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:
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>
40 static int thaw_super_locked(struct super_block *sb);
42 static LIST_HEAD(super_blocks);
43 static DEFINE_SPINLOCK(sb_lock);
45 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
52 * One thing we have to be careful of with a per-sb shrinker is that we don't
53 * drop the last active reference to the superblock from within the shrinker.
54 * If that happens we could trigger unregistering the shrinker from within the
55 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
56 * take a passive reference to the superblock to avoid this from occurring.
58 static unsigned long super_cache_scan(struct shrinker *shrink,
59 struct shrink_control *sc)
61 struct super_block *sb;
68 sb = container_of(shrink, struct super_block, s_shrink);
71 * Deadlock avoidance. We may hold various FS locks, and we don't want
72 * to recurse into the FS that called us in clear_inode() and friends..
74 if (!(sc->gfp_mask & __GFP_FS))
77 if (!trylock_super(sb))
80 if (sb->s_op->nr_cached_objects)
81 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
83 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
84 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
85 total_objects = dentries + inodes + fs_objects + 1;
89 /* proportion the scan between the caches */
90 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
91 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
92 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
95 * prune the dcache first as the icache is pinned by it, then
96 * prune the icache, followed by the filesystem specific caches
98 * Ensure that we always scan at least one object - memcg kmem
99 * accounting uses this to fully empty the caches.
101 sc->nr_to_scan = dentries + 1;
102 freed = prune_dcache_sb(sb, sc);
103 sc->nr_to_scan = inodes + 1;
104 freed += prune_icache_sb(sb, sc);
107 sc->nr_to_scan = fs_objects + 1;
108 freed += sb->s_op->free_cached_objects(sb, sc);
111 up_read(&sb->s_umount);
115 static unsigned long super_cache_count(struct shrinker *shrink,
116 struct shrink_control *sc)
118 struct super_block *sb;
119 long total_objects = 0;
121 sb = container_of(shrink, struct super_block, s_shrink);
124 * We don't call trylock_super() here as it is a scalability bottleneck,
125 * so we're exposed to partial setup state. The shrinker rwsem does not
126 * protect filesystem operations backing list_lru_shrink_count() or
127 * s_op->nr_cached_objects(). Counts can change between
128 * super_cache_count and super_cache_scan, so we really don't need locks
131 * However, if we are currently mounting the superblock, the underlying
132 * filesystem might be in a state of partial construction and hence it
133 * is dangerous to access it. trylock_super() uses a SB_BORN check to
134 * avoid this situation, so do the same here. The memory barrier is
135 * matched with the one in mount_fs() as we don't hold locks here.
137 if (!(sb->s_flags & SB_BORN))
141 if (sb->s_op && sb->s_op->nr_cached_objects)
142 total_objects = sb->s_op->nr_cached_objects(sb, sc);
144 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
145 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
147 total_objects = vfs_pressure_ratio(total_objects);
148 return total_objects;
151 static void destroy_super_work(struct work_struct *work)
153 struct super_block *s = container_of(work, struct super_block,
157 for (i = 0; i < SB_FREEZE_LEVELS; i++)
158 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
162 static void destroy_super_rcu(struct rcu_head *head)
164 struct super_block *s = container_of(head, struct super_block, rcu);
165 INIT_WORK(&s->destroy_work, destroy_super_work);
166 schedule_work(&s->destroy_work);
169 /* Free a superblock that has never been seen by anyone */
170 static void destroy_unused_super(struct super_block *s)
174 up_write(&s->s_umount);
175 list_lru_destroy(&s->s_dentry_lru);
176 list_lru_destroy(&s->s_inode_lru);
178 put_user_ns(s->s_user_ns);
180 free_prealloced_shrinker(&s->s_shrink);
181 /* no delays needed */
182 destroy_super_work(&s->destroy_work);
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);
206 init_rwsem(&s->s_umount);
207 lockdep_set_class(&s->s_umount, &type->s_umount_key);
209 * sget() can have s_umount recursion.
211 * When it cannot find a suitable sb, it allocates a new
212 * one (this one), and tries again to find a suitable old
215 * In case that succeeds, it will acquire the s_umount
216 * lock of the old one. Since these are clearly distrinct
217 * locks, and this object isn't exposed yet, there's no
220 * Annotate this by putting this lock in a different
223 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
225 if (security_sb_alloc(s))
228 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
229 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
231 &type->s_writers_key[i]))
234 init_waitqueue_head(&s->s_writers.wait_unfrozen);
235 s->s_bdi = &noop_backing_dev_info;
237 if (s->s_user_ns != &init_user_ns)
238 s->s_iflags |= SB_I_NODEV;
239 INIT_HLIST_NODE(&s->s_instances);
240 INIT_HLIST_BL_HEAD(&s->s_roots);
241 mutex_init(&s->s_sync_lock);
242 INIT_LIST_HEAD(&s->s_inodes);
243 spin_lock_init(&s->s_inode_list_lock);
244 INIT_LIST_HEAD(&s->s_inodes_wb);
245 spin_lock_init(&s->s_inode_wblist_lock);
247 if (list_lru_init_memcg(&s->s_dentry_lru))
249 if (list_lru_init_memcg(&s->s_inode_lru))
252 atomic_set(&s->s_active, 1);
253 mutex_init(&s->s_vfs_rename_mutex);
254 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
255 init_rwsem(&s->s_dquot.dqio_sem);
256 s->s_maxbytes = MAX_NON_LFS;
257 s->s_op = &default_op;
258 s->s_time_gran = 1000000000;
259 s->cleancache_poolid = CLEANCACHE_NO_POOL;
261 s->s_shrink.seeks = DEFAULT_SEEKS;
262 s->s_shrink.scan_objects = super_cache_scan;
263 s->s_shrink.count_objects = super_cache_count;
264 s->s_shrink.batch = 1024;
265 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
266 if (prealloc_shrinker(&s->s_shrink))
271 destroy_unused_super(s);
275 /* Superblock refcounting */
278 * Drop a superblock's refcount. The caller must hold sb_lock.
280 static void __put_super(struct super_block *s)
283 list_del_init(&s->s_list);
284 WARN_ON(s->s_dentry_lru.node);
285 WARN_ON(s->s_inode_lru.node);
286 WARN_ON(!list_empty(&s->s_mounts));
288 put_user_ns(s->s_user_ns);
290 call_rcu(&s->rcu, destroy_super_rcu);
295 * put_super - drop a temporary reference to superblock
296 * @sb: superblock in question
298 * Drops a temporary reference, frees superblock if there's no
301 static void put_super(struct super_block *sb)
305 spin_unlock(&sb_lock);
310 * deactivate_locked_super - drop an active reference to superblock
311 * @s: superblock to deactivate
313 * Drops an active reference to superblock, converting it into a temporary
314 * one if there is no other active references left. In that case we
315 * tell fs driver to shut it down and drop the temporary reference we
318 * Caller holds exclusive lock on superblock; that lock is released.
320 void deactivate_locked_super(struct super_block *s)
322 struct file_system_type *fs = s->s_type;
323 if (atomic_dec_and_test(&s->s_active)) {
324 cleancache_invalidate_fs(s);
325 unregister_shrinker(&s->s_shrink);
329 * Since list_lru_destroy() may sleep, we cannot call it from
330 * put_super(), where we hold the sb_lock. Therefore we destroy
331 * the lru lists right now.
333 list_lru_destroy(&s->s_dentry_lru);
334 list_lru_destroy(&s->s_inode_lru);
339 up_write(&s->s_umount);
343 EXPORT_SYMBOL(deactivate_locked_super);
346 * deactivate_super - drop an active reference to superblock
347 * @s: superblock to deactivate
349 * Variant of deactivate_locked_super(), except that superblock is *not*
350 * locked by caller. If we are going to drop the final active reference,
351 * lock will be acquired prior to that.
353 void deactivate_super(struct super_block *s)
355 if (!atomic_add_unless(&s->s_active, -1, 1)) {
356 down_write(&s->s_umount);
357 deactivate_locked_super(s);
361 EXPORT_SYMBOL(deactivate_super);
364 * grab_super - acquire an active reference
365 * @s: reference we are trying to make active
367 * Tries to acquire an active reference. grab_super() is used when we
368 * had just found a superblock in super_blocks or fs_type->fs_supers
369 * and want to turn it into a full-blown active reference. grab_super()
370 * is called with sb_lock held and drops it. Returns 1 in case of
371 * success, 0 if we had failed (superblock contents was already dead or
372 * dying when grab_super() had been called). Note that this is only
373 * called for superblocks not in rundown mode (== ones still on ->fs_supers
374 * of their type), so increment of ->s_count is OK here.
376 static int grab_super(struct super_block *s) __releases(sb_lock)
379 spin_unlock(&sb_lock);
380 down_write(&s->s_umount);
381 if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
385 up_write(&s->s_umount);
391 * trylock_super - try to grab ->s_umount shared
392 * @sb: reference we are trying to grab
394 * Try to prevent fs shutdown. This is used in places where we
395 * cannot take an active reference but we need to ensure that the
396 * filesystem is not shut down while we are working on it. It returns
397 * false if we cannot acquire s_umount or if we lose the race and
398 * filesystem already got into shutdown, and returns true with the s_umount
399 * lock held in read mode in case of success. On successful return,
400 * the caller must drop the s_umount lock when done.
402 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
403 * The reason why it's safe is that we are OK with doing trylock instead
404 * of down_read(). There's a couple of places that are OK with that, but
405 * it's very much not a general-purpose interface.
407 bool trylock_super(struct super_block *sb)
409 if (down_read_trylock(&sb->s_umount)) {
410 if (!hlist_unhashed(&sb->s_instances) &&
411 sb->s_root && (sb->s_flags & SB_BORN))
413 up_read(&sb->s_umount);
420 * generic_shutdown_super - common helper for ->kill_sb()
421 * @sb: superblock to kill
423 * generic_shutdown_super() does all fs-independent work on superblock
424 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
425 * that need destruction out of superblock, call generic_shutdown_super()
426 * and release aforementioned objects. Note: dentries and inodes _are_
427 * taken care of and do not need specific handling.
429 * Upon calling this function, the filesystem may no longer alter or
430 * rearrange the set of dentries belonging to this super_block, nor may it
431 * change the attachments of dentries to inodes.
433 void generic_shutdown_super(struct super_block *sb)
435 const struct super_operations *sop = sb->s_op;
438 shrink_dcache_for_umount(sb);
440 sb->s_flags &= ~SB_ACTIVE;
442 fsnotify_unmount_inodes(sb);
443 cgroup_writeback_umount();
447 if (sb->s_dio_done_wq) {
448 destroy_workqueue(sb->s_dio_done_wq);
449 sb->s_dio_done_wq = NULL;
455 if (!list_empty(&sb->s_inodes)) {
456 printk("VFS: Busy inodes after unmount of %s. "
457 "Self-destruct in 5 seconds. Have a nice day...\n",
462 /* should be initialized for __put_super_and_need_restart() */
463 hlist_del_init(&sb->s_instances);
464 spin_unlock(&sb_lock);
465 up_write(&sb->s_umount);
466 if (sb->s_bdi != &noop_backing_dev_info) {
468 sb->s_bdi = &noop_backing_dev_info;
472 EXPORT_SYMBOL(generic_shutdown_super);
475 * sget_userns - find or create a superblock
476 * @type: filesystem type superblock should belong to
477 * @test: comparison callback
478 * @set: setup callback
479 * @flags: mount flags
480 * @user_ns: User namespace for the super_block
481 * @data: argument to each of them
483 struct super_block *sget_userns(struct file_system_type *type,
484 int (*test)(struct super_block *,void *),
485 int (*set)(struct super_block *,void *),
486 int flags, struct user_namespace *user_ns,
489 struct super_block *s = NULL;
490 struct super_block *old;
493 if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) &&
494 !(type->fs_flags & FS_USERNS_MOUNT) &&
495 !capable(CAP_SYS_ADMIN))
496 return ERR_PTR(-EPERM);
500 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
501 if (!test(old, data))
503 if (user_ns != old->s_user_ns) {
504 spin_unlock(&sb_lock);
505 destroy_unused_super(s);
506 return ERR_PTR(-EBUSY);
508 if (!grab_super(old))
510 destroy_unused_super(s);
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 destroy_unused_super(s);
529 strlcpy(s->s_id, type->name, sizeof(s->s_id));
530 list_add_tail(&s->s_list, &super_blocks);
531 hlist_add_head(&s->s_instances, &type->fs_supers);
532 spin_unlock(&sb_lock);
533 get_filesystem(type);
534 register_shrinker_prepared(&s->s_shrink);
538 EXPORT_SYMBOL(sget_userns);
541 * sget - find or create a superblock
542 * @type: filesystem type superblock should belong to
543 * @test: comparison callback
544 * @set: setup callback
545 * @flags: mount flags
546 * @data: argument to each of them
548 struct super_block *sget(struct file_system_type *type,
549 int (*test)(struct super_block *,void *),
550 int (*set)(struct super_block *,void *),
554 struct user_namespace *user_ns = current_user_ns();
556 /* We don't yet pass the user namespace of the parent
557 * mount through to here so always use &init_user_ns
558 * until that changes.
560 if (flags & SB_SUBMOUNT)
561 user_ns = &init_user_ns;
563 /* Ensure the requestor has permissions over the target filesystem */
564 if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) && !ns_capable(user_ns, CAP_SYS_ADMIN))
565 return ERR_PTR(-EPERM);
567 return sget_userns(type, test, set, flags, user_ns, data);
572 void drop_super(struct super_block *sb)
574 up_read(&sb->s_umount);
578 EXPORT_SYMBOL(drop_super);
580 void drop_super_exclusive(struct super_block *sb)
582 up_write(&sb->s_umount);
585 EXPORT_SYMBOL(drop_super_exclusive);
587 static void __iterate_supers(void (*f)(struct super_block *))
589 struct super_block *sb, *p = NULL;
592 list_for_each_entry(sb, &super_blocks, s_list) {
593 if (hlist_unhashed(&sb->s_instances))
596 spin_unlock(&sb_lock);
607 spin_unlock(&sb_lock);
610 * iterate_supers - call function for all active superblocks
611 * @f: function to call
612 * @arg: argument to pass to it
614 * Scans the superblock list and calls given function, passing it
615 * locked superblock and given argument.
617 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
619 struct super_block *sb, *p = NULL;
622 list_for_each_entry(sb, &super_blocks, s_list) {
623 if (hlist_unhashed(&sb->s_instances))
626 spin_unlock(&sb_lock);
628 down_read(&sb->s_umount);
629 if (sb->s_root && (sb->s_flags & SB_BORN))
631 up_read(&sb->s_umount);
640 spin_unlock(&sb_lock);
644 * iterate_supers_type - call function for superblocks of given type
646 * @f: function to call
647 * @arg: argument to pass to it
649 * Scans the superblock list and calls given function, passing it
650 * locked superblock and given argument.
652 void iterate_supers_type(struct file_system_type *type,
653 void (*f)(struct super_block *, void *), void *arg)
655 struct super_block *sb, *p = NULL;
658 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
660 spin_unlock(&sb_lock);
662 down_read(&sb->s_umount);
663 if (sb->s_root && (sb->s_flags & SB_BORN))
665 up_read(&sb->s_umount);
674 spin_unlock(&sb_lock);
677 EXPORT_SYMBOL(iterate_supers_type);
679 static struct super_block *__get_super(struct block_device *bdev, bool excl)
681 struct super_block *sb;
688 list_for_each_entry(sb, &super_blocks, s_list) {
689 if (hlist_unhashed(&sb->s_instances))
691 if (sb->s_bdev == bdev) {
693 spin_unlock(&sb_lock);
695 down_read(&sb->s_umount);
697 down_write(&sb->s_umount);
699 if (sb->s_root && (sb->s_flags & SB_BORN))
702 up_read(&sb->s_umount);
704 up_write(&sb->s_umount);
705 /* nope, got unmounted */
711 spin_unlock(&sb_lock);
716 * get_super - get the superblock of a device
717 * @bdev: device to get the superblock for
719 * Scans the superblock list and finds the superblock of the file system
720 * mounted on the device given. %NULL is returned if no match is found.
722 struct super_block *get_super(struct block_device *bdev)
724 return __get_super(bdev, false);
726 EXPORT_SYMBOL(get_super);
728 static struct super_block *__get_super_thawed(struct block_device *bdev,
732 struct super_block *s = __get_super(bdev, excl);
733 if (!s || s->s_writers.frozen == SB_UNFROZEN)
736 up_read(&s->s_umount);
738 up_write(&s->s_umount);
739 wait_event(s->s_writers.wait_unfrozen,
740 s->s_writers.frozen == SB_UNFROZEN);
746 * get_super_thawed - get thawed superblock of a device
747 * @bdev: device to get the superblock for
749 * Scans the superblock list and finds the superblock of the file system
750 * mounted on the device. The superblock is returned once it is thawed
751 * (or immediately if it was not frozen). %NULL is returned if no match
754 struct super_block *get_super_thawed(struct block_device *bdev)
756 return __get_super_thawed(bdev, false);
758 EXPORT_SYMBOL(get_super_thawed);
761 * get_super_exclusive_thawed - get thawed superblock of a device
762 * @bdev: device to get the superblock for
764 * Scans the superblock list and finds the superblock of the file system
765 * mounted on the device. The superblock is returned once it is thawed
766 * (or immediately if it was not frozen) and s_umount semaphore is held
767 * in exclusive mode. %NULL is returned if no match is found.
769 struct super_block *get_super_exclusive_thawed(struct block_device *bdev)
771 return __get_super_thawed(bdev, true);
773 EXPORT_SYMBOL(get_super_exclusive_thawed);
776 * get_active_super - get an active reference to the superblock of a device
777 * @bdev: device to get the superblock for
779 * Scans the superblock list and finds the superblock of the file system
780 * mounted on the device given. Returns the superblock with an active
781 * reference or %NULL if none was found.
783 struct super_block *get_active_super(struct block_device *bdev)
785 struct super_block *sb;
792 list_for_each_entry(sb, &super_blocks, s_list) {
793 if (hlist_unhashed(&sb->s_instances))
795 if (sb->s_bdev == bdev) {
798 up_write(&sb->s_umount);
802 spin_unlock(&sb_lock);
806 struct super_block *user_get_super(dev_t dev)
808 struct super_block *sb;
812 list_for_each_entry(sb, &super_blocks, s_list) {
813 if (hlist_unhashed(&sb->s_instances))
815 if (sb->s_dev == dev) {
817 spin_unlock(&sb_lock);
818 down_read(&sb->s_umount);
820 if (sb->s_root && (sb->s_flags & SB_BORN))
822 up_read(&sb->s_umount);
823 /* nope, got unmounted */
829 spin_unlock(&sb_lock);
834 * do_remount_sb - asks filesystem to change mount options.
835 * @sb: superblock in question
836 * @sb_flags: revised superblock flags
837 * @data: the rest of options
838 * @force: whether or not to force the change
840 * Alters the mount options of a mounted file system.
842 int do_remount_sb(struct super_block *sb, int sb_flags, void *data, int force)
847 if (sb->s_writers.frozen != SB_UNFROZEN)
851 if (!(sb_flags & SB_RDONLY) && bdev_read_only(sb->s_bdev))
855 remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
858 if (!hlist_empty(&sb->s_pins)) {
859 up_write(&sb->s_umount);
860 group_pin_kill(&sb->s_pins);
861 down_write(&sb->s_umount);
864 if (sb->s_writers.frozen != SB_UNFROZEN)
866 remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb);
869 shrink_dcache_sb(sb);
871 /* If we are remounting RDONLY and current sb is read/write,
872 make sure there are no rw files opened */
875 sb->s_readonly_remount = 1;
878 retval = sb_prepare_remount_readonly(sb);
884 if (sb->s_op->remount_fs) {
885 retval = sb->s_op->remount_fs(sb, &sb_flags, data);
888 goto cancel_readonly;
889 /* If forced remount, go ahead despite any errors */
890 WARN(1, "forced remount of a %s fs returned %i\n",
891 sb->s_type->name, retval);
894 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (sb_flags & MS_RMT_MASK);
895 /* Needs to be ordered wrt mnt_is_readonly() */
897 sb->s_readonly_remount = 0;
900 * Some filesystems modify their metadata via some other path than the
901 * bdev buffer cache (eg. use a private mapping, or directories in
902 * pagecache, etc). Also file data modifications go via their own
903 * mappings. So If we try to mount readonly then copy the filesystem
904 * from bdev, we could get stale data, so invalidate it to give a best
905 * effort at coherency.
907 if (remount_ro && sb->s_bdev)
908 invalidate_bdev(sb->s_bdev);
912 sb->s_readonly_remount = 0;
916 static void do_emergency_remount_callback(struct super_block *sb)
918 down_write(&sb->s_umount);
919 if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
922 * What lock protects sb->s_flags??
924 do_remount_sb(sb, SB_RDONLY, NULL, 1);
926 up_write(&sb->s_umount);
929 static void do_emergency_remount(struct work_struct *work)
931 __iterate_supers(do_emergency_remount_callback);
933 printk("Emergency Remount complete\n");
936 void emergency_remount(void)
938 struct work_struct *work;
940 work = kmalloc(sizeof(*work), GFP_ATOMIC);
942 INIT_WORK(work, do_emergency_remount);
947 static void do_thaw_all_callback(struct super_block *sb)
949 down_write(&sb->s_umount);
950 if (sb->s_root && sb->s_flags & SB_BORN) {
951 emergency_thaw_bdev(sb);
952 thaw_super_locked(sb);
954 up_write(&sb->s_umount);
958 static void do_thaw_all(struct work_struct *work)
960 __iterate_supers(do_thaw_all_callback);
962 printk(KERN_WARNING "Emergency Thaw complete\n");
966 * emergency_thaw_all -- forcibly thaw every frozen filesystem
968 * Used for emergency unfreeze of all filesystems via SysRq
970 void emergency_thaw_all(void)
972 struct work_struct *work;
974 work = kmalloc(sizeof(*work), GFP_ATOMIC);
976 INIT_WORK(work, do_thaw_all);
982 * Unnamed block devices are dummy devices used by virtual
983 * filesystems which don't use real block-devices. -- jrs
986 static DEFINE_IDA(unnamed_dev_ida);
987 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
988 /* Many userspace utilities consider an FSID of 0 invalid.
989 * Always return at least 1 from get_anon_bdev.
991 static int unnamed_dev_start = 1;
993 int get_anon_bdev(dev_t *p)
999 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
1001 spin_lock(&unnamed_dev_lock);
1002 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
1004 unnamed_dev_start = dev + 1;
1005 spin_unlock(&unnamed_dev_lock);
1006 if (error == -EAGAIN)
1007 /* We raced and lost with another CPU. */
1012 if (dev >= (1 << MINORBITS)) {
1013 spin_lock(&unnamed_dev_lock);
1014 ida_remove(&unnamed_dev_ida, dev);
1015 if (unnamed_dev_start > dev)
1016 unnamed_dev_start = dev;
1017 spin_unlock(&unnamed_dev_lock);
1020 *p = MKDEV(0, dev & MINORMASK);
1023 EXPORT_SYMBOL(get_anon_bdev);
1025 void free_anon_bdev(dev_t dev)
1027 int slot = MINOR(dev);
1028 spin_lock(&unnamed_dev_lock);
1029 ida_remove(&unnamed_dev_ida, slot);
1030 if (slot < unnamed_dev_start)
1031 unnamed_dev_start = slot;
1032 spin_unlock(&unnamed_dev_lock);
1034 EXPORT_SYMBOL(free_anon_bdev);
1036 int set_anon_super(struct super_block *s, void *data)
1038 return get_anon_bdev(&s->s_dev);
1041 EXPORT_SYMBOL(set_anon_super);
1043 void kill_anon_super(struct super_block *sb)
1045 dev_t dev = sb->s_dev;
1046 generic_shutdown_super(sb);
1047 free_anon_bdev(dev);
1050 EXPORT_SYMBOL(kill_anon_super);
1052 void kill_litter_super(struct super_block *sb)
1055 d_genocide(sb->s_root);
1056 kill_anon_super(sb);
1059 EXPORT_SYMBOL(kill_litter_super);
1061 static int ns_test_super(struct super_block *sb, void *data)
1063 return sb->s_fs_info == data;
1066 static int ns_set_super(struct super_block *sb, void *data)
1068 sb->s_fs_info = data;
1069 return set_anon_super(sb, NULL);
1072 struct dentry *mount_ns(struct file_system_type *fs_type,
1073 int flags, void *data, void *ns, struct user_namespace *user_ns,
1074 int (*fill_super)(struct super_block *, void *, int))
1076 struct super_block *sb;
1078 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN
1079 * over the namespace.
1081 if (!(flags & SB_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN))
1082 return ERR_PTR(-EPERM);
1084 sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags,
1087 return ERR_CAST(sb);
1091 err = fill_super(sb, data, flags & SB_SILENT ? 1 : 0);
1093 deactivate_locked_super(sb);
1094 return ERR_PTR(err);
1097 sb->s_flags |= SB_ACTIVE;
1100 return dget(sb->s_root);
1103 EXPORT_SYMBOL(mount_ns);
1106 static int set_bdev_super(struct super_block *s, void *data)
1109 s->s_dev = s->s_bdev->bd_dev;
1110 s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
1115 static int test_bdev_super(struct super_block *s, void *data)
1117 return (void *)s->s_bdev == data;
1120 struct dentry *mount_bdev(struct file_system_type *fs_type,
1121 int flags, const char *dev_name, void *data,
1122 int (*fill_super)(struct super_block *, void *, int))
1124 struct block_device *bdev;
1125 struct super_block *s;
1126 fmode_t mode = FMODE_READ | FMODE_EXCL;
1129 if (!(flags & SB_RDONLY))
1130 mode |= FMODE_WRITE;
1132 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1134 return ERR_CAST(bdev);
1137 * once the super is inserted into the list by sget, s_umount
1138 * will protect the lockfs code from trying to start a snapshot
1139 * while we are mounting
1141 mutex_lock(&bdev->bd_fsfreeze_mutex);
1142 if (bdev->bd_fsfreeze_count > 0) {
1143 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1147 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
1149 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1154 if ((flags ^ s->s_flags) & SB_RDONLY) {
1155 deactivate_locked_super(s);
1161 * s_umount nests inside bd_mutex during
1162 * __invalidate_device(). blkdev_put() acquires
1163 * bd_mutex and can't be called under s_umount. Drop
1164 * s_umount temporarily. This is safe as we're
1165 * holding an active reference.
1167 up_write(&s->s_umount);
1168 blkdev_put(bdev, mode);
1169 down_write(&s->s_umount);
1172 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1173 sb_set_blocksize(s, block_size(bdev));
1174 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1176 deactivate_locked_super(s);
1180 s->s_flags |= SB_ACTIVE;
1184 return dget(s->s_root);
1189 blkdev_put(bdev, mode);
1191 return ERR_PTR(error);
1193 EXPORT_SYMBOL(mount_bdev);
1195 void kill_block_super(struct super_block *sb)
1197 struct block_device *bdev = sb->s_bdev;
1198 fmode_t mode = sb->s_mode;
1200 bdev->bd_super = NULL;
1201 generic_shutdown_super(sb);
1202 sync_blockdev(bdev);
1203 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1204 blkdev_put(bdev, mode | FMODE_EXCL);
1207 EXPORT_SYMBOL(kill_block_super);
1210 struct dentry *mount_nodev(struct file_system_type *fs_type,
1211 int flags, void *data,
1212 int (*fill_super)(struct super_block *, void *, int))
1215 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1220 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1222 deactivate_locked_super(s);
1223 return ERR_PTR(error);
1225 s->s_flags |= SB_ACTIVE;
1226 return dget(s->s_root);
1228 EXPORT_SYMBOL(mount_nodev);
1230 static int compare_single(struct super_block *s, void *p)
1235 struct dentry *mount_single(struct file_system_type *fs_type,
1236 int flags, void *data,
1237 int (*fill_super)(struct super_block *, void *, int))
1239 struct super_block *s;
1242 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1246 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1248 deactivate_locked_super(s);
1249 return ERR_PTR(error);
1251 s->s_flags |= SB_ACTIVE;
1253 do_remount_sb(s, flags, data, 0);
1255 return dget(s->s_root);
1257 EXPORT_SYMBOL(mount_single);
1260 mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1262 struct dentry *root;
1263 struct super_block *sb;
1264 char *secdata = NULL;
1265 int error = -ENOMEM;
1267 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1268 secdata = alloc_secdata();
1272 error = security_sb_copy_data(data, secdata);
1274 goto out_free_secdata;
1277 root = type->mount(type, flags, name, data);
1279 error = PTR_ERR(root);
1280 goto out_free_secdata;
1284 WARN_ON(!sb->s_bdi);
1287 * Write barrier is for super_cache_count(). We place it before setting
1288 * SB_BORN as the data dependency between the two functions is the
1289 * superblock structure contents that we just set up, not the SB_BORN
1293 sb->s_flags |= SB_BORN;
1295 error = security_sb_kern_mount(sb, flags, secdata);
1300 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1301 * but s_maxbytes was an unsigned long long for many releases. Throw
1302 * this warning for a little while to try and catch filesystems that
1303 * violate this rule.
1305 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1306 "negative value (%lld)\n", type->name, sb->s_maxbytes);
1308 up_write(&sb->s_umount);
1309 free_secdata(secdata);
1313 deactivate_locked_super(sb);
1315 free_secdata(secdata);
1317 return ERR_PTR(error);
1321 * Setup private BDI for given superblock. It gets automatically cleaned up
1322 * in generic_shutdown_super().
1324 int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1326 struct backing_dev_info *bdi;
1330 bdi = bdi_alloc(GFP_KERNEL);
1334 bdi->name = sb->s_type->name;
1336 va_start(args, fmt);
1337 err = bdi_register_va(bdi, fmt, args);
1343 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1348 EXPORT_SYMBOL(super_setup_bdi_name);
1351 * Setup private BDI for given superblock. I gets automatically cleaned up
1352 * in generic_shutdown_super().
1354 int super_setup_bdi(struct super_block *sb)
1356 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1358 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1359 atomic_long_inc_return(&bdi_seq));
1361 EXPORT_SYMBOL(super_setup_bdi);
1364 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1367 void __sb_end_write(struct super_block *sb, int level)
1369 percpu_up_read(sb->s_writers.rw_sem + level-1);
1371 EXPORT_SYMBOL(__sb_end_write);
1374 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1377 int __sb_start_write(struct super_block *sb, int level, bool wait)
1379 bool force_trylock = false;
1382 #ifdef CONFIG_LOCKDEP
1384 * We want lockdep to tell us about possible deadlocks with freezing
1385 * but it's it bit tricky to properly instrument it. Getting a freeze
1386 * protection works as getting a read lock but there are subtle
1387 * problems. XFS for example gets freeze protection on internal level
1388 * twice in some cases, which is OK only because we already hold a
1389 * freeze protection also on higher level. Due to these cases we have
1390 * to use wait == F (trylock mode) which must not fail.
1395 for (i = 0; i < level - 1; i++)
1396 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) {
1397 force_trylock = true;
1402 if (wait && !force_trylock)
1403 percpu_down_read(sb->s_writers.rw_sem + level-1);
1405 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1);
1407 WARN_ON(force_trylock && !ret);
1410 EXPORT_SYMBOL(__sb_start_write);
1413 * sb_wait_write - wait until all writers to given file system finish
1414 * @sb: the super for which we wait
1415 * @level: type of writers we wait for (normal vs page fault)
1417 * This function waits until there are no writers of given type to given file
1420 static void sb_wait_write(struct super_block *sb, int level)
1422 percpu_down_write(sb->s_writers.rw_sem + level-1);
1426 * We are going to return to userspace and forget about these locks, the
1427 * ownership goes to the caller of thaw_super() which does unlock().
1429 static void lockdep_sb_freeze_release(struct super_block *sb)
1433 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1434 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1438 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1440 static void lockdep_sb_freeze_acquire(struct super_block *sb)
1444 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1445 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1448 static void sb_freeze_unlock(struct super_block *sb)
1452 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1453 percpu_up_write(sb->s_writers.rw_sem + level);
1457 * freeze_super - lock the filesystem and force it into a consistent state
1458 * @sb: the super to lock
1460 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1461 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1464 * During this function, sb->s_writers.frozen goes through these values:
1466 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1468 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1469 * writes should be blocked, though page faults are still allowed. We wait for
1470 * all writes to complete and then proceed to the next stage.
1472 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1473 * but internal fs threads can still modify the filesystem (although they
1474 * should not dirty new pages or inodes), writeback can run etc. After waiting
1475 * for all running page faults we sync the filesystem which will clean all
1476 * dirty pages and inodes (no new dirty pages or inodes can be created when
1479 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1480 * modification are blocked (e.g. XFS preallocation truncation on inode
1481 * reclaim). This is usually implemented by blocking new transactions for
1482 * filesystems that have them and need this additional guard. After all
1483 * internal writers are finished we call ->freeze_fs() to finish filesystem
1484 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1485 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1487 * sb->s_writers.frozen is protected by sb->s_umount.
1489 int freeze_super(struct super_block *sb)
1493 atomic_inc(&sb->s_active);
1494 down_write(&sb->s_umount);
1495 if (sb->s_writers.frozen != SB_UNFROZEN) {
1496 deactivate_locked_super(sb);
1500 if (!(sb->s_flags & SB_BORN)) {
1501 up_write(&sb->s_umount);
1502 return 0; /* sic - it's "nothing to do" */
1505 if (sb_rdonly(sb)) {
1506 /* Nothing to do really... */
1507 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1508 up_write(&sb->s_umount);
1512 sb->s_writers.frozen = SB_FREEZE_WRITE;
1513 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1514 up_write(&sb->s_umount);
1515 sb_wait_write(sb, SB_FREEZE_WRITE);
1516 down_write(&sb->s_umount);
1518 /* Now we go and block page faults... */
1519 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1520 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1522 /* All writers are done so after syncing there won't be dirty data */
1523 sync_filesystem(sb);
1525 /* Now wait for internal filesystem counter */
1526 sb->s_writers.frozen = SB_FREEZE_FS;
1527 sb_wait_write(sb, SB_FREEZE_FS);
1529 if (sb->s_op->freeze_fs) {
1530 ret = sb->s_op->freeze_fs(sb);
1533 "VFS:Filesystem freeze failed\n");
1534 sb->s_writers.frozen = SB_UNFROZEN;
1535 sb_freeze_unlock(sb);
1536 wake_up(&sb->s_writers.wait_unfrozen);
1537 deactivate_locked_super(sb);
1542 * For debugging purposes so that fs can warn if it sees write activity
1543 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1545 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1546 lockdep_sb_freeze_release(sb);
1547 up_write(&sb->s_umount);
1550 EXPORT_SYMBOL(freeze_super);
1553 * thaw_super -- unlock filesystem
1554 * @sb: the super to thaw
1556 * Unlocks the filesystem and marks it writeable again after freeze_super().
1558 static int thaw_super_locked(struct super_block *sb)
1562 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1563 up_write(&sb->s_umount);
1567 if (sb_rdonly(sb)) {
1568 sb->s_writers.frozen = SB_UNFROZEN;
1572 lockdep_sb_freeze_acquire(sb);
1574 if (sb->s_op->unfreeze_fs) {
1575 error = sb->s_op->unfreeze_fs(sb);
1578 "VFS:Filesystem thaw failed\n");
1579 lockdep_sb_freeze_release(sb);
1580 up_write(&sb->s_umount);
1585 sb->s_writers.frozen = SB_UNFROZEN;
1586 sb_freeze_unlock(sb);
1588 wake_up(&sb->s_writers.wait_unfrozen);
1589 deactivate_locked_super(sb);
1593 int thaw_super(struct super_block *sb)
1595 down_write(&sb->s_umount);
1596 return thaw_super_locked(sb);
1598 EXPORT_SYMBOL(thaw_super);