If forced remount, go ahead despite any errors */
777 WARN(1, "forced remount of a %s fs returned %i\n",
778 sb->s_type->name, retval);
780 } else if (sb->s_op->remount_fs) {
781 retval = sb->s_op->remount_fs(sb, &flags, data);
784 goto cancel_readonly;
785 /* If forced remount, go ahead despite any errors */
786 WARN(1, "forced remount of a %s fs returned %i\n",
787 sb->s_type->name, retval);
790 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
791 /* Needs to be ordered wrt mnt_is_readonly() */
793 sb->s_readonly_remount = 0;
796 * Some filesystems modify their metadata via some other path than the
797 * bdev buffer cache (eg. use a private mapping, or directories in
798 * pagecache, etc). Also file data modifications go via their own
799 * mappings. So If we try to mount readonly then copy the filesystem
800 * from bdev, we could get stale data, so invalidate it to give a best
801 * effort at coherency.
803 if (remount_ro && sb->s_bdev)
804 invalidate_bdev(sb->s_bdev);
808 sb->s_readonly_remount = 0;
812 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
814 return do_remount_sb2(NULL, sb, flags, data, force);
817 static void do_emergency_remount(struct work_struct *work)
819 struct super_block *sb, *p = NULL;
822 list_for_each_entry_reverse(sb, &super_blocks, s_list) {
823 if (hlist_unhashed(&sb->s_instances))
826 spin_unlock(&sb_lock);
827 down_write(&sb->s_umount);
828 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
829 !(sb->s_flags & MS_RDONLY)) {
831 * What lock protects sb->s_flags??
833 do_remount_sb(sb, MS_RDONLY, NULL, 1);
835 up_write(&sb->s_umount);
843 spin_unlock(&sb_lock);
845 printk("Emergency Remount complete\n");
848 void emergency_remount(void)
850 struct work_struct *work;
852 work = kmalloc(sizeof(*work), GFP_ATOMIC);
854 INIT_WORK(work, do_emergency_remount);
860 * Unnamed block devices are dummy devices used by virtual
861 * filesystems which don't use real block-devices. -- jrs
864 static DEFINE_IDA(unnamed_dev_ida);
865 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
866 /* Many userspace utilities consider an FSID of 0 invalid.
867 * Always return at least 1 from get_anon_bdev.
869 static int unnamed_dev_start = 1;
871 int get_anon_bdev(dev_t *p)
877 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
879 spin_lock(&unnamed_dev_lock);
880 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
882 unnamed_dev_start = dev + 1;
883 spin_unlock(&unnamed_dev_lock);
884 if (error == -EAGAIN)
885 /* We raced and lost with another CPU. */
890 if (dev >= (1 << MINORBITS)) {
891 spin_lock(&unnamed_dev_lock);
892 ida_remove(&unnamed_dev_ida, dev);
893 if (unnamed_dev_start > dev)
894 unnamed_dev_start = dev;
895 spin_unlock(&unnamed_dev_lock);
898 *p = MKDEV(0, dev & MINORMASK);
901 EXPORT_SYMBOL(get_anon_bdev);
903 void free_anon_bdev(dev_t dev)
905 int slot = MINOR(dev);
906 spin_lock(&unnamed_dev_lock);
907 ida_remove(&unnamed_dev_ida, slot);
908 if (slot < unnamed_dev_start)
909 unnamed_dev_start = slot;
910 spin_unlock(&unnamed_dev_lock);
912 EXPORT_SYMBOL(free_anon_bdev);
914 int set_anon_super(struct super_block *s, void *data)
916 return get_anon_bdev(&s->s_dev);
919 EXPORT_SYMBOL(set_anon_super);
921 void kill_anon_super(struct super_block *sb)
923 dev_t dev = sb->s_dev;
924 generic_shutdown_super(sb);
928 EXPORT_SYMBOL(kill_anon_super);
930 void kill_litter_super(struct super_block *sb)
933 d_genocide(sb->s_root);
937 EXPORT_SYMBOL(kill_litter_super);
939 static int ns_test_super(struct super_block *sb, void *data)
941 return sb->s_fs_info == data;
944 static int ns_set_super(struct super_block *sb, void *data)
946 sb->s_fs_info = data;
947 return set_anon_super(sb, NULL);
950 struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
951 void *data, int (*fill_super)(struct super_block *, void *, int))
953 struct super_block *sb;
955 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
961 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
963 deactivate_locked_super(sb);
967 sb->s_flags |= MS_ACTIVE;
970 return dget(sb->s_root);
973 EXPORT_SYMBOL(mount_ns);
976 static int set_bdev_super(struct super_block *s, void *data)
979 s->s_dev = s->s_bdev->bd_dev;
982 * We set the bdi here to the queue backing, file systems can
983 * overwrite this in ->fill_super()
985 s->s_bdi = bdev_get_queue(s->s_bdev)->backing_dev_info;
989 static int test_bdev_super(struct super_block *s, void *data)
991 return (void *)s->s_bdev == data;
994 struct dentry *mount_bdev(struct file_system_type *fs_type,
995 int flags, const char *dev_name, void *data,
996 int (*fill_super)(struct super_block *, void *, int))
998 struct block_device *bdev;
999 struct super_block *s;
1000 fmode_t mode = FMODE_READ | FMODE_EXCL;
1003 if (!(flags & MS_RDONLY))
1004 mode |= FMODE_WRITE;
1006 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1008 return ERR_CAST(bdev);
1011 * once the super is inserted into the list by sget, s_umount
1012 * will protect the lockfs code from trying to start a snapshot
1013 * while we are mounting
1015 mutex_lock(&bdev->bd_fsfreeze_mutex);
1016 if (bdev->bd_fsfreeze_count > 0) {
1017 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1021 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
1023 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1028 if ((flags ^ s->s_flags) & MS_RDONLY) {
1029 deactivate_locked_super(s);
1035 * s_umount nests inside bd_mutex during
1036 * __invalidate_device(). blkdev_put() acquires
1037 * bd_mutex and can't be called under s_umount. Drop
1038 * s_umount temporarily. This is safe as we're
1039 * holding an active reference.
1041 up_write(&s->s_umount);
1042 blkdev_put(bdev, mode);
1043 down_write(&s->s_umount);
1045 char b[BDEVNAME_SIZE];
1048 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1049 sb_set_blocksize(s, block_size(bdev));
1050 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1052 deactivate_locked_super(s);
1056 s->s_flags |= MS_ACTIVE;
1060 return dget(s->s_root);
1065 blkdev_put(bdev, mode);
1067 return ERR_PTR(error);
1069 EXPORT_SYMBOL(mount_bdev);
1071 void kill_block_super(struct super_block *sb)
1073 struct block_device *bdev = sb->s_bdev;
1074 fmode_t mode = sb->s_mode;
1076 bdev->bd_super = NULL;
1077 generic_shutdown_super(sb);
1078 sync_blockdev(bdev);
1079 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1080 blkdev_put(bdev, mode | FMODE_EXCL);
1083 EXPORT_SYMBOL(kill_block_super);
1086 struct dentry *mount_nodev(struct file_system_type *fs_type,
1087 int flags, void *data,
1088 int (*fill_super)(struct super_block *, void *, int))
1091 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1096 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1098 deactivate_locked_super(s);
1099 return ERR_PTR(error);
1101 s->s_flags |= MS_ACTIVE;
1102 return dget(s->s_root);
1104 EXPORT_SYMBOL(mount_nodev);
1106 static int compare_single(struct super_block *s, void *p)
1111 struct dentry *mount_single(struct file_system_type *fs_type,
1112 int flags, void *data,
1113 int (*fill_super)(struct super_block *, void *, int))
1115 struct super_block *s;
1118 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1122 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1124 deactivate_locked_super(s);
1125 return ERR_PTR(error);
1127 s->s_flags |= MS_ACTIVE;
1129 do_remount_sb(s, flags, data, 0);
1131 return dget(s->s_root);
1133 EXPORT_SYMBOL(mount_single);
1136 mount_fs(struct file_system_type *type, int flags, const char *name, struct vfsmount *mnt, void *data)
1138 struct dentry *root;
1139 struct super_block *sb;
1140 char *secdata = NULL;
1141 int error = -ENOMEM;
1143 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1144 secdata = alloc_secdata();
1148 error = security_sb_copy_data(data, secdata);
1150 goto out_free_secdata;
1154 root = type->mount2(mnt, type, flags, name, data);
1156 root = type->mount(type, flags, name, data);
1158 error = PTR_ERR(root);
1159 goto out_free_secdata;
1163 WARN_ON(!sb->s_bdi);
1166 * Write barrier is for super_cache_count(). We place it before setting
1167 * MS_BORN as the data dependency between the two functions is the
1168 * superblock structure contents that we just set up, not the MS_BORN
1172 sb->s_flags |= MS_BORN;
1174 error = security_sb_kern_mount(sb, flags, secdata);
1179 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1180 * but s_maxbytes was an unsigned long long for many releases. Throw
1181 * this warning for a little while to try and catch filesystems that
1182 * violate this rule.
1184 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1185 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1187 up_write(&sb->s_umount);
1188 free_secdata(secdata);
1192 deactivate_locked_super(sb);
1194 free_secdata(secdata);
1196 return ERR_PTR(error);
1200 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1203 void __sb_end_write(struct super_block *sb, int level)
1205 percpu_up_read(sb->s_writers.rw_sem + level-1);
1207 EXPORT_SYMBOL(__sb_end_write);
1210 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1213 int __sb_start_write(struct super_block *sb, int level, bool wait)
1215 bool force_trylock = false;
1218 #ifdef CONFIG_LOCKDEP
1220 * We want lockdep to tell us about possible deadlocks with freezing
1221 * but it's it bit tricky to properly instrument it. Getting a freeze
1222 * protection works as getting a read lock but there are subtle
1223 * problems. XFS for example gets freeze protection on internal level
1224 * twice in some cases, which is OK only because we already hold a
1225 * freeze protection also on higher level. Due to these cases we have
1226 * to use wait == F (trylock mode) which must not fail.
1231 for (i = 0; i < level - 1; i++)
1232 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
1233 force_trylock = true;
1238 if (wait && !force_trylock)
1239 percpu_down_read(sb->s_writers.rw_sem + level-1);
1241 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1243 WARN_ON(force_trylock & !ret);
1246 EXPORT_SYMBOL(__sb_start_write);
1249 * sb_wait_write - wait until all writers to given file system finish
1250 * @sb: the super for which we wait
1251 * @level: type of writers we wait for (normal vs page fault)
1253 * This function waits until there are no writers of given type to given file
1256 static void sb_wait_write(struct super_block *sb, int level)
1258 percpu_down_write(sb->s_writers.rw_sem + level-1);
1260 * We are going to return to userspace and forget about this lock, the
1261 * ownership goes to the caller of thaw_super() which does unlock.
1263 * FIXME: we should do this before return from freeze_super() after we
1264 * called sync_filesystem(sb) and s_op->freeze_fs(sb), and thaw_super()
1265 * should re-acquire these locks before s_op->unfreeze_fs(sb). However
1266 * this leads to lockdep false-positives, so currently we do the early
1267 * release right after acquire.
1269 percpu_rwsem_release(sb->s_writers.rw_sem + level-1, 0, _THIS_IP_);
1272 static void sb_freeze_unlock(struct super_block *sb)
1276 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1277 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1279 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1280 percpu_up_write(sb->s_writers.rw_sem + level);
1284 * freeze_super - lock the filesystem and force it into a consistent state
1285 * @sb: the super to lock
1287 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1288 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1291 * During this function, sb->s_writers.frozen goes through these values:
1293 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1295 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1296 * writes should be blocked, though page faults are still allowed. We wait for
1297 * all writes to complete and then proceed to the next stage.
1299 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1300 * but internal fs threads can still modify the filesystem (although they
1301 * should not dirty new pages or inodes), writeback can run etc. After waiting
1302 * for all running page faults we sync the filesystem which will clean all
1303 * dirty pages and inodes (no new dirty pages or inodes can be created when
1306 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1307 * modification are blocked (e.g. XFS preallocation truncation on inode
1308 * reclaim). This is usually implemented by blocking new transactions for
1309 * filesystems that have them and need this additional guard. After all
1310 * internal writers are finished we call ->freeze_fs() to finish filesystem
1311 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1312 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1314 * sb->s_writers.frozen is protected by sb->s_umount.
1316 int freeze_super(struct super_block *sb)
1320 atomic_inc(&sb->s_active);
1321 down_write(&sb->s_umount);
1322 if (sb->s_writers.frozen != SB_UNFROZEN) {
1323 deactivate_locked_super(sb);
1327 if (!(sb->s_flags & MS_BORN)) {
1328 up_write(&sb->s_umount);
1329 return 0; /* sic - it's "nothing to do" */
1332 if (sb->s_flags & MS_RDONLY) {
1333 /* Nothing to do really... */
1334 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1335 up_write(&sb->s_umount);
1339 sb->s_writers.frozen = SB_FREEZE_WRITE;
1340 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1341 up_write(&sb->s_umount);
1342 sb_wait_write(sb, SB_FREEZE_WRITE);
1343 down_write(&sb->s_umount);
1345 /* Now we go and block page faults... */
1346 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1347 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1349 /* All writers are done so after syncing there won't be dirty data */
1350 sync_filesystem(sb);
1352 /* Now wait for internal filesystem counter */
1353 sb->s_writers.frozen = SB_FREEZE_FS;
1354 sb_wait_write(sb, SB_FREEZE_FS);
1356 if (sb->s_op->freeze_fs) {
1357 ret = sb->s_op->freeze_fs(sb);
1360 "VFS:Filesystem freeze failed\n");
1361 sb->s_writers.frozen = SB_UNFROZEN;
1362 sb_freeze_unlock(sb);
1363 wake_up(&sb->s_writers.wait_unfrozen);
1364 deactivate_locked_super(sb);
1369 * For debugging purposes so that fs can warn if it sees write activity
1370 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1372 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1373 up_write(&sb->s_umount);
1376 EXPORT_SYMBOL(freeze_super);
1379 * thaw_super -- unlock filesystem
1380 * @sb: the super to thaw
1382 * Unlocks the filesystem and marks it writeable again after freeze_super().
1384 int thaw_super(struct super_block *sb)
1388 down_write(&sb->s_umount);
1389 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1390 up_write(&sb->s_umount);
1394 if (sb->s_flags & MS_RDONLY) {
1395 sb->s_writers.frozen = SB_UNFROZEN;
1399 if (sb->s_op->unfreeze_fs) {
1400 error = sb->s_op->unfreeze_fs(sb);
1403 "VFS:Filesystem thaw failed\n");
1404 up_write(&sb->s_umount);
1409 sb->s_writers.frozen = SB_UNFROZEN;
1410 sb_freeze_unlock(sb);
1412 wake_up(&sb->s_writers.wait_unfrozen);
1413 deactivate_locked_super(sb);
1416 EXPORT_SYMBOL(thaw_super);