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/fscrypt.h>
36 #include <linux/fsnotify.h>
37 #include <linux/lockdep.h>
38 #include <linux/user_namespace.h>
39 #include <linux/fs_context.h>
40 #include <uapi/linux/mount.h>
43 static int thaw_super_locked(struct super_block *sb);
45 static LIST_HEAD(super_blocks);
46 static DEFINE_SPINLOCK(sb_lock);
48 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
55 * One thing we have to be careful of with a per-sb shrinker is that we don't
56 * drop the last active reference to the superblock from within the shrinker.
57 * If that happens we could trigger unregistering the shrinker from within the
58 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
59 * take a passive reference to the superblock to avoid this from occurring.
61 static unsigned long super_cache_scan(struct shrinker *shrink,
62 struct shrink_control *sc)
64 struct super_block *sb;
71 sb = container_of(shrink, struct super_block, s_shrink);
74 * Deadlock avoidance. We may hold various FS locks, and we don't want
75 * to recurse into the FS that called us in clear_inode() and friends..
77 if (!(sc->gfp_mask & __GFP_FS))
80 if (!trylock_super(sb))
83 if (sb->s_op->nr_cached_objects)
84 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
86 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
87 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
88 total_objects = dentries + inodes + fs_objects + 1;
92 /* proportion the scan between the caches */
93 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
94 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
95 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
98 * prune the dcache first as the icache is pinned by it, then
99 * prune the icache, followed by the filesystem specific caches
101 * Ensure that we always scan at least one object - memcg kmem
102 * accounting uses this to fully empty the caches.
104 sc->nr_to_scan = dentries + 1;
105 freed = prune_dcache_sb(sb, sc);
106 sc->nr_to_scan = inodes + 1;
107 freed += prune_icache_sb(sb, sc);
110 sc->nr_to_scan = fs_objects + 1;
111 freed += sb->s_op->free_cached_objects(sb, sc);
114 up_read(&sb->s_umount);
118 static unsigned long super_cache_count(struct shrinker *shrink,
119 struct shrink_control *sc)
121 struct super_block *sb;
122 long total_objects = 0;
124 sb = container_of(shrink, struct super_block, s_shrink);
127 * We don't call trylock_super() here as it is a scalability bottleneck,
128 * so we're exposed to partial setup state. The shrinker rwsem does not
129 * protect filesystem operations backing list_lru_shrink_count() or
130 * s_op->nr_cached_objects(). Counts can change between
131 * super_cache_count and super_cache_scan, so we really don't need locks
134 * However, if we are currently mounting the superblock, the underlying
135 * filesystem might be in a state of partial construction and hence it
136 * is dangerous to access it. trylock_super() uses a SB_BORN check to
137 * avoid this situation, so do the same here. The memory barrier is
138 * matched with the one in mount_fs() as we don't hold locks here.
140 if (!(sb->s_flags & SB_BORN))
144 if (sb->s_op && sb->s_op->nr_cached_objects)
145 total_objects = sb->s_op->nr_cached_objects(sb, sc);
147 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
148 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
153 total_objects = vfs_pressure_ratio(total_objects);
154 return total_objects;
157 static void destroy_super_work(struct work_struct *work)
159 struct super_block *s = container_of(work, struct super_block,
163 for (i = 0; i < SB_FREEZE_LEVELS; i++)
164 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
168 static void destroy_super_rcu(struct rcu_head *head)
170 struct super_block *s = container_of(head, struct super_block, rcu);
171 INIT_WORK(&s->destroy_work, destroy_super_work);
172 schedule_work(&s->destroy_work);
175 /* Free a superblock that has never been seen by anyone */
176 static void destroy_unused_super(struct super_block *s)
180 up_write(&s->s_umount);
181 list_lru_destroy(&s->s_dentry_lru);
182 list_lru_destroy(&s->s_inode_lru);
184 put_user_ns(s->s_user_ns);
186 free_prealloced_shrinker(&s->s_shrink);
187 /* no delays needed */
188 destroy_super_work(&s->destroy_work);
192 * alloc_super - create new superblock
193 * @type: filesystem type superblock should belong to
194 * @flags: the mount flags
195 * @user_ns: User namespace for the super_block
197 * Allocates and initializes a new &struct super_block. alloc_super()
198 * returns a pointer new superblock or %NULL if allocation had failed.
200 static struct super_block *alloc_super(struct file_system_type *type, int flags,
201 struct user_namespace *user_ns)
203 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
204 static const struct super_operations default_op;
210 INIT_LIST_HEAD(&s->s_mounts);
211 s->s_user_ns = get_user_ns(user_ns);
212 init_rwsem(&s->s_umount);
213 lockdep_set_class(&s->s_umount, &type->s_umount_key);
215 * sget() can have s_umount recursion.
217 * When it cannot find a suitable sb, it allocates a new
218 * one (this one), and tries again to find a suitable old
221 * In case that succeeds, it will acquire the s_umount
222 * lock of the old one. Since these are clearly distrinct
223 * locks, and this object isn't exposed yet, there's no
226 * Annotate this by putting this lock in a different
229 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
231 if (security_sb_alloc(s))
234 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
235 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
237 &type->s_writers_key[i]))
240 init_waitqueue_head(&s->s_writers.wait_unfrozen);
241 s->s_bdi = &noop_backing_dev_info;
243 if (s->s_user_ns != &init_user_ns)
244 s->s_iflags |= SB_I_NODEV;
245 INIT_HLIST_NODE(&s->s_instances);
246 INIT_HLIST_BL_HEAD(&s->s_roots);
247 mutex_init(&s->s_sync_lock);
248 INIT_LIST_HEAD(&s->s_inodes);
249 spin_lock_init(&s->s_inode_list_lock);
250 INIT_LIST_HEAD(&s->s_inodes_wb);
251 spin_lock_init(&s->s_inode_wblist_lock);
254 atomic_set(&s->s_active, 1);
255 mutex_init(&s->s_vfs_rename_mutex);
256 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
257 init_rwsem(&s->s_dquot.dqio_sem);
258 s->s_maxbytes = MAX_NON_LFS;
259 s->s_op = &default_op;
260 s->s_time_gran = 1000000000;
261 s->s_time_min = TIME64_MIN;
262 s->s_time_max = TIME64_MAX;
263 s->cleancache_poolid = CLEANCACHE_NO_POOL;
265 s->s_shrink.seeks = DEFAULT_SEEKS;
266 s->s_shrink.scan_objects = super_cache_scan;
267 s->s_shrink.count_objects = super_cache_count;
268 s->s_shrink.batch = 1024;
269 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
270 if (prealloc_shrinker(&s->s_shrink))
272 if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink))
274 if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink))
279 destroy_unused_super(s);
283 /* Superblock refcounting */
286 * Drop a superblock's refcount. The caller must hold sb_lock.
288 static void __put_super(struct super_block *s)
291 list_del_init(&s->s_list);
292 WARN_ON(s->s_dentry_lru.node);
293 WARN_ON(s->s_inode_lru.node);
294 WARN_ON(!list_empty(&s->s_mounts));
297 put_user_ns(s->s_user_ns);
299 call_rcu(&s->rcu, destroy_super_rcu);
304 * put_super - drop a temporary reference to superblock
305 * @sb: superblock in question
307 * Drops a temporary reference, frees superblock if there's no
310 void put_super(struct super_block *sb)
314 spin_unlock(&sb_lock);
319 * deactivate_locked_super - drop an active reference to superblock
320 * @s: superblock to deactivate
322 * Drops an active reference to superblock, converting it into a temporary
323 * one if there is no other active references left. In that case we
324 * tell fs driver to shut it down and drop the temporary reference we
327 * Caller holds exclusive lock on superblock; that lock is released.
329 void deactivate_locked_super(struct super_block *s)
331 struct file_system_type *fs = s->s_type;
332 if (atomic_dec_and_test(&s->s_active)) {
333 cleancache_invalidate_fs(s);
334 unregister_shrinker(&s->s_shrink);
338 * Since list_lru_destroy() may sleep, we cannot call it from
339 * put_super(), where we hold the sb_lock. Therefore we destroy
340 * the lru lists right now.
342 list_lru_destroy(&s->s_dentry_lru);
343 list_lru_destroy(&s->s_inode_lru);
348 up_write(&s->s_umount);
352 EXPORT_SYMBOL(deactivate_locked_super);
355 * deactivate_super - drop an active reference to superblock
356 * @s: superblock to deactivate
358 * Variant of deactivate_locked_super(), except that superblock is *not*
359 * locked by caller. If we are going to drop the final active reference,
360 * lock will be acquired prior to that.
362 void deactivate_super(struct super_block *s)
364 if (!atomic_add_unless(&s->s_active, -1, 1)) {
365 down_write(&s->s_umount);
366 deactivate_locked_super(s);
370 EXPORT_SYMBOL(deactivate_super);
373 * grab_super - acquire an active reference
374 * @s: reference we are trying to make active
376 * Tries to acquire an active reference. grab_super() is used when we
377 * had just found a superblock in super_blocks or fs_type->fs_supers
378 * and want to turn it into a full-blown active reference. grab_super()
379 * is called with sb_lock held and drops it. Returns 1 in case of
380 * success, 0 if we had failed (superblock contents was already dead or
381 * dying when grab_super() had been called). Note that this is only
382 * called for superblocks not in rundown mode (== ones still on ->fs_supers
383 * of their type), so increment of ->s_count is OK here.
385 static int grab_super(struct super_block *s) __releases(sb_lock)
388 spin_unlock(&sb_lock);
389 down_write(&s->s_umount);
390 if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
394 up_write(&s->s_umount);
400 * trylock_super - try to grab ->s_umount shared
401 * @sb: reference we are trying to grab
403 * Try to prevent fs shutdown. This is used in places where we
404 * cannot take an active reference but we need to ensure that the
405 * filesystem is not shut down while we are working on it. It returns
406 * false if we cannot acquire s_umount or if we lose the race and
407 * filesystem already got into shutdown, and returns true with the s_umount
408 * lock held in read mode in case of success. On successful return,
409 * the caller must drop the s_umount lock when done.
411 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
412 * The reason why it's safe is that we are OK with doing trylock instead
413 * of down_read(). There's a couple of places that are OK with that, but
414 * it's very much not a general-purpose interface.
416 bool trylock_super(struct super_block *sb)
418 if (down_read_trylock(&sb->s_umount)) {
419 if (!hlist_unhashed(&sb->s_instances) &&
420 sb->s_root && (sb->s_flags & SB_BORN))
422 up_read(&sb->s_umount);
429 * generic_shutdown_super - common helper for ->kill_sb()
430 * @sb: superblock to kill
432 * generic_shutdown_super() does all fs-independent work on superblock
433 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
434 * that need destruction out of superblock, call generic_shutdown_super()
435 * and release aforementioned objects. Note: dentries and inodes _are_
436 * taken care of and do not need specific handling.
438 * Upon calling this function, the filesystem may no longer alter or
439 * rearrange the set of dentries belonging to this super_block, nor may it
440 * change the attachments of dentries to inodes.
442 void generic_shutdown_super(struct super_block *sb)
444 const struct super_operations *sop = sb->s_op;
447 shrink_dcache_for_umount(sb);
449 sb->s_flags &= ~SB_ACTIVE;
451 cgroup_writeback_umount();
453 /* evict all inodes with zero refcount */
455 /* only nonzero refcount inodes can have marks */
456 fsnotify_sb_delete(sb);
457 security_sb_delete(sb);
459 if (sb->s_dio_done_wq) {
460 destroy_workqueue(sb->s_dio_done_wq);
461 sb->s_dio_done_wq = NULL;
467 if (!list_empty(&sb->s_inodes)) {
468 printk("VFS: Busy inodes after unmount of %s. "
469 "Self-destruct in 5 seconds. Have a nice day...\n",
474 /* should be initialized for __put_super_and_need_restart() */
475 hlist_del_init(&sb->s_instances);
476 spin_unlock(&sb_lock);
477 up_write(&sb->s_umount);
478 if (sb->s_bdi != &noop_backing_dev_info) {
480 sb->s_bdi = &noop_backing_dev_info;
484 EXPORT_SYMBOL(generic_shutdown_super);
486 bool mount_capable(struct fs_context *fc)
488 if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT))
489 return capable(CAP_SYS_ADMIN);
491 return ns_capable(fc->user_ns, CAP_SYS_ADMIN);
495 * sget_fc - Find or create a superblock
496 * @fc: Filesystem context.
497 * @test: Comparison callback
498 * @set: Setup callback
500 * Find or create a superblock using the parameters stored in the filesystem
501 * context and the two callback functions.
503 * If an extant superblock is matched, then that will be returned with an
504 * elevated reference count that the caller must transfer or discard.
506 * If no match is made, a new superblock will be allocated and basic
507 * initialisation will be performed (s_type, s_fs_info and s_id will be set and
508 * the set() callback will be invoked), the superblock will be published and it
509 * will be returned in a partially constructed state with SB_BORN and SB_ACTIVE
512 struct super_block *sget_fc(struct fs_context *fc,
513 int (*test)(struct super_block *, struct fs_context *),
514 int (*set)(struct super_block *, struct fs_context *))
516 struct super_block *s = NULL;
517 struct super_block *old;
518 struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns;
524 hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) {
526 goto share_extant_sb;
530 spin_unlock(&sb_lock);
531 s = alloc_super(fc->fs_type, fc->sb_flags, user_ns);
533 return ERR_PTR(-ENOMEM);
537 s->s_fs_info = fc->s_fs_info;
541 spin_unlock(&sb_lock);
542 destroy_unused_super(s);
545 fc->s_fs_info = NULL;
546 s->s_type = fc->fs_type;
547 s->s_iflags |= fc->s_iflags;
548 strlcpy(s->s_id, s->s_type->name, sizeof(s->s_id));
549 list_add_tail(&s->s_list, &super_blocks);
550 hlist_add_head(&s->s_instances, &s->s_type->fs_supers);
551 spin_unlock(&sb_lock);
552 get_filesystem(s->s_type);
553 register_shrinker_prepared(&s->s_shrink);
557 if (user_ns != old->s_user_ns) {
558 spin_unlock(&sb_lock);
559 destroy_unused_super(s);
560 return ERR_PTR(-EBUSY);
562 if (!grab_super(old))
564 destroy_unused_super(s);
567 EXPORT_SYMBOL(sget_fc);
570 * sget - find or create a superblock
571 * @type: filesystem type superblock should belong to
572 * @test: comparison callback
573 * @set: setup callback
574 * @flags: mount flags
575 * @data: argument to each of them
577 struct super_block *sget(struct file_system_type *type,
578 int (*test)(struct super_block *,void *),
579 int (*set)(struct super_block *,void *),
583 struct user_namespace *user_ns = current_user_ns();
584 struct super_block *s = NULL;
585 struct super_block *old;
588 /* We don't yet pass the user namespace of the parent
589 * mount through to here so always use &init_user_ns
590 * until that changes.
592 if (flags & SB_SUBMOUNT)
593 user_ns = &init_user_ns;
598 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
599 if (!test(old, data))
601 if (user_ns != old->s_user_ns) {
602 spin_unlock(&sb_lock);
603 destroy_unused_super(s);
604 return ERR_PTR(-EBUSY);
606 if (!grab_super(old))
608 destroy_unused_super(s);
613 spin_unlock(&sb_lock);
614 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
616 return ERR_PTR(-ENOMEM);
622 spin_unlock(&sb_lock);
623 destroy_unused_super(s);
627 strlcpy(s->s_id, type->name, sizeof(s->s_id));
628 list_add_tail(&s->s_list, &super_blocks);
629 hlist_add_head(&s->s_instances, &type->fs_supers);
630 spin_unlock(&sb_lock);
631 get_filesystem(type);
632 register_shrinker_prepared(&s->s_shrink);
637 void drop_super(struct super_block *sb)
639 up_read(&sb->s_umount);
643 EXPORT_SYMBOL(drop_super);
645 void drop_super_exclusive(struct super_block *sb)
647 up_write(&sb->s_umount);
650 EXPORT_SYMBOL(drop_super_exclusive);
652 static void __iterate_supers(void (*f)(struct super_block *))
654 struct super_block *sb, *p = NULL;
657 list_for_each_entry(sb, &super_blocks, s_list) {
658 if (hlist_unhashed(&sb->s_instances))
661 spin_unlock(&sb_lock);
672 spin_unlock(&sb_lock);
675 * iterate_supers - call function for all active superblocks
676 * @f: function to call
677 * @arg: argument to pass to it
679 * Scans the superblock list and calls given function, passing it
680 * locked superblock and given argument.
682 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
684 struct super_block *sb, *p = NULL;
687 list_for_each_entry(sb, &super_blocks, s_list) {
688 if (hlist_unhashed(&sb->s_instances))
691 spin_unlock(&sb_lock);
693 down_read(&sb->s_umount);
694 if (sb->s_root && (sb->s_flags & SB_BORN))
696 up_read(&sb->s_umount);
705 spin_unlock(&sb_lock);
709 * iterate_supers_type - call function for superblocks of given type
711 * @f: function to call
712 * @arg: argument to pass to it
714 * Scans the superblock list and calls given function, passing it
715 * locked superblock and given argument.
717 void iterate_supers_type(struct file_system_type *type,
718 void (*f)(struct super_block *, void *), void *arg)
720 struct super_block *sb, *p = NULL;
723 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
725 spin_unlock(&sb_lock);
727 down_read(&sb->s_umount);
728 if (sb->s_root && (sb->s_flags & SB_BORN))
730 up_read(&sb->s_umount);
739 spin_unlock(&sb_lock);
742 EXPORT_SYMBOL(iterate_supers_type);
745 * get_super - get the superblock of a device
746 * @bdev: device to get the superblock for
748 * Scans the superblock list and finds the superblock of the file system
749 * mounted on the device given. %NULL is returned if no match is found.
751 struct super_block *get_super(struct block_device *bdev)
753 struct super_block *sb;
760 list_for_each_entry(sb, &super_blocks, s_list) {
761 if (hlist_unhashed(&sb->s_instances))
763 if (sb->s_bdev == bdev) {
765 spin_unlock(&sb_lock);
766 down_read(&sb->s_umount);
768 if (sb->s_root && (sb->s_flags & SB_BORN))
770 up_read(&sb->s_umount);
771 /* nope, got unmounted */
777 spin_unlock(&sb_lock);
782 * get_active_super - get an active reference to the superblock of a device
783 * @bdev: device to get the superblock for
785 * Scans the superblock list and finds the superblock of the file system
786 * mounted on the device given. Returns the superblock with an active
787 * reference or %NULL if none was found.
789 struct super_block *get_active_super(struct block_device *bdev)
791 struct super_block *sb;
798 list_for_each_entry(sb, &super_blocks, s_list) {
799 if (hlist_unhashed(&sb->s_instances))
801 if (sb->s_bdev == bdev) {
804 up_write(&sb->s_umount);
808 spin_unlock(&sb_lock);
812 struct super_block *user_get_super(dev_t dev, bool excl)
814 struct super_block *sb;
818 list_for_each_entry(sb, &super_blocks, s_list) {
819 if (hlist_unhashed(&sb->s_instances))
821 if (sb->s_dev == dev) {
823 spin_unlock(&sb_lock);
825 down_write(&sb->s_umount);
827 down_read(&sb->s_umount);
829 if (sb->s_root && (sb->s_flags & SB_BORN))
832 up_write(&sb->s_umount);
834 up_read(&sb->s_umount);
835 /* nope, got unmounted */
841 spin_unlock(&sb_lock);
846 * reconfigure_super - asks filesystem to change superblock parameters
847 * @fc: The superblock and configuration
849 * Alters the configuration parameters of a live superblock.
851 int reconfigure_super(struct fs_context *fc)
853 struct super_block *sb = fc->root->d_sb;
855 bool remount_ro = false;
856 bool force = fc->sb_flags & SB_FORCE;
858 if (fc->sb_flags_mask & ~MS_RMT_MASK)
860 if (sb->s_writers.frozen != SB_UNFROZEN)
863 retval = security_sb_remount(sb, fc->security);
867 if (fc->sb_flags_mask & SB_RDONLY) {
869 if (!(fc->sb_flags & SB_RDONLY) && sb->s_bdev &&
870 bdev_read_only(sb->s_bdev))
874 remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb);
878 if (!hlist_empty(&sb->s_pins)) {
879 up_write(&sb->s_umount);
880 group_pin_kill(&sb->s_pins);
881 down_write(&sb->s_umount);
884 if (sb->s_writers.frozen != SB_UNFROZEN)
886 remount_ro = !sb_rdonly(sb);
889 shrink_dcache_sb(sb);
891 /* If we are reconfiguring to RDONLY and current sb is read/write,
892 * make sure there are no files open for writing.
896 sb->s_readonly_remount = 1;
899 retval = sb_prepare_remount_readonly(sb);
905 if (fc->ops->reconfigure) {
906 retval = fc->ops->reconfigure(fc);
909 goto cancel_readonly;
910 /* If forced remount, go ahead despite any errors */
911 WARN(1, "forced remount of a %s fs returned %i\n",
912 sb->s_type->name, retval);
916 WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) |
917 (fc->sb_flags & fc->sb_flags_mask)));
918 /* Needs to be ordered wrt mnt_is_readonly() */
920 sb->s_readonly_remount = 0;
923 * Some filesystems modify their metadata via some other path than the
924 * bdev buffer cache (eg. use a private mapping, or directories in
925 * pagecache, etc). Also file data modifications go via their own
926 * mappings. So If we try to mount readonly then copy the filesystem
927 * from bdev, we could get stale data, so invalidate it to give a best
928 * effort at coherency.
930 if (remount_ro && sb->s_bdev)
931 invalidate_bdev(sb->s_bdev);
935 sb->s_readonly_remount = 0;
939 static void do_emergency_remount_callback(struct super_block *sb)
941 down_write(&sb->s_umount);
942 if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
944 struct fs_context *fc;
946 fc = fs_context_for_reconfigure(sb->s_root,
947 SB_RDONLY | SB_FORCE, SB_RDONLY);
949 if (parse_monolithic_mount_data(fc, NULL) == 0)
950 (void)reconfigure_super(fc);
954 up_write(&sb->s_umount);
957 static void do_emergency_remount(struct work_struct *work)
959 __iterate_supers(do_emergency_remount_callback);
961 printk("Emergency Remount complete\n");
964 void emergency_remount(void)
966 struct work_struct *work;
968 work = kmalloc(sizeof(*work), GFP_ATOMIC);
970 INIT_WORK(work, do_emergency_remount);
975 static void do_thaw_all_callback(struct super_block *sb)
977 down_write(&sb->s_umount);
978 if (sb->s_root && sb->s_flags & SB_BORN) {
979 emergency_thaw_bdev(sb);
980 thaw_super_locked(sb);
982 up_write(&sb->s_umount);
986 static void do_thaw_all(struct work_struct *work)
988 __iterate_supers(do_thaw_all_callback);
990 printk(KERN_WARNING "Emergency Thaw complete\n");
994 * emergency_thaw_all -- forcibly thaw every frozen filesystem
996 * Used for emergency unfreeze of all filesystems via SysRq
998 void emergency_thaw_all(void)
1000 struct work_struct *work;
1002 work = kmalloc(sizeof(*work), GFP_ATOMIC);
1004 INIT_WORK(work, do_thaw_all);
1005 schedule_work(work);
1009 static DEFINE_IDA(unnamed_dev_ida);
1012 * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1013 * @p: Pointer to a dev_t.
1015 * Filesystems which don't use real block devices can call this function
1016 * to allocate a virtual block device.
1018 * Context: Any context. Frequently called while holding sb_lock.
1019 * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1020 * or -ENOMEM if memory allocation failed.
1022 int get_anon_bdev(dev_t *p)
1027 * Many userspace utilities consider an FSID of 0 invalid.
1028 * Always return at least 1 from get_anon_bdev.
1030 dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1040 EXPORT_SYMBOL(get_anon_bdev);
1042 void free_anon_bdev(dev_t dev)
1044 ida_free(&unnamed_dev_ida, MINOR(dev));
1046 EXPORT_SYMBOL(free_anon_bdev);
1048 int set_anon_super(struct super_block *s, void *data)
1050 return get_anon_bdev(&s->s_dev);
1052 EXPORT_SYMBOL(set_anon_super);
1054 void kill_anon_super(struct super_block *sb)
1056 dev_t dev = sb->s_dev;
1057 generic_shutdown_super(sb);
1058 free_anon_bdev(dev);
1060 EXPORT_SYMBOL(kill_anon_super);
1062 void kill_litter_super(struct super_block *sb)
1065 d_genocide(sb->s_root);
1066 kill_anon_super(sb);
1068 EXPORT_SYMBOL(kill_litter_super);
1070 int set_anon_super_fc(struct super_block *sb, struct fs_context *fc)
1072 return set_anon_super(sb, NULL);
1074 EXPORT_SYMBOL(set_anon_super_fc);
1076 static int test_keyed_super(struct super_block *sb, struct fs_context *fc)
1078 return sb->s_fs_info == fc->s_fs_info;
1081 static int test_single_super(struct super_block *s, struct fs_context *fc)
1087 * vfs_get_super - Get a superblock with a search key set in s_fs_info.
1088 * @fc: The filesystem context holding the parameters
1089 * @keying: How to distinguish superblocks
1090 * @fill_super: Helper to initialise a new superblock
1092 * Search for a superblock and create a new one if not found. The search
1093 * criterion is controlled by @keying. If the search fails, a new superblock
1094 * is created and @fill_super() is called to initialise it.
1096 * @keying can take one of a number of values:
1098 * (1) vfs_get_single_super - Only one superblock of this type may exist on the
1099 * system. This is typically used for special system filesystems.
1101 * (2) vfs_get_keyed_super - Multiple superblocks may exist, but they must have
1102 * distinct keys (where the key is in s_fs_info). Searching for the same
1103 * key again will turn up the superblock for that key.
1105 * (3) vfs_get_independent_super - Multiple superblocks may exist and are
1106 * unkeyed. Each call will get a new superblock.
1108 * A permissions check is made by sget_fc() unless we're getting a superblock
1109 * for a kernel-internal mount or a submount.
1111 int vfs_get_super(struct fs_context *fc,
1112 enum vfs_get_super_keying keying,
1113 int (*fill_super)(struct super_block *sb,
1114 struct fs_context *fc))
1116 int (*test)(struct super_block *, struct fs_context *);
1117 struct super_block *sb;
1121 case vfs_get_single_super:
1122 case vfs_get_single_reconf_super:
1123 test = test_single_super;
1125 case vfs_get_keyed_super:
1126 test = test_keyed_super;
1128 case vfs_get_independent_super:
1135 sb = sget_fc(fc, test, set_anon_super_fc);
1140 err = fill_super(sb, fc);
1144 sb->s_flags |= SB_ACTIVE;
1145 fc->root = dget(sb->s_root);
1147 fc->root = dget(sb->s_root);
1148 if (keying == vfs_get_single_reconf_super) {
1149 err = reconfigure_super(fc);
1161 deactivate_locked_super(sb);
1164 EXPORT_SYMBOL(vfs_get_super);
1166 int get_tree_nodev(struct fs_context *fc,
1167 int (*fill_super)(struct super_block *sb,
1168 struct fs_context *fc))
1170 return vfs_get_super(fc, vfs_get_independent_super, fill_super);
1172 EXPORT_SYMBOL(get_tree_nodev);
1174 int get_tree_single(struct fs_context *fc,
1175 int (*fill_super)(struct super_block *sb,
1176 struct fs_context *fc))
1178 return vfs_get_super(fc, vfs_get_single_super, fill_super);
1180 EXPORT_SYMBOL(get_tree_single);
1182 int get_tree_single_reconf(struct fs_context *fc,
1183 int (*fill_super)(struct super_block *sb,
1184 struct fs_context *fc))
1186 return vfs_get_super(fc, vfs_get_single_reconf_super, fill_super);
1188 EXPORT_SYMBOL(get_tree_single_reconf);
1190 int get_tree_keyed(struct fs_context *fc,
1191 int (*fill_super)(struct super_block *sb,
1192 struct fs_context *fc),
1195 fc->s_fs_info = key;
1196 return vfs_get_super(fc, vfs_get_keyed_super, fill_super);
1198 EXPORT_SYMBOL(get_tree_keyed);
1202 static int set_bdev_super(struct super_block *s, void *data)
1205 s->s_dev = s->s_bdev->bd_dev;
1206 s->s_bdi = bdi_get(s->s_bdev->bd_disk->bdi);
1208 if (blk_queue_stable_writes(s->s_bdev->bd_disk->queue))
1209 s->s_iflags |= SB_I_STABLE_WRITES;
1213 static int set_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1215 return set_bdev_super(s, fc->sget_key);
1218 static int test_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1220 return s->s_bdev == fc->sget_key;
1224 * get_tree_bdev - Get a superblock based on a single block device
1225 * @fc: The filesystem context holding the parameters
1226 * @fill_super: Helper to initialise a new superblock
1228 int get_tree_bdev(struct fs_context *fc,
1229 int (*fill_super)(struct super_block *,
1230 struct fs_context *))
1232 struct block_device *bdev;
1233 struct super_block *s;
1234 fmode_t mode = FMODE_READ | FMODE_EXCL;
1237 if (!(fc->sb_flags & SB_RDONLY))
1238 mode |= FMODE_WRITE;
1241 return invalf(fc, "No source specified");
1243 bdev = blkdev_get_by_path(fc->source, mode, fc->fs_type);
1245 errorf(fc, "%s: Can't open blockdev", fc->source);
1246 return PTR_ERR(bdev);
1249 /* Once the superblock is inserted into the list by sget_fc(), s_umount
1250 * will protect the lockfs code from trying to start a snapshot while
1253 mutex_lock(&bdev->bd_fsfreeze_mutex);
1254 if (bdev->bd_fsfreeze_count > 0) {
1255 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1256 warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
1257 blkdev_put(bdev, mode);
1261 fc->sb_flags |= SB_NOSEC;
1262 fc->sget_key = bdev;
1263 s = sget_fc(fc, test_bdev_super_fc, set_bdev_super_fc);
1264 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1266 blkdev_put(bdev, mode);
1271 /* Don't summarily change the RO/RW state. */
1272 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1273 warnf(fc, "%pg: Can't mount, would change RO state", bdev);
1274 deactivate_locked_super(s);
1275 blkdev_put(bdev, mode);
1280 * s_umount nests inside open_mutex during
1281 * __invalidate_device(). blkdev_put() acquires
1282 * open_mutex and can't be called under s_umount. Drop
1283 * s_umount temporarily. This is safe as we're
1284 * holding an active reference.
1286 up_write(&s->s_umount);
1287 blkdev_put(bdev, mode);
1288 down_write(&s->s_umount);
1291 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1292 sb_set_blocksize(s, block_size(bdev));
1293 error = fill_super(s, fc);
1295 deactivate_locked_super(s);
1299 s->s_flags |= SB_ACTIVE;
1304 fc->root = dget(s->s_root);
1307 EXPORT_SYMBOL(get_tree_bdev);
1309 static int test_bdev_super(struct super_block *s, void *data)
1311 return (void *)s->s_bdev == data;
1314 struct dentry *mount_bdev(struct file_system_type *fs_type,
1315 int flags, const char *dev_name, void *data,
1316 int (*fill_super)(struct super_block *, void *, int))
1318 struct block_device *bdev;
1319 struct super_block *s;
1320 fmode_t mode = FMODE_READ | FMODE_EXCL;
1323 if (!(flags & SB_RDONLY))
1324 mode |= FMODE_WRITE;
1326 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1328 return ERR_CAST(bdev);
1331 * once the super is inserted into the list by sget, s_umount
1332 * will protect the lockfs code from trying to start a snapshot
1333 * while we are mounting
1335 mutex_lock(&bdev->bd_fsfreeze_mutex);
1336 if (bdev->bd_fsfreeze_count > 0) {
1337 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1341 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
1343 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1348 if ((flags ^ s->s_flags) & SB_RDONLY) {
1349 deactivate_locked_super(s);
1355 * s_umount nests inside open_mutex during
1356 * __invalidate_device(). blkdev_put() acquires
1357 * open_mutex and can't be called under s_umount. Drop
1358 * s_umount temporarily. This is safe as we're
1359 * holding an active reference.
1361 up_write(&s->s_umount);
1362 blkdev_put(bdev, mode);
1363 down_write(&s->s_umount);
1366 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1367 sb_set_blocksize(s, block_size(bdev));
1368 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1370 deactivate_locked_super(s);
1374 s->s_flags |= SB_ACTIVE;
1378 return dget(s->s_root);
1383 blkdev_put(bdev, mode);
1385 return ERR_PTR(error);
1387 EXPORT_SYMBOL(mount_bdev);
1389 void kill_block_super(struct super_block *sb)
1391 struct block_device *bdev = sb->s_bdev;
1392 fmode_t mode = sb->s_mode;
1394 bdev->bd_super = NULL;
1395 generic_shutdown_super(sb);
1396 sync_blockdev(bdev);
1397 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1398 blkdev_put(bdev, mode | FMODE_EXCL);
1401 EXPORT_SYMBOL(kill_block_super);
1404 struct dentry *mount_nodev(struct file_system_type *fs_type,
1405 int flags, void *data,
1406 int (*fill_super)(struct super_block *, void *, int))
1409 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1414 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1416 deactivate_locked_super(s);
1417 return ERR_PTR(error);
1419 s->s_flags |= SB_ACTIVE;
1420 return dget(s->s_root);
1422 EXPORT_SYMBOL(mount_nodev);
1424 static int reconfigure_single(struct super_block *s,
1425 int flags, void *data)
1427 struct fs_context *fc;
1430 /* The caller really need to be passing fc down into mount_single(),
1431 * then a chunk of this can be removed. [Bollocks -- AV]
1432 * Better yet, reconfiguration shouldn't happen, but rather the second
1433 * mount should be rejected if the parameters are not compatible.
1435 fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK);
1439 ret = parse_monolithic_mount_data(fc, data);
1443 ret = reconfigure_super(fc);
1449 static int compare_single(struct super_block *s, void *p)
1454 struct dentry *mount_single(struct file_system_type *fs_type,
1455 int flags, void *data,
1456 int (*fill_super)(struct super_block *, void *, int))
1458 struct super_block *s;
1461 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1465 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1467 s->s_flags |= SB_ACTIVE;
1469 error = reconfigure_single(s, flags, data);
1471 if (unlikely(error)) {
1472 deactivate_locked_super(s);
1473 return ERR_PTR(error);
1475 return dget(s->s_root);
1477 EXPORT_SYMBOL(mount_single);
1480 * vfs_get_tree - Get the mountable root
1481 * @fc: The superblock configuration context.
1483 * The filesystem is invoked to get or create a superblock which can then later
1484 * be used for mounting. The filesystem places a pointer to the root to be
1485 * used for mounting in @fc->root.
1487 int vfs_get_tree(struct fs_context *fc)
1489 struct super_block *sb;
1495 /* Get the mountable root in fc->root, with a ref on the root and a ref
1496 * on the superblock.
1498 error = fc->ops->get_tree(fc);
1503 pr_err("Filesystem %s get_tree() didn't set fc->root\n",
1505 /* We don't know what the locking state of the superblock is -
1506 * if there is a superblock.
1511 sb = fc->root->d_sb;
1512 WARN_ON(!sb->s_bdi);
1515 * Write barrier is for super_cache_count(). We place it before setting
1516 * SB_BORN as the data dependency between the two functions is the
1517 * superblock structure contents that we just set up, not the SB_BORN
1521 sb->s_flags |= SB_BORN;
1523 error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
1524 if (unlikely(error)) {
1530 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1531 * but s_maxbytes was an unsigned long long for many releases. Throw
1532 * this warning for a little while to try and catch filesystems that
1533 * violate this rule.
1535 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1536 "negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes);
1540 EXPORT_SYMBOL(vfs_get_tree);
1543 * Setup private BDI for given superblock. It gets automatically cleaned up
1544 * in generic_shutdown_super().
1546 int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1548 struct backing_dev_info *bdi;
1552 bdi = bdi_alloc(NUMA_NO_NODE);
1556 va_start(args, fmt);
1557 err = bdi_register_va(bdi, fmt, args);
1563 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1568 EXPORT_SYMBOL(super_setup_bdi_name);
1571 * Setup private BDI for given superblock. I gets automatically cleaned up
1572 * in generic_shutdown_super().
1574 int super_setup_bdi(struct super_block *sb)
1576 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
1578 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name,
1579 atomic_long_inc_return(&bdi_seq));
1581 EXPORT_SYMBOL(super_setup_bdi);
1584 * sb_wait_write - wait until all writers to given file system finish
1585 * @sb: the super for which we wait
1586 * @level: type of writers we wait for (normal vs page fault)
1588 * This function waits until there are no writers of given type to given file
1591 static void sb_wait_write(struct super_block *sb, int level)
1593 percpu_down_write(sb->s_writers.rw_sem + level-1);
1597 * We are going to return to userspace and forget about these locks, the
1598 * ownership goes to the caller of thaw_super() which does unlock().
1600 static void lockdep_sb_freeze_release(struct super_block *sb)
1604 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1605 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1609 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb).
1611 static void lockdep_sb_freeze_acquire(struct super_block *sb)
1615 for (level = 0; level < SB_FREEZE_LEVELS; ++level)
1616 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_);
1619 static void sb_freeze_unlock(struct super_block *sb)
1623 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--)
1624 percpu_up_write(sb->s_writers.rw_sem + level);
1628 * freeze_super - lock the filesystem and force it into a consistent state
1629 * @sb: the super to lock
1631 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1632 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1635 * During this function, sb->s_writers.frozen goes through these values:
1637 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1639 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1640 * writes should be blocked, though page faults are still allowed. We wait for
1641 * all writes to complete and then proceed to the next stage.
1643 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1644 * but internal fs threads can still modify the filesystem (although they
1645 * should not dirty new pages or inodes), writeback can run etc. After waiting
1646 * for all running page faults we sync the filesystem which will clean all
1647 * dirty pages and inodes (no new dirty pages or inodes can be created when
1650 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1651 * modification are blocked (e.g. XFS preallocation truncation on inode
1652 * reclaim). This is usually implemented by blocking new transactions for
1653 * filesystems that have them and need this additional guard. After all
1654 * internal writers are finished we call ->freeze_fs() to finish filesystem
1655 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1656 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1658 * sb->s_writers.frozen is protected by sb->s_umount.
1660 int freeze_super(struct super_block *sb)
1664 atomic_inc(&sb->s_active);
1665 down_write(&sb->s_umount);
1666 if (sb->s_writers.frozen != SB_UNFROZEN) {
1667 deactivate_locked_super(sb);
1671 if (!(sb->s_flags & SB_BORN)) {
1672 up_write(&sb->s_umount);
1673 return 0; /* sic - it's "nothing to do" */
1676 if (sb_rdonly(sb)) {
1677 /* Nothing to do really... */
1678 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1679 up_write(&sb->s_umount);
1683 sb->s_writers.frozen = SB_FREEZE_WRITE;
1684 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1685 up_write(&sb->s_umount);
1686 sb_wait_write(sb, SB_FREEZE_WRITE);
1687 down_write(&sb->s_umount);
1689 /* Now we go and block page faults... */
1690 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1691 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1693 /* All writers are done so after syncing there won't be dirty data */
1694 sync_filesystem(sb);
1696 /* Now wait for internal filesystem counter */
1697 sb->s_writers.frozen = SB_FREEZE_FS;
1698 sb_wait_write(sb, SB_FREEZE_FS);
1700 if (sb->s_op->freeze_fs) {
1701 ret = sb->s_op->freeze_fs(sb);
1704 "VFS:Filesystem freeze failed\n");
1705 sb->s_writers.frozen = SB_UNFROZEN;
1706 sb_freeze_unlock(sb);
1707 wake_up(&sb->s_writers.wait_unfrozen);
1708 deactivate_locked_super(sb);
1713 * For debugging purposes so that fs can warn if it sees write activity
1714 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1716 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1717 lockdep_sb_freeze_release(sb);
1718 up_write(&sb->s_umount);
1721 EXPORT_SYMBOL(freeze_super);
1723 static int thaw_super_locked(struct super_block *sb)
1727 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1728 up_write(&sb->s_umount);
1732 if (sb_rdonly(sb)) {
1733 sb->s_writers.frozen = SB_UNFROZEN;
1737 lockdep_sb_freeze_acquire(sb);
1739 if (sb->s_op->unfreeze_fs) {
1740 error = sb->s_op->unfreeze_fs(sb);
1743 "VFS:Filesystem thaw failed\n");
1744 lockdep_sb_freeze_release(sb);
1745 up_write(&sb->s_umount);
1750 sb->s_writers.frozen = SB_UNFROZEN;
1751 sb_freeze_unlock(sb);
1753 wake_up(&sb->s_writers.wait_unfrozen);
1754 deactivate_locked_super(sb);
1759 * thaw_super -- unlock filesystem
1760 * @sb: the super to thaw
1762 * Unlocks the filesystem and marks it writeable again after freeze_super().
1764 int thaw_super(struct super_block *sb)
1766 down_write(&sb->s_umount);
1767 return thaw_super_locked(sb);
1769 EXPORT_SYMBOL(thaw_super);