1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
7 * super.c contains code to handle: - mount structures
9 * - filesystem drivers list
11 * - umount system call
14 * GK 2/5/95 - Changed to support mounting the root fs via NFS
16 * Added kerneld support: Jacques Gelinas and Bjorn Ekwall
17 * Added change_root: Werner Almesberger & Hans Lermen, Feb '96
18 * Added options to /proc/mounts:
19 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
20 * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
21 * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
24 #include <linux/export.h>
25 #include <linux/slab.h>
26 #include <linux/blkdev.h>
27 #include <linux/mount.h>
28 #include <linux/security.h>
29 #include <linux/writeback.h> /* for the emergency remount stuff */
30 #include <linux/idr.h>
31 #include <linux/mutex.h>
32 #include <linux/backing-dev.h>
33 #include <linux/rculist_bl.h>
34 #include <linux/cleancache.h>
35 #include <linux/fsnotify.h>
36 #include <linux/lockdep.h>
37 #include <linux/user_namespace.h>
41 static LIST_HEAD(super_blocks);
42 static DEFINE_SPINLOCK(sb_lock);
44 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
51 * One thing we have to be careful of with a per-sb shrinker is that we don't
52 * drop the last active reference to the superblock from within the shrinker.
53 * If that happens we could trigger unregistering the shrinker from within the
54 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
55 * take a passive reference to the superblock to avoid this from occurring.
57 static unsigned long super_cache_scan(struct shrinker *shrink,
58 struct shrink_control *sc)
60 struct super_block *sb;
67 sb = container_of(shrink, struct super_block, s_shrink);
70 * Deadlock avoidance. We may hold various FS locks, and we don't want
71 * to recurse into the FS that called us in clear_inode() and friends..
73 if (!(sc->gfp_mask & __GFP_FS))
76 if (!trylock_super(sb))
79 if (sb->s_op->nr_cached_objects)
80 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
82 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
83 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
84 total_objects = dentries + inodes + fs_objects + 1;
88 /* proportion the scan between the caches */
89 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
90 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
91 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
94 * prune the dcache first as the icache is pinned by it, then
95 * prune the icache, followed by the filesystem specific caches
97 * Ensure that we always scan at least one object - memcg kmem
98 * accounting uses this to fully empty the caches.
100 sc->nr_to_scan = dentries + 1;
101 freed = prune_dcache_sb(sb, sc);
102 sc->nr_to_scan = inodes + 1;
103 freed += prune_icache_sb(sb, sc);
106 sc->nr_to_scan = fs_objects + 1;
107 freed += sb->s_op->free_cached_objects(sb, sc);
110 up_read(&sb->s_umount);
114 static unsigned long super_cache_count(struct shrinker *shrink,
115 struct shrink_control *sc)
117 struct super_block *sb;
118 long total_objects = 0;
120 sb = container_of(shrink, struct super_block, s_shrink);
123 * We don't call trylock_super() here as it is a scalability bottleneck,
124 * so we're exposed to partial setup state. The shrinker rwsem does not
125 * protect filesystem operations backing list_lru_shrink_count() or
126 * s_op->nr_cached_objects(). Counts can change between
127 * super_cache_count and super_cache_scan, so we really don't need locks
130 * However, if we are currently mounting the superblock, the underlying
131 * filesystem might be in a state of partial construction and hence it
132 * is dangerous to access it. trylock_super() uses a SB_BORN check to
133 * avoid this situation, so do the same here. The memory barrier is
134 * matched with the one in mount_fs() as we don't hold locks here.
136 if (!(sb->s_flags & SB_BORN))
140 if (sb->s_op && sb->s_op->nr_cached_objects)
141 total_objects = sb->s_op->nr_cached_objects(sb, sc);
143 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
144 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
146 total_objects = vfs_pressure_ratio(total_objects);
147 return total_objects;
150 static void destroy_super_work(struct work_struct *work)
152 struct super_block *s = container_of(work, struct super_block,
156 for (i = 0; i < SB_FREEZE_LEVELS; i++)
157 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
161 static void destroy_super_rcu(struct rcu_head *head)
163 struct super_block *s = container_of(head, struct super_block, rcu);
164 INIT_WORK(&s->destroy_work, destroy_super_work);
165 schedule_work(&s->destroy_work);
169 * destroy_super - frees a superblock
170 * @s: superblock to free
172 * Frees a superblock.
174 static void destroy_super(struct super_block *s)
176 list_lru_destroy(&s->s_dentry_lru);
177 list_lru_destroy(&s->s_inode_lru);
179 WARN_ON(!list_empty(&s->s_mounts));
180 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 init_rwsem(&s->s_dquot.dqio_sem);
257 s->s_maxbytes = MAX_NON_LFS;
258 s->s_op = &default_op;
259 s->s_time_gran = 1000000000;
260 s->cleancache_poolid = CLEANCACHE_NO_POOL;
262 s->s_shrink.seeks = DEFAULT_SEEKS;
263 s->s_shrink.scan_objects = super_cache_scan;
264 s->s_shrink.count_objects = super_cache_count;
265 s->s_shrink.batch = 1024;
266 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
274 /* Superblock refcounting */
277 * Drop a superblock's refcount. The caller must hold sb_lock.
279 static void __put_super(struct super_block *sb)
281 if (!--sb->s_count) {
282 list_del_init(&sb->s_list);
288 * put_super - drop a temporary reference to superblock
289 * @sb: superblock in question
291 * Drops a temporary reference, frees superblock if there's no
294 static void put_super(struct super_block *sb)
298 spin_unlock(&sb_lock);
303 * deactivate_locked_super - drop an active reference to superblock
304 * @s: superblock to deactivate
306 * Drops an active reference to superblock, converting it into a temporary
307 * one if there is no other active references left. In that case we
308 * tell fs driver to shut it down and drop the temporary reference we
311 * Caller holds exclusive lock on superblock; that lock is released.
313 void deactivate_locked_super(struct super_block *s)
315 struct file_system_type *fs = s->s_type;
316 if (atomic_dec_and_test(&s->s_active)) {
317 cleancache_invalidate_fs(s);
318 unregister_shrinker(&s->s_shrink);
322 * Since list_lru_destroy() may sleep, we cannot call it from
323 * put_super(), where we hold the sb_lock. Therefore we destroy
324 * the lru lists right now.
326 list_lru_destroy(&s->s_dentry_lru);
327 list_lru_destroy(&s->s_inode_lru);
332 up_write(&s->s_umount);
336 EXPORT_SYMBOL(deactivate_locked_super);
339 * deactivate_super - drop an active reference to superblock
340 * @s: superblock to deactivate
342 * Variant of deactivate_locked_super(), except that superblock is *not*
343 * locked by caller. If we are going to drop the final active reference,
344 * lock will be acquired prior to that.
346 void deactivate_super(struct super_block *s)
348 if (!atomic_add_unless(&s->s_active, -1, 1)) {
349 down_write(&s->s_umount);
350 deactivate_locked_super(s);
354 EXPORT_SYMBOL(deactivate_super);
357 * grab_super - acquire an active reference
358 * @s: reference we are trying to make active
360 * Tries to acquire an active reference. grab_super() is used when we
361 * had just found a superblock in super_blocks or fs_type->fs_supers
362 * and want to turn it into a full-blown active reference. grab_super()
363 * is called with sb_lock held and drops it. Returns 1 in case of
364 * success, 0 if we had failed (superblock contents was already dead or
365 * dying when grab_super() had been called). Note that this is only
366 * called for superblocks not in rundown mode (== ones still on ->fs_supers
367 * of their type), so increment of ->s_count is OK here.
369 static int grab_super(struct super_block *s) __releases(sb_lock)
372 spin_unlock(&sb_lock);
373 down_write(&s->s_umount);
374 if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
378 up_write(&s->s_umount);
384 * trylock_super - try to grab ->s_umount shared
385 * @sb: reference we are trying to grab
387 * Try to prevent fs shutdown. This is used in places where we
388 * cannot take an active reference but we need to ensure that the
389 * filesystem is not shut down while we are working on it. It returns
390 * false if we cannot acquire s_umount or if we lose the race and
391 * filesystem already got into shutdown, and returns true with the s_umount
392 * lock held in read mode in case of success. On successful return,
393 * the caller must drop the s_umount lock when done.
395 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
396 * The reason why it's safe is that we are OK with doing trylock instead
397 * of down_read(). There's a couple of places that are OK with that, but
398 * it's very much not a general-purpose interface.
400 bool trylock_super(struct super_block *sb)
402 if (down_read_trylock(&sb->s_umount)) {
403 if (!hlist_unhashed(&sb->s_instances) &&
404 sb->s_root && (sb->s_flags & SB_BORN))
406 up_read(&sb->s_umount);
413 * generic_shutdown_super - common helper for ->kill_sb()
414 * @sb: superblock to kill
416 * generic_shutdown_super() does all fs-independent work on superblock
417 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
418 * that need destruction out of superblock, call generic_shutdown_super()
419 * and release aforementioned objects. Note: dentries and inodes _are_
420 * taken care of and do not need specific handling.
422 * Upon calling this function, the filesystem may no longer alter or
423 * rearrange the set of dentries belonging to this super_block, nor may it
424 * change the attachments of dentries to inodes.
426 void generic_shutdown_super(struct super_block *sb)
428 const struct super_operations *sop = sb->s_op;
431 shrink_dcache_for_umount(sb);
433 sb->s_flags &= ~SB_ACTIVE;
435 fsnotify_unmount_inodes(sb);
436 cgroup_writeback_umount();
440 if (sb->s_dio_done_wq) {
441 destroy_workqueue(sb->s_dio_done_wq);
442 sb->s_dio_done_wq = NULL;
448 if (!list_empty(&sb->s_inodes)) {
449 printk("VFS: Busy inodes after unmount of %s. "
450 "Self-destruct in 5 seconds. Have a nice day...\n",
455 /* should be initialized for __put_super_and_need_restart() */
456 hlist_del_init(&sb->s_instances);
457 spin_unlock(&sb_lock);
458 up_write(&sb->s_umount);
459 if (sb->s_bdi != &noop_backing_dev_info) {
461 sb->s_bdi = &noop_backing_dev_info;
465 EXPORT_SYMBOL(generic_shutdown_super);
468 * sget_userns - find or create a superblock
469 * @type: filesystem type superblock should belong to
470 * @test: comparison callback
471 * @set: setup callback
472 * @flags: mount flags
473 * @user_ns: User namespace for the super_block
474 * @data: argument to each of them
476 struct super_block *sget_userns(struct file_system_type *type,
477 int (*test)(struct super_block *,void *),
478 int (*set)(struct super_block *,void *),
479 int flags, struct user_namespace *user_ns,
482 struct super_block *s = NULL;
483 struct super_block *old;
486 if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) &&
487 !(type->fs_flags & FS_USERNS_MOUNT) &&
488 !capable(CAP_SYS_ADMIN))
489 return ERR_PTR(-EPERM);
493 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
494 if (!test(old, data))
496 if (user_ns != old->s_user_ns) {
497 spin_unlock(&sb_lock);
499 up_write(&s->s_umount);
502 return ERR_PTR(-EBUSY);
504 if (!grab_super(old))
507 up_write(&s->s_umount);
515 spin_unlock(&sb_lock);
516 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
518 return ERR_PTR(-ENOMEM);
524 spin_unlock(&sb_lock);
525 up_write(&s->s_umount);
530 strlcpy(s->s_id, type->name, sizeof(s->s_id));
531 list_add_tail(&s->s_list, &super_blocks);
532 hlist_add_head(&s->s_instances, &type->fs_supers);
533 spin_unlock(&sb_lock);
534 get_filesystem(type);
535 err = register_shrinker(&s->s_shrink);
537 deactivate_locked_super(s);
543 EXPORT_SYMBOL(sget_userns);
546 * sget - find or create a superblock
547 * @type: filesystem type superblock should belong to
548 * @test: comparison callback
549 * @set: setup callback
550 * @flags: mount flags
551 * @data: argument to each of them
553 struct super_block *sget(struct file_system_type *type,
554 int (*test)(struct super_block *,void *),
555 int (*set)(struct super_block *,void *),
559 struct user_namespace *user_ns = current_user_ns();
561 /* We don't yet pass the user namespace of the parent
562 * mount through to here so always use &init_user_ns
563 * until that changes.
565 if (flags & SB_SUBMOUNT)
566 user_ns = &init_user_ns;
568 /* Ensure the requestor has permissions over the target filesystem */
569 if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) && !ns_capable(user_ns, CAP_SYS_ADMIN))
570 return ERR_PTR(-EPERM);
572 return sget_userns(type, test, set, flags, user_ns, data);
577 void drop_super(struct super_block *sb)
579 up_read(&sb->s_umount);
583 EXPORT_SYMBOL(drop_super);
585 void drop_super_exclusive(struct super_block *sb)
587 up_write(&sb->s_umount);
590 EXPORT_SYMBOL(drop_super_exclusive);
593 * iterate_supers - call function for all active superblocks
594 * @f: function to call
595 * @arg: argument to pass to it
597 * Scans the superblock list and calls given function, passing it
598 * locked superblock and given argument.
600 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
602 struct super_block *sb, *p = NULL;
605 list_for_each_entry(sb, &super_blocks, s_list) {
606 if (hlist_unhashed(&sb->s_instances))
609 spin_unlock(&sb_lock);
611 down_read(&sb->s_umount);
612 if (sb->s_root && (sb->s_flags & SB_BORN))
614 up_read(&sb->s_umount);
623 spin_unlock(&sb_lock);
627 * iterate_supers_type - call function for superblocks of given type
629 * @f: function to call
630 * @arg: argument to pass to it
632 * Scans the superblock list and calls given function, passing it
633 * locked superblock and given argument.
635 void iterate_supers_type(struct file_system_type *type,
636 void (*f)(struct super_block *, void *), void *arg)
638 struct super_block *sb, *p = NULL;
641 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
643 spin_unlock(&sb_lock);
645 down_read(&sb->s_umount);
646 if (sb->s_root && (sb->s_flags & SB_BORN))
648 up_read(&sb->s_umount);
657 spin_unlock(&sb_lock);
660 EXPORT_SYMBOL(iterate_supers_type);
662 static struct super_block *__get_super(struct block_device *bdev, bool excl)
664 struct super_block *sb;
671 list_for_each_entry(sb, &super_blocks, s_list) {
672 if (hlist_unhashed(&sb->s_instances))
674 if (sb->s_bdev == bdev) {
676 spin_unlock(&sb_lock);
678 down_read(&sb->s_umount);
680 down_write(&sb->s_umount);
682 if (sb->s_root && (sb->s_flags & SB_BORN))
685 up_read(&sb->s_umount);
687 up_write(&sb->s_umount);
688 /* nope, got unmounted */
694 spin_unlock(&sb_lock);
699 * get_super - get the superblock of a device
700 * @bdev: device to get the superblock for
702 * Scans the superblock list and finds the superblock of the file system
703 * mounted on the device given. %NULL is returned if no match is found.
705 struct super_block *get_super(struct block_device *bdev)
707 return __get_super(bdev, false);
709 EXPORT_SYMBOL(get_super);
711 static struct super_block *__get_super_thawed(struct block_device *bdev,
715 struct super_block *s = __get_super(bdev, excl);
716 if (!s || s->s_writers.frozen == SB_UNFROZEN)
719 up_read(&s->s_umount);
721 up_write(&s->s_umount);
722 wait_event(s->s_writers.wait_unfrozen,
723 s->s_writers.frozen == SB_UNFROZEN);
729 * get_super_thawed - get thawed superblock of a device
730 * @bdev: device to get the superblock for
732 * Scans the superblock list and finds the superblock of the file system
733 * mounted on the device. The superblock is returned once it is thawed
734 * (or immediately if it was not frozen). %NULL is returned if no match
737 struct super_block *get_super_thawed(struct block_device *bdev)
739 return __get_super_thawed(bdev, false);
741 EXPORT_SYMBOL(get_super_thawed);
744 * get_super_exclusive_thawed - get thawed superblock of a device
745 * @bdev: device to get the superblock for
747 * Scans the superblock list and finds the superblock of the file system
748 * mounted on the device. The superblock is returned once it is thawed
749 * (or immediately if it was not frozen) and s_umount semaphore is held
750 * in exclusive mode. %NULL is returned if no match is found.
752 struct super_block *get_super_exclusive_thawed(struct block_device *bdev)
754 return __get_super_thawed(bdev, true);
756 EXPORT_SYMBOL(get_super_exclusive_thawed);
759 * get_active_super - get an active reference to the superblock of a device
760 * @bdev: device to get the superblock for
762 * Scans the superblock list and finds the superblock of the file system
763 * mounted on the device given. Returns the superblock with an active
764 * reference or %NULL if none was found.
766 struct super_block *get_active_super(struct block_device *bdev)
768 struct super_block *sb;
775 list_for_each_entry(sb, &super_blocks, s_list) {
776 if (hlist_unhashed(&sb->s_instances))
778 if (sb->s_bdev == bdev) {
781 up_write(&sb->s_umount);
785 spin_unlock(&sb_lock);
789 struct super_block *user_get_super(dev_t dev)
791 struct super_block *sb;
795 list_for_each_entry(sb, &super_blocks, s_list) {
796 if (hlist_unhashed(&sb->s_instances))
798 if (sb->s_dev == dev) {
800 spin_unlock(&sb_lock);
801 down_read(&sb->s_umount);
803 if (sb->s_root && (sb->s_flags & SB_BORN))
805 up_read(&sb->s_umount);
806 /* nope, got unmounted */
812 spin_unlock(&sb_lock);
817 * do_remount_sb - asks filesystem to change mount options.
818 * @sb: superblock in question
819 * @sb_flags: revised superblock flags
820 * @data: the rest of options
821 * @force: whether or not to force the change
823 * Alters the mount options of a mounted file system.
825 int do_remount_sb(struct super_block *sb, int sb_flags, void *data, int force)
830 if (sb->s_writers.frozen != SB_UNFROZEN)
834 if (!(sb_flags & SB_RDONLY) && bdev_read_only(sb->s_bdev))
838 remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
841 if (!hlist_empty(&sb->s_pins)) {
842 up_write(&sb->s_umount);
843 group_pin_kill(&sb->s_pins);
844 down_write(&sb->s_umount);
847 if (sb->s_writers.frozen != SB_UNFROZEN)
849 remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
852 shrink_dcache_sb(sb);
854 /* If we are remounting RDONLY and current sb is read/write,
855 make sure there are no rw files opened */
858 sb->s_readonly_remount = 1;
861 retval = sb_prepare_remount_readonly(sb);
867 if (sb->s_op->remount_fs) {
868 retval = sb->s_op->remount_fs(sb, &sb_flags, data);
871 goto cancel_readonly;
872 /* If forced remount, go ahead despite any errors */
873 WARN(1, "forced remount of a %s fs returned %i\n",
874 sb->s_type->name, retval);
877 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (sb_flags & MS_RMT_MASK);
878 /* Needs to be ordered wrt mnt_is_readonly() */
880 sb->s_readonly_remount = 0;
883 * Some filesystems modify their metadata via some other path than the
884 * bdev buffer cache (eg. use a private mapping, or directories in
885 * pagecache, etc). Also file data modifications go via their own
886 * mappings. So If we try to mount readonly then copy the filesystem
887 * from bdev, we could get stale data, so invalidate it to give a best
888 * effort at coherency.
890 if (remount_ro && sb->s_bdev)
891 invalidate_bdev(sb->s_bdev);
895 sb->s_readonly_remount = 0;
899 static void do_emergency_remount(struct work_struct *work)
901 struct super_block *sb, *p = NULL;
904 list_for_each_entry(sb, &super_blocks, s_list) {
905 if (hlist_unhashed(&sb->s_instances))
908 spin_unlock(&sb_lock);
909 down_write(&sb->s_umount);
910 if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
913 * What lock protects sb->s_flags??
915 do_remount_sb(sb, SB_RDONLY, NULL, 1);
917 up_write(&sb->s_umount);
925 spin_unlock(&sb_lock);
927 printk("Emergency Remount complete\n");
930 void emergency_remount(void)
932 struct work_struct *work;
934 work = kmalloc(sizeof(*work), GFP_ATOMIC);
936 INIT_WORK(work, do_emergency_remount);
942 * Unnamed block devices are dummy devices used by virtual
943 * filesystems which don't use real block-devices. -- jrs
946 static DEFINE_IDA(unnamed_dev_ida);
947 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
948 /* Many userspace utilities consider an FSID of 0 invalid.
949 * Always return at least 1 from get_anon_bdev.
951 static int unnamed_dev_start = 1;
953 int get_anon_bdev(dev_t *p)
959 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
961 spin_lock(&unnamed_dev_lock);
962 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
964 unnamed_dev_start = dev + 1;
965 spin_unlock(&unnamed_dev_lock);
966 if (error == -EAGAIN)
967 /* We raced and lost with another CPU. */
972 if (dev >= (1 << MINORBITS)) {
973 spin_lock(&unnamed_dev_lock);
974 ida_remove(&unnamed_dev_ida, dev);
975 if (unnamed_dev_start > dev)
976 unnamed_dev_start = dev;
977 spin_unlock(&unnamed_dev_lock);
980 *p = MKDEV(0, dev & MINORMASK);
983 EXPORT_SYMBOL(get_anon_bdev);
985 void free_anon_bdev(dev_t dev)
987 int slot = MINOR(dev);
988 spin_lock(&unnamed_dev_lock);
989 ida_remove(&unnamed_dev_ida, slot);
990 if (slot < unnamed_dev_start)
991 unnamed_dev_start = slot;
992 spin_unlock(&unnamed_dev_lock);
994 EXPORT_SYMBOL(free_anon_bdev);
996 int set_anon_super(struct super_block *s, void *data)
998 return get_anon_bdev(&s->s_dev);
1001 EXPORT_SYMBOL(set_anon_super);
1003 void kill_anon_super(struct super_block *sb)
1005 dev_t dev = sb->s_dev;
1006 generic_shutdown_super(sb);
1007 free_anon_bdev(dev);
1010 EXPORT_SYMBOL(kill_anon_super);
1012 void kill_litter_super(struct super_block *sb)
1015 d_genocide(sb->s_root);
1016 kill_anon_super(sb);
1019 EXPORT_SYMBOL(kill_litter_super);
1021 static int ns_test_super(struct super_block *sb, void *data)
1023 return sb->s_fs_info == data;
1026 static int ns_set_super(struct super_block *sb, void *data)
1028 sb->s_fs_info = data;
1029 return set_anon_super(sb, NULL);
1032 struct dentry *mount_ns(struct file_system_type *fs_type,
1033 int flags, void *data, void *ns, struct user_namespace *user_ns,
1034 int (*fill_super)(struct super_block *, void *, int))
1036 struct super_block *sb;
1038 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
1039 * over the namespace.
1041 if (!(flags & SB_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
1042 return ERR_PTR(-EPERM);
1044 sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
1047 return ERR_CAST(sb);
1051 err = fill_super(sb, data, flags & SB_SILENT ? 1 : 0);
1053 deactivate_locked_super(sb);
1054 return ERR_PTR(err);
1057 sb->s_flags |= SB_ACTIVE;
1060 return dget(sb->s_root);
1063 EXPORT_SYMBOL(mount_ns);
1066 static int set_bdev_super(struct super_block *s, void *data)
1069 s->s_dev = s->s_bdev->bd_dev;
1070 s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
1075 static int test_bdev_super(struct super_block *s, void *data)
1077 return (void *)s->s_bdev == data;
1080 struct dentry *mount_bdev(struct file_system_type *fs_type,
1081 int flags, const char *dev_name, void *data,
1082 int (*fill_super)(struct super_block *, void *, int))
1084 struct block_device *bdev;
1085 struct super_block *s;
1086 fmode_t mode = FMODE_READ | FMODE_EXCL;
1089 if (!(flags & SB_RDONLY))
1090 mode |= FMODE_WRITE;
1092 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1094 return ERR_CAST(bdev);
1097 * once the super is inserted into the list by sget, s_umount
1098 * will protect the lockfs code from trying to start a snapshot
1099 * while we are mounting
1101 mutex_lock(&bdev->bd_fsfreeze_mutex);
1102 if (bdev->bd_fsfreeze_count > 0) {
1103 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1107 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
1109 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1114 if ((flags ^ s->s_flags) & SB_RDONLY) {
1115 deactivate_locked_super(s);
1121 * s_umount nests inside bd_mutex during
1122 * __invalidate_device(). blkdev_put() acquires
1123 * bd_mutex and can't be called under s_umount. Drop
1124 * s_umount temporarily. This is safe as we're
1125 * holding an active reference.
1127 up_write(&s->s_umount);
1128 blkdev_put(bdev, mode);
1129 down_write(&s->s_umount);
1132 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1133 sb_set_blocksize(s, block_size(bdev));
1134 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1136 deactivate_locked_super(s);
1140 s->s_flags |= SB_ACTIVE;
1144 return dget(s->s_root);
1149 blkdev_put(bdev, mode);
1151 return ERR_PTR(error);
1153 EXPORT_SYMBOL(mount_bdev);
1155 void kill_block_super(struct super_block *sb)
1157 struct block_device *bdev = sb->s_bdev;
1158 fmode_t mode = sb->s_mode;
1160 bdev->bd_super = NULL;
1161 generic_shutdown_super(sb);
1162 sync_blockdev(bdev);
1163 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1164 blkdev_put(bdev, mode | FMODE_EXCL);
1167 EXPORT_SYMBOL(kill_block_super);
1170 struct dentry *mount_nodev(struct file_system_type *fs_type,
1171 int flags, void *data,
1172 int (*fill_super)(struct super_block *, void *, int))
1175 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1180 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1182 deactivate_locked_super(s);
1183 return ERR_PTR(error);
1185 s->s_flags |= SB_ACTIVE;
1186 return dget(s->s_root);
1188 EXPORT_SYMBOL(mount_nodev);
1190 static int compare_single(struct super_block *s, void *p)
1195 struct dentry *mount_single(struct file_system_type *fs_type,
1196 int flags, void *data,
1197 int (*fill_super)(struct super_block *, void *, int))
1199 struct super_block *s;
1202 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1206 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1208 deactivate_locked_super(s);
1209 return ERR_PTR(error);
1211 s->s_flags |= SB_ACTIVE;
1213 do_remount_sb(s, flags, data, 0);
1215 return dget(s->s_root);
1217 EXPORT_SYMBOL(mount_single);
1220 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1222 struct dentry *root;
1223 struct super_block *sb;
1224 char *secdata = NULL;
1225 int error = -ENOMEM;
1227 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1228 secdata = alloc_secdata();
1232 error = security_sb_copy_data(data, secdata);
1234 goto out_free_secdata;
1237 root = type->mount(type, flags, name, data);
1239 error = PTR_ERR(root);
1240 goto out_free_secdata;
1244 WARN_ON(!sb->s_bdi);
1247 * Write barrier is for super_cache_count(). We place it before setting
1248 * SB_BORN as the data dependency between the two functions is the
1249 * superblock structure contents that we just set up, not the SB_BORN
1253 sb->s_flags |= SB_BORN;
1255 error = security_sb_kern_mount(sb, flags, secdata);
1260 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1261 * but s_maxbytes was an unsigned long long for many releases. Throw
1262 * this warning for a little while to try and catch filesystems that
1263 * violate this rule.
1265 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1266 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1268 up_write(&sb->s_umount);
1269 free_secdata(secdata);
1273 deactivate_locked_super(sb);
1275 free_secdata(secdata);
1277 return ERR_PTR(error);
1281 * Setup private BDI for given superblock. It gets automatically cleaned up
1282 * in generic_shutdown_super().
1284 int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1286 struct backing_dev_info *bdi;
1290 bdi = bdi_alloc(GFP_KERNEL);
1294 bdi->name = sb->s_type->name;
1296 va_start(args, fmt);
1297 err = bdi_register_va(bdi, fmt, args);
1303 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1308 EXPORT_SYMBOL(super_setup_bdi_name);
1311 * Setup private BDI for given superblock. I gets automatically cleaned up
1312 * in generic_shutdown_super().
1314 int super_setup_bdi(struct super_block *sb)
1316 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1318 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1319 atomic_long_inc_return(&bdi_seq));
1321 EXPORT_SYMBOL(super_setup_bdi);
1324 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1327 void __sb_end_write(struct super_block *sb, int level)
1329 percpu_up_read(sb->s_writers.rw_sem + level-1);
1331 EXPORT_SYMBOL(__sb_end_write);
1334 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1337 int __sb_start_write(struct super_block *sb, int level, bool wait)
1340 return percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1342 percpu_down_read(sb->s_writers.rw_sem + level-1);
1345 EXPORT_SYMBOL(__sb_start_write);
1348 * sb_wait_write - wait until all writers to given file system finish
1349 * @sb: the super for which we wait
1350 * @level: type of writers we wait for (normal vs page fault)
1352 * This function waits until there are no writers of given type to given file
1355 static void sb_wait_write(struct super_block *sb, int level)
1357 percpu_down_write(sb->s_writers.rw_sem + level-1);
1361 * We are going to return to userspace and forget about these locks, the
1362 * ownership goes to the caller of thaw_super() which does unlock().
1364 static void lockdep_sb_freeze_release(struct super_block *sb)
1368 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1369 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1373 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1375 static void lockdep_sb_freeze_acquire(struct super_block *sb)
1379 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1380 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1383 static void sb_freeze_unlock(struct super_block *sb, int level)
1385 for (level--; level >= 0; level--)
1386 percpu_up_write(sb->s_writers.rw_sem + level);
1390 * freeze_super - lock the filesystem and force it into a consistent state
1391 * @sb: the super to lock
1393 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1394 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1397 * During this function, sb->s_writers.frozen goes through these values:
1399 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1401 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1402 * writes should be blocked, though page faults are still allowed. We wait for
1403 * all writes to complete and then proceed to the next stage.
1405 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1406 * but internal fs threads can still modify the filesystem (although they
1407 * should not dirty new pages or inodes), writeback can run etc. After waiting
1408 * for all running page faults we sync the filesystem which will clean all
1409 * dirty pages and inodes (no new dirty pages or inodes can be created when
1412 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1413 * modification are blocked (e.g. XFS preallocation truncation on inode
1414 * reclaim). This is usually implemented by blocking new transactions for
1415 * filesystems that have them and need this additional guard. After all
1416 * internal writers are finished we call ->freeze_fs() to finish filesystem
1417 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1418 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1420 * sb->s_writers.frozen is protected by sb->s_umount.
1422 int freeze_super(struct super_block *sb)
1426 atomic_inc(&sb->s_active);
1427 down_write(&sb->s_umount);
1428 if (sb->s_writers.frozen != SB_UNFROZEN) {
1429 deactivate_locked_super(sb);
1433 if (!(sb->s_flags & SB_BORN)) {
1434 up_write(&sb->s_umount);
1435 return 0; /* sic - it's "nothing to do" */
1438 if (sb_rdonly(sb)) {
1439 /* Nothing to do really... */
1440 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1441 up_write(&sb->s_umount);
1445 sb->s_writers.frozen = SB_FREEZE_WRITE;
1446 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1447 up_write(&sb->s_umount);
1448 sb_wait_write(sb, SB_FREEZE_WRITE);
1449 down_write(&sb->s_umount);
1451 /* Now we go and block page faults... */
1452 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1453 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1455 /* All writers are done so after syncing there won't be dirty data */
1456 ret = sync_filesystem(sb);
1458 sb->s_writers.frozen = SB_UNFROZEN;
1459 sb_freeze_unlock(sb, SB_FREEZE_PAGEFAULT);
1460 wake_up(&sb->s_writers.wait_unfrozen);
1461 deactivate_locked_super(sb);
1465 /* Now wait for internal filesystem counter */
1466 sb->s_writers.frozen = SB_FREEZE_FS;
1467 sb_wait_write(sb, SB_FREEZE_FS);
1469 if (sb->s_op->freeze_fs) {
1470 ret = sb->s_op->freeze_fs(sb);
1473 "VFS:Filesystem freeze failed\n");
1474 sb->s_writers.frozen = SB_UNFROZEN;
1475 sb_freeze_unlock(sb, SB_FREEZE_FS);
1476 wake_up(&sb->s_writers.wait_unfrozen);
1477 deactivate_locked_super(sb);
1482 * For debugging purposes so that fs can warn if it sees write activity
1483 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1485 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1486 lockdep_sb_freeze_release(sb);
1487 up_write(&sb->s_umount);
1490 EXPORT_SYMBOL(freeze_super);
1493 * thaw_super -- unlock filesystem
1494 * @sb: the super to thaw
1496 * Unlocks the filesystem and marks it writeable again after freeze_super().
1498 int thaw_super(struct super_block *sb)
1502 down_write(&sb->s_umount);
1503 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1504 up_write(&sb->s_umount);
1508 if (sb_rdonly(sb)) {
1509 sb->s_writers.frozen = SB_UNFROZEN;
1513 lockdep_sb_freeze_acquire(sb);
1515 if (sb->s_op->unfreeze_fs) {
1516 error = sb->s_op->unfreeze_fs(sb);
1519 "VFS:Filesystem thaw failed\n");
1520 lockdep_sb_freeze_release(sb);
1521 up_write(&sb->s_umount);
1526 sb->s_writers.frozen = SB_UNFROZEN;
1527 sb_freeze_unlock(sb, SB_FREEZE_FS);
1529 wake_up(&sb->s_writers.wait_unfrozen);
1530 deactivate_locked_super(sb);
1533 EXPORT_SYMBOL(thaw_super);