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 * Don't call trylock_super as it is a potential
123 * scalability bottleneck. The counts could get updated
124 * between super_cache_count and super_cache_scan anyway.
125 * Call to super_cache_count with shrinker_rwsem held
126 * ensures the safety of call to list_lru_shrink_count() and
127 * s_op->nr_cached_objects().
129 if (sb->s_op && sb->s_op->nr_cached_objects)
130 total_objects = sb->s_op->nr_cached_objects(sb, sc);
132 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
133 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
135 total_objects = vfs_pressure_ratio(total_objects);
136 return total_objects;
139 static void destroy_super_work(struct work_struct *work)
141 struct super_block *s = container_of(work, struct super_block,
145 for (i = 0; i < SB_FREEZE_LEVELS; i++)
146 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
150 static void destroy_super_rcu(struct rcu_head *head)
152 struct super_block *s = container_of(head, struct super_block, rcu);
153 INIT_WORK(&s->destroy_work, destroy_super_work);
154 schedule_work(&s->destroy_work);
158 * destroy_super - frees a superblock
159 * @s: superblock to free
161 * Frees a superblock.
163 static void destroy_super(struct super_block *s)
165 list_lru_destroy(&s->s_dentry_lru);
166 list_lru_destroy(&s->s_inode_lru);
168 WARN_ON(!list_empty(&s->s_mounts));
169 put_user_ns(s->s_user_ns);
172 call_rcu(&s->rcu, destroy_super_rcu);
176 * alloc_super - create new superblock
177 * @type: filesystem type superblock should belong to
178 * @flags: the mount flags
179 * @user_ns: User namespace for the super_block
181 * Allocates and initializes a new &struct super_block. alloc_super()
182 * returns a pointer new superblock or %NULL if allocation had failed.
184 static struct super_block *alloc_super(struct file_system_type *type, int flags,
185 struct user_namespace *user_ns)
187 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
188 static const struct super_operations default_op;
194 INIT_LIST_HEAD(&s->s_mounts);
195 s->s_user_ns = get_user_ns(user_ns);
197 if (security_sb_alloc(s))
200 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
201 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
203 &type->s_writers_key[i]))
206 init_waitqueue_head(&s->s_writers.wait_unfrozen);
207 s->s_bdi = &noop_backing_dev_info;
209 INIT_HLIST_NODE(&s->s_instances);
210 INIT_HLIST_BL_HEAD(&s->s_anon);
211 mutex_init(&s->s_sync_lock);
212 INIT_LIST_HEAD(&s->s_inodes);
213 spin_lock_init(&s->s_inode_list_lock);
215 if (list_lru_init_memcg(&s->s_dentry_lru))
217 if (list_lru_init_memcg(&s->s_inode_lru))
220 init_rwsem(&s->s_umount);
221 lockdep_set_class(&s->s_umount, &type->s_umount_key);
223 * sget() can have s_umount recursion.
225 * When it cannot find a suitable sb, it allocates a new
226 * one (this one), and tries again to find a suitable old
229 * In case that succeeds, it will acquire the s_umount
230 * lock of the old one. Since these are clearly distrinct
231 * locks, and this object isn't exposed yet, there's no
234 * Annotate this by putting this lock in a different
237 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
239 atomic_set(&s->s_active, 1);
240 mutex_init(&s->s_vfs_rename_mutex);
241 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
242 mutex_init(&s->s_dquot.dqio_mutex);
243 mutex_init(&s->s_dquot.dqonoff_mutex);
244 s->s_maxbytes = MAX_NON_LFS;
245 s->s_op = &default_op;
246 s->s_time_gran = 1000000000;
247 s->cleancache_poolid = CLEANCACHE_NO_POOL;
249 s->s_shrink.seeks = DEFAULT_SEEKS;
250 s->s_shrink.scan_objects = super_cache_scan;
251 s->s_shrink.count_objects = super_cache_count;
252 s->s_shrink.batch = 1024;
253 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
261 /* Superblock refcounting */
264 * Drop a superblock's refcount. The caller must hold sb_lock.
266 static void __put_super(struct super_block *sb)
268 if (!--sb->s_count) {
269 list_del_init(&sb->s_list);
275 * put_super - drop a temporary reference to superblock
276 * @sb: superblock in question
278 * Drops a temporary reference, frees superblock if there's no
281 static void put_super(struct super_block *sb)
285 spin_unlock(&sb_lock);
290 * deactivate_locked_super - drop an active reference to superblock
291 * @s: superblock to deactivate
293 * Drops an active reference to superblock, converting it into a temporary
294 * one if there is no other active references left. In that case we
295 * tell fs driver to shut it down and drop the temporary reference we
298 * Caller holds exclusive lock on superblock; that lock is released.
300 void deactivate_locked_super(struct super_block *s)
302 struct file_system_type *fs = s->s_type;
303 if (atomic_dec_and_test(&s->s_active)) {
304 cleancache_invalidate_fs(s);
305 unregister_shrinker(&s->s_shrink);
309 * Since list_lru_destroy() may sleep, we cannot call it from
310 * put_super(), where we hold the sb_lock. Therefore we destroy
311 * the lru lists right now.
313 list_lru_destroy(&s->s_dentry_lru);
314 list_lru_destroy(&s->s_inode_lru);
319 up_write(&s->s_umount);
323 EXPORT_SYMBOL(deactivate_locked_super);
326 * deactivate_super - drop an active reference to superblock
327 * @s: superblock to deactivate
329 * Variant of deactivate_locked_super(), except that superblock is *not*
330 * locked by caller. If we are going to drop the final active reference,
331 * lock will be acquired prior to that.
333 void deactivate_super(struct super_block *s)
335 if (!atomic_add_unless(&s->s_active, -1, 1)) {
336 down_write(&s->s_umount);
337 deactivate_locked_super(s);
341 EXPORT_SYMBOL(deactivate_super);
344 * grab_super - acquire an active reference
345 * @s: reference we are trying to make active
347 * Tries to acquire an active reference. grab_super() is used when we
348 * had just found a superblock in super_blocks or fs_type->fs_supers
349 * and want to turn it into a full-blown active reference. grab_super()
350 * is called with sb_lock held and drops it. Returns 1 in case of
351 * success, 0 if we had failed (superblock contents was already dead or
352 * dying when grab_super() had been called). Note that this is only
353 * called for superblocks not in rundown mode (== ones still on ->fs_supers
354 * of their type), so increment of ->s_count is OK here.
356 static int grab_super(struct super_block *s) __releases(sb_lock)
359 spin_unlock(&sb_lock);
360 down_write(&s->s_umount);
361 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
365 up_write(&s->s_umount);
371 * trylock_super - try to grab ->s_umount shared
372 * @sb: reference we are trying to grab
374 * Try to prevent fs shutdown. This is used in places where we
375 * cannot take an active reference but we need to ensure that the
376 * filesystem is not shut down while we are working on it. It returns
377 * false if we cannot acquire s_umount or if we lose the race and
378 * filesystem already got into shutdown, and returns true with the s_umount
379 * lock held in read mode in case of success. On successful return,
380 * the caller must drop the s_umount lock when done.
382 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
383 * The reason why it's safe is that we are OK with doing trylock instead
384 * of down_read(). There's a couple of places that are OK with that, but
385 * it's very much not a general-purpose interface.
387 bool trylock_super(struct super_block *sb)
389 if (down_read_trylock(&sb->s_umount)) {
390 if (!hlist_unhashed(&sb->s_instances) &&
391 sb->s_root && (sb->s_flags & MS_BORN))
393 up_read(&sb->s_umount);
400 * generic_shutdown_super - common helper for ->kill_sb()
401 * @sb: superblock to kill
403 * generic_shutdown_super() does all fs-independent work on superblock
404 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
405 * that need destruction out of superblock, call generic_shutdown_super()
406 * and release aforementioned objects. Note: dentries and inodes _are_
407 * taken care of and do not need specific handling.
409 * Upon calling this function, the filesystem may no longer alter or
410 * rearrange the set of dentries belonging to this super_block, nor may it
411 * change the attachments of dentries to inodes.
413 void generic_shutdown_super(struct super_block *sb)
415 const struct super_operations *sop = sb->s_op;
418 shrink_dcache_for_umount(sb);
420 sb->s_flags &= ~MS_ACTIVE;
422 fsnotify_unmount_inodes(sb);
423 cgroup_writeback_umount();
427 if (sb->s_dio_done_wq) {
428 destroy_workqueue(sb->s_dio_done_wq);
429 sb->s_dio_done_wq = NULL;
435 if (!list_empty(&sb->s_inodes)) {
436 printk("VFS: Busy inodes after unmount of %s. "
437 "Self-destruct in 5 seconds. Have a nice day...\n",
442 /* should be initialized for __put_super_and_need_restart() */
443 hlist_del_init(&sb->s_instances);
444 spin_unlock(&sb_lock);
445 up_write(&sb->s_umount);
448 EXPORT_SYMBOL(generic_shutdown_super);
451 * sget_userns - find or create a superblock
452 * @type: filesystem type superblock should belong to
453 * @test: comparison callback
454 * @set: setup callback
455 * @flags: mount flags
456 * @user_ns: User namespace for the super_block
457 * @data: argument to each of them
459 struct super_block *sget_userns(struct file_system_type *type,
460 int (*test)(struct super_block *,void *),
461 int (*set)(struct super_block *,void *),
462 int flags, struct user_namespace *user_ns,
465 struct super_block *s = NULL;
466 struct super_block *old;
472 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
473 if (!test(old, data))
475 if (user_ns != old->s_user_ns) {
476 spin_unlock(&sb_lock);
478 up_write(&s->s_umount);
481 return ERR_PTR(-EBUSY);
483 if (!grab_super(old))
486 up_write(&s->s_umount);
494 spin_unlock(&sb_lock);
495 s = alloc_super(type, flags, user_ns);
497 return ERR_PTR(-ENOMEM);
503 spin_unlock(&sb_lock);
504 up_write(&s->s_umount);
509 strlcpy(s->s_id, type->name, sizeof(s->s_id));
510 list_add_tail(&s->s_list, &super_blocks);
511 hlist_add_head(&s->s_instances, &type->fs_supers);
512 spin_unlock(&sb_lock);
513 get_filesystem(type);
514 register_shrinker(&s->s_shrink);
518 EXPORT_SYMBOL(sget_userns);
521 * sget - find or create a superblock
522 * @type: filesystem type superblock should belong to
523 * @test: comparison callback
524 * @set: setup callback
525 * @flags: mount flags
526 * @data: argument to each of them
528 struct super_block *sget(struct file_system_type *type,
529 int (*test)(struct super_block *,void *),
530 int (*set)(struct super_block *,void *),
534 struct user_namespace *user_ns = current_user_ns();
536 /* Ensure the requestor has permissions over the target filesystem */
537 if (!(flags & MS_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
538 return ERR_PTR(-EPERM);
540 return sget_userns(type, test, set, flags, user_ns, data);
545 void drop_super(struct super_block *sb)
547 up_read(&sb->s_umount);
551 EXPORT_SYMBOL(drop_super);
554 * iterate_supers - call function for all active superblocks
555 * @f: function to call
556 * @arg: argument to pass to it
558 * Scans the superblock list and calls given function, passing it
559 * locked superblock and given argument.
561 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
563 struct super_block *sb, *p = NULL;
566 list_for_each_entry(sb, &super_blocks, s_list) {
567 if (hlist_unhashed(&sb->s_instances))
570 spin_unlock(&sb_lock);
572 down_read(&sb->s_umount);
573 if (sb->s_root && (sb->s_flags & MS_BORN))
575 up_read(&sb->s_umount);
584 spin_unlock(&sb_lock);
588 * iterate_supers_type - call function for superblocks of given type
590 * @f: function to call
591 * @arg: argument to pass to it
593 * Scans the superblock list and calls given function, passing it
594 * locked superblock and given argument.
596 void iterate_supers_type(struct file_system_type *type,
597 void (*f)(struct super_block *, void *), void *arg)
599 struct super_block *sb, *p = NULL;
602 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
604 spin_unlock(&sb_lock);
606 down_read(&sb->s_umount);
607 if (sb->s_root && (sb->s_flags & MS_BORN))
609 up_read(&sb->s_umount);
618 spin_unlock(&sb_lock);
621 EXPORT_SYMBOL(iterate_supers_type);
624 * get_super - get the 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 given. %NULL is returned if no match is found.
631 struct super_block *get_super(struct block_device *bdev)
633 struct super_block *sb;
640 list_for_each_entry(sb, &super_blocks, s_list) {
641 if (hlist_unhashed(&sb->s_instances))
643 if (sb->s_bdev == bdev) {
645 spin_unlock(&sb_lock);
646 down_read(&sb->s_umount);
648 if (sb->s_root && (sb->s_flags & MS_BORN))
650 up_read(&sb->s_umount);
651 /* nope, got unmounted */
657 spin_unlock(&sb_lock);
661 EXPORT_SYMBOL(get_super);
664 * get_super_thawed - get thawed superblock of a device
665 * @bdev: device to get the superblock for
667 * Scans the superblock list and finds the superblock of the file system
668 * mounted on the device. The superblock is returned once it is thawed
669 * (or immediately if it was not frozen). %NULL is returned if no match
672 struct super_block *get_super_thawed(struct block_device *bdev)
675 struct super_block *s = get_super(bdev);
676 if (!s || s->s_writers.frozen == SB_UNFROZEN)
678 up_read(&s->s_umount);
679 wait_event(s->s_writers.wait_unfrozen,
680 s->s_writers.frozen == SB_UNFROZEN);
684 EXPORT_SYMBOL(get_super_thawed);
687 * get_active_super - get an active reference to the superblock of a device
688 * @bdev: device to get the superblock for
690 * Scans the superblock list and finds the superblock of the file system
691 * mounted on the device given. Returns the superblock with an active
692 * reference or %NULL if none was found.
694 struct super_block *get_active_super(struct block_device *bdev)
696 struct super_block *sb;
703 list_for_each_entry(sb, &super_blocks, s_list) {
704 if (hlist_unhashed(&sb->s_instances))
706 if (sb->s_bdev == bdev) {
709 up_write(&sb->s_umount);
713 spin_unlock(&sb_lock);
717 struct super_block *user_get_super(dev_t dev)
719 struct super_block *sb;
723 list_for_each_entry(sb, &super_blocks, s_list) {
724 if (hlist_unhashed(&sb->s_instances))
726 if (sb->s_dev == dev) {
728 spin_unlock(&sb_lock);
729 down_read(&sb->s_umount);
731 if (sb->s_root && (sb->s_flags & MS_BORN))
733 up_read(&sb->s_umount);
734 /* nope, got unmounted */
740 spin_unlock(&sb_lock);
745 * do_remount_sb - asks filesystem to change mount options.
746 * @sb: superblock in question
747 * @flags: numeric part of options
748 * @data: the rest of options
749 * @force: whether or not to force the change
751 * Alters the mount options of a mounted file system.
753 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
758 if (sb->s_writers.frozen != SB_UNFROZEN)
762 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
766 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
769 if (!hlist_empty(&sb->s_pins)) {
770 up_write(&sb->s_umount);
771 group_pin_kill(&sb->s_pins);
772 down_write(&sb->s_umount);
775 if (sb->s_writers.frozen != SB_UNFROZEN)
777 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
780 shrink_dcache_sb(sb);
782 /* If we are remounting RDONLY and current sb is read/write,
783 make sure there are no rw files opened */
786 sb->s_readonly_remount = 1;
789 retval = sb_prepare_remount_readonly(sb);
795 if (sb->s_op->remount_fs) {
796 retval = sb->s_op->remount_fs(sb, &flags, data);
799 goto cancel_readonly;
800 /* If forced remount, go ahead despite any errors */
801 WARN(1, "forced remount of a %s fs returned %i\n",
802 sb->s_type->name, retval);
805 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
806 /* Needs to be ordered wrt mnt_is_readonly() */
808 sb->s_readonly_remount = 0;
811 * Some filesystems modify their metadata via some other path than the
812 * bdev buffer cache (eg. use a private mapping, or directories in
813 * pagecache, etc). Also file data modifications go via their own
814 * mappings. So If we try to mount readonly then copy the filesystem
815 * from bdev, we could get stale data, so invalidate it to give a best
816 * effort at coherency.
818 if (remount_ro && sb->s_bdev)
819 invalidate_bdev(sb->s_bdev);
823 sb->s_readonly_remount = 0;
827 static void do_emergency_remount(struct work_struct *work)
829 struct super_block *sb, *p = NULL;
832 list_for_each_entry(sb, &super_blocks, s_list) {
833 if (hlist_unhashed(&sb->s_instances))
836 spin_unlock(&sb_lock);
837 down_write(&sb->s_umount);
838 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
839 !(sb->s_flags & MS_RDONLY)) {
841 * What lock protects sb->s_flags??
843 do_remount_sb(sb, MS_RDONLY, NULL, 1);
845 up_write(&sb->s_umount);
853 spin_unlock(&sb_lock);
855 printk("Emergency Remount complete\n");
858 void emergency_remount(void)
860 struct work_struct *work;
862 work = kmalloc(sizeof(*work), GFP_ATOMIC);
864 INIT_WORK(work, do_emergency_remount);
870 * Unnamed block devices are dummy devices used by virtual
871 * filesystems which don't use real block-devices. -- jrs
874 static DEFINE_IDA(unnamed_dev_ida);
875 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
876 /* Many userspace utilities consider an FSID of 0 invalid.
877 * Always return at least 1 from get_anon_bdev.
879 static int unnamed_dev_start = 1;
881 int get_anon_bdev(dev_t *p)
887 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
889 spin_lock(&unnamed_dev_lock);
890 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
892 unnamed_dev_start = dev + 1;
893 spin_unlock(&unnamed_dev_lock);
894 if (error == -EAGAIN)
895 /* We raced and lost with another CPU. */
900 if (dev >= (1 << MINORBITS)) {
901 spin_lock(&unnamed_dev_lock);
902 ida_remove(&unnamed_dev_ida, dev);
903 if (unnamed_dev_start > dev)
904 unnamed_dev_start = dev;
905 spin_unlock(&unnamed_dev_lock);
908 *p = MKDEV(0, dev & MINORMASK);
911 EXPORT_SYMBOL(get_anon_bdev);
913 void free_anon_bdev(dev_t dev)
915 int slot = MINOR(dev);
916 spin_lock(&unnamed_dev_lock);
917 ida_remove(&unnamed_dev_ida, slot);
918 if (slot < unnamed_dev_start)
919 unnamed_dev_start = slot;
920 spin_unlock(&unnamed_dev_lock);
922 EXPORT_SYMBOL(free_anon_bdev);
924 int set_anon_super(struct super_block *s, void *data)
926 return get_anon_bdev(&s->s_dev);
929 EXPORT_SYMBOL(set_anon_super);
931 void kill_anon_super(struct super_block *sb)
933 dev_t dev = sb->s_dev;
934 generic_shutdown_super(sb);
938 EXPORT_SYMBOL(kill_anon_super);
940 void kill_litter_super(struct super_block *sb)
943 d_genocide(sb->s_root);
947 EXPORT_SYMBOL(kill_litter_super);
949 static int ns_test_super(struct super_block *sb, void *data)
951 return sb->s_fs_info == data;
954 static int ns_set_super(struct super_block *sb, void *data)
956 sb->s_fs_info = data;
957 return set_anon_super(sb, NULL);
960 struct dentry *mount_ns(struct file_system_type *fs_type,
961 int flags, void *data, void *ns, struct user_namespace *user_ns,
962 int (*fill_super)(struct super_block *, void *, int))
964 struct super_block *sb;
966 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
967 * over the namespace.
969 if (!(flags & MS_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
970 return ERR_PTR(-EPERM);
972 sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
979 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
981 deactivate_locked_super(sb);
985 sb->s_flags |= MS_ACTIVE;
988 return dget(sb->s_root);
991 EXPORT_SYMBOL(mount_ns);
994 static int set_bdev_super(struct super_block *s, void *data)
997 s->s_dev = s->s_bdev->bd_dev;
1000 * We set the bdi here to the queue backing, file systems can
1001 * overwrite this in ->fill_super()
1003 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1007 static int test_bdev_super(struct super_block *s, void *data)
1009 return (void *)s->s_bdev == data;
1012 struct dentry *mount_bdev(struct file_system_type *fs_type,
1013 int flags, const char *dev_name, void *data,
1014 int (*fill_super)(struct super_block *, void *, int))
1016 struct block_device *bdev;
1017 struct super_block *s;
1018 fmode_t mode = FMODE_READ | FMODE_EXCL;
1021 if (!(flags & MS_RDONLY))
1022 mode |= FMODE_WRITE;
1024 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1026 return ERR_CAST(bdev);
1029 * once the super is inserted into the list by sget, s_umount
1030 * will protect the lockfs code from trying to start a snapshot
1031 * while we are mounting
1033 mutex_lock(&bdev->bd_fsfreeze_mutex);
1034 if (bdev->bd_fsfreeze_count > 0) {
1035 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1039 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
1041 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1046 if ((flags ^ s->s_flags) & MS_RDONLY) {
1047 deactivate_locked_super(s);
1053 * s_umount nests inside bd_mutex during
1054 * __invalidate_device(). blkdev_put() acquires
1055 * bd_mutex and can't be called under s_umount. Drop
1056 * s_umount temporarily. This is safe as we're
1057 * holding an active reference.
1059 up_write(&s->s_umount);
1060 blkdev_put(bdev, mode);
1061 down_write(&s->s_umount);
1064 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1065 sb_set_blocksize(s, block_size(bdev));
1066 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1068 deactivate_locked_super(s);
1072 s->s_flags |= MS_ACTIVE;
1076 return dget(s->s_root);
1081 blkdev_put(bdev, mode);
1083 return ERR_PTR(error);
1085 EXPORT_SYMBOL(mount_bdev);
1087 void kill_block_super(struct super_block *sb)
1089 struct block_device *bdev = sb->s_bdev;
1090 fmode_t mode = sb->s_mode;
1092 bdev->bd_super = NULL;
1093 generic_shutdown_super(sb);
1094 sync_blockdev(bdev);
1095 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1096 blkdev_put(bdev, mode | FMODE_EXCL);
1099 EXPORT_SYMBOL(kill_block_super);
1102 struct dentry *mount_nodev(struct file_system_type *fs_type,
1103 int flags, void *data,
1104 int (*fill_super)(struct super_block *, void *, int))
1107 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1112 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1114 deactivate_locked_super(s);
1115 return ERR_PTR(error);
1117 s->s_flags |= MS_ACTIVE;
1118 return dget(s->s_root);
1120 EXPORT_SYMBOL(mount_nodev);
1122 static int compare_single(struct super_block *s, void *p)
1127 struct dentry *mount_single(struct file_system_type *fs_type,
1128 int flags, void *data,
1129 int (*fill_super)(struct super_block *, void *, int))
1131 struct super_block *s;
1134 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1138 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1140 deactivate_locked_super(s);
1141 return ERR_PTR(error);
1143 s->s_flags |= MS_ACTIVE;
1145 do_remount_sb(s, flags, data, 0);
1147 return dget(s->s_root);
1149 EXPORT_SYMBOL(mount_single);
1152 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1154 struct dentry *root;
1155 struct super_block *sb;
1156 char *secdata = NULL;
1157 int error = -ENOMEM;
1159 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1160 secdata = alloc_secdata();
1164 error = security_sb_copy_data(data, secdata);
1166 goto out_free_secdata;
1169 root = type->mount(type, flags, name, data);
1171 error = PTR_ERR(root);
1172 goto out_free_secdata;
1176 WARN_ON(!sb->s_bdi);
1177 sb->s_flags |= MS_BORN;
1179 error = security_sb_kern_mount(sb, flags, secdata);
1184 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1185 * but s_maxbytes was an unsigned long long for many releases. Throw
1186 * this warning for a little while to try and catch filesystems that
1187 * violate this rule.
1189 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1190 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1192 up_write(&sb->s_umount);
1193 free_secdata(secdata);
1197 deactivate_locked_super(sb);
1199 free_secdata(secdata);
1201 return ERR_PTR(error);
1205 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1208 void __sb_end_write(struct super_block *sb, int level)
1210 percpu_up_read(sb->s_writers.rw_sem + level-1);
1212 EXPORT_SYMBOL(__sb_end_write);
1215 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1218 int __sb_start_write(struct super_block *sb, int level, bool wait)
1220 bool force_trylock = false;
1223 #ifdef CONFIG_LOCKDEP
1225 * We want lockdep to tell us about possible deadlocks with freezing
1226 * but it's it bit tricky to properly instrument it. Getting a freeze
1227 * protection works as getting a read lock but there are subtle
1228 * problems. XFS for example gets freeze protection on internal level
1229 * twice in some cases, which is OK only because we already hold a
1230 * freeze protection also on higher level. Due to these cases we have
1231 * to use wait == F (trylock mode) which must not fail.
1236 for (i = 0; i < level - 1; i++)
1237 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
1238 force_trylock = true;
1243 if (wait && !force_trylock)
1244 percpu_down_read(sb->s_writers.rw_sem + level-1);
1246 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1248 WARN_ON(force_trylock && !ret);
1251 EXPORT_SYMBOL(__sb_start_write);
1254 * sb_wait_write - wait until all writers to given file system finish
1255 * @sb: the super for which we wait
1256 * @level: type of writers we wait for (normal vs page fault)
1258 * This function waits until there are no writers of given type to given file
1261 static void sb_wait_write(struct super_block *sb, int level)
1263 percpu_down_write(sb->s_writers.rw_sem + level-1);
1265 * We are going to return to userspace and forget about this lock, the
1266 * ownership goes to the caller of thaw_super() which does unlock.
1268 * FIXME: we should do this before return from freeze_super() after we
1269 * called sync_filesystem(sb) and s_op->freeze_fs(sb), and thaw_super()
1270 * should re-acquire these locks before s_op->unfreeze_fs(sb). However
1271 * this leads to lockdep false-positives, so currently we do the early
1272 * release right after acquire.
1274 percpu_rwsem_release(sb->s_writers.rw_sem + level-1, 0, _THIS_IP_);
1277 static void sb_freeze_unlock(struct super_block *sb)
1281 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1282 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1284 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1285 percpu_up_write(sb->s_writers.rw_sem + level);
1289 * freeze_super - lock the filesystem and force it into a consistent state
1290 * @sb: the super to lock
1292 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1293 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1296 * During this function, sb->s_writers.frozen goes through these values:
1298 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1300 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1301 * writes should be blocked, though page faults are still allowed. We wait for
1302 * all writes to complete and then proceed to the next stage.
1304 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1305 * but internal fs threads can still modify the filesystem (although they
1306 * should not dirty new pages or inodes), writeback can run etc. After waiting
1307 * for all running page faults we sync the filesystem which will clean all
1308 * dirty pages and inodes (no new dirty pages or inodes can be created when
1311 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1312 * modification are blocked (e.g. XFS preallocation truncation on inode
1313 * reclaim). This is usually implemented by blocking new transactions for
1314 * filesystems that have them and need this additional guard. After all
1315 * internal writers are finished we call ->freeze_fs() to finish filesystem
1316 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1317 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1319 * sb->s_writers.frozen is protected by sb->s_umount.
1321 int freeze_super(struct super_block *sb)
1325 atomic_inc(&sb->s_active);
1326 down_write(&sb->s_umount);
1327 if (sb->s_writers.frozen != SB_UNFROZEN) {
1328 deactivate_locked_super(sb);
1332 if (!(sb->s_flags & MS_BORN)) {
1333 up_write(&sb->s_umount);
1334 return 0; /* sic - it's "nothing to do" */
1337 if (sb->s_flags & MS_RDONLY) {
1338 /* Nothing to do really... */
1339 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1340 up_write(&sb->s_umount);
1344 sb->s_writers.frozen = SB_FREEZE_WRITE;
1345 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1346 up_write(&sb->s_umount);
1347 sb_wait_write(sb, SB_FREEZE_WRITE);
1348 down_write(&sb->s_umount);
1350 /* Now we go and block page faults... */
1351 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1352 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1354 /* All writers are done so after syncing there won't be dirty data */
1355 sync_filesystem(sb);
1357 /* Now wait for internal filesystem counter */
1358 sb->s_writers.frozen = SB_FREEZE_FS;
1359 sb_wait_write(sb, SB_FREEZE_FS);
1361 if (sb->s_op->freeze_fs) {
1362 ret = sb->s_op->freeze_fs(sb);
1365 "VFS:Filesystem freeze failed\n");
1366 sb->s_writers.frozen = SB_UNFROZEN;
1367 sb_freeze_unlock(sb);
1368 wake_up(&sb->s_writers.wait_unfrozen);
1369 deactivate_locked_super(sb);
1374 * This is just for debugging purposes so that fs can warn if it
1375 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1377 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1378 up_write(&sb->s_umount);
1381 EXPORT_SYMBOL(freeze_super);
1384 * thaw_super -- unlock filesystem
1385 * @sb: the super to thaw
1387 * Unlocks the filesystem and marks it writeable again after freeze_super().
1389 int thaw_super(struct super_block *sb)
1393 down_write(&sb->s_umount);
1394 if (sb->s_writers.frozen == SB_UNFROZEN) {
1395 up_write(&sb->s_umount);
1399 if (sb->s_flags & MS_RDONLY) {
1400 sb->s_writers.frozen = SB_UNFROZEN;
1404 if (sb->s_op->unfreeze_fs) {
1405 error = sb->s_op->unfreeze_fs(sb);
1408 "VFS:Filesystem thaw failed\n");
1409 up_write(&sb->s_umount);
1414 sb->s_writers.frozen = SB_UNFROZEN;
1415 sb_freeze_unlock(sb);
1417 wake_up(&sb->s_writers.wait_unfrozen);
1418 deactivate_locked_super(sb);
1421 EXPORT_SYMBOL(thaw_super);