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 mutex_init(&s->s_dquot.dqonoff_mutex);
248 s->s_maxbytes = MAX_NON_LFS;
249 s->s_op = &default_op;
250 s->s_time_gran = 1000000000;
251 s->cleancache_poolid = CLEANCACHE_NO_POOL;
253 s->s_shrink.seeks = DEFAULT_SEEKS;
254 s->s_shrink.scan_objects = super_cache_scan;
255 s->s_shrink.count_objects = super_cache_count;
256 s->s_shrink.batch = 1024;
257 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
265 /* Superblock refcounting */
268 * Drop a superblock's refcount. The caller must hold sb_lock.
270 static void __put_super(struct super_block *sb)
272 if (!--sb->s_count) {
273 list_del_init(&sb->s_list);
279 * put_super - drop a temporary reference to superblock
280 * @sb: superblock in question
282 * Drops a temporary reference, frees superblock if there's no
285 static void put_super(struct super_block *sb)
289 spin_unlock(&sb_lock);
294 * deactivate_locked_super - drop an active reference to superblock
295 * @s: superblock to deactivate
297 * Drops an active reference to superblock, converting it into a temporary
298 * one if there is no other active references left. In that case we
299 * tell fs driver to shut it down and drop the temporary reference we
302 * Caller holds exclusive lock on superblock; that lock is released.
304 void deactivate_locked_super(struct super_block *s)
306 struct file_system_type *fs = s->s_type;
307 if (atomic_dec_and_test(&s->s_active)) {
308 cleancache_invalidate_fs(s);
309 unregister_shrinker(&s->s_shrink);
313 * Since list_lru_destroy() may sleep, we cannot call it from
314 * put_super(), where we hold the sb_lock. Therefore we destroy
315 * the lru lists right now.
317 list_lru_destroy(&s->s_dentry_lru);
318 list_lru_destroy(&s->s_inode_lru);
323 up_write(&s->s_umount);
327 EXPORT_SYMBOL(deactivate_locked_super);
330 * deactivate_super - drop an active reference to superblock
331 * @s: superblock to deactivate
333 * Variant of deactivate_locked_super(), except that superblock is *not*
334 * locked by caller. If we are going to drop the final active reference,
335 * lock will be acquired prior to that.
337 void deactivate_super(struct super_block *s)
339 if (!atomic_add_unless(&s->s_active, -1, 1)) {
340 down_write(&s->s_umount);
341 deactivate_locked_super(s);
345 EXPORT_SYMBOL(deactivate_super);
348 * grab_super - acquire an active reference
349 * @s: reference we are trying to make active
351 * Tries to acquire an active reference. grab_super() is used when we
352 * had just found a superblock in super_blocks or fs_type->fs_supers
353 * and want to turn it into a full-blown active reference. grab_super()
354 * is called with sb_lock held and drops it. Returns 1 in case of
355 * success, 0 if we had failed (superblock contents was already dead or
356 * dying when grab_super() had been called). Note that this is only
357 * called for superblocks not in rundown mode (== ones still on ->fs_supers
358 * of their type), so increment of ->s_count is OK here.
360 static int grab_super(struct super_block *s) __releases(sb_lock)
363 spin_unlock(&sb_lock);
364 down_write(&s->s_umount);
365 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
369 up_write(&s->s_umount);
375 * trylock_super - try to grab ->s_umount shared
376 * @sb: reference we are trying to grab
378 * Try to prevent fs shutdown. This is used in places where we
379 * cannot take an active reference but we need to ensure that the
380 * filesystem is not shut down while we are working on it. It returns
381 * false if we cannot acquire s_umount or if we lose the race and
382 * filesystem already got into shutdown, and returns true with the s_umount
383 * lock held in read mode in case of success. On successful return,
384 * the caller must drop the s_umount lock when done.
386 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
387 * The reason why it's safe is that we are OK with doing trylock instead
388 * of down_read(). There's a couple of places that are OK with that, but
389 * it's very much not a general-purpose interface.
391 bool trylock_super(struct super_block *sb)
393 if (down_read_trylock(&sb->s_umount)) {
394 if (!hlist_unhashed(&sb->s_instances) &&
395 sb->s_root && (sb->s_flags & MS_BORN))
397 up_read(&sb->s_umount);
404 * generic_shutdown_super - common helper for ->kill_sb()
405 * @sb: superblock to kill
407 * generic_shutdown_super() does all fs-independent work on superblock
408 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
409 * that need destruction out of superblock, call generic_shutdown_super()
410 * and release aforementioned objects. Note: dentries and inodes _are_
411 * taken care of and do not need specific handling.
413 * Upon calling this function, the filesystem may no longer alter or
414 * rearrange the set of dentries belonging to this super_block, nor may it
415 * change the attachments of dentries to inodes.
417 void generic_shutdown_super(struct super_block *sb)
419 const struct super_operations *sop = sb->s_op;
422 shrink_dcache_for_umount(sb);
424 sb->s_flags &= ~MS_ACTIVE;
426 fsnotify_unmount_inodes(sb);
427 cgroup_writeback_umount();
431 if (sb->s_dio_done_wq) {
432 destroy_workqueue(sb->s_dio_done_wq);
433 sb->s_dio_done_wq = NULL;
439 if (!list_empty(&sb->s_inodes)) {
440 printk("VFS: Busy inodes after unmount of %s. "
441 "Self-destruct in 5 seconds. Have a nice day...\n",
446 /* should be initialized for __put_super_and_need_restart() */
447 hlist_del_init(&sb->s_instances);
448 spin_unlock(&sb_lock);
449 up_write(&sb->s_umount);
452 EXPORT_SYMBOL(generic_shutdown_super);
455 * sget_userns - find or create a superblock
456 * @type: filesystem type superblock should belong to
457 * @test: comparison callback
458 * @set: setup callback
459 * @flags: mount flags
460 * @user_ns: User namespace for the super_block
461 * @data: argument to each of them
463 struct super_block *sget_userns(struct file_system_type *type,
464 int (*test)(struct super_block *,void *),
465 int (*set)(struct super_block *,void *),
466 int flags, struct user_namespace *user_ns,
469 struct super_block *s = NULL;
470 struct super_block *old;
473 if (!(flags & MS_KERNMOUNT) &&
474 !(type->fs_flags & FS_USERNS_MOUNT) &&
475 !capable(CAP_SYS_ADMIN))
476 return ERR_PTR(-EPERM);
480 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
481 if (!test(old, data))
483 if (user_ns != old->s_user_ns) {
484 spin_unlock(&sb_lock);
486 up_write(&s->s_umount);
489 return ERR_PTR(-EBUSY);
491 if (!grab_super(old))
494 up_write(&s->s_umount);
502 spin_unlock(&sb_lock);
503 s = alloc_super(type, flags, user_ns);
505 return ERR_PTR(-ENOMEM);
511 spin_unlock(&sb_lock);
512 up_write(&s->s_umount);
517 strlcpy(s->s_id, type->name, sizeof(s->s_id));
518 list_add_tail(&s->s_list, &super_blocks);
519 hlist_add_head(&s->s_instances, &type->fs_supers);
520 spin_unlock(&sb_lock);
521 get_filesystem(type);
522 register_shrinker(&s->s_shrink);
526 EXPORT_SYMBOL(sget_userns);
529 * sget - find or create a superblock
530 * @type: filesystem type superblock should belong to
531 * @test: comparison callback
532 * @set: setup callback
533 * @flags: mount flags
534 * @data: argument to each of them
536 struct super_block *sget(struct file_system_type *type,
537 int (*test)(struct super_block *,void *),
538 int (*set)(struct super_block *,void *),
542 struct user_namespace *user_ns = current_user_ns();
544 /* Ensure the requestor has permissions over the target filesystem */
545 if (!(flags & MS_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
546 return ERR_PTR(-EPERM);
548 return sget_userns(type, test, set, flags, user_ns, data);
553 void drop_super(struct super_block *sb)
555 up_read(&sb->s_umount);
559 EXPORT_SYMBOL(drop_super);
562 * iterate_supers - call function for all active superblocks
563 * @f: function to call
564 * @arg: argument to pass to it
566 * Scans the superblock list and calls given function, passing it
567 * locked superblock and given argument.
569 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
571 struct super_block *sb, *p = NULL;
574 list_for_each_entry(sb, &super_blocks, s_list) {
575 if (hlist_unhashed(&sb->s_instances))
578 spin_unlock(&sb_lock);
580 down_read(&sb->s_umount);
581 if (sb->s_root && (sb->s_flags & MS_BORN))
583 up_read(&sb->s_umount);
592 spin_unlock(&sb_lock);
596 * iterate_supers_type - call function for superblocks of given type
598 * @f: function to call
599 * @arg: argument to pass to it
601 * Scans the superblock list and calls given function, passing it
602 * locked superblock and given argument.
604 void iterate_supers_type(struct file_system_type *type,
605 void (*f)(struct super_block *, void *), void *arg)
607 struct super_block *sb, *p = NULL;
610 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
612 spin_unlock(&sb_lock);
614 down_read(&sb->s_umount);
615 if (sb->s_root && (sb->s_flags & MS_BORN))
617 up_read(&sb->s_umount);
626 spin_unlock(&sb_lock);
629 EXPORT_SYMBOL(iterate_supers_type);
632 * get_super - get the superblock of a device
633 * @bdev: device to get the superblock for
635 * Scans the superblock list and finds the superblock of the file system
636 * mounted on the device given. %NULL is returned if no match is found.
639 struct super_block *get_super(struct block_device *bdev)
641 struct super_block *sb;
648 list_for_each_entry(sb, &super_blocks, s_list) {
649 if (hlist_unhashed(&sb->s_instances))
651 if (sb->s_bdev == bdev) {
653 spin_unlock(&sb_lock);
654 down_read(&sb->s_umount);
656 if (sb->s_root && (sb->s_flags & MS_BORN))
658 up_read(&sb->s_umount);
659 /* nope, got unmounted */
665 spin_unlock(&sb_lock);
669 EXPORT_SYMBOL(get_super);
672 * get_super_thawed - get thawed superblock of a device
673 * @bdev: device to get the superblock for
675 * Scans the superblock list and finds the superblock of the file system
676 * mounted on the device. The superblock is returned once it is thawed
677 * (or immediately if it was not frozen). %NULL is returned if no match
680 struct super_block *get_super_thawed(struct block_device *bdev)
683 struct super_block *s = get_super(bdev);
684 if (!s || s->s_writers.frozen == SB_UNFROZEN)
686 up_read(&s->s_umount);
687 wait_event(s->s_writers.wait_unfrozen,
688 s->s_writers.frozen == SB_UNFROZEN);
692 EXPORT_SYMBOL(get_super_thawed);
695 * get_active_super - get an active reference to the superblock of a device
696 * @bdev: device to get the superblock for
698 * Scans the superblock list and finds the superblock of the file system
699 * mounted on the device given. Returns the superblock with an active
700 * reference or %NULL if none was found.
702 struct super_block *get_active_super(struct block_device *bdev)
704 struct super_block *sb;
711 list_for_each_entry(sb, &super_blocks, s_list) {
712 if (hlist_unhashed(&sb->s_instances))
714 if (sb->s_bdev == bdev) {
717 up_write(&sb->s_umount);
721 spin_unlock(&sb_lock);
725 struct super_block *user_get_super(dev_t dev)
727 struct super_block *sb;
731 list_for_each_entry(sb, &super_blocks, s_list) {
732 if (hlist_unhashed(&sb->s_instances))
734 if (sb->s_dev == dev) {
736 spin_unlock(&sb_lock);
737 down_read(&sb->s_umount);
739 if (sb->s_root && (sb->s_flags & MS_BORN))
741 up_read(&sb->s_umount);
742 /* nope, got unmounted */
748 spin_unlock(&sb_lock);
753 * do_remount_sb - asks filesystem to change mount options.
754 * @sb: superblock in question
755 * @flags: numeric part of options
756 * @data: the rest of options
757 * @force: whether or not to force the change
759 * Alters the mount options of a mounted file system.
761 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
766 if (sb->s_writers.frozen != SB_UNFROZEN)
770 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
774 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
777 if (!hlist_empty(&sb->s_pins)) {
778 up_write(&sb->s_umount);
779 group_pin_kill(&sb->s_pins);
780 down_write(&sb->s_umount);
783 if (sb->s_writers.frozen != SB_UNFROZEN)
785 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
788 shrink_dcache_sb(sb);
790 /* If we are remounting RDONLY and current sb is read/write,
791 make sure there are no rw files opened */
794 sb->s_readonly_remount = 1;
797 retval = sb_prepare_remount_readonly(sb);
803 if (sb->s_op->remount_fs) {
804 retval = sb->s_op->remount_fs(sb, &flags, data);
807 goto cancel_readonly;
808 /* If forced remount, go ahead despite any errors */
809 WARN(1, "forced remount of a %s fs returned %i\n",
810 sb->s_type->name, retval);
813 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
814 /* Needs to be ordered wrt mnt_is_readonly() */
816 sb->s_readonly_remount = 0;
819 * Some filesystems modify their metadata via some other path than the
820 * bdev buffer cache (eg. use a private mapping, or directories in
821 * pagecache, etc). Also file data modifications go via their own
822 * mappings. So If we try to mount readonly then copy the filesystem
823 * from bdev, we could get stale data, so invalidate it to give a best
824 * effort at coherency.
826 if (remount_ro && sb->s_bdev)
827 invalidate_bdev(sb->s_bdev);
831 sb->s_readonly_remount = 0;
835 static void do_emergency_remount(struct work_struct *work)
837 struct super_block *sb, *p = NULL;
840 list_for_each_entry(sb, &super_blocks, s_list) {
841 if (hlist_unhashed(&sb->s_instances))
844 spin_unlock(&sb_lock);
845 down_write(&sb->s_umount);
846 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
847 !(sb->s_flags & MS_RDONLY)) {
849 * What lock protects sb->s_flags??
851 do_remount_sb(sb, MS_RDONLY, NULL, 1);
853 up_write(&sb->s_umount);
861 spin_unlock(&sb_lock);
863 printk("Emergency Remount complete\n");
866 void emergency_remount(void)
868 struct work_struct *work;
870 work = kmalloc(sizeof(*work), GFP_ATOMIC);
872 INIT_WORK(work, do_emergency_remount);
878 * Unnamed block devices are dummy devices used by virtual
879 * filesystems which don't use real block-devices. -- jrs
882 static DEFINE_IDA(unnamed_dev_ida);
883 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
884 /* Many userspace utilities consider an FSID of 0 invalid.
885 * Always return at least 1 from get_anon_bdev.
887 static int unnamed_dev_start = 1;
889 int get_anon_bdev(dev_t *p)
895 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
897 spin_lock(&unnamed_dev_lock);
898 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
900 unnamed_dev_start = dev + 1;
901 spin_unlock(&unnamed_dev_lock);
902 if (error == -EAGAIN)
903 /* We raced and lost with another CPU. */
908 if (dev >= (1 << MINORBITS)) {
909 spin_lock(&unnamed_dev_lock);
910 ida_remove(&unnamed_dev_ida, dev);
911 if (unnamed_dev_start > dev)
912 unnamed_dev_start = dev;
913 spin_unlock(&unnamed_dev_lock);
916 *p = MKDEV(0, dev & MINORMASK);
919 EXPORT_SYMBOL(get_anon_bdev);
921 void free_anon_bdev(dev_t dev)
923 int slot = MINOR(dev);
924 spin_lock(&unnamed_dev_lock);
925 ida_remove(&unnamed_dev_ida, slot);
926 if (slot < unnamed_dev_start)
927 unnamed_dev_start = slot;
928 spin_unlock(&unnamed_dev_lock);
930 EXPORT_SYMBOL(free_anon_bdev);
932 int set_anon_super(struct super_block *s, void *data)
934 return get_anon_bdev(&s->s_dev);
937 EXPORT_SYMBOL(set_anon_super);
939 void kill_anon_super(struct super_block *sb)
941 dev_t dev = sb->s_dev;
942 generic_shutdown_super(sb);
946 EXPORT_SYMBOL(kill_anon_super);
948 void kill_litter_super(struct super_block *sb)
951 d_genocide(sb->s_root);
955 EXPORT_SYMBOL(kill_litter_super);
957 static int ns_test_super(struct super_block *sb, void *data)
959 return sb->s_fs_info == data;
962 static int ns_set_super(struct super_block *sb, void *data)
964 sb->s_fs_info = data;
965 return set_anon_super(sb, NULL);
968 struct dentry *mount_ns(struct file_system_type *fs_type,
969 int flags, void *data, void *ns, struct user_namespace *user_ns,
970 int (*fill_super)(struct super_block *, void *, int))
972 struct super_block *sb;
974 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
975 * over the namespace.
977 if (!(flags & MS_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
978 return ERR_PTR(-EPERM);
980 sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
987 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
989 deactivate_locked_super(sb);
993 sb->s_flags |= MS_ACTIVE;
996 return dget(sb->s_root);
999 EXPORT_SYMBOL(mount_ns);
1002 static int set_bdev_super(struct super_block *s, void *data)
1005 s->s_dev = s->s_bdev->bd_dev;
1008 * We set the bdi here to the queue backing, file systems can
1009 * overwrite this in ->fill_super()
1011 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1015 static int test_bdev_super(struct super_block *s, void *data)
1017 return (void *)s->s_bdev == data;
1020 struct dentry *mount_bdev(struct file_system_type *fs_type,
1021 int flags, const char *dev_name, void *data,
1022 int (*fill_super)(struct super_block *, void *, int))
1024 struct block_device *bdev;
1025 struct super_block *s;
1026 fmode_t mode = FMODE_READ | FMODE_EXCL;
1029 if (!(flags & MS_RDONLY))
1030 mode |= FMODE_WRITE;
1032 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1034 return ERR_CAST(bdev);
1037 * once the super is inserted into the list by sget, s_umount
1038 * will protect the lockfs code from trying to start a snapshot
1039 * while we are mounting
1041 mutex_lock(&bdev->bd_fsfreeze_mutex);
1042 if (bdev->bd_fsfreeze_count > 0) {
1043 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1047 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
1049 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1054 if ((flags ^ s->s_flags) & MS_RDONLY) {
1055 deactivate_locked_super(s);
1061 * s_umount nests inside bd_mutex during
1062 * __invalidate_device(). blkdev_put() acquires
1063 * bd_mutex and can't be called under s_umount. Drop
1064 * s_umount temporarily. This is safe as we're
1065 * holding an active reference.
1067 up_write(&s->s_umount);
1068 blkdev_put(bdev, mode);
1069 down_write(&s->s_umount);
1072 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1073 sb_set_blocksize(s, block_size(bdev));
1074 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1076 deactivate_locked_super(s);
1080 s->s_flags |= MS_ACTIVE;
1084 return dget(s->s_root);
1089 blkdev_put(bdev, mode);
1091 return ERR_PTR(error);
1093 EXPORT_SYMBOL(mount_bdev);
1095 void kill_block_super(struct super_block *sb)
1097 struct block_device *bdev = sb->s_bdev;
1098 fmode_t mode = sb->s_mode;
1100 bdev->bd_super = NULL;
1101 generic_shutdown_super(sb);
1102 sync_blockdev(bdev);
1103 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1104 blkdev_put(bdev, mode | FMODE_EXCL);
1107 EXPORT_SYMBOL(kill_block_super);
1110 struct dentry *mount_nodev(struct file_system_type *fs_type,
1111 int flags, void *data,
1112 int (*fill_super)(struct super_block *, void *, int))
1115 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1120 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1122 deactivate_locked_super(s);
1123 return ERR_PTR(error);
1125 s->s_flags |= MS_ACTIVE;
1126 return dget(s->s_root);
1128 EXPORT_SYMBOL(mount_nodev);
1130 static int compare_single(struct super_block *s, void *p)
1135 struct dentry *mount_single(struct file_system_type *fs_type,
1136 int flags, void *data,
1137 int (*fill_super)(struct super_block *, void *, int))
1139 struct super_block *s;
1142 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1146 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1148 deactivate_locked_super(s);
1149 return ERR_PTR(error);
1151 s->s_flags |= MS_ACTIVE;
1153 do_remount_sb(s, flags, data, 0);
1155 return dget(s->s_root);
1157 EXPORT_SYMBOL(mount_single);
1160 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1162 struct dentry *root;
1163 struct super_block *sb;
1164 char *secdata = NULL;
1165 int error = -ENOMEM;
1167 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1168 secdata = alloc_secdata();
1172 error = security_sb_copy_data(data, secdata);
1174 goto out_free_secdata;
1177 root = type->mount(type, flags, name, data);
1179 error = PTR_ERR(root);
1180 goto out_free_secdata;
1184 WARN_ON(!sb->s_bdi);
1185 sb->s_flags |= MS_BORN;
1187 error = security_sb_kern_mount(sb, flags, secdata);
1192 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1193 * but s_maxbytes was an unsigned long long for many releases. Throw
1194 * this warning for a little while to try and catch filesystems that
1195 * violate this rule.
1197 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1198 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1200 up_write(&sb->s_umount);
1201 free_secdata(secdata);
1205 deactivate_locked_super(sb);
1207 free_secdata(secdata);
1209 return ERR_PTR(error);
1213 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1216 void __sb_end_write(struct super_block *sb, int level)
1218 percpu_up_read(sb->s_writers.rw_sem + level-1);
1220 EXPORT_SYMBOL(__sb_end_write);
1223 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1226 int __sb_start_write(struct super_block *sb, int level, bool wait)
1228 bool force_trylock = false;
1231 #ifdef CONFIG_LOCKDEP
1233 * We want lockdep to tell us about possible deadlocks with freezing
1234 * but it's it bit tricky to properly instrument it. Getting a freeze
1235 * protection works as getting a read lock but there are subtle
1236 * problems. XFS for example gets freeze protection on internal level
1237 * twice in some cases, which is OK only because we already hold a
1238 * freeze protection also on higher level. Due to these cases we have
1239 * to use wait == F (trylock mode) which must not fail.
1244 for (i = 0; i < level - 1; i++)
1245 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
1246 force_trylock = true;
1251 if (wait && !force_trylock)
1252 percpu_down_read(sb->s_writers.rw_sem + level-1);
1254 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1256 WARN_ON(force_trylock && !ret);
1259 EXPORT_SYMBOL(__sb_start_write);
1262 * sb_wait_write - wait until all writers to given file system finish
1263 * @sb: the super for which we wait
1264 * @level: type of writers we wait for (normal vs page fault)
1266 * This function waits until there are no writers of given type to given file
1269 static void sb_wait_write(struct super_block *sb, int level)
1271 percpu_down_write(sb->s_writers.rw_sem + level-1);
1273 * We are going to return to userspace and forget about this lock, the
1274 * ownership goes to the caller of thaw_super() which does unlock.
1276 * FIXME: we should do this before return from freeze_super() after we
1277 * called sync_filesystem(sb) and s_op->freeze_fs(sb), and thaw_super()
1278 * should re-acquire these locks before s_op->unfreeze_fs(sb). However
1279 * this leads to lockdep false-positives, so currently we do the early
1280 * release right after acquire.
1282 percpu_rwsem_release(sb->s_writers.rw_sem + level-1, 0, _THIS_IP_);
1285 static void sb_freeze_unlock(struct super_block *sb)
1289 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1290 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1292 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1293 percpu_up_write(sb->s_writers.rw_sem + level);
1297 * freeze_super - lock the filesystem and force it into a consistent state
1298 * @sb: the super to lock
1300 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1301 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1304 * During this function, sb->s_writers.frozen goes through these values:
1306 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1308 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1309 * writes should be blocked, though page faults are still allowed. We wait for
1310 * all writes to complete and then proceed to the next stage.
1312 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1313 * but internal fs threads can still modify the filesystem (although they
1314 * should not dirty new pages or inodes), writeback can run etc. After waiting
1315 * for all running page faults we sync the filesystem which will clean all
1316 * dirty pages and inodes (no new dirty pages or inodes can be created when
1319 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1320 * modification are blocked (e.g. XFS preallocation truncation on inode
1321 * reclaim). This is usually implemented by blocking new transactions for
1322 * filesystems that have them and need this additional guard. After all
1323 * internal writers are finished we call ->freeze_fs() to finish filesystem
1324 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1325 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1327 * sb->s_writers.frozen is protected by sb->s_umount.
1329 int freeze_super(struct super_block *sb)
1333 atomic_inc(&sb->s_active);
1334 down_write(&sb->s_umount);
1335 if (sb->s_writers.frozen != SB_UNFROZEN) {
1336 deactivate_locked_super(sb);
1340 if (!(sb->s_flags & MS_BORN)) {
1341 up_write(&sb->s_umount);
1342 return 0; /* sic - it's "nothing to do" */
1345 if (sb->s_flags & MS_RDONLY) {
1346 /* Nothing to do really... */
1347 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1348 up_write(&sb->s_umount);
1352 sb->s_writers.frozen = SB_FREEZE_WRITE;
1353 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1354 up_write(&sb->s_umount);
1355 sb_wait_write(sb, SB_FREEZE_WRITE);
1356 down_write(&sb->s_umount);
1358 /* Now we go and block page faults... */
1359 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1360 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1362 /* All writers are done so after syncing there won't be dirty data */
1363 sync_filesystem(sb);
1365 /* Now wait for internal filesystem counter */
1366 sb->s_writers.frozen = SB_FREEZE_FS;
1367 sb_wait_write(sb, SB_FREEZE_FS);
1369 if (sb->s_op->freeze_fs) {
1370 ret = sb->s_op->freeze_fs(sb);
1373 "VFS:Filesystem freeze failed\n");
1374 sb->s_writers.frozen = SB_UNFROZEN;
1375 sb_freeze_unlock(sb);
1376 wake_up(&sb->s_writers.wait_unfrozen);
1377 deactivate_locked_super(sb);
1382 * This is just for debugging purposes so that fs can warn if it
1383 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1385 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1386 up_write(&sb->s_umount);
1389 EXPORT_SYMBOL(freeze_super);
1392 * thaw_super -- unlock filesystem
1393 * @sb: the super to thaw
1395 * Unlocks the filesystem and marks it writeable again after freeze_super().
1397 int thaw_super(struct super_block *sb)
1401 down_write(&sb->s_umount);
1402 if (sb->s_writers.frozen == SB_UNFROZEN) {
1403 up_write(&sb->s_umount);
1407 if (sb->s_flags & MS_RDONLY) {
1408 sb->s_writers.frozen = SB_UNFROZEN;
1412 if (sb->s_op->unfreeze_fs) {
1413 error = sb->s_op->unfreeze_fs(sb);
1416 "VFS:Filesystem thaw failed\n");
1417 up_write(&sb->s_umount);
1422 sb->s_writers.frozen = SB_UNFROZEN;
1423 sb_freeze_unlock(sb);
1425 wake_up(&sb->s_writers.wait_unfrozen);
1426 deactivate_locked_super(sb);
1429 EXPORT_SYMBOL(thaw_super);