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/fscrypt.h>
35 #include <linux/fsnotify.h>
36 #include <linux/lockdep.h>
37 #include <linux/user_namespace.h>
38 #include <linux/fs_context.h>
39 #include <uapi/linux/mount.h>
42 static int thaw_super_locked(struct super_block *sb);
44 static LIST_HEAD(super_blocks);
45 static DEFINE_SPINLOCK(sb_lock);
47 static char *sb_writers_name[SB_FREEZE_LEVELS] = {
54 * One thing we have to be careful of with a per-sb shrinker is that we don't
55 * drop the last active reference to the superblock from within the shrinker.
56 * If that happens we could trigger unregistering the shrinker from within the
57 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
58 * take a passive reference to the superblock to avoid this from occurring.
60 static unsigned long super_cache_scan(struct shrinker *shrink,
61 struct shrink_control *sc)
63 struct super_block *sb;
70 sb = container_of(shrink, struct super_block, s_shrink);
73 * Deadlock avoidance. We may hold various FS locks, and we don't want
74 * to recurse into the FS that called us in clear_inode() and friends..
76 if (!(sc->gfp_mask & __GFP_FS))
79 if (!trylock_super(sb))
82 if (sb->s_op->nr_cached_objects)
83 fs_objects = sb->s_op->nr_cached_objects(sb, sc);
85 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc);
86 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc);
87 total_objects = dentries + inodes + fs_objects + 1;
91 /* proportion the scan between the caches */
92 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
93 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
94 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects);
97 * prune the dcache first as the icache is pinned by it, then
98 * prune the icache, followed by the filesystem specific caches
100 * Ensure that we always scan at least one object - memcg kmem
101 * accounting uses this to fully empty the caches.
103 sc->nr_to_scan = dentries + 1;
104 freed = prune_dcache_sb(sb, sc);
105 sc->nr_to_scan = inodes + 1;
106 freed += prune_icache_sb(sb, sc);
109 sc->nr_to_scan = fs_objects + 1;
110 freed += sb->s_op->free_cached_objects(sb, sc);
113 up_read(&sb->s_umount);
117 static unsigned long super_cache_count(struct shrinker *shrink,
118 struct shrink_control *sc)
120 struct super_block *sb;
121 long total_objects = 0;
123 sb = container_of(shrink, struct super_block, s_shrink);
126 * We don't call trylock_super() here as it is a scalability bottleneck,
127 * so we're exposed to partial setup state. The shrinker rwsem does not
128 * protect filesystem operations backing list_lru_shrink_count() or
129 * s_op->nr_cached_objects(). Counts can change between
130 * super_cache_count and super_cache_scan, so we really don't need locks
133 * However, if we are currently mounting the superblock, the underlying
134 * filesystem might be in a state of partial construction and hence it
135 * is dangerous to access it. trylock_super() uses a SB_BORN check to
136 * avoid this situation, so do the same here. The memory barrier is
137 * matched with the one in mount_fs() as we don't hold locks here.
139 if (!(sb->s_flags & SB_BORN))
143 if (sb->s_op && sb->s_op->nr_cached_objects)
144 total_objects = sb->s_op->nr_cached_objects(sb, sc);
146 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc);
147 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc);
152 total_objects = vfs_pressure_ratio(total_objects);
153 return total_objects;
156 static void destroy_super_work(struct work_struct *work)
158 struct super_block *s = container_of(work, struct super_block,
162 for (i = 0; i < SB_FREEZE_LEVELS; i++)
163 percpu_free_rwsem(&s->s_writers.rw_sem[i]);
167 static void destroy_super_rcu(struct rcu_head *head)
169 struct super_block *s = container_of(head, struct super_block, rcu);
170 INIT_WORK(&s->destroy_work, destroy_super_work);
171 schedule_work(&s->destroy_work);
174 /* Free a superblock that has never been seen by anyone */
175 static void destroy_unused_super(struct super_block *s)
179 up_write(&s->s_umount);
180 list_lru_destroy(&s->s_dentry_lru);
181 list_lru_destroy(&s->s_inode_lru);
183 put_user_ns(s->s_user_ns);
185 free_prealloced_shrinker(&s->s_shrink);
186 /* no delays needed */
187 destroy_super_work(&s->destroy_work);
191 * alloc_super - create new superblock
192 * @type: filesystem type superblock should belong to
193 * @flags: the mount flags
194 * @user_ns: User namespace for the super_block
196 * Allocates and initializes a new &struct super_block. alloc_super()
197 * returns a pointer new superblock or %NULL if allocation had failed.
199 static struct super_block *alloc_super(struct file_system_type *type, int flags,
200 struct user_namespace *user_ns)
202 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
203 static const struct super_operations default_op;
209 INIT_LIST_HEAD(&s->s_mounts);
210 s->s_user_ns = get_user_ns(user_ns);
211 init_rwsem(&s->s_umount);
212 lockdep_set_class(&s->s_umount, &type->s_umount_key);
214 * sget() can have s_umount recursion.
216 * When it cannot find a suitable sb, it allocates a new
217 * one (this one), and tries again to find a suitable old
220 * In case that succeeds, it will acquire the s_umount
221 * lock of the old one. Since these are clearly distrinct
222 * locks, and this object isn't exposed yet, there's no
225 * Annotate this by putting this lock in a different
228 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
230 if (security_sb_alloc(s))
233 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
234 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i],
236 &type->s_writers_key[i]))
239 init_waitqueue_head(&s->s_writers.wait_unfrozen);
240 s->s_bdi = &noop_backing_dev_info;
242 if (s->s_user_ns != &init_user_ns)
243 s->s_iflags |= SB_I_NODEV;
244 INIT_HLIST_NODE(&s->s_instances);
245 INIT_HLIST_BL_HEAD(&s->s_roots);
246 mutex_init(&s->s_sync_lock);
247 INIT_LIST_HEAD(&s->s_inodes);
248 spin_lock_init(&s->s_inode_list_lock);
249 INIT_LIST_HEAD(&s->s_inodes_wb);
250 spin_lock_init(&s->s_inode_wblist_lock);
253 atomic_set(&s->s_active, 1);
254 mutex_init(&s->s_vfs_rename_mutex);
255 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
256 init_rwsem(&s->s_dquot.dqio_sem);
257 s->s_maxbytes = MAX_NON_LFS;
258 s->s_op = &default_op;
259 s->s_time_gran = 1000000000;
260 s->s_time_min = TIME64_MIN;
261 s->s_time_max = TIME64_MAX;
263 s->s_shrink.seeks = DEFAULT_SEEKS;
264 s->s_shrink.scan_objects = super_cache_scan;
265 s->s_shrink.count_objects = super_cache_count;
266 s->s_shrink.batch = 1024;
267 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE;
268 if (prealloc_shrinker(&s->s_shrink))
270 if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink))
272 if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink))
277 destroy_unused_super(s);
281 /* Superblock refcounting */
284 * Drop a superblock's refcount. The caller must hold sb_lock.
286 static void __put_super(struct super_block *s)
289 list_del_init(&s->s_list);
290 WARN_ON(s->s_dentry_lru.node);
291 WARN_ON(s->s_inode_lru.node);
292 WARN_ON(!list_empty(&s->s_mounts));
295 put_user_ns(s->s_user_ns);
297 call_rcu(&s->rcu, destroy_super_rcu);
302 * put_super - drop a temporary reference to superblock
303 * @sb: superblock in question
305 * Drops a temporary reference, frees superblock if there's no
308 void put_super(struct super_block *sb)
312 spin_unlock(&sb_lock);
317 * deactivate_locked_super - drop an active reference to superblock
318 * @s: superblock to deactivate
320 * Drops an active reference to superblock, converting it into a temporary
321 * one if there is no other active references left. In that case we
322 * tell fs driver to shut it down and drop the temporary reference we
325 * Caller holds exclusive lock on superblock; that lock is released.
327 void deactivate_locked_super(struct super_block *s)
329 struct file_system_type *fs = s->s_type;
330 if (atomic_dec_and_test(&s->s_active)) {
331 unregister_shrinker(&s->s_shrink);
335 * Since list_lru_destroy() may sleep, we cannot call it from
336 * put_super(), where we hold the sb_lock. Therefore we destroy
337 * the lru lists right now.
339 list_lru_destroy(&s->s_dentry_lru);
340 list_lru_destroy(&s->s_inode_lru);
345 up_write(&s->s_umount);
349 EXPORT_SYMBOL(deactivate_locked_super);
352 * deactivate_super - drop an active reference to superblock
353 * @s: superblock to deactivate
355 * Variant of deactivate_locked_super(), except that superblock is *not*
356 * locked by caller. If we are going to drop the final active reference,
357 * lock will be acquired prior to that.
359 void deactivate_super(struct super_block *s)
361 if (!atomic_add_unless(&s->s_active, -1, 1)) {
362 down_write(&s->s_umount);
363 deactivate_locked_super(s);
367 EXPORT_SYMBOL(deactivate_super);
370 * grab_super - acquire an active reference
371 * @s: reference we are trying to make active
373 * Tries to acquire an active reference. grab_super() is used when we
374 * had just found a superblock in super_blocks or fs_type->fs_supers
375 * and want to turn it into a full-blown active reference. grab_super()
376 * is called with sb_lock held and drops it. Returns 1 in case of
377 * success, 0 if we had failed (superblock contents was already dead or
378 * dying when grab_super() had been called). Note that this is only
379 * called for superblocks not in rundown mode (== ones still on ->fs_supers
380 * of their type), so increment of ->s_count is OK here.
382 static int grab_super(struct super_block *s) __releases(sb_lock)
385 spin_unlock(&sb_lock);
386 down_write(&s->s_umount);
387 if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) {
391 up_write(&s->s_umount);
397 * trylock_super - try to grab ->s_umount shared
398 * @sb: reference we are trying to grab
400 * Try to prevent fs shutdown. This is used in places where we
401 * cannot take an active reference but we need to ensure that the
402 * filesystem is not shut down while we are working on it. It returns
403 * false if we cannot acquire s_umount or if we lose the race and
404 * filesystem already got into shutdown, and returns true with the s_umount
405 * lock held in read mode in case of success. On successful return,
406 * the caller must drop the s_umount lock when done.
408 * Note that unlike get_super() et.al. this one does *not* bump ->s_count.
409 * The reason why it's safe is that we are OK with doing trylock instead
410 * of down_read(). There's a couple of places that are OK with that, but
411 * it's very much not a general-purpose interface.
413 bool trylock_super(struct super_block *sb)
415 if (down_read_trylock(&sb->s_umount)) {
416 if (!hlist_unhashed(&sb->s_instances) &&
417 sb->s_root && (sb->s_flags & SB_BORN))
419 up_read(&sb->s_umount);
426 * generic_shutdown_super - common helper for ->kill_sb()
427 * @sb: superblock to kill
429 * generic_shutdown_super() does all fs-independent work on superblock
430 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
431 * that need destruction out of superblock, call generic_shutdown_super()
432 * and release aforementioned objects. Note: dentries and inodes _are_
433 * taken care of and do not need specific handling.
435 * Upon calling this function, the filesystem may no longer alter or
436 * rearrange the set of dentries belonging to this super_block, nor may it
437 * change the attachments of dentries to inodes.
439 void generic_shutdown_super(struct super_block *sb)
441 const struct super_operations *sop = sb->s_op;
444 shrink_dcache_for_umount(sb);
446 sb->s_flags &= ~SB_ACTIVE;
448 cgroup_writeback_umount();
450 /* evict all inodes with zero refcount */
452 /* only nonzero refcount inodes can have marks */
453 fsnotify_sb_delete(sb);
454 security_sb_delete(sb);
456 if (sb->s_dio_done_wq) {
457 destroy_workqueue(sb->s_dio_done_wq);
458 sb->s_dio_done_wq = NULL;
464 if (!list_empty(&sb->s_inodes)) {
465 printk("VFS: Busy inodes after unmount of %s. "
466 "Self-destruct in 5 seconds. Have a nice day...\n",
471 /* should be initialized for __put_super_and_need_restart() */
472 hlist_del_init(&sb->s_instances);
473 spin_unlock(&sb_lock);
474 up_write(&sb->s_umount);
475 if (sb->s_bdi != &noop_backing_dev_info) {
476 if (sb->s_iflags & SB_I_PERSB_BDI)
477 bdi_unregister(sb->s_bdi);
479 sb->s_bdi = &noop_backing_dev_info;
483 EXPORT_SYMBOL(generic_shutdown_super);
485 bool mount_capable(struct fs_context *fc)
487 if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT))
488 return capable(CAP_SYS_ADMIN);
490 return ns_capable(fc->user_ns, CAP_SYS_ADMIN);
494 * sget_fc - Find or create a superblock
495 * @fc: Filesystem context.
496 * @test: Comparison callback
497 * @set: Setup callback
499 * Find or create a superblock using the parameters stored in the filesystem
500 * context and the two callback functions.
502 * If an extant superblock is matched, then that will be returned with an
503 * elevated reference count that the caller must transfer or discard.
505 * If no match is made, a new superblock will be allocated and basic
506 * initialisation will be performed (s_type, s_fs_info and s_id will be set and
507 * the set() callback will be invoked), the superblock will be published and it
508 * will be returned in a partially constructed state with SB_BORN and SB_ACTIVE
511 struct super_block *sget_fc(struct fs_context *fc,
512 int (*test)(struct super_block *, struct fs_context *),
513 int (*set)(struct super_block *, struct fs_context *))
515 struct super_block *s = NULL;
516 struct super_block *old;
517 struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns;
523 hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) {
525 goto share_extant_sb;
529 spin_unlock(&sb_lock);
530 s = alloc_super(fc->fs_type, fc->sb_flags, user_ns);
532 return ERR_PTR(-ENOMEM);
536 s->s_fs_info = fc->s_fs_info;
540 spin_unlock(&sb_lock);
541 destroy_unused_super(s);
544 fc->s_fs_info = NULL;
545 s->s_type = fc->fs_type;
546 s->s_iflags |= fc->s_iflags;
547 strlcpy(s->s_id, s->s_type->name, sizeof(s->s_id));
548 list_add_tail(&s->s_list, &super_blocks);
549 hlist_add_head(&s->s_instances, &s->s_type->fs_supers);
550 spin_unlock(&sb_lock);
551 get_filesystem(s->s_type);
552 register_shrinker_prepared(&s->s_shrink);
556 if (user_ns != old->s_user_ns) {
557 spin_unlock(&sb_lock);
558 destroy_unused_super(s);
559 return ERR_PTR(-EBUSY);
561 if (!grab_super(old))
563 destroy_unused_super(s);
566 EXPORT_SYMBOL(sget_fc);
569 * sget - find or create a superblock
570 * @type: filesystem type superblock should belong to
571 * @test: comparison callback
572 * @set: setup callback
573 * @flags: mount flags
574 * @data: argument to each of them
576 struct super_block *sget(struct file_system_type *type,
577 int (*test)(struct super_block *,void *),
578 int (*set)(struct super_block *,void *),
582 struct user_namespace *user_ns = current_user_ns();
583 struct super_block *s = NULL;
584 struct super_block *old;
587 /* We don't yet pass the user namespace of the parent
588 * mount through to here so always use &init_user_ns
589 * until that changes.
591 if (flags & SB_SUBMOUNT)
592 user_ns = &init_user_ns;
597 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
598 if (!test(old, data))
600 if (user_ns != old->s_user_ns) {
601 spin_unlock(&sb_lock);
602 destroy_unused_super(s);
603 return ERR_PTR(-EBUSY);
605 if (!grab_super(old))
607 destroy_unused_super(s);
612 spin_unlock(&sb_lock);
613 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns);
615 return ERR_PTR(-ENOMEM);
621 spin_unlock(&sb_lock);
622 destroy_unused_super(s);
626 strlcpy(s->s_id, type->name, sizeof(s->s_id));
627 list_add_tail(&s->s_list, &super_blocks);
628 hlist_add_head(&s->s_instances, &type->fs_supers);
629 spin_unlock(&sb_lock);
630 get_filesystem(type);
631 register_shrinker_prepared(&s->s_shrink);
636 void drop_super(struct super_block *sb)
638 up_read(&sb->s_umount);
642 EXPORT_SYMBOL(drop_super);
644 void drop_super_exclusive(struct super_block *sb)
646 up_write(&sb->s_umount);
649 EXPORT_SYMBOL(drop_super_exclusive);
651 static void __iterate_supers(void (*f)(struct super_block *))
653 struct super_block *sb, *p = NULL;
656 list_for_each_entry(sb, &super_blocks, s_list) {
657 if (hlist_unhashed(&sb->s_instances))
660 spin_unlock(&sb_lock);
671 spin_unlock(&sb_lock);
674 * iterate_supers - call function for all active superblocks
675 * @f: function to call
676 * @arg: argument to pass to it
678 * Scans the superblock list and calls given function, passing it
679 * locked superblock and given argument.
681 void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
683 struct super_block *sb, *p = NULL;
686 list_for_each_entry(sb, &super_blocks, s_list) {
687 if (hlist_unhashed(&sb->s_instances))
690 spin_unlock(&sb_lock);
692 down_read(&sb->s_umount);
693 if (sb->s_root && (sb->s_flags & SB_BORN))
695 up_read(&sb->s_umount);
704 spin_unlock(&sb_lock);
708 * iterate_supers_type - call function for superblocks of given type
710 * @f: function to call
711 * @arg: argument to pass to it
713 * Scans the superblock list and calls given function, passing it
714 * locked superblock and given argument.
716 void iterate_supers_type(struct file_system_type *type,
717 void (*f)(struct super_block *, void *), void *arg)
719 struct super_block *sb, *p = NULL;
722 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
724 spin_unlock(&sb_lock);
726 down_read(&sb->s_umount);
727 if (sb->s_root && (sb->s_flags & SB_BORN))
729 up_read(&sb->s_umount);
738 spin_unlock(&sb_lock);
741 EXPORT_SYMBOL(iterate_supers_type);
744 * get_super - get the superblock of a device
745 * @bdev: device to get the superblock for
747 * Scans the superblock list and finds the superblock of the file system
748 * mounted on the device given. %NULL is returned if no match is found.
750 struct super_block *get_super(struct block_device *bdev)
752 struct super_block *sb;
759 list_for_each_entry(sb, &super_blocks, s_list) {
760 if (hlist_unhashed(&sb->s_instances))
762 if (sb->s_bdev == bdev) {
764 spin_unlock(&sb_lock);
765 down_read(&sb->s_umount);
767 if (sb->s_root && (sb->s_flags & SB_BORN))
769 up_read(&sb->s_umount);
770 /* nope, got unmounted */
776 spin_unlock(&sb_lock);
781 * get_active_super - get an active reference to the superblock of a device
782 * @bdev: device to get the superblock for
784 * Scans the superblock list and finds the superblock of the file system
785 * mounted on the device given. Returns the superblock with an active
786 * reference or %NULL if none was found.
788 struct super_block *get_active_super(struct block_device *bdev)
790 struct super_block *sb;
797 list_for_each_entry(sb, &super_blocks, s_list) {
798 if (hlist_unhashed(&sb->s_instances))
800 if (sb->s_bdev == bdev) {
803 up_write(&sb->s_umount);
807 spin_unlock(&sb_lock);
811 struct super_block *user_get_super(dev_t dev, bool excl)
813 struct super_block *sb;
817 list_for_each_entry(sb, &super_blocks, s_list) {
818 if (hlist_unhashed(&sb->s_instances))
820 if (sb->s_dev == dev) {
822 spin_unlock(&sb_lock);
824 down_write(&sb->s_umount);
826 down_read(&sb->s_umount);
828 if (sb->s_root && (sb->s_flags & SB_BORN))
831 up_write(&sb->s_umount);
833 up_read(&sb->s_umount);
834 /* nope, got unmounted */
840 spin_unlock(&sb_lock);
845 * reconfigure_super - asks filesystem to change superblock parameters
846 * @fc: The superblock and configuration
848 * Alters the configuration parameters of a live superblock.
850 int reconfigure_super(struct fs_context *fc)
852 struct super_block *sb = fc->root->d_sb;
854 bool remount_ro = false;
855 bool force = fc->sb_flags & SB_FORCE;
857 if (fc->sb_flags_mask & ~MS_RMT_MASK)
859 if (sb->s_writers.frozen != SB_UNFROZEN)
862 retval = security_sb_remount(sb, fc->security);
866 if (fc->sb_flags_mask & SB_RDONLY) {
868 if (!(fc->sb_flags & SB_RDONLY) && sb->s_bdev &&
869 bdev_read_only(sb->s_bdev))
873 remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb);
877 if (!hlist_empty(&sb->s_pins)) {
878 up_write(&sb->s_umount);
879 group_pin_kill(&sb->s_pins);
880 down_write(&sb->s_umount);
883 if (sb->s_writers.frozen != SB_UNFROZEN)
885 remount_ro = !sb_rdonly(sb);
888 shrink_dcache_sb(sb);
890 /* If we are reconfiguring to RDONLY and current sb is read/write,
891 * make sure there are no files open for writing.
895 sb->s_readonly_remount = 1;
898 retval = sb_prepare_remount_readonly(sb);
904 if (fc->ops->reconfigure) {
905 retval = fc->ops->reconfigure(fc);
908 goto cancel_readonly;
909 /* If forced remount, go ahead despite any errors */
910 WARN(1, "forced remount of a %s fs returned %i\n",
911 sb->s_type->name, retval);
915 WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) |
916 (fc->sb_flags & fc->sb_flags_mask)));
917 /* Needs to be ordered wrt mnt_is_readonly() */
919 sb->s_readonly_remount = 0;
922 * Some filesystems modify their metadata via some other path than the
923 * bdev buffer cache (eg. use a private mapping, or directories in
924 * pagecache, etc). Also file data modifications go via their own
925 * mappings. So If we try to mount readonly then copy the filesystem
926 * from bdev, we could get stale data, so invalidate it to give a best
927 * effort at coherency.
929 if (remount_ro && sb->s_bdev)
930 invalidate_bdev(sb->s_bdev);
934 sb->s_readonly_remount = 0;
938 static void do_emergency_remount_callback(struct super_block *sb)
940 down_write(&sb->s_umount);
941 if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) &&
943 struct fs_context *fc;
945 fc = fs_context_for_reconfigure(sb->s_root,
946 SB_RDONLY | SB_FORCE, SB_RDONLY);
948 if (parse_monolithic_mount_data(fc, NULL) == 0)
949 (void)reconfigure_super(fc);
953 up_write(&sb->s_umount);
956 static void do_emergency_remount(struct work_struct *work)
958 __iterate_supers(do_emergency_remount_callback);
960 printk("Emergency Remount complete\n");
963 void emergency_remount(void)
965 struct work_struct *work;
967 work = kmalloc(sizeof(*work), GFP_ATOMIC);
969 INIT_WORK(work, do_emergency_remount);
974 static void do_thaw_all_callback(struct super_block *sb)
976 down_write(&sb->s_umount);
977 if (sb->s_root && sb->s_flags & SB_BORN) {
978 emergency_thaw_bdev(sb);
979 thaw_super_locked(sb);
981 up_write(&sb->s_umount);
985 static void do_thaw_all(struct work_struct *work)
987 __iterate_supers(do_thaw_all_callback);
989 printk(KERN_WARNING "Emergency Thaw complete\n");
993 * emergency_thaw_all -- forcibly thaw every frozen filesystem
995 * Used for emergency unfreeze of all filesystems via SysRq
997 void emergency_thaw_all(void)
999 struct work_struct *work;
1001 work = kmalloc(sizeof(*work), GFP_ATOMIC);
1003 INIT_WORK(work, do_thaw_all);
1004 schedule_work(work);
1008 static DEFINE_IDA(unnamed_dev_ida);
1011 * get_anon_bdev - Allocate a block device for filesystems which don't have one.
1012 * @p: Pointer to a dev_t.
1014 * Filesystems which don't use real block devices can call this function
1015 * to allocate a virtual block device.
1017 * Context: Any context. Frequently called while holding sb_lock.
1018 * Return: 0 on success, -EMFILE if there are no anonymous bdevs left
1019 * or -ENOMEM if memory allocation failed.
1021 int get_anon_bdev(dev_t *p)
1026 * Many userspace utilities consider an FSID of 0 invalid.
1027 * Always return at least 1 from get_anon_bdev.
1029 dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1,
1039 EXPORT_SYMBOL(get_anon_bdev);
1041 void free_anon_bdev(dev_t dev)
1043 ida_free(&unnamed_dev_ida, MINOR(dev));
1045 EXPORT_SYMBOL(free_anon_bdev);
1047 int set_anon_super(struct super_block *s, void *data)
1049 return get_anon_bdev(&s->s_dev);
1051 EXPORT_SYMBOL(set_anon_super);
1053 void kill_anon_super(struct super_block *sb)
1055 dev_t dev = sb->s_dev;
1056 generic_shutdown_super(sb);
1057 free_anon_bdev(dev);
1059 EXPORT_SYMBOL(kill_anon_super);
1061 void kill_litter_super(struct super_block *sb)
1064 d_genocide(sb->s_root);
1065 kill_anon_super(sb);
1067 EXPORT_SYMBOL(kill_litter_super);
1069 int set_anon_super_fc(struct super_block *sb, struct fs_context *fc)
1071 return set_anon_super(sb, NULL);
1073 EXPORT_SYMBOL(set_anon_super_fc);
1075 static int test_keyed_super(struct super_block *sb, struct fs_context *fc)
1077 return sb->s_fs_info == fc->s_fs_info;
1080 static int test_single_super(struct super_block *s, struct fs_context *fc)
1086 * vfs_get_super - Get a superblock with a search key set in s_fs_info.
1087 * @fc: The filesystem context holding the parameters
1088 * @keying: How to distinguish superblocks
1089 * @fill_super: Helper to initialise a new superblock
1091 * Search for a superblock and create a new one if not found. The search
1092 * criterion is controlled by @keying. If the search fails, a new superblock
1093 * is created and @fill_super() is called to initialise it.
1095 * @keying can take one of a number of values:
1097 * (1) vfs_get_single_super - Only one superblock of this type may exist on the
1098 * system. This is typically used for special system filesystems.
1100 * (2) vfs_get_keyed_super - Multiple superblocks may exist, but they must have
1101 * distinct keys (where the key is in s_fs_info). Searching for the same
1102 * key again will turn up the superblock for that key.
1104 * (3) vfs_get_independent_super - Multiple superblocks may exist and are
1105 * unkeyed. Each call will get a new superblock.
1107 * A permissions check is made by sget_fc() unless we're getting a superblock
1108 * for a kernel-internal mount or a submount.
1110 int vfs_get_super(struct fs_context *fc,
1111 enum vfs_get_super_keying keying,
1112 int (*fill_super)(struct super_block *sb,
1113 struct fs_context *fc))
1115 int (*test)(struct super_block *, struct fs_context *);
1116 struct super_block *sb;
1120 case vfs_get_single_super:
1121 case vfs_get_single_reconf_super:
1122 test = test_single_super;
1124 case vfs_get_keyed_super:
1125 test = test_keyed_super;
1127 case vfs_get_independent_super:
1134 sb = sget_fc(fc, test, set_anon_super_fc);
1139 err = fill_super(sb, fc);
1143 sb->s_flags |= SB_ACTIVE;
1144 fc->root = dget(sb->s_root);
1146 fc->root = dget(sb->s_root);
1147 if (keying == vfs_get_single_reconf_super) {
1148 err = reconfigure_super(fc);
1160 deactivate_locked_super(sb);
1163 EXPORT_SYMBOL(vfs_get_super);
1165 int get_tree_nodev(struct fs_context *fc,
1166 int (*fill_super)(struct super_block *sb,
1167 struct fs_context *fc))
1169 return vfs_get_super(fc, vfs_get_independent_super, fill_super);
1171 EXPORT_SYMBOL(get_tree_nodev);
1173 int get_tree_single(struct fs_context *fc,
1174 int (*fill_super)(struct super_block *sb,
1175 struct fs_context *fc))
1177 return vfs_get_super(fc, vfs_get_single_super, fill_super);
1179 EXPORT_SYMBOL(get_tree_single);
1181 int get_tree_single_reconf(struct fs_context *fc,
1182 int (*fill_super)(struct super_block *sb,
1183 struct fs_context *fc))
1185 return vfs_get_super(fc, vfs_get_single_reconf_super, fill_super);
1187 EXPORT_SYMBOL(get_tree_single_reconf);
1189 int get_tree_keyed(struct fs_context *fc,
1190 int (*fill_super)(struct super_block *sb,
1191 struct fs_context *fc),
1194 fc->s_fs_info = key;
1195 return vfs_get_super(fc, vfs_get_keyed_super, fill_super);
1197 EXPORT_SYMBOL(get_tree_keyed);
1201 static int set_bdev_super(struct super_block *s, void *data)
1204 s->s_dev = s->s_bdev->bd_dev;
1205 s->s_bdi = bdi_get(s->s_bdev->bd_disk->bdi);
1207 if (bdev_stable_writes(s->s_bdev))
1208 s->s_iflags |= SB_I_STABLE_WRITES;
1212 static int set_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1214 return set_bdev_super(s, fc->sget_key);
1217 static int test_bdev_super_fc(struct super_block *s, struct fs_context *fc)
1219 return s->s_bdev == fc->sget_key;
1223 * get_tree_bdev - Get a superblock based on a single block device
1224 * @fc: The filesystem context holding the parameters
1225 * @fill_super: Helper to initialise a new superblock
1227 int get_tree_bdev(struct fs_context *fc,
1228 int (*fill_super)(struct super_block *,
1229 struct fs_context *))
1231 struct block_device *bdev;
1232 struct super_block *s;
1233 fmode_t mode = FMODE_READ | FMODE_EXCL;
1236 if (!(fc->sb_flags & SB_RDONLY))
1237 mode |= FMODE_WRITE;
1240 return invalf(fc, "No source specified");
1242 bdev = blkdev_get_by_path(fc->source, mode, fc->fs_type);
1244 errorf(fc, "%s: Can't open blockdev", fc->source);
1245 return PTR_ERR(bdev);
1248 /* Once the superblock is inserted into the list by sget_fc(), s_umount
1249 * will protect the lockfs code from trying to start a snapshot while
1252 mutex_lock(&bdev->bd_fsfreeze_mutex);
1253 if (bdev->bd_fsfreeze_count > 0) {
1254 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1255 warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev);
1256 blkdev_put(bdev, mode);
1260 fc->sb_flags |= SB_NOSEC;
1261 fc->sget_key = bdev;
1262 s = sget_fc(fc, test_bdev_super_fc, set_bdev_super_fc);
1263 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1265 blkdev_put(bdev, mode);
1270 /* Don't summarily change the RO/RW state. */
1271 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1272 warnf(fc, "%pg: Can't mount, would change RO state", bdev);
1273 deactivate_locked_super(s);
1274 blkdev_put(bdev, mode);
1279 * s_umount nests inside open_mutex during
1280 * __invalidate_device(). blkdev_put() acquires
1281 * open_mutex and can't be called under s_umount. Drop
1282 * s_umount temporarily. This is safe as we're
1283 * holding an active reference.
1285 up_write(&s->s_umount);
1286 blkdev_put(bdev, mode);
1287 down_write(&s->s_umount);
1290 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1291 sb_set_blocksize(s, block_size(bdev));
1292 error = fill_super(s, fc);
1294 deactivate_locked_super(s);
1298 s->s_flags |= SB_ACTIVE;
1303 fc->root = dget(s->s_root);
1306 EXPORT_SYMBOL(get_tree_bdev);
1308 static int test_bdev_super(struct super_block *s, void *data)
1310 return (void *)s->s_bdev == data;
1313 struct dentry *mount_bdev(struct file_system_type *fs_type,
1314 int flags, const char *dev_name, void *data,
1315 int (*fill_super)(struct super_block *, void *, int))
1317 struct block_device *bdev;
1318 struct super_block *s;
1319 fmode_t mode = FMODE_READ | FMODE_EXCL;
1322 if (!(flags & SB_RDONLY))
1323 mode |= FMODE_WRITE;
1325 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1327 return ERR_CAST(bdev);
1330 * once the super is inserted into the list by sget, s_umount
1331 * will protect the lockfs code from trying to start a snapshot
1332 * while we are mounting
1334 mutex_lock(&bdev->bd_fsfreeze_mutex);
1335 if (bdev->bd_fsfreeze_count > 0) {
1336 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1340 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC,
1342 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1347 if ((flags ^ s->s_flags) & SB_RDONLY) {
1348 deactivate_locked_super(s);
1354 * s_umount nests inside open_mutex during
1355 * __invalidate_device(). blkdev_put() acquires
1356 * open_mutex and can't be called under s_umount. Drop
1357 * s_umount temporarily. This is safe as we're
1358 * holding an active reference.
1360 up_write(&s->s_umount);
1361 blkdev_put(bdev, mode);
1362 down_write(&s->s_umount);
1365 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1366 sb_set_blocksize(s, block_size(bdev));
1367 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1369 deactivate_locked_super(s);
1373 s->s_flags |= SB_ACTIVE;
1377 return dget(s->s_root);
1382 blkdev_put(bdev, mode);
1384 return ERR_PTR(error);
1386 EXPORT_SYMBOL(mount_bdev);
1388 void kill_block_super(struct super_block *sb)
1390 struct block_device *bdev = sb->s_bdev;
1391 fmode_t mode = sb->s_mode;
1393 bdev->bd_super = NULL;
1394 generic_shutdown_super(sb);
1395 sync_blockdev(bdev);
1396 WARN_ON_ONCE(!(mode & FMODE_EXCL));
1397 blkdev_put(bdev, mode | FMODE_EXCL);
1400 EXPORT_SYMBOL(kill_block_super);
1403 struct dentry *mount_nodev(struct file_system_type *fs_type,
1404 int flags, void *data,
1405 int (*fill_super)(struct super_block *, void *, int))
1408 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
1413 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1415 deactivate_locked_super(s);
1416 return ERR_PTR(error);
1418 s->s_flags |= SB_ACTIVE;
1419 return dget(s->s_root);
1421 EXPORT_SYMBOL(mount_nodev);
1423 int reconfigure_single(struct super_block *s,
1424 int flags, void *data)
1426 struct fs_context *fc;
1429 /* The caller really need to be passing fc down into mount_single(),
1430 * then a chunk of this can be removed. [Bollocks -- AV]
1431 * Better yet, reconfiguration shouldn't happen, but rather the second
1432 * mount should be rejected if the parameters are not compatible.
1434 fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK);
1438 ret = parse_monolithic_mount_data(fc, data);
1442 ret = reconfigure_super(fc);
1448 static int compare_single(struct super_block *s, void *p)
1453 struct dentry *mount_single(struct file_system_type *fs_type,
1454 int flags, void *data,
1455 int (*fill_super)(struct super_block *, void *, int))
1457 struct super_block *s;
1460 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
1464 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
1466 s->s_flags |= SB_ACTIVE;
1468 error = reconfigure_single(s, flags, data);
1470 if (unlikely(error)) {
1471 deactivate_locked_super(s);
1472 return ERR_PTR(error);
1474 return dget(s->s_root);
1476 EXPORT_SYMBOL(mount_single);
1479 * vfs_get_tree - Get the mountable root
1480 * @fc: The superblock configuration context.
1482 * The filesystem is invoked to get or create a superblock which can then later
1483 * be used for mounting. The filesystem places a pointer to the root to be
1484 * used for mounting in @fc->root.
1486 int vfs_get_tree(struct fs_context *fc)
1488 struct super_block *sb;
1494 /* Get the mountable root in fc->root, with a ref on the root and a ref
1495 * on the superblock.
1497 error = fc->ops->get_tree(fc);
1502 pr_err("Filesystem %s get_tree() didn't set fc->root\n",
1504 /* We don't know what the locking state of the superblock is -
1505 * if there is a superblock.
1510 sb = fc->root->d_sb;
1511 WARN_ON(!sb->s_bdi);
1514 * Write barrier is for super_cache_count(). We place it before setting
1515 * SB_BORN as the data dependency between the two functions is the
1516 * superblock structure contents that we just set up, not the SB_BORN
1520 sb->s_flags |= SB_BORN;
1522 error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL);
1523 if (unlikely(error)) {
1529 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1530 * but s_maxbytes was an unsigned long long for many releases. Throw
1531 * this warning for a little while to try and catch filesystems that
1532 * violate this rule.
1534 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1535 "negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes);
1539 EXPORT_SYMBOL(vfs_get_tree);
1542 * Setup private BDI for given superblock. It gets automatically cleaned up
1543 * in generic_shutdown_super().
1545 int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1547 struct backing_dev_info *bdi;
1551 bdi = bdi_alloc(NUMA_NO_NODE);
1555 va_start(args, fmt);
1556 err = bdi_register_va(bdi, fmt, args);
1562 WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1564 sb->s_iflags |= SB_I_PERSB_BDI;
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, int level)
1621 for (level--; level >= 0; level--)
1622 percpu_up_write(sb->s_writers.rw_sem + level);
1626 * freeze_super - lock the filesystem and force it into a consistent state
1627 * @sb: the super to lock
1629 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1630 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1633 * During this function, sb->s_writers.frozen goes through these values:
1635 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1637 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1638 * writes should be blocked, though page faults are still allowed. We wait for
1639 * all writes to complete and then proceed to the next stage.
1641 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1642 * but internal fs threads can still modify the filesystem (although they
1643 * should not dirty new pages or inodes), writeback can run etc. After waiting
1644 * for all running page faults we sync the filesystem which will clean all
1645 * dirty pages and inodes (no new dirty pages or inodes can be created when
1648 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1649 * modification are blocked (e.g. XFS preallocation truncation on inode
1650 * reclaim). This is usually implemented by blocking new transactions for
1651 * filesystems that have them and need this additional guard. After all
1652 * internal writers are finished we call ->freeze_fs() to finish filesystem
1653 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1654 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1656 * sb->s_writers.frozen is protected by sb->s_umount.
1658 int freeze_super(struct super_block *sb)
1662 atomic_inc(&sb->s_active);
1663 down_write(&sb->s_umount);
1664 if (sb->s_writers.frozen != SB_UNFROZEN) {
1665 deactivate_locked_super(sb);
1669 if (!(sb->s_flags & SB_BORN)) {
1670 up_write(&sb->s_umount);
1671 return 0; /* sic - it's "nothing to do" */
1674 if (sb_rdonly(sb)) {
1675 /* Nothing to do really... */
1676 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1677 up_write(&sb->s_umount);
1681 sb->s_writers.frozen = SB_FREEZE_WRITE;
1682 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1683 up_write(&sb->s_umount);
1684 sb_wait_write(sb, SB_FREEZE_WRITE);
1685 down_write(&sb->s_umount);
1687 /* Now we go and block page faults... */
1688 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1689 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1691 /* All writers are done so after syncing there won't be dirty data */
1692 ret = sync_filesystem(sb);
1694 sb->s_writers.frozen = SB_UNFROZEN;
1695 sb_freeze_unlock(sb, SB_FREEZE_PAGEFAULT);
1696 wake_up(&sb->s_writers.wait_unfrozen);
1697 deactivate_locked_super(sb);
1701 /* Now wait for internal filesystem counter */
1702 sb->s_writers.frozen = SB_FREEZE_FS;
1703 sb_wait_write(sb, SB_FREEZE_FS);
1705 if (sb->s_op->freeze_fs) {
1706 ret = sb->s_op->freeze_fs(sb);
1709 "VFS:Filesystem freeze failed\n");
1710 sb->s_writers.frozen = SB_UNFROZEN;
1711 sb_freeze_unlock(sb, SB_FREEZE_FS);
1712 wake_up(&sb->s_writers.wait_unfrozen);
1713 deactivate_locked_super(sb);
1718 * For debugging purposes so that fs can warn if it sees write activity
1719 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super().
1721 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
1722 lockdep_sb_freeze_release(sb);
1723 up_write(&sb->s_umount);
1726 EXPORT_SYMBOL(freeze_super);
1728 static int thaw_super_locked(struct super_block *sb)
1732 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) {
1733 up_write(&sb->s_umount);
1737 if (sb_rdonly(sb)) {
1738 sb->s_writers.frozen = SB_UNFROZEN;
1742 lockdep_sb_freeze_acquire(sb);
1744 if (sb->s_op->unfreeze_fs) {
1745 error = sb->s_op->unfreeze_fs(sb);
1748 "VFS:Filesystem thaw failed\n");
1749 lockdep_sb_freeze_release(sb);
1750 up_write(&sb->s_umount);
1755 sb->s_writers.frozen = SB_UNFROZEN;
1756 sb_freeze_unlock(sb, SB_FREEZE_FS);
1758 wake_up(&sb->s_writers.wait_unfrozen);
1759 deactivate_locked_super(sb);
1764 * thaw_super -- unlock filesystem
1765 * @sb: the super to thaw
1767 * Unlocks the filesystem and marks it writeable again after freeze_super().
1769 int thaw_super(struct super_block *sb)
1771 down_write(&sb->s_umount);
1772 return thaw_super_locked(sb);
1774 EXPORT_SYMBOL(thaw_super);