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>
36 #include <linux/user_namespace.h>
40 static LIST_HEAD(super_blocks);
41 static DEFINE_SPINLOCK(sb_lock);
43 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
50 * One thing we have to be careful of with a per-sb shrinker is that we don't
51 * drop the last active reference to the superblock from within the shrinker.
52 * If that happens we could trigger unregistering the shrinker from within the
53 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
54 * take a passive reference to the superblock to avoid this from occurring.
56 static unsigned long super_cache_scan(struct shrinker *shrink,
57 struct shrink_control *sc)
59 struct super_block *sb;
66 sb = container_of(shrink, struct super_block, s_shrink);
69 * Deadlock avoidance. We may hold various FS locks, and we don't want
70 * to recurse into the FS that called us in clear_inode() and friends..
72 if (!(sc->gfp_mask & __GFP_FS))
75 if (!trylock_super(sb))
78 if (sb->s_op->nr_cached_objects)
79 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
81 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
82 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
83 total_objects = dentries + inodes + fs_objects + 1;
87 /* proportion the scan between the caches */
88 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
89 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
90 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
93 * prune the dcache first as the icache is pinned by it, then
94 * prune the icache, followed by the filesystem specific caches
96 * Ensure that we always scan at least one object - memcg kmem
97 * accounting uses this to fully empty the caches.
99 sc->nr_to_scan = dentries + 1;
100 freed = prune_dcache_sb(sb, sc);
101 sc->nr_to_scan = inodes + 1;
102 freed += prune_icache_sb(sb, sc);
105 sc->nr_to_scan = fs_objects + 1;
106 freed += sb->s_op->free_cached_objects(sb, sc);
109 up_read(&sb->s_umount);
113 static unsigned long super_cache_count(struct shrinker *shrink,
114 struct shrink_control *sc)
116 struct super_block *sb;
117 long total_objects = 0;
119 sb = container_of(shrink, struct super_block, s_shrink);
122 * We don't call trylock_super() here as it is a scalability bottleneck,
123 * so we're exposed to partial setup state. The shrinker rwsem does not
124 * protect filesystem operations backing list_lru_shrink_count() or
125 * s_op->nr_cached_objects(). Counts can change between
126 * super_cache_count and super_cache_scan, so we really don't need locks
129 * However, if we are currently mounting the superblock, the underlying
130 * filesystem might be in a state of partial construction and hence it
131 * is dangerous to access it. trylock_super() uses a MS_BORN check to
132 * avoid this situation, so do the same here. The memory barrier is
133 * matched with the one in mount_fs() as we don't hold locks here.
135 if (!(sb->s_flags & MS_BORN))
139 if (sb->s_op && sb->s_op->nr_cached_objects)
140 total_objects = sb->s_op->nr_cached_objects(sb, sc);
142 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
143 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
145 total_objects = vfs_pressure_ratio(total_objects);
146 return total_objects;
149 static void destroy_super_work(struct work_struct *work)
151 struct super_block *s = container_of(work, struct super_block,
155 for (i = 0; i < SB_FREEZE_LEVELS; i++)
156 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
160 static void destroy_super_rcu(struct rcu_head *head)
162 struct super_block *s = container_of(head, struct super_block, rcu);
163 INIT_WORK(&s->destroy_work, destroy_super_work);
164 schedule_work(&s->destroy_work);
168 * destroy_super - frees a superblock
169 * @s: superblock to free
171 * Frees a superblock.
173 static void destroy_super(struct super_block *s)
175 list_lru_destroy(&s->s_dentry_lru);
176 list_lru_destroy(&s->s_inode_lru);
178 WARN_ON(!list_empty(&s->s_mounts));
179 put_user_ns(s->s_user_ns);
182 call_rcu(&s->rcu, destroy_super_rcu);
186 * alloc_super - create new superblock
187 * @type: filesystem type superblock should belong to
188 * @flags: the mount flags
189 * @user_ns: User namespace for the super_block
191 * Allocates and initializes a new &struct super_block. alloc_super()
192 * returns a pointer new superblock or %NULL if allocation had failed.
194 static struct super_block *alloc_super(struct file_system_type *type, int flags,
195 struct user_namespace *user_ns)
197 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
198 static const struct super_operations default_op;
204 INIT_LIST_HEAD(&s->s_mounts);
205 s->s_user_ns = get_user_ns(user_ns);
207 if (security_sb_alloc(s))
210 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
211 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
213 &type->s_writers_key[i]))
216 init_waitqueue_head(&s->s_writers.wait_unfrozen);
217 s->s_bdi = &noop_backing_dev_info;
219 if (s->s_user_ns != &init_user_ns)
220 s->s_iflags |= SB_I_NODEV;
221 INIT_HLIST_NODE(&s->s_instances);
222 INIT_HLIST_BL_HEAD(&s->s_anon);
223 mutex_init(&s->s_sync_lock);
224 INIT_LIST_HEAD(&s->s_inodes);
225 spin_lock_init(&s->s_inode_list_lock);
226 INIT_LIST_HEAD(&s->s_inodes_wb);
227 spin_lock_init(&s->s_inode_wblist_lock);
229 if (list_lru_init_memcg(&s->s_dentry_lru))
231 if (list_lru_init_memcg(&s->s_inode_lru))
234 init_rwsem(&s->s_umount);
235 lockdep_set_class(&s->s_umount, &type->s_umount_key);
237 * sget() can have s_umount recursion.
239 * When it cannot find a suitable sb, it allocates a new
240 * one (this one), and tries again to find a suitable old
243 * In case that succeeds, it will acquire the s_umount
244 * lock of the old one. Since these are clearly distrinct
245 * locks, and this object isn't exposed yet, there's no
248 * Annotate this by putting this lock in a different
251 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
253 atomic_set(&s->s_active, 1);
254 mutex_init(&s->s_vfs_rename_mutex);
255 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
256 mutex_init(&s->s_dquot.dqio_mutex);
257 mutex_init(&s->s_dquot.dqonoff_mutex);
258 s->s_maxbytes = MAX_NON_LFS;
259 s->s_op = &default_op;
260 s->s_time_gran = 1000000000;
261 s->cleancache_poolid = CLEANCACHE_NO_POOL;
263 s->s_shrink.seeks = DEFAULT_SEEKS;
264 s->s_shrink.scan_objects = super_cache_scan;
265 s->s_shrink.count_objects = super_cache_count;
266 s->s_shrink.batch = 1024;
267 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
275 /* Superblock refcounting */
278 * Drop a superblock's refcount. The caller must hold sb_lock.
280 static void __put_super(struct super_block *sb)
282 if (!--sb->s_count) {
283 list_del_init(&sb->s_list);
289 * put_super - drop a temporary reference to superblock
290 * @sb: superblock in question
292 * Drops a temporary reference, frees superblock if there's no
295 static void put_super(struct super_block *sb)
299 spin_unlock(&sb_lock);
304 * deactivate_locked_super - drop an active reference to superblock
305 * @s: superblock to deactivate
307 * Drops an active reference to superblock, converting it into a temporary
308 * one if there is no other active references left. In that case we
309 * tell fs driver to shut it down and drop the temporary reference we
312 * Caller holds exclusive lock on superblock; that lock is released.
314 void deactivate_locked_super(struct super_block *s)
316 struct file_system_type *fs = s->s_type;
317 if (atomic_dec_and_test(&s->s_active)) {
318 cleancache_invalidate_fs(s);
319 unregister_shrinker(&s->s_shrink);
323 * Since list_lru_destroy() may sleep, we cannot call it from
324 * put_super(), where we hold the sb_lock. Therefore we destroy
325 * the lru lists right now.
327 list_lru_destroy(&s->s_dentry_lru);
328 list_lru_destroy(&s->s_inode_lru);
333 up_write(&s->s_umount);
337 EXPORT_SYMBOL(deactivate_locked_super);
340 * deactivate_super - drop an active reference to superblock
341 * @s: superblock to deactivate
343 * Variant of deactivate_locked_super(), except that superblock is *not*
344 * locked by caller. If we are going to drop the final active reference,
345 * lock will be acquired prior to that.
347 void deactivate_super(struct super_block *s)
349 if (!atomic_add_unless(&s->s_active, -1, 1)) {
350 down_write(&s->s_umount);
351 deactivate_locked_super(s);
355 EXPORT_SYMBOL(deactivate_super);
358 * grab_super - acquire an active reference
359 * @s: reference we are trying to make active
361 * Tries to acquire an active reference. grab_super() is used when we
362 * had just found a superblock in super_blocks or fs_type->fs_supers
363 * and want to turn it into a full-blown active reference. grab_super()
364 * is called with sb_lock held and drops it. Returns 1 in case of
365 * success, 0 if we had failed (superblock contents was already dead or
366 * dying when grab_super() had been called). Note that this is only
367 * called for superblocks not in rundown mode (== ones still on ->fs_supers
368 * of their type), so increment of ->s_count is OK here.
370 static int grab_super(struct super_block *s) __releases(sb_lock)
373 spin_unlock(&sb_lock);
374 down_write(&s->s_umount);
375 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
379 up_write(&s->s_umount);
385 * trylock_super - try to grab ->s_umount shared
386 * @sb: reference we are trying to grab
388 * Try to prevent fs shutdown. This is used in places where we
389 * cannot take an active reference but we need to ensure that the
390 * filesystem is not shut down while we are working on it. It returns
391 * false if we cannot acquire s_umount or if we lose the race and
392 * filesystem already got into shutdown, and returns true with the s_umount
393 * lock held in read mode in case of success. On successful return,
394 * the caller must drop the s_umount lock when done.
396 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
397 * The reason why it's safe is that we are OK with doing trylock instead
398 * of down_read(). There's a couple of places that are OK with that, but
399 * it's very much not a general-purpose interface.
401 bool trylock_super(struct super_block *sb)
403 if (down_read_trylock(&sb->s_umount)) {
404 if (!hlist_unhashed(&sb->s_instances) &&
405 sb->s_root && (sb->s_flags & MS_BORN))
407 up_read(&sb->s_umount);
414 * generic_shutdown_super - common helper for ->kill_sb()
415 * @sb: superblock to kill
417 * generic_shutdown_super() does all fs-independent work on superblock
418 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
419 * that need destruction out of superblock, call generic_shutdown_super()
420 * and release aforementioned objects. Note: dentries and inodes _are_
421 * taken care of and do not need specific handling.
423 * Upon calling this function, the filesystem may no longer alter or
424 * rearrange the set of dentries belonging to this super_block, nor may it
425 * change the attachments of dentries to inodes.
427 void generic_shutdown_super(struct super_block *sb)
429 const struct super_operations *sop = sb->s_op;
432 shrink_dcache_for_umount(sb);
434 sb->s_flags &= ~MS_ACTIVE;
436 fsnotify_unmount_inodes(sb);
437 cgroup_writeback_umount();
441 if (sb->s_dio_done_wq) {
442 destroy_workqueue(sb->s_dio_done_wq);
443 sb->s_dio_done_wq = NULL;
449 if (!list_empty(&sb->s_inodes)) {
450 printk("VFS: Busy inodes after unmount of %s. "
451 "Self-destruct in 5 seconds. Have a nice day...\n",
456 /* should be initialized for __put_super_and_need_restart() */
457 hlist_del_init(&sb->s_instances);
458 spin_unlock(&sb_lock);
459 up_write(&sb->s_umount);
462 EXPORT_SYMBOL(generic_shutdown_super);
465 * sget_userns - find or create a superblock
466 * @type: filesystem type superblock should belong to
467 * @test: comparison callback
468 * @set: setup callback
469 * @flags: mount flags
470 * @user_ns: User namespace for the super_block
471 * @data: argument to each of them
473 struct super_block *sget_userns(struct file_system_type *type,
474 int (*test)(struct super_block *,void *),
475 int (*set)(struct super_block *,void *),
476 int flags, struct user_namespace *user_ns,
479 struct super_block *s = NULL;
480 struct super_block *old;
483 if (!(flags & (MS_KERNMOUNT|MS_SUBMOUNT)) &&
484 !(type->fs_flags & FS_USERNS_MOUNT) &&
485 !capable(CAP_SYS_ADMIN))
486 return ERR_PTR(-EPERM);
490 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
491 if (!test(old, data))
493 if (user_ns != old->s_user_ns) {
494 spin_unlock(&sb_lock);
496 up_write(&s->s_umount);
499 return ERR_PTR(-EBUSY);
501 if (!grab_super(old))
504 up_write(&s->s_umount);
512 spin_unlock(&sb_lock);
513 s = alloc_super(type, (flags & ~MS_SUBMOUNT), user_ns);
515 return ERR_PTR(-ENOMEM);
521 spin_unlock(&sb_lock);
522 up_write(&s->s_umount);
527 strlcpy(s->s_id, type->name, sizeof(s->s_id));
528 list_add_tail(&s->s_list, &super_blocks);
529 hlist_add_head(&s->s_instances, &type->fs_supers);
530 spin_unlock(&sb_lock);
531 get_filesystem(type);
532 err = register_shrinker(&s->s_shrink);
534 deactivate_locked_super(s);
540 EXPORT_SYMBOL(sget_userns);
543 * sget - find or create a superblock
544 * @type: filesystem type superblock should belong to
545 * @test: comparison callback
546 * @set: setup callback
547 * @flags: mount flags
548 * @data: argument to each of them
550 struct super_block *sget(struct file_system_type *type,
551 int (*test)(struct super_block *,void *),
552 int (*set)(struct super_block *,void *),
556 struct user_namespace *user_ns = current_user_ns();
558 /* We don't yet pass the user namespace of the parent
559 * mount through to here so always use &init_user_ns
560 * until that changes.
562 if (flags & MS_SUBMOUNT)
563 user_ns = &init_user_ns;
565 /* Ensure the requestor has permissions over the target filesystem */
566 if (!(flags & (MS_KERNMOUNT|MS_SUBMOUNT)) && !ns_capable(user_ns, CAP_SYS_ADMIN))
567 return ERR_PTR(-EPERM);
569 return sget_userns(type, test, set, flags, user_ns, data);
574 void drop_super(struct super_block *sb)
576 up_read(&sb->s_umount);
580 EXPORT_SYMBOL(drop_super);
583 * iterate_supers - call function for all active superblocks
584 * @f: function to call
585 * @arg: argument to pass to it
587 * Scans the superblock list and calls given function, passing it
588 * locked superblock and given argument.
590 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
592 struct super_block *sb, *p = NULL;
595 list_for_each_entry(sb, &super_blocks, s_list) {
596 if (hlist_unhashed(&sb->s_instances))
599 spin_unlock(&sb_lock);
601 down_read(&sb->s_umount);
602 if (sb->s_root && (sb->s_flags & MS_BORN))
604 up_read(&sb->s_umount);
613 spin_unlock(&sb_lock);
617 * iterate_supers_type - call function for superblocks of given type
619 * @f: function to call
620 * @arg: argument to pass to it
622 * Scans the superblock list and calls given function, passing it
623 * locked superblock and given argument.
625 void iterate_supers_type(struct file_system_type *type,
626 void (*f)(struct super_block *, void *), void *arg)
628 struct super_block *sb, *p = NULL;
631 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
633 spin_unlock(&sb_lock);
635 down_read(&sb->s_umount);
636 if (sb->s_root && (sb->s_flags & MS_BORN))
638 up_read(&sb->s_umount);
647 spin_unlock(&sb_lock);
650 EXPORT_SYMBOL(iterate_supers_type);
653 * get_super - get the superblock of a device
654 * @bdev: device to get the superblock for
656 * Scans the superblock list and finds the superblock of the file system
657 * mounted on the device given. %NULL is returned if no match is found.
660 struct super_block *get_super(struct block_device *bdev)
662 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_bdev == bdev) {
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);
690 EXPORT_SYMBOL(get_super);
693 * get_super_thawed - get thawed superblock of a device
694 * @bdev: device to get the superblock for
696 * Scans the superblock list and finds the superblock of the file system
697 * mounted on the device. The superblock is returned once it is thawed
698 * (or immediately if it was not frozen). %NULL is returned if no match
701 struct super_block *get_super_thawed(struct block_device *bdev)
704 struct super_block *s = get_super(bdev);
705 if (!s || s->s_writers.frozen == SB_UNFROZEN)
707 up_read(&s->s_umount);
708 wait_event(s->s_writers.wait_unfrozen,
709 s->s_writers.frozen == SB_UNFROZEN);
713 EXPORT_SYMBOL(get_super_thawed);
716 * get_active_super - get an active reference to the superblock of a device
717 * @bdev: device to get the superblock for
719 * Scans the superblock list and finds the superblock of the file system
720 * mounted on the device given. Returns the superblock with an active
721 * reference or %NULL if none was found.
723 struct super_block *get_active_super(struct block_device *bdev)
725 struct super_block *sb;
732 list_for_each_entry(sb, &super_blocks, s_list) {
733 if (hlist_unhashed(&sb->s_instances))
735 if (sb->s_bdev == bdev) {
738 up_write(&sb->s_umount);
742 spin_unlock(&sb_lock);
746 struct super_block *user_get_super(dev_t dev)
748 struct super_block *sb;
752 list_for_each_entry(sb, &super_blocks, s_list) {
753 if (hlist_unhashed(&sb->s_instances))
755 if (sb->s_dev == dev) {
757 spin_unlock(&sb_lock);
758 down_read(&sb->s_umount);
760 if (sb->s_root && (sb->s_flags & MS_BORN))
762 up_read(&sb->s_umount);
763 /* nope, got unmounted */
769 spin_unlock(&sb_lock);
774 * do_remount_sb2 - asks filesystem to change mount options.
775 * @mnt: mount we are looking at
776 * @sb: superblock in question
777 * @flags: numeric part of options
778 * @data: the rest of options
779 * @force: whether or not to force the change
781 * Alters the mount options of a mounted file system.
783 int do_remount_sb2(struct vfsmount *mnt, struct super_block *sb, int flags, void *data, int force)
788 if (sb->s_writers.frozen != SB_UNFROZEN)
792 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
796 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
799 if (!hlist_empty(&sb->s_pins)) {
800 up_write(&sb->s_umount);
801 group_pin_kill(&sb->s_pins);
802 down_write(&sb->s_umount);
805 if (sb->s_writers.frozen != SB_UNFROZEN)
807 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
810 shrink_dcache_sb(sb);
812 /* If we are remounting RDONLY and current sb is read/write,
813 make sure there are no rw files opened */
816 sb->s_readonly_remount = 1;
819 retval = sb_prepare_remount_readonly(sb);
825 if (mnt && sb->s_op->remount_fs2) {
826 retval = sb->s_op->remount_fs2(mnt, sb, &flags, data);
829 goto cancel_readonly;
830 /* If forced remount, go ahead despite any errors */
831 WARN(1, "forced remount of a %s fs returned %i\n",
832 sb->s_type->name, retval);
834 } else if (sb->s_op->remount_fs) {
835 retval = sb->s_op->remount_fs(sb, &flags, data);
838 goto cancel_readonly;
839 /* If forced remount, go ahead despite any errors */
840 WARN(1, "forced remount of a %s fs returned %i\n",
841 sb->s_type->name, retval);
844 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
845 /* Needs to be ordered wrt mnt_is_readonly() */
847 sb->s_readonly_remount = 0;
850 * Some filesystems modify their metadata via some other path than the
851 * bdev buffer cache (eg. use a private mapping, or directories in
852 * pagecache, etc). Also file data modifications go via their own
853 * mappings. So If we try to mount readonly then copy the filesystem
854 * from bdev, we could get stale data, so invalidate it to give a best
855 * effort at coherency.
857 if (remount_ro && sb->s_bdev)
858 invalidate_bdev(sb->s_bdev);
862 sb->s_readonly_remount = 0;
866 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
868 return do_remount_sb2(NULL, sb, flags, data, force);
871 static void do_emergency_remount(struct work_struct *work)
873 struct super_block *sb, *p = NULL;
876 list_for_each_entry_reverse(sb, &super_blocks, s_list) {
877 if (hlist_unhashed(&sb->s_instances))
880 spin_unlock(&sb_lock);
881 down_write(&sb->s_umount);
882 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
883 !(sb->s_flags & MS_RDONLY)) {
885 * What lock protects sb->s_flags??
887 do_remount_sb(sb, MS_RDONLY, NULL, 1);
889 up_write(&sb->s_umount);
897 spin_unlock(&sb_lock);
899 printk("Emergency Remount complete\n");
902 void emergency_remount(void)
904 struct work_struct *work;
906 work = kmalloc(sizeof(*work), GFP_ATOMIC);
908 INIT_WORK(work, do_emergency_remount);
914 * Unnamed block devices are dummy devices used by virtual
915 * filesystems which don't use real block-devices. -- jrs
918 static DEFINE_IDA(unnamed_dev_ida);
919 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
920 /* Many userspace utilities consider an FSID of 0 invalid.
921 * Always return at least 1 from get_anon_bdev.
923 static int unnamed_dev_start = 1;
925 int get_anon_bdev(dev_t *p)
931 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
933 spin_lock(&unnamed_dev_lock);
934 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
936 unnamed_dev_start = dev + 1;
937 spin_unlock(&unnamed_dev_lock);
938 if (error == -EAGAIN)
939 /* We raced and lost with another CPU. */
944 if (dev >= (1 << MINORBITS)) {
945 spin_lock(&unnamed_dev_lock);
946 ida_remove(&unnamed_dev_ida, dev);
947 if (unnamed_dev_start > dev)
948 unnamed_dev_start = dev;
949 spin_unlock(&unnamed_dev_lock);
952 *p = MKDEV(0, dev & MINORMASK);
955 EXPORT_SYMBOL(get_anon_bdev);
957 void free_anon_bdev(dev_t dev)
959 int slot = MINOR(dev);
960 spin_lock(&unnamed_dev_lock);
961 ida_remove(&unnamed_dev_ida, slot);
962 if (slot < unnamed_dev_start)
963 unnamed_dev_start = slot;
964 spin_unlock(&unnamed_dev_lock);
966 EXPORT_SYMBOL(free_anon_bdev);
968 int set_anon_super(struct super_block *s, void *data)
970 return get_anon_bdev(&s->s_dev);
973 EXPORT_SYMBOL(set_anon_super);
975 void kill_anon_super(struct super_block *sb)
977 dev_t dev = sb->s_dev;
978 generic_shutdown_super(sb);
982 EXPORT_SYMBOL(kill_anon_super);
984 void kill_litter_super(struct super_block *sb)
987 d_genocide(sb->s_root);
991 EXPORT_SYMBOL(kill_litter_super);
993 static int ns_test_super(struct super_block *sb, void *data)
995 return sb->s_fs_info == data;
998 static int ns_set_super(struct super_block *sb, void *data)
1000 sb->s_fs_info = data;
1001 return set_anon_super(sb, NULL);
1004 struct dentry *mount_ns(struct file_system_type *fs_type,
1005 int flags, void *data, void *ns, struct user_namespace *user_ns,
1006 int (*fill_super)(struct super_block *, void *, int))
1008 struct super_block *sb;
1010 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
1011 * over the namespace.
1013 if (!(flags & MS_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
1014 return ERR_PTR(-EPERM);
1016 sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
1019 return ERR_CAST(sb);
1023 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
1025 deactivate_locked_super(sb);
1026 return ERR_PTR(err);
1029 sb->s_flags |= MS_ACTIVE;
1032 return dget(sb->s_root);
1035 EXPORT_SYMBOL(mount_ns);
1038 static int set_bdev_super(struct super_block *s, void *data)
1041 s->s_dev = s->s_bdev->bd_dev;
1044 * We set the bdi here to the queue backing, file systems can
1045 * overwrite this in ->fill_super()
1047 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1051 static int test_bdev_super(struct super_block *s, void *data)
1053 return (void *)s->s_bdev == data;
1056 struct dentry *mount_bdev(struct file_system_type *fs_type,
1057 int flags, const char *dev_name, void *data,
1058 int (*fill_super)(struct super_block *, void *, int))
1060 struct block_device *bdev;
1061 struct super_block *s;
1062 fmode_t mode = FMODE_READ | FMODE_EXCL;
1065 if (!(flags & MS_RDONLY))
1066 mode |= FMODE_WRITE;
1068 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1070 return ERR_CAST(bdev);
1073 * once the super is inserted into the list by sget, s_umount
1074 * will protect the lockfs code from trying to start a snapshot
1075 * while we are mounting
1077 mutex_lock(&bdev->bd_fsfreeze_mutex);
1078 if (bdev->bd_fsfreeze_count > 0) {
1079 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1083 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
1085 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1090 if ((flags ^ s->s_flags) & MS_RDONLY) {
1091 deactivate_locked_super(s);
1097 * s_umount nests inside bd_mutex during
1098 * __invalidate_device(). blkdev_put() acquires
1099 * bd_mutex and can't be called under s_umount. Drop
1100 * s_umount temporarily. This is safe as we're
1101 * holding an active reference.
1103 up_write(&s->s_umount);
1104 blkdev_put(bdev, mode);
1105 down_write(&s->s_umount);
1108 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1109 sb_set_blocksize(s, block_size(bdev));
1110 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1112 deactivate_locked_super(s);
1116 s->s_flags |= MS_ACTIVE;
1120 return dget(s->s_root);
1125 blkdev_put(bdev, mode);
1127 return ERR_PTR(error);
1129 EXPORT_SYMBOL(mount_bdev);
1131 void kill_block_super(struct super_block *sb)
1133 struct block_device *bdev = sb->s_bdev;
1134 fmode_t mode = sb->s_mode;
1136 bdev->bd_super = NULL;
1137 generic_shutdown_super(sb);
1138 sync_blockdev(bdev);
1139 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1140 blkdev_put(bdev, mode | FMODE_EXCL);
1143 EXPORT_SYMBOL(kill_block_super);
1146 struct dentry *mount_nodev(struct file_system_type *fs_type,
1147 int flags, void *data,
1148 int (*fill_super)(struct super_block *, void *, int))
1151 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1156 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1158 deactivate_locked_super(s);
1159 return ERR_PTR(error);
1161 s->s_flags |= MS_ACTIVE;
1162 return dget(s->s_root);
1164 EXPORT_SYMBOL(mount_nodev);
1166 static int compare_single(struct super_block *s, void *p)
1171 struct dentry *mount_single(struct file_system_type *fs_type,
1172 int flags, void *data,
1173 int (*fill_super)(struct super_block *, void *, int))
1175 struct super_block *s;
1178 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1182 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1184 deactivate_locked_super(s);
1185 return ERR_PTR(error);
1187 s->s_flags |= MS_ACTIVE;
1189 do_remount_sb(s, flags, data, 0);
1191 return dget(s->s_root);
1193 EXPORT_SYMBOL(mount_single);
1196 mount_fs(struct file_system_type *type, int flags, const char *name, struct vfsmount *mnt, void *data)
1198 struct dentry *root;
1199 struct super_block *sb;
1200 char *secdata = NULL;
1201 int error = -ENOMEM;
1203 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1204 secdata = alloc_secdata();
1208 error = security_sb_copy_data(data, secdata);
1210 goto out_free_secdata;
1214 root = type->mount2(mnt, type, flags, name, data);
1216 root = type->mount(type, flags, name, data);
1218 error = PTR_ERR(root);
1219 goto out_free_secdata;
1223 WARN_ON(!sb->s_bdi);
1226 * Write barrier is for super_cache_count(). We place it before setting
1227 * MS_BORN as the data dependency between the two functions is the
1228 * superblock structure contents that we just set up, not the MS_BORN
1232 sb->s_flags |= MS_BORN;
1234 error = security_sb_kern_mount(sb, flags, secdata);
1239 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1240 * but s_maxbytes was an unsigned long long for many releases. Throw
1241 * this warning for a little while to try and catch filesystems that
1242 * violate this rule.
1244 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1245 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1247 up_write(&sb->s_umount);
1248 free_secdata(secdata);
1252 deactivate_locked_super(sb);
1254 free_secdata(secdata);
1256 return ERR_PTR(error);
1260 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1263 void __sb_end_write(struct super_block *sb, int level)
1265 percpu_up_read(sb->s_writers.rw_sem + level-1);
1267 EXPORT_SYMBOL(__sb_end_write);
1270 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1273 int __sb_start_write(struct super_block *sb, int level, bool wait)
1275 bool force_trylock = false;
1278 #ifdef CONFIG_LOCKDEP
1280 * We want lockdep to tell us about possible deadlocks with freezing
1281 * but it's it bit tricky to properly instrument it. Getting a freeze
1282 * protection works as getting a read lock but there are subtle
1283 * problems. XFS for example gets freeze protection on internal level
1284 * twice in some cases, which is OK only because we already hold a
1285 * freeze protection also on higher level. Due to these cases we have
1286 * to use wait == F (trylock mode) which must not fail.
1291 for (i = 0; i < level - 1; i++)
1292 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
1293 force_trylock = true;
1298 if (wait && !force_trylock)
1299 percpu_down_read(sb->s_writers.rw_sem + level-1);
1301 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1303 WARN_ON(force_trylock && !ret);
1306 EXPORT_SYMBOL(__sb_start_write);
1309 * sb_wait_write - wait until all writers to given file system finish
1310 * @sb: the super for which we wait
1311 * @level: type of writers we wait for (normal vs page fault)
1313 * This function waits until there are no writers of given type to given file
1316 static void sb_wait_write(struct super_block *sb, int level)
1318 percpu_down_write(sb->s_writers.rw_sem + level-1);
1322 * We are going to return to userspace and forget about these locks, the
1323 * ownership goes to the caller of thaw_super() which does unlock().
1325 static void lockdep_sb_freeze_release(struct super_block *sb)
1329 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1330 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1334 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1336 static void lockdep_sb_freeze_acquire(struct super_block *sb)
1340 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1341 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1344 static void sb_freeze_unlock(struct super_block *sb)
1348 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1349 percpu_up_write(sb->s_writers.rw_sem + level);
1353 * freeze_super - lock the filesystem and force it into a consistent state
1354 * @sb: the super to lock
1356 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1357 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1360 * During this function, sb->s_writers.frozen goes through these values:
1362 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1364 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1365 * writes should be blocked, though page faults are still allowed. We wait for
1366 * all writes to complete and then proceed to the next stage.
1368 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1369 * but internal fs threads can still modify the filesystem (although they
1370 * should not dirty new pages or inodes), writeback can run etc. After waiting
1371 * for all running page faults we sync the filesystem which will clean all
1372 * dirty pages and inodes (no new dirty pages or inodes can be created when
1375 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1376 * modification are blocked (e.g. XFS preallocation truncation on inode
1377 * reclaim). This is usually implemented by blocking new transactions for
1378 * filesystems that have them and need this additional guard. After all
1379 * internal writers are finished we call ->freeze_fs() to finish filesystem
1380 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1381 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1383 * sb->s_writers.frozen is protected by sb->s_umount.
1385 int freeze_super(struct super_block *sb)
1389 atomic_inc(&sb->s_active);
1390 down_write(&sb->s_umount);
1391 if (sb->s_writers.frozen != SB_UNFROZEN) {
1392 deactivate_locked_super(sb);
1396 if (!(sb->s_flags & MS_BORN)) {
1397 up_write(&sb->s_umount);
1398 return 0; /* sic - it's "nothing to do" */
1401 if (sb->s_flags & MS_RDONLY) {
1402 /* Nothing to do really... */
1403 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1404 up_write(&sb->s_umount);
1408 sb->s_writers.frozen = SB_FREEZE_WRITE;
1409 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1410 up_write(&sb->s_umount);
1411 sb_wait_write(sb, SB_FREEZE_WRITE);
1412 down_write(&sb->s_umount);
1414 /* Now we go and block page faults... */
1415 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1416 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1418 /* All writers are done so after syncing there won't be dirty data */
1419 sync_filesystem(sb);
1421 /* Now wait for internal filesystem counter */
1422 sb->s_writers.frozen = SB_FREEZE_FS;
1423 sb_wait_write(sb, SB_FREEZE_FS);
1425 if (sb->s_op->freeze_fs) {
1426 ret = sb->s_op->freeze_fs(sb);
1429 "VFS:Filesystem freeze failed\n");
1430 sb->s_writers.frozen = SB_UNFROZEN;
1431 sb_freeze_unlock(sb);
1432 wake_up(&sb->s_writers.wait_unfrozen);
1433 deactivate_locked_super(sb);
1438 * For debugging purposes so that fs can warn if it sees write activity
1439 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1441 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1442 lockdep_sb_freeze_release(sb);
1443 up_write(&sb->s_umount);
1446 EXPORT_SYMBOL(freeze_super);
1449 * thaw_super -- unlock filesystem
1450 * @sb: the super to thaw
1452 * Unlocks the filesystem and marks it writeable again after freeze_super().
1454 int thaw_super(struct super_block *sb)
1458 down_write(&sb->s_umount);
1459 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1460 up_write(&sb->s_umount);
1464 if (sb->s_flags & MS_RDONLY) {
1465 sb->s_writers.frozen = SB_UNFROZEN;
1469 lockdep_sb_freeze_acquire(sb);
1471 if (sb->s_op->unfreeze_fs) {
1472 error = sb->s_op->unfreeze_fs(sb);
1475 "VFS:Filesystem thaw failed\n");
1476 lockdep_sb_freeze_release(sb);
1477 up_write(&sb->s_umount);
1482 sb->s_writers.frozen = SB_UNFROZEN;
1483 sb_freeze_unlock(sb);
1485 wake_up(&sb->s_writers.wait_unfrozen);
1486 deactivate_locked_super(sb);
1489 EXPORT_SYMBOL(thaw_super);