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 if (s->s_user_ns != &init_user_ns)
210 s->s_iflags |= SB_I_NODEV;
211 INIT_HLIST_NODE(&s->s_instances);
212 INIT_HLIST_BL_HEAD(&s->s_anon);
213 mutex_init(&s->s_sync_lock);
214 INIT_LIST_HEAD(&s->s_inodes);
215 spin_lock_init(&s->s_inode_list_lock);
216 INIT_LIST_HEAD(&s->s_inodes_wb);
217 spin_lock_init(&s->s_inode_wblist_lock);
219 if (list_lru_init_memcg(&s->s_dentry_lru))
221 if (list_lru_init_memcg(&s->s_inode_lru))
224 init_rwsem(&s->s_umount);
225 lockdep_set_class(&s->s_umount, &type->s_umount_key);
227 * sget() can have s_umount recursion.
229 * When it cannot find a suitable sb, it allocates a new
230 * one (this one), and tries again to find a suitable old
233 * In case that succeeds, it will acquire the s_umount
234 * lock of the old one. Since these are clearly distrinct
235 * locks, and this object isn't exposed yet, there's no
238 * Annotate this by putting this lock in a different
241 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
243 atomic_set(&s->s_active, 1);
244 mutex_init(&s->s_vfs_rename_mutex);
245 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
246 mutex_init(&s->s_dquot.dqio_mutex);
247 s->s_maxbytes = MAX_NON_LFS;
248 s->s_op = &default_op;
249 s->s_time_gran = 1000000000;
250 s->cleancache_poolid = CLEANCACHE_NO_POOL;
252 s->s_shrink.seeks = DEFAULT_SEEKS;
253 s->s_shrink.scan_objects = super_cache_scan;
254 s->s_shrink.count_objects = super_cache_count;
255 s->s_shrink.batch = 1024;
256 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
264 /* Superblock refcounting */
267 * Drop a superblock's refcount. The caller must hold sb_lock.
269 static void __put_super(struct super_block *sb)
271 if (!--sb->s_count) {
272 list_del_init(&sb->s_list);
278 * put_super - drop a temporary reference to superblock
279 * @sb: superblock in question
281 * Drops a temporary reference, frees superblock if there's no
284 static void put_super(struct super_block *sb)
288 spin_unlock(&sb_lock);
293 * deactivate_locked_super - drop an active reference to superblock
294 * @s: superblock to deactivate
296 * Drops an active reference to superblock, converting it into a temporary
297 * one if there is no other active references left. In that case we
298 * tell fs driver to shut it down and drop the temporary reference we
301 * Caller holds exclusive lock on superblock; that lock is released.
303 void deactivate_locked_super(struct super_block *s)
305 struct file_system_type *fs = s->s_type;
306 if (atomic_dec_and_test(&s->s_active)) {
307 cleancache_invalidate_fs(s);
308 unregister_shrinker(&s->s_shrink);
312 * Since list_lru_destroy() may sleep, we cannot call it from
313 * put_super(), where we hold the sb_lock. Therefore we destroy
314 * the lru lists right now.
316 list_lru_destroy(&s->s_dentry_lru);
317 list_lru_destroy(&s->s_inode_lru);
322 up_write(&s->s_umount);
326 EXPORT_SYMBOL(deactivate_locked_super);
329 * deactivate_super - drop an active reference to superblock
330 * @s: superblock to deactivate
332 * Variant of deactivate_locked_super(), except that superblock is *not*
333 * locked by caller. If we are going to drop the final active reference,
334 * lock will be acquired prior to that.
336 void deactivate_super(struct super_block *s)
338 if (!atomic_add_unless(&s->s_active, -1, 1)) {
339 down_write(&s->s_umount);
340 deactivate_locked_super(s);
344 EXPORT_SYMBOL(deactivate_super);
347 * grab_super - acquire an active reference
348 * @s: reference we are trying to make active
350 * Tries to acquire an active reference. grab_super() is used when we
351 * had just found a superblock in super_blocks or fs_type->fs_supers
352 * and want to turn it into a full-blown active reference. grab_super()
353 * is called with sb_lock held and drops it. Returns 1 in case of
354 * success, 0 if we had failed (superblock contents was already dead or
355 * dying when grab_super() had been called). Note that this is only
356 * called for superblocks not in rundown mode (== ones still on ->fs_supers
357 * of their type), so increment of ->s_count is OK here.
359 static int grab_super(struct super_block *s) __releases(sb_lock)
362 spin_unlock(&sb_lock);
363 down_write(&s->s_umount);
364 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
368 up_write(&s->s_umount);
374 * trylock_super - try to grab ->s_umount shared
375 * @sb: reference we are trying to grab
377 * Try to prevent fs shutdown. This is used in places where we
378 * cannot take an active reference but we need to ensure that the
379 * filesystem is not shut down while we are working on it. It returns
380 * false if we cannot acquire s_umount or if we lose the race and
381 * filesystem already got into shutdown, and returns true with the s_umount
382 * lock held in read mode in case of success. On successful return,
383 * the caller must drop the s_umount lock when done.
385 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
386 * The reason why it's safe is that we are OK with doing trylock instead
387 * of down_read(). There's a couple of places that are OK with that, but
388 * it's very much not a general-purpose interface.
390 bool trylock_super(struct super_block *sb)
392 if (down_read_trylock(&sb->s_umount)) {
393 if (!hlist_unhashed(&sb->s_instances) &&
394 sb->s_root && (sb->s_flags & MS_BORN))
396 up_read(&sb->s_umount);
403 * generic_shutdown_super - common helper for ->kill_sb()
404 * @sb: superblock to kill
406 * generic_shutdown_super() does all fs-independent work on superblock
407 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
408 * that need destruction out of superblock, call generic_shutdown_super()
409 * and release aforementioned objects. Note: dentries and inodes _are_
410 * taken care of and do not need specific handling.
412 * Upon calling this function, the filesystem may no longer alter or
413 * rearrange the set of dentries belonging to this super_block, nor may it
414 * change the attachments of dentries to inodes.
416 void generic_shutdown_super(struct super_block *sb)
418 const struct super_operations *sop = sb->s_op;
421 shrink_dcache_for_umount(sb);
423 sb->s_flags &= ~MS_ACTIVE;
425 fsnotify_unmount_inodes(sb);
426 cgroup_writeback_umount();
430 if (sb->s_dio_done_wq) {
431 destroy_workqueue(sb->s_dio_done_wq);
432 sb->s_dio_done_wq = NULL;
438 if (!list_empty(&sb->s_inodes)) {
439 printk("VFS: Busy inodes after unmount of %s. "
440 "Self-destruct in 5 seconds. Have a nice day...\n",
445 /* should be initialized for __put_super_and_need_restart() */
446 hlist_del_init(&sb->s_instances);
447 spin_unlock(&sb_lock);
448 up_write(&sb->s_umount);
449 if (sb->s_bdi != &noop_backing_dev_info) {
451 sb->s_bdi = &noop_backing_dev_info;
455 EXPORT_SYMBOL(generic_shutdown_super);
458 * sget_userns - find or create a superblock
459 * @type: filesystem type superblock should belong to
460 * @test: comparison callback
461 * @set: setup callback
462 * @flags: mount flags
463 * @user_ns: User namespace for the super_block
464 * @data: argument to each of them
466 struct super_block *sget_userns(struct file_system_type *type,
467 int (*test)(struct super_block *,void *),
468 int (*set)(struct super_block *,void *),
469 int flags, struct user_namespace *user_ns,
472 struct super_block *s = NULL;
473 struct super_block *old;
476 if (!(flags & (MS_KERNMOUNT|MS_SUBMOUNT)) &&
477 !(type->fs_flags & FS_USERNS_MOUNT) &&
478 !capable(CAP_SYS_ADMIN))
479 return ERR_PTR(-EPERM);
483 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
484 if (!test(old, data))
486 if (user_ns != old->s_user_ns) {
487 spin_unlock(&sb_lock);
489 up_write(&s->s_umount);
492 return ERR_PTR(-EBUSY);
494 if (!grab_super(old))
497 up_write(&s->s_umount);
505 spin_unlock(&sb_lock);
506 s = alloc_super(type, (flags & ~MS_SUBMOUNT), user_ns);
508 return ERR_PTR(-ENOMEM);
514 spin_unlock(&sb_lock);
515 up_write(&s->s_umount);
520 strlcpy(s->s_id, type->name, sizeof(s->s_id));
521 list_add_tail(&s->s_list, &super_blocks);
522 hlist_add_head(&s->s_instances, &type->fs_supers);
523 spin_unlock(&sb_lock);
524 get_filesystem(type);
525 register_shrinker(&s->s_shrink);
529 EXPORT_SYMBOL(sget_userns);
532 * sget - find or create a superblock
533 * @type: filesystem type superblock should belong to
534 * @test: comparison callback
535 * @set: setup callback
536 * @flags: mount flags
537 * @data: argument to each of them
539 struct super_block *sget(struct file_system_type *type,
540 int (*test)(struct super_block *,void *),
541 int (*set)(struct super_block *,void *),
545 struct user_namespace *user_ns = current_user_ns();
547 /* We don't yet pass the user namespace of the parent
548 * mount through to here so always use &init_user_ns
549 * until that changes.
551 if (flags & MS_SUBMOUNT)
552 user_ns = &init_user_ns;
554 /* Ensure the requestor has permissions over the target filesystem */
555 if (!(flags & (MS_KERNMOUNT|MS_SUBMOUNT)) && !ns_capable(user_ns, CAP_SYS_ADMIN))
556 return ERR_PTR(-EPERM);
558 return sget_userns(type, test, set, flags, user_ns, data);
563 void drop_super(struct super_block *sb)
565 up_read(&sb->s_umount);
569 EXPORT_SYMBOL(drop_super);
571 void drop_super_exclusive(struct super_block *sb)
573 up_write(&sb->s_umount);
576 EXPORT_SYMBOL(drop_super_exclusive);
579 * iterate_supers - call function for all active superblocks
580 * @f: function to call
581 * @arg: argument to pass to it
583 * Scans the superblock list and calls given function, passing it
584 * locked superblock and given argument.
586 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
588 struct super_block *sb, *p = NULL;
591 list_for_each_entry(sb, &super_blocks, s_list) {
592 if (hlist_unhashed(&sb->s_instances))
595 spin_unlock(&sb_lock);
597 down_read(&sb->s_umount);
598 if (sb->s_root && (sb->s_flags & MS_BORN))
600 up_read(&sb->s_umount);
609 spin_unlock(&sb_lock);
613 * iterate_supers_type - call function for superblocks of given type
615 * @f: function to call
616 * @arg: argument to pass to it
618 * Scans the superblock list and calls given function, passing it
619 * locked superblock and given argument.
621 void iterate_supers_type(struct file_system_type *type,
622 void (*f)(struct super_block *, void *), void *arg)
624 struct super_block *sb, *p = NULL;
627 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
629 spin_unlock(&sb_lock);
631 down_read(&sb->s_umount);
632 if (sb->s_root && (sb->s_flags & MS_BORN))
634 up_read(&sb->s_umount);
643 spin_unlock(&sb_lock);
646 EXPORT_SYMBOL(iterate_supers_type);
648 static struct super_block *__get_super(struct block_device *bdev, bool excl)
650 struct super_block *sb;
657 list_for_each_entry(sb, &super_blocks, s_list) {
658 if (hlist_unhashed(&sb->s_instances))
660 if (sb->s_bdev == bdev) {
662 spin_unlock(&sb_lock);
664 down_read(&sb->s_umount);
666 down_write(&sb->s_umount);
668 if (sb->s_root && (sb->s_flags & MS_BORN))
671 up_read(&sb->s_umount);
673 up_write(&sb->s_umount);
674 /* nope, got unmounted */
680 spin_unlock(&sb_lock);
685 * get_super - get the superblock of a device
686 * @bdev: device to get the superblock for
688 * Scans the superblock list and finds the superblock of the file system
689 * mounted on the device given. %NULL is returned if no match is found.
691 struct super_block *get_super(struct block_device *bdev)
693 return __get_super(bdev, false);
695 EXPORT_SYMBOL(get_super);
697 static struct super_block *__get_super_thawed(struct block_device *bdev,
701 struct super_block *s = __get_super(bdev, excl);
702 if (!s || s->s_writers.frozen == SB_UNFROZEN)
705 up_read(&s->s_umount);
707 up_write(&s->s_umount);
708 wait_event(s->s_writers.wait_unfrozen,
709 s->s_writers.frozen == SB_UNFROZEN);
715 * get_super_thawed - get thawed superblock of a device
716 * @bdev: device to get the superblock for
718 * Scans the superblock list and finds the superblock of the file system
719 * mounted on the device. The superblock is returned once it is thawed
720 * (or immediately if it was not frozen). %NULL is returned if no match
723 struct super_block *get_super_thawed(struct block_device *bdev)
725 return __get_super_thawed(bdev, false);
727 EXPORT_SYMBOL(get_super_thawed);
730 * get_super_exclusive_thawed - get thawed superblock of a device
731 * @bdev: device to get the superblock for
733 * Scans the superblock list and finds the superblock of the file system
734 * mounted on the device. The superblock is returned once it is thawed
735 * (or immediately if it was not frozen) and s_umount semaphore is held
736 * in exclusive mode. %NULL is returned if no match is found.
738 struct super_block *get_super_exclusive_thawed(struct block_device *bdev)
740 return __get_super_thawed(bdev, true);
742 EXPORT_SYMBOL(get_super_exclusive_thawed);
745 * get_active_super - get an active reference to the superblock of a device
746 * @bdev: device to get the superblock for
748 * Scans the superblock list and finds the superblock of the file system
749 * mounted on the device given. Returns the superblock with an active
750 * reference or %NULL if none was found.
752 struct super_block *get_active_super(struct block_device *bdev)
754 struct super_block *sb;
761 list_for_each_entry(sb, &super_blocks, s_list) {
762 if (hlist_unhashed(&sb->s_instances))
764 if (sb->s_bdev == bdev) {
767 up_write(&sb->s_umount);
771 spin_unlock(&sb_lock);
775 struct super_block *user_get_super(dev_t dev)
777 struct super_block *sb;
781 list_for_each_entry(sb, &super_blocks, s_list) {
782 if (hlist_unhashed(&sb->s_instances))
784 if (sb->s_dev == dev) {
786 spin_unlock(&sb_lock);
787 down_read(&sb->s_umount);
789 if (sb->s_root && (sb->s_flags & MS_BORN))
791 up_read(&sb->s_umount);
792 /* nope, got unmounted */
798 spin_unlock(&sb_lock);
803 * do_remount_sb - asks filesystem to change mount options.
804 * @sb: superblock in question
805 * @flags: numeric part of options
806 * @data: the rest of options
807 * @force: whether or not to force the change
809 * Alters the mount options of a mounted file system.
811 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
816 if (sb->s_writers.frozen != SB_UNFROZEN)
820 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
824 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
827 if (!hlist_empty(&sb->s_pins)) {
828 up_write(&sb->s_umount);
829 group_pin_kill(&sb->s_pins);
830 down_write(&sb->s_umount);
833 if (sb->s_writers.frozen != SB_UNFROZEN)
835 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
838 shrink_dcache_sb(sb);
840 /* If we are remounting RDONLY and current sb is read/write,
841 make sure there are no rw files opened */
844 sb->s_readonly_remount = 1;
847 retval = sb_prepare_remount_readonly(sb);
853 if (sb->s_op->remount_fs) {
854 retval = sb->s_op->remount_fs(sb, &flags, data);
857 goto cancel_readonly;
858 /* If forced remount, go ahead despite any errors */
859 WARN(1, "forced remount of a %s fs returned %i\n",
860 sb->s_type->name, retval);
863 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
864 /* Needs to be ordered wrt mnt_is_readonly() */
866 sb->s_readonly_remount = 0;
869 * Some filesystems modify their metadata via some other path than the
870 * bdev buffer cache (eg. use a private mapping, or directories in
871 * pagecache, etc). Also file data modifications go via their own
872 * mappings. So If we try to mount readonly then copy the filesystem
873 * from bdev, we could get stale data, so invalidate it to give a best
874 * effort at coherency.
876 if (remount_ro && sb->s_bdev)
877 invalidate_bdev(sb->s_bdev);
881 sb->s_readonly_remount = 0;
885 static void do_emergency_remount(struct work_struct *work)
887 struct super_block *sb, *p = NULL;
890 list_for_each_entry(sb, &super_blocks, s_list) {
891 if (hlist_unhashed(&sb->s_instances))
894 spin_unlock(&sb_lock);
895 down_write(&sb->s_umount);
896 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
897 !(sb->s_flags & MS_RDONLY)) {
899 * What lock protects sb->s_flags??
901 do_remount_sb(sb, MS_RDONLY, NULL, 1);
903 up_write(&sb->s_umount);
911 spin_unlock(&sb_lock);
913 printk("Emergency Remount complete\n");
916 void emergency_remount(void)
918 struct work_struct *work;
920 work = kmalloc(sizeof(*work), GFP_ATOMIC);
922 INIT_WORK(work, do_emergency_remount);
928 * Unnamed block devices are dummy devices used by virtual
929 * filesystems which don't use real block-devices. -- jrs
932 static DEFINE_IDA(unnamed_dev_ida);
933 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
934 /* Many userspace utilities consider an FSID of 0 invalid.
935 * Always return at least 1 from get_anon_bdev.
937 static int unnamed_dev_start = 1;
939 int get_anon_bdev(dev_t *p)
945 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
947 spin_lock(&unnamed_dev_lock);
948 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
950 unnamed_dev_start = dev + 1;
951 spin_unlock(&unnamed_dev_lock);
952 if (error == -EAGAIN)
953 /* We raced and lost with another CPU. */
958 if (dev >= (1 << MINORBITS)) {
959 spin_lock(&unnamed_dev_lock);
960 ida_remove(&unnamed_dev_ida, dev);
961 if (unnamed_dev_start > dev)
962 unnamed_dev_start = dev;
963 spin_unlock(&unnamed_dev_lock);
966 *p = MKDEV(0, dev & MINORMASK);
969 EXPORT_SYMBOL(get_anon_bdev);
971 void free_anon_bdev(dev_t dev)
973 int slot = MINOR(dev);
974 spin_lock(&unnamed_dev_lock);
975 ida_remove(&unnamed_dev_ida, slot);
976 if (slot < unnamed_dev_start)
977 unnamed_dev_start = slot;
978 spin_unlock(&unnamed_dev_lock);
980 EXPORT_SYMBOL(free_anon_bdev);
982 int set_anon_super(struct super_block *s, void *data)
984 return get_anon_bdev(&s->s_dev);
987 EXPORT_SYMBOL(set_anon_super);
989 void kill_anon_super(struct super_block *sb)
991 dev_t dev = sb->s_dev;
992 generic_shutdown_super(sb);
996 EXPORT_SYMBOL(kill_anon_super);
998 void kill_litter_super(struct super_block *sb)
1001 d_genocide(sb->s_root);
1002 kill_anon_super(sb);
1005 EXPORT_SYMBOL(kill_litter_super);
1007 static int ns_test_super(struct super_block *sb, void *data)
1009 return sb->s_fs_info == data;
1012 static int ns_set_super(struct super_block *sb, void *data)
1014 sb->s_fs_info = data;
1015 return set_anon_super(sb, NULL);
1018 struct dentry *mount_ns(struct file_system_type *fs_type,
1019 int flags, void *data, void *ns, struct user_namespace *user_ns,
1020 int (*fill_super)(struct super_block *, void *, int))
1022 struct super_block *sb;
1024 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
1025 * over the namespace.
1027 if (!(flags & MS_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
1028 return ERR_PTR(-EPERM);
1030 sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
1033 return ERR_CAST(sb);
1037 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
1039 deactivate_locked_super(sb);
1040 return ERR_PTR(err);
1043 sb->s_flags |= MS_ACTIVE;
1046 return dget(sb->s_root);
1049 EXPORT_SYMBOL(mount_ns);
1052 static int set_bdev_super(struct super_block *s, void *data)
1055 s->s_dev = s->s_bdev->bd_dev;
1056 s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
1061 static int test_bdev_super(struct super_block *s, void *data)
1063 return (void *)s->s_bdev == data;
1066 struct dentry *mount_bdev(struct file_system_type *fs_type,
1067 int flags, const char *dev_name, void *data,
1068 int (*fill_super)(struct super_block *, void *, int))
1070 struct block_device *bdev;
1071 struct super_block *s;
1072 fmode_t mode = FMODE_READ | FMODE_EXCL;
1075 if (!(flags & MS_RDONLY))
1076 mode |= FMODE_WRITE;
1078 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1080 return ERR_CAST(bdev);
1083 * once the super is inserted into the list by sget, s_umount
1084 * will protect the lockfs code from trying to start a snapshot
1085 * while we are mounting
1087 mutex_lock(&bdev->bd_fsfreeze_mutex);
1088 if (bdev->bd_fsfreeze_count > 0) {
1089 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1093 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
1095 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1100 if ((flags ^ s->s_flags) & MS_RDONLY) {
1101 deactivate_locked_super(s);
1107 * s_umount nests inside bd_mutex during
1108 * __invalidate_device(). blkdev_put() acquires
1109 * bd_mutex and can't be called under s_umount. Drop
1110 * s_umount temporarily. This is safe as we're
1111 * holding an active reference.
1113 up_write(&s->s_umount);
1114 blkdev_put(bdev, mode);
1115 down_write(&s->s_umount);
1118 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1119 sb_set_blocksize(s, block_size(bdev));
1120 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1122 deactivate_locked_super(s);
1126 s->s_flags |= MS_ACTIVE;
1130 return dget(s->s_root);
1135 blkdev_put(bdev, mode);
1137 return ERR_PTR(error);
1139 EXPORT_SYMBOL(mount_bdev);
1141 void kill_block_super(struct super_block *sb)
1143 struct block_device *bdev = sb->s_bdev;
1144 fmode_t mode = sb->s_mode;
1146 bdev->bd_super = NULL;
1147 generic_shutdown_super(sb);
1148 sync_blockdev(bdev);
1149 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1150 blkdev_put(bdev, mode | FMODE_EXCL);
1153 EXPORT_SYMBOL(kill_block_super);
1156 struct dentry *mount_nodev(struct file_system_type *fs_type,
1157 int flags, void *data,
1158 int (*fill_super)(struct super_block *, void *, int))
1161 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1166 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1168 deactivate_locked_super(s);
1169 return ERR_PTR(error);
1171 s->s_flags |= MS_ACTIVE;
1172 return dget(s->s_root);
1174 EXPORT_SYMBOL(mount_nodev);
1176 static int compare_single(struct super_block *s, void *p)
1181 struct dentry *mount_single(struct file_system_type *fs_type,
1182 int flags, void *data,
1183 int (*fill_super)(struct super_block *, void *, int))
1185 struct super_block *s;
1188 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1192 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1194 deactivate_locked_super(s);
1195 return ERR_PTR(error);
1197 s->s_flags |= MS_ACTIVE;
1199 do_remount_sb(s, flags, data, 0);
1201 return dget(s->s_root);
1203 EXPORT_SYMBOL(mount_single);
1206 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1208 struct dentry *root;
1209 struct super_block *sb;
1210 char *secdata = NULL;
1211 int error = -ENOMEM;
1213 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1214 secdata = alloc_secdata();
1218 error = security_sb_copy_data(data, secdata);
1220 goto out_free_secdata;
1223 root = type->mount(type, flags, name, data);
1225 error = PTR_ERR(root);
1226 goto out_free_secdata;
1230 WARN_ON(!sb->s_bdi);
1231 sb->s_flags |= MS_BORN;
1233 error = security_sb_kern_mount(sb, flags, secdata);
1238 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1239 * but s_maxbytes was an unsigned long long for many releases. Throw
1240 * this warning for a little while to try and catch filesystems that
1241 * violate this rule.
1243 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1244 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1246 up_write(&sb->s_umount);
1247 free_secdata(secdata);
1251 deactivate_locked_super(sb);
1253 free_secdata(secdata);
1255 return ERR_PTR(error);
1259 * Setup private BDI for given superblock. It gets automatically cleaned up
1260 * in generic_shutdown_super().
1262 int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1264 struct backing_dev_info *bdi;
1268 bdi = bdi_alloc(GFP_KERNEL);
1272 bdi->name = sb->s_type->name;
1274 va_start(args, fmt);
1275 err = bdi_register_va(bdi, fmt, args);
1281 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1286 EXPORT_SYMBOL(super_setup_bdi_name);
1289 * Setup private BDI for given superblock. I gets automatically cleaned up
1290 * in generic_shutdown_super().
1292 int super_setup_bdi(struct super_block *sb)
1294 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1296 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1297 atomic_long_inc_return(&bdi_seq));
1299 EXPORT_SYMBOL(super_setup_bdi);
1302 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1305 void __sb_end_write(struct super_block *sb, int level)
1307 percpu_up_read(sb->s_writers.rw_sem + level-1);
1309 EXPORT_SYMBOL(__sb_end_write);
1312 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1315 int __sb_start_write(struct super_block *sb, int level, bool wait)
1317 bool force_trylock = false;
1320 #ifdef CONFIG_LOCKDEP
1322 * We want lockdep to tell us about possible deadlocks with freezing
1323 * but it's it bit tricky to properly instrument it. Getting a freeze
1324 * protection works as getting a read lock but there are subtle
1325 * problems. XFS for example gets freeze protection on internal level
1326 * twice in some cases, which is OK only because we already hold a
1327 * freeze protection also on higher level. Due to these cases we have
1328 * to use wait == F (trylock mode) which must not fail.
1333 for (i = 0; i < level - 1; i++)
1334 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
1335 force_trylock = true;
1340 if (wait && !force_trylock)
1341 percpu_down_read(sb->s_writers.rw_sem + level-1);
1343 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1345 WARN_ON(force_trylock && !ret);
1348 EXPORT_SYMBOL(__sb_start_write);
1351 * sb_wait_write - wait until all writers to given file system finish
1352 * @sb: the super for which we wait
1353 * @level: type of writers we wait for (normal vs page fault)
1355 * This function waits until there are no writers of given type to given file
1358 static void sb_wait_write(struct super_block *sb, int level)
1360 percpu_down_write(sb->s_writers.rw_sem + level-1);
1364 * We are going to return to userspace and forget about these locks, the
1365 * ownership goes to the caller of thaw_super() which does unlock().
1367 static void lockdep_sb_freeze_release(struct super_block *sb)
1371 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1372 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1376 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1378 static void lockdep_sb_freeze_acquire(struct super_block *sb)
1382 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1383 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1386 static void sb_freeze_unlock(struct super_block *sb)
1390 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1391 percpu_up_write(sb->s_writers.rw_sem + level);
1395 * freeze_super - lock the filesystem and force it into a consistent state
1396 * @sb: the super to lock
1398 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1399 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1402 * During this function, sb->s_writers.frozen goes through these values:
1404 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1406 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1407 * writes should be blocked, though page faults are still allowed. We wait for
1408 * all writes to complete and then proceed to the next stage.
1410 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1411 * but internal fs threads can still modify the filesystem (although they
1412 * should not dirty new pages or inodes), writeback can run etc. After waiting
1413 * for all running page faults we sync the filesystem which will clean all
1414 * dirty pages and inodes (no new dirty pages or inodes can be created when
1417 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1418 * modification are blocked (e.g. XFS preallocation truncation on inode
1419 * reclaim). This is usually implemented by blocking new transactions for
1420 * filesystems that have them and need this additional guard. After all
1421 * internal writers are finished we call ->freeze_fs() to finish filesystem
1422 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1423 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1425 * sb->s_writers.frozen is protected by sb->s_umount.
1427 int freeze_super(struct super_block *sb)
1431 atomic_inc(&sb->s_active);
1432 down_write(&sb->s_umount);
1433 if (sb->s_writers.frozen != SB_UNFROZEN) {
1434 deactivate_locked_super(sb);
1438 if (!(sb->s_flags & MS_BORN)) {
1439 up_write(&sb->s_umount);
1440 return 0; /* sic - it's "nothing to do" */
1443 if (sb->s_flags & MS_RDONLY) {
1444 /* Nothing to do really... */
1445 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1446 up_write(&sb->s_umount);
1450 sb->s_writers.frozen = SB_FREEZE_WRITE;
1451 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1452 up_write(&sb->s_umount);
1453 sb_wait_write(sb, SB_FREEZE_WRITE);
1454 down_write(&sb->s_umount);
1456 /* Now we go and block page faults... */
1457 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1458 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1460 /* All writers are done so after syncing there won't be dirty data */
1461 sync_filesystem(sb);
1463 /* Now wait for internal filesystem counter */
1464 sb->s_writers.frozen = SB_FREEZE_FS;
1465 sb_wait_write(sb, SB_FREEZE_FS);
1467 if (sb->s_op->freeze_fs) {
1468 ret = sb->s_op->freeze_fs(sb);
1471 "VFS:Filesystem freeze failed\n");
1472 sb->s_writers.frozen = SB_UNFROZEN;
1473 sb_freeze_unlock(sb);
1474 wake_up(&sb->s_writers.wait_unfrozen);
1475 deactivate_locked_super(sb);
1480 * For debugging purposes so that fs can warn if it sees write activity
1481 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1483 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1484 lockdep_sb_freeze_release(sb);
1485 up_write(&sb->s_umount);
1488 EXPORT_SYMBOL(freeze_super);
1491 * thaw_super -- unlock filesystem
1492 * @sb: the super to thaw
1494 * Unlocks the filesystem and marks it writeable again after freeze_super().
1496 int thaw_super(struct super_block *sb)
1500 down_write(&sb->s_umount);
1501 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1502 up_write(&sb->s_umount);
1506 if (sb->s_flags & MS_RDONLY) {
1507 sb->s_writers.frozen = SB_UNFROZEN;
1511 lockdep_sb_freeze_acquire(sb);
1513 if (sb->s_op->unfreeze_fs) {
1514 error = sb->s_op->unfreeze_fs(sb);
1517 "VFS:Filesystem thaw failed\n");
1518 lockdep_sb_freeze_release(sb);
1519 up_write(&sb->s_umount);
1524 sb->s_writers.frozen = SB_UNFROZEN;
1525 sb_freeze_unlock(sb);
1527 wake_up(&sb->s_writers.wait_unfrozen);
1528 deactivate_locked_super(sb);
1531 EXPORT_SYMBOL(thaw_super);