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 * Don't call trylock_super as it is a potential
122 * scalability bottleneck. The counts could get updated
123 * between super_cache_count and super_cache_scan anyway.
124 * Call to super_cache_count with shrinker_rwsem held
125 * ensures the safety of call to list_lru_shrink_count() and
126 * s_op->nr_cached_objects().
128 if (sb->s_op && sb->s_op->nr_cached_objects)
129 total_objects = sb->s_op->nr_cached_objects(sb, sc);
131 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
132 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
134 total_objects = vfs_pressure_ratio(total_objects);
135 return total_objects;
138 static void destroy_super_work(struct work_struct *work)
140 struct super_block *s = container_of(work, struct super_block,
144 for (i = 0; i < SB_FREEZE_LEVELS; i++)
145 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
149 static void destroy_super_rcu(struct rcu_head *head)
151 struct super_block *s = container_of(head, struct super_block, rcu);
152 INIT_WORK(&s->destroy_work, destroy_super_work);
153 schedule_work(&s->destroy_work);
157 * destroy_super - frees a superblock
158 * @s: superblock to free
160 * Frees a superblock.
162 static void destroy_super(struct super_block *s)
164 list_lru_destroy(&s->s_dentry_lru);
165 list_lru_destroy(&s->s_inode_lru);
167 WARN_ON(!list_empty(&s->s_mounts));
170 call_rcu(&s->rcu, destroy_super_rcu);
174 * alloc_super - create new superblock
175 * @type: filesystem type superblock should belong to
176 * @flags: the mount flags
178 * Allocates and initializes a new &struct super_block. alloc_super()
179 * returns a pointer new superblock or %NULL if allocation had failed.
181 static struct super_block *alloc_super(struct file_system_type *type, int flags)
183 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
184 static const struct super_operations default_op;
190 INIT_LIST_HEAD(&s->s_mounts);
192 if (security_sb_alloc(s))
195 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
196 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
198 &type->s_writers_key[i]))
201 init_waitqueue_head(&s->s_writers.wait_unfrozen);
202 s->s_bdi = &noop_backing_dev_info;
204 INIT_HLIST_NODE(&s->s_instances);
205 INIT_HLIST_BL_HEAD(&s->s_anon);
206 mutex_init(&s->s_sync_lock);
207 INIT_LIST_HEAD(&s->s_inodes);
208 spin_lock_init(&s->s_inode_list_lock);
210 if (list_lru_init_memcg(&s->s_dentry_lru))
212 if (list_lru_init_memcg(&s->s_inode_lru))
215 init_rwsem(&s->s_umount);
216 lockdep_set_class(&s->s_umount, &type->s_umount_key);
218 * sget() can have s_umount recursion.
220 * When it cannot find a suitable sb, it allocates a new
221 * one (this one), and tries again to find a suitable old
224 * In case that succeeds, it will acquire the s_umount
225 * lock of the old one. Since these are clearly distrinct
226 * locks, and this object isn't exposed yet, there's no
229 * Annotate this by putting this lock in a different
232 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
234 atomic_set(&s->s_active, 1);
235 mutex_init(&s->s_vfs_rename_mutex);
236 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
237 mutex_init(&s->s_dquot.dqio_mutex);
238 mutex_init(&s->s_dquot.dqonoff_mutex);
239 s->s_maxbytes = MAX_NON_LFS;
240 s->s_op = &default_op;
241 s->s_time_gran = 1000000000;
242 s->cleancache_poolid = CLEANCACHE_NO_POOL;
244 s->s_shrink.seeks = DEFAULT_SEEKS;
245 s->s_shrink.scan_objects = super_cache_scan;
246 s->s_shrink.count_objects = super_cache_count;
247 s->s_shrink.batch = 1024;
248 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
256 /* Superblock refcounting */
259 * Drop a superblock's refcount. The caller must hold sb_lock.
261 static void __put_super(struct super_block *sb)
263 if (!--sb->s_count) {
264 list_del_init(&sb->s_list);
270 * put_super - drop a temporary reference to superblock
271 * @sb: superblock in question
273 * Drops a temporary reference, frees superblock if there's no
276 static void put_super(struct super_block *sb)
280 spin_unlock(&sb_lock);
285 * deactivate_locked_super - drop an active reference to superblock
286 * @s: superblock to deactivate
288 * Drops an active reference to superblock, converting it into a temprory
289 * one if there is no other active references left. In that case we
290 * tell fs driver to shut it down and drop the temporary reference we
293 * Caller holds exclusive lock on superblock; that lock is released.
295 void deactivate_locked_super(struct super_block *s)
297 struct file_system_type *fs = s->s_type;
298 if (atomic_dec_and_test(&s->s_active)) {
299 cleancache_invalidate_fs(s);
300 unregister_shrinker(&s->s_shrink);
304 * Since list_lru_destroy() may sleep, we cannot call it from
305 * put_super(), where we hold the sb_lock. Therefore we destroy
306 * the lru lists right now.
308 list_lru_destroy(&s->s_dentry_lru);
309 list_lru_destroy(&s->s_inode_lru);
314 up_write(&s->s_umount);
318 EXPORT_SYMBOL(deactivate_locked_super);
321 * deactivate_super - drop an active reference to superblock
322 * @s: superblock to deactivate
324 * Variant of deactivate_locked_super(), except that superblock is *not*
325 * locked by caller. If we are going to drop the final active reference,
326 * lock will be acquired prior to that.
328 void deactivate_super(struct super_block *s)
330 if (!atomic_add_unless(&s->s_active, -1, 1)) {
331 down_write(&s->s_umount);
332 deactivate_locked_super(s);
336 EXPORT_SYMBOL(deactivate_super);
339 * grab_super - acquire an active reference
340 * @s: reference we are trying to make active
342 * Tries to acquire an active reference. grab_super() is used when we
343 * had just found a superblock in super_blocks or fs_type->fs_supers
344 * and want to turn it into a full-blown active reference. grab_super()
345 * is called with sb_lock held and drops it. Returns 1 in case of
346 * success, 0 if we had failed (superblock contents was already dead or
347 * dying when grab_super() had been called). Note that this is only
348 * called for superblocks not in rundown mode (== ones still on ->fs_supers
349 * of their type), so increment of ->s_count is OK here.
351 static int grab_super(struct super_block *s) __releases(sb_lock)
354 spin_unlock(&sb_lock);
355 down_write(&s->s_umount);
356 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
360 up_write(&s->s_umount);
366 * trylock_super - try to grab ->s_umount shared
367 * @sb: reference we are trying to grab
369 * Try to prevent fs shutdown. This is used in places where we
370 * cannot take an active reference but we need to ensure that the
371 * filesystem is not shut down while we are working on it. It returns
372 * false if we cannot acquire s_umount or if we lose the race and
373 * filesystem already got into shutdown, and returns true with the s_umount
374 * lock held in read mode in case of success. On successful return,
375 * the caller must drop the s_umount lock when done.
377 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
378 * The reason why it's safe is that we are OK with doing trylock instead
379 * of down_read(). There's a couple of places that are OK with that, but
380 * it's very much not a general-purpose interface.
382 bool trylock_super(struct super_block *sb)
384 if (down_read_trylock(&sb->s_umount)) {
385 if (!hlist_unhashed(&sb->s_instances) &&
386 sb->s_root && (sb->s_flags & MS_BORN))
388 up_read(&sb->s_umount);
395 * generic_shutdown_super - common helper for ->kill_sb()
396 * @sb: superblock to kill
398 * generic_shutdown_super() does all fs-independent work on superblock
399 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
400 * that need destruction out of superblock, call generic_shutdown_super()
401 * and release aforementioned objects. Note: dentries and inodes _are_
402 * taken care of and do not need specific handling.
404 * Upon calling this function, the filesystem may no longer alter or
405 * rearrange the set of dentries belonging to this super_block, nor may it
406 * change the attachments of dentries to inodes.
408 void generic_shutdown_super(struct super_block *sb)
410 const struct super_operations *sop = sb->s_op;
413 shrink_dcache_for_umount(sb);
415 sb->s_flags &= ~MS_ACTIVE;
417 fsnotify_unmount_inodes(sb);
421 if (sb->s_dio_done_wq) {
422 destroy_workqueue(sb->s_dio_done_wq);
423 sb->s_dio_done_wq = NULL;
429 if (!list_empty(&sb->s_inodes)) {
430 printk("VFS: Busy inodes after unmount of %s. "
431 "Self-destruct in 5 seconds. Have a nice day...\n",
436 /* should be initialized for __put_super_and_need_restart() */
437 hlist_del_init(&sb->s_instances);
438 spin_unlock(&sb_lock);
439 up_write(&sb->s_umount);
442 EXPORT_SYMBOL(generic_shutdown_super);
445 * sget - find or create a superblock
446 * @type: filesystem type superblock should belong to
447 * @test: comparison callback
448 * @set: setup callback
449 * @flags: mount flags
450 * @data: argument to each of them
452 struct super_block *sget(struct file_system_type *type,
453 int (*test)(struct super_block *,void *),
454 int (*set)(struct super_block *,void *),
458 struct super_block *s = NULL;
459 struct super_block *old;
465 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
466 if (!test(old, data))
468 if (!grab_super(old))
471 up_write(&s->s_umount);
479 spin_unlock(&sb_lock);
480 s = alloc_super(type, flags);
482 return ERR_PTR(-ENOMEM);
488 spin_unlock(&sb_lock);
489 up_write(&s->s_umount);
494 strlcpy(s->s_id, type->name, sizeof(s->s_id));
495 list_add_tail(&s->s_list, &super_blocks);
496 hlist_add_head(&s->s_instances, &type->fs_supers);
497 spin_unlock(&sb_lock);
498 get_filesystem(type);
499 register_shrinker(&s->s_shrink);
505 void drop_super(struct super_block *sb)
507 up_read(&sb->s_umount);
511 EXPORT_SYMBOL(drop_super);
514 * iterate_supers - call function for all active superblocks
515 * @f: function to call
516 * @arg: argument to pass to it
518 * Scans the superblock list and calls given function, passing it
519 * locked superblock and given argument.
521 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
523 struct super_block *sb, *p = NULL;
526 list_for_each_entry(sb, &super_blocks, s_list) {
527 if (hlist_unhashed(&sb->s_instances))
530 spin_unlock(&sb_lock);
532 down_read(&sb->s_umount);
533 if (sb->s_root && (sb->s_flags & MS_BORN))
535 up_read(&sb->s_umount);
544 spin_unlock(&sb_lock);
548 * iterate_supers_type - call function for superblocks of given type
550 * @f: function to call
551 * @arg: argument to pass to it
553 * Scans the superblock list and calls given function, passing it
554 * locked superblock and given argument.
556 void iterate_supers_type(struct file_system_type *type,
557 void (*f)(struct super_block *, void *), void *arg)
559 struct super_block *sb, *p = NULL;
562 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
564 spin_unlock(&sb_lock);
566 down_read(&sb->s_umount);
567 if (sb->s_root && (sb->s_flags & MS_BORN))
569 up_read(&sb->s_umount);
578 spin_unlock(&sb_lock);
581 EXPORT_SYMBOL(iterate_supers_type);
584 * get_super - get the superblock of a device
585 * @bdev: device to get the superblock for
587 * Scans the superblock list and finds the superblock of the file system
588 * mounted on the device given. %NULL is returned if no match is found.
591 struct super_block *get_super(struct block_device *bdev)
593 struct super_block *sb;
600 list_for_each_entry(sb, &super_blocks, s_list) {
601 if (hlist_unhashed(&sb->s_instances))
603 if (sb->s_bdev == bdev) {
605 spin_unlock(&sb_lock);
606 down_read(&sb->s_umount);
608 if (sb->s_root && (sb->s_flags & MS_BORN))
610 up_read(&sb->s_umount);
611 /* nope, got unmounted */
617 spin_unlock(&sb_lock);
621 EXPORT_SYMBOL(get_super);
624 * get_super_thawed - get thawed superblock of a device
625 * @bdev: device to get the superblock for
627 * Scans the superblock list and finds the superblock of the file system
628 * mounted on the device. The superblock is returned once it is thawed
629 * (or immediately if it was not frozen). %NULL is returned if no match
632 struct super_block *get_super_thawed(struct block_device *bdev)
635 struct super_block *s = get_super(bdev);
636 if (!s || s->s_writers.frozen == SB_UNFROZEN)
638 up_read(&s->s_umount);
639 wait_event(s->s_writers.wait_unfrozen,
640 s->s_writers.frozen == SB_UNFROZEN);
644 EXPORT_SYMBOL(get_super_thawed);
647 * get_active_super - get an active reference to the superblock of a device
648 * @bdev: device to get the superblock for
650 * Scans the superblock list and finds the superblock of the file system
651 * mounted on the device given. Returns the superblock with an active
652 * reference or %NULL if none was found.
654 struct super_block *get_active_super(struct block_device *bdev)
656 struct super_block *sb;
663 list_for_each_entry(sb, &super_blocks, s_list) {
664 if (hlist_unhashed(&sb->s_instances))
666 if (sb->s_bdev == bdev) {
669 up_write(&sb->s_umount);
673 spin_unlock(&sb_lock);
677 struct super_block *user_get_super(dev_t dev)
679 struct super_block *sb;
683 list_for_each_entry(sb, &super_blocks, s_list) {
684 if (hlist_unhashed(&sb->s_instances))
686 if (sb->s_dev == dev) {
688 spin_unlock(&sb_lock);
689 down_read(&sb->s_umount);
691 if (sb->s_root && (sb->s_flags & MS_BORN))
693 up_read(&sb->s_umount);
694 /* nope, got unmounted */
700 spin_unlock(&sb_lock);
705 * do_remount_sb - asks filesystem to change mount options.
706 * @sb: superblock in question
707 * @flags: numeric part of options
708 * @data: the rest of options
709 * @force: whether or not to force the change
711 * Alters the mount options of a mounted file system.
713 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
718 if (sb->s_writers.frozen != SB_UNFROZEN)
722 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
726 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
729 if (!hlist_empty(&sb->s_pins)) {
730 up_write(&sb->s_umount);
731 group_pin_kill(&sb->s_pins);
732 down_write(&sb->s_umount);
735 if (sb->s_writers.frozen != SB_UNFROZEN)
737 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
740 shrink_dcache_sb(sb);
742 /* If we are remounting RDONLY and current sb is read/write,
743 make sure there are no rw files opened */
746 sb->s_readonly_remount = 1;
749 retval = sb_prepare_remount_readonly(sb);
755 if (sb->s_op->remount_fs) {
756 retval = sb->s_op->remount_fs(sb, &flags, data);
759 goto cancel_readonly;
760 /* If forced remount, go ahead despite any errors */
761 WARN(1, "forced remount of a %s fs returned %i\n",
762 sb->s_type->name, retval);
765 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
766 /* Needs to be ordered wrt mnt_is_readonly() */
768 sb->s_readonly_remount = 0;
771 * Some filesystems modify their metadata via some other path than the
772 * bdev buffer cache (eg. use a private mapping, or directories in
773 * pagecache, etc). Also file data modifications go via their own
774 * mappings. So If we try to mount readonly then copy the filesystem
775 * from bdev, we could get stale data, so invalidate it to give a best
776 * effort at coherency.
778 if (remount_ro && sb->s_bdev)
779 invalidate_bdev(sb->s_bdev);
783 sb->s_readonly_remount = 0;
787 static void do_emergency_remount(struct work_struct *work)
789 struct super_block *sb, *p = NULL;
792 list_for_each_entry(sb, &super_blocks, s_list) {
793 if (hlist_unhashed(&sb->s_instances))
796 spin_unlock(&sb_lock);
797 down_write(&sb->s_umount);
798 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
799 !(sb->s_flags & MS_RDONLY)) {
801 * What lock protects sb->s_flags??
803 do_remount_sb(sb, MS_RDONLY, NULL, 1);
805 up_write(&sb->s_umount);
813 spin_unlock(&sb_lock);
815 printk("Emergency Remount complete\n");
818 void emergency_remount(void)
820 struct work_struct *work;
822 work = kmalloc(sizeof(*work), GFP_ATOMIC);
824 INIT_WORK(work, do_emergency_remount);
830 * Unnamed block devices are dummy devices used by virtual
831 * filesystems which don't use real block-devices. -- jrs
834 static DEFINE_IDA(unnamed_dev_ida);
835 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
836 /* Many userspace utilities consider an FSID of 0 invalid.
837 * Always return at least 1 from get_anon_bdev.
839 static int unnamed_dev_start = 1;
841 int get_anon_bdev(dev_t *p)
847 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
849 spin_lock(&unnamed_dev_lock);
850 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
852 unnamed_dev_start = dev + 1;
853 spin_unlock(&unnamed_dev_lock);
854 if (error == -EAGAIN)
855 /* We raced and lost with another CPU. */
860 if (dev >= (1 << MINORBITS)) {
861 spin_lock(&unnamed_dev_lock);
862 ida_remove(&unnamed_dev_ida, dev);
863 if (unnamed_dev_start > dev)
864 unnamed_dev_start = dev;
865 spin_unlock(&unnamed_dev_lock);
868 *p = MKDEV(0, dev & MINORMASK);
871 EXPORT_SYMBOL(get_anon_bdev);
873 void free_anon_bdev(dev_t dev)
875 int slot = MINOR(dev);
876 spin_lock(&unnamed_dev_lock);
877 ida_remove(&unnamed_dev_ida, slot);
878 if (slot < unnamed_dev_start)
879 unnamed_dev_start = slot;
880 spin_unlock(&unnamed_dev_lock);
882 EXPORT_SYMBOL(free_anon_bdev);
884 int set_anon_super(struct super_block *s, void *data)
886 return get_anon_bdev(&s->s_dev);
889 EXPORT_SYMBOL(set_anon_super);
891 void kill_anon_super(struct super_block *sb)
893 dev_t dev = sb->s_dev;
894 generic_shutdown_super(sb);
898 EXPORT_SYMBOL(kill_anon_super);
900 void kill_litter_super(struct super_block *sb)
903 d_genocide(sb->s_root);
907 EXPORT_SYMBOL(kill_litter_super);
909 static int ns_test_super(struct super_block *sb, void *data)
911 return sb->s_fs_info == data;
914 static int ns_set_super(struct super_block *sb, void *data)
916 sb->s_fs_info = data;
917 return set_anon_super(sb, NULL);
920 struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
921 void *data, int (*fill_super)(struct super_block *, void *, int))
923 struct super_block *sb;
925 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
931 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
933 deactivate_locked_super(sb);
937 sb->s_flags |= MS_ACTIVE;
940 return dget(sb->s_root);
943 EXPORT_SYMBOL(mount_ns);
946 static int set_bdev_super(struct super_block *s, void *data)
949 s->s_dev = s->s_bdev->bd_dev;
952 * We set the bdi here to the queue backing, file systems can
953 * overwrite this in ->fill_super()
955 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
959 static int test_bdev_super(struct super_block *s, void *data)
961 return (void *)s->s_bdev == data;
964 struct dentry *mount_bdev(struct file_system_type *fs_type,
965 int flags, const char *dev_name, void *data,
966 int (*fill_super)(struct super_block *, void *, int))
968 struct block_device *bdev;
969 struct super_block *s;
970 fmode_t mode = FMODE_READ | FMODE_EXCL;
973 if (!(flags & MS_RDONLY))
976 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
978 return ERR_CAST(bdev);
981 * once the super is inserted into the list by sget, s_umount
982 * will protect the lockfs code from trying to start a snapshot
983 * while we are mounting
985 mutex_lock(&bdev->bd_fsfreeze_mutex);
986 if (bdev->bd_fsfreeze_count > 0) {
987 mutex_unlock(&bdev->bd_fsfreeze_mutex);
991 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
993 mutex_unlock(&bdev->bd_fsfreeze_mutex);
998 if ((flags ^ s->s_flags) & MS_RDONLY) {
999 deactivate_locked_super(s);
1005 * s_umount nests inside bd_mutex during
1006 * __invalidate_device(). blkdev_put() acquires
1007 * bd_mutex and can't be called under s_umount. Drop
1008 * s_umount temporarily. This is safe as we're
1009 * holding an active reference.
1011 up_write(&s->s_umount);
1012 blkdev_put(bdev, mode);
1013 down_write(&s->s_umount);
1015 char b[BDEVNAME_SIZE];
1018 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1019 sb_set_blocksize(s, block_size(bdev));
1020 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1022 deactivate_locked_super(s);
1026 s->s_flags |= MS_ACTIVE;
1030 return dget(s->s_root);
1035 blkdev_put(bdev, mode);
1037 return ERR_PTR(error);
1039 EXPORT_SYMBOL(mount_bdev);
1041 void kill_block_super(struct super_block *sb)
1043 struct block_device *bdev = sb->s_bdev;
1044 fmode_t mode = sb->s_mode;
1046 bdev->bd_super = NULL;
1047 generic_shutdown_super(sb);
1048 sync_blockdev(bdev);
1049 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1050 blkdev_put(bdev, mode | FMODE_EXCL);
1053 EXPORT_SYMBOL(kill_block_super);
1056 struct dentry *mount_nodev(struct file_system_type *fs_type,
1057 int flags, void *data,
1058 int (*fill_super)(struct super_block *, void *, int))
1061 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1066 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1068 deactivate_locked_super(s);
1069 return ERR_PTR(error);
1071 s->s_flags |= MS_ACTIVE;
1072 return dget(s->s_root);
1074 EXPORT_SYMBOL(mount_nodev);
1076 static int compare_single(struct super_block *s, void *p)
1081 struct dentry *mount_single(struct file_system_type *fs_type,
1082 int flags, void *data,
1083 int (*fill_super)(struct super_block *, void *, int))
1085 struct super_block *s;
1088 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1092 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1094 deactivate_locked_super(s);
1095 return ERR_PTR(error);
1097 s->s_flags |= MS_ACTIVE;
1099 do_remount_sb(s, flags, data, 0);
1101 return dget(s->s_root);
1103 EXPORT_SYMBOL(mount_single);
1106 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1108 struct dentry *root;
1109 struct super_block *sb;
1110 char *secdata = NULL;
1111 int error = -ENOMEM;
1113 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1114 secdata = alloc_secdata();
1118 error = security_sb_copy_data(data, secdata);
1120 goto out_free_secdata;
1123 root = type->mount(type, flags, name, data);
1125 error = PTR_ERR(root);
1126 goto out_free_secdata;
1130 WARN_ON(!sb->s_bdi);
1131 sb->s_flags |= MS_BORN;
1133 error = security_sb_kern_mount(sb, flags, secdata);
1138 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1139 * but s_maxbytes was an unsigned long long for many releases. Throw
1140 * this warning for a little while to try and catch filesystems that
1141 * violate this rule.
1143 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1144 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1146 up_write(&sb->s_umount);
1147 free_secdata(secdata);
1151 deactivate_locked_super(sb);
1153 free_secdata(secdata);
1155 return ERR_PTR(error);
1159 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1162 void __sb_end_write(struct super_block *sb, int level)
1164 percpu_up_read(sb->s_writers.rw_sem + level-1);
1166 EXPORT_SYMBOL(__sb_end_write);
1169 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1172 int __sb_start_write(struct super_block *sb, int level, bool wait)
1174 bool force_trylock = false;
1177 #ifdef CONFIG_LOCKDEP
1179 * We want lockdep to tell us about possible deadlocks with freezing
1180 * but it's it bit tricky to properly instrument it. Getting a freeze
1181 * protection works as getting a read lock but there are subtle
1182 * problems. XFS for example gets freeze protection on internal level
1183 * twice in some cases, which is OK only because we already hold a
1184 * freeze protection also on higher level. Due to these cases we have
1185 * to use wait == F (trylock mode) which must not fail.
1190 for (i = 0; i < level - 1; i++)
1191 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
1192 force_trylock = true;
1197 if (wait && !force_trylock)
1198 percpu_down_read(sb->s_writers.rw_sem + level-1);
1200 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1202 WARN_ON(force_trylock && !ret);
1205 EXPORT_SYMBOL(__sb_start_write);
1208 * sb_wait_write - wait until all writers to given file system finish
1209 * @sb: the super for which we wait
1210 * @level: type of writers we wait for (normal vs page fault)
1212 * This function waits until there are no writers of given type to given file
1215 static void sb_wait_write(struct super_block *sb, int level)
1217 percpu_down_write(sb->s_writers.rw_sem + level-1);
1219 * We are going to return to userspace and forget about this lock, the
1220 * ownership goes to the caller of thaw_super() which does unlock.
1222 * FIXME: we should do this before return from freeze_super() after we
1223 * called sync_filesystem(sb) and s_op->freeze_fs(sb), and thaw_super()
1224 * should re-acquire these locks before s_op->unfreeze_fs(sb). However
1225 * this leads to lockdep false-positives, so currently we do the early
1226 * release right after acquire.
1228 percpu_rwsem_release(sb->s_writers.rw_sem + level-1, 0, _THIS_IP_);
1231 static void sb_freeze_unlock(struct super_block *sb)
1235 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1236 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1238 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1239 percpu_up_write(sb->s_writers.rw_sem + level);
1243 * freeze_super - lock the filesystem and force it into a consistent state
1244 * @sb: the super to lock
1246 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1247 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1250 * During this function, sb->s_writers.frozen goes through these values:
1252 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1254 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1255 * writes should be blocked, though page faults are still allowed. We wait for
1256 * all writes to complete and then proceed to the next stage.
1258 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1259 * but internal fs threads can still modify the filesystem (although they
1260 * should not dirty new pages or inodes), writeback can run etc. After waiting
1261 * for all running page faults we sync the filesystem which will clean all
1262 * dirty pages and inodes (no new dirty pages or inodes can be created when
1265 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1266 * modification are blocked (e.g. XFS preallocation truncation on inode
1267 * reclaim). This is usually implemented by blocking new transactions for
1268 * filesystems that have them and need this additional guard. After all
1269 * internal writers are finished we call ->freeze_fs() to finish filesystem
1270 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1271 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1273 * sb->s_writers.frozen is protected by sb->s_umount.
1275 int freeze_super(struct super_block *sb)
1279 atomic_inc(&sb->s_active);
1280 down_write(&sb->s_umount);
1281 if (sb->s_writers.frozen != SB_UNFROZEN) {
1282 deactivate_locked_super(sb);
1286 if (!(sb->s_flags & MS_BORN)) {
1287 up_write(&sb->s_umount);
1288 return 0; /* sic - it's "nothing to do" */
1291 if (sb->s_flags & MS_RDONLY) {
1292 /* Nothing to do really... */
1293 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1294 up_write(&sb->s_umount);
1298 sb->s_writers.frozen = SB_FREEZE_WRITE;
1299 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1300 up_write(&sb->s_umount);
1301 sb_wait_write(sb, SB_FREEZE_WRITE);
1302 down_write(&sb->s_umount);
1304 /* Now we go and block page faults... */
1305 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1306 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1308 /* All writers are done so after syncing there won't be dirty data */
1309 sync_filesystem(sb);
1311 /* Now wait for internal filesystem counter */
1312 sb->s_writers.frozen = SB_FREEZE_FS;
1313 sb_wait_write(sb, SB_FREEZE_FS);
1315 if (sb->s_op->freeze_fs) {
1316 ret = sb->s_op->freeze_fs(sb);
1319 "VFS:Filesystem freeze failed\n");
1320 sb->s_writers.frozen = SB_UNFROZEN;
1321 sb_freeze_unlock(sb);
1322 wake_up(&sb->s_writers.wait_unfrozen);
1323 deactivate_locked_super(sb);
1328 * This is just for debugging purposes so that fs can warn if it
1329 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1331 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1332 up_write(&sb->s_umount);
1335 EXPORT_SYMBOL(freeze_super);
1338 * thaw_super -- unlock filesystem
1339 * @sb: the super to thaw
1341 * Unlocks the filesystem and marks it writeable again after freeze_super().
1343 int thaw_super(struct super_block *sb)
1347 down_write(&sb->s_umount);
1348 if (sb->s_writers.frozen == SB_UNFROZEN) {
1349 up_write(&sb->s_umount);
1353 if (sb->s_flags & MS_RDONLY) {
1354 sb->s_writers.frozen = SB_UNFROZEN;
1358 if (sb->s_op->unfreeze_fs) {
1359 error = sb->s_op->unfreeze_fs(sb);
1362 "VFS:Filesystem thaw failed\n");
1363 up_write(&sb->s_umount);
1368 sb->s_writers.frozen = SB_UNFROZEN;
1369 sb_freeze_unlock(sb);
1371 wake_up(&sb->s_writers.wait_unfrozen);
1372 deactivate_locked_super(sb);
1375 EXPORT_SYMBOL(thaw_super);